HEALTH MONITORING OF ELECTRICAL ACTUATORS FOR …

HEALTH MONITORING OF ELECTRICALACTUATORS FOR LANDING GEARS

A thesis submitted to the University of Manchester for the degree ofDoctor of Engineering in the Faculty of Engineering and Physical

Science

2012

Paul Anthony Phillips

The School of Mechanical, Aerospace and Civil Engineering

TABLE OF CONTENTS

LIST OF NOTATION vii

ABSTRACT viii

DECLERATION ix

COPYRIGHT x

ACKNOWLEDGEMENTS xi

THE AUTHOR xii

CHAPTER 1: INTRODUCTION 1

1.1 The Degree of Doctor of Engineering 1

1.2 The Sponsoring Company 2

1.3 Division History 3

1.4 Research and Technology 4

1.5 Problem Statement and Research Motivation 5

1.6 Research Objectives 8

1.7 Project Management 9

1.8 Thesis Layout 9

CHAPTER 2: LANDING GEAR, ELECTROMECHANICAL ACTUATORAND SYSTEMS BACKGROUND 12

2.1 Introduction 12

2.2 The “More Electric Aircraft” 12

2.3 More Electric Aircraft Research History 14

2.4 Current Relevant EU Electric Aircraft Research projects 172.4.1 Clean Sky Joint Technology Initiative 17

HEALTH MONITORING OF ELECTRICAL ACTUATORS FOR LANDING GEARS

DEGREE OF DOCTOR OF ENGINEERING (ENGD)

ii

2.4.2 More Open Electrical Technologies 172.4.3 Power Optimised Aircraft 172.4.4 Electric Landing Gear Extension and Retraction (ELGEAR) 18

2.5 Overview of Landing Gears 18

2.6 Actuator Types 212.6.1 Lead Screw 232.6.2 Ball screw 232.6.3 Roller screw 24

2.7 Regulation on Landing Gear Retraction Mechanisms 25

2.8 Messier Dowty Actuator Design 27

2.9 Main Retraction Actuator Control and Performance Requirements 29

2.10 Reliability and Safety issues 32

2.11 Actuator Component Failures 342.11.1 Bearing Faults 342.11.2 Gear Faults 352.11.3 Roller Screw Failure 35

2.12 Conclusion 36

2.13 References 37

CHAPTER 3: CONDITION BASED MAINTENANCE FOR ENGINEERINGSYSTEMS 41

3.1 Introduction 41

3.2 Fault and Failure Definitions 44

3.3 Diagnostics and Prognostic Definitions 46

3.4 Review of Condition Based Maintenance System Requirements 49

3.5 Systems Based Strategy for Condition Based Maintenance 51

3.6 Open Systems Architecture for Condition Based Maintenance 54

3.7 Sensor Systems 563.7.1 Technology Aspects of Sensors 573.7.2 Wireless and Smart Sensors 583.7.3 Multiple Sensor Networks 59

3.8 Data Fusion Overview 603.8.1 Fusion Processes 623.8.2 Data Fusion Models 633.8.3 Architecture Selection 65



iii

3.8.4 The JDL Data Fusion Process Model 65

3.9 Health Monitoring Techniques 673.9.1 Vibration Monitoring 683.9.2 Lubricant Wear Debris 693.9.3 Motor Current Signature Monitoring 703.9.4 Thermal Monitoring 713.9.5 Acoustic Emission Monitoring 713.9.6 Performance Monitoring 723.9.7 Corrosion Monitoring 72

3.10 Critical Review of Monitoring Methods 72

3.11 Expert Systems 753.11.1 Model-Based Expert Systems 753.11.2 Knowledge Based Rule Systems 773.11.3 Neural Networks 793.11.4 Fuzzy Systems 793.11.5 Uncertainty in Expert Systems 80

3.12 Critical Review of Health Monitoring Strategies 80

3.13 Motor-Driven Actuator Health Monitoring Review 833.13.1 Overview 833.13.2 Aerospace 843.13.3 Automotive 853.13.4 Rail 863.13.5 Power Industry 87

3.14 Conclusion 88

3.15 References 89

CHAPTER 4: UNDERSTANDING THE COMMERCIAL BENEFITS OFAEROSPACE HEALTH MONITORING 101

4.1 Introduction 101

4.2 Current Aerospace Maintenance Practice 101

4.3 Changing Maintenance Practice 106

4.4 Predictive Maintenance 108

4.5 Value potential of Predictive Maintenance 110

4.6 Developing, Integrating and Pricing the Technology 1124.6.1 Technical Challenges to Integrating Health Monitoring 1124.6.2 Commercial Integration Challenges 1144.6.3 Pricing Deployment Strategies 116

4.7 SWOT Analysis: Actuator Health Monitoring Technology 117



iv

4.7.1 Strengths 1184.7.2 Weaknesses 1194.7.3 Opportunities 1194.7.4 Threats 120

4.8 Conclusion 121

4.9 References 122

CHAPTER 5: HEALTH MONITORING SYSTEMS METHODOLOGY ANDFRAMEWORK 124


5.2 Framework Objectives 124

5.3 Overview of the Health Monitoring Data Fusion Framework 125

5.4 Inputs to the Fusion Centre 1285.4.1 Objectives 1285.4.2 Part Trees 1305.4.3 Fault Trees 1305.4.4 Observables 132

5.5 The Fusion Process 1345.5.1 Alignment 1345.5.2 Association 1345.5.3 Hypothesis Generation 1365.5.4 Hypothesis Evaluation 1375.5.5 Hypothesis Selection 138

5.6 Estimation 138

5.7 Proposed Decision Support Outputs 139

5.8 Considerations for Practical Implementation 1395.8.1 Service bay implemented 1405.8.2 Embedded deployment 141

5.9 Health Monitoring Acceptance Criteria and Metrics 1425.9.1 Validation Procedure 1425.9.2 Fault Diagnostic Performance Metrics 1435.9.3 Technical Value 146

5.10 Conclusions 147

5.11 References 148

CHAPTER6: APPLICATION OF FUZZY LOGIC AND PRINCIPALCOMPONENT ANALYSIS FOR DETERMINING PROCESS QUALITY 150



v


6.2 Estimation of Actuator Process Quality 1516.2.1 Data Redundancy 1516.2.2 Principal Component Analysis 1526.2.3 Choosing the Principle Components 1546.2.4 Generating Performance Statistics 1556.2.5 Fuzzy Logic Classification 1566.2.6 Fuzzy Rule Base 1586.2.7 Obtaining a Quantitative Quality Index 160

6.3 Experimental Demonstration 1616.3.1 Experimental Objectives 1616.3.2 Experimental Setup, DataAcquisition and Post-Processing 1626.3.3 Implementing a Lubrication Fault 1646.3.4 Actuator Responses 1656.3.5 Estimating the Nominal PCA Model 1676.3.6 Fuzzy Inference 1696.3.7 Estimation of the Actuator Quality Index 170

6.4 Experimental Testing Limitations 171

6.5 Practical Considerations 173

6.6 Conclusions 174

6.7 References 175

CHAPTER 7: FORMULISATION OF A PROPOSED ACTUATOR HEALTHMONITORING ALGORITHM 177


7.2 Fault Diagnostics 1777.2.1 Model-Based Fault Monitoring 1777.2.2 Formulating Parity Relations 1787.2.3 Defining Residual Thresholds 1807.2.4 Traditional Threshold Evaluation 1827.2.5 Evidential Reasoning 1857.2.6 Introducing Residual Uncertainty 1877.2.7 Combining Multiple Evidential Intervals 1887.2.8 Combining Rules for Comprehensive Diagnostics 189

7.3 Advantages of the Proposed Methodology 190

7.4 Conclusions 191

7.5 References 192

CHAPTER 8: SYSTEM MODELLING, SIMULATION AND DIAGNOSTICSDEMONSTRATION 194



vi


8.2 Modelling the landing Gear Actuator 195

8.3 Simulation 1978.3.1 Overview of the Actuator SIMULINK Model 1978.3.2 Actuator Loading 1998.3.3 Actuator Performance Simulation 200

8.4 Fault Cases 202

8.5 Actuator Performance Assessment 204

8.6 Parity residuals 2058.6.1 Nominal Test 2068.6.2 Residual Fault Sensitivity 2078.6.3 Combining Residual BPA's 208

8.7 Diagnostics Algorithm Demonstration 2088.7.1 Overview of the Simulation Process 2088.7.2 Simulation Results 210

8.8 Conclusions 213

8.9 References 215

CHAPTER 9: CONCLUSIONS 216

9.1 Summary 216

9.2 Research Conclusions 2179.2.1 Objective 1 2179.2.2 Objective 2 2189.2.3 Objective 3 2209.2.4 Objective 4 222

9.3 Contribution to Knowledge 223

9.4 Further Work 2239.4.1 Experimental Landing Gear Test bed 2249.4.2 Uncertainty and Performance Metrics 2249.4.3 New Sensor Technology and Systems Integration 2259.4.4 Cost Modelling 2259.4.5 Remaining Life Models 226

9.5 Published Research Papers 2269.5.1 Journal Papers 2269.5.2 Peer Reviewed Conference Contributions 227



vii

LIST OF NOTATION

AE. Acoustic Emissions

ANN. Artificial Neural Network

BIT. Built In Tests

BPM. Basic Probability Mass

CBM. Condition Based Maintenance

CEng. Chartered Engineer

CPV. Cumulative Percentage Variance

ECS. Environmental Control System

ELGEAR. Electric Landing Gear Extension and Retraction

EMA. Electro-Mechanical Actuator

ERP. Enterprise Resource Planning

ETA. Event Tree Analysis

FMEA. Failure Mode and Event Analysis

FMECA. Failure Mode and Event Critical Analysis

FTA. Fault Tree Analysis

IET. Institute of Engineering and Technology

IMechE. Institute of Mechanical Engineering

IVHM. Integrated Vehicle Health Monitoring

MCSA. Motor Current Signature Analysis

MEA. More Electric Aircraft

MRO. Maintenance Repair and Overhaul



viii

MRP. Materials Resource Planning

O&S. Operations and Support

OEM. Original Equipment Manufacturer

PCA. Principle Component Analysis

POA. Power Optimised Aircraft

POD. Probability of Detection

POFA. Probability of False Alarm

REACTS. Reliable Electrical Actuation Systems

RTF. Run to Failure

RUL. Remaining Useful Life

SPE. Squared Prediction Error

SVD. Singular Value Decomposition

SWOT. Strengths, Weaknesses, Opportunities and Threats

TBPM. Time Based Preventative Maintenance



ix

ABSTRACT

There are numerous benefits associated with replacing hydraulic actuators with

electrical counterparts as part of an all electric landing gear including reduced

consumption of non-propulsive engine power, reduced weight, reduced cost and the

elimination of hydraulic systems. The development of health monitoring systems to

support the introduction of electrical actuation systems into landing gears will aid in

guaranteeing reliability and to optimise landing gear maintenance activities.

One of the difficulties with designing health monitoring for industrial integration

involves the large number of subject areas involved, ranging from architectural

design, software and signal processing design, hardware selection and business

modelling. The reason that many health monitoring systems never reach full

development maturity is that there is a failure in realising a holistic design process.

The purpose of this thesis and the overall contribution which has been made is to

bring together a combined understanding of landing gear design, health monitoring

and the business environment for aircraft maintenance in order for a holistic design

process for landing gear health monitoring to be realised.



x

DECLERATIONS

No portion of the work referred to in this thesis has been submitted in support of an

application for another degree or qualification of this or any other university or

institute of learning.



xi

COPYRIGHT

i. The author of this thesis (including any appendices and /or schedules to this

thesis owns certain copyright or related rights in it (the “Copyright”) and he

has given The University of Manchester certain rights to use such Copyright,

including for administrative purposes.

ii. Copies of this thesis, either in full or in extracts and whether in hard or

electronic copy, may be made only in accordance with the Copyright, Designs

and Patents Act 1988 (as amended) and regulations issued under it or, where

appropriate, in accordance with licensing agreements which the University has

from time to time. This page must form part of such copies made.

iii. The ownership of certain Copyright, patents, designs, trademarks and other

intellectual property (the “Intellectual property”) and any reproductions of

copyright works in the thesis, for example graphs and tables

(“Reproductions”), which may be described in this thesis, may not be owned

by the author and may be owned by third parties. Such Intellectual Property

and Reproductions cannot and must not be made available for use without the

prior written permission of the owner(s) of the relevant Intellectual Property

and/or Reproductions

iv. Further information on the conditions under which disclosure, publication and

commercialisation of this thesis, the Copyright and any Intellectual property

and/or Reproductions described in it may take place is available in the

University IP Policy (see

http://documents.manchester.ac.uk/DocuInfo.aspx?docID=487) , in any

relevant Thesis restriction declarations deposited in the University Library,

The University library’s regulations

(see http://www.manchester.ac.uk/library/aboutus/regulations), and in The

University’s policy on presentation of Theses.



xii

ACKNOWLEDGMENTS

The author acknowledges that this thesis would not have been possible without the

help and support from fellow researchers, industrial experts and academic peers. The

following individuals have provided invaluable support.

Firstly, I would like to express my gratitude to the project supervisors Dr Dominic

Diston at the University of Manchester, Julia Payne and Satish Pandya at Messier

Dowty Ltd. In addition I would like to thank Professor Andrew Starr from Cranfield

University for his sharing of expertise, providing commentary on my research reports

and his contributions to conference and journal publications.

I also gratefully acknowledge the assistance and support of the Manchester

Engineering Doctorate Centre in particular the advice, guidance and support offered

by Dr David Stanley and Janet Wade.

Finally I would like to thank my family for their continued support and

encouragement.



xiii

THE AUTHOR

After obtaining his Bachelor’s degree in Physics and Astrophysics from the University

of Keele in 2004 Paul worked in a variety of technical manufacturing roles before

obtaining his postgraduate Masters degree in Control Systems Engineering from the

University of Sheffield in 2006. Immediately after this he took up an appointment as

an Engineering Doctorate (EngD) research engineer at the University of Manchester.

During his doctorate Paul was elected as a full member of the Institute of Engineering

and Technology (MIET) in 2007 and obtained his postgraduate diploma in

Management Sciences from Manchester Business School in 2008.

In 2008 and 2009 Paul worked closely with Spirit Aerosystems Inc. (KS, USA),

Wichita State University (KS, USA) and the Northwest Composite Centre (UK)

investigating novel manufacturing techniques and technologies for composite fuselage

and wing structures. In addition to this during his EngD time Paul was engaged in

several high profile projects including the DTI funded ELGEAR project along with

the EU Framework 6 projects TATEM and DYNAMITE. Post EngD Paul worked for

12 months as a Research Associate in the Rail Technology Unit (RTU) in the School

of Engineering at Manchester Metropolitan University (MMU). His research there

included enhancing the use of simulation in the homologation process for rail vehicles

across Europe, modelling future high-speed freight vehicles, and investigations into

the business case for sustainable freight transportation as part of the EU FP7 projects

DYNOTRAIN, SUSTRAIL and SPECTRUM.

He is currently a Research Project Manager with Cranfield Defence and Security, as

part of a new £10.5m EPSRC Centre for Innovative Manufacturing in Through-Life

Engineering Services hosted by Cranfield and Durham Universities. His main focus

is in the management of research projects supported by the MOD, civilian aircraft

operators and the rail industry developing technology and processes to mitigate the

No-Faults Found (NFF) problem for new and ageing complex engineered products.

C h a p t e r 1 : I n t r o d u c t i o n

This chapter aims to introduce the engineering doctorate programme, the sponsoring

organisation, the project background, objectives and management. Finally a

description of the thesis layout, in terms of individual chapters is given.

1.1 The Degree of Doctor of Engineering

The Engineering Doctorate (EngD) is a flagship programme which provides the

opportunity for outstanding engineers to work within industry whilst obtaining a

doctoral level qualification. The student termed a research engineer is jointly funded

by the Engineering and Physical Science Research Council (EPSRC) and a

collaborating organisation. The aim of the EngD is to provide the research engineers

with an intensive and broadly based research training experience. This ensures that

the EngD research engineers who often aspire to senior management roles in industry

are able to gain practical experience of working within industry whilst expanding their

knowledge through further academic study.

As well as the research the EngD has a number of other aspects which must be

completed satisfactory. These include supporting technical courses, a diploma in

management science and various monitored professional development elements. The

University of Manchester’s EngD professional development scheme is unique within

the UK with it being accredited by the Institute of Mechanical Engineers (IMechE)

and the Institution of Engineering and Technology (IET). A professional development

mentor and advisor are appointed for each research engineer, usually within another

industrial organisation, to offer advice and guidance on professional development.

The aim of the EngD professional development program is for the research engineer

to reach professional chartered status (CEng) upon completion of the EngD. The

nature and objectives of the EngD programme mean that for the success of the project,

research activities must be aligned to the sponsor’s objectives and the commercial

implications of the research project must also be considered.

.



2

1.2 The Sponsoring Company

The sponsoring company, Messier – Dowty are suppliers of aircraft landing gear

systems to aircraft constructors worldwide. They are involved in all aspects of

landing gear, starting from the design, right through to the development and

manufacture of fully integrated systems for all types of aircraft. Messier-Dowty are a

SAFRAN Group company; an international high technology group involved in

aerospace propulsion, aircraft equipment, defence security and communications. They

are the world leaders in the design, development, manufacture and support of landing

gear systems. These gear systems are in service on more than 19,500 aircraft making

over 35,000 landings every day. The company supplies 33 airframe manufacturers and

supports 2,000 operators of large commercial aircraft, regional and business aircraft,

military aircraft and helicopters. Approximately 4500 personnel are employed across

13 sites in Europe, North America and Asia

Messier Dowty takes a holistic view of their product life cycle in order to meet the

challenges of today’s dynamic aerospace environment. Their focus is on providing

landing gear systems which are not only reliable and robust, but increasingly weight

efficient and environmentally responsible, thus providing overall value across the

full life of an aircraft program. As part of the SAFRAN Group’s landing gear

systems integration capability that covers the full ATA Chapter 32 of commercial

landing gear systems. Messier Dowty’s capability offers air framers a single source

for their needs. Saving considerable time and cost in terms of design, technical

interface and supplier management. They lead systems activity on a number of

development programs, coordinating the integration of sub-systems provided by

specialist partners, allowing customers to reduce management responsibilities, lead

times and acquisition costs.

In the commercial sector Messier–Dowty supply landing gear for the entire Airbus

range of aircraft and the nose and main landing gear for the Boeing 787 Dreamliner.

In addition to this they also supply gears for one-third of the worlds regional

business jet programmes. In the military sector contributions are made to the world’s



3

most advanced military programs, including, Boeing’s F/A-18E/F, the Eurofighter,

the Airbus A400M military transport aircraft and Dassault’s Mirage and Rafale. This

product range also extends to helicopters and tiltrotors, where Messier-Dowty

supports both the BA 609 and the Eurocopter Tiger programs.

1.3 Division History

The EngD project is run in conjunction with Messier-Dowty’s UK facility in

Gloucester. The Gloucester facility has been at the forefront of landing gear

technology for over 70 years, dating from innovative landing solutions in the 1930’s

with the development of the first internally sprung wheel, to the advanced landing

gears on most of today’s aircraft.

Today the 44,000m² Gloucester facilities employ approximately 1000 people.

Activities include a total capability from concept to in-service support including

design, development, test, production, processing (heat treatment, surface finishing,

etc), assembly and product support. Core competencies at Gloucester include a strong

engineering function involved in design, research, development & test and systems

integration together with comprehensive state of the art production capability.

The Gloucester test facility includes extensive capability for strength, fatigue, drop,

endurance, environmental testing and systems integration. Production activities focus

on critical components including complex major structures such as large main fittings

and bogies/truck beams for large commercial aircraft, plus main fittings and the larger

components for military and commuter aircraft. Programs supported at Gloucester

include: - Airbus A330/340 family of main gears; A318/319/320/321 family of main

gears; A310 bogies; Airbus Military A400M nose gear; Boeing 787 truck beams;

Eurofighter/Typhoon landing gear system; Boeing T-45 Goshawk main gear and BAE

Nimrod MRA4 nose & main gears. Landing gear spares support includes: BAE

Harrier & Boeing AV-8B nose & main gears; Panavia Tornado nose & main gears;

Sepecat Jaguar nose gear; Avro RJ (146) main & nose gears; Fokker 50 nose and main

gears; Fokker 70 & 100 nose gears and BAE ATP nose and main gears.



4

1.4 Research and Technology

Messier-Dowty is actively pursuing new opportunities to optimise landing gear

technology throughout every stage of the product life cycle. Two of the research

projects in which Messier Dowty have been recently engaged in and which the EngD

project has helped to support are ELGEAR (Electric Landing Gear Extension and

Retraction) and TATEM (Technologies and Techniques for New Maintenance

concepts).

TATEM (Techniques and Technologies for New Maintenance Concepts) was a 4-

year EU framework 6 research project which began in March 2004 and finished in

2008, costing €40 million. The project brought together a consortium of 57

contractors from 12 countries across Europe, Israel and Australia. The purpose of the

project was to investigate methods for reducing the cost of maintenance on both fixed

wing and rotary wing aircraft. The objectives aimed at ensuring that the European

aerospace industry remains competitive in the design and support of current and next

generation aircraft. The research included new maintenance philosophies,

technologies and techniques, to develop new approaches for maintaining aircraft

structure, avionics, utilities, landing gear and engines. The project demonstrated the

means to achieve a 20% reduction in airline maintenance related costs within 10 years

and a 50% reduction over 20 years. The technical focus of the TATEM project

assessed the following maintenance philosophies, technologies and techniques:

Maintenance-free avionics that require no scheduled maintenance work.

Signal processing techniques which can be used to convert data into

information about the health of the systems.

Novel onboard sensor technology to gather data from the aircraft (avionics,

utilities, actuation, engines and structures), to feed prognostic or diagnostic

systems.



5

Diagnostic methods to identify and locate failures and malfunctions and so

reduce the incidence of no fault found alarms.

Prognostic methods to provide support for preventative maintenance actions.

Decision support techniques to generate process-oriented information and

guidance (instructions) for the maintenance engineer.

Human interface technologies to provide the ground crew with information,

data and advice at the point of work.

ELGEAR (Electric Landing Gear Extension and Retraction) was a £11 million UK

part DTI (Department of Trade & Industry) funded programme that commenced in

February 2006 until late 2009. The project involved four major manufacturers –

Airbus, Goodrich, General Electric and Messier-Dowty. The programme aimed to

design and model a complex electrical system for the control and actuation of an all

electric landing gear. This would be followed by the manufacture of the electric

actuators for the landing gears. Goodrich, GE and Messier-Dowty are responsible for

the design and manufacture of the electric actuation systems for one landing gear

each. The requirements/constraints for the landing gear/electric system designs are

provided by Airbus along with the vehicle testing rig for validating the all electric

landing gear. The test vehicle would be an Airbus A320, however the all electric

landing gear is intended to be used in the next generation single aisle aircraft.

1.5 Problem Statement and Research Motivation

Electric motor driven actuation is now very widespread. In automotive products, for

example, electric windows, locks, aerials, seat/lamp/mirror adjustment are common.

Drive-by-wire introduces motor-actuated steering and the starting circuit is a heavy

motor-driven actuation system. Similar situations are encountered in railway point

mechanisms, heavy electrical switch gear, and valve actuation. Many similar



6

applications are proposed for the “more electric aircraft” for example in future civil

aircraft landing gears.

There are numerous benefits associated with an all electric landing gear

including reduced consumption of non-propulsive engine power, reduced weight,

reduced cost and the elimination of hydraulic systems. Elimination of the hydraulic

systems of the landing gears and brakes leads to reduced aircraft turn-around

times and the toxic, hydraulic fluids which require significant maintenance effort to

contain, are no longer needed. Challenges associated with the actuator design

include the space/weight constraints (it will be difficult to fit electric motors

into the available space) and being able to anticipate and provide solutions for all

of the possible failure modes associated with the completely new all-electric design.

The fact that hydraulic actuation has been used in aerospace successfully for many

decades, proving to be reliable and hence gaining the confidence of aircraft operators

means that any replacement drive will need to provide insurances that they are of

equal robustness and reliability to the preceding system. An all new electric landing

gear actuation system will therefore require the use of health monitoring to help

guarantee reliability and ensure customer confidence in the replacement actuator.

The development of a health monitoring system for landing gear actuators would aim

at providing a diagnostic/prognostic health monitoring capability, which will enable

decisions to be taken regarding aircraft flight worthiness. However the development

of a health monitoring system poses several significant challenges in the choice of

monitoring approach. These actuators can be considered to have a variety of ‘normal’

operating modes and experience varying loads, speed, friction and operating

environment. These factors which lead to widely varying data from typical

measurements can mask the actuator faults until the severity has increased to the point

of failure.



7

Further issues relating to the design of the health monitoring system are related to its

implementation into the landing gears. The drive for reductions in landing gear mass

and volume combined with issues relating to the cost and complexity of the health

monitoring system, with all data requiring on-board processing means that

performance data will be limited placing restrictions on the use of additional sensor

equipment on the landing gears.

The overall motivation for developing a health monitoring system for

electromechanical actuators is to provide diagnostic and prognostic information

regarding the health state of the actuator for the purpose of:

1. Avoiding faults that may lead to in-flight actuator failures impacting upon the

landing gear reliability

2. Improving the availability of the aircraft and to reduce maintenance support

costs by investigating the use of prognostics for the actuator system.

Actuator health monitoring would ultimately result in reductions in scheduled

maintenance where serviceable items would remain on the landing gears for longer

periods. Maintenance operations would be able to be optimised, reducing the cost of

replacement parts (legislation dictates that certain components are replaced at regular

intervals regardless of condition) and also increasing aircraft availability. There are

both these commercial benefits and also disadvantages as health monitoring would

result in reduced sales revenues for the landing gear manufacturer from serviceable

parts.

Designing a health monitoring system requires the use of a variety of multidisciplinary

approaches, and requires the use of a systems based methodology. Key top level

design requirements include:

1. The use of existing data will be used wherever possible



8

2. The impact of data acquisition on cost, weight, size, reliability, power

consumption and the operation of the landing gear system shall be minimal.

3. Any additional transducers will be capable of effectively operating in the

landing gear environment.

4. Monitored data and decisions can be provided with a measure of data quality

and performance assessment.

1.6 Research Objectives

The following of research objectives acted as a framework in which the research has

been conducted

Objective 1: Assess the current state of the art health monitoring techniques and

show that established techniques exist which are viable for a landing gear actuator

application without the need for additional sensor equipment, placed upon the

actuators.

Objective 2: Define a systems architectural framework for EMA diagnostics and

prognostics, with identification of key nodes which will:

Identify abnormal behaviour

Incorporate performance metrics

Allow analytical and heuristic symptoms to be used effectively alongside

process history, costs and risks.

Be accessible for additional sensor/heuristic data, for health monitoring

purposes, to be incorporated without architectural alterations.

Objective 3: Define and demonstrate a health monitoring algorithm for component

level actuator fault detection and diagnostics.



9

Objective 5: Demonstrate and assess the commercial benefits of incorporating health

monitoring systems into aircraft landing gears.

1.7 Project Management

As part of the EngD training objectives it is essential for the research engineer to

obtain the necessary skills to efficiently plan and manage the doctorate research in a

manner which is expected for an industrial focused project. The project management

and organisation of the budget for this thesis work has been the sole responsibility of

the author. A detailed project plan was created to translate the industrial sponsors top

level requirements into identifying feasible research objectives and project scope

incorporated into several discrete work packages, with associated milestones and

deliverables. The project plan was formulated with close considerations of the

industrial sponsors’ outcome requirements and planning standards. As with any

industrial project, it is often the case that things do not always progress as planned.

This was the case with certain elements of this research project, but through the use of

initial project risk assessments with identifiable mitigation processes and the use of a

dynamic project plan all major research deliverables were met on time.

1.8 Thesis Layout

Chapter one has introduced the reader to the degree of Doctor of Engineering and

highlighted how it is different from the conventional PhD system. Information

regarding the sponsoring company has been presented, including information specific

to the UK facility in which this thesis is attributed. The motivation and background

for the research is emphasised and a framework based upon a set of research

objectives in which the research has been carried out has been provided, along with

notes concerning the project management and planning.

Chapter two presents a review relating to the technical aspects of the projects

application. It demonstrates a high level understanding of the more electric aircraft,

landing gears and actuation devices prior to subsequent health monitoring system

design. A brief review is given on related industrial projects connected with the more



10

electric aircraft concept. A comparative review on varying types of electromechanical

actuation provides the justification and insight into the actuator design being

developed by Messier-Dowty. Chapter 2 also highlights the key failure modes as

identified from the development of Failure Mode and Event Analysis (FMEA) and

Event/Fault Tree Analysis (FTA/ETA).

Chapter three describes, in detail, condition based maintenance and individual

monitoring techniques, including the state of the art sensor selection and signal

processing methods. The use of expert systems lends themselves well to aerospace

health monitoring and different approaches are presented and critically reviewed.

The use of fusion architectures combined with the health monitoring standard ISO-

13374 is considered in detail. Finally an industry wide review actuator health

monitoring is provided along.

Chapter four covers the commercial implications of actuator health monitoring. The

chapter starts by looking at the global landing gear market from the sponsoring

companies’ viewpoint. Current maintenance practices are used along with potential

changes to these operations which health monitoring would create is used to determine

how aerospace actuator health monitoring should be packaged. This is important to

ensure that all key players involved in aircraft maintenance and repair can obtain

maximum benefit from the technology. Integration of health monitoring technology is

discussed from both a technological and business model perspective and a variety of

pricing models are proposed. The chapter concludes by presenting a detailed review

of strengths, weaknesses, opportunities and threats associated with actuator health

monitoring.

Chapter five presents a generic framework and methodology for the development of a

monitoring system. The framework provides concise descriptions of individual

system modules, and how they will operate in a health monitoring context. Different

deployment strategies are also reviewed highlighting their advantages and

disadvantages



11

Chapter six introduces the first part of a proposed health monitoring algorithm based

within the concept of fault detection. The approach is aimed at providing an overall

assessment of actuator operating quality, utilising only the available control signals

with no requirement for extensive efficiency modelling. An experimental

demonstration of the fault detection, using a bench top actuator with varying levels of

degrading lubrication conditions.

Chapter seven builds upon the fault detection presented in chapter seven into a

diagnostics algorithm. This component level diagnostics approach is based upon the

use model-based parity equations, and coupled with, an evidential reasoning inference

process. The chapter outlines the diagnostic process and mathematical formulisation

from symptom generation through to fault identification.

Chapter eight verifies the diagnostic algorithms potential through the development of

an actuator model and simulation certified by Messier Dowty Ltd as representative of

the predicted performance of the ELGEAR landing gear retraction system. Utilising

published information on parametric symptom to fault relationships a variety of faulty

residuals are used to obtain evidence relating to a selection of different faults. The

simulation demonstrates the effectiveness of combining evidence sets in reducing

diagnostic uncertainty and providing strong diagnostic results. The chapter concludes

with comments on the practical application of the proposed approach.

Finally, chapter nine presents the summary and conclusions of the thesis. The novelty

and contribution to knowledge is highlighted here. The main focus for this chapter is

to reflect on the original objectives by providing a review of each and how the

research thesis has achieved these objectives. The chapter concludes by proposing a

number of future research opportunities.



12

C h a p t e r 2 : L a n d i n g G e a r , E l e c t r o m e c h a n i c a lA c t u a t o r a n d S y s t e m s B a c k g r o u n d

2.1 Introduction

The aim of the chapter is to provide the necessary background information for the

novel landing gear actuators. Understanding of the application area is essential for the

development of any health monitoring system. The chapter begins by setting the

research context which is aligned to the ‘More Electric Aircraft’ concept. This

involves a description of the key benefits of electrical powered systems and a brief

review of relevant industrial research projects. An overview of landing gear systems

with emphasis placed on landing gear retraction methods is given. This is intended

not too provide the reader with an in depth study of all the main systems involved with

the gears operation. Rather it is aimed at highlighting how the landing gears are an

essential safety and mission critical part of any aircraft that must be guaranteed to

operate.

Leading on from the discussion on landing gears a comparative review of different

actuator types is given, focusing on electromechanical actuator assemblies such as

lead, roller and ball screw in the context of their structure, performance, industry

application and failure modes. This provides an insight into the reasoning behind the

choice of actuator design for the main landing gear retraction actuator. An overview

of this system is also discussed, focusing on air worthiness directive design

requirements, control, reliability and safety issues, critical failure modes and

performance requirements.

2.2 The “More Electric Aircraft”

Today civil and military aircraft secondary power which is used for on board systems

falls into the following categories:



13

Hydraulic power is used in most aircraft actuation systems. Including engine

actuation; primary and secondary flight control; and landing gear deployment,

retraction and braking.

Pneumatic power is used for Environmental Control Systems (ECS) and ice

protection

Electrical Power is used to power the avionics and most of the aircrafts utility

functions.

Figure (2.1) illustrates the range of aircraft systems which are currently powered

through a mix of pneumatic, hydraulic and electrical systems.

Figure 2.1: Schematic of Aircraft Power Systems(SBAC 2007).

In aerospace there is a move towards developing a More Electric Aircraft (MEA)

where a large part of this secondary power is electrical in nature. There are the

following reasons which can be used to support the introduction of electrical systems:



14

1. Reduced weight- hydraulic and bleed air systems generally contain bulky,

heavy hardware which contribute significantly to the weight of the aircraft.

2. Reduced fuel consumption - bleed air systems rely on compressed air

from the engine which is produced at the expense of fuel consumption (i.e.

fuel is consumed to produce the bleed air which then does not contribute to

engine thrust).

3. Increased efficiency - losses in hydraulic or pneumatic piping are higher

than in electrical cabling, plus these systems do not have the ability to

provide power on demand (they cannot simply be switched on and off as

required like electric systems) hence resulting in higher quiescent losses.

This means that the peak power consumption of hydraulic and bleed air

systems is higher than necessary.

4. Compromise of optimal component design - as the equipment using the bleed

air systems requires certain pressures to operate the optimal design of an

engine component may be compromised in order to provide the required air

pressure for the bleed air system (Faleiro 2005). This can lead to non-

optimal performance characteristics including fuel burn.

5. The power off-takes at the engine from all of the aircraft systems are typically

responsible for 3-5% of the total power produced by the engine (Faleiro 2005).

By developing a more electric aircraft, this power requirement can be

significantly reduced, enabling lower fuel burn and emissions.

2.3 More Electric Aircraft Research History

The concept of a complete electrically powered aircraft is not a new concept and the

considerations for military aircraft to be more electrically powered aircraft have

existed since World War 2 when debates had been begun on the best way to distribute



15

power around an aircraft. Hydraulic power systems began to rapidly develop in the

1970s and the use of electrical power onboard aircraft began to be restricted to

electronic systems only. The idea of using electric power as the only secondary power

began to gain ground with early research performed jointly by NASA and The Lewis

Research Centre (Spitzer 1984). There are two steps which are being taken to further

the MEA. The first is the removing of current air and hydraulic secondary power

supplied by the engines and increasing electrical power generation. The second is the

replacing of hydraulic and pneumatic actuators with electrical counterpart.

Different approaches including the use of electro-hydrostatic, hybrid and electro-

mechanical actuators have been considered for use as alternative actuation system for

the actuation of primary and secondary flight controls; braking; nacelle actuation and

new landing gear extension and retraction. In the past decade the use of Electro-

Hydrostatic Actuator (EHA) technology has rapidly developed and they are now

replacing hydraulic circuits on new aircraft such as the Airbus A380 and Boeing

B787.

The feasibility of using EMA was shown in research by NASA in the early 1980’s in

cooperation with Boeing Commercial Airplane Company (Holmadhl 1983). In this

research an EMA was installed on the inboard flight spoiler of a small research

aircraft. The performance of which was shown to match that of its hydraulic

counterparts. The use of EMA gained more favourable results in a second NASA

project where a large EMA was used to drive the Space Shuttle elevon (Pond and

Wyllie 1983).

The Power-By-Wire (PBW)/Fly-By-Light (FBL) research carried out by the NASA

Lewis Research Centre aimed at looking at the potentials of electric actuators in

aircraft with some specific focus on trends and tradeoffs involved in the selection of a

particular motor drive technology. In particular DC motor drives, switched reluctance

motor drives (Elbuluk and Kankam 1995) and induction drives (Elbuluk and Kankam

1996).



16

Early research into electric actuation for military aircraft included research by

Lockheed-Georgia (Bradbury 1987) into the potentials of using electric actuation for

flight controls onboard the C-130 (Alden 1991). Jensen et al. (2000) describes the test

results of an EMA fitted onto the aileron of an F/A-18. The performance of which

was shown to be virtually identical to that of a standard hydraulic actuator and

therefore validating the potential use of aileron EMAs. The tests did however

highlight problems with the actuators thermal performance.

In the early 1990s research into aircraft power systems had advanced to the stage of

replacing centralised onboard aircraft hydraulics with electrical power. This led to the

United States Air Force More Electric Aircraft program (MEA) (Cloyd 1998) which

aims at increasing a fighter aircrafts electrical capability.

There have been several DTI funded technology programmes under the Civil Aircraft

Research And technology Demonstration (CARAD) over the last decade as part of the

UK “More Electric” initiative. These programmes have helped technological

advances for electric power generation, distribution and flight surface actuation.

Example programmes include Reliable Electric ACTuation Systems (REACTS)

(Dixon et al. 1999) which investigated the use of a smart EMA which would be

suitable for use on civil gas turbine aero-engines. The Distributed Electrical Power

Management Architecture (DEPMA) (Bailey et al. 1999) consortium investigated

alternative electrical power distribution architectures on both civil and military

aircraft. In the Electric Actuated Braking SYSTem (EABSYS) (Collins 1999)

programme the aim was to design and develop an electrically actuated braking system

to replace conventional hydraulics. The Totally Integrated More Electrical Systems

(TIMES) (Cutts 2002) programme was devoted to using systems which have been

previously developed in electrical aircraft.



17

2.4 Current Relevant EU Electric Aircraft Research projects

2.4.1 Clean Sky Joint Technology Initiative

The CLEAN SKY Joint Technology Initiative is a large scale EU-wide research

programme designed to integrate and further develop the promising results of

many technology programmes currently underway. The €1.6 billion, seven year

project aims to develop the breakthrough technologies and operating practices

required for the industry to achieve the ACARE 2020 targets.

2.4.2 More Open Electrical Technologies

The More Open Electrical Technologies (MOET) project is a consortium of 46

companies and 15 research centres and universities from 14 countries in the EU. The

three year programme co-ordinated by Airbus France with a budget of around €70

million, was launched in July 2006 and aims to establish a new industrial standard

for the design of commercial aircraft electrical systems. This new standard, Power-

by-Wire (PbW) will improve aircraft design and utilisation through power source

rationalisation and electrical power flexibility.

2.4.3 Power Optimised Aircraft

The Power Optimised Aircraft (POA) is an EU programme that began in January

2002 and involved 46 partners. The €99.2 million project was led by the German

company Liebherr-Aerospace and examined novel ways of generating, distributing

and using electric power so that non-propulsive power consumption could be

minimised. Specifically, POA aimed to validate, at aircraft level, the ability

of next generation aircraft equipment systems to:

1. Reduce peak non-propulsive power by 25%

2. Reduce total non-propulsive power

3. Reduce fuel consumption by 5%



18

4. Reduce total equipment weight

2.4.4 Electric Landing Gear Extension and Retraction (ELGEAR)

This DTI funded technology programme began in 2006 with a consortium consisting

of four major partners, Airbus, Smiths Aerospace, Goodrich Actuation Services and

Messier-Dowty. This aims of which are to design and model a complex system for

control and actuation of an all electric landing gear system for the future replacements

to single aisle aircraft such as those in the size range of the Airbus A320 and Boeings

B737.

2.5 Overview of Landing Gears

Landing gears are an essential part of the aircraft even though they remain redundant

for most of the flight. The main task of the landing gear is to absorb the horizontal

and vertical energy of the touchdown as well as ensuring a smooth ride before takeoff.

Jenkins (1985) and Young (1986) have published overviews of landing gear design

with text books such as Conway (1958) and Currey (1988) giving in depth details on

landing gear development.

Most modern transport aircraft are designed with retractable landing gears positioned

in a tricycle configuration with a nose gear and two main gears. Conventional

tricycle configured landing gears have become the best solution to taxiing, taking off

and landing with unconventional designs such as skids or air cushion landing gears

being reserved for special applications. Landing gears must be positioned relative to

the aircrafts centre of gravity to prevent them from being easily overturned or from

tipping back onto its tail under static and dynamic loads. The geometry of the

landing gears must also provide clearance of the aircraft with the ground during all

operational conditions. During flight most modern aircraft have their landing gears

retracted and stowed. A prime task for the design of landing gears is to minimise the

volume of the stowed gears and provide the lowest weight possible. This can pose

restrictions on the positioning of the gears with the volume and installed weight

having adverse effects on the performance of the aircraft.



19

Landing gear occupy significant amounts of volume and accounts for about 3% of the

overall mass on military aircraft and about 4% on civil aircraft. In landing gear

design there is a continued effort to reduce this mass through development of new

materials with advanced design and manufacturing techniques allowing for

optimization of aircraft mass (Jenkins 1989). At the heart of the landing gear unit is

the shock absorber. This is designed to absorb the energy generated by the impact

between the gears and the runway upon landing. There are many different types of

shock absorber construction. The most popular is as an oleo-pneumatic shock strut

which combines a gas spring with oil and additional friction damping

The landing gear brakes are required to bring the aircraft to a halt and to help control

the aircraft as it taxis to the runway. The design of the brakes must be able to support

very high temperatures. The brakes add substantial mass to the landing gears and are

generally fixed to the main gears only. As an example the brakes make up 24% of the

total 3626kg mass of the AIRBUS A300 main landing gears (Kruger et al. 1997). The

temperature of the brakes must be monitored as take off cannot be carried out if the

brakes are hot. The reasons for this are that there is a risk of fire if the high

temperatures on the brakes come into contact with hydraulic fluid in the land gear bay.

The number and size of the aircraft tyres is an important design consideration and is

dictated by the aircrafts weight and the strength of the runway which will vary

depending on the airport. Larger civil aircraft such as the A320 and Boeing 747 reach

loads of 20 tons per tyre on each of the main gears (Krüger 2000).

In order to retract landing gears into the smallest possible space, complex motions are

required which will put the landing gear into the assigned space without colliding with

other structures. The main methods of retracting landing gear include mechanical

drives such as pneumatic, hydraulic and electrical actuation. The landing gear

retraction geometry is effectively dependent upon the position of the lowered wheels

(which is governed by ground stability, load distribution and clearance angles),



20

stowage position in the airframe and the attachment points to the airframe. An

example the A310 main landing gear in both retracted and extended positions is

shown in Figure (2.2).

Figure 2.2: A310 Main Gears (Young 1986).

On the landing gear there are several actuating elements which are currently hydraulic.

A typical landing gear may comprise a retraction actuator, a down and up lock, a door

actuator and a door lock. Retraction mechanisms on the majority of the world’s

aircraft however are powered by a hydraulic actuator which acts about the pivot axis

in order to raise the landing gear against weight and aerodynamic loads. The other

two actuators are the lock-stay actuator, which locks the landing gear in place once

extended and the door actuator that ensures that the bay doors are successfully opened

and closed for landing gear deployment. Figure (2.3) shows a typical arrangement of

the down lock and main retraction actuator positions.



21

Figure 2.3 Airbus A320 Main Gears

General future requirements for aircraft landing gear include (i) longer life, (ii) lower

mass/volume and (iii) lower support costs. Longer life and lower support costs can be

met through the use of advanced maintenance such as a future prognostic/diagnostic

health monitoring system.

2.6 Actuator Types

Historically there have been numerous difficulties associated with using electrically

powered actuators in aircraft (Wijekoon 2009). These have attributed directly to the

many reasons why hydraulics dominate actuation in aircraft and have done for a

number of decades. The extensive use of hydraulics by aircraft constructors has led to

wide experience with hydraulics and familiarity with their pros and cons. The wide

use of hydraulic systems immediately indicates a reliable and safe system, creating

confidence in the systems. The most serious criticisms of hydraulics are the potential

fire risk from inflammable hydraulic fluids and their general messiness. The

worldwide use of hydraulic equipment has created an extensive and specialised

industry. Through competitive effort a high level of technical design and production



22

and supply chains have been achieved. Since hydraulics has set the pace competitive

systems lag behind in the solution to certain actuation problems.

Hydraulic equipment is naturally adequately lubricated by the operating fluid this is

not the case with a pneumatic or electric systems. Here the piston seals are lubricated

with grease which can become solid at low temperatures meaning that maintaining an

adequate film of grease at all times is difficult. However the need to reduce aircraft

frame noise and aircraft weight has led to alternative electrical actuation being

considered as permanent replacements to hydraulics. Electrical actuators operate

much quieter than hydraulic cylinders and also have the potential to reduce mass and

overall volume of the system. Electrical actuation units are however often heavier

than hydraulic cylinders but hydraulic system consists of a number of individual

additional components connected by piping which can require a large space whilst

electric systems use much smaller wires. It is therefore predicted that the removal of

the overall hydraulic systems will result in a beneficial weight reduction. This

reduction in landing gear weight will help reduce the consumption of fuel and hence a

reduction in polluting emissions.

The use of electrical actuators also offers potential maintenance savings and

production costs. They run more efficiently than hydraulic cylinders at low ambient

temperatures, low temperatures tend to cause hydraulic fluid to become more viscous

making the operation of the cylinders sluggish. The most important reason however

for investigating replacement of hydraulic actuation is to optimise engine power.

Actuation is regarded as secondary power systems, the power for which is generated

within the engines. Increases in airline fuel taxes make it desirable to optimise engine

power usage. Electrical actuation will aid in reducing this dependency on power

generated from the engines reducing fuel costs. There is also a distinct and real

possibility that future aircraft engines will not produce hydraulic power.

When deciding whether to use electromechanical actuators in a particular application

the sole deciding factors are: which actuator type best meets the technical and



23

economic demands of the application. There are three main types of linear actuation,

which operate either by a lead, roller or ball screw.

2.6.1 Lead Screw

The basic leadscrew illustrated in Figure (2.4) assembly is a simple nut and screw

mating with rubbing surfaces. Consequently they have a relatively high friction and

stiction compared to mechanical parts which mate with rolling surfaces and bearings

(i.e. roller and ball screws).

Figure 2.4: Example of a Lead Screw Assembly*

2.6.2 Ball screw

In a ball screw assembly illustrated in Figure (2.5) a threaded shaft provides a spiral

raceway for ball bearings which act as a precision screw. As well as being able to

apply or withstand high thrust loads they can do so with minimum internal friction.

They are made to close tolerances and are therefore suitable for use in situations in

which high precision is necessary. The ball assembly acts as the nut while the

threaded shaft acts as the screw.

*Illustration obtained from http://www.servo-drive.com (accessed 19/7/2011)



24

Figure 2.5: Example of a Ball Screw Assembly*

2.6.3 Roller screw

The difference in the roller screw design, illustrated in Figure (2.6) from the ball

screw design for transmitting forces is that multiple threaded helical rollers are

assembled in a planetary arrangement around a threaded shaft which converts the

motor's rotary motion into linear movement of the shaft or nut.

Figure 2.6: Example of a Roller Screw Assembly †

Table 2.1 provides a comparison of lead, roller, and ball screw assemblies which are

commonly used for electromechanical actuation. As can be seen each assembly has

its own set of advantages and disadvantages, so the choice of assembly is significantly

application dependant.

* *Illustration obtained from http://www.servo-drive.com (accessed 19/7/2011)



25

Table 2.1: Comparison of Actuator Types

Roller screw Lead screw Ball screw Hydraulic PneumaticLoad ratings Very High High High Very High High

Speed Very high Low Moderate Moderate Very highAcceleration Very high Low Moderate Very high Very high

Lifetime

Very long,many timesgreater thanball screw

Very low, due tohigh friction and

wearModerate

Can be longwith propermaintenance

Can be longwith propermaintenance

ElectronicPositioning

Easy Moderate Easy Difficult Very Difficult

Stiffness Very high Very high Moderate Very high Very lowRelative SpaceRequirements

Minimum Moderate Moderate High High

Friction Low High Low High ModerateShock Loads Very high Very high Moderate Very high High

Efficiency >90% approx 40% >90% <50% <50%Installation Compatible

withstandard

servoelectroniccontrols

User may haveto engineer a

motion/actuatorinterface

Compatiblewith

standardservo

electroniccontrols

Complex,requires

servo-valves,high pressure

plumbing,filtering

pumps, linearpositioningand sensing

Verycomplex,requires

servo-valves,plumbing,filtering,

compressors,linear

positioningand sensing

Maintenance Very lowHigh due to poor

wearcharacteristics

Moderate Very high High

Environmental Minimal Minimal MinimalHydraulic

fluid leaks &disposal

High noiselevels

2.7 Regulation on Landing Gear Retraction Mechanisms

So that any potential unsafe conditions can be identified and addressed, the country of

aircraft registration and the civil aviation authority of the manufacturing country,

generate a set of mandatory guidelines for all aspects of design, manufacture,

operation and aircraft maintenance known as airworthiness directives. These

directives notify the aircraft operators or designers that their aircraft may not conform

to the appropriate standards and if there are any actions (i.e. maintenance) that must be

taken. It is a legal requirement that operators follow the airworthiness directives and

country specific authorities closely regulate them. Such authorities include the Federal



26

Aviation Administration (USA), The Civil Aviation Safety Authority (Australia) and

The Joint Aviation Authorities (Europe).For aircraft with retractable landing gear, the

following common directives apply:

Each landing gear retracting mechanism and its supporting structure must be

designed for maximum flight load factors with the gear retracted. They must

be designed to handle the combination of friction, inertia, brake torque, and air

loads, occurring during retraction at any airspeed up to 1.6 V with flaps

retracted.

The landing gear and retracting mechanism, including the wheel doors, must

withstand flight loads, including loads resulting from all yawing conditions

with the landing gear extended at any speed up to at least 1.6 V with the flaps

retracted.

There must be positive means (other than the use of hydraulic pressure) to

keep the landing gear extended

For a landplane having retractable landing gear that cannot be extended

manually, there must be means to extend the landing gear in the event of

either:

1. Any reasonably probable failure in the normal landing gear

operation system;

2. Any reasonably probable failure in a power source that would

prevent the operation of the normal landing gear operation system.

If a retractable landing gear is used, there must be a landing gear position

indicator (as well as necessary switches to actuate the indicator) or other

means to inform the pilot that each gear is secured in the extended or retracted

positions.



27

2.8 Messier Dowty Actuator Design

The actuator designs for the retraction and lock-stay actuators currently under

development by Messier-Dowty Ltd for use on the aircrafts main gears is based

around that of a roller screw. The other possible consideration was a ball screw

assembly, but with the roller screws ability to handle larger shock loads, reduced

friction, smaller space requirements, easy control and longer lifespan; the roller screw

has been selected as the most appropriate arrangement to meet the specific landing

gear requirement.

One of the primary difficulties in designing electromechanical landing gear retraction

actuators is the legal requirement that there is an emergency means of lowering the

gears in the advent of a full systems failure. In hydraulic systems this is achieved

simply by unlocking the gears and allowing the systems to fall under gravity, the nose

gear is usually retracted to the front so that if they are emergency released the air flow

will help push them into position as shown in the case of an Airbus A300 nose gear in

Figure (2.7).

Figure 2.7: A300 Nose Gear (Krüger 2000).

Front of plane



28

In an electromechanically actuator system, the very nature of the screw/nut assembly

means that a simple release and gravity freefall is not possible. Instead in built

redundancy must be incorporated into the design to ensure a controlled freefall, should

the primary systems fail. Figure (2.8) provides a basic schematic for the main

retraction actuator arrangement. A primary duplex motor is connected to a gearbox

which linearly displaces the nut by rotation of the screw. This in turn moves a lever

arm about a pivot achieving retraction/extension of the landing gears. The actuator

has in built redundancy to ensure the system operates safely. If the primary motor

fails, or the primary roller screw jams, then there is an emergency system consisting of

a backup motor, gear box and rollerscrew that will ensure successful displacement of

the actuator.

Figure 2.8: The EMA Retraction Actuator

Figure (2.9) provides a basic schematic for the lock-stay actuator. The lock-stay

actuator consists of a single rollerscrew, duplex motor and gear box.



29

Figure 2.9: The Unlocking EMA

2.9 Main Retraction Actuator Control and Performance Requirements

The main landing gear actuator control is achieved by local control systems (Active

standby) with internal sensing for snubbing and under/over end of travel position

sensors. The locations of the end of travel hall sensors are identified in Figure (2.10).

Figure 2.10: Main Landing Gear Retraction Cycle*

* Source: Messier Dowty ELGEAR presentation (March 2007)



30

Position control will be achieved through counting of motor commutation hall sensor

states. Hall sensors are available to detect when the actuator has reached the desired

position to stop accelerating to the maximum rate of speed and also when to begin

decelerating to the end of travel position. This is shown in Figure (2.11). The motor

also contains thermocouples to avoid operation at overly high temperatures. External

aircraft proximity sensors are used to indicate up lock and down lock positions.

Figure (2.10) illustrates a retraction cycle of the main gears, demonstrating the

direction of travel for the retraction and locking actuators. The landing gear retraction

mechanisms also contain snubbing devices used to suppress high voltage transients in

the electrical systems. The range of these sensors is also indicated in Figure (2.11).

Figure 2.11: Position Control Strategy*

As described in the previous section, the actuator is designed as to operate reliably in

the advent of the primary actuator failing to extend/retract the gears. If a jam does




31

occur, for example if during landing gear deployment a down and locked signal is not

observed, then the pilot has two options. Firstly the pilot can select to retract the

gears, followed by re-extending them to see if the jam clears. If the jam is bi-

directional and the gears cannot be retracted then the pilot will select the emergency

deployment control safely deploying the gears. This cycle is illustrated in Figure

(2.12).

Figure 2.12: Illustration of the Main landing GearDeployment When Experiencing Actuator Jamming*




32

2.10 Reliability and Safety issues

Landing gear actuators are primarily mechanical/electrical systems with moving parts

and as with all mechanical/electrical systems they are subject to failure. The design of

the actuator therefore must be designed sufficiently robust as to maximise the

probability that it will perform satisfactory for a specific period of time under

specified conditions. The main failure modes which are of primary concerns are as

follows.

1. Failure to retract/extend the landing gear in the systems normal operating

mode.

2. Failure to damp the landing gear at the end of the retraction/extension cycle in

the systems normal operating mode.

3. Inadvertent retraction/extension.

4. Slow or jerky retraction.

5. Failure to permit the full extension of the landing gear in the freefall mode.

6. Failure to damp the landing gear at the end of the extension in the freefall

mode.

If any of these modes occur then it can be deemed that the actuator system is “lost”.

This means that it can no longer perform satisfactory and is unsafe to fly the aircraft

until rectified. In order to avoid these failure modes than the system must be designed

to meet specific reliability criteria. Table 2.2 gives an example of the actuator

reliability prediction as calculated during the early stages of development. This

reliability prediction is estimated based upon standard material components. The use

of aerospace grade materials and components however is likely to significantly

increase the reliability of the actuator system.



33

Table 2.2: Actuator Reliability Prediction*

Assembly Item/Sub-assembly Failure rate per 106

hours

% of total failure

Rate

Extend/RetractActuator

MechanicalComponents

10 2

Duplex Motor 16.66 4Simplex Emergency

Motor8.33 2

Brake (TriplexSolenoid)

10 2

Sensors, Duplex X 4 20 5

Unlock Actuator

MechanicalComponents

5 1

Duplex Motor 16.66 4Sensors, Duplex X 3 15 4

ECU

MCU’s 940A) X 2 102 24MCU’s (3A) X 2 102 24

Abnqalogue & DigitalI/O Interface

100 23

Solid State PowerControllers X 2

1.53 0

DC/DC Power SuppliesX 2

20 5

Total failure per610 hours

427.20

MTBF 2,341

Flights Before Failure 1232

Failure modes in actuators will vary depending upon the type of actuator and the

application for which it is being used. Experience in engineering has demonstrated to

us that actuators do however have common failure modes which are of specific

interest in failure diagnosis. Electrical actuators have failure modes dominated by

mechanical failures such as within the gear and bearings. These are often caused

through inadequate lubrication, overloading, corrosion and poor maintenance. Other

components within the actuator system can also regularly fail; for example electrical

components; sensors and the control system. Table 2.3 shows potential failures in

electrical actuators.

* Source: Messier Dowty ELGEAR presentation given to the author March 2007. Data calculated by Claverham whowere initially responsible for the design and development of the actuator systems.



34

Table 2.3: Potential Electromechanical Actuator

Faults

EMA Component Function of EMA Component Failure Modes

Motor Transforms electrical signalsinto mechanical rotationalmotion

Seized bearings Misaligned Shaft Windings open/shorted

Controller Controls the motor output basedon position feedback

Loss of power Sensor failure Switch/connector

failureGearbox Transforms rotational energy

and helps with speed reduction Fatigue cracking Gear stripping

Acme Configuration Transforms rotational motion(gears) to linear motion (leadscrew)

Screw cracked Nut cracked Nut and screw seized

togetherLead screw/ Ball screw Provides linear displacement to

the actuated system Jammed lead screw Bearing seizure

A full detailed study of actuator failure modes has been carried out in this research

work through the construction of a Failure Mode and Event Analysis (FMEA). Each

component failure mode is given a rating based on occurrence, severity and current

control ability to detect the failure mode. It should be noted that the scores given are a

best guess judgement, designed through published engineering experience and

discussions with human experts and are therefore subjective. They do however give a

strong indication on which components are more likely to occur and which have the

greatest impact on the actuators primary function.

2.11 Actuator Component Failures

From the FMEA studies there are several individual actuator components which are

most likely to lead to loss of the actuator system. The causes of these failure modes

are most likely to manifest as faults within bearings, gears, actuator screw, and

lubrication or within the control sensors. The background to faults in these

components is briefly described below.

2.11.1 Bearing Faults

Bearings are among the most important components in the vast majority of machines

and exacting demands are made upon their carrying capacity and reliability. Even



35

though bearings are precision engineered to operate reliability for a certain calculated

useful life it sometimes happens that they do not attain to this calculated rating and

can fail. There are many reasons for this such as heavier than usual loading,

inadequate/unsuitable lubrication, ineffective sealing or poor installation. Each of

these factors creates its own damage type within the bearing, categorised as the

primary damage including wear, indentations, smearing, surface distress and

corrosion. This can then lead onto secondary damage resulting in metallic flaking and

cracks. Damaged bearings will affect the smooth running of the actuator creating

jerks within the motion and if left unmaintained they will eventually seize leading to

the total loss of the actuator system.

2.11.2 Gear Faults

Metal gears fail for numerous reasons, some, in part, independent from the gears

themselves. Assessing gear damage can be a challenge, especially in industrial

equipment. Unlike lab tests designed to isolate a particular failure mode, field failures

may combine several modes. The more common failures include bending fatigue

failure, which is the result of cyclic bending stress at the tooth root. The damage

process follows three stages: crack nucleation, crack propagation, and final unstable

fracture. Pitting or macropitting is surface damage from cyclic contact stress

transmitted through a lubrication film that is in or near the elastohydrodynamic

regime. Pitting is one of the most common causes of gear failure. Scuffing, also

termed is a severe type of adhesive wear which instantly damages tooth surfaces that

are in relative motion. In fact, a single overload can lead to catastrophic failure. Wear

is a continuous, abrasive process of material removal from mating gear teeth that

happens with or without abrasive particles in the oil. For example, hard asperities on

gear flanks can remove material from mating flanks. Removal of the hardened layer

from surface-hardened gears accelerates wear.

2.11.3 Roller Screw Failure

Roller screws, in contrast to conventional lead screws tend to much bulkier due to the

need to have a mechanism for recirculation of the balls. To maintain the inherent



36

accuracy associated with a roller screw and too ensure minimal risk of failure great

care is needed to avoid contamination with dirt and abrasive particles. For instance, if

metal chips get into the nut assembly, the roller screws may grind them into a lapping

compound, which will cause spalling and eventual catastrophic failure. The result of

such contamination will lead to rapid degradation of the screws lubrication fluids,

increasing friction resulting in higher than normal loading and inevitable wearing of

the screw thread along with potential seizure of the roller assembly. Also, the

inadequate use of the correct lubricant to protect the assembly from heat build up,

caused through natural friction between mating metal surfaces will also lead to a

similar failure result. Improper loading is another possible cause of roller screw

failure. As a general rule, screws do not tolerate applied moments or side loading,

which can be caused through incorrect installation and may result in the warping of

the threaded screw further increasing the side loading.

2.12 Conclusion

For over 60 years the aerospace industry has been dominated by the use of hydraulic

and pneumatic drives with electrical drives being reserved for low power and load

applications on small light aircraft. The pneumatic and hydraulic power for these

drives originates as secondary engine power and therefore has a direct impact upon

the aircrafts fuel consumption and polluting emissions. With advances in electrical

power generation technology research across the aerospace sector has begun to seek to

reduce dependence of engine generated hydraulic secondary power for actuation

systems and more towards electrical powered systems. This has led to many advances

in electromechanical actuation technology for aerospace applications such as primary

flight control surfaces, engine actuation, cargo doors and landing gear retraction

mechanisms.

The issues with replacing hydraulic actuators with electromechanically counterparts

are mainly related to reliability. The actuators which are under development by

Messier Dowty utilise technology which is unproven in a landing gear application for

medium to large aircraft. They therefore have to be engineered to include redundancy



37

ensuring that they are robust enough to operate in the extreme environment of the

landing gear bay and are reliable enough to ensure successful operation.

The main concerns of actuator failure are mechanical jamming. There are a variety of

underlying faults which can increase in severity and lead to actuator jamming. These

have been identified through a Failure Mode Event Analysis of an actuator system as

bearing damage, gear seizure or physical actuator screw damage. Through the

elicitation of expert knowledge during the course of studying the actuator system, the

primary and most common cause for each of these failures begins with the

degradation of the lubrication system. It is therefore considered that in order to detect

incipient faults as early as possible, for this application the ability to detect changes to

the lubrication within the actuator system is of paramount importance.

The use of aerospace grade components can increase the reliability of individual

actuator components but failures will still inevitably occur. Further health monitoring

systems incorporated into the actuator system would offer the potential to increase this

reliability even further. This would aid in justifying the need for additional automated

fault detection and diagnostic health monitoring systems but the choice of acceptable

approaches will be limited through design restrictions, regulation and costs.

2.13 References

Alden, R. ‘C-141 and C-130 power-by-wire flight control systems’, (1991) Aerospace

and Electronics Conference: NAECON, Dayton, OH , USA

Bailey, M., Hale, N., Ucerpi, G., Hunt, J-A., Mollov, S., Forsyth, A. (1999),Distributed electrical power management architecture', IEEE Colloquium onElectrical Machines and Systems for the More Electric Aircraft, London, UK

Bradbury, G. 'Development of an advanced primary flight control electromechanical

actuator', (1987), Dayton, OH, USA: IEEE, New York, NY, USA.



38

Cloyd, J. (1998), 'Status of the United States Air Force's more electric aircraft

initiative', IEEE Aerospace and Electronic Systems Magazine, Vol 13, No 4, pp. 17-

22.

Collins, A. (1999), 'EABSYS: electrically actuated braking system'. IEE Colloquium

on Electrical Machines and Systems for the More Electric Aircraft, London, UK: IEE.

Conway, H.G., (1958), 'Landing gear design', The Royal Aeronautical Society.

Cutts, S.J. (2002), 'A collaborative approach to the more electric aircraft',

International Conference on Power Electronics, Machines and Drives, Bath, UK

Currey, N., (1998), 'Aircraft landing gear design: principles and practices', AIAA.

Dixon, R., Gifford, N., Sewell, C., Spalton, M. (1999), 'REACTS: Reliable electrical

actuation systems', IEE Colloquium: Electrical Machines and Systems for the More

Electric Aircraft, Vol 1999, No 180, pp. 23-38.

Elbuluk, M., Kankam, M. (1995a), 'Motor drive technologies for the power-by-wire

(PBW) program: options, trends and tradeoffs', Aerospace and Electronics

Conference, NAECON,

Elbuluk, M., Kankam, M. (1995b), 'Motor drive technologies for the power-by-wire

(PBW) program: options, trends and tradeoffs. I. Motors and controllers', Aerospace

and Electronic Systems Magazine, IEEE, Vol 10, No 11, pp 37-42

Faleiro, L. (2005), 'Beyond the more electric aircraft' Aerospace America,

Holmdahl, M. (1983), 'Putting new all electric technology development to the test',

NAECON, Dayton, OH, USA,



39

Jenkins, S. (1989), 'Landing gear design and development', Proceedings of the

Institution of Mechanical Engineers, Vol 203, No G1/1989, pp. 67-73

Jensen, S., Raymond, B., Dawson, D. (2000), 'Flight test experience with an

electromechanical actuator on the F-18 systems research aircraft', The 19th

Proceedings on Digital Avionics Systems Conferences, IEEE, Philadelphia, PA, USA

Krüger, W., Besselink, I., Cowling, D., Boan, D., Kortüm, W., Krabacher, W. (1997),

'Aircraft Landing Gear Dynamics: Simulation and Control', Vehicle System Dynamics,

Vol 28, No 2, pp. 119-158.

Krüger, W., (2000), 'Integrated design process for the development of semi-active

landing gears for transport aircraft', Doctorate Thesis

Pond, C,. Wyllie, C. (1983), 'Test results of a unique high power electric motor

actuator designed for space shuttle applications', IEEE

SBAC. (2007), 'Aircraft and environmental briefing papers', Aircraft Technology and

Emissions

Spitzer, C,. (1984), 'The all-electric aircraft: a systems view and proposed NASA

research programs', IEEE: Transactions on Aerospace and Electronic Systems,

Dayton, OH, USA.

Wijekoon, L., Wheeler, P., Clare, J., Whitley, C., Towers, G. (2009), 'Aircraft

electrical landing gear actuation using dual-output power converter with mutual power

circuit components', The 24th IEEE Applied Power Electronics Conference and

Exposition, Washington, DC, USA



40

Young, D. (1986), 'Aircraft landing gears - The past, present and future', Proceedings

of the Institution of Mechanical Engineers, Part D: Transport Engineering, Vol 200,

No D2, pp. 75-92.



41

C h a p t e r 3 : C o n d i t i o n B a s e d M a i n t e n a n c e f o rE n g i n e e r i n g S y s t e m s

3.1 Introduction

This chapter presents a general review within the area of engineering Condition Based

Maintenance (CBM) and related topics. The review covers a wide range of literature

including general methodologies, design guidelines, open systems architectural

frameworks, sensor systems, feature selection and health assessment strategies. An

industry wide review of the subject area applied to actuation systems is provided

along with critical reviews of the main fault diagnosis techniques and strategies, in the

context of aerospace electromechanical actuator health monitoring (Phillips et al.

2008).

There are several conventional maintenance strategies which are aimed at

safeguarding against sudden machine breakdowns. These include Run-to-Failure

(RTF), Time-Based Preventative Maintenance (TBPM) and Condition Based

Maintenance (CBM), each of which is described below:

1. Run-to-Failure Maintenance (RTF) is a method which allows any maintenance

action to be postponed until an actual machine breakdown occurs. This has the

advantage of requiring less planning and scheduling of maintenance activities.

RTF maintenance is at its most effective when the cost of a breakdown and risk to

human or environmental safety is negligible. Reliance upon RTF practice can lead

to unexpected and costly machine breakdowns. There is the possibility of

inexpensive component faults leading to further costly damage to other machine

systems. RTF maintenance also has a large requirement on maintenance

personnel and spares inventories.

2. Time-Based Preventative Maintenance (TBPM) operates on the principle of

carrying out periodic checkups based upon monitoring the number of machine

operating hours. During TBPM the machine is dismantled, cleaned and any



42

deteriorated parts are replaced. Overall this may not be cost efficient as the

process consumes large numbers of man hours. Components may also not have

fully deteriorated at the time of the maintenance activity and do not necessarily

require replacements, even though the philosophy and procedure dictates that they

must be. The approach can therefore lead to high costs associated with a higher

expenditure on spare parts and a loss of production whilst machine maintenance is

performed. TBPM therefore requires that optimised maintenance plans are

efficiently organised.

3. Condition Based Maintenance (CBM) is often regarded as the most advanced

maintenance strategy. CBM is aims at reducing the number of breakdowns by

monitoring the machinery with the purpose that faults may be detected at an early

incipient stage and in order for corrective actions to be scheduled for a convenient

time. CBM makes use of measurements of some physical parameter and through

monitoring the trends of these parameters over time, any indication of abnormal

behaviour can be identified. The tool for achieving this is widely known as

condition monitoring, even though in aerospace it is often given the more human

analogous name of 'health monitoring'. Threshold warning levels are constructed

to trigger maintenance activity if a specific parameter shows measurements

outside of the threshold regions. One difficulty however is dealing with false

alarms and when a large number of alarms are triggered prioritising them can

often be a difficult and time consuming task.

There is currently a drive in the majority of industries to turn away from the more

traditional RTF maintenance and TBPM and incorporate CBM systems supported

through health monitoring tools into engineering systems. CBM is proven to help

minimise maintenance costs, improve operational safety and it is effective at reducing

the severity and number of machine failures. The health monitoring of relevant

components or equipment offers the advantage of extending operating life and

increasing the availability of the machine or system. Some of the more generic

benefits which a CBM programme can provide are:



43

1. Change of maintenance philosophy

On condition

Opportunistic

Not “on failure” nor “per schedule”

Less interruption of mission schedule

2. Reduction in test equipment

Less intermediate and flight-line test equipment

35 percent less peculiar support equipment during system design and

development

Eliminated O-level test equipment

3. Benefits to the maintainer

Unprecedented insight into vehicle/fleet health

Less time spent on inspections

Better ability to plan maintenance

Simplified training

Improved fault detection

A CBM programme can be aimed at either fault diagnostics or prognostics (Jardine et

al. 2006). Diagnostics refers to a posterior event analysis and deals with fault

detection (indicates a fault has occurred), fault isolation (faulty component is

identified) and fault identification (the nature of the fault is determined). Prognosis is

a prior event analysis and deals with fault prediction before failure occurs.

It should be noted that a CBM programme is only worthwhile if the benefits can

significantly outweigh the costs of its introduction and upkeep. There are four generic

steps to CBM. These are the acquiring of data, the processing of the gathered data



44

with extraction of suitable features, diagnostics/prognostics and finally the

maintenance decision making as illustrated in Figure (3.1).

Figure 3.1: A Generic CBM Process

3.2 Fault and Failure Definitions

A fault can be defined as an un-permitted deviation off at least one characteristic

property resulting in abnormal behaviour of the machine, system or process as

illustrated in Figure (3.2). It should be noted that this abnormal behaviour, even

though it will be observably different to the nominal specifications, the machine,

system or process will still be able to perform its specified task, albeit at a reduced

level.

Figure 3.2: Schematic of the Fault Concept

A fault can be regarded as developing into a failure is when the machine or process no

longer operates to a defined specification. The effects of a failure can be calculated in

terms of costs such as spare parts, labour and lost production and can be described in

relative terms such as

FeatureExtraction

FaultProgression/

Trending

MaintenanceDecision

DataAcquisition

Process Change ofbehaviour

Faults



45

High (expensive)

Medium (moderate)

Low (cheap)

The effects of failure can further be subdivided into four main categories

1. Safety/environmental failures may have associated risks of injury to people or to

the damage environment. There are often high risks in the petrochemical industry,

transport industry and within heavy manufacturing industries.

2. Lost production can have a serious consequence for a manufacturing process, such

as a car assembly line or the inability to fly an aircraft. Stopped operations can be

difficult and expensive to recover.

3. Secondary damage often results when a supporting piece of equipment such as an

oil lubrication pump fails causing further damage to other system components.

4. Replacement/scrap costs may be significant high if equipment performs poorly

and or if it fails. Specialist equipment can be very expensive to replace or repair.

Failure mechanisms can be described in two ways. The first is that the failure is only

dependant on the condition variables reflected by a predetermined fault level. This

gives a failure definition as ‘a failure occurs when the fault reaches a predetermined

level’. The second builds a model for the failure mechanism using historical data. A

definition of a failure can thus be ‘the event that the machine is operating at an

unsatisfactory level’ or it can be described as a functional failure where ‘the machine

cannot perform its intended function’ (Jardine et al 2006). Faults within a system are

dependent upon time. There dependencies may take the form of abrupt faults,

incipient faults or intermittent faults as illustrated in Figure (3.3).



46

Figure 3.3: a) abrupt fault, b) incipient fault, c)

intermittent

3.3 Diagnostics and Prognostic Definitions

Diagnostic capabilities traditionally relate to the area between a fault developing and

complete system catastrophic failure. Advances in diagnostic technologies have

recently enabled detections to be made at much earlier incipient fault stages. In order

to maximise the benefits of continued operational life, maintenance will often be

delayed until the early incipient fault progresses to a more severe state, but still

maintained before catastrophic failure occurs. This idea pushes out of the realms of

fault diagnostics and further towards the concept of employing a prognostic strategy.

If an incipient fault has been detected by the machine operator, and the decision has

been made to continue to run the machine. Then the operator will want to know how

long they can actually run the machine for before a failure event occurs. This is

known as the Remaining Useful Life (RUL) (Engel et. al 2000; Kothamasu et al.

2006). Figure (3.4) represents a failure progression timeline (Vachtsevanos et al.

2006).

f f f

t t t

ab

c



47

Figure 3.4: Failure Progression Time Line

(Vachtsevanos 2006)

If a new fully functioning machine begins to show signs of an incipient fault which

will further develop into an eventual catastrophic failure as the machine is continued

to be used. The RUL is the point along this particular failure progression timeline in

which to stop usage and carry out maintenance. The aim of estimating the RUL is to

maximise usage, optimise maintenance operations, minimise downtime and reduce

costs of spare parts and revenue losses. This is the real domain of prognostics and

requires a large array of tools and knowledge. There must be sufficient data or

knowledge on both the fault propagation process and on the failure mechanisms but

understanding this suffers from at the following difficulties.

State Awareness Detection

Determine effects onthe rest of aircraft

DIAGNOSTICS

Predicted remaining useful life

Secondary damage,catastrophic failure

Need: better modelsto determine failure

effects acrosssubsystems

Develop: useful liferemaining prediction

models

Desire: Advancedsensors and detectiontechniques to “see”

incipient faults

Need: understandingof fault to failureprogression rate

System, Subsystemor component

failure

Very early incipientfault

New workingorder

PROGNOSTICS

The goal is to detect “state changes” as far to the left as possible



48

The idea of trending has become a widely accepted approach to interpreting long term

behaviour thought the use of analyzing past behaviour. However there is also an

underlying assumption that the path from the initial fault and catastrophic failure is

predictable and smoothly changing. This is unfortunately rarely true and a range of

typical behaviours is shown in Figure (3.5).

Figure 3.5: Failure Progression Trends from Incipient

Fault to Catastrophic Failure (Starr 2005)

The choice of monitoring architecture and methodologies will heavily depend upon

the given application coupled with reliability requirements, safety issues, restrictions

Remaining Useful Life

Initial fault

Catastrophicfailure

HealthStatus

Time

B

AC

D

A The ideal transition between detection and failureB Trend from a high sensitivity sensor – stays highC The trend is a cumulative count with stepsD Trend from a low sensitivity sensor - late



49

on additional equipment and sensor limitations. Some monitoring techniques react too

slowly to an initial event, this is due to limitations of the sensor sensitivity, the sensor

may be physically remote from the fault or because the processing and analysis of the

signal is insensitive (Starr 2005). Factors such as this need to be considered in the

initial design stages of any health monitoring system an effective integrated systems

based strategy should therefore be adopted (Kacprzynski et al. 2002).

3.4 Review of Condition Based Maintenance System Requirements

Every individual CBM systems will have to meet specific performance measures for a

given characteristic. However a set of general requirements applicable to all CBM

systems will need to be met. A list of such requirements may include (Vachtsevanos

et al. 2006):

The CBM system must ensure enhanced maintainability and reduced

Operational and Support (O&S) costs over the lifetime of the monitored

systems.

The CBM system must be designed as an open-system architecture that

maximises ease of subsystem and component changes and replacements with

minimising system/process change.

The weight and complexity of the CBM system must be closely controlled

The CBM system must meet reliability, availability, maintainability and

durability requirements.

Structural and environmental requirements

Cost requirements



50

Power requirements

Compatibility with existing hardware.

The implementation of a CBM system can be expensive. If this initial investment is to

be made then there must be guaranteed significant O&S costs. If these savings do not

materialise then the use of the CBM system is not cost effective and consequently not

worthwhile. Most large scale engineering systems, such as aircraft continuously have

individual subsystems and components upgraded as new technology advances are

made. This ensures that the lifetime of the aircraft, efficiency and safety is

maximised. The CBM system must be able to adapt to these often small changes.

The total redesign of an actuator CBM system in order to allow it to work on an

upgraded actuator with different specifications would not be acceptable in terms costs,

in particular time and financial funding. The implementation of CBM should have no

adverse affects on the performance of the system in which it is attributed to. This

means that if diagnostic information matching the CBM performance requirements

cannot be gained without adversely affecting the systems performance or the CBM

equipment reliably operating in a given environment then CBM will not be suitable.

It should be noted that even though CBM practitioners would like to see their systems

on all engineering systems, it is certainly not applicable for all engineering systems.

One of the main concerns of CBM systems is their performance. In some cases a false

alarm will have huge financial and safety implications. If for example a CBM system

on an aircraft engine did not correctly raise an alarm, there are two implications. The

first is that passenger safety is affected; the second is that the aircraft may end up

being grounded in an undesirable location, where maintenance expertise or spare parts

are not easily available. Conversely if the alarm is triggered unnecessarily then the

aircraft would be grounded whilst inspection work is carried out leading to flight

delays, disgruntled passengers and lost revenue.



51

3.5 Systems Based Strategy for Condition Based Maintenance

The design of a CBM will be heavily application dependant and no single architecture

exists or specific guidelines on CBM design. There is however generalised

frameworks which are useful and follow systems based design strategies. That is the

ideal design method is to incorporate CBM into systems rather than systems into

CBM. Figure (3.6) depicts the main modules of an integrated approach to CBM

design with brief description of the systems based components of this architecture

described below:

Figure 3.6: An Integrated Approach to CBM Design (Vachtsevanos 2006)

Trade Studies: The scope of a trade study is to arrive at the best or at least the most

balanced solution to the diagnosis/prognosis of the system. This should lead to

optimal CBM practices. Specific objectives include:

Establish the need

Design andtrade studies

FMECA

CBMtesting

DataCollection

DataAnalysis

AlgorithmDevelopment

Implementationvalidation and

verification



52

Define the problem

Establish value objectives

Generate feasible alternatives

Make decisions

Failure Modes and Effects Analysis (FMEA): The cornerstone of a good CBM

system design is the understanding the physics of failure mechanisms. The use of a

FMEA is aimed at providing the designer with the tools and procedures that will lead

towards a systematic and thorough framework for design. A FMEA has the following

properties:

Identifying failure modes, their location, severity, frequency of occurrence and

testability

Relates failure events to their root causes

Explain the impact of faults/failures on the system, subsystem or component

performance

Make suggestions for the sensors/monitoring equipment required to detect and

track a particular fault

System CBM Test Plan: The objective of a CBM test plan is to operate the system

under controlled conditions on a laboratory testing rig or within a simulation

environment if an appropriate model describing the process exists. The testing can

also be performed under real operating regimes if possible to obtain baseline fault

data. Baseline fault data can be used to eventually train and validate



53

diagnostic/prognostic algorithms. A systematic approach to planning a CBM test

procedure would be as follows:

Determine systems operating modes

Decide on a set of fault modes that can be seeded safely (it will rarely be

possible to test all failure modes due to practical reason such as equipment

costs and testing time scales)

Determine sensors and data acquisition equipment

Decide the number and nature of test runs for both baseline and fault data.

Performance Assessment: CBM systems are designed to meet multiple objectives by

providing useful information to a number of end users including the maintainer, the

operator or the process manager. Performance assessment studies are conducted to

evaluate the technical and economic feasibility of various diagnostic/prognostic

technologies. Technical performance metrics (Orsagh and Roemer 2000; Byington et

al. 2003) are created for all of the algorithms in the CBM framework from sensing and

feature extraction to diagnostics and prognostics.

Verification and validation of CBM Systems: Verification is defined as ensuring

that the system as built can meet the performance specifications as stated. Validation

can be defined as determining if the system is the correct system. Verification and

validation techniques for CBM technologies serve to ensure that delivered capabilities

meet the system design requirements the system performance metrics are useful at this

point and can serve as a foundation for verification and validation and can be used for

system evolutionary design. There are currently few formal or accepted techniques

exist (Vachtsevanos 2006).



54

3.6 Open Systems Architecture for Condition Based Maintenance

Several organisations such as the Machine Information Management Open Systems

Alliances (MIMOSA) and the Open System Architecture for Condition Based

Maintenance (OSA-CBM) have developed standards and architectures for CBM

systems. The most popular architecture has been proposed by OSA-CBM and is

based on seven levels of health monitoring functionality (Swearingen (2007). Figure

(3.7) outlines the flow of information between the seven layers in the system. A

general description of the seven layers is also given below

Figure 3.7: OSA-CBM Architecture

Level 1 Data Acquisition At this level the CBM system will be provided with

analytical data obtained through sensors positioned on the equipment to be monitored.

Data Acquisition

Data Manipulation

Health Monitor

Assessment

Prognostics

Presentation

Prognostics

Decision



55

Level 2 Data Manipulation Data obtained from level 1 will be received and

processed at this level and will output digitally filtered data and for example frequency

spectra or other extracted CBM features.

Level 3 Health Monitor The health monitoring level will be focused on comparing

the processed data against what would be expected in a healthy system. The

capability of generating CBM alarms through pre-set threshold levels will also be

available at this level.

Level 4 Health Assessment At this level the health assessment will be able to

determine if the monitored components health has degraded. The health assessment

will utilise and align fault symptoms generated at the health monitoring level to

propose fault possibilities and give a measure of fault severity.

Level 5 Prognostics Data and information available from all other levels is available

together with failure models and prognostic algorithms to calculate the future health of

the system. The prognostics level will return the health status for a specific time and

as a measure of the remaining useful life.

Level 6 Decision The focus here will be to generate a set of recommended actions.

Maintenance requirements, legislations, costs and risks will be taken into

consideration in order for decisions on effective landing gear maintenance scheduling

to be made.

Level 7 Presentation The presentation level will provide a transparent view of the

information obtained from the previous levels. The most important of which will be

the health assessment, prognostics and the decision levels. The ability to see the other

levels will be a crucial factor if any justifications are required.



56

3.7 Sensor Systems

The use of sensor equipment is fundamental to successful CBM. A sensor can be

defined as an electrical or mechanical device which maps the value of some

environmental attribute to a quantifiable measurement. The use of sensor suites

employed to collect data is what leads to an online realisation of diagnostic and

prognostic algorithms. Strategic issues which must be addressed include the type and

number of sensors to be used, location, cost, weight and dynamic range. Optimising

the best suite of sensors and in what location and capacity to use them is no trivial

task. There are however guidelines available for choosing sensor locations in military

and aerospace systems (Padula and Kincaid 1999). Traditionally sensors will already

have been placed as part of a control system or for the purpose of monitoring

performance objectives.

Every sensor detects some aspect of its environment in which it is operating. There are

numerous types of conventional sensors available and sensor technology progresses

rapidly in terms of specialisation, miniaturisation and performance characteristics

(Kanoun and Trankler 2004). It is not in the scope of this thesis to give a full detailed

review on the operating principles of individual sensor type but a brief non exhaustive

list is given below.

Mechanical Sensors Pressure sensors, accelerometers, displacement

transducers, strain gauges, force sensors, ultrasonic sensors, angular velocity

sensors and acoustic wave sensors.

Thermal Sensors Thermocouples, diode and transistor temperature sensors,

thermistors, pyroelectric and piezoelectric thermometers.

Optical Sensors Photoconductors, photodiodes, phototransistors, position-

sensitive photodetectors.



57

Magnetic Sensors Megneto resistors, hall-effect devices and magnetometers

Other Sensors Viscosity sensors, proximity sensors, altimeters, gas sensors,

humidity sensors and acoustic velocity sensors

3.7.1 Technology Aspects of Sensors

There are a variety of aspects to consider when choosing appropriate sensors; some of

the more important are reviewed below (Fraden 1993):

Sensitivity: the minimum magnitude of the input signal required to produce a

specified output signal.

Measuring range: the range of values that the sensor can measure effectively.

Resolution: the smallest change the sensor can detect in the parameter that is

being measured.

Stability: changes within the sensors performance over a period of minutes,

hours or days is known as the short-term stability. Long-term stability

depends upon the operating conditions of the sensor and is related to the aging

of the sensors electrical, mechanical or thermal material properties. Long-term

stability is an important aspect to consider for sensors required to perform

precision measurements.

Reliability: the ability of a sensor to perform a required function under the

given operating conditions for a given period of time.

Accuracy: this is measured as the highest deviation of a value represented by

the sensor from an ideal or true value at its input.

Selectivity: the ability of the sensor to reject other nearby signals



58

Repeatability: the ability of the sensor to represent the same measured value

under identical conditions.

Response speed: the time between the measured phenomena occurs and the

sensor providing an output measurement

Cost, size, weight: these are geared to specific requirements for a given

application. Cost is a cornerstone in sensor design for a broad range of

applications; however it may be a secondary issue when the sensors reliability

and accuracy are of paramount importance, for example, in life support

equipment, weapons or spacecraft.

3.7.2 Wireless and Smart Sensors

With recent developments in sensor technology there is much emphasis on the use of

wireless technologies such as Bluetooth and Wi-Fi as well as the use of intelligent

sensors (smart sensors) within health monitoring (Starr et al. 2007). An intelligent and

wireless sensor offers a variety of advanced functionalities that are beyond that of

conventional sensors, such as on-board processing. They have a capability of

adapting to changes within the local environment, they can be autonomous and self

adjust to effects caused through faults making them more robust. Sensors added to a

system as part of a health monitoring system must be reliable and robust to avoid the

transmitting of false information.

Wireless intelligent sensors are also capable of communicating accurate, self validated

and reliable signals to higher-level systems for the purposes of information fusion;

tracking or estimation (Pietruszkiewicz et al. 2006). Often it can be the case within an

industrial environment where there are literally hundreds of wires connected to

various components. This creates a huge risk of fire and also with every human’s

experience of the difficulties and nuisances that just a few entangled wires can cause;



59

industry takes very serious the possibility use of wireless smart sensing. Smart

wireless sensors in a maintenance role provide the ability to remotely monitor

machinery from potentially any global position outside of the machines location. This

has recently been investigated through a sizeable European Union 6th Framework

Programme known as Dynamite (Dynamic Decisions in Maintenance)*.

Remote health monitoring is particularly suitable to the monitoring of aerospace

systems by the very nature of the systems. Aircraft are rarely in one geographical

location for very long and each airport must have suitably trained personnel to inspect

key aircraft systems. With remote health monitoring data can be transmitted to any

location reducing the number of maintenance personnel stationed at each airport.

3.7.3 Multiple Sensor Networks

Experience in industry has shown that there are significant advantages offered by

utilising multiple sensors in an application in the form of sensor networks. These

advantages can include:

Robust system: If a health monitoring system depends upon a single sensor

source then decisions may not be robust and reliable. If the single data source

fails, then the output of the monitoring system will be adversely affected.

Fusion of several sources of data can have a higher fault-tolerance rate.

Situation awareness: This can increase reaction time by taking advantage of

different sensor sources. By fusing the results would give a better situation

awareness which would lead to better decision making.

Improved data accuracy: Fusing of multiple data sources can remove data

ambiguity improving data accuracy and reducing uncertainty.

* The DYNAMITE project ran between 2005and 2008, consisted of a budget of 6.3 million Euros and 17 partners. Ofwhich the University of Manchester was a major academic partner.



60

Extended parameter coverage: More data sources provide extended

information regarding an object, generating a more complete picture of a

situation.

The number of sensors introduced into a system requires consideration to be given to

aspects such as costs and complexity. Costs calculated in terms of time, money and

effort. Sensor networks can provide a reduction in measurement time, and therefore

reduce costs; increased numbers of sensors however require more computational effort

and add complexity to a system. System complexity can have adverse effects upon

the reliability of the system, the higher the number of components, the higher the

number and frequency of possible failures. These factors require a trade-off between

sensor numbers and system level requirements. If different types of sensors within a

sensor network are used then the data needs to be aligned to a common form. The

number and type of sensors required depends on the application and the results

required. Data reliability will depend on the availability of sensors and the fusion

methodology; of which there are several.

If large sensor arrays are to be used in a health monitoring context, then it is essential

that appropriate frameworks are in place to deal with the vast quantities of data.

Issues particularly arise with the merging of data to obtain useful information from a

range of sensors which provide non-commensurate data. One multidisciplinary

approach is loosely termed data or information fusion and provides a variety of

generic frameworks and tools suitable for health monitoring.

3.8 Data Fusion Overview

Data fusion is the process of using collaborative or competitive information obtained

from multiple sources to deduce a more confident and informed decision regarding a

situation. The data fusion paradigm is that the whole is greater than the sum of its

parts (Iyenger et al 2003). That is, fusing data from multiple sources, results in an

output which is much more enriched in information than from a single source. Data

fusion converts data into information which is subsequently combined with



61

knowledge and logic to aid in inference and decision making. It is this ability which

makes the use of data fusion an attractive model in which to base a decision critical

health monitoring system upon. Data fusion has been an active area of research and

systems development for about 20 years and is analogous to the human brain which

fuses information regarding an environment from sensory perceptions such as touch,

vision, hearing and smell, this is then used to derive knowledge or draw conclusions.

Figure (3.8) illustrates this analogy.

Figure3.8: The Intelligent Fusion Process

Data fusion incorporates many of the techniques common with disciplines such as

resource management, sensor management, correlation and data mining. The fusion

process is implemented into a system in order to provide support for the system

operators, who, without the support of a data fusion system, would need to manually

examine the information to achieve timely, robust and reliable situation assessments

or projections (Steinberg et al 1999). Data fusion has found uses in a wide range of

applications; Table 3.1 provides a small overview.

Acousticsensor

Hearing

Temperaturesensor

Pressuresensor

EO sensor

Radar

Vision

Touch

Chemicalsensor

Intelligent ProcessingInference and Decisions

Smell



62

Table 3.1: Overview of Data Fusion ApplicationsApplication Inference Primary Data

Medical Diagnostics(Pattichis et al. 2001)

The identificationand location ofabnormalities anddisease

X-rays Temperature NMR Chemical/biological

data Visual inspections

Robot Navigation (Luo et al1988)

The identificationand location ofobstacles and objects.

Acoustic signals EM signals Infrared signals

Condition BasedMaintenance (Starr et al.2002a, 2002b)

Detection, diagnosisand isolation ofmachine faults andcorrectivemaintenancerecommendations

Acoustic signals Vibrations Temperature Wear debris Current signatures

Transport Systems (ElFauzi 2004)

Fusion ofinformation toprovide better travelrelated information

CCTV Weather reports Incident reports Vehicle location

3.8.1 Fusion Processes

Before discussing fusion strategies it should be noted that the fusing of multiple data

streams is by no means a trivial process, the decision to implement data fusion into a

system has its associated difficulties, some of which are outlined below:

Sensor diversity: sensors will be of differing types, all have different

properties including outputs, synchronisation, and location.

Data diversity: different sensors make different measurements and it is highly

likely that they will have different characteristics and dimensionality (e.g.

miles feet and inches).

Sensor calibration: sensors will need to be correctly calibrated to cope with

errors and uncertainties.



63

Sensor limitations: sensors all have limitations on the environments that they

can operate in and measurement ranges etc

This is a non-exclusive list of difficulties that arise during data fusion implementation.

A prime architectural issue is what level fusion is to take place. Fusion can take place

on three levels, depending on the application requirements; these are fusing at the

data, feature and decision levels.

Data level fusion where information obtained from sensor-arrays are fused

together in order to validate signals and create features.

Feature level fusion combines features are extracted from the raw sensor data

and are then combined to obtain diagnostic information.

Decision level fusion incorporates experience-based information such as

physical model predictions, failure rates, management data and heuristic

knowledge to provide confident decisions.

3.8.2 Data Fusion Models

In a centralised fusion all of the available data is collected together into a central

processor where it is fused together, allowing decisions to be made. The fusion

process is performed on raw data and is usually commensurate, for example, infrared

images and satellite images and fused using pattern recognition, estimation and

statistical techniques which require high computational requirements. A centralised

fusion model is shown in Figure (3.9).



64

Figure 3.9: Centralised Fusion

Decentralised fusion systems, Figure (3.10), have no central processing facility as in

centralised fusion. They are composed of a series of nodes, each of which has its own

individual processing facility where state vectors are generated from individual

signals. These form the inputs to the fusion centre where a fused identity is generated

using techniques such as Dempster-Shafer and Bayesian theory. Decentralised fusion

fuses data at a higher level than centralised fusion which fuses at a data level whereas

decentralised fusion fuses features or decisions.

Figure 3.10: Decentralised Fusion

It is on occasions necessary to take advantages of both centralised and decentralised

data fusion. In this case combinations of the two are used as hybrid architectures. In

this case state vector fusion is performed to reduce the computational workload and

communications demand, with data level fusion being performed on demand when

Sensor 2

Sensor 1

Sensor n

FusionCentre

DecisionLevel

FeatureExtraction

FusionCentre

DecisionLevel

Sensor 2

Sensor 1

Sensor n



65

more accuracy is required. Alternatively a combination of state vectors and raw data

may be fused depending upon available sensors.

3.8.3 Architecture Selection

The centralised fusion architecture is generally more often used than the decentralised

or hybrid options. The use of centralised architectures may be computationally

intensive but they carry the advantage of developing a picture of an object from the

raw data. Original raw data contains information regarding data quality, which may be

lost or diminished in higher fusion levels used in decentralised fusion (Linas and Hall

1998). Decentralised architectures reduce the amount of data and hence require much

less computational effort and data handling capabilities, but at the cost of adding a

signal processing unit to each sensor increases the overall system complexity. The

hybrid architecture is the most complex architecture to use but it does offer the most

flexible fusion approach, a hybrid fusion system requires a monitoring system/person

to select between data and state vector fusion.

There is no individually best architecture to adopt when designing a data fusion

system, the choice of fusion architecture is a matter of the requirements of the

application and is therefore a system-engineering problem. These include data

exploitation level, data availability, strategic planning, computing constraints, cost

along with temporal and special issues.

3.8.4 The JDL Data Fusion Process Model

Many generic data process models exist, several in depth reviews of the various

models have been provided in the literature (Esteban eta al 2005). However the most

intuitive and applicable model to consider for the current application is that proposed

by the Joint Directors of Laboratories (JDL) data fusion sub-panel within the US

Department of Defence to aid in the development of military systems.

The JDL model is a generalised framework which has found popular use in many data

fusion applications including the development of intelligent health monitoring. The



66

framework provides no formal rigorous rule system for the development of a fusion

system, with no guidelines on appropriate fusion levels, or selection of fusion

algorithms or techniques. It does however provide a multi-level framework allowing

for the attention and refinement of the key elements in the fusion system. Figure

(3.11) depicts the JDL data fusion process model.

Figure 3.11: The JDL Data Fusion Process Model

Low Level Fusion

Level 0 - Source Pre-Processing

The lowest level of fusion is referred to as the fusion at the signal level (after signal

conditioning) and pixel level fusion (imagery fusion). This involves individual

sensors multiple detection inputs to the signal-processing unit. This stage of fusion

also looks at reducing the quantity of data whilst retaining the useful information for

higher level processing

Level 1 – Object Refinement

The attributes of the multiple data sources obtained through processing at level 0 are

fused here in an attempt to locate and identify objects. The process involves data

alignment, association and correlation, state estimation (i.e. position, speed etc) and

refinement of an entities identity. The output at this level will be object classification

and identification; state and orientation.



67

Higher Level Fusion

Level 2 – Situation Assessment

Level 2 processing aims to provide a higher level of understanding of the output from

level 1 processing. It identifies the most probable situation based on the observed data

and events through establishing relationships among objects and assessing these

relationships to identify the meaning of entities in a specific environment. The output

at this level would for example be a collection of higher-order inferences providing a

view of what is happening.

Level 3 – Impact Assessment

The purpose of processing at this level is to assess what the impact of the events

derived from level 2 processing. This could include assessing the threat or danger

level, predicting the possible outcome, assessing asset vulnerability and analysing the

advantages and disadvantages of taking one course of action over another

Level 4 – Process Refinement

Level 4 processing is the refinement stage. This is often implemented with the

purpose of improving the fusion process between levels 0 to 3. This level can identify

potential sources of information enhancement, manage resources such as the sensors

and prioritise tasks.

Implementing all levels of fusion is a complicated task, low and higher level fusion

processes both have distinct operational functions; low level fusion must address the

complexity of association, short term history process and structural issues. High-level

fusion handles intelligently the integration of present information, long-term history

and the problems with recalling information

3.9 Health Monitoring Techniques

Once the decision to design a health monitoring system has been made, the next step

is to determine the choice of variables to be monitored. There are a multitude of



68

techniques available for measuring and associating variables with faults. Some of

these require specialist equipment, specialist training and some require complex signal

processing techniques each of these factors affect the monitoring choice. Each

technique is appropriate for different machine or system types and the detecting of

particular failures and this section reviews the main signal based approaches.

3.9.1 Vibration Monitoring

The health monitoring market has for 40 years been dominated by frequency analysis

of machine vibrations. All dynamical systems will generate vibrations whilst in

operation (Rao 1996). Individual machine components will usually have distinctive

vibration patterns which are symptomatic of their health. These vibration patterns if

monitored and analysed can provide a good indication of deterioration in machine

health. A machines vibration signature is a complex signal which is a mixture of

sinusoidal waveforms all of different amplitudes, frequencies and phase differences

which relate to fundamental rotational speed. Vibration can be measured in terms of

various parameters some of the most popular include displacement, velocity,

acceleration, frequency, bandwidth, spike energy or power spectral density.

The use of vibration signals in health monitoring depends upon the quality of the

measured signal. This means that the choice and location of measuring transducers is

of particular importance if the health monitoring system is to be based upon vibration

analysis. There are a variety of vibration signal processing techniques and methods

used to distinguish faults such as imbalance, looseness, misalignment, wear, poor

lubrication and structural cracks in rotating machinery (Wua 2009; Poyhonen et al.

2004), these include cepstrum analysis, spectrum analysis and autoregressive

modelling. Other methods include the measurement of spike energy and shock pulse

methods. These can be used when there is no deformation present and is successful at

detecting a lack of lubrication and wear in aerospace engine bearings (Byington et al.

2004). Vibration analysis has also found popular use in Structural health Monitoring

of composite materials (Montalvao et al. 2006) and in the health monitoring of



69

traditional aerospace structures (Bovio and Lecce 2006) but is applied extensively to

rotating machinery.

3.9.2 Lubricant Wear Debris

Lubrication in a system is usually required for reducing friction, cooling components

and to clean load bearing surfaces. In systems where it is difficult or impossible to

mount sensors it can be beneficial to monitor the actual lubricant which is in direct

contact with moving parts. This can be an effective way of not only assessing the oil

quality but also the condition of the components in which the fluids have come into

contact with. As time passes the lubricant will begin to degrade and components will

begin experience increased friction due to metal-metal contact leading to wear. The

result of which will be a build-up of both metallic and non metallic particles within the

system. Other effects which can lead to the presence of wear debris are the failure of

filters or the corrosion of the metallic components. By observing the size, quantity,

material composition and the shape of wear debris a number of identifiable faults may

be classified (Khan et al 2008).

The material composition can identify which particular component the particles have

originated from. Their shape can help to indicate the mechanism which is causing the

build up of debris and the quantity and size of the particles can give a good indication

of the rate at which any damage is occurring (Khan and Starr 2006). The monitoring

of the lubrication also offers the ability to remove any large particles through the use

of filters and magnets before they can cause more wear and damage to seals through

abrasive effects. When monitoring lubricant wear debris measurements should be

taken from the path of the lubrications flow to get accurate results. These

measurements are then analysed using techniques such as the use of magnetic chip

detectors, ferrography, spectrography or visual inspections (both manual and

automated).



70

3.9.3 Motor Current Signature Monitoring

Motor Current Signature Analysis (MCSA) is a technology developed for the specific

task of monitoring electrical machinery (Nandi et al. 2005). Many of the vibration

techniques outlined earlier can be used to effectively diagnose most faults within

electrical motors. It is argued that a better and fuller understanding of the motors

electrical and mechanical condition can be obtained by monitoring the current passing

through it. MCSA is based upon small time independent motor load variations

generated from within the mechanical system and converting them into electrical

signals that flow along the cable supplying the power to the motor. These signals can

be extracted and used as an indicator to the motors condition. Measured current

signatures can be analysed in both the time and the frequency domain, with analysis in

the time domain can be useful in the initial or final stages of the motors operation.

Monitoring motor current has a number of beneficial strengths and can provide a non-

intrusive monitoring capability. Through the use of Fourier analysis the current

signature can provide good degradation and diagnostic information on damage such as

broken rotors, unbalanced magnetic forces, winding problems, mechanical unbalance

and bent shafts amongst others. The monitoring of the electrical current has a high

level of sensitivity to a variety of mechanical disorders affecting the machines

operation and it is cost effective and can be performed by relatively unskilled

personnel. Literature surveys have shown that MCSA is a popular tool in the fault

diagnosis of electrical machines and has found extensive use in the monitoring of

induction drives used for driving of a vast range of high load bearing machinery

(Thomson and Fenger 2001; Bendouzid 2000). A current based detection of faults has

been shown as a viable and cheaper than vibration analysis tool for use with brushless

DC motors to detect faults in rotors (Rajagopalan 2004) and has been demonstrated by

successfully detecting broken gear teeth, a lack of lubrication and wear particles in the

gear lubrication (Rajagopalan 2006).



71

3.9.4 Thermal Monitoring

An early indication of component failure can be achieved through the monitoring of

temperature changes. This is a particularly applicable technique for components

which generate, transfer or store energy as heat. These include thermal insulation,

motors, coolant/lubricant, electrical wiring and bearing housings. A temperature rise

in these components will be indicative of a developing fault and may be caused by

such things as increased friction due to lack of lubricant, incorrect electrical loading

and damaged thermal or electrical insulation.

There are two categories of thermal monitoring methods: contact and non-contact.

Contact methods will require a temperature measuring device such as a visual

indicator (mercury thermometers, temperature paints), thermocouple or resistive

devices to be placed on or within the surface of the monitored component. The use of

non-contact methods have the advantage that large areas can be surveyed quickly and

at a distance and is particularly useful where access is difficult such as inside rotating

equipment. This approach uses the principle that all bodies radiate energy in

proportion to their temperature and that it is possible to relate the wavelength of the

radiation to the temperature of the radiating body.

3.9.5 Acoustic Emission Monitoring

Acoustic Emissions (AE) are produced from the rapid release of energy from sources

within a material. These waves are converted into voltage signals by the use of small

piezoelectric sensors mounted in suitable locations. The sensor response and front

end filters remove noise below 100 kHz, this includes most audible noise. This allows

acoustic emissions to be used to measure the condition of a structure even when

ambient noise levels are extremely high. Acoustic emission sources include fractures,

plastic deformation, impacts, and friction along with many other processes. Acoustic

emission is sensitive enough to detect cracks and fractures down to a few hundred

square micrometers or less (Finlayson et al. 2001).



72

3.9.6 Performance Monitoring

Performance monitoring makes use of process information to indicate the efficiency

of equipment. Any changes will be an indication that the condition is deteriorating.

The advantages of performance monitoring are that it is cheap and simple to

implement because most of the required instrumentation already exists. The main

disadvantage is that most of the parameters which can be measured will depend on

some other variable making it difficult to calculate a performance indicator.

3.9.7 Corrosion Monitoring

Any fluid present in a process or in the operating environment can lead to a corrosion

of parts. Severe corrosion can then lead to a gradual or sudden breakdown. A

deterioration in performance caused by corrosion may however be detected through

other monitoring techniques. Corrosion is usually a problem more associated with

structural components and techniques used to determine its extent include chemical

analysis, ultrasonic testing and electrical methods.

3.10 Critical Review of Monitoring Methods

Vibration analysis works well for continuously rotating machinery where vibrations

provide good stable symptoms of health. The electric actuation as part of the new ‘all

electric landing gear’ which is the subject of this thesis operates in a start and stop

fashion at often irregular intervals. The actuation mechanisms have varying speeds

throughout the operational cycle with varying friction, environment and mechanical

advantage. Vibration is dependent upon the rotating speed of the machine and if it

operates under varying speed, its vibrations will become non-stationary. The

rotational angle does not remain directly proportional with time and so conventional

methods of signal processing become inappropriate when monitoring the vibrations of

varying speed machinery.

There are various methods which can be used for motor current signature monitoring

(Kliman and Stein 1992) which has been shown to work well in fault detection

schemes. However they all assume that the load remains constant and does not vary



73

with rotational speed. Electromechanical actuation devices work by unlocking a

mechanism and causing the displacement of a lever-actuated or gear transmission,

linear displacement of an object, with it being stopped against a locking mechanism.

This leads to higher loads at the beginning and end of its operation. Variable friction

also creates non constant loads. If the load varies with rotor position such as it does in

a reciprocating compressor or in the proposed motor drives, then the current will

contain spectral components which will coincide and overlap with those caused by

any fault condition. This complicates any fault detection scheme. The problem of

motor current spectral analysis in the presence of time-varying loads in AC induction

machines is addressed by (Schoen and Habetler 1995). Other treatments such as

Wavelet Packet Decomposition with application to induction drives (Zhongming et al.

2003) has successfully differentiated between healthy and faulty conditions by giving

good feature representations of multiple frequency resolutions for faulty conditions. It

has been shown to give a better treatment of the stator current than the currently used

Fourier techniques in the detecting of bearing faults under varying speeds and loads

(Eren and Devaney 2002). There is however very little published work on the use of

MCSA for DC motors operating in a similar fashion to the proposed drives.

Vibration analysis has proven to be the most reliable and popular method of gear fault

diagnosis but the cost of sensors such as accelerometers and the associated wiring has

made the measuring of vibration a disadvantage in cost-sensitive areas. MCSA

however has shown to be a cost effective alternative to vibration analysis in the

detection of gear faults and has the potential to overtake vibration in detecting faulty

gears due to the advantage of it being cheap and easy to use.

In aircraft landing gear there are also problems associated with the use of health

monitoring based upon performance monitoring. Warm motor starts are as common

as cold motor starts due to varying operating environments which will lead to widely

varying current signatures meaning that conventional threshold warning levels will not

work. This is also the case with thermal monitoring of the motors, where the ambient

temperatures can affect the measurements. The temperatures in the landing gear bay



74

can change rapidly requiring regular adjustments to the sensor readings. This will

make it difficult to use thermal monitoring as a health monitoring technique when the

temperature may vary rapidly. To obtain a temperature reading devices such as a

pyrometer or infrared devices focus the energy from the source onto a sensor where

the reading can be processed and displayed as a temperature. Difficulties arise with

this method because different materials emit different energies when at the same

temperatures. Infrared measurements are very sensitive to the ambient conditions so

care must be taken to ensure that this is correctly compensated for when acquiring

thermal data. Also it may be difficult to use thermal monitoring effectively as a health

monitoring technique when the temperature may vary rapidly. The use of intelligent

sensors which may be able to self-adjust to fluctuating environmental changes may

offer benefits and solutions to the condition monitoring problems associated with

varying environments.

Where lubricant or hydraulic fluid is present, the monitoring of these can be seen as a

complimentary role to other monitoring techniques such as vibration analysis. This

can give information regarding the results of exposure to variable duty cycles.

Lubricant monitoring works well for variable speeds, variable loads such as engines

and for mobile vehicles. The acquisition of Lubricant wear debris data within motor

driven actuators is often difficult or impossible to obtain. This is typically due to the

motor bearings being greased and sealed.

The landing gear retraction and extension drives contain many materials which are

susceptible to fatigue, fracture, friction and corrosion which can all be detected

through AE. The use of AE are already regarded as reliable and is in wide spread use

as a structural health monitoring technology in the aerospace industry for the

monitoring of composite structures and helicopter drive trains and gears (Finlayson et

al. 2001; Saniger et al. 2002; Hood and Pines 2001). AE can detect structural defects

long before a possible catastrophic failure and therefore some interesting possibilities

may exist for them to be used in a complimentary role to other condition monitoring

techniques in the overall health monitoring of landing gear systems. The main



75

disadvantage however is the number of required sensors and associated costs and

weight.

3.11 Expert Systems

The use of expert systems in health monitoring provides an effective tool for decision

making. Expert systems can be classified as either data or knowledge orientated.

Data-driven methods have a high dependency on data collection and analysis whereas

the second knowledge-driven class of methods have a lower dependency on measured

data and much more on what is already known about the physics of the system, what

has happened before and from knowledge captured from human experts. Table 3.2

provides an example list of different approaches.

Table 3.2: Data and Knowledge Based Methods

Data Knowledge

Statistical methods Knowledge Models

Optimisers Physical Models

Neural Networks Case Based Reasoning

Pattern Classifiers Rule Based Reasoning

3.11.1 Model-Based Expert Systems

There are a variety of different model types which can be used for health monitoring.

For example physics based stochastic models have been used for gas turbines fault

diagnostics (Roemer and Kacprzynski 2000). The use of thermodynamic has been

built for health monitoring of diesel engines (Hountalas et al. 1999) and for the

monitoring of traction motors (Sen and Muttey 1999). The use of finite element

modelling has shown effective results and has been used to model torsion vibration

analysis in power trains (Crowther and Zhang 2003) and for fault identification in

turbochargers (Pantelelis et al. 2000).



76

The most popular model based approach is the use of dynamic process models and is

illustrated in Figure (3.12).

Figure 3.12: The Model-Based Fault DiagnosticsProcess

A variety of different methods for health monitoring based upon dynamic models

have been developed over the past two decades (Abidin et al. 2000; Isermann 2005;

Pedragal and Carneo 2006). Several common approaches include the use of

parameter estimation, observers and parity equations. All of these methods operate by

generating a set of residuals which can be compared to the systems nominal behaviour

and hence used to indicate any faults which are present or developing. These residuals

can be analysed and machine faults can be detected, isolated and identified.

Parameters, features,states etc

Normal behaviour

Symptom generation

Diagnose

Detectchange

Featuregeneration

Processmodel

Actuators SensorsProcessOutput

FaultsFaults

Input

Faults



77

In most practical cases the parameters describing a dynamic system will be unknown.

Parameter estimation algorithms such as least squares, recursive least squares,

instrumental variables, prediction error methods or optimisation techniques are used to

minimise the errors between a measured input and output signal and signals obtained

from a model describing the process to estimate the system parameters. Faults which

have already occurred or are beginning to develop within the system will often

manifest within the parameters making the tracking of these changes ideal for fault

detection work.

If the process parameters are already known then fault detection may be achieved

through the use of a classical state or output observer. Fault detection is achieved

through the calculation of the error between measured and estimated outputs. If a

fault can be detected through state variable changes then classical observers can be

used in fault detection. The use of output observers can be used if reconstructing the

state variables is not of use. A linear transformation will then lead to new state

variables. Observers have been demonstrated as offering a means to effectively

identify signal offsets indicative of sensor faults but are less effective at identifying

parametric faults (Patton and Chen 1997; Hammouri et al. 1999).

The principle of using parity equations (equations linking different variables) in fault

detection is to check for parity between the measurements generated by the process

and a set of residuals by comparing the model and the process behaviours. Parity

equations have advantages in detecting additive faults and are feasible for

corresponding faults in the sensors or actuators. The use of parity relations are also

easier to implement than output or state observers and lead to similar comparable

results (Patton and Chen 1992).

3.11.2 Knowledge Based Rule Systems

Knowledge-based systems (Harris-Jones 1995; Ignizio 1991; Hopgood 1993) are a

type of expert system designed to embody expertise in a particular specialised domain.

This system is intended to act like a human expert who can be consulted on a range of



78

problems which fall within their area of expertise. The basic concept of knowledge

base is that the user supplies facts or other information to the system and receives

expert advice (expertise) in response as illustrated in Figure (3.13).

Figure3.13: Basic Concept of a Rule Based Expert

System

A knowledge based system should be capable of responding at a level of competency

equal to or better than an expert within that field. This must be performed in a

reasonable time and the systems performance must be reliable and the users have

confidence in the received expert advice. Human safety and security may be

dependable upon the answers provided by the expert system.

The explanation facility incorporated into the expert system may be a simple set of

rules which led to a particular decision or may consist of more elaborate and complex

explanations. The large amount of knowledge that an expert system may have means

that it is important to have an efficient mechanism for adding, changing and deleting

knowledge.

The attempt to design knowledge-bases has been based upon the attempt to emulate

human thinking. One such characteristic is human’s ability to recall previous

experience to deal with a similar situation. This has led to the utilisation of case-based

reasoning in a heath monitoring context. If a problem that has been previously

User

Facts

Expertise

Knowledge Base

Interface Engine



79

diagnosed by the expert system it will store the information for use if a similar event

re-occurs. Therefore the diagnosis can be completed by simply recalling the previous

solution.

3.11.3 Neural Networks

Artificial Neural Networks (ANN) belongs to a family of numerical learning

techniques. They are models which are designed to emulate a biological neural

network. The inputs which the artificial neuron receives are analogue to the

electrochemical impulses that biological neurons receive from other neurons.

Practically they are however much simpler than biological ones so it is not expected

for them to produce the sophisticated behaviour of humans. They can however

perform certain tasks very effectively in particular classifications.

The artificial neural network is built up of individual nodes which each independently

perform a simple computation. This means that neural networks have a highly parallel

structure allowing them to explore many competing hypothesis simultaneously. The

most commonly used neural network is the Multi-Layer Perceptron (MLP) network

which is composed of an input layer, a number of hidden layers and output nodes. The

processing is done at each node (neuron) and consists of multiplying each input by a

weight, adding the weighted inputs together and passing the sum through a sigmoid

activation function.

Many authors have focused on the use of neural networks in health monitoring

applications for solving fault detection and isolation tasks such as in industrial

processes (Patan and Parisini 2005; Koeppin-Seliger and Frank 1995; Korbicz and

Janczak 2002).

3.11.4 Fuzzy Systems

Whilst probability theory is useful and works well in measuring what the likelihood of

a hypothesis is it will not say anything about the meaning of the hypothesis. Fuzzy

logic (also known as possibility theory) can address this problem providing a



80

systematic framework for the processing of vague qualitative knowledge by allowing

a “degree of truth” rather than the usual “true or false” offered by neural networks.

Fuzzy systems are therefore suitable for uncertain or approximate reasoning.

3.11.5 Uncertainty in Expert Systems

In the real world situations are never clearly true or false and human knowledge is

often vague, inexact and incomplete. This leads uncertainty within diagnostic

decisions and it has long been recognised that uncertainty should be included into an

expert system. There are two forms of uncertainty in an expert system, the first is the

uncertainty about the validity of the rule and the second is the uncertainty in the expert

systems user response. There are several common techniques used when dealing with

uncertainty which include Bayesian updating, Dempster-Shafer theory and fuzzy

logic/sets.

Bayesian updating is a technique for reasoning with uncertainty and has a rigorous

derivation which is built upon probability theory. However the underlying

assumptions which are made my not be true in practical situations. It assumes that it

is possible to give every hypothesis a probability and that this probability can be

updated in the light of new evidence. The Dempster-Shafer theory is also built upon

probability and is often regarded as a generalisation of the Bayesian methods.

Dempster-Shafer assigns a degree of belief as a measure of evidence that supports the

hypothesis; therefore a decision is made in favour of the hypothesis which contains

the most believable evidence.

3.12 Critical Review of Health Monitoring Strategies

Relying heavily on data-driven methods would require large amounts of data to be

collected directly from measurements taken from the landing gear actuators. Measured

data can be utilised from several sources such as the addition of transducers on the

actuators, continuously monitoring various parameters. Data can be obtained from the

actuator control system and the on-board flight data which the aircraft monitors and

records during all phases of its flight, including pre-flight taxi, take-off, airborne,



81

landing and post-flight taxi, data is also available through the aircrafts Built In Tests

(BIT). Applying data driven methods such as neural networks to actuator fault

diagnostics can provide good and accurate decisions, they are extremely good at

pattern classification but are generally regarded as ‘black box’, they are therefore only

focused on inputs and outputs and do not allow the internal logic to be accessed and

inspected. This leads to difficulties in justifying maintenance decisions to aircraft

operators due to the large financial losses that can be occurred when incorrectly

grounding aircraft and the risk to passenger safety if the decision to fly the aircraft is

wrong. There is therefore the need to provide justifications reasons for a decision.

Due to legislations and aircraft construction requirements the use of sensor equipment

can be limited, for example in landing gears where weight minimisation is currently

the major driving force, large sensor arrays are not possible. They add complexity,

weight and volume to the system and require power, calibration, wiring and

processing time. The use of existing flight data will help reduce the need for

additional sensor equipment and can provide important information on the actuators

operating conditions. Flight data can also be incorporated into dynamic landing gear

loading models, reducing the need for load measuring sensors. Due to the various

operating modes, often threshold levels can be breached even though no fault has

occurred. Flight data can be particularly useful in deducing if the observed symptoms

originate due to the actuator’s operating mode or through any impending failures.

Model-based health management of the landing gear actuators can offer robust early

fault detection and is based upon accurately modelling the dynamics of the actuator

system. The use of dynamic modelling techniques is an attractive approach to health

monitoring for the electric landing gear actuator system. For example, the actuators

do not require complicated modelling and can provide fault diagnosis relating to the

systems physical states or parameters. The ability to model failure progression further

enhances confidence in fault classification and has been shown to be able to

distinguish between gear and bearing failure modes for electromechanical actuators

used in a variety of other applications. They offer a non-intrusive approach, requiring



82

little addition sensor equipment incorporated into the landing gear system by acting

only upon command and response data obtained through the control systems. Data

driven methods however do not require system modelling, this creates faster algorithm

processing time than that which is required for models, the use of models can provide

better actuator fault diagnostics but at the cost of higher computational needs.

Rule-based systems encompass a set of rules which can be encoded similar to human

logic. This leads to transparency in any health assessment, a crucial factor in

decisions relating to aerospace maintenance. Case-based reasoning approaches are

often used when a rule base would be too large to construct or model-based

diagnostics are impractical. Case-based diagnostic reasoning is particularly useful at

capturing qualitative information which can be incorporated into the system and

decisions be made based on past case history (Frank 1990). Incorporating case based

reasoning into the health monitoring system, can be designed as fault trees. This

allows maintenance engineers to take an event which occurs at the platform level and

using extracted features from signals, knowledge of past events, and actuator usage

data, then a logical progression can be followed through the system and subsystem

levels to arrive at possible component faults (Raheja et al. 2006).

With the issues outlined earlier, in order to achieve the best possible level of reliability

in a monitoring system, health management can be achieved through extracting

information from a combination of sources and techniques. For example, data patterns

relating to known information can be provided through the combination of monitored

data, knowledge and models. Such approaches are gaining popularity in

electromechanical actuator health monitoring within the aerospace industry due to the

ability to achieve auditable and robust decisions (Keller et al. 2006; Watson and

Byington 2006). Figure (3.14) illustrates this monitoring concept.



83

Figure 3.14: Health Monitoring Concept Based on

Multiple Strategies

3.13 Motor-Driven Actuator Health Monitoring Review

3.13.1 Overview

The electric actuators which are the focus of this research must be robust and reliable

to avoid potential risk of accidents. Aerospace lags behind in the use of electric

actuators whereas other industries such as automotive, rail and process industries have

used these drives for many years. Much of the applications in these industries are just

as cost and safety critical as they are in aerospace. In aerospace actuation systems are

used for primary and secondary flight controls; braking, cargo doors, pressure valves,

weapons systems and landing gear extension and retraction. Most of which are

critical for the successful and safe operation of the aircraft.

In the automotive industry electric motor driven actuation is now commonplace. Such

applications include electric windows, locks, aerials and seat/lamp/mirror adjustments.

Drive-by-wire introduces motor-actuated steering and the starting circuit is a heavy

motor-driven actuation system. The rail industry actuators are one such application of

electrical actuation is in railway points. These are mechanical devices that allow a

train to move from one set of tracks to another. Point mechanisms are track elements

which frequently fail causing delays and increased costs. Such failures have also been

MonitoredData

PhysicalModels

ExpertKnowledge

CBMSystem Decisions



84

known to cause fatal accidents. The principle components of these point mechanisms

must be reliable and are subject to extreme environmental changes, varying loads,

stresses and very large mechanical forces. Modern power/process plants and motor

driven process control valves which control fluid-circulating systems are a regular

feature. These control valves are essential for plant operation and safety. A loss of

these systems can have catastrophic consequences. This section will review some

key health monitoring systems and techniques which have been developed specifically

for the monitoring of electrical actuation in the aerospace, automotive, rail and power

industry.

3.13.2 Aerospace

The use of model-based fault detection schemes based upon the use of parameter

estimation for use in both on-line and off-line diagnosing of actuator faults has been

demonstrated as an intuitive and non-intrusive approach (Moseler and Isermann 1998,

2000). This was further developed as a monitoring application for an air pressure

opening and closing valve in a passenger aircraft operated by a DC motor driven

actuator (Juricic et al. 2001). The model-based approach here was based upon the use

of parity equations supported by an approximate reasoning technique known as the

Transfer Belief Model (TBM) (Smets and Kennes 1994) which helped achieve high

diagnostic resolution, stability and accuracy over more traditional Boolean logic

methods. The use of a model–based fault detection scheme to support the reliability

of electromechanical replacements for civil aircraft engine actuators was also used by

Dixon and Pike (2002). It was shown that by measuring key parameters and testing

them against pre-determined baselines, the use of a fuzzy logic change detector and

fault classifier yielded promising results on determining significant parameter

changes.

Byington et al. (2004a) as part of the REACTS project developed a health

management methodology for the health monitoring of electrical flight control

actuators. They showed that model-based methods for prognosis and health

management offered a means for robust health monitoring and early fault detection.



85

Byington et al.( 2004b) furthered this health monitoring strategy for application for the

health management of the F/A-18 stabilator electro-hydraulic servo valves. The

development of this methodology took into consideration sensor and processing

limitations for onboard implementation. It made use of processing the

command/response signals and hydraulic pressure data along with neural networks,

automated reasoning, classification and advanced knowledge fusion to provide a real-

time assessment of the current and future actuator health.

The Kennedy Space Centre concerned with the health of mission critical

electromechanical actuator systems developed a valve health monitoring system

(Perotti et al. 2006). This system based upon a smart current signature sensor that

monitors the valves in a nonintrusive way. In the development of this the effects of

external features on the algorithm were analysed. It was highlighted that fluctuating

temperatures had an effect on all of the extracted features and there are further

challenges in regard to parameter nonlinearity.

3.13.3 Automotive

The health monitoring of a diesel engine coolant mechanisms was achieved by

Twiddle and Jones (2002) through the use a few low cost sensors. Residuals were

generated by system models and extracted features classified through fuzzy rules as

part of an overall engine health management system. In a car the air mass flow

through the intake manifold into the cylinders is controlled by a DC motor driven

throttle valve actuator. These must operate robustly and reliably because any

malfunction can cause unintentional stopping of the engine. The application of a

neuro-fuzzy hybrid system for the diagnosis of technical faults in such a throttle valve

actuator has been addressed by Pfeufer et al. (1997). This used a knowledge base in

the form of rules making the diagnosis understandable to human operators. It was

found that the use of a neuro-fuzzy system could significantly enhance the mapping of

observed symptoms to the underlying faults through the use of a parameterised logical

operator.



86

It is quite possible in a system that multiple failures in both the actuators and sensors

may occur simultaneously further complicating any fault diagnosis and isolation. Hsu

et al. (1995) uses a hexadecimal decision table to relate all possible identified failure

patterns to the residual code. This was successfully applied both in simulations and

experimental work. However the performance of the diagnostic system was degraded

with the presence of noise and system uncertainty.

3.13.4 Rail

Since various major rail accidents in Europe over the past few decades a provision of a

reliable infrastructure has become paramount in achieving good levels of safety. The

rail industry has always searched for new means to improve the performance of

subsystems to ensure both safety and reliability of services. One such subsystem is

railway points. Point mechanisms are track elements which frequently fail causing

delays and increased costs. Such failures have also been known to cause fatal

accidents. Roberts et al. (2002) considered a case study of an electro-pneumatic

railway point machine and proposed a fault detection and isolation process across an

interconnected geographical area. This used field bus data communication networks

allowing fault detection and isolation to be achieved on a number of close proximity

assets. Fault detection was based upon abstract static models. Fault isolation was

achieved through the use of a neuro-fuzzy system.

Oyebande and Renfrew (2002) used a net energy analysis technique relying upon

measured armature current and voltage waveforms and developed a system for the

health monitoring of point mechanisms which uses a large range of measured data

including distances, driving force, current and voltage, electrical noise, temperature

and state changes. The use of Finite Impulse Response (FIR) systems based upon a

normH 2 criterion was used by Zattoni (2006) as a means for guaranteeing

robustness with respect to disturbance inputs. This proposed algorithm could be used



87

for successful incipient fault detection for systems where the actuators operation could

with sufficient accuracy be described by linear time invariant models.

Wear is one of the most important factors affecting rail point mechanisms. Garcia

Marquez et al. (2003) put forward a model for the remote health monitoring of rail

mechanisms and highlighted the need for models to adapt to dynamic external

environments. This was applied to a predictive maintenance system known as RCM²

(Marquez et al. 2003). Further algorithms were put forward including the use of

Kalman filtering (Marquez et al. 2007) and the development of unobserved

component models approach for detecting wear and the behaviour of a worn set of

points based. The model was used to search for significant correlations between a

reference signal and new information coming in from critical components updating

model parameters on a continuous basis.

3.13.5 Power Industry

A common component found in modern power plants and process plants such as in

the petrochemical or nuclear industry are control valves which control fluid-

circulating systems. These valves are often driven by pneumatic actuators but are

being replaced with electromechanical drives. These control valves are essential for

plant operation and safety. A loss of these systems can have catastrophic

consequences. It has been shown that the use of motor current signature analysis can

be a selective and early indicator of developing mechanical and electrical

abnormalities in actuated valves and is becoming a commonly used technique. Non-

intrusive methods based upon motor current signature analysis have been proposed by

as part of an expert system for use in preventative maintenance in nuclear power

stations (Mukhopadhyay and Chaudhuri 1995). On-line health monitoring and using

electrical measurements to estimate the electric torque of the induction motor attached

to the MOV was outlined in Chai et al. (2004). This was reviewed against current

signature and power signature methods and was shown to have advantages including

better resolution, accuracy and sensitivity to load changes which are very useful in the



88

early detection of faults. A model free on-line fault detection technique based upon a

spectral analysis technique known as the Squared Coherency Function designed to be

sensitive to disturbances in plant dynamics was used by Previdi and Parisini (2006) to

provide effective actuator fault detection.

3.14 Conclusion

There are typically three kinds of maintenance regimes, time-based preventative, run-

to-failure and condition based maintenance. The potential benefits of moving towards

condition based maintenance is based around the concept of optimisation of

maintenance scheduling. These include reducing the lifetime costs of a machine or

system, minimising the effects upon the environment and increasing safety to

personnel. The design of a condition based maintenance system is application specific

and a strategy should be adopted that incorporates condition based maintenance into

systems rather than systems into condition based maintenance. This strategy should

be systems based and follow a logical progression from design and trade studies,

where the objectives and requirements are used to select appropriate equipment and

monitoring strategies, through to experimental testing, algorithm refinement and

implementation.

The uses of sensors are fundamental to a condition based maintenance system and

must be selected carefully to ensure the system meets the specific performance

requirements. The number of sensors introduced into a system requires consideration

to be given to aspects such as costs and complexity. Sensor networks can provide a

reduction in measurement time, and therefore reduce costs. Increased numbers of

sensors however require more computational effort and add complexity to a system.

System complexity can have adverse effects upon the reliability of the system, the

higher the number of components, the higher the number and frequency of possible

failures. These factors require a trade-off between sensor numbers and system level

requirements.



89

There exists a large tool box of health monitoring techniques, the most popular of

which are vibration analysis, wear debris analysis, motor current analysis and

performance analysis. Each of these techniques offers particular advantages and

disadvantages depending on the application. The most popular strategies for decision

reasoning are confined to the domain of expert systems. These are classified as data

driven or knowledge driven. A fusion of these approaches however can limit

uncertainty and maximise their decision making abilities.

Health monitoring of electrical actuation systems is not a new topic but nor is it a fully

advanced subject. As electrical actuation sees more and more introductions in the

aerospace, automotive, rail and process and power industries in safety critical

applications, numerous combinations of mature health monitoring techniques are

being applied to electrical actuation. In the aerospace industry there appears to be a

strong reliance upon transparent dynamical models. The automotive industry where

decision justifications are not so necessary utilise heavily neural networks. The rail

and process industry where limitations on sensors are not as paramount as in

aerospace rely heavily on signal processing methods.

3.15 References

Abidin, M.S.Z., Yusof, R., Kahlid, M., Amin, S. (2002), 'Application of a model-

based fault detection and diagnosis using parameter estimation and fuzzy inference to

a DC-servomotor', Proceedings of the 2002 IEEE International Symposium on

Intelligent Control, pp.783-787

Benbouzid, M.E.H. (2000), 'Review of induction motors signature analysis as a

medium for faults detection', IEEE Transactions on Industrial Electronics, Vol 47, No

5, pp. 984-993.

Bovio, I., Lecce, L. (2006), 'Health Monitoring of aeronautical structures based on

vibrations measurements', International Society for Optical Engineering,



90

Busch-Vishniac, I.J. (1999), 'Electromechanical sensors and actuators'

Byington, C., Safa-Bakhsh, R., Watson, M., Kalgren, P. (2003),'Metrics evaluation

and tool development for health and usage Monitoring technology'. in AHS Forum.

Phoenix Arizona.

Byington, C.S. (2006), 'Recent case studies in bearing fault detection and prognosis',

IEEE Aerospace Conference.

Byington, C.S., Watson, M., Edwards, D., Stoelting, P. (2004), 'A model-based

approach to prognostics and health management for flight control actuators', Big Sky,

MT, United States: Institute of Electrical and Electronics Engineers Computer

Society, Piscataway, NJ 08855-1331, United States.

Byington, C.S., Watson, M., Edwards, D., (2004), 'Data-driven neural network

methodology to remaining life predictions for aircraft actuator components', in IEEE

Aerospace Conference Proceedings.

Chai, J., Kang, S., Park, S., Hong, S., Lim, C. (2004), 'Condition monitoring of a

motor-operated valve using estimated motor torque', JSME International Journal,

Series C: Mechanical Systems, Machine Elements and Manufacturing, Vol 47, No 1,

pp.305-313.

Crowther, A., Zhang, N. (2003), 'Powertrain Vibration Analysis Using Torsional

Finite Elements', in The 10th Asia-Pacific vibration Conference, Gold Coast,

Australia.

De Silva, C.W. (1989), 'Control sensors and actuators'



91

Dixon, R., Pike, A.W. (2002), 'Application of condition monitoring to an

electromechanical actuator: A parameter estimation based approach', Computing and

Control Engineering Journal, Vol 13, No 3, pp71-81.

El Faouzi, N. (2004), 'Data fusion in road traffic engineering: an overview', SPIE-Int.

Soc. Opt. Eng.

Engel, S.J., et al. (2000), 'Prognostics, the real issues involved with predicting life

remaining', 2000. Big Sky, MT.

Eren, L., Devaney, M., J. (2002), 'Bearing damage detection via wavelet packet'

Anchorage, AK, USA: IEEE.

Esteban, J., et al., (2005), 'A review of data fusion models and architectures: towards

engineering guidelines', Neural Computing Applications, Vol 14, No 4, pp. 273-81.

Fraden, J., (1993), 'AIP handbook of modern sensors: physics, designs and

applications',

Frank, P.M., (1990), 'Fault diagnosis in dynamic systems using analytical and

knowledge-based redundancy: A survey and some new results', Automatica, Vol 26,

No 3, pp. 459-474.

Garcia Marquez, Schmid, F., Collado, J. (2003), 'Wear assessment employing remote

condition monitoring: A case study', Wear, Vol 255, No 7-12, pp. 1209-1220.

Garcia Marquez, F.P., Schmid, F., Collado, J., (2003), 'A reliability centred approach

to remote condition monitoring. A railway points case study', Reliability Engineering

and System Safety, Vol 80, No 1, pp. 33-40.



92

Garcia Marquez, F.P., Schmid, F., (2007), 'A digital filter-based approach to the

remote condition monitoring of railway turnouts', Reliability Engineering and System

Safety, Vol 92, No 6, pp. 830-840.

Garcia Marquez, F.P., Pedregal Tercero, D.J. Schmid, F., (2007), 'Unobserved

Component models applied to the assessment of wear in railway points: A case study',

European Journal of Operational Research, Vol 176, No 3, pp. 1703-1712.

Hammouri, H., Kinnaert, M., El Yaagoubi, E.H. (1999), ' Observer-based approach to

fault detection and isolation for nonlinear systems', IEEE Transactions on Automatic

Control, Vol 44, No 10, pp. 1879-84.

Harris-Jones, C., (1995), 'Knowledge based systems methods', The Practitioner

Series, Prentice Hall.

Hood, A., Pines, D. (2001), 'Feasibility of a linear phased acoustic array for health

monitoring of gears', Society of Photo-Optical Instrumentation Engineers, Newport

Beach, CA:

Hopgood, A. (1993), 'Knowledge-based systems for engineers and scientists', CRC

Press, Inc.

Hountalas, D. (1999), 'Development and Application of a Fully Automatic

Troubleshooting method for Large Marine Diesel Engines', Applied Thermal

Engineering, Vol 19, pp. 299-324.

Hsu, P.-L., Lin, K.-L. Shen, L.-C. (1995), 'Diagnosis of multiple sensor and actuator

failures in automotive engines', IEEE Transactions on Vehicular Technology, Vol 44,

No 4, pp. 779-789.



93

Ignizio, J. (1991), Iintroduction to expert systems: The development and

implementation of rule-based expert systems', McGraw-Hill, Inc.

Isermann, R. (2005), 'Model-based fault-detection and diagnosis - Status and

applications', Annual Reviews in Control, Vol 29, No 1, pp. 71-85.

Iyengar, S., Sastry, S., Balakrishnan, N. (2003), 'Foundations of data fusion for

automation', IEE Instrumentation & Measurement Magazine, Vol 6, No4, pp.35-41

Jardine, A.K.S., Lin, D., Banjevic, D. (2006), 'A review on machinery diagnostics and

prognostics implementing condition-based maintenance', Mechanical Systems and

Signal Processing, Vol 20, No 7, p. 1483-1510.

Juricic, D., Moseler, O., Rakar, A. (2001), 'Model-based condition monitoring of an

actuator system driven by a brushless DC motor', Control Engineering Practice, Vol 9,

No 5, pp. 545-554.

Keller, K., (2006), 'Aircraft electrical power systems prognostics and health

management' in IEEE Aerospace Conference,

Kacprzynski, G., Roemer, M., Hess, A., (2002), 'Health Management System Design:

Development, Simulation and Cost/Benefit Optimization', IEE

Kammer, D., (1991), 'Sensor Placement for on-orbit modal identification and

correlation of large space structures', Journal of Guidance, Control and Dynamics,

Vol 14, pp. 251-259.

Kanoun, O., trankler, H., (2004), 'Sensor technology advances and future trends',

IEEE Transactions on Instrumentation and Measurement, Vol 53, No 6, pp. 1497-

1501.



94

Khan, M.A., Starr, A., Cooper, D., (2008), 'A methodology for online wear debris

morphology and composition analysis', Proceedings of the Institution of Mechanical

Engineers, Part J: Journal of Engineering Tribology, Vol 222, No 7, pp. 785-796.

Khan, M.A., Starr, A., (2006), 'Wear debris: Basic features and machine diagnostics',

Insight: Non-Destructive Testing and Condition Monitoring, Vol 48, No 8, pp. 470-

476.

Kliman, G., Stein, J., (1992), 'Methods of motor current signature analysis', Electric

Machines and Power Systems, Vol 20, No 5, pp. 463-474.

Koeppen-Seliger, B., Frank. P., (1995), 'Fault detection and isolation in technical

processes with neural networks', New Orleans, LA, USA: IEEE, Piscataway, NJ,

USA.

Korbicz, J., Janczak, A., (2002), 'Artificial neural network models for fault detection

and isolation of industrial processes', Computer Assisted Mechanics and Engineering

Sciences, Vol 9, No 1, pp. 55-69.

Kothamasu, R., Huang, S., Verduin, W. (2006), 'System health monitoring and

prognostics - A review of current paradigms and practices', International Journal of

Advanced Manufacturing Technology, Vol 28, No 9, pp. 1012-1024.

Linas, J., Hall, D. (1998), 'An introduction to multi-sensor data fusion', Monterey, CA,

USA: IEEE.

Luo, R., Lin, M., and Scherp, R. (1988), 'Dynamic multi-sensor data fusion system for

intelligent robots', IEEE Journal of Robotics and Automation, Vol 4. No4, pp. 386-

396



95

Montalvao, D., Maia, N., Ribeiro, A., (2006), 'A review of vibration-based structural

health monitoring with special emphasis on composite materials', Shock and Vibration

Digest, Vol 38, No 4, pp. 295-324.

Moseler, O., Isermann, R. (2000), 'Application of model-based fault detection to a

brushless DC motor', IEEE Transactions on Industrial Electronics, Vol 47, No 5, pp.

1015-1020.

Moseler, O., Isermann, R. (1998), 'Model-based fault detection for brushless dc motor

using parameter estimation', Aachen, Ger: IEEE Comp Soc, Los Alamitos, CA, USA.

Mukhopadhyay, S., Chaudhuri, S. (1995), 'A feature-based approach to monitor

motor-operated valves used in nuclear power plants', IEEE Transactions on Nuclear

Science, Vol 42, No 6, pp. 2209-2220.

Nandi, S., Toliyat, H., Xiaodong, L. (2005), 'Condition monitoring and fault

diagnosis of electrical motors-a review', IEEE Transactions on Energy Conversion,

Vol 20, No 4, pp. 719-29.

Orsagh, R., Roemer, M. (2000), 'Development of metrics for mechanical diagnostic

technique qualification and validation', in COMADEM. Houston, TX.

Oyebande, B., Renfrew, A., (2002), 'Condition monitoring of railway electric point

machines', IEE Proceedings: Electric Power Applications, Vol 149, No 6, pp. 465-

473.

Padula, S., Kincaid, R., (1999), 'Optimization strategies for sensor and actuator

placement', NASA Report.



96

Pantelelis, N., Kanarachos, A., Gotzias, N., (2001), 'Neural networks and simple

models for the fault diagnostics of naval turbochargers', Mathematics and Computers

in Simulation, Vol 51, pp. 387-397.

Patan, K., Parisini, T. (2005), 'Identification of neural dynamic models for fault

detection and isolation: The case of a real sugar evaporation process', Journal of

Process Control, Vol 15, No 1, pp. 67-79.

Pattichis, C., Pattichis, M., Micheli-Tzanakou, E. (2001), 'Medical imaging fusion

applications: An overview', Institute of Electrical and Electronics Engineers

Computer Society, Pacific Grove, CA:

Patton, R., Chen, J., (1997), 'Observer-based fault detection and isolation: robustness

and applications', Control Engineering Practice, Vol 5, No 5, pp. 671-82.

Patton, R., Chen, J., 'Review of parity space approaches to fault diagnosis', Baden-

Baden, Ger: Publ by Pergamon Press Inc, Tarrytown, NY, USA.

Pedregal, D., Carnero, M. (2006), 'State space models for condition monitoring: A

case study', Reliability Engineering and System Safety, Vol 91, No 2, pp. 171-180.

Perotti, J., Lucena, A., Burns, B. (2006), 'Smart actuators - Valve Health Monitor

(VHM) system', Kissimmee, FL, United States: International Society for Optical

Engineering, Bellingham WA, WA 98227-0010, United States.

Pfeufer, T., Ayoubi, M. (1997), 'Application of a hybrid neuro-fuzzy system to the

fault diagnosis of an automotive electromechanical actuator', Fuzzy Sets and Systems,

Vol 89, No 3, pp. 351-360.

Phillips, P., Diston, D., Payne, J., Pandya, S., Starr, A. (2008), 'The application of

condition monitoring methodologies for certification of reliability in electric landing



97

gear actuators', in The 5th International Conference on Condition Monitoring and

Machine Failure Technologies. Edinburgh, UK

Pietruszkiewicz, R., Starr, A., Albarbar, A. Mattheys, R. (2006), 'Remote sensing in

condition monitoring' in 1st World Congress on Engineering Asset Management, Gold

Coast, Australia.

Poyhonen, S., Jover, P., Hyotyniemi. H., (2004), 'Signal processing of vibrations for

condition monitoring of an induction motor', Institute of Electrical and Electronics

Engineers Inc., New York, United States.

Previdi, F., Parisini, T. (2006), 'Model-free actuator fault detection using a spectral

estimation approach: The case of the DAMADICS benchmark problem', Control

Engineering Practice, Vol 14. No 6, pp. 635-644.

Raheja, D., et al., (2006), 'Data fusion/data mining-based architecture for condition-

based maintenance', International Journal of Production Research, Vol 44, No 14, pp.

2869-2887.

Rajagopalan, S., et al. (2004), 'Diagnosis of potential rotor faults in brushless DC

machines', Institution of Electrical Engineers, United Kingdom.

Rajagopalan, S., et al. (2006), 'Current/Voltage-based detection of faults in gears

coupled to electric motors', IEEE Transactions on Industry Applications, Vol 42, No

6, pp. 1412-1420.

Rao, B. (1996), 'The handbook of condition monitoring' 1st ed. Elsevier Advanced

Science Ltd.



98

Roberts, C., et al. (2002), 'Distributed quantitative and qualitative fault diagnosis:

Railway junction case study', Control Engineering Practice, Vol 10, No 4, pp. 419-

429.

Zhou, F.B., et al. (2002), 'Remote condition monitoring for railway point machine',

Institute of Electrical and Electronics Engineers Inc. Washington, DC:

Roemer, M., Kacprzynski, G. (2000), 'Advanced diagnostics and prognostics for gas

turbine engine risk assessment', in IGTI/ASME Turbo Expo, Munich, Germany.

Saniger, J., et al. (2001), 'Structural health monitoring methodology for aircraft

condition based maintenance', Society of Photo-Optical Instrumentation Engineers.

Schoen, R., Habetler, T. (1995), 'Effects of time-varying loads on rotor fault detection

in induction machines', IEEE Transactions on Industry Applications, Vol 31, No 4,

pp. 900-906.

Sen, A., Murty A., (1999), 'Parallel processing application in traction motor fault

diagnosis', Electrical power Systems Research, Vol 52, pp. 241-249.

Smets, P., Kennes, R. (1994), 'Transferable belief mode', Artificial Intelligence, Vol

66, No 2, pp. 191-234.

Starr, A., Willets, R., Hannah, P., Banjevic, D., Jardine, A. (2002a), 'Data fusion

applications in intelligent condition monitoring', Recent Advances in Computers,

Computing and Communications, pp. 110-115.

Starr, A., Hannah, P., Esteban, J., Willets, R. (2002b), 'Data fusion as a model for

advanced condition-based maintenance', Insight-Non-Destructive Testing and

Condition Monitoring, Vol 44, No 8, pp. 503-507.



99

Starr, A. (2005), 'Merging condition monitoring data and knowledge: fusion and

neural approaches', in Maintenance, Condition Monitoring and Diagnostics, POHTO,

Oulu, Finland, POHTO.

Starr, A., Albarbar, A., Pietruszkiewicz, R., Mekid, S., (2007), 'Developments in

wireless sensing for condition monitoring', Condition Monitor,

Steinberg, A., Bowman, C., White, F. (1999), 'Revisions to the JDL data fusion

model', Proc. SPIE Algorithms: Issues and Approaches II, Vol 37

Swearingen, K., et al. (2007), 'An open system architecture for condition based

maintenance overview', Institute of Electrical and Electronics Engineers Computer

Society, Piscataway, NJ 08855-1331, United States.

Thomson, W., Fenger, M. (2001), 'Current signature analysis to detect induction

motor faults', IEEE Industry Applications Magazine, Vol 7, No 4, pp. 26-34.

Twiddle, J., Jones, N. (2002), 'Fuzzy model-based condition monitoring and fault

diagnosis of a diesel engine cooling system', Proceedings of the Institution of

Mechanical Engineers. Part I: Journal of Systems and Control Engineering, Vol 216,

No 3, pp. 215-224.

Vachtsevanos, G., Lewis, F., Roemer, M., Hess, A., WU, B. (2006), 'Intelligent Fault

Diagnosis and Prognosis for Engineering Systems' John Wiley & Sons , INC.

Watson, M., Byington, C. (2006), 'Improving the maintenance process and enabling

prognostics for control actuators using CAHM software', Institute of Electrical and

Electronics Engineers Computer Society, Piscataway, United States.



100

Wu, Y. (2009), 'Misalignment diagnosis of rotating machinery through vibration

analysis via the hybrid EEMD and EMD approach', Smart Materials and Structures,

Vol 18, No 9.

Zattoni, E. (2006), 'Detection of incipient failures by using an H2- norm criterion:

Application to railway switching points', Control Engineering Practice, Vol 14, No 8,

pp. 885-895.

Zhongming, Y., Bin, W., Sadeghian, A. (2003), 'Current signature analysis of

induction motor mechanical faults by wavelet packet decomposition', IEEE

Transactions on Industrial Electronics, Vol 50, No 6, pp. 1217-28.



101

C h a p t e r 4 : U n d e r s t a n d i n g t h e C o m m e r c i a lB e n e f i t s o f A e r o s p a c e H e a l t h M o n i t o r i n g

4.1 Introduction

There are a number of questions that must be addressed before a realistic commercial

strategy for aerospace actuator health monitoring can be proposed. These include

providing definitions on the customer, the supply chain networks and any supporting

technologies. But most importantly, what the commercial drivers are? It is important,

that there is a clearly identified need for the actuator health monitoring. If there is no

existing need then any attempt for marketing health monitoring as a necessary future

technological requirement is doomed for failure.

In this chapter the commercial need for actuator health monitoring is explored in the

context of the role of health monitoring in the changing aerospace maintenance

industry. Focuses on the reasons behind changes to the maintenance market and the

drive towards innovative maintenance support concepts is discussed at length. This

discussion includes the key benefits and value potential of predictive maintenance and

through life support of aerospace products. Challenges to integrating new monitoring

technologies into existing products are identified as extending to not only technical

challenges but also challenges to commercial integration. A number of pricing

strategies are explored in the context of commercial integration, and finally a SWOT

analysis of the technology is provided.

4.2 Current Aerospace Maintenance Practice

The airline industry is considered as one of the most unique businesses in the world

which suffers from a variety of complex operations. These include moving aircraft

loaded with passengers and cargo over large distances and the scheduling of flights,

crews and maintenance. These all lead up to substantial costs measured in time and

money. Aircraft maintenance forms an essential part of an aircrafts airworthiness

criteria, with its main objective being to ensure a fully serviced, operational and safe



102

aircraft. If an aircraft is not maintained to the required level then this inevitably risks

passenger and crew safety. Table 4.1 lists examples of incidents that have occurred

due to insufficient maintenance (Gramopadyhe and Drury 2000). There is a

substantial risk, if maintenance is not correctly performed, that the aircraft may be

unable to take-off leading to passenger dissatisfaction. Likewise it is plausible that the

aircraft may be forced to land in undesirable locations, where spare parts or

maintenance expertise is unavailable. Maintenance actions therefore have to be

carried out at regular scheduled intervals, but ideally be performed with minimum cost

to the operator, whilst maximising revenue to the maintenance providers.

Table 4.1: Aircraft Maintenance Related Accidents

Airline Location Year Incident

Aloha Airlines 737 Hawaii 1988 Inspection failure led to fuselage

failure

BM AirTours 737 Manchester 1989 Wrong bolts led to windshieldblowout

United AirlinesDC10

Iowa 1989 Engine inspection failure led to loss ofsystems

ContinentalExpress

Texas 1991 Tail failure as task not completedbefore flight

Northwest Airlines Tokyo 1994 Incomplete assembly led to engineseparation

ValueJet Florida 1996 Fire in hold due to incendiary cargo

Maintenance programmes for key systems such as the engines and landing gears are

made up of several activities based around preventive, corrective, on-condition and

redesign maintenance. Preventive actions are taken at pre-determined intervals based

upon the number of operating hours, or often in the case of landing gears, the number

of landings. This is supported by regularly scheduled inspections and tests in which

on-condition maintenance is performed based upon observations and test results.

Each of these activities is finally supported by corrective maintenance conducted in

response to discrepancies or failures within the aircraft during service. The final

action type, redesign maintenance takes the form of engineering modifications that are



103

made in order to address arising safety or reliability issues, which were unanticipated

in the original design.

Much of the major maintenance and repair work performed on aircraft is provided

through service providers who carry out Maintenance, Repair and Overhaul (MRO)

operations for the aircraft operators. The landing gear is a critical assembly and a

major key to maintaining the overall aircraft value. Operators cannot afford, or are

willing to risk compromising their landing gear MRO activities and will look for the

best combination of affordability, expertise, flexibility and the ability to offer the best

solutions when faced with the choice of MRO provider.

An example of how maintenance support of landing gears would be as follows. In the

event of a series of incidents such as ‘hard landings’ reported by the operators, major

repair operations, or complete gear overhauls will be conducted at a MRO provider’s

maintenance site. The operators themselves can carry out, minor repairs and on-wing

maintenance, also at pre-determined intervals. Once the aircraft has been received at

the MRO maintenance facility, the landing gears will be dismantled and individual

parts will be put through a serious of non-destructive tests. This testing will identify

any developing failures, such as structural fatigues or internal corrosion. The results

of which will determine if the parts are repaired, replaced, scrapped or recycled

(Patkai et. al 2007). Landing gears are complex systems with a vast number of parts o

which need to be maintained and inspected, which results in costly maintenance

operations, in terms of time. An example of key inspection areas along with typical

timescales would be:

1. After 300 hours or after 1 year in service inspection

- Shock absorber Nitrogen Pressure check

2. After 600 hour inspections

- Landing gear hinge points visual inspections

- Leak inspection (oil, hydraulic fluid etc)



104

- Inspection of torque link play

3. After 7 years or 5000 cycles : Landing gear overhaul

To understand maintenance costs it is necessary to look at the elements of

maintenance in terms of time. Figure (4.1) gives a breakdown of the time elements

covering typical maintenance actions. A breakdown such as this can show designers

the areas in which they can influence related activity times. In corrective maintenance

much of the time is spent on locating a defect which often requires a sequence of

disassembly and reassembly. Being able to predict fault location times is extremely

difficult using traditional inspection techniques. The ability to automate this fault

diagnosis, with advanced technologies and techniques, can help accurately predict the

downtime required (Knotts 1999).

The market for landing gear overhauls, new gear-sets, exchange gears and spare parts

has become somewhat chaotic, with some new gear lead times running at up to four

years and overhaul queues lengthening by the day. This has been somewhat

unexpected by Original Equipment Manufacturers (OEM) and overhaul providers

even though landing gear maintenance intervals are widely known and plans put into

place.



105

Figure 4.1: Civil Aircraft Maintenance Time

Relationships (Knotts 1999)

There is a growing feeling that current business models will eventually no longer

cope and the following coinciding factors are forcing changes in maintenance

strategy.

The large number of aircraft sales between 1998 and 2000 which, given the

typical 18,000-cycle or eight to ten years time between overhauls, has

created an unprecedented demand for landing gear overhauls on both long-

and short-haul aircraft.

Time

Up Time Down Time

Flying Time Available forFlying Time

Flight Prep.Time

MaintenanceTime

ModificationTime

Pre-flightinspection

Time

Turn AroundTime

PreventiveMaintenance

CorrectiveMaintenance

Access Time Inspection Time PreparationTime

Defect LocationTime

DefectRectification

Time

BITE effectivenessFault diagnostic aids

Equipment test/Read out capabilityTechnician skill, experience & training

Rectify by adjustment timeIn-situ repair time

Remove repair & refit timeRemove and replace time



106

Record aircraft production following unprecedented sales in 2006 and 2007.

Higher utilization of short-haul aircraft, which has shortened time between

overhauls for some airlines by one or more years.

The growing number of aircraft in service, including some older types

returned from desert storage, bringing more and more landing gear into the

market. This also has increased the number leakage or heavy corrosion

findings in line maintenance which also drives early overhauls.

A lack of landing gear overhauls capacity created by the high cost of setting

up an overhaul facility. The number of service providers has not grown

enough to meet the increasing demand.

Seasonal cycles have overloaded winter overhaul slots, leaving some

summer slots unfilled.

A worldwide shortage of raw materials like rubber, high strength steel and

titanium, exacerbated by increasing demand from the burgeoning economies

of China and India.

The on-going conflicts in Iraq and Afghanistan, which make significant

demands on spare-part production and material supply.

Complacent operators who either ignored the "need-to-plan" warnings from

the OEMs or simply have left landing gear overhauls to the last minute

4.3 Changing Maintenance Practice

Currently the European market holds a 26% share of the worldwide MRO business

compared to 39% held by North America and is expected to experience further



107

dramatic worldwide growth during the next 10 years (Jenson 2008). There are

however several hurdles which must be overcome by these MRO providers in order to

continue their leading global market shares (Fitzsimons 2007). Examples of which

include:

Growing competition from the Middle East.

Greater competition from original equipment manufacturers.

Continuing pressure from airlines to reduce costs.

These hurdles coupled with increased demand for airline MRO are forcing changes in

the global aviation maintenance industries, including:

MRO providers are expanding their geographical reach and capabilities in a

bid to become regional and global full service providers.

Spending on MRO is expected to universally increase.

Airlines are now seeking how to make the next level of savings, which has

raise the demand for more predictive maintenance strategies, with more

reliability and material solutions to compliment outsourced maintenance repair

work.

To drive further cost reductions, airlines are seeking to incorporate

sophisticated maintenance management solutions into their aircraft, reducing

investments in inventory and to aid in improvements in airline operations and

reliability.

Such factors have begun to dictate a change in maintenance strategy for operators and

the solutions in the services that the MRO suppliers can provide. These will aid in



108

reducing the levels of scheduled maintenance and hence optimising maintenance on

aircraft fleets and recommendations and techniques on selecting the best maintenance

strategy are currently being developed (Labib 2010). In terms of landing gear, much

of the current business offered to the customers is contracted in the form of ‘time and

materials’, which can be an expensive option for operators. The changing face of the

aviation industry requires that maintenance management become increasingly tailored

towards individual customers needs with cost-effective solutions being found, offering

compromises between customer involvement and the level of commitment required

from the providers. Figure (4.2) shows a matrix with different maintenance solutions

and the level of commitment and partnerships required by the operators and MRO

providers (Phillips et al 2009).

Figure 4.2: Maintenance Support Concepts

4.4 Predictive Maintenance

The desire is such that in order to remain competitive and meet the demands and

challenges facing operators and suppliers new maintenance support concepts should

offer several gains. For the operators these should be reductions in unscheduled

maintenance activity, lower total cost of ownership, reductions in administrative

burdens and overall optimisation of maintenance activities. This can be achieved by

Through LifeSupport

High Medium Low

Aircraft Operator Involvement

MROSupport

PredictiveMaintenance

PreventiveMaintenance

CustomisedPayment Scheme

Time andMaterials

All InclusiveOverhaulsHigh

Medium

Low



109

moving away from the scheduled preventive maintenance actions by introducing new

systems that can provide details on the in-service operation and condition of landing

gear mechanisms, such as brakes, shock absorbers and actuators. Such systems

known as health monitoring systems (Kothamasu et. al 2006) utilise a variety of data

gained from on-board sensors in order to extract meaningful information. This

information when combined with expert knowledge such as component reliabilities,

failure mechanisms and service/maintenance history will provide a quantification of

system/subsystem/component health. Based upon this information future corrective

maintenance actions can be predicted and allow for the optimisation of aircraft

maintenance. Incorporating health monitoring systems into aircraft landing gears in

order to employ a predictive maintenance strategy (Mobeley 2002) in place of

preventive maintenance, offers benefits to both the operators, MRO providers and

landing gear manufacturers as described in Table 2.2.

Table 2.2: Benefits of a Predictive maintenance

Strategy

Operator MRO provider Landing gear manufacturer

Optimised maintenancescheduling

Reductions in maintenance costs

Reduced risk of in-servicefailures

Increased aircraft availability

Optimisation of spare partsstockpiling

Minimisation of scrap

Elimination of bottlenecks inmachine usage during MROoperations

Reduction in turnaround times

Information available from on-board health monitoring sensorscan be used as a marketing tool

Evaluation of in-serviceperformance of landing gearsystems

Extensive knowledge of in-service performance can beincorporated into re-designs.

Aids in increasing operatorconfidence in incorporating newreplacement technologies.

However it should be noted that innovative predictive maintenance solutions

supported by health monitoring can only provide each of the key players the necessary

benefits if the necessary commitments are made. A smooth flow of information is



110

required between the operators, maintenance providers and the manufacturers. It

could also be questionable if operators would really want to commit to a long term

innovative maintenance solution, due to the added commitment requirements on their

behalf. They may be hesitant to uptake the offer of health monitoring systems if the

manufacturers have not listened to the specific requirements for their aircraft, most

notably component reliability and minimal effects on weight and complexity. The

operators will also be wary of the need for the probable handling of vast quantities of

extra data and information generated from the health monitoring systems. Support

with this should therefore be offered within any innovative maintenance service, or

systems that can provide automatic health, related decisions are essential if health

monitoring is to be accepted. Operators must also be willing to follow a long-term

commitment as a support partner and be willing to exchange failure data with the

manufacturers in order for increased reliability in future designs. This flow of

information is illustrated in Figure (4.3)

Figure 4.3: The Process of Information Flow

4.5 Value potential of Predictive Maintenance

The value of incorporating health monitoring systems is most likely to arise in savings

in operating costs. The use of health monitoring systems for landing gear retraction



111

mechanisms, or other aircraft systems, will offer a very competitive advantage in

maintenance decision-making, which is crucial for both military and commercial

aerospace users. This will help manufacturers retain customers and attract new

business; these aspects will mean that health monitoring solutions will become a key

part of formulating future maintenance strategies. The airline industry has seen a

rapid increase in operators over the past decade, particularly in low cost short haul

operators. The nature of the budget airlines business succeeds in the ability to operate

large aircraft fleets, coupled with high aircraft availability and short turn-around times

whilst keeping ticket costs low. For such factors to remain and for airlines to create a

business winning advantage, then strategic maintenance management has to become

one of the significant factors in their operations management. The adoption of health

monitoring and overall predictive maintenance can help push an aircraft operators

business forward as illustrated in Figure (4.4).



112

Figure 4.4: Potential Effects of PredictiveMaintenance on an Aircraft Operators Business

4.6 Developing, Integrating and Pricing the Technology

4.6.1 Technical Challenges to Integrating Health Monitoring

Health monitoring is a disruptive technology – in that large-scale integration will

cause disruptive changes within well defined and established working practices. But

once established it can quickly go on to become a fully performance competitive

system. Health monitoring systems are aimed at improving the performance of the

aircraft, which will be achieved on the lines of ‘evolutionary’ changes whilst

demonstrating reliability, validated cost benefits and reduces operational risks. The

integration of new technologies inevitably face difficulties and a number of

challenges face the community of engineers and technical specialists as they seek to

Give an Operations andbusiness winningAdvantage

Link maintenance withoperations strategy

Adopt best practice

Correct the worst problems

STAGE 1 STAGE 2 STAGE 3 STAGE 4

Organisationheld back

Be as good ascompetitors

Be clearly thebest in theindustry

Redefineexpectations

The ability toimplement

strategy

The ability tosupportstrategy

The ability todrive strategy

Increasing contributionof predictivemaintenance



113

utilise health monitoring for aerospace usage, a non-exhaustive list of these

difficulties include:.

1. The technology and frameworks are available but under utilised.

2. Performance characteristics are usually untested, leading to a lack of confidence

3. There is often a wealth of data available from the end users, but access to this

data can be limited and much is yet to be converted to ‘meaningful information’

Health monitoring systems for aerospace applications differ from those for other

applications such as industrial machine monitoring or the monitoring of civil

structures due to hardware restrictions and the difficulties associated with certification.

Also, in many areas of aerospace health monitoring system development, often the

state-of-the-art monitoring technique being developed are restricted by a variety of

limitations. This affects their use in a real operational situation’, for example, many of

the sensor based methods under development for the monitoring of fuselage

structures, based upon such methods as acoustics or vibration patterns require vast

sensor arrays. Much of the information gained requires high levels of signal

processing with the results being very subjective and consequently they may not be

applicable for an on-line real time aerospace monitoring system, even though the

fundamentals of the techniques work well in other applications. This will potentially

lead to a case where the state of the art has difficulties in matching the necessary

requirements for aerospace integration. This the author believes is the reason for the

current slow integration of health monitoring on civil aircraft, despite the vast wealth

of academic research detailing monitoring methods, industry drive and potential areas

for application.

Figure (4.5) illustrates this hypothesis; (Phillips et al. 2009) it demonstrates how the

current health monitoring state-of-the-art trend is progressing with respect to the



114

capability requirements for health monitoring for aerospace usage. The hypothesis

indicates that the current state-of-the-art is advanced enough for most industry uses;

offering leaps in performance and capabilities. But is far below what is required for

aerospace applications, and will require further innovations, amongst others, in terms

of hardware minimisation, data reduction techniques and the use of fusion to merge

multiple techniques to reduce individual limitations and maximise advantages.

Figure 4.5: Aerospace Health MonitoringRequirements as Compared to the State-of-the-Art

4.6.2 Commercial Integration Challenges

Aerospace OEM will have well defined business models and practices. For example,

this will usually follow two integrated paths which can be regarded as the product

lifecycle, which begins with product innovation, design and development,

manufacture, production and finally through life support. The second path is the

business supply chain which begins with forecasts of landing gear sales, received

orders, scheduling procurement, production and finally distribution and after sales

support. For a OEM which has no history of supplying health monitoring systems the

integration into these processes, which are illustrated in Figure (4.6) will not be an

Time

HM system requirements for an‘enabling’ technology

HM system requirements foraerospace applications

Current HM state of the art trend

Capability

Desired HM state of the art trendfor aerospace applications



115

easy task. Integration of health monitoring for instance will directly affect how after

sales service and through life support is conducted. For successful commercial

integration it is very probable that a whole new structure to the suppliers’ business

model will be required.

Figure 4.6: Logistics Chain and Business Management

Some of the more general issues (Raheja et al. 2006) relate to the conceptualisation of

strategies for decision making and goal setting across multiple 'component or system'

levels and time periods. A unique business methodology for incorporating cost as a

factor to be considered in setting up these decision making strategies is a fundamental

requirement for the successful integration of health monitoring technologies. Other

considerations involve the linking of the maintenance system with Enterprise and

Materials Resource Planning (ERP/MRP) to enable optimisation of spare parts

ordering, and therefore ensuring the after sales and through life support elements of

the business remain at their optimal capabilities.

Salesforecasts

Orders Schedulingprocurement

Distributionand after sales

service

Productinnovation

Designand

development

Manufacture

Production

Throughlife

support

Demand forHM

Integration

Logistics chain /Businessmanagement

Product / ProcessLifecycle



116

4.6.3 Pricing Deployment Strategies

There are very few published works which detail proposed pricing models for health

monitoring technology. The reasons for this are generally due to the secretive nature

of organisations have over their business models, one example however with direct

relevance to this current work is provided by Kidd 2006. This Engineering Doctorate

thesis provides an extensive discussion on business and pricing models for health

monitoring for automotive vehicles as seen from the vehicle OEM perspective. The

nature of the automotive and aerospace commercial business operate on very different

principles, but the models in this work provide a framework in developing pricing

models for aerospace actuator health monitoring solutions.

A number of pricing models are therefore proposed in order to generate revenue for

the provider of the health monitoring system, which is aimed at adding significant

added value to their landing gear products. As has been identified there are several key

players in aircraft maintenance, which all must be included in any deployment/pricing

models, if health monitoring on landing gears are to be successful. It is logical that

any integration of health monitoring hardware as a standard addition to landing gears

are technically sold to the aircraft OEM directly. However, as any additional costs

met by the aircraft OEM, will almost certainly be met by the aircraft purchasers

(operators), and because it is the operator’s responsibility for maintaining their aircraft

the health monitoring customer is simplified to be the aircraft operators in this case.

1. The first pricing model generates revenue based upon sales volumes (variable

costs).A unit cost is paid by the landing gear customer to the health monitoring

system provider for every landing gear set using the monitoring technology.

2. A second pricing model is based upon per landing gear unit (fixed cost). A

one off payment is made to the health monitoring provider for any given

monitoring solution per landing gear set. This ensures that the customer

retains the ownership for the life of that product.



117

3. The third model is based upon the health monitoring technology being

incorporated into the landing gear as standard equipment (consolidated costs)

by the OEM. No direct costs are passed onto the customer, but indirect

revenue for the OEM can be generated through product differentiation.

It should be noted that in all the pricing models the health monitoring tools would be

sold under licence, with the health monitoring system providers maintaining control

over the background intellectual property.

4.7 SWOT Analysis: Actuator Health Monitoring Technology

The SWOT analysis is a useful tool for understanding and decision-making for a

multitude of situations in businesses and organisations. SWOT is an acronym for

Strengths, Weaknesses, Opportunities and Threats. This is illustrated in Figure (4.7).

These headings provide a good framework for reviewing strategy, position and

direction of a company or product idea. It is a subjective assessment of data which

when put into the SWOT format enables understanding, presentation, discussion and

decision making. In this chapter a SWOT analysis is presented for the actuator health

monitoring technology is presented.

Figure 4.7: Illustration of a SWOT Analysis

ActuatorHealth

Monitoring

THREATS

STRENGTHS

OPPORTUNITIES

WEAKNESSES



118

It should be noted that other market analysis tools exist. For example one could have

used a PEST analysis. PEST, an acronym for Political, Environmental, Social and

Technological is a useful tool for understanding market growth or decline. As a PEST

analysis most commonly measures an existing market, whilst a SWOT analysis is

used to measure a proposition or business idea, the SWOT analysis was chosen to be

used to assess commercial applicability of actuator health monitoring.

4.7.1 Strengths

Aerospace OEM, MRO providers and aircraft operators have expressed

interests in aerospace health monitoring technologies.

There are a number of advantages offered to OEM, MRO providers and

aircraft operators.

Combining information generated from the actuator health monitoring system

with other aerospace monitoring systems as part of a Integrated vehicle Health

Monitoring (IVHM) system will increase overall aircraft safety, reliability and

operational lifespan

Actuator health monitoring will aid in increasing customer confidence in new

replacement all electric actuator technology.

OEM can make extensive use of in-service performance data generated from

the monitoring system for re-design improvements.

Electromechanical actuators used in different applications all share the same

common component types and general operating procedure. This means that

an actuator monitoring system could be packaged and sold off-the-shelf. With

customers only required to perform simple tuning for their individual

application.



119

4.7.2 Weaknesses

Aerospace certification procedures for new hardware/software coupled with

the requirements for aircraft weight reduction may restrict the addition of a

health monitoring systems.

There may be a reluctance to accept the monitoring technology by operators as

a new decision-making tool until the system has proven itself as an in-service

reliable technology.

In general the incorporation of health monitoring would allow serviceable

components to remain in service for longer periods. This may result in OEM’s

and third party providers loosing revenue generated by periodic maintenance.

Once health monitoring systems are in place they must be reliable. Unreliable

monitoring will result in reductions in customer confidence. This would lead

to reluctance for future customers to invest in the technology.

4.7.3 Opportunities

Environmental factors have led to new governmental legislation demanding

cleaner and more fuel efficient aircraft. This has led designers and

manufacturers to begin incorporating innovative replacement technologies, for

example, the new all electric actuation. If these are accepted then a market for

health monitoring systems will open up to support their reliability and gain

customer confidence in replacement technologies.

Electromechanical actuation is not just confined to aerospace applications. The

rail, automotive, shipping and power processing industries all make use of

them. More often than not in mission or safety critical applications. This

offers the potential for a cross-market business.



120

Health monitoring offers the potential to improve current maintenance

operations. Allowing the provider to supply the customers with a range of

innovative maintenance packages. These could be tailored to individual

customer requirements, offering a competitive business winning advantage

over competitors.

4.7.4 Threats

Customers may see the introduction of health monitoring to enhance product

reliability as an admittance of inherent product unreliability.

There are currently several key European aerospace companies investigating

the potential use of electromechanical actuation for landing gears. It is highly

unlikely that these are oblivious to the potential advantages posed by health

monitoring. There is also a more advanced drive for similar actuators for other

aerospace applications such as control surfaces. There is a risk therefore that a

competitor could be the first to the market.

It is difficult to evaluate the cost of manufacture, implementation and upkeep

of a health monitoring system in development. It may be the case that aircraft

operators will not see health monitoring as an economically viable option they

may therefore seek other maintenance solutions.

Aerospace certification procedures may lead to the technologies which health

monitoring is aimed at supporting not being accepted for incorporation into

aircraft. This would make the designed monitoring system instantly

redundant.

Landing gear actuator health monitoring is just one monitoring system that is

likely to see introduction onto aircraft in the future. For all of these systems to



121

be optimally effective it would be desirable for them to work together.

However this will be unlikely the case due to each supplier using incompatible

hardware and software. This would make the case for a fully IVHM system

unlikely, with operators choosing to optimise their selection of monitoring

systems, leaving some key items on the aircraft as unmonitored. Landing gear

actuation may be one of these items.

4.8 Conclusion

Health monitoring technology is intractably tied up with aerospace maintenance

activities as a whole. The aerospace maintenance industry is currently facing a time of

unprecedented demand for spare parts, complete overhauls and general servicing.

This is due to, amongst other reasons, a sudden increase in aircraft numbers in the last

decade or so which now have key systems such as landing gears reaching the end of

their life. This is therefore putting a strain on overhaul providers and Original

Equipment Manufacturers (OEM). This has begun to force operators, OEM and

overhaul providers to begin seeking new innovative maintenance solutions, to meet

rising demands and costs. The nature of this therefore opens up the possibility of

integrating health monitoring technologies into OEM business models.

It is envisioned as part of this thesis work that health monitoring technology will play

a crucial role in revolutionising aircraft maintenance practice. This it will be proposed

will not come about entirely as a direct result of the implementation of the technology.

But rather by a set of unique customised solutions and support packages offered as a

result of mature health monitoring technology. It has long been the tradition that

operators are secretive when it comes to information regarding the maintenance of

their fleets, information that suppliers often have to purchase in order to optimise their

spare parts inventories. It is also proposed that for customised maintenance support

packages to be successful, there must be a beneficial trade off for all players involved,

and this will require changes to the way the key players share information.



122

4.9 References

Fitzsimons., B. (2007), 'The BIG picture: Airline MRO in a global context'. Airline

Fleet & Network Management, Vol 52, pp. 46-54.

Gramopadyhe, A.., Drury, C. (2000), 'Human factors in aviation maintenance: How

we got to where we are', International Journal of Industrial Ergonomics, Vol 26, pp.

125-131.

Jenson, D. (2008), 'Europe’s Challenges In a Dynamic MRO Market'. [cited 4th April

2009]; Available from: http://www.aviationtoday.com/.

Kidd, M., (2006), 'Automotive condition monitoring using standard vehicle

Architecture', Engineering Doctorate Thesis, University of Manchester, UK

Kothamasu, R., Huang, S., VerDuin, W. (2006), 'System health monitoring and

prognostics - a review of current paradigms and practices', in International Journal of

Advanced Manufacturing Technology. Springer-Verlag. pp. 1012-24.

Knotts., R.M., (1999), 'Civil aircraft maintenance and support fault diagnosis from a

bussiness perspective', Journal of Quality in Maintenance Engineering, Vol 5, No 4,

pp. 335-348.

Labib, A., (2010), 'Maintenance strategies: A systematic approach for selection of the

right strategies', Proceedings of the 4th World Congress on Engineering Asset

Management, Athens, Greece

Mobley, R. (2002). 'An introduction to predictive maintenance,' Materials &

Mechanical. Elsevier Butterworth-Heinemann.



123

Patkai, B., Theodorou, L., McFarlane, D,. Schmidt, K. (2007), 'Requirements for

RFID-based Sensor Integration in Landing Gear Monitoring - A Case Study', Auto-ID

Lab, University of Cambridge.

P. Phillips, D. Diston, A. Starr, J. Payne and S. Pandya, (2009), ‘A review on the

optimisation of aircraft maintenance with application to landing gears’, World

Congress in Engineering Asset Management, Athens, Greece

Raheja, D., llinas, J., Romanowski, C. (2006), 'Data fusion/data mining-based

architecture for condition-based maintenance', International Journal of Production

Research, Vol. 44, no. 14, pp. 2869-2887



124

C h a p t e r 5 : H e a l t h M o n i t o r i n g S y s t e m sM e t h o d o l o g y a n d F r a m e w o r k

5.1 Introduction

Before undertaking the design of any health monitoring system, a study on what is

required and by what process it can be or will be achieved is required. This does not

simply mean attempting to force the system to be monitored into a popular health

monitoring technique or into off the shelf monitoring systems. Rather an approach is

required that takes into consideration all of the necessary information regarding the

specific system to be monitored (i.e. failure history, design requirements, costs,

quality of information), utilising this information to develop and integrate application

specific health monitoring. In order to achieve this then an appropriate framework is

required which provides a set of generic guidelines for the development of the health

monitoring system.

The purpose of this chapter is to provide such a framework based upon the concept of

data fusion for the design of the actuator health monitoring system. The purpose of

which is not to provide a detailed specification of the hardware or software, but rather

to provide a conceptual description of the modules that will aid in the systems

development. Descriptions are enhanced through the use of relevant examples

throughout this chapter. A discussion is also provided on the engineering need for the

monitoring capability to satisfy verification criteria in the form of diagnostic

performance metrics and technical value.

5.2 Framework Objectives

The objectives of the framework presented as part of this research are defined as

follows (Phillips et al. 2008):

Provide a generic framework for electromechanical health monitoring which

will demonstrate how qualitative, quantitative and heuristic information can

be used together to achieve maintenance objectives.



125

Illustrate the importance and applicability of using data fusion for health

monitoring in this application.

To show the application and benefits of utilising data fusion alongside

conventional health monitoring techniques, methodologies and standards.

To provide a methodology for the conversion of high-level organisational

maintenance objectives into objectives pertaining to the specific actuator

system and subsystem components. These can then be used to aid in:

- Developing component descriptions, such as ‘part trees’

- Identifying failure modes for the relevant components in the form of

‘fault trees’.

To provide and understand the interrelationships of the various

architectural modules, which contain the functions and processes that

interrelate to accomplish the overall system goals.

Provide a generic framework and module descriptors, which are

applicable to a wide range of health monitoring techniques and allow the

architecture to be accessible for additional sensor data, information

extraction and decision making tools.

5.3 Overview of the Health Monitoring Data Fusion Framework

Based upon reviews of data fusion models, health monitoring standards, requirements

and standards a top level architecture has been chosen for the current application. In

order to maximise the information obtainable from multiple data sources then the data

must be used together effectively. Data obtained through on-board acquisition



126

systems and that derived through modelling techniques can be used to generate an

actuator health status, this can then be used in conjunction with management data such

as costs, risks and usage requirements to achieve an effective maintenance schedule.

Figure (5.1) illustrates the CBM system concept.

Figure 5.1: CBM System Concept

The health monitoring framework is based upon a standard ‘hierarchy’ which

partitions the CBM into several levels:

Platform - Landing gear

System - Electromechanical actuator

Subsystem - Gear box, Motor, roller screw assembly, control etc

Component - Gears, bearings, lubricant, wiring etc

At every level of the hierarchy there will be the need for the following information to

exist:

HealthStatus

PerformanceData

Past History

ManagementRequirements

AircraftMaintenance

Database

MaintenanceSchedulerManagement

Database

CBM System

MaintenanceSchedule

MaintenanceAction



127

1. Component identification

2. Identified component failure modes

3. CBM objectives pertaining to each group of system/subsystem/component

levels

4. Health estimates relating to the state of the failure mode,

Figure (5.2) is an illustrated representation of the proposed architecture along with the

corresponding OSA-CBM framework modules lain out by Raheja et al. (2006). It

should be noted that it is likely that a combination of fusion levels will be required, as

described by the JDL data fusion framework, reviewed in Chapter 3.



128

Figure 5.2: CBM Data Fusion Architecture

5.4 Inputs to the Fusion Centre

5.4.1 Objectives

5.4.1.1 CBM Objectives

The generic objectives/goals for CBM programme are usually defined by top level

management at the platform level; these are then broken down into increasingly

specific for each of the lower system/subsystem/component levels.

5.4.1.2 Top level Objectives

Management level objectives are usually measured in terms of costs, in particular

revenue losses or gains, for example:

Dataalignment

Hypothesisgeneration

Hypothesisselection

Hypothesisevaluation Estimation

Fusion Process

MaintenanceDatabase

ObjectivesPart treesFault trees

On-boardsensors

Signal Processing

Decision Support

Presentation

DataAcquisitionmodule

DataManipulationModule

DecisionSupport/PresentationModules

Datacleaning

Signalcondition

Featureextraction

HealthMonitoring/AssessmentModules



129

1. Reduce aircraft maintenance costs.

2. Maximise the number of aircraft flight hours.

3. Increase aircraft safety

5.4.1.3 Platform Level CBM Objectives

The platform level objectives, relate directly to the electromechanical actuators

mounted upon the landing gears. These objectives are normally associated with

reducing the probability of an event occurring. They relate to the failure of the entire

actuation system to perform its required function leading to a loss of the landing gears

normal operation such as the following:

1. Prevent the risk of the actuator from jamming leading to a failure to

retract/extend the landing gears, impacting upon aircraft safety.

2. Prevent inadvertent retraction/extension of the landing gears.

3. Prevent failure to damp the landing gears during extension/retraction, risking

increased structural damage.

5.4.1.4 Lower System/Component Level Objectives

Once the platform level objectives have been designed objectives for the lower

system, subsystem and component levels can be derived. These objectives will also

be based around improvements in the individual system and subsystems. For example

if an event such as actuator jamming is a result of a gear box failure, then the objective

at the system level may be to increase the Mean Times Between Failures (MTBF) of

the gear box. Likewise if the gear box failures are a result of the gears failing due to



130

tooth fractures the component level CBM objectives could be to reduce tooth fractures

in gear box gears for example by 10%.

5.4.2 Part Trees

In order for a system under maintenance to be capable of being associated with all of

the hierarchy levels, and to understand its component assemblythe construction of a

‘parts tree’ is useful. The part tree essentially identifies all of the parts which are

assembled to create the actuator system. Part trees that are used for the benefit of

CBM usually however only contain the parts which require maintenance or

monitoring. Figure (5.3) illustrates an example of a parts tree for the

electromechanical actuators gear train.

Figure 5.3: Example Parts Tree

5.4.3 Fault Trees

The development of a fault tree can aid in identifying the parts of the actuator system

which need to be maintained. This is done through associating individual components

with specific failure modes. The first step in developing a fault tree is to utilise the

Gear Box Assembly

ElectromechanicalActuator

GearHousing Gear Shaft

GearTeeth

GearBearing

Lubricant

Electrical Motor Roller Screw

Platformlevel

Systemlevel

Subsystemlevel

Componentlevel



131

parts tree to identify individual component failure modes and there effects upon the

system. The identification of the failure modes offers the following benefits:

1. Aids in setting detailed tangible objectives for the subsystem components.

2. The fault tree can be used to determine the sensing requirements needed to aid

in achieving the objectives. This includes the sensor type and suggestions on

the correct sensor placement.

Identification of the actuator failure modes is achieved through the use of a Failure

Mode and Events Analysis (FMEA) or a Failure Mode Event and Criticality Analysis

(FMECA) which incorporates the charting of the probability of failure against the

severity of their consequences. The FMEA/FEMCA ranks each potential failure

mode according to the level of severity (the greater the impact on the normal operation

the higher the severity), the rate of occurrence and the ability of the current measures

to detect the failure.

Fault tree construction is highly dependent upon expert knowledge which is captured

within the failure mode analysis. Once incorporated within a fault tree; the

information can allow informed decisions to be made regarding potential

abnormalities which occur during normal operation. This is achieved by means of the

fault tree acting as a steering tool towards a specific diagnosis. By applying rules to a

set of symptoms describing a general problem, such as an ‘actuator jam’ at the root of

the tree then progress can be made along the tree branches until a specific diagnosis

can be made such as ‘broken gear tooth’. An advantage of utilising tools such as fault

trees into the health monitoring architecture is that a partial solution can be found at

every stage of the reasoning process. A partial diagnosis can always be generated

given a symptom even if there is insufficient knowledge or data to provide a complete

diagnosis. Figure (5.4) illustrates an example of a fault tree.



132

Figure 5.4: Example of a Fault Tree

5.4.4 Observables

The effectiveness of any health monitoring system is dependent upon the ability to

obtain information regarding the actuators operating health. This is derived from

available data, and a particular set of algorithms are used to extract any information

indicating the onset of failure. These measurements or derived parameters are effects

of phenomena and are known classified as observables. An example in the current

application would be current signatures and temperature data (observables) to

ascertain winding failure (phenomena) in the motor. Observables will be either direct

or indirect indications of a failure mode.

Indirect indications require pre-processing to convert them into useful representations

of the phenomena; this can only be achieved if an adequate model describing the

relationship is available. The use of a fault tree provides the information on the

failures which can be measured, as well as information on the sub-systems and

components involved in the failure and what the observable effects would be. This



133

allows partitioning of the observables as direct or indirect and aids in the choice of the

most appropriate observables, depending on the criticality of the failure mode, in

which to measure and hence the choice of sensor and sensor location. Table 5.1 shows

common sensor measurements which can be used as either ‘direct’ or ‘indirect’

estimates of selected EMA failures

Table 5.1: EMA Faults and Associated Observables

Damagedgears

Bearingdamage

Electricalmotor faults

Roller-screwstructuraldamage

Inadequate/poorlubrication

Excessiveloading

Corrosion

Vibrationanalysis • • • •

Thermalmonitoring

• • •

Motor currentsignatures

• • • • • •

Corrosionmonitoring

•

Wear debris /lubricantanalysis

• • • •

Acousticsemissions andultrasonicwaves

• • • • •

Visualinspections

• • • • •

Performancemonitoring

• • • • • •



134

Observables do not have to just be information gained from a sensor placed upon the

actuation system. Information gained from control sensors for example can be used

with physical actuator models to derive information which help aid in the diagnoses of

particular failures. Examples include

Roller screw or motor efficiency.

Positions/speeds/accelerations.

Dynamic model parameters.

5.5 The Fusion Process

5.5.1 Alignment

For different data sets to be successfully combined together they must be consistent in

terms of measurement units and co-ordinate system. The purpose of the alignment

function within the architecture (Figure (5.2)) is to put the various data streams into a

common data and time basis. Various methods exist which can be used to unify these

type of symptoms (Raheja et al. 2006). An example of this (Equation 5.1) could

include the utilisation of physical coefficients ip estimated through the use of a

mathematical model which when normalised generates a set of symptoms with a

common unit’s basis in the range [0→1]:

(min)(max)

(min)

ii

ii

ipp

ppS

Equation 5.1

5.5.2 Association

Once the necessary data has been aligned it is necessary to associate each of the

important observables with a phenomena. It is not beneficial to try to fuse together

information relating to separate incidences so emphasis must be placed on

determining the correct association rule for each observable. There are difficulties in



135

developing association rules, also called hypotheses, as in ‘real world’ situations it is

more often than not the case that the same observed variable may be associated with

multiple events.

A simple example highlighting association rules would begin at defining health

monitoring objectives at the component level as being that all gearbox housing unit

seals must remain undamaged and air/water tight. Damaged seals will allow

contaminants such as condensation and dirt to degrade components such as gears

through corrosion and wear. Excess humidity in the gear box housing can also freeze

at high altitude temperatures which may impede the required interaction between

gears, leading to the actuator jamming. If it is assumed that there is humidity sensing

equipment mounted within the gear box, which then raises an alarm alerting to the

presence of water, then utilising a ‘fault tree’, illustrated in Figure (5.5) can help to

determine a set of likely associations for this observable, each association having a

given probability of occurrence.

Figure 5.5: Example of Symptom/Fault Association

This example shows two possible associations with the presence of humidity, the first

‘seal damage’ has a higher confidence than the second ‘cracked housing’ generated

via knowledge of failure occurrences. The association of an observable with a failure

mode aided by the fault/part trees and each probable association can be referred to as a



136

likely hypothesis. Determining the associations is one of the initial tasks in data

fusion problem development; this is then followed by the following sub-tasks

1. Hypothesis generation: the use of sensor data to generate possible associations

2. Hypothesis evaluation: here the level of support for each association is

determined

3. Hypothesis selection: the optimal association rule is selected

5.5.3 Hypothesis Generation

The use of data mining methods is a popular choice in determining optimal

associations. These include case based reasoning and decision trees. The use of

decision trees as a data mining methodology provides optimal associations of any

sensor data to the observed phenomena. Unlike case-based reasoning approaches to

fault diagnoses, decision trees cover all anticipated faults and failures regardless if

they have occurred before or not. This is important if the system is new and

insufficient historical data concerning past events is available. As the lifetime of the

system increases and more failure occurrences are recorded case-based reasoning can

become a more optimal way of generating hypothesis. Data mining forms a ‘feedback

loop’ with the health monitoring modules by searching known information to form

hypothesis and identifies the important sensor data to be evaluated.

There are two knowledge engineering methods used in conjunction with data mining

techniques are useful in generating association rules (hypothesis), these are backward

and forward chaining. To highlight these methods let it be considered that there is a

defined list of rules associating a set of symptoms (extracted from sensor signals, such

as temperature or vibrations) to failure modes in the form

R₁: IF Gearbox has failed–THEN Actuator jams



137

R₂: IF Motor has failed – THEN Actuator jams

R₃: IF Gear tooth has broken –THEN Gearbox has failed

R₄: IF– Motor bearings are seized THEN Motor has failed

R₅: IF symptoms s₁, s₂ and s₃ are observed - THEN Gear tooth has broken

R₆: IF symptoms s₄, s₅ and s₆ are observed – THEN Motor bearings are seized

The approach known as backward chaining where the mining of the rule base begins

with a consequence which is a known occurred failure and searches out the rules until

an antecedent (fault evidence) is found to be true. In this example the fact that the

actuator has jammed is known to have occurred and is the consequence of a failed

component. Backward chaining will work through the rules in the following way

R₁→R₃→R₅, or R₂→ R₄→R₆, hence in this example we have two possible

hypotheses each with a set of symptoms to be evaluated. As this example shows

backward-chaining is a goal driven process, in this case the goal is determining the

cause of an actuator jam.

In a landing gear system however it would be desirable to be able to detect these

symptoms to aid in determining the onset of damage before the event actually occurs.

Following a reverse process known as forward chaining which is much more data-

driven, that is using data to infer faults and further possible consequences. Using the

above example rules the system begins with a set of observed symptoms which are

known to be true; the rule base is then searched until a matching consequence (failure

mode) is found giving association rules following the paths R₅ → R₃→ R₁ and

R₆→R₄→ R₂.

5.5.4 Hypothesis Evaluation

Once a set of hypotheses have been formulated then they need to be evaluated to

determine the level of support for each of them, this is given in the form of a

probability of occurrence. This support is determined through domain knowledge and

the modelling of specific sensor-to-phenomena relationships. Let’s say, for example,



138

that the proximity sensors present on the landing gears do not detect a landing gear

uplock, indicating that the retraction actuator has jammed. There are a variety of

faults at the component level which may have attributed to this platform level failure

mode. Following on from hypothesis generation, failure rates/probabilities of

components can be used in tangent with the decision tree to generate a measure of

confidence, in the form of a probability that that fault mode has occurred.

For simplistic example, the generated hypothesis of the cause of an actuator jam is

either broken gears housed within the gear box, or due to seized bearings within the

roller screw system. Therefore the following hypothesis can be evaluated:

Landing gear failed to retract due to an actuator jam resulting from gear box failure

due to seized gears caused through a broken gear tooth – support X

OR

Landing gear failed to retract due to an actuator jam resulting from roller screw

failure due to a seized actuator nut caused through bearing damage – support Y

5.5.5 Hypothesis Selection

Measures of support and uncertainty in the diagnostics can be used to rank the various

hypotheses in order of the level of support. The most likely hypothesis would then be

the diagnosis with the highest support and lowest uncertainty. If a clear diagnosis

cannot be made, then at this stage it may be necessary to improve the confidence level

by incorporating heuristic symptoms, past maintenance histories and/or further fault

statistics

5.6 Estimation

Once the most probable hypothesis has been selected, it is then necessary to make an

estimate representing the degree of damage. This is performed by a model or

algorithm capable of merging the different data sets containing representations of the

most probable fault selected through the association process. The output therefore

will be a fused estimated representation of the condition of a particular actuator



139

component, sub-system or system. There are a variety of methods which can be used,

including statistical mathematical techniques such as Dempster-Shafer theory,

Bayesian reasoning, neural networks and fuzzy logic

5.7 Proposed Decision Support Outputs

The monitoring system is not aimed at telling the maintenance personnel when to

perform maintenance operations. But rather to offer an efficient tool supporting them

in making optimised maintenance decisions. This is because aerospace legislation and

procedures are not ready to except a fully automated intelligent decision making

system, without human input, for safety critical operations. It would be beneficial to

the maintenance engineer if the diagnostic results were combined to give a global

actuator quality index. This value could then be correlated to actuator performance

loss as shown in Figure (5.6). If maintenance scheduling is based upon system

degradation, then a measure of performance loss will be more explainable in terms of

decision making.

Figure 5.6: Illustration of Diagnostics Output for

Decision Support

5.8 Considerations for Practical Implementation

The conceptual design presented her in this chapter is aimed at providing a template

for the creation of a corresponding implementation strategy for health monitoring.

There are two distinct implementation strategies which can be identified for vehicle

Fault ClassificationResults

Fault 1

Fault 2

Fault N

PerformanceLoss (%)

0

20

40

60

80

Actuator QualityIndex

1

0.8

0.6

0.4

0.2



140

health monitoring (Kidd 2006) and adapted to the specific case of landing gear

actuator monitoring.

5.8.1 Service bay implemented

One of the options for health monitoring implementation is the case where all of the

health monitoring and diagnostics is performed at a maintenance service bay. In this

strategy the aircrafts landing gear would be put through a series of on wing tests as is

currently the case, for example, multiple retraction/extension cycles under a variety of

loading conditions. The data which is then obtained would be processed by the health

monitoring system and algorithms, which in this case can be conveniently located on a

local computer, or even on the technicians Personal Digital Assistant (PDA). This

combined with data and information gained from the aircrafts standard Built in Tests

(BIT) and flight records, the status of the landing gear actuators health can be

assessed. The key advantages and disadvantages to this model can be summarised as

follows:

Advantages:

The health monitoring solution is not restricted by hardware issues such as

sensor weight/complexity/ability to operate in the harsh environment of the

landing gear bay. For example, specific sensors could be attached to the

actuator for the duration of the maintenance testing only. Therefore, such

methods as acoustic monitoring which are restricted for embedded health

monitoring, despite the value of information they can provide, can be utilised.

Diagnostic results can be instantly verified by the maintenance technicians and

any anomalies in the results traced back to the source. In flight embedded

monitoring requires that the on-board computing makes the decision with no

immediate human input or validation. This reliance on automated decision

making for such a safety critical process will almost certainly prove difficult to

certify under strict aerospace regulations.



141

The use of automated diagnostic tools during service bay maintenance will

almost certainly speed up the maintenance procedure. Minimising the time

that the aircraft is out of operational service.

Disadvantages

One of the key aims for health monitoring is for the reduction in scheduled and

unscheduled maintenance. In bay monitoring will almost certainly aid in

reducing unscheduled maintenance activities, but scheduled maintenance may

remain unaffected.

This model will not allow for any in flight warning system, pertaining to the

imminent failure of the landing gear actuator. However the mechanical and

electrical redundancy within the actuator system would mean that this is not

necessary as the gears are guaranteed to operate.

5.8.2 Embedded deployment

The second approach differs from the service bay deployment in that the monitoring

algorithms would be embedded within the aircraft alongside control algorithms and

BIT testing. This deployment strategy would allow for all of the processing to occur

on-board the aircraft, which would require strict consideration for hardware

requirements.

Advantages

Embedded health monitoring would allow for pre-flight warnings on the

landing gears or other monitored systems.

It would almost certainly be the most effective monitoring approach to reduce

the levels of scheduled maintenance. Maintenance operations would also be

quicker as the technical engineer can be directed immediately to the faulty

system/component based upon pre-processed information.



142

Disadvantages

The aircraft would have to be recalled every time the health monitoring

software requires updating.

There are restrictions on the approaches which can be used due to hardware

weight/certification procedures and signal processing capabilities.

5.9 Health Monitoring Acceptance Criteria and Metrics

5.9.1 Validation Procedure

An important issue in health monitoring systems is the ability to assess the

effectiveness of the system in performing its task of timely and correct diagnosis of

faults. This can be split into both technical and economic feasibility system metrics.

In terms of technical performance any health monitoring system would need to be

validated against set performance metrics. in order to do this the engineer would need

to decide on the most effective metrics for the application and set the necessary

thresholds.

The diagnostic performance would first need to be tested using experimental data

from a representative landing gear testing rig. As part of the landing gear actuators

acceptance tests it is known that they are put through a serious of destructive testing

and are tested through 10,000 on wing retraction/extension cycles. The data obtained

is useful in health monitoring validation for two reasons. Firstly if the underlying

algorithms are based upon system models, performance and control data from these

cycle tests can be used to estimate the actuators dynamic parameters and therefore

validate the model, as well as generating a database of nominal/faulty parameters.

Secondly as experimental faults are incorporated into the actuator and tested on the

landing gear rig, the ability for the health monitoring to detect/diagnose those faults

can be evaluated without risk to aircraft safety. Once these test are completed and the

health monitoring has been verified as the correct system for the application, it would

then be implemented onto a variety of aircraft for in-flight testing. In this stage of

validation the aircraft is in no way dependant on the health monitoring system for



143

safety or reliability but the aim would be to generate data on the operational

performance of the system - this is also useful in evaluating the technical value in

commercial terms of implementing the system.

There are several key aspects which need to be considered in validating the landing

gear health monitoring system,

1) Different aircraft need to be validated seperately. Failure modes in high altitude

long haul aircraft are likely to be different to low altitude short haul aircraft.

2) The operating location of the aircraft to be implemented with the health monitoring

system must be considered. Some failure modes, may generate more false alarms in

colder climates, for example lubrication will become sticky possibly generating a

false alarm indicating mechanical damage.

5.9.2 Fault Diagnostic Performance Metrics

Major accepted performance metrics for a diagnostic process are given as False

Negatives or False Positives. False negatives may present risks to the health of the

machine, with the missed fault condition leading to catastrophic failure. Also, a high

percentage of false positives is likely to result in a loss of confidence in the diagnostic

algorithms by the system operator. diagnostic algorithm performance requirements

would need to specigfy the maximum number of acceptable false positives and

negatives as a percentage of the total faultsin the monitored system over its expected

life.

Diagnostic events are therefore evaluated through a decision matrix (Liu and Motoda

1998) as shown in table 5.2



144

Table 5.2: Decision Matrix for Fault DetectionEvaluation

Outcome Fault 1F No Fault 0F Total

Positive (detected)

1Da

Number of detectedfaults

bNumber of false

alarms

a + bTotal Number of

alarms

Negative (notdetected) 0D

cNumber of missed

faults

dNumber of

correct rejections

c + dTotal number of

non-alarms

a + cTotal number of

faults

b + dTotal number ofcorrect rejections

a + b + c + dTotal number of

cases

From the matrix in table 5.2 the following metrics can be calculated. The probability

of detection(POD) given a fault assess the detected faults over all potential fault cases.

ca

aFDPPOD

)/( 11 Equation 5.2

The probability of false alarm (POFA) considers the proportion of all fault-free cases

that trigger a fault alarm.

db

bFDPPOFA

)/( 01 Equation 5.3

A metric of accuracy would need to be used to measure the effectiveness of any

diagnostics system or algorithm in its ability to correctly distinguish between a fault-

present and fault-free condition. The accuracy metric (Equation 5.4) uses all of the

data available for analysis (both fault and no-fault)

dcba

daFDFDPAccuracy

)/&/( 0011 Equation 5.4



145

The two metrics, false positives and false negatives may change if the detection

thresholds vary or if tradeoffs are required. It is therefore necessary to assess how

well a fault is actually diagnosed. One method which is proposed is the use of what is

termed the Receiver Operating Characteristic (ROC). The ROC provides a

comprehensive overview of the tradeoffs between false positives and false negatives,

an example of a ROC curve is given in Figure 5.7. A guide to constructing ROC

curves are given in Vachtsevanos 2006.

From Figure 5.7 a straight curve signifies that the diagnosis has 50/50 odds of being

correct. As the curve bows more to the left, it indicates a greater accuracy (i.e. a

higher ratio of true positives to false negatives). The accuracy of the diagnosis is more

precisely measured by the area under the curve, which increases as the curve bows to

the left Using these technical performance measures will require the system operator

to design specific thresholds, for example diagnostic accuracy of 80%, which the

health monitoring system would have to match or exceed for validation.

Figure 5.7: The ROC Curve



146

5.9.3 Technical Value

The benefits which are achieved through accurate detection and diagnosis of faults

need to be weighed against the costs associated with false alarms, innacurate

diagnosis, costs and resource rquirements of implementing and operating the health

monitoring system. A metric for the technical value of a health monitoring system in a

particular application is the summation of all the benefits which it provides over all of

the failure modes which it has been designed to diagnose minus the

implementation/operating/maintenance costs. A total value equation (Vachtsevanos

2006) is given as:

))(1()(ValueTech 1 PPPIDP Dff Equation 5.5

fP =probability of failure occurrence

D = overall detection confidence score

=savings realised by detecting a fault prior to failure

I = overall isolation confidence metric score

= savings realised through automated isolation of a fault

DP = false positive detection metric score

= cost associated with false positive detection

1P = false positive isolation metric score

= cost associated with a false positive isolation

Therefore the total value of the health monitoring system is given by:

ModesFailure

)1(ValueTech cPOATOTAL Equation 5.6

A = acquisition and implementation costs

O = life-cycle operation and maintenance costs

cP = computer resource and requirements score

= cost of computing systems



147

For all of these metrics, a low score would indicate an undesirable result and a high

score indicates a desirable one. Equations 5.5 and 5.6 therefore illustrate the necessity

in applying tradeoffs between differing metrics in order to obtain an acceptable overall

value of the health monitoring system.

5.10 Conclusions

Before a practical implementation strategy can be implemented a conceptual

framework which provides descriptions of the key modules and guidelines to their use

is created as a health monitoring system template. In this work this template has been

constructed with considerations on the requirements for the actuator health monitoring

system to allow the integration of an application specific health monitoring system.

The framework discussed in this chapter is based around the concept of data fusion

and discusses the key areas such as generic inputs, data association, hypothesis

generation/evaluation/selection and proposed decision outputs. The frameworks

objective does not provide detailed recommendations on specific hardware or software

based strategies but rather acts as a kind of methodology in health monitoring design.

There are still a variety of key design issues which remain unanswered and will need

to be tackled as part of this framework these can be identified are as follows:

Providing a detailed and accurate analysis of all the potential failure modes

relevant to a particular component or assembly.

Modelling the influence of one component failure over another. This therefore

requires a comprehensive understanding of component - component

interaction.

For the introduction of health monitoring into a vehicle platform considerations need

to be made on the deployment strategy. There are two separate possibilities for this.

The first is dependent on the monitoring algorithms and decision support to be made

at a maintenance service bay, where the monitoring software is located. The second



148

approach would see the health monitoring algorithms embedded onto the aircraft

alongside control algorithms and BIT testing. Both deployment strategies have a

variety of advantages and disadvantages which have been discussed but no single

strategy is committed to as part of this work. It is highly likely that for effective

aerospace vehicle monitoring a hybrid of the two deployment strategies would be a

natural approach.

Before a health monitoring system can be accepted it must be verified against a set of

pre-determined performance metrics in order to justify its acceptance. Diagnostic

performance metrics, which is the ability to accurately diagnose faults, assessment

would be through consideration of the number of missed alarms or false alarms in

order to generate a ROC curve. The use of the ROC curve allows the health

monitoring designer to gauge the overall diagnostic accuracy of the system - which

would need to match a set threshold criteria. The second metric which has been

discussed is a method of providing the overall technical value of the health monitoring

system. This finds the difference between cost savings over all possible detectable

failure modes and the costs of implementation and upkeep. This is an important

metric which must exceed the value of not having the monitoring system.

5.11 References

Kidd, M., (2006), 'Automotive condition monitoring using standard vehicle

Architecture', Engineering Doctorate Thesis, University of Manchester, UK

Liu, H., Motoda, H., (1998), "Feature selection for knowledge discovery and data

mining", Boston: Kluwer Academic.

Phillips, P., Diston, D., Payne, J., Pandya, S., Starr, A. (2008), 'The application of

condition monitoring methodologies for certification of reliability in electric landing

gear actuators', in The 5th International Conference on Condition Monitoring and

Machine Failure Technologies. Edinburgh, UK.



149

Raheja, D. (2006), 'Data fusion/data mining-based architecture for condition-based

maintenance', International Journal of Production Research, Vol 44. No 14. pp. 2869-

2887.

Vachtsevanos, G., Lewis, F., Roemer, M., Hess, A., WU, B. (2006), 'Intelligent Fault

Diagnosis and Prognosis for Engineering Systems' John Wiley & Sons , INC.



150

C h a p t e r 6 : A p p l i c a t i o n o f F u z z y L o g i c a n d

P r i n c i p a l C o m p o n e n t A n a l y s i s f o r D e t e r m i n i n g

P r o c e s s Q u a l i t y

6.1 Introduction

The previous chapter provided the framework constructed for use in developing

actuator fault diagnostics. The framework reviewed the factors relating to the

information which should be gained from a monitoring system for effective

maintenance support. This chapter proposes a mathematical algorithm which

addresses one way in which these outputs can be achieved, whilst keeping within the

design requirements for the monitoring system (Phillips and Diston 2010).

The monitoring outputs for decision support are categorised as 'fault detection',

represented by overall actuator operating quality and 'fault diagnostics' represented by

a ranking of probabilities of specific faults occurring. Through estimating the overall

actuator process quality, then the affects that individual faults are having on the

actuator systems ability to perform to a specific requirement are identifiable. The fault

detection within this research is based upon the use of Principal Component Analysis

(PCA) which uses only the actuators dynamic process data, obtained as part of the

actuators control system. The use of PCA allows reductions in the size of the data set

and generates statistics, which are analysed through the use of a fuzzy logic rule base

to provide a quantified performance quality. This actuator quality can then be

monitored to ensure that it does not fall below a predefined quality threshold, which if

it does full fault diagnostics would be initiated.

This chapter presents the formulisation of the fault detection algorithm and provides

the results of an experimental demonstration. The experiments make use of an off-

the-shelf bench top actuator with several degrees of lubricant degradation

incorporated into the actuator gears and screw assembly. The experiment highlights

the potential of the algorithm in fault detection for lubrication faults in the actuators.



151

6.2 Estimation of Actuator Process Quality

6.2.1 Data Redundancy

The monitoring of a machine or process often involves large volumes of data, both in

terms of the number of samples and data sets, the latter of which indicates a

requirement for a large number of sensors. However the meaningful information

contained within this data can be significantly less than the volume of the data may

suggest. When recording measurements from multiple sensors the issue of

redundancy arises. What is meant by this is the question: do the individual data sets

record the same dynamic information and therefore is it necessary to record all of

these sets? Figures (6.1) and (6.2) illustrate two possible plots between two variables.

Figure (6.1) shows how the variables are highly correlated with each other (high

redundancy) and Figure (6.2) shows the case where the variables are uncorrelated

(low redundancy).

Figure 6.1: Example of Two Data Sets With High

Redundancy



152

Figure 6.2: Example of Two Data Sets with Low

Redundancy

High redundancy indicates that the two variable A and B are statistically dependable,

likewise low redundancy indicates that the two variables are statistically independent

and cannot easily be predicted from oneanother. The highly correlated variables can

be predicted by simply analysing the best fit line so therefore only one variable

requires measurement reducing the data sets 12 , this is the concept of data

reduction. If there is a high measure of redundancy between multiple variables that are

susceptible to variation under faulty conditions then the concept of data reduction can

be applied, for monitoring purposes. The case with aerospace actuation, the dynamic

operating data which consists of control/power inputs, motor current, rotational

speeds, loads and position should under healthy conditions have a high degree of

correlation. For fault detection purposes, each of these measurements can be assessed

individually against some pre-requisite requirements. However, the use of data

reduction techniques may therefore be a more applicable approach to simplifying this

task (Kambhatla et al 1997).

6.2.2 Principal Component Analysis

Principal component analysis is a vector space transformation often used to transform

multivariable space into a subspace which preserves maximum variances of the



153

original space, but with a reduced number of dimensions (Jackson 2003). The

measured variables in the original space are usually highly correlated, so PCA can be

defined as a linear transformation of the original correlated data into a new set of

uncorrelated data that explains the trend in the process. The formulation of a PCA is

as follows. First a new mn matrix is defined as Tm1 xxX . Each row of

X corresponds to measurements of a particular type )( ix and the columns correspond

to a set of measurements from a trial. The covariance matrix of this data set can be

written as

TX XXC .

1

1

nEquation 6.1

Equation 1

The factor1

1

nis a constant for normalisation. The covariance matrix XC captures

all of the correlations between all possible pairs of measurements and reflects the

noise and redundancy in the measurements. It has the following properties:

XC is a square mm matrix

The diagonal terms of XC are the variance of particular measurement types.

Large values correspond to interesting dynamics and small values correspond

to noise.

The off-diagonal terms of XC are the covariance between measurement types.

High values correspond to high redundancy and low values low redundancy.

The first step is to calculate PCA is to construct the covariance matrix given

inEquation 6.1 and performe the Singular Value Decomposition (SVD) on the matrix

as follows:



154

TΛVVCX . Equation 6.2

Where Λ is a diagonal matrix containing the eigenvalues of XC in decreasing order

( 021 m ). The columns of the matrix Vare the eigenvectors of XC . A

transformation matrix P of dimensions am , where a is the chosen number of

eigenvectors which are used as the principal components, can be generated to

transform the original measured variable space into the reduced dimension space

PXT . Equation 6.3

The matrix P is known as the loading matrix and T as the score matrix. The elements

of T are the values of the original measured variables that have been transformed into

the reduced dimension space.

6.2.3 Choosing the Principle Components

For PCA to be successful in a health monitoring context then the appropriate

number of principle components must be selected. In this instance a method known

as Cumulative Percent Variance (CPV) is used to calculate the correct number in

which to project the data onto. This approach chooses the number of principle

components (a) for a particular measure of percentage variance %)95)(i.e.( aCPV .

The CPV is calculated using:

m

jj

a

jj

CPV

1

1

Equation 6.4



155

6.2.4 Generating Performance Statistics

Traditional fault detection based upon PCA is achieved through the generation of a

statistic known as a Squared Prediction Error (SPE). The SPE is aimed at capturing

any small variations between the PCA model data and the next set of process

measurements. Under nominal conditions the SPE will remain close to zero but any

variation in the process will lead to a positive or negative deviation, which if a set

threshold is breached an alarm is raised. The SPE can be calculated in the following

manner.

From Equation 6.3 it can be seen that the scores can be transformed into m-

dimensional observational space by:

TPTX .ˆ Equation 6.5

The difference between the observation space X̂ and the original data space

X represented by the residual matrix E which captures the variations in the

observation space spanned by the loading vectors associated with the m-smallest

singular values.

XXE ˆ Equation 6.6

The SPE is therefore defined by:

EEQ TT Equation 6.7

Using a SPE the process is considered normal if the process is within an upper

control limit represented by

2δQ Equation 6.8



156

A second statistic is calculated using a 2T - Hotellings statistic. In this work we can

consider the scores to be bounded by an elliptical threshold, which is regarded as the

confidence region which is defined as (Palma et al. 2005).

2

2

22

1

21

Ttt

Equation 6.9

In Equation 6.9 nt are the projections along the orthogonal axes defined by the

loading vectors, 1 and 2 are the principle eigenvalues (Note that Equation 6.9

represents the case where the number of principle components a=2). The 2T

statistics threshold is defined by Equation 6.10. The value ),( anaF is obtainable

from an F-distribution table for a certain level of significance (Jackson 2003).

),()(

)1)(1(2 anaFann

nnaT

Equation 6.10

In this work both the Q and 2T statistics will be combined together in order to

generate an overall measure of process quality.

6.2.5 Fuzzy Logic Classification

After the appropriate features have been identified then an inference to their meanings

has to be made. An appropriate technique, which fits well here, is the use of fuzzy

logic classification (Bartys et al. 2005) which can deal with reasoning that is

approximate rather than precise. There are three stages to classification via fuzzy logic

(Leekwijick & kerre 1999). The first is to convert the input variables into fuzzy

variables (fuzzification); the second is the evaluation of a fuzzy rule base and finally

converting the fuzzy outputs back into crisp numbers (defuzzification). The first stage

begins with the designing of individual membership functions. Here the variables are

mapped to predetermined membership functions and are converted into linguistic

variables, such as Okay, Medium or High as illustrated in Figure (6.3).



157

Figure 6.3: Example of a Fuzzy Membership

Function

A membership function represents information in a vague and somewhat ambiguous

way. There is a variety of differing shaped membership functions, which include

(non-exhaustively) triangular, waveform, trapezoidal, Gaussian, bell-shaped and

sigmoidal. The choice of membership function is often subjective, but in general for

systems that require significant dynamic variation in a short time period triangular or

trapezoidal functions are used. The method of choosing the boundaries for the

membership function is illustrated in Figure (6.4). A variety of threshold boundaries

are created around the input residual, these boundaries are indicative of regions which

are to varying levels acceptable or not.



158

Figure 6.4: A Subjective Approach to Designing a

Fuzzy Membership Function

In this methodology the inputs for the fuzzy rule base will be the fuzzy Q and

2T statistics obtained from the PCA of the system. The output will be a measure of

the quality Q of the actuators operation

6.2.6 Fuzzy Rule Base

The second stage is the evaluation of the fuzzy inputs against a set of fuzzy rules to

provide a set of fuzzy output representing qualitative assessments of the system. An

example of the rule base will be constructed from the following form:

IF 1x is 1B and nx is nB THEN y is C

Where represent the input variables and the output variables.

The terms B and C are the variables characteristic of the membership functions.

These rules offer a convenient way of expressing reasoning simply and transparently.

A rule base such as this does have limitations and is usually not a sufficient approach

to be considered for systems, or processes that have a large array of variables each

with intrinsic correlated relationships leading to overly sized rule bases.

Input value

-0.8

0.8

0.2

-0.2

1

80 %

-80%

20%

-20%Time

OK

LN

LP

Dangerous

Dangerous



159

The application of a rule base in fuzzy logic must assess the fuzzy inputs and project

these onto the consequences of that rule. Table 6.1 shows an example of the fuzzy rule

base:

Table 6.1: The Fuzzy Rule Base

Okay Medium High

OK VG G M

Medium G M P

High M P VP

As can be seen in Table 6.1 each rule has a specific output, in this case they have been

categorised as Very Poor (VP), Poor, (P), Medium (M), Good (G) or Very Good

(VG). The output values for each rule fired are determined by the use of the AND

operation which takes the minimum membership valueand maps this value onto a

specific fuzzy output function, illustrated in Figure (6.5).

Figure 6.5: Example of a Fuzzy Output Function

SPE

2T



160

For example consider the SPE was measured at 0.7and the 2T measured as 0.2. These

values are then mapped onto their respective membership functions which according

to Table 6.1 the following two rules are fired:

RULE A: If SPE is HIGH (0.78) AND 2T is MED (0.1) Then Quality is POOR (0.1)

RULE B: If SPE is MEDIUM (0.3) AND 2T is OK (0.9) Then Quality is GOOD

(0.3)

Figure (6.6) graphically shows the application of these rules applied to the

membership and output function in Figure (6.3) and (6.5). The quality inference is

taken as the output with the largest area; in this example the system quality is GOOD.

Figure 6.6: Combining Symptoms According to

Fuzzy Rules

6.2.7 Obtaining a Quantitative Quality Index

In the case where several of the rules fired provide the same qualitative output, with

varying degrees of membership, then the first step too defuzzification is calculation of

0 0.6SPE

High0.7

2T

Med

0.3

RULE B

RULE APoor

0.3

-100 0

0.1

Good

Min = 0.1

Min = 0.3

SPE0.60 +100

Med

0.1

0.20

2T

0.2

0.9

OK

00



161

the output strengths. In this case a Root-Sum-Squares (RSS) approach is used which

combines the effects of all applicable rules, scales the functions at their respective

magnitudes, and computes the fuzzy centroid of the composite area. This method is

more complicated mathematically than other methods, but was selected for this

application since it seemed to give the best weighted influence to all firing rules. The

calculation of the fuzzy centroid is given by:

strengthsoutput

strength)output).(centreoutput(Centroid Equation 6.11

Figure (6.7) illustrates the defuzzification process with consideration of the example

in the previous section.

Figure 6.7: Example of the Defuzzification Process

Using the 'Centroid' Algorithm

6.3 Experimental Demonstration

6.3.1 Experimental Objectives

In order to demonstrate the potential usage of the proposed method for estimating

actuator quality for fault detection purposes, a bench top actuator is used with varying

degree of lubrication faults.

Poor

-100% +100%

V. Poor Good V.Good

0

25% GOOD

Strength of Outputs

)(3.03.0 2 Good

)(1.01.0 2 Poor

Fuzzy Centroid

)%(253.01.0

)3.050()1.050(Good



162

The purpose of the experimental demonstration are as follows:

1) Demonstrate the effects which degraded lubrication can have on the actuator

performance - illustrating that for this fault case there are directly observable changes

in the actuator speed (cycle time) and actuator current.

2) Illustrate the potential of the proposed PCA and fuzzy logic based fault detection

process. Highlighting that lubrication based faults can be easily detected with the use

of control system based sensors.

3) Test the integrity of the proposed approach in providing a global actuator quality

index which can be regarded as a realistic representation under varying fault

conditions.

6.3.2 Experimental Setup, DataAcquisition and Post-Processing

The actuator selected for the testing demonstration is a MecVel ALI3 24VDC single

phase motor operated leadscrew actuator. The actuators maximum stroke length was

0.45 m and a lead of 3102 m. This particular actuator had been selected primarily

due to its low cost and availability. The actuator stroke length control was achieved

through the use of electronic limit switches which break the power circuit when

tripped. The stroke length was varied through manually adjustiment of the CAM

which was directly connected to the actuator screw shaft via a worm screw gear. The

signals which were available for this setup were the position of the actuator nut and

the current drawn from the actuators motor. Figure (6.8) provides a schematic of the

experimental setup.



163

Figure 6.8: Schematic of the Experimental Setup

The position of the actuator nut was obtained through the use of an angular

potentiometer. A potentiometer is a commonly used position sensing device which

operates as a variable resistor with relation to the position of a wiper mechanically

coupled to the actuator shaft. Figure (6.9) shows the relation of the potentiometer to

the Cam shaft and electronic limiting switches.

Figure 6.9: The Actuator Potentiometer*

* Diagram obtained from MecVel ALI3 24VDC single phase motor operated leadscrew actuator data sheet

ComputerNI-USB6009 DAQ

Unloaded MecVELElectrical Actuator

Potentiometer andlimit switches

Current sensor

Current

Position

Data



164

The potentiometer rotated between 0 º and 340º dependant upon the direction of the

actuator travel returning a voltage output of 10V at its maximum rotation and 0V at its

minimum rotation. This represents the actuator position as a function of voltage and

can either be used directly in this form or it can be used to provide information

regarding the actuator position and speed in more conventional units.

Measurements on the current drawn by the motor were available through the use of

LEM type LTA50P/SP1 closed loop current transducer employing the principles of

the Hall Effect to measure D.C currents. The data from the potentiometer and hall

sensor was obtained through the use of a NI USB-6009 low-cost multifunctional DAQ

connected to a laptop computer with the LABVIEW software suite installed. After the

current and positioning signals were obtained excess noise was removed from the

signals using an inbuilt LABVIEW low-pass filter. The data from the potentiometer

at this stage was also converted from a voltage time based signal into a distance

(metres) time signal using a lookup table which mapped potentiometer volts to

actuator stroke length in metres.

Post-processing for the task of fault detection was achieved offline within the

MATLAB environment. The PCA and fuzzy logic algorithm were programmed into a

MATLAB script which read in the saved current and positioning data performing the

necessary calculations wirthin the algorithm. The MATLAB program provided saved

data files containing the outputs from the fuzzy membership functions and the final

quality estimation as functions of time ready for plotting.

6.3.3 Implementing a Lubrication Fault

In order to simulate lubricant failure into actuator hardware it is a relatively simple

task of cleaning lubricant from the actuators various mechanical elements. It is

difficult to assign a numerical measure representing fault severity when implementing

lubrication degradation into an experimental actuator. So in this case a systematic

qualitative measure will be used for a succession of five separate test through gradual



165

removal and insertion of abrasive material on key mechanical areas as shown in Table

(6.2) are used to simulate lubricant degradation and increases in friction.

Table 6.2: Implementing Lubrication Fault

Fault Fault severity

Lubricantfailure

No removal

Partial removal from gearsand screw

Dry gear

Dry rollerscrew/nut andgears

Insertion of abrasive materialonto gears and between

screw/nut

6.3.4 Actuator Responses

Figure (6.10) provides the data gained from the angular positioning potentiometer

attached to the actuator demonstrating the cycle time for both a full 0.45m extension

and retraction cycle under fault free conditions at room temperature. The signals are

filtered to remove excessve noise, and the position output from has been converted

into a position measured in metres hrough the use of an appropriate lookup table.



166

Figure 6.10: Nominal Actuator Position and CurrentResponses

Figure (6.11) provides an example of two cases which were used to represent

lubrication faults. The first case (Faulty 1) lubrication was totally removed from the

mechanical elements (gears, screw, nut) and the second case (Faulty 2) the addition of

abrasive material (sand) has been inserted between the nut and screw mating surfaces.

This case represents an extreme fault case where total losss of lubrication has resulted

in material spalling. As the fault cases are increased the effects on the actuator

performance are slower cycle speeds and slight increases in current.



167

Figure 6.11: Example Position and CurrentResponses with Removed Lubricant

6.3.5 Estimating the Nominal PCA Model

Table 6.4 and 6.5 show the nominal eigenvectors and Eigen values based upon the

nominal data in Figure (6.10). Choosing a CPV >90% from Equation 6.4 results in

the requirement for data to be projected onto the first eigenvector (principle

component) reducing the data set dimensions from 2→1.

Table 6.4: Eigenvectors

1V 2V

0.3152 0.94900.3152 0.3152

Table 6.5: Eigenvalues

1 2

27.76 1.5972



168

After projecting the data presented in Figure (6.10) and (6.11) onto the principle

components the next stage is to calculate the squared prediction error and the

Hotellings 2T statistic using Equations 6.9 and 6.7 respectively. the results of which

are shown in Figure (6.12) illustrating a variation in both statistics, for each of the

three cases tested.

Figure 6.12: Performance Statistics



169

6.3.6 Fuzzy Inference

Examination of how the performance statistics change with removal of the actuators

lubricant is used to define two triangular fuzzy membership functions similar to

Figure (6.3). The centres of the triangular membership functions used are given in

Table 6.6. Figure (6.13) shows the transformations of the SPE and 2T statistics into

fuzzy inputs.

Table 6.6: Input Membership Function Centres

Okay Medium High

2T 0 0.025 0.05

SPE 0 0.05 0.1

The fuzzy inoputs in are evaluated against a fuzzy rule base, which has five possible

outputs. These are quality is Very Good (VG), Good (G), Medium (M), Poor (P)

and Very Poor (VP). The rule base is shown in Table 6.1 and are evaluated by

calculation of a root mean square for each of the possible outcomes.



170

Figure 6.13:2T and SPE Fuzzy Inputs

6.3.7 Estimation of the Actuator Quality Index

Figure (6.3) show the fuzzy output function which converts the outputs of the fuzzy

rule base into a useful measure of actuator process quality. For the three cases used

in this demonstration the results of the quality estimation are presented in Figure

(6.14). The nominal case which was tested, where the actuator is in a healthy

working condition with no lubrication removed, reflected a strong positive quality

index within the Very Good → Good regions. Fault case 1 reflected the case where



171

paartial removal of the actuators lubricant was made, resulting in a decreased quality

index which transcended the Good → Medium output regions. For fault case 2

which had significant removal of lubricant and the addition of abrasive material, the

estimation of the quality further reduces extending into the Poor → Very Poor

quality regions.

Figure 6.14: Quality Estimation - Nominal, FaultCase 1, Fault Case 2

The reduction of quality index as the implemented fault condirtion is increased

illusrates the potential for using the PCA and fuzzy logic approach to assigning a

realistic value for the global quality index for the actuator process quality.

6.4 Experimental Testing Limitations

The experimental tests presented in this chapter have illustrated the potential for the

use of PCA combined with a fuzzy logic knowledge base to detect the presence of



172

lubrication loss within an actuator system. there are however several limitations to the

procedure presented, which are as follows:

1) The actuator was not considered under any loading conditions - differing loads may

affect the operation of the actuator, so the diagnostic results under an unloaded

actuator may not be representative of a loaded actuator.

2) Considerations for the actuators operating environment, for example the actuator

has only been tested at room temperature and viscosity of lubrication is temperature

dependant and tests would need to be conducted on the range of operating

temperatures in order to assess the robustness of the fuzzy logic membership functions

and PCA performance statistics.

3) The experiment has only considered the issue of lubrication loss at three extreme

intervals, i.e. no loss, full loss, full loss plus abrasive material. This has limited the

assessment of the sensitivity of the approach and further testing would be required

which takes into account gradual lubrication loss or other fault development.

4) Only one fault condition has been considered, for full acceptance of the proposed

approaches experimental testing on a range of electrical and mechanical faults are

required.

5) With the current bench top actuator setup, it is not possible to up scale the

experimental results to the actuator system on the aircraft landing gears. A modified

rig is therefore required which is more representative of the landing gear application

so that experimental results can be upscaled.

6) it is very difficult to assign a numerical measure representing the severity of an

implemented mechanical fault. This affects the repeatability of the experiments as it

is very difficult to reproduce the same fault at the exact same magnitude.



173

6.5 Practical Considerations

One of the limitations of principle component analysis is that most aerospace

actuators operate under differing environmental conditions and modes. Therefore

conventional principle component analysis as proposed may produce false alarms

and the possibility of inaccurate results. Transitions from one operating mode to

another can have the effect of breaking the correlation between variables. Also the

loading of the landing gear actuator varies during a cycle. This may lead also lead to

inaccuracies with the approach.

There are several solutions to these issues:

1. Generate a PCA model for each operating mode or possible flight condition

2. Update the PCA model to reflect the changes in the operating modes

3. Develop PCA models to account for all operational changes

The advantages of the outlined approach as a practical solution are as follows:

Non-intrusive method with no requirement for additional sensing equipment

Reductions in data set size

The fuzzy approach allows a transparent progression from the initial data to

the final output. With decisions traceable due to the nature of a fuzzy IF-

THEN rule base.

The nature of rule bases means that in the presence of new information they

can easily be updated.



174

6.6 Conclusions

It is known that a strong correlation between the actuators process data will exist

under nominal operating conditions indicating a high degree of data redundancy.

The actuator process signals can all be used individually to generate performance

statistics, but the high degree of redundancy indicates that the same information can

be achieved through reductions in the data set. Based upon healthy data it is

proposed that a nominal PCA model is generated by selecting the number of

principle components which capture the largest amount of variability within the

data. The data is then projected onto these components reducing the dimensions of

the data set.

When suspected faulty process data is available, this data is projected onto the

nominal principle component model and two performance statistics are generated

know as the squared prediction error and a 2T - Hotellings statistic. These statistics

are the sole information which is then used for determining a measure of actuator

quality, the advantage of this approach over conventional statistical techniques on

each set of actuator process data is reduced processing and data size. It is proposed

that the use of fuzzy logic classification applied to the performance statistics is an

adequate and relatively simple, transparent way of inferring the process quality. The

inference process follows a three stage fuzzy process which includes the

fuzzification of the statistics inputs, evaluation of the fuzzy inputs against a fuzzy

rule base and finally the defuzzification of the outputs from the fuzzy rule base. The

application of the fuzzy logic approach provides the user of the monitoring system a

crisp numerical output describing the actuators process quality for use in aiding in

maintenance decisions.

The approach aims at focusing on what the system should do rather than trying to

understand how it works. The approach has the advantage that allows the

concentration on solving the problem rather than trying to model the system, or fault

relationships mathematically. This almost invariably leads to quicker, cheaper

solutions. The effectiveness of the methodology has been demonstrated through



175

experimental procedures utilising a bench top actuator with varying degrees of

lubrication removed. The demonstration highlights that with correct tuning and

classification, an overall index representing the actuators process quality can be

obtained with relative ease.

6.7 References

Bartys, M., Koscielny, J., P. Rzepiejewski, P. (2005), 'Fuzzy logic application for fault

isolation of actuators'. Computer Assisted Mechanics and Engineering Sciences, Vol

12, No. 2-3, p89.

Jackson, J., (2003), ‘A users guide to principle components’, Wiley.

Kambhatla, N., Todd K., Leen, T., (1997), ‘Dimension Reduction by Local Principal

Component Analysis’, Neural Computation, Vol. 9, No. 7, pp. 1493-1516.

Leekwijick, W. and Kerre, E. (1999), ‘Defuzzification: criteria and classification',

Fuzzy Sets and Systems, Vol. 108, pp. 159-178.

Palma, L., Coito, F., Silva, R. (2005), ‘A combined approach to fault diagnosis based

on principle component analysis and influence matrix’, Faro, Portugal.

Phillip, P., Diston, D., (2010b) ‘An Intelligent Health Monitoring Framework for a

Motor-Driven Actuator’, Proceedings for The 7th International Conference on

Condition Monitoring and Machine Failure Technologies, Edinburgh, UK



176



177

C h a p t e r 7 : F o r m u l i s a t i o n o f a P r o p o s e d A c t u a t o r

H e a l t h M o n i t o r i n g A l g o r i t h m

7.1 Introduction

As part of the output of the monitoring system the framework described in the

previous chapter has proposed that information regarding component faults should be

provided. A model-based diagnostics algorithm utilising evidential reasoning aimed

at achieving this (Phillips et al. 2010a) is therefore proposed. The approach utilises

the concept that if an accurate model is known then faults can be diagnosed through

the estimation of fault symptoms, in the form of physical parameters and signal

offsets. These symptoms are then combined using the Dempster Shafer theory of

evidence. The result is a ranking system of probable faults based upon a measure of

belief and plausibility, representing a confidence metric in the diagnostics.

7.2 Fault Diagnostics

7.2.1 Model-Based Fault Monitoring

Fault diagnostic processes based upon models rely on the ability of the system to

make a comparison between what the true system observes and that generated by a

system model. This relies heavily on the availability of an accurate model which can

be reliably tracked. This requires a definition of what is normal for that system, such

as what are the nominal operating parameters and what are the acceptable parameter

tolerances.

The basis of using a physical actuator model is that faults within a process are

indicated by internal non-measurable state variables. These variables are represented

by either the systems physical parameters which are directly affected by the health of

the actuator, or by offsets in the parity of the response inputs and inputs and outputs

described by the model. These offsets are good indications of faults within the

feedback sensors. Through experimental testing and the use of classification

algorithms, relationships between these variables (fault symptoms) and faults can then

be established (Ji and Bals 2009). Once this relationship is known, as these symptoms



178

are derived, they can be automatically mapped to corresponding fault cases. Potential

fault candidates would then be ranked in order of likelihood of occurrence, severity or

by a measure of believability (Phillips et al 2010b). Figure (7.1) illustrates this

approach.

Figure7.1: Model Based Fault Diagnostics Scheme

Control sensors are an essential part of the overall actuator system and are the core

source of data for the proposed health monitoring system. Therefore sensor fault

detection/diagnosis must also be included in the design. The most appropriate model-

based approach to sensor fault detection is to utilise a set of residuals derived from the

parity between the actuators governing equations.

7.2.2 Formulating Parity Relations

Prior to formulating a set of parity equations two vector sets of parameters must be

specified. The fault to be diagnosed denoted as vector Tnaaa ,, 21A and a

set of the process measurements which are available denoted as the

vector Tkmmm ,, 21M . The fault set A actually includes information

describing sensor failures, actuator failures, external disturbances, degradation of

equipment etc. Following on from the definition of these two vectors a set of parity

relations can be expressed as:



179

mknm

jknj

kn

kn

rmmmmaaaac

rmmmmaaaac

rmmmmaaaac

rmmmmaaaac

),,,,,,,,,(

),,,,,,,,,(

),,,,,,,,,(

),,,,,,,,,(

321321

321321

23213212

13213211

RMAC ),( Equation 7.1

These equations are typically a set of nonlinear relations which under nominal

operating conditions the RHS is equal to zero 021 mrrr for

Tsn

sss aaa ,, 21A where the superscript s denotes the nominal steady state

allowing a set of linear algebraic equations too be derived as:

mmnmnimimm

jjnjnijijj

nnii

nnii

rkapapapap

rkapapapap

rkapapapap

rkapapapap

2211

2211

2222222121

1111212111

Or RKAP . Equation 7.2

It should be noticed that from the above set of equations that iji ap , and jk are all

functions of process measurements and the system parameters. Here jip can be

viewed as the sensitivity to the ith fault ( ia ) with respect to the jth parity equation at a

nominal steady state. Mathematically this can be described as:

si

j

jia

cp

Equation 7.3



180

Again it must be emphasised that at nominal steady-state conditions the residuals

generated from the parity equations are zero. That is when nominal steady-state

values are substituted for the fault set sii aa and process measurements

sii mm and the equations yields zero residuals:

0)()( sss mkamp Equation 7.4

When a fault occurs isii aaa * , the fault leads to a new set of process

measurements M . However, at this point no knowledge of which possible change

constituting as the fault is known. Substituting sA and M into the parity relations

leading to an inconsistency in the relations:

0RMKAMP )()( * Equation 7.5

For the case where there is a failure of a sensor and ia is a process measurement, then

the explanation is slightly different. The correct process measurement is

imeasij aaa ,* Equation 7.2

Where measia , is the measurement reading and ia represents the sensor bias. Actuator

and sensor faults are not the only source of errors present in such residuals. In order to

achieve robustness in any fault diagnosis provisions must be made for the electing of

satisfactory violations in the parity relations.

7.2.3 Defining Residual Thresholds

Most diagnostic systems based upon model-based techniques define tolerance

thresholds solely for each parity equation. Problems can arise in the diagnostic

resolution with this approach, for example, a fixed threshold may be too large to be

violated by one fault and too small for another. A natural approach would be to define



181

thresholds not just for each equation but also for each fault possibility. For example, a

system with m parity equations and n faults to be diagnosed would have nm

tolerances denoted as ji where mj ,2,1( and ),2,1 ni . Therefore a fault of

origin ia would have an upper bound of sia)1( and a lower bound s

ia)1( which

when applied to the jth residual becomes:

0),,,,)1(,,,( 121 Msn

si

si

ssj

Hji aaaaac Equation 7.3

),,,,)1(,,,(0 121 Msn

si

si

ssj

Lji aaaaac Equation 7.4

It should be noted that tolerances for both an upper and lower bound should be

defined but that these do not necessarily have to be the same.

Lji

Hji Equation 7.5

One of the characteristics of start/stop motor driven actuation is that high voltage

spikes can occur during the initial actuator start up and at the stopping phases. The

nature of the actuators varying operation regime and environmental changes mean that

the magnitude of these spikes will not be constant at each actuator cycle. Even though

snubbing systems are designed to limit these voltage spikes some spiking is still likely

to occur. The starting and stopping phases are therefore nonlinear and will be

translated into the residuals. Fixed constant thresholds therefore would immediately

be violated at the start and stop phases of the operation, triggering alarms

unnecessarily. Residuals could be enhanced through effective modelling of these

stages, to stabilise the process of change detection. However, this task can be

simplified by designing thresholds to take into consideration the sensitivity of the

residual with respect to the symptoms creating a threshold which will encompass all

stages of the systems nominal operation. This is demonstrated in Figure (7.2).



182

Figure 7.2: Example of Upper and Lower Thresholds

7.2.4 Traditional Threshold Evaluation

Evaluation of residual violation with respect to a threshold has traditionally been

achieved through the use of Boolean logic. In the traditional approach an incidence

matrix consisting of a set of 0’s and 1’s at given positions is used as given in Equation

7.10. The rows of the matrix represent individual residuals and the columns represent

individual faults.

mnmm

n

21

11211

Λ Equation 7.10

Where 1ji , if the ith fault affects the jth residual and 0ji if the ith fault does not

affect the jth residual. This matrix in effect provides a distinct model for each fault,

which can be termed the fault signature. The task of traditional Boolean logical based

fault diagnostics therefore is to attempt to reconstruct these fault signatures. This is

simply done by evaluating the residual against the threshold using the criteria given in

Equations 7.21 and 7.22 and comparing to the model matrix in Equation 7.10.

ijjr , , 1, ij Equation 7.11



183

ijjr , , 0, ij Equation 7.12

Thus, under the ideal circumstances the signature of a fault should be identical to the

respective column of the model matrix. This implies that for each fault/failure to be

detectable, no column of the incidence matrix should contain only zero elements, and

for each signature to be unique all of the columns must be different. There are three

different types of incidence matrix as illustrated in Figure (7.4).

1011

0111

1101

1011

0111

1110

1101

1011

0111

(a) Not isolable (b) Deterministically isolable (c) Statistically isolable

Figure 7.4: The Various Forms of a Fault Incidence

Matrix

The first example is non-isolable, which means that there are multiple identical

columns, making reliable fault diagnosis based on this approach impossible. The

second is known as deterministically isolable where all of the respective columns are

different allowing for effective fault diagnosis, providing that there is no misfiring of

the rules in Equations 7.11 and 7.12. For example, a medium sized fault may occur

and cause the firing of a matrix element whilst others that should fire are not. The

resulting signature is a degraded version for the respective column (with some 1’s

replaced by 0’s). This partial firing therefore leads to a mis-isolation of faults. There

is however a way in which this can be avoided to obtain a statistically isolable

incidence matrix. The incidence matrix must be constructed in such a fashion that no

column can be obtained from any other column through degradation. These structures

take a column canonical form, where each column has the same number of zeros in a

different pattern.



184

The use of this traditional Boolean method can provide undesirable outcomes in the

diagnostic results. Frequent small deviations in the residuals about a threshold point

will lead to diagnostic instabilities. This is possible if there are modelling errors or

high levels of noise. This affect is illustrated in Figure (7.5). Ideally incremental

changes in the residuals should be accompanied by incremental alterations within the

diagnostic outputs (Juricic et al. 2001).

Figure 7.5: Boolean Logic Diagnostic Instability

If the use of standard Boolean classification of constraints is used to express belief

in the presence of a fault; the diagnoses will always be unstable, regardless of

statistical criterion used to classify the constraints. To avoid this it is desirable to

smooth the threshold function which is best achieved through the use of approximate

reasoning techniques. Through these a residual is no longer qualified as zero (0) or

non-zero (1) but is assigned a degree of being non-zero in the range [0, 1]. An

appropriate function to provide this assignment would be a sigmoidal function

(Juricic et al. 2001) as follows:

2

)/)1((1

1)(

ji

j

ijr

am Equation 7.13



185

This describes a sigmoid function with and representing smoothing parameters to

be selected depending on the system design and is illustrated in Figure (7.6) along

with a more stable diagnostic result applied to Figure (7.5).

Figure 7.6: Sigmoid Smoothing Function and Stable

Diagnostic Results

The result of this is used to represent an assignment known as the Basic Probability

Assignment (BPA) of the fault assumption occurring. A BPA expresses a kind of

human’s judgment on the degree in which the assumption has caused a violation of

the threshold. It is then necessary to interpret these BPM into meaningful information

regarding the evidence in which particular fault has occurred.

7.2.5 Evidential Reasoning

The Dempster-Shafer theory of evidence is a mathematical technique which takes

evidences of events and combines them in order to calculate the belief of an event

occurring. Dempster-Shafer theory is important and useful in this application as it

takes into account what is unknown and what is precisely known. In the theory,

evidence of the likelihood of an occurrence is represented by the interval

)(),( AplABel which is a subset of the interval 1,0 . The parameter

)(ABel represents the evidential support for the proposition A, whilst the parameter



186

)(Apl represents the plausibility of the proposition. An interpretation of this is the

probability of A is bounded by defined as )(ABel and )(Apl .

)()()( AplAPABel Equation 7.14

The uncertainty in A is given by

)()()( ABelAplAu Equation 7.15

In the Dempster Shafer theory hypotheses are represented by a subset of all relevant

hypotheses. This is known as the frame of discernment . In the context of fault

diagnostics the frame of discernment will contain all faults and the state of normal

operation. It should be noted that the state of normal operation is equal to the negation

of all other faults.

The evidential intervals )(),( AplABel are derived through the manipulation of the

BPA which distributes a unit of belief over the set of hypotheses. In effect the BPA

of A represented by )(Am is the portion of belief assigned to the hypothesis A , where

A is any subset of . If it is not possible to assign a portion of belief to any particular

subset of based upon the available evidence then this residual belief is assigned

directly to . This has the effect of introducing uncertainty into the system. The

formulation of the theory of evidence is as follows.

The belief in the hypothesis A is the sum of all of the BPA assigned to A and all of

the subsets of A :

)()( iAmABel , AAi Equation 7.16

The plausibility of the proposition is given by:



187

)(1)( ii BmApl , 'ABi Equation 7.17

The set 'A is the set with all of the elements of A removed.

For a given application there will be several sources of information which will

contribute various degrees of belief to a given proposition under a common frame of

discernment. In the current model based fault diagnosis approach this is highlighted

by considering that multiple residuals are sensitive to the common faults. Dempster

Shafer provides an efficient rule of combination to deal with this. This area of the

reasoning process is discussed in later sections.

7.2.6 Introducing Residual Uncertainty

As with all systems a certain degree of uncertainty will be inherently visible in the

residuals, due to noise, modelling or measurement errors, even under fault free

conditions. The assignment of appropriate belief masses therefore must take into

consideration this uncertainty, and Equation 7.13 is modified as follows:

2

)/)1((1

11)(

ji

j

jiijr

am Equation 7.18

Here the term ji1 has permitted the introduction of residual errors into the belief

function when the reliability of the information derived from the constraint is less than

unity. The evidential intervals based upon Equation 7.14 are now described by

Equations 7.19 and 7.20. Where 0ji ,

ji , ji represents the conditions of no threshold,

upper threshold and lower threshold breach respectively, with regards to the jth

residual and ith fault.

iiiji mm , , iij ,0, iiiij mm 1,10 for 0jr Equation 7.19



188

iji ,0 , iiiij mm , , iiiij mm 1,10 for 0jr Equation 7.20

The interpretation of the above intervals can be visualised in Figure (7.7) and is

synonymous with fuzzy logic membership functions (Bartys et al. 2005).

Figure 7.7: Application Evidential Intervals

7.2.7 Combining Multiple Evidential Intervals

At this stage each residual will have provided a set of intervals for each possible fault

symptom. It is clear that evidence for common symptoms will be present in multiple

residuals. The Dempster-Shafer theory of evidence provides a combination rule given

in Equation 7.31 and 7.32 to combine the available evidences. Consider two sources

of BPA 1m and 2m , then they can be combined according to the following rule:

K

BmAmCm

ji

1

)(.)()(

21, CBA ji Equation 7.21

)().( 21 ji BmAmK , ji BA Equation 7.22



189

The factor )1( K is a normalization factor that keeps the total belief equal to unity.

The adjustment is necessary by the presence of propositions in 1m and 2m whose

intersection is empty, resulting in the donation of a portion of belief to the empty set

. In terms of decision making, consider a fault symptom Ff , the evidential interval

)(),( FF fplfBel represents the truth of the proposition indicated by the following

examples (Kramer 1987).

]1,0[Ff → No information exists regarding Ff

]0,0[Ff → Ff is false

]1,1[Ff → Ff is true

]1,3.0[Ff → The evidence partially supports Ff

]7.0,0[Ff → The evidence partially supports the negation of Ff

]7.0,3.0[Ff → The evidence partially supports both the presence and negation of

Ff .The probability of Ff is between 0.3 and 0.7 with uncertainty

0.4.

7.2.8 Combining Rules for Comprehensive Diagnostics

After the belief intervals for the propositions have been found and the uncertainties in

all propositions in the frame of discernment then the diagnostic results can be

determined through the application of a few simple decision rules:

Rule 1: )(max)( ii

a fBelfBel

Rule 2: 0,)()(,)()( mfBelfBelfBel aia

Rule 3: 0,)( m



190

Rule 1 indicates that the result of the diagnostics must be the proposition af with

the biggest proportion of belief. Rule 2 indicates that the possibility of the

diagnostic result must be bigger than all o the propositions by an amount . Rule 3

indicates that the uncertainty in the evidence for a proposition must be less than a

value . The values and are too be chosen according to the practical

application. If the diagnostic result af cannot be ascertained then a new frame of

discernment should be constructed or more evidence bodies should be added to the

computation (Yang and Wu 2007).

7.3 Advantages of the Proposed Methodology

The model-based diagnostics approach has the following advantages:

Requires no additional sensing equipment acting only on dynamic data already

available as part of the actuator control.

The actuator does not require complicated modelling, making a model-based

diagnostic approach ideal.

The combination of data into parity residuals reduces the number of signals for

analysis.

Faults are traced back to meaningful physical parameters, which can often be

more easily understood than many of the subjective features available from

alternative diagnostic techniques.

The approach allows the introduction of uncertainty into the system which

relates to measurement or modelling errors.



191

The final output of the approach provides the user with information on the

extent to which a fault has occurred coupled with how believable the evidence

is. The difference between the two provides a measure of uncertainty useful in

decision making.

7.4 Conclusions

The second part of the health monitoring algorithm forms the basis of a model based

fault diagnostics process. This approach requires the use of an accurate system

model and combining the actuators dynamic process data in the form of a set of

parity equations which are used to form a set of diagnostic residuals, sensitive to

fault occurrences. The residuals are naturally designed to remain close to zero in

fault-free conditions. In the presence of a fault the residuals will deviate either

positively or negatively.

Unlike traditional parity equation based fault diagnostics, the residuals are not

assessed through the use of ‘one threshold per residual’. Each residual will contain

information regarding multiple fault symptoms. An untraditional approach is therefore

proposed utilising multiple thresholds for each residual each representing an

individual fault. The nature of the actuators operation means that there are several

nonlinear regions in the dynamic al data, most notably at the start and stopping phases.

Fixed static thresholds cannot be used in this case as there will most likely be an

instant threshold violation in these regions. This is overcome by designing thresholds

by utilising the model equations sensitivities to faults. This creates a threshold which

is not fixed but follows the dynamic profiles of the data.

Threshold violations are traditionally detected through Boolean Logic. The

shortcomings of which is that in the presence of noise and modelling errors Boolean

logic based diagnostics suffer from instability. Also there is no indication to the level

of severity in the threshold violation. This is overcome by utilising approximate

reasoning. This approach provides a measure on the extent to which a threshold has

been breached allowing estimation to the magnitude of the fault occurrence. With this



192

approach incremental changes in the residuals about a threshold will result in

incremental changes to the diagnostic results.

The Dempster Shafer theory of evidence is used to produce intervals based on

symptoms categorised as High, Low or Okay. The interval provides a measure of

belief in the symptom occurring and a measure of how plausible the reasoning is. The

probability of the symptom is bounded by the belief and plausibility and the difference

magnitude of the space between the bounds represents the uncertainty in the

diagnostics. Each residual includes information regarding multiple symptoms, many

of which will be common throughout the set of residuals. Dempster Shafers theory of

evidence provides a rule of combination which is utilised to combine these common

residuals, leaving an individual interval for the values of High, Low, and Okay for

each symptom. These intervals are then compared to a set of three diagnostic

performance rules before a diagnostic certainty is made.

7.5 References

Ji, Y., Bals, J., (2009), ‘Application of Model Detection Techniques to Health

Monitoring for the Electrical Network of More Electric Aircraft’, Proceedings of the

World Congress on Engineering and Computer Science, San Francisco, USA.

Juricic, D., O. Moseler, and A. Rakar, Model-based condition monitoring of an

actuator system driven by a brushless DC motor. Control Engineering Practice, 2001.

9(5): p. 545-554.

Kramer, M., (1987), 'Malfunction diagnosis using quantitative models with non-

boolean reasoning in expert systems'. AIChE Journal, Vol. 33, No. 1, pp. 130-140

Phillips, P., Diston, D., Payne, J., Pandya, S. (2010a), 'Evidential Reasoning Applied

to Model-Based Diagnostics in Landing gear Actuators' in Machine Failure and

Prevention Technology Conference. AL, USA.



193

Phillip, P., Diston, D., (2010b) ‘An Intelligent Health Monitoring Framework for a

Motor-Driven Actuator’, Proceedings for The 7th International Conference on

Condition Monitoring and Machine Failure Technologies, Edinburgh, UK



194

C h a p t e r 8 : S y s t e m M o d e l l i n g , S i m u l a t i o n a n dD i a g n o s t i c s D e m o n s t r a t i o n

8.1 Introduction

Chapter seven provided the formulisation of a proposed model-based approach to

actuator fault diagnostics. This chapter seeks to justify the applicability of this

algorithm and demonstrate its usage and potential effectiveness. The landing gear

retraction actuators focused on in this research are currently still under development

and have not yet been put into production for operational usage or testing. A

consequence of which is that there does not exist any operational data on the actuator

system whilst a fault is developing. The economic costs associated with the actuator

development would also prevent artificial faults being introduced. This inevitably

creates challenges in the development of a health monitoring system. In the absence

of a landing gear experimental testing rig this demonstration must be sought through

actuator modelling and simulation.

In order to keep the demonstration applicable to the specific landing gear application,

a representative model of the actuator system has been developed and its performance

simulated within the MATLAB/SIMULINK environment. Though the use of expert

knowledge on how faults manifest themselves within changes to model parameters,

the use of a simulated model can be used to generate representative data relative to

faulty behaviour.

Based upon the actuator model equations a set of nominal fault sensitive residuals are

derived. Faults are simulated through a range of physical parameter changes and

signal offsets based in accordance with information obtained from published

literature. Threshold levels are identified based upon the effects model changes have

upon the actuator cycle time, with upper and lower boundary cycle times in

accordance with the actuator developer’s requirements. The simulation is run for the

case of changes to the frictional constant and illustrates how combinations of

additional evidence can aid in clarifying an uncertain diagnostic result.



195

8.2 Modelling the landing Gear Actuator

Figure (8.1) provides a schematic of an actuator arrangement identifying the key

actuator features.

Figure 8.1: An Example of a Typical Screw Actuator

Assembly*

The actuator model has two key elements, the electrical subsystem model and the

screw and gear mechanical subsystems. The electrical component of a DC motor can

be described by the following:

dt

tdiLtKtRitV rb

)()()()( Equation 8.1

Where i (t) is the motor current, R is the internal resistance, L is the inductance, bK is

the motors back emf constant, V(t) is the input signal and )(tr is the angular velocity

of the motors rotor. The magnetic field caused by stator currents will result in a torque

generated through the electrical properties of the motor described by:

* Diagram modified from MecVel ALI3 24VDC single phase motor operated leadscrew actuator data sheet



196

iTelec tiKt ).()( Equation 8.2

The torque constant TK in an ideal motor will tend to be bT KK . The term i

represents the efficiency of the motors current controller. Through the use of

conservation of momentum the torque will be described as:

)()()()( tttt lossesloadelec Equation 8.3

The torque load , results from work being done against the systems mechanical load

and losses through friction (coulomb and viscous) within the system. Equation 8.3 is

therefore described as:

)()()()( tBttiKtJ rloadbr Equation 8.4

Where J is the inertia of the system which is concentrated at the motor and B is given

as the friction constant.

For a rigid screw without backlash, the compatibility condition between the rotational

position of the gear )(tr and the rotational position of the ballscrew )(tbs is given

by:

rr t )( Equation 8.5

)(1

)( tN

t bsr Equation 8.6

Where N is the gear ratio and the linear distance moved by the nut in one complete

revolution of the screw is the lead l , which can be described as the effective radius of

the screw as:



197

2

lR Equation 8.7

The total linear distance moved in one actuator cycle is therefore described as:

)(.2

.)(.)( t

N

ltRtx rbs

Equation 8.9

The load load can be calculated through measurement of the actuator load F as:

screwgear

loadN

FRt

..

.)( Equation 8.10

8.3 Simulation

8.3.1 Overview of the Actuator SIMULINK Model

The model of the actuator was designed within the SIMULINK environment, with

MATLAB program files containing the SIMULINK model parameters. The

simulation of the actuator was achieved by running a MATLAB run file which

requests the user select the following options:

1) Initial start position of the actuator (i.e. extended or retracted)

2) Actuator load case

3) Damped or un-damped performance

The SIMULINK model for the main actuator model (Figure 8.2) is automatically run

for one actuator cycle and the user is requested to select the performance response

data to save as a data file.



198

max demand

10V

xmax

Kt

torque constant

at 20 deg C (Nm/A)

1/N 1/effscrew

rollerscrew

efficiency

N

motor:rollerscrew

1000

m to mm

init_pos_m

1/effgear

gear

efficiency

Kb

emf constant

eff_curr_cont

s

1.0e-03s+1

differentiator

ViLim

>=

Step

R

R

1

2

*3,

1

J.s+B

Load Transfer Fcn

Load Look-Up

Table

250knt_1.3g

Load Look-Up

Table

170knt_1.1g

Load Look-Up

Table

0knt_1g

K1V

1

s

H2

-K-

Vb

Vi

im

Current Controller

damped_mode

load_case_selector

0

1/R

1/N

1/(motor:rollerscrew)

back emf (V)

torque

current (A)

motor angle

(rad)

rollerscrew

angle (rad) xm (m)

xm (mm)

(Position)

Act_Load

(N)

torque

(N/m)

torque

(N/m)

initial

position (m)

actuator rate (m/s)

rollerscrew

rotation speed

(rad/s)

motor_speed

(rad/s)

motor_speed

(rad/s)

motor speed

(rad/s)

input

v oltage

Rate

Figure 8.2: SIMULINK Block Diagram of the MainActuator Model

The actuator simulation provides the option to be run with or without actuator

damping. The case where damping is included is representative of the actuator

deployed with snubbing to reduce harsh voltage spikes at the transition phases of the

actuator cycle (start/stop phases). This is achieved in the simulation through the

addition of an additional resistance as indicated in the current controller subsystem as

the variable Rm1 shown in Figure (8.3).



199

1

im

VsLim

>=

>=

Rm1

Rm

K3

1

s

H3

damped_mode

damped_mode

0

1/Lm

2

Vi

1

Vb

current

resistance

Figure 8.3: SIMULINK Block Diagram of the MotorCurrent Controller

8.3.2 Actuator Loading

Landing gear actuator must be able to perform to satisfactory criteria under a variety

of loading conditions, and during the initial modelling and design stages a variety of

load cases are identified. The load cases are integrated into the actuator simulation

model through the use of three look up tables supplied by Messier-Dowty. The

lookup tables contain a range of values which represent combined static and

aerodynamic loading of an A320 landing gear which have been generated from

landing gear kinematic models. The three possible load cases are defined as:

Load Case 1: 0 knots, 1 g

Load Case 2: 170 knots, 1.1g

Load Case 3: 250 knots, 1.3g

Load case 1 represents the actuator on the bench top with no external aerodynamic

forces acting upon it. Load case 2 represents the nominal operation during flight,

whilst load case 3 represents an extreme operating scenario. In terms of simulation,



200

the lookup tables are used to provide the correct loading torque for feedback into the

electrical motor model. This is achieved by taking the measured actuator stroke

length, at a given time interval, and then interpolating the data provided in the look up

table to obtain the correct loading value for the actuator at that current position.

The use of these validated loading lookup tables reduces the need for modelling and

simulation of the landing gear kinematics as part of this research, reducing the

complexity of the model for simulation of the health monitoring algorithm. This

loading is then converted into the motor loading torque using Equation 8.10.

When the landing gears are not deployed, the actuator is at a fully extended stroke

position. As the gears are then extended through to deployment, the actuator

essentially pulls against a lever arm acting about a pivot lowering the gears.

Conversely, when the landing gear is retracted the actuator extends and hence pushes

against the lever arm.

When the landing gears are being extended, the applied loading drives the landing

gear motion, pushing the landing gears down, with higher loads resulting in faster

cycle times. In the case of a retraction cycle, then the applied load works against the

landing gears meaning that higher loads will result in slower cycle times.

8.3.3 Actuator Performance Simulation

Table 8.1 provides the nominal actuator physical data parameters used within the

simulation model



201

Table 8.1: Actuator Nominal Parameters

Parameter Value Units

Gearbox ratio 54.7Screw lead 6 MmGear efficiency 90%Motor Control efficiency 80%Screw efficiency 80%Inductance 3102.1 H

Supply voltage 540 VMotor resistance 0.395 Ω EMF constant 0.221 11.. sradVTorque constant 0.221 1. ANmMotor inertia 31002.1 2.mkgMotor viscous load 41021.2 1... radsmNActuator full extension 0.234 M

The actuator responses for load cases 2 and 3 are provided in Figure (8.4) for an

extension of the landing gear with dampening. As the responses show for the case of

the actuator extension, the higher load case result in reduced cycle times due to the

load helping to drive the motion of the landing gears.

Controlling the cycle speed of the landing gears is an important factor in the actuator

design. Rapid retraction can cause control difficulties at the end of the cycle risking

potential damage which caused to the landing gear structures or landing gear bay.

Conversely if the actuator draws significantly reduced power or retracts at an overly

slow speed, there is the potential that the actuator will be unable to overcome the

necessary loads for full retraction.



202

Figure 8.4: Simulated Actuator Responses for Load

Cases 1 and 2

8.4 Fault Cases

For the purpose of simulating faults within the actuator model information published

by other authors working in similar areas is utilised. There is a wealth of published

work showing methods of simulating faults into an actuator model, but for this

demonstration actuator fault classification work performed and published by Impact



203

Technologies, LLC has been used. Byington and Stoetling (2004) identify after

substantial experimental testing four key parameters of interest for fault diagnostics of

electrical actuators. Examples of these are identified as frictional constants, local gear

stiffness, torque constant and motor temperature which can all be matched to the

following potential faults.

Gear Slippage

Decrease in local gear stiffness

Small increases in frictional damping coefficient

Bearing Seizure

Large increases in frictional coefficient

Small increases in motor temp

Motor Failure

Decreases in torque constant

Large increases in motor temp

It has also been acknowledged that simulation of sensor faults can be achieved

through the introduction of a positive or negative bias within a particular signal of

interest (Byington et al. 2003). In the case of this work the following faults will be

considered:

Fault 1 = Changes in the torque constant

Fault 2 = Changes in the motor resistance

Fault 3 = Changes in the frictional coefficient

Fault 4 = Bias in the speed sensor signal

Fault 5 = Bias in the current sensor signal



204

It should be noted that this is a simplified example of real world faults. For example,

even though a bearing fault is identifiable as significant changes to the frictional

coefficient, other small parameter changes will also occur. This highlights the

importance of robust fault - symptom classification for successful diagnostic

capability.

8.5 Actuator Performance Assessment

In this application, the significant performance characteristics of the actuator system

are defined by the actuator cycle time (Phillips et al. 2010). This upper and lower

boundary conditions applied to the actuator cycle time are used for determining the

fault thresholds. For the case of the actuator operating in a nominal operating mode,

simulated as an extension cycle under load case 2, the cycle time is 11.7 seconds and

the performance boundaries are defined as ± 30% of this value. Examples of these

performance requirements are shown in Figure (8.5) as changes in faults 1 and 2.

Figure 8.5: Example of the Effects on the Actuator

Cycle for Simulation of Faults 1 and 2

In the same manner the upper and lower boundaries for all five fault conditions are

computed and provided in Table 8.2



205

Table 8.2: Upper and Lower Boundaries for Faults

Fault 1

TK

Fault 2

mR

Fault 3

B

Fault 4(S)

biasii

Fault 5

biasrr ,

Upper 0.32 0.5 0.7 0.35 0.25

Lower -0.35 -0.25 0.4 0.32 -0.25

8.6 Parity residuals

A set of four residuals are derived directly from the equations describing the actuator

model given in section 8.2. These are structured to be insensitive to certain

parameters or signals.

Residual 1:

0)(

)()(.)()(1 dt

tdiLtKtiRmtUtr mrb Equation 8.11

Residual 2:

0)()(.)(..)(

)(2 ttBtiKdt

tdJtr loadrit

r

Equation 8.13

Residual 3:

)()(.)()...(1

)()(.)()(

2

2

3

ttUBtiRBKKK

dt

tdU

dt

tdi

J

LBR

dt

tidL

K

Jtr

loadmitb

b

mmm

b

Equation 8.14



206

Residual 4

)()(.

)(..)(..)(

)(4

ttUR

K

tBR

K

dt

tdi

R

LK

dt

tdJtr

load

m

it

r

m

it

m

mitr

Equation 8.15

In the case of the residuals derived here the following applies:

Residual 1r is independent of the signals load and x

Residual 2r is independent of the signals V and x

Residual 3r is independent of the signals load , x , V and r

Residual 4r is independent of the signals V and x

8.6.1 Nominal Test

Figure (8.6) shows the residual responses under nominal simulation conditions. Even

though the residuals have been designed to reduce to zero in fault free conditions,

there is still some deviation, particularly evident in residuals 2 and 3. The cause of

this can be attributed to the residuals dependency on second order differentials, which

have exaggerated the high order dynamics. The mean of this deviation can be taken as

the error within the residual as described by Equation 7.28.



207

Figure 8.6: Nominal Residual Responses

8.6.2 Residual Fault Sensitivity

For this application the frame of discernment describing the five fault cases is given

as. ,5,4,3,2,1 FFFFF . Each of the four residuals will be affected by

different fault set. The faults which each of the four residuals contribute belief to are

as follows:

)(

)(

,5,3,1)(

,5,4,2)(

44

33

222

111

mr

mr

FFFmr

FFFmr

These represent the total uncertainty in each of the residuals and are calculated using

the OR operator. For example consider 111 ,5,4,2)( FFFmr any deviation in the

first residual is attributed to one of these potential faults, described as:



208

)()()(,5,4,2 1151412154211 mFmFmFmFFFmFFF

The value )( jjm is the mass attributed to modelling errors or residual noise. For

simplicity this has been chosen to take the value of 0.001 for all four residuals.

8.6.3 Combining Residual BPA's

The basic probability assignment for each fault is divided between the three cases of

High (H), Low (L) or Okay (O) as described in Equations 7.29 and 7.30. The

intersections between two evidence sets are shown in Table 8.3 and are combined to

form fault belief intervals using Equation 7.31.

Table 8.3: Intersections Between Two Evidence Sets

1m / 2m μL,O,H, H O L μ

μL,O,H, ,,, LOH H O L µ

H H H H

O O O O

L L L L

μ µ H O L µ

8.7 Diagnostics Algorithm Demonstration

8.7.1 Overview of the Simulation Process

For the simulation of the diagnostics process a MATLAB program was written which

incorporated the actuator SIMULINK model described in section 8.2 along with

additional MATLAB scripts for the diagnostics analysis. Within these MATLAB

scripts the threshold arrays for each of the four residuals, were stored. These had been

previously obtained through simulation of the actuator under the maximum and

minimum conditions illustrated in Table 8.2.



209

For the diagnostics simulation the user is requested to select the specific fault and its

magnitude as a percentage of its nominal value. The program then automatically

simulates the actuator model with the selected parameter/signal offset changes,

generating the necessary residuals and carrying out the evidential reasoning

diagnostics analysis.

For the purpose of demonstrating the diagnostics, changes to Fault 3 (frictional

constant) were made in the range of %40%40 . Basic probabilities are assigned

for each fault condition by automatically comparing the residual responses with the

stored fault threshold arrays resulting in a set of four evidences, relating to the

individual residuals. The calculation of the belief intervals first combined evidences

1m and 2m then individually combined the result with 3m and 4m . The purpose of

which was to illustrate how the diagnostics becomes clearer with reduced uncertainty

with additional evidences. At this point it is worth revisiting the explanation of the

evidential intervals )(),( FF fplfBel as presented in Chapter 7 for the purpose of

interpreting the diagnostics.

]1,0[Ff → No information exists regarding Ff

]0,0[Ff → Ff is false

]1,1[Ff → Ff is true

]1,3.0[Ff → The evidence partially supports Ff

]7.0,0[Ff → The evidence partially supports the negation of Ff

]7.0,3.0[Ff → The evidence partially supports both the presence and negation of

Ff .The probability of Ff is between 0.3 and 0.7 with uncertainty

0.4.



210

8.7.2 Simulation Results

Combining the evidences 1m and 2m shown in Figure (8.7) it is evident that no clear

diagnosis can be determined. Even though there is higher belief in Fault 3 than any of

the other possible faults the levels of uncertainty shown in Figure (8.10) is

significantly high.

Figure 8.7: Belief Intervals for Two Evidence

Combination



211

Combining the results for the two evidence combination with additional evidence 3m ,

the belief interval begins to correctly support the presence of high or low fault 3,

shown in Figure (8.8) with further reductions in the uncertainty shown in Figure

(8.10).

Figure 8.8: Belief Intervals for Three EvidenceCombination



212

Combining additional evidence 4m the uncertainty further reduces and the belief

interval moves further towards ]1,1[B supporting the presence of fault 3, Figure (8.9).

It however can be seen that there is also strong support for the presence of fault 1.

Even though the support for fault 3 is much stronger care would need to be taken with

this diagnosis. In a scenario such as this there is a strong case for employing an on

demand parameter estimation scheme.

Figure 8.9: Belief Intervals for Four EvidenceCombination



213

Figure (8.10) shows the decreasing uncertainty betwen two, three and four evidence

combinations.

Figure 8.10: Diagnostic Uncertainty in Two, Threeand Four Evidence Combination

8.8 Conclusions

In the absence of the ability to perform experimental testing on the landing gear

actuators, the need to demonstrate the potential of the diagnostics approach falls into

the realm of simulation. Through the modelling and simulation of the dynamic

characteristics of the actuator and utilising published expertise in fault manifestation

within physical parameters a variety of fault representative responses can be

simulated. These include changes to the frictional coefficient, torque constants, motor

resistances and signal bias'. Through application of the proposed model-based



214

diagnosis approach to simulation of the actuator it is clearly demonstrated there is the

ability to distinguish between this selection of parameter and signal changes.

Testing the diagnostic ability by simulating a range of negative and positive changes

to the frictional coefficient, which has been identified in the literature as a key

symptom of bearing faults, within a simulation model of the actuator model, verified

by the sponsor’s designers as representative of the current design, it has been shown

that the use of the proposed evidential reasoning process, increases the diagnostic

confidence as more evidence combinations are added.

One of the key issues which is identified is that some faults may affect the residuals in

near identical ways, for example resistance changes are not distinguishable from a

simulated bias in the speed sensor. In these cases further evidence is required, and the

use of an online parameter estimation to distinguish between sensor faults and actuator

faults is proposed.

The use of simulation to demonstrate diagnostics however does have a variety of

limiting factors in algorithm verification. Without the experimental classification of

parameter/signal changes in the presence of faults, it is difficult to accurately simulate

these faulty conditions. Also, full validation of the model would be required against

significant amounts of test data before the actuator process model could be relied

upon.

An approach such as that proposed at this stage modelling errors can only be

estimated. The simulation has assumed that the errors within each of the signals are

common, which in practice will not be the case due to measurement errors inherent in

different sensors. The first two residuals for example are more attuned to the

electrical and mechanical parts of the model; these may have differing modelling

errors affecting the individual residuals. In a similar way some common symptoms

may be stronger for one fault than for another. This insinuates that investigations into

the use of weightings for symptom to fault classification will be required.



215

8.9 References

Byington, C., Safa-Bakhsh, R., Watson, M., Kalgren, P. (2003), 'Metrics evaluation

and tool development for health and Usage Monitoring Technology', in AHS Forum.

Phoenix Arizona.

Byington, C., Watson, M. ,Edwards, D., Stoetling, P. (2004), ‘A model-based

approach to prognostics and health management for flight control actuators’,

Proceedings of the IEE Aerospace Conference,

Phillips, P., Diston, D., Payne, J., Pandya, S. (2010), ' Evidential Reasoning Applied

to Model-Based Diagnostics in Landing gear Actuators', in Machine Failure and

Prevention Technology Conference. AL, USA.



216

C h a p t e r 9 : C o n c l u s i o n s

9.1 Summary

Health monitoring technology is intractably tied up with aerospace maintenance

activities as a whole. The aerospace maintenance industry is currently facing a time of

unprecedented demand for spare parts, complete overhauls and general servicing.

This is therefore putting a strain on overhaul providers, suppliers and Original

Equipment Manufacturers (OEM) which have begun seeking new innovative

maintenance solutions, to meet rising demands and costs. The nature of this therefore

opens up the possibility of integrating health monitoring technologies into unique

customised maintenance solutions and support packages offered to aircraft operators.

It has however been the tradition that there is an inherent secretive nature regarding

aerospace maintenance. For successful tailored maintenance solutions focused upon a

more predictive approach requires a transparent flow of information from the aircraft

operators, aiding in suppliers and manufacturers optimising their spare part inventories

and developing efficient maintenance procedures/schedules.

Aerospace is often considered to be at the forefront of high technology requiring the

strictest of safety standards and criteria. This however is not necessary the case when

it comes to health monitoring technology. Very little is actually documented about

aerospace monitoring research and surprisingly little has been implemented into

aircraft systems. In the UK the railway industry has recognised the need to improve

maintenance practices after a variety of high profile rail disasters. Like aerospace, the

implementation of rail health monitoring is in an infant stage, but there is much more

transparency on the progress of research and the industry practices, likewise in the

power industry.

Many of the approaches illustrated within this research thesis have been used to

certain varying degrees of success individually within a variety of industries and this

research has sought to combine differing techniques together to form a single



217

algorithmic approach. A study of techniques and their applicability to actuation

systems across a range of industries, which included aerospace, rail, automotive and

the power industry has been carried out. The results of which identify that health

monitoring in each of these industries have in some form or another either been

implemented to some extent, or at the very least the desire to implement and develop

them, with early stages of research being evident.

9.2 Research Conclusions

At the start of this thesis a number of clear objectives were defined. In order to

conclude that each of these has been satisfied these can now be revisited.

9.2.1 Objective 1

Assess the current state of the art health monitoring techniques and show that

established techniques exist which are viable for a landing gear actuator application

without impacting upon the weight, volume or complexity of the actuators.

Health monitoring has over the past several decades become a well established

discipline with a multitude of tools and techniques. The state of the art in health

monitoring for industrial applications for example vibration analysis or acoustic

monitoring with advanced signal processing and methods such as wear debris analysis

using optical imaging and fuzzy classification all perform well. However the nature

of the actuator design and operation generates a unique set of challenges in the choice

of actuator monitoring approach. Most monitoring research aimed at linear electrical

actuator technology has converged towards the classification of estimated physical

parameters to underlying faults. This offers several advantages, being that very little

additional equipment is required, there is less dependency on measured data and the

methods can be implemented on line or offline. The downsides include the need for

an accurate process model must be developed and the accuracy of parameter

estimation techniques. The presentation of a full health monitoring technology review

presented to the sponsors as a part deliverable for the EU framework 6 projects



218

TATEM assessed the varying techniques, in terms of hardware requirements, fault

sensitivities and general issues. The review studies confirmed that the most applicable

approach to follow was to consider the use of multiple methodologies including

model-based, performance monitoring and knowledge based methods.

The reasoning behind developing a health monitoring system based upon multiple

techniques is to maximise the amount of data and knowledge available for use. It has

been identified that a fundamental problem which arises when attempting to combine

information from multiple sources is that it will usually be non-commensurate. That

unit scales and time periods will differ. Techniques and frameworks must therefore be

available to combine this data together. The research has made use of the concept of

data fusion and existing frameworks in order to identify the correct methodology for

achieving this. The use of data fusion is gaining popularity within health monitoring

community but its application to actuation systems is still regarded as in its infancy.

9.2.2 Objective 2

Define a systems architectural framework for EMA diagnostics and prognostics, with

identification of key nodes which will enable the following:

Identification of abnormal behaviour

Incorporate performance metrics

Allow analytical and heuristic symptoms to be used effectively alongside

process history, costs and risks.

Be accessible for additional sensor/heuristic data, for health monitoring

purposes, to be incorporated without architectural alterations.

The development of a conceptual framework has the purpose of acting as a template

or set of guidelines in the selection and development of individual algorithms. The

framework presented in this research combines the different levels of the OSA-CBM

health monitoring standards and the hierarchal levels of the JDL data fusion

framework in order to visually identify the important modules required in the system.

The framework design is not application specific and the individual elements have



219

been identified on a generic level. The purpose of such a framework is to ensure that

the health monitoring system development will integrate into the actuator application

rather than trying to force generic health monitoring into the actuator system.

The architectural framework, which does not directly make any commitment to

particular data streams, feature extraction or inference processes, does however

identify certain key areas of knowledge which are of importance. These are the

hierarchical objectives, from the management level through platform, system, and

subsystem and component levels. Without the formation of objectives, assessment of

the health monitoring capability and performance would be meaningless. Identifying

these objective ensure that performance metrics can be identified and incorporated

into the decision making process.

Next it is important that knowledge on the failure of components and their effects on

the overall system is known. Understanding the impact of failure effects it is best

achieved through the construction of FMEA. This heuristic knowledge can then be

used to form failure event trees. Both of these areas have been a key development part

of this research in close collaboration with the sponsors.

The knowledge of how a faulty component affects the system leads onto the notion

that a rule base is an essential part of the systems consequence inference. Rule bases

have come under much criticism due to the number of rules required and their

potential clumsiness. However it should be noted, for any monitoring system which

will provide a consequence of failure, a rule base in some form or another will be

essential. Rule bases also allow for additional heuristic or historical data or

knowledge on new previously unseen events to be easily incorporated into the

monitoring.

The framework descriptions on the fusion processes, identified as data association,

hypothesis generation/evaluation/selection and fault estimation have been used to

select the process by which the diagnostics algorithm has been formulated. The very



220

nature of basing the framework upon data fusion means that it is possible to combine

different information sources or evidences without significant architectural changes.

As long as additional information is aligned to a common reference frame it can be

combined to increase the confidence in the diagnostics reducing uncertainty.

For the introduction of health monitoring into a vehicle platform considerations need

to be made on the deployment strategy. Two separate possibilities for this have been

identified. The first is dependent on the monitoring algorithms and decision support

to be made at a maintenance service bay, where the monitoring software is located.

The second approach would see the health monitoring algorithms embedded onto the

aircraft alongside control algorithms and BIT testing. It is highly likely that for

effective aerospace vehicle monitoring a hybrid of the two deployment strategies

would be a natural approach.

9.2.3 Objective 3

Define and demonstrate a health monitoring algorithm for component level actuator

fault detection and diagnostics.

It has been an essential part of this research project that a close working dialogue with

the sponsoring organisation has been maintained to understand the necessary elements

of actuator design requirements, performance expectations and design restrictions.

Understanding these key factors has been identified as essential for the design of any

application specific aerospace health monitoring system. The health monitoring

algorithm derived in this research incorporates separated fault detection and diagnostic

approaches, with the view that diagnostic computing resources would be used only

depending upon appropriate warnings from the fault detection scheme.

It has been illustrated that to simplify the fault detection task, it would be beneficial to

utilise an estimation of the overall operating performance/quality of the actuator. To

achieve this, the algorithm makes use of the knowledge that in a healthy system there

will be strong correlations between the actuators process data. The algorithm performs



221

data reduction by identifying the most significant eigenvalues (principle components)

which capture the highest variability and projecting the data sets onto these principal

components. This means that the same information can be achieved through

reductions in the data. The fault detection is reliant upon two performance statistics

referred to as squared prediction error and a 2T - Hotellings statistics. When suspected

faulty process data is available, this data is projected onto the nominal principle

component model and these statistics are calculated. The use of fuzzy logic

classification is then used to interpret these, to extract the necessary information

regarding the actuator process quality. The application of the fuzzy logic approach

provides the user of the monitoring system a crisp numerical output describing the

actuators process quality for use in aiding in maintenance decisions.

The proposed diagnostics approach utilises a system model which acts upon the

actuators dynamic process data to generate a set of diagnostic residuals. A model of

the landing gear actuator certified by the sponsoring organisations has been developed

and the four key residuals based upon this model have been defined. The approach to

threshold evaluation used which moves away from the traditional use of ‘one

threshold per residual’ approach. Rather proposing the use of multiple thresholds for

each residual each sensitive to an individual fault. Diagnostic inaccuracies associated

with fixed static thresholds are overcome by designing thresholds which follow

dynamic profiles.

Evaluation of the residuals with respect to thresholds also moves away from the use of

traditional Boolean Logic utilising the concepts of approximate reasoning. The

approximate reasoning approach provides a measure on the extent to which a

threshold has been breached, allowing estimation to the magnitude of the fault

occurrence. With this approach incremental changes in the residuals about a threshold

will result in incremental changes to the diagnostic results. The result is recorded as a

measure known as basic probability assignments. These refer to the evidence for the



222

presence of individual faults. The use of approximate reasoning ensures that

diagnostic stability issues related to modelling errors or noise are removed

The Dempster Shafer theory of evidence is used to combine fault evidences, which is

demonstrated in this research thesis to increase the belief in the diagnostic results

whilst reducing uncertainty, with increased levels of evidence. The results of fusing

multiple evidences are a set of ranked evidential intervals based upon belief and

plausibility which are evaluated against a set of defined fusion rules.

Importantly the algorithm has been designed to only require data which will be

available from the actuators control system. This therefore meets the key design

criteria of minimising the levels of additional sensor equipment and complexities.

9.2.4 Objective 4

Demonstrate and assess the commercial benefits of incorporating health monitoring

systems into aircraft landing gears from the viewpoints of for O Manufacturers

(OEM), Suppliers, Maintenance providers and aircraft operators.

The commercial benefits of incorporating health monitoring have been assessed

through dialogue with the sponsors, in order to understand aircraft maintenance from a

holistic viewpoint. There are a variety of differing organisations whose business is

affected by aircraft maintenance practice. These include manufactures, maintenance

providers, suppliers, and aircraft operators. The nature of current maintenance practice

has been clearly identified, along with the direction towards a predictive maintenance

practice and the role in which health monitoring will play. In addition to technical

challenges associated with health monitoring there are a variety of business integration

challenges which must be understood. The key challenges are identified as the

following



223

Cost drivers: The cost drivers for each business associated with maintenance will

differ and will often be competitive. Understanding these and the differing

organisations relationships are the key to developing essential predictive maintenance

costing model. The cost drivers for OEM, MRO and aircraft operators have been

clearly identified and an approach to maintenance based upon unique customer

tailored maintenance packages, requiring information transparency is proposed.

Business Models: The sponsoring company which sees itself as a future landing gear

health monitoring solutions provider has well defined business models for their

products. Challenges include how to market the technology alongside these existing

products. Three pricing models are therefore proposed. The first considers that health

monitoring is given away at a consolidated cost differentiating the landing gear

product from those of a competitor. The second involves generating revenue based

upon sales volumes. A unit cost is paid by the landing gear customer to the health

monitoring system provider for every landing gear set using the monitoring

technology. The final model is based upon per landing gear unit. A one off payment

is made to the health monitoring provider for any given monitoring solution per

landing gear set. This ensures that the customer retains the ownership for the life of

that product.

9.3 Contribution to Knowledge

The purpose of this thesis and the overall contribution which has been made is to

bring together a combined understanding of landing gear design, health monitoring

and the business environment for aircraft maintenance in order for a holistic design

process for landing gear health monitoring to be realised.

9.4 Further Work

The development a health monitoring system for an industrial application such as

electrical actuators for Landing gears is very multidisciplinary and a holistic viewpoint

must be taken. As with all research programmes time constraints will limit further



224

development and narrow the research focus. There is however a number of

opportunities and areas, both from a technical and business perspective which must

still be investigated and developed before landing gear health monitoring can be

developed into a market ready system. A selection of these includes:

9.4.1 Experimental Landing Gear Test bed

One of the major limitations of implementing health monitoring into a landing gear

electrical actuator will be the lack of data from the new actuators. Extensive testing

would be needed to be performed prior to any in service implementation and a set of

generic fault cases will need to be developed. Data from fault case testing will enable

the classification of faults enabling the diagnostics approach to be validated

experimentally on a landing gear testing rig. Experimental classification would

enable diagnostic residuals and the fault sensitive thresholds to be designed to be

robust. In any fault related case, it is known that common symptoms will occur

between faults, with some symptoms being more prominent depending upon the

particular fault present; this brings into focus the requirement for weighting

symptoms. Experimental fault testing on a landing gear rig, under representative

operating conditions would enable fault - symptom weights to be assigned, increasing

diagnostic certainty.

9.4.2 Uncertainty and Performance Metrics

The use of a model based diagnostics system will suffer from a variety of uncertainties

which will need to be taken into account, to reduce uncertainty in the diagnostics.

Research would focus on understanding and dealing with the various sources of

uncertainty. Quantifying modelling errors, signal errors and dealing with uncertainties

in the initial lack of information. Techniques would need to be developed to handle

historical and service data as it becomes available throughout the actuators service

life. Difficulties with dealing with such information are usually associated with

incomplete or missing records.



225

It is essential that the performance of a monitoring system is known. Dialogue with

designers and experimental testing is required to quantify individual component

failure probabilities. This would allow specific component level objectives to be set

for the monitoring system. Performance metrics would then be employed, likely to be

based upon a measure of accuracy in the diagnosis. This measure of accuracy would

be calculated through measures such as probability of false alarm, correct alarm,

missed alarm etc. If a monitoring system cannot meet specified objectives then

reliance upon it would have an adverse effect upon the aircraft safety.

9.4.3 New Sensor Technology and Systems Integration

This research has put a strong emphasis on the need to keep a diagnostics systems

affects on the actuators weight and complexity to a minimum. The thesis review work

briefly highlights the emergence of new sensor technologies. Future research areas

should focus on monitoring solutions using wireless smart sensors with inbuilt signal

processing that can perform all monitoring tasks. The benefits of these are that they

are lightweight, do not require additional cabling, creates a reduction in the demand

for aircraft computing resources. These advantages opens new doors for the

investigation of monitoring techniques which have been deemed inapplicable with

conventional sensor technology.

Systems integration research would need to be undertaken to ensure that any future

monitoring system can not only integrate into the landing gear subsystems, but can

also integrate seamlessly into the aircraft systems as a whole and if necessary work

alongside aircraft BIT and any other local monitoring systems.

9.4.4 Cost Modelling

The introduction of a health monitoring based maintenance system would have a

direct impact upon the through life support costs and processes of the landing gear

equipment. The research into potential novel approaches to predict the through-life

manufacture and repair costs for long life actuator products would be fundamental for

decision making. Information from a monitoring diagnostics system incorporated into



226

costing models would be aimed at enabling the supply chain to predict, evaluate and

optimise their operations for repair and maintenance. Costing research would require

a focus on developing risk and uncertainty service metrics, assessments of

organisational relationships and the identification and quantification of the various

cost drivers.

9.4.5 Remaining Life Models

This research has been focused entirely upon diagnostics. However it is the industries

desire to extend diagnostics into the realm of fault prognosis, returning remaining

useful life estimations. A case based empirical study would be required to produce

data sets observed from in service reports. The empirical data would then be used to

generate an in service benchmark model the value of which would be to establish the

remaining useful life of the system, or individual component. Remaining useful life

models typically use measures such as operating hours, however for the landing gear

application a remaining useful life measured in terms of 'number of cycles' would be

more appropriate. Such a remaining useful life model would offer invaluable

information in regards to generating an aircraft maintenance strategy.

9.5 Published Research Papers

As part of this research the following peer reviewed papers have been published or

submitted for review.

9.5.1 Journal Papers

Phillips, P., Diston, D., (2011), "A knowledge driven approach to aerospace condition

monitoring", Knowledge-Based Systems, Vol 24 (6), pp. 915 - 927

Phillips, P., Diston, D., Starr, A., (2011), "perspectives on the commercial

development of landing gear health monitoring systems", Transportation Research

Part C: Emerging Technologies, Article in Press



227

Phillips, P., Diston, D., Starr, A., (2011), "Towards a business enterprise strategy to

support the integration of proactive aerospace maintenance technologies", INSIGHT:

non-Destructive Testing and Condition Monitoring", Article in Press

9.5.2 Peer Reviewed Conference Contributions

Phillips, P., Diston, D., Payne, J., Pandya, S., Starr, A. (2008) “The application of

condition monitoring methodologies for the certification of reliability in electric

landing gear actuators”, The 5th International Conference on Condition Monitoring

and Machine Failure Prevention Technologies, Edinburgh, UK

Phillips, P., Diston, D., Starr, A., Payne, J., Pandya, S. (2009) “A review on the

optimisation of aircraft maintenance with application to landing gears”, The 4th World

Congress on Engineering Asset Management and Intelligent Maintenance Systems,

Athens, Greece

Phillips, P., Diston, D., Payne, J., Pandya S., (2010) ‘Evidential reasoning applied to

model-based diagnostics in landing gear actuators’, Machine Failure and Prevention

Technology Conference, AL, USA

Phillips, P., Diston, D., (2010) ‘An intelligent health monitoring framework for a

motor-driven actuator’, Proceeding on the 7th International Conference on Condition

Monitoring and Machine Failure Technologies, Edinburgh, UK

Documents

HEALTH MONITORING OF ELECTRICAL ACTUATORS FOR …