Predictive Analytics with Artificial Intelligence...Case Study: Manufacturing Conveyer Catch PRiSM identified a conveyor motor drawing a higher than expected amount of current (amps)

© 2019 AVEVA Group plc and its subsidiaries. All rights reserved.

Predictive Analyticswith Artificial IntelligenceDeep-dive

Jim Chappell – VP Information Solutions

October 2019

Agenda


• Introduction to Predictive Analytics w/real-world examples

• How does it work?

• More real-world case studies

• How to deploy & integrate it

• How to get the most out of it

• What’s next with AVEVA AI?

• Putting it all together

• Key take-aways

• AI questionnaire

What Does It Do?


DetectAbnormalBehavior

PrescribeSolutions

Understand When to

Take Action


How Does It Help?

• Reduces Unscheduled Downtime

• Improves Operations

• Improves Quality

• Prevents Equipment Failure

• Optimizes Maintenance Strategy

• Reduces Costs & Risk

• Increases Line & Asset Utilization

• Extends Equipment Life

• Identifies Underperforming Assets

• Improves Safety


Enhanced Maintenance


Enhanced Operations

Alarms• Alarms are only sounded for critical issues• Process performance has been decreasing• Quality of output has decreased

Inspection• Employees lack guidance when inspecting issues• Complex issues are difficult to detect manually

Alerts• Process specific notifications are sent upon early

warning detection • Ample time is provided to prepare for inspection

Root-Cause Analysis• Employees understand exactly what needs to be

assessed• Issues are tended to in a brief amount of time

Traditional Operations Predictive Operations

Cost• Loss in income when your process is not at peak

performance• Loss in income from frequent unscheduled downtime

Savings• Assess issues at an optimal time that minimizes the

cost of labor and amount of downtime

Cutting-edge

What’s Happening

Real-Time

• Processing of real-time operational data

• Rule based inference for causal analysis

• First principles

Real-Time Domain

What Happened

Historical

• Assessment and exploration of historical operational data

• Trends, KPIs, Dashboards to present abstracted views

Historical Domain

What If

Predictive

• Comprehensive model based assessment of operational data ranges to determine potential outcomes.

• Deterministic or non-deterministic models

• Open-loop simulations

Machine Learning Domain

What to do

Prescriptive

• Systems that synthesize, predict and provide scenario-based guidance

• Fault diagnostics and knowledge capture

• Sensor contribution & root-cause analysis

• Extensive library of prescriptive actions

Maintenance Action Domain

How bad will it get

Prognostics

• Forecast future state of assets and sensor values

• Determine if you can make it to the next planned maintenance outage

• Provide key input for risk assessment

Artificial Intelligence Domain


Real-world Examples


Case Study: Manufacturing Conveyer

Catch PRiSM identified a conveyor motor drawing a higher than expected amount of current (amps) for the given level of operation.

Cause In an effort to improve tracking, a mechanic changed the tension of the belt by increasing the air pressure.

Avoided If this hadn’t been caught, a bearing, roller, motor, or some combination of these would have been compromised, resulting in significant downtime. In addition, the belt would have likely separated.

Issue Monitor product quality by detecting multi-variate sensor changes which could indicate a problem with a motor, gearbox, or bearing.

Predictive Models

Roller Mechanical

• Amps drawn by motors

• Speed of roller motors

• Gap between top & bottom rollers

Conveyor Mechanical

• Amps drawn by conveyor motors

• Speed of conveyor motors

• Speed of production line


ObservationBearing metal temperature spikes were observed.

ResultsSite investigation found oil reservoir filled with half water and half oil. It was determined that the intricate valving was supplying too much pressure to the seals resulting in water flowing to the bearings.

Example Predictive “catch”

Case Study: Pumps & Valving


Predictive, Prescriptive, & Prognostic


DEMO

Predictive Maintenance with AI



PrescribeSolutions

Understand When to

Take Action


How Does It Do It?

EARLY WARNING DETECTIONDeviations from normal operation

identified and displayed3

HISTORICAL DATAApplication learns normal

operation from historical data1

PATTERN RECOGNITIONAdvanced algorithms automatically create

and organize operational profiles2

• On-premises

• Cloud / SaaS

• Hybrid architectures


Monitoring Without Advanced Analytics


Subtle Changes

Actual Value

Predicted ValueOu

tbo

ard

Be

arin

g T

em

p (

°F)

Date and Time


A Smaller Box

Ou

tbo

ard

Be

arin

g T

em

p (

°F)

Oil Drain Temp (°F)

Traditional Alarms

PRiSM


Case Management

Core Features

Diagnostics Advisor

(Fault Diagnostics)

Early Warning Event

Detection & Management

(Alarm Manager)

Trend Analysis Reporting

Significance of the Deviation from Normal

Operation

Signal Contribution to Performance Anomaly – Normal/Predicted vs.

Actual

Likely Fault Condition

Signal Contribution to

Change in Health Status

Probability of Fault Match

Asset Health Trend


Prescriptive

• Signal impact

• Pinpoint problem

• Root cause analysis

• Upstream/downstream

line impacts

• Case management

Likely Fault Condition Probability of Fault Match

Asset Health Trend

Signal Contribution to

Change in Health Status

• Signal impact

• Pinpoint problem

• Root cause analysis

• Upstream/downstream

line impacts

• Case management


Prescriptive

• Leveraging library of

over 900 asset types

and 10,000 faults with

prescriptive actions


Prognostics

• How fast will the situation

worsen?

• What-if analysis

• Can you make it to the next

planned maintenance outage?

ACTUAL data

stops here

FORECAST future values;

there is no actual value at

these times

➢ Neural Net ➢ Deep Learning ➢ Easy to Use


Ease of Deployment

Templatized:

• Models

• Digital signatures

• Alarms

• Filters

Templates for fast deployment and easy administration


Intelligent Alert List



Data Playback

• Can have “n” number of training

data sets

• Determine when PRiSM would

catch the problems

• Determine when to alarm

• Easy to retrain model


Case Management (Knowledge Capture)


Advanced Calculations


Data Visualization


Transient Analysis (Startup/Shutdown)


Vertical Markets and Equipment Types

• Turbine

• Compressor

• Conveyor

• HVAC

• Electric Generator

• Pumps

• VFD’s

• Fans, blowers

• Heat Exchanger, Boiler, Oven, Kiln

• Air Heaters

• Water Heaters

• Pulverizer, Crusher

• Condenser

• Transformers, Breakers, Capacitors

• Agitators, Blenders, Mixers

• Gearbox

• Chillers

• Seal systems

• Renewable energy

Power Generation

Power T&D

Oil & GasWater Management

MiningProcess,

Manufacturing, and

F&B

Etc…

More Real-world Case Studies



Real-world: Ovens

Predictive model

• Top temp of each zone

• Bottom temp of each zone

• Damper valve %

• Extraction fan speed

• Etc…

• Zone humidity

• Zone pressure

• Extraction fan current

• Oxidizer pressure

Issue

• Each oven zone needs to maintain certain heating

characteristics to drive product specifications, such as stack

height, color, % moisture content, etc. Predictive analytics

is used to minimize line stoppages.


Case Study: Irregular Motor Operation

Observation

• Motor current increased from 14 to 18 amps for a given

load.

Result

• Plant found a leak on the floor above that saturated the

insulation, causing expansion issues with the shroud.


Real-world: Cogeneration

Issue

• Pasta production line creates methane gas which is used as

fuel in a cogen facility. The plant uses the methane to

generate steam in order to produce heat and electricity, both of

which are subsequently used back in the production process.

Predictive analytics is used to optimize maintenance practices

to avoid unplanned outages.

Predictive model

• Inlet/outlet water temps

• Inlet/outlet water pressures

• Inlet/outlet steam temps

• Inlet/outlet steam pressures

• Etc…


Case Study: Steam Turbine

Due to the advanced pattern recognition and alarming

features of the Avantis PRiSM software, catastrophic

equipment damage and potential significant personnel injury

were averted.

After reviewing the events and all actions taken to remedy

the situation, a conservative estimate of the equipment

avoided costs was determined to be approximately

$34.5 million USD.

Summary



Case Study: Pump Control Issue

• Vacuum fell from 27.8 to 25.9 inches within an hour

• Hotwell temp increased from 98 degrees F to 123 degrees

F within an hour

Found water in the air lines to the suction valves at the

vacuum pumps, which was causing them to close.

Predictive Observation:

Result:


Case Study: Gas Turbine

Unaccounted for turbine vibrations for the given generation

level, but nowhere near sufficient to cause an alarm.

After investigation, a turbine blade was found to have a chip

that was reducing the efficiency. This issue was getting worse

over time and would have eventually resulted in a failure.

Predictive Observation:

Result:

The plant replaced the blade and estimated the cost avoidance at over $17 million USD.


Case Study: Compressor

Compressors are extensively used and critical to the operation of

one of the largest manufacturers of industrial gases in the world.

The site found a cracked impeller.

Issue:

Cause:

The customer estimated that over $500K USD in costs were

avoided by preventing reactive maintenance and unplanned

downtime. Based on the speed of the crack propagation, our

Predictive Analytics detected this issue approximately 3 months

before operators would have noticed it.

System identified higher than expected vibrations on the second

stage of a compressor in relation to the current level of operation.

Based on this detected anomaly, the site investigated the issue

during an upcoming planned maintenance outage.

Catch:

Avoided Impacts:

SummaryThe oven oxidiser is responsible for reducing the environmentally harmful emissions created during the baking process. This is done by burning the oven exhaust gases at a high temperature (oxidising) and then running the hot exhaust over a catalyst to help induce a chemical reaction.

Predictive “catch”The system identified a higher than expected differential pressure across the catalyst bed, indicating a clog which can disrupt the oxidation process. However, this issue only occurred once in awhile.

The site determined that this predictive notification only occurred when the weather was extremely cold, causing the lines that bleed moisture to the outside to freeze. Maintenance personnel unthawed the lines, allowing the moisture to dissipate which, in turn, reduced the pressure and eliminated the predictive analytics alarm.

Case Study: Oven Oxidizer



Case Study: Renewable Energy - Wind

Identified asset health issues around a wind turbine resulting from

bearing vibrations being too high for the level of power generated.

The wind turbine was then inspected and PRiSM results

confirmed.

The wind turbine was shut down immediately. Maintenance found

a roller cocked 180 degrees and the retainer had failed.

Significant damage to the turbine was avoided, and it was

repaired and put back in service.

Observation:

Result:


Case Study: Renewable Energy - Solar

Photovoltaic performance diminished resulting in less

current (amps) produced.

A cracked solar panel was identified and repaired,

resulting in normal operations once again. Additional

issues were found around an inverter.

Observation:

Result:


Case Study: Non-rotating Machinery

Southern Company generates more & more energy from renewable resources

each year. They use wind, solar and biomass to meet customer demand and

regulatory requirements.

Observation:Southern Company found a major transformer issue during Hurricane Matthew in

October 2016. One of the PRiSM transformer health indicators (dissolved gas

ratios) went extremely high, indicating problems. Southern investigated, finding

a transformer that was fully charged with no load.

Result:An explosion could have occurred in this situation if loads were brought on

quickly. They successfully prevented catastrophe. Typical replacement cost of a

major transmission transformer ($10M USD).

Summary



Line 8 Oven• Tag determining if the oven is ON/OFF

• Pounds per minute produced while running

Line 8 AccumulatorPercent fullness of accumulator

Max capacity of accumulator in pounds

Packing Line 13Tag ON/OFF

Lbs packed per minute








Lbs packed per min


Lbs packed per min

Production Line Operation


Production Line Operation

Periods of increasing

accumulator levels

correspond with depressions

in the ‘Time Full’ chart.

Alarms typically configured for

when the ‘Time Full’ tag is less

than 5 minutes

• Accurately determine when

to start/stop operations due

to starved or blocked lines

• Improve Operational

Efficiency

• Analyze a production line in

its entirety to optimize

Quality

How Do You Deploy & Integrate Predictive Analytics?



Easy Integration

Predictive Analytics& beyond

SCADA

• System Platform• InTouch• Citect• Etc..

WonderwareHistorian

On-premises Cloud


Easy Integration


SCADA

• System Platform• InTouch• Citect• Etc..

WonderwareHistorian

On-premises Cloud

MES


Any Data Historian

Any Control System


Any Database

Any Data Lake

On-premises Cloud

Easy Integration


Any Control System


AVEVA Insight

On-premises Cloud

Easy Integration


Questions?


Getting the Most out of Predictive Analytics


AVEVA’s Engineering/Analytics Center of Excellence

• Standards & Best Practices

• Technical Training/Support

• Modeling Services

• Global Monitoring Centers

• Chicago, IL, US

• Hyderabad, India

• Monitoring & Tech Transfer Services

• Consulting Services

CoE


The MDC

reviews and

notifies the site of

relevant alarms

The site receives

a report from the

MDC highlighting

potential issues

detected by the

system

AVEVA’s Engineering/Analytics Center of Excellence


What’s Next with AVEVA AI?

Human UnderstandingGap in Understanding

The complexity of AI creates a lack of understanding for

humans. This lack of understanding causes people to be wary of AI technology.

AI TechnologyUnlimited Potential

AI is quickly being integrated into the workplace.

Technology is constantly evolving and holds vast

opportunities for businesses.

Bridging the GapPrescriptive Solutions

Prescriptive bridges the gap between AI technology and

human understanding. • In context• Useful• Actionable

Bridging the AI Gap


How to Bridge the Gap

ExpertiseFaults & Prescriptive

Actions AIEarly

WarningDetection

FuturePredictions

ClearSteps

ToAction


Amplified

Workforce

Productivity

Prescriptive: The AI Bridge

Actionable Context

• Enhances workforce productivity

• Reduces risk

• Improves safety

• Improves security

• Improves:

Availability – Performance - Quality

• Improves reliability

• Optimizes budgets

Clear actions

Schedule optimization


$100,000,000’s in avoided costs (per major customer)

Workforce Optimization

Prescriptive

Risk Mitigation• Manage time remaining before action is required • Optimize schedules• Minimize impact on operations• Reduce costs

Early Detection• Production problems• Inefficiencies• Asset maintenance issues• Data governance (bad sensors, lost connections, etc.)

Understanding• When the issue began to occur (early warning)• Where the issue is located (root cause)• Severity and potential impacts• Prescriptive solutions


Predictive & Beyond

MaintainIncreased longevity and performance of assets while ensuring a safe, reliable environment for the workforce through predictive, prescriptive, prognostic analytics

Monitor | Control Predictive and autonomous operational control to ensure safety, performance

Operate | OptimizeImproved Optimization & OEE for safe and profitable operations within constraints and regulatory norms

Plan | ScheduleSelf-learning approach to model planning, scheduling in orderto maximize profitability and efficiency

Artificial Intelligence

EngineerAnalysis and automated design generation enabling lower total cost, time, risk in capital project engineering

• Predictive

• Performance

• Prescriptive

• Prognostic



E3

AI-Driven End to End Engineering


• Design engineering

• Operations engineering

• Maintenance engineering



Simulation with AI: Gray-box Modeling

Machine

Learning

Statistically

Correlated

Inputs*

Predicted

Output(s)

First Principles

Algorithms

Predicted

Output(s)

*sensor data / calculated data / measurements

Sensors

First Principles

Algorithms


Simulation with AI: Enhanced Anomaly Detection

Asset

tinpin

min

tout

pout

mout

toutcalc

poutcalc

moutcalc

tout vs toutcalc

pout vs poutcalc

mout vs moutcalc

First Principles Analytics

Automated

Performance Curves / Tuning

(machine learning)

sensors sensors


Predictive Data Generation for Enhanced Templates

DynSimInput

Data Set

Simulated

Output

Data Sets

Machine

Learning

Predicted Values

Anomaly Detection


Asset Health with Schedule Optimization

Machine

Learning /

Neural Net /

Deep Learning

Sensor

Data

Planning /

Scheduling

Optimization

(Reinforcement

Learning)

Remaining

Useful Life

EstimateEAM


Comprehensive Alert Infrastructure (all in one)


Asset Remaining Useful Life with Schedule Optimization

DEMO


SCADA AI

SCADAOperator

Control

Machine

LearningSCADA

data

Detection of issues

Machine

Vision

Video


SCADA AI

DEMO


Putting It All Together

AVEVA Insight

Cloud Mobile

IIoT


AVEVA Insight

Cloud

On Premise

Hybrid

Simple – Intuitive – Frictionless

Cloud

On Premises

Hybrid


Common IntelligentModel

Asset / Entity

Raw or

production

material

Prognostics

Calculations

Alarms &

events

Condition

triggers

Video

Prescriptive

actionsAsset

health

Performance

OEE

HMI/

Process

Graphics

Sensor

data

Location

Predictive

alerts


Integrated Digital Asset


Real-world Integrated Solution

Technician makes repairs and closes out Work Order > Case Management

System recommends procedure to rectify, providing step-by-step

video

> Prescriptive

> Mobile Workforce

System informs technician the pump is likely to fail within 7 days.

Emergency Work Order issued.> Prognostics

Technician finds oil reservoir filled with half water / half oil.

Determines valving supplying too much pressure to the seals,

resulting in water flowing to the bearings.

System triggers Work Request> Condition-based

> EAM

Technician receives alert on mobile device > Mobile

System determines bearing temp too high for conditions > Predictive

Insight collects various sensor data in the cloud > Big data

> Troubleshoot

REVIEW: Predictive Maintenance Fundamentals



PrescribeSolutions

Understand When to

Take Action


Key Predictive Analytics Take-aways

1. Improves Industrial Operations & Maintenance• O&G

• F&B / CPG

• Water / wastewater

2. Extends the value of AVEVA SCADA systems, MES, and Historian

3. Works with any historian or database (on-prem, cloud, data lake, etc.)

4. Offered in the cloud (SaaS) or on-premises

• Manufacturing

• Power

• Mining, Etc…


https://www.surveymonkey.com/r/H28VZHB

AI Survey: We need your feedback!

https://www.surveymonkey.com/r/H28VZHB


Questions?

linkedin.com/company/aveva

@avevagroup

ABOUT AVEVA

AVEVA is a global leader in engineering and industrial software driving digital transformation across the entire asset and operational life cycle of capital-intensive industries.

The company’s engineering, planning and operations, asset performance, and monitoring and control solutions deliver proven results to over 16,000 customers across the globe. Its customers are supported by the largest industrial software ecosystem, including 4,200 partners and 5,700 certified developers. AVEVA is headquartered in Cambridge, UK, with over 4,400 employees at 80 locations in over 40 countries.

aveva.com


Documents

Predictive Analytics with Artificial Intelligence...Case Study: Manufacturing Conveyer Catch PRiSM identified a conveyor motor drawing a higher than expected amount of current (amps)