43
1 Reghu Anguswamy Reghu Anguswamy MS – Systems Engineering MS – Systems Engineering Missouri University of Science and Technology Missouri University of Science and Technology 15 May 2008 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING HAND-HELD TOOLS IN NETWORK ENABLED MANUFACTURING HAND-HELD TOOLS IN NETWORK ENABLED MANUFACTURING ENVIRONMENTS ENVIRONMENTS

1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

Embed Size (px)

Citation preview

Page 1: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

1

Reghu AnguswamyReghu Anguswamy

MS – Systems EngineeringMS – Systems Engineering

Missouri University of Science and TechnologyMissouri University of Science and Technology

15 May 200815 May 2008

WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING HAND-HELD TOOLS IN NETWORK ENABLED USING HAND-HELD TOOLS IN NETWORK ENABLED

MANUFACTURING ENVIRONMENTSMANUFACTURING ENVIRONMENTS

Page 2: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

2

PAPERS

I. “In Process Detection of Fastener Grip Length using Embedded Mobile Wireless Sensor Network-based Pull Type Tools,” Reghu Anguswamy, Can Saygin and Jagannathan Sarangapani

• INTERNATIONAL JOURNAL OF MANUFACTURING RESEARCH

• ASME International Mechanical Engineering Congress and Exposition, Seattle, Washington, November 10-16, 2007

II. “A Multi-channel Routing Protocol for Ad Hoc Wireless Networks,” Reghu Anguswamy, Maciej Zawodniok and Jagannathan Sarangapani

• INTERNATIONAL JOURNAL OF DISTRIBUTED SENSOR NETWORKS

Page 3: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

3

Introduction

• Network-Enabled Manufacturing (NEM)– Remote monitoring of processes– Provides a platform for diagnostics in real-time

at both component and the system levels – In aerospace manufacturing environments,

• Hand-held tools widely used for fastening• Post process inspection – cost ineffective and

time consuming• On board diagnostics using NEM enables in-

process monitoring and removes the need for post process inspection

Page 4: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

4

NEM using Wireless Networks

Multi-Hop Wireless Networking

Mobile Devices (hand-held tools) and

Wireless SensorsApplication

Monitoring and Diagnostics at System Level

Monitoring at Component and Equipment Level

(Diagnostics on the mote)

Page 5: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

5

Problem Definition

• Two major issues have been addressed in this thesis for NEM shop floor systems using hand-held tools:– Incorporating intelligent behavior– hand-held tools

equipped with embedded sensors and on-board processing/computing capabilities; also, wireless capability for the tools add significant advantages to the network-enabled environment:

• increases the mobility of the tools in the shop floor which may not be the case if they are wired. Wireless ensures safety

• the hand-held tools become more complex and cumbersome to use if additional wires are added to the tool

– Data management • a typical shop floor environment will have many tools

operating simultaneously forming an ad hoc wireless networks

• the network requires efficient protocols such as routing and scheduling for handling the voluminous data generated.

Page 6: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

6

Paper I - OverviewPaper I - Overview• Introduction Introduction • Pull-type tool: Fastening operationsPull-type tool: Fastening operations• Grip length deviationGrip length deviation• Pull-type tool analysisPull-type tool analysis• Methodology: Real-time feature Methodology: Real-time feature

extraction and decision makingextraction and decision making• Experimental SetupExperimental Setup• ResultsResults• ConclusionsConclusions

Page 7: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

7

IntroductionIntroduction• Hand-held fastening tools in manufacturing industry:Hand-held fastening tools in manufacturing industry:

– Installation of fasteners may incur up to 80% of the Installation of fasteners may incur up to 80% of the total fastening costs while the remaining 20% is for total fastening costs while the remaining 20% is for procurement of fastenersprocurement of fasteners

– Fastening process and its inspection can be a very time Fastening process and its inspection can be a very time consuming taskconsuming task

– No inspection data is typically collected during the No inspection data is typically collected during the process unless a major problem is encounteredprocess unless a major problem is encountered

– Real-time monitoring of the fastening process and Real-time monitoring of the fastening process and verification of joint quality are two important factors to verification of joint quality are two important factors to reduce the manufacturing lead time while ensuring reduce the manufacturing lead time while ensuring safety and qualitysafety and quality

– Existing techniques include detection using measuring Existing techniques include detection using measuring probes, image processing systems or measuring gages probes, image processing systems or measuring gages

Page 8: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

8

Pull-type tool: Fastening OperationPull-type tool: Fastening Operation

Page 9: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

9

Pull-type tool: Fastening OperationPull-type tool: Fastening Operation

1.1. Fastener is inserted into the hole and the collar is placed over Fastener is inserted into the hole and the collar is placed over the fastener pintailthe fastener pintail

2.2. Tool nosepiece is placed on the shank of the fastener.Tool nosepiece is placed on the shank of the fastener.– When the tool is actuated, the pintail is pulled against the push on the When the tool is actuated, the pintail is pulled against the push on the

collar collar – The head of the fastener is seated into the hole and the tool starts The head of the fastener is seated into the hole and the tool starts

swaging the collarswaging the collar

3.3. Continued swaging of the collar forces the collar material into Continued swaging of the collar forces the collar material into the pin locking grooves creating an extrusion action that the pin locking grooves creating an extrusion action that stretches the collar, which results in a corresponding stretch stretches the collar, which results in a corresponding stretch of the pintail.of the pintail.

4.4. When collar is completely swaged (i.e., collar is resting When collar is completely swaged (i.e., collar is resting against the plate), the tool continues pulling the pintail until it against the plate), the tool continues pulling the pintail until it breaks at the break neck groovebreaks at the break neck groove

Page 10: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

10

Grip Length DefinitionGrip Length Definition

ThreadsGrip

Length

Normal Grip

Under GripOver Grip

• Normal Grip : The thickness of the material to be joined is Normal Grip : The thickness of the material to be joined is equal to the grip length of the fastener being usedequal to the grip length of the fastener being used

• Over Grip : The thickness of the material to be joined is less Over Grip : The thickness of the material to be joined is less than the grip length of the fastener being usedthan the grip length of the fastener being used

• Under Grip : The thickness of the material to be joined is more Under Grip : The thickness of the material to be joined is more than the grip length of the fastener being usedthan the grip length of the fastener being used

Page 11: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

11

Pull-type Tool AnalysisPull-type Tool Analysis

• Detect the grip length of the joint being Detect the grip length of the joint being fastened using a pneumatic, pull-type fastened using a pneumatic, pull-type tool tool

• Sensors to the system areSensors to the system are– StrainStrain– DisplacementDisplacement– PressurePressure

• Strain-displacement signatures are Strain-displacement signatures are analyzedanalyzed

Page 12: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

12

• Approach based on extracting unique features from process Approach based on extracting unique features from process signatures generated in real-time with data obtained from signatures generated in real-time with data obtained from strain and displacement sensors; such an approach has not strain and displacement sensors; such an approach has not been investigated previouslybeen investigated previously

• Sensors are installed on the exterior of the toolSensors are installed on the exterior of the tool

– No redesign of the tool is necessaryNo redesign of the tool is necessary• Fastening tool is also equipped with real-time, on-the-tool Fastening tool is also equipped with real-time, on-the-tool

decision making and wireless communication capabilities to decision making and wireless communication capabilities to communicate the result to a base-stationcommunicate the result to a base-station

• By combining the fastening feature with inspection capability By combining the fastening feature with inspection capability on a single tool, the proposed approach allows for data on a single tool, the proposed approach allows for data collection on 100% of the joints in real-time as opposed to collection on 100% of the joints in real-time as opposed to sampling-based statistical process control methods that are sampling-based statistical process control methods that are applied post-process as stand-alone operationsapplied post-process as stand-alone operations

Uniqueness in approachUniqueness in approach

Page 13: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

13

Preliminary StudyPreliminary Study

• Different combinations of plates Different combinations of plates were testedwere tested

• Plates with rubber washers in Plates with rubber washers in between were also tested to between were also tested to investigate the impact of material investigate the impact of material properties on the process properties on the process signaturesignature

• For a given combination of plates For a given combination of plates and fasteners, there is a unique and fasteners, there is a unique process signatureprocess signature

• Any deviation from this unique Any deviation from this unique signature is a sign of abnormality, signature is a sign of abnormality, which could be a quality problem which could be a quality problem

Page 14: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

14

Process Signature AnalysisProcess Signature Analysis

• Due to variations within each grip Due to variations within each grip length category, strain-time and length category, strain-time and displacement-time signatures displacement-time signatures alone are not robust enough to alone are not robust enough to detect normal, over, and under detect normal, over, and under grip conditions grip conditions

• A new parameter, namely A new parameter, namely Strain/Displacement ratio (S/D), is Strain/Displacement ratio (S/D), is introduced, using the primary introduced, using the primary sensor data (i.e., nosepiece strain sensor data (i.e., nosepiece strain and displacement), a compound and displacement), a compound process signature process signature Strain/Displacement versus Strain/Displacement versus Displacement is developedDisplacement is developed

• Normal grip data exhibits a Normal grip data exhibits a unique “bowl-shaped dip” unique “bowl-shaped dip”

Page 15: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

15

Real-time Feature ExtractionReal-time Feature Extraction

• Three unique features that Three unique features that are used to detect grip are used to detect grip length variations length variations • Peak strain on the strain-Peak strain on the strain-

time signature, which time signature, which occurs right before the occurs right before the tool completes the tool completes the fastening operationfastening operation

• The other two features The other two features are on the are on the strain/displacement-strain/displacement-displacement signaturedisplacement signature

• Depth of the “bowl” Depth of the “bowl” • Width of the “bowl” Width of the “bowl”

Page 16: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

16

Continuous Monitoring of

Air Pressure, P(t)

NORMAL GRIP

Dlow_limit < D < Dhigh_ limit

andWlow_limit < W < Whigh_ limit

NO

YESYES

NO

Displacement(t)

Strain(t)

S/D(t)

P(t), W, D, εp

?

εp < εp_under grip

?

UNDER GRIP OVER GRIP

?P(t): Air pressureW: Width of bowlD: Depth of bowlεp: Peak strain

Real-time Feature Decision-makingReal-time Feature Decision-making

Page 17: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

17

• Variation in depth (D), Variation in depth (D), width (W), peak strain width (W), peak strain (εp), and pressure have (εp), and pressure have been studied based on 15 been studied based on 15 samples (i.e., data from samples (i.e., data from 15 fasteners) 15 fasteners) – Limits for depth of Limits for depth of

bowl-shaped dip for bowl-shaped dip for normal data are 5 mm normal data are 5 mm and 25 mm and 25 mm

– Limits for width is 0.25 Limits for width is 0.25 mm and 2 mmmm and 2 mm

– To differentiate over To differentiate over grip from under grip, grip from under grip, peak strain for under peak strain for under grip is used, which has grip is used, which has limit of 930 limit of 930 microstrainmicrostrain

Variability And Setting LimitsVariability And Setting Limits

Setting the LimitsNormal Grip:

Dlow_limit = 5 microstrain/mmDhigh_ limit= 25 microstrain/mmWlow_limit = 0.25 mmWhigh_ limit= 2 mm

Under Grip:

εp_under grip = 930 microstrain

Page 18: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

18

Variability And Setting LimitsVariability And Setting Limits

• Several factors can contribute to variability in dataSeveral factors can contribute to variability in data– Variability in plate thicknessVariability in plate thickness

• Manufacturers specify a tolerance range for Manufacturers specify a tolerance range for platesplates

– Sensitivity of the sensors Sensitivity of the sensors • Strain gage used on the nosepiece has a 120 Strain gage used on the nosepiece has a 120

ohm configuration with ± 0.0015 % variability in ohm configuration with ± 0.0015 % variability in resistance. resistance.

• LVDT used has a sensitivity of ± 6.2 nanometer LVDT used has a sensitivity of ± 6.2 nanometer per mm per mm

– Variability due to operatorVariability due to operator• Holding the tool at different anglesHolding the tool at different angles• Using both hands or pressing harder on the Using both hands or pressing harder on the

nosepiece using body weightnosepiece using body weight

Page 19: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

19

Experimental SetupExperimental Setup

• Three thicknesses have been tested: Three thicknesses have been tested: – 3.7 mm (normal grip)3.7 mm (normal grip)– 5 mm (under grip) 5 mm (under grip) – 7.4 mm (over grip)7.4 mm (over grip)

• Sensor data are acquired using a USB based Data Sensor data are acquired using a USB based Data Acquisition Card with 8 analog input channels Acquisition Card with 8 analog input channels – Data have been sampled and stored at 5,000 samples Data have been sampled and stored at 5,000 samples

per second for each sensor data per second for each sensor data

• Tool is integrated with embedded mobile Tool is integrated with embedded mobile hardware (mote) and is powered by batteries on-hardware (mote) and is powered by batteries on-board the tool board the tool

Page 20: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

20

• The proposed methodology has been implemented on a 5-layer The proposed methodology has been implemented on a 5-layer stack motestack mote ::

– Power layer: Regulated power supply of ±5V to the sensors Power layer: Regulated power supply of ±5V to the sensors and 3.3V to the processing layer and the communication layerand 3.3V to the processing layer and the communication layer

– Sensor input layer: Output voltages from the sensors are fed to Sensor input layer: Output voltages from the sensors are fed to the processing layer through this layerthe processing layer through this layer

– Strain gage conditioning layer: Strain gage conditioning is done Strain gage conditioning layer: Strain gage conditioning is done on this board; strain gage placed on the nosepiece is operated on this board; strain gage placed on the nosepiece is operated in a quarter bridge configurationin a quarter bridge configuration

– Processing layer: based on a Silabs® C8051F120 processor for Processing layer: based on a Silabs® C8051F120 processor for processing the sensor data and provide the decision to the processing the sensor data and provide the decision to the communication layercommunication layer

– Communication layer: based on a MaxStream® XBee wireless Communication layer: based on a MaxStream® XBee wireless radio communication, sends the decision received from the radio communication, sends the decision received from the processing layer to the serverprocessing layer to the server

HardwareHardware

Page 21: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

21

ResultsResults

• 100% correct detection of normal grip condition, while 4% of 100% correct detection of normal grip condition, while 4% of over grip cases and 6% of under grip cases are falsely over grip cases and 6% of under grip cases are falsely detected detected

• False detections can be attributed to the variability caused False detections can be attributed to the variability caused by the tool, sensor sensitivity or human error while by the tool, sensor sensitivity or human error while operating the tool, as well as due to the size of initial operating the tool, as well as due to the size of initial training sample set training sample set

Page 22: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

22

ConclusionsConclusions• Feature extraction-based methodology detects normal, over, and under grip Feature extraction-based methodology detects normal, over, and under grip

cases and is practical to implement as a real-time decision-making tool cases and is practical to implement as a real-time decision-making tool • Experimental results have shown that over 96% of the grip lengths have Experimental results have shown that over 96% of the grip lengths have

been detected with 100% of the data transferred successfully via the motebeen detected with 100% of the data transferred successfully via the mote• Wireless implementation also shows that such a methodology is practical Wireless implementation also shows that such a methodology is practical

and reliable for in-process quality monitoring in a shop floor environment, and reliable for in-process quality monitoring in a shop floor environment, such real-time quality control techniques:such real-time quality control techniques:• Reduce the amount of post-process quality control thereby saving Reduce the amount of post-process quality control thereby saving

expended capitalexpended capital• Can be very effective in other manufacturing environments as well Can be very effective in other manufacturing environments as well

• The proposed architecture has merits to The proposed architecture has merits to • Detect and report quality problems in real-time during the process Detect and report quality problems in real-time during the process • Implement without complexity by extracting useful features from Implement without complexity by extracting useful features from

process signatures and performing simple rule based methodology,process signatures and performing simple rule based methodology,• Reduce post-process inspection, thereby improving quality while Reduce post-process inspection, thereby improving quality while

reducing cost and man power reducing cost and man power

Page 23: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

23

VideoVideo

Page 24: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

24

Paper II

• Motivation and challenges• Proposed routing protocol• Optimality analysis• Results and Discussion• Conclusions

Page 25: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

25

Motivation

• Multi-channel– IEEE wireless standards, for example the IEEE

802.11b/g, IEEE 802.11a and IEEE 802.15.4, offer up to 16 non-overlapping frequency channels for simultaneous communication

• Multi-radio– Single radio interfaces can switch between

channels, but only one channel at a time– Multiple radios enable communications in

multiple channels at a time

Page 26: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

26

Challenges

• Selection of the best route from the source to the destination based on a more appropriate routing metric– Routing metric must consider the three major

measures of QoS namely the throughput, energy efficiency and the end to end delay for data transfer

• To fully exploit the advantages of the multi-channel environment, the route elected must be in multiple channels over multiple hops

Page 27: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

27

Proposed Routing Protocol

• Proactive and asynchronous• Independent of channel assignment

– Receiver based channel assignment• MPR (multi-point relay) based protocol• Multi-channel multi-radio

Page 28: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

28

Routing Metric

Nomenclature: N : Set of nodes in the network s : Source node d : Destination node

)(sN : Set of one-hop neighbors of node s

)(2 sN : Set of two-hop neighbors of node s )(sMPR : Selected Multipoint Relay (MPR) set of

node s

S N2N1

Page 29: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

29

Routing Metric

Utilization metric ( MPRnsU 2, ) of the link from node s to node

2n through the chosen MPR ( 1

n ) and )(2

2 sNn ; is given by:

DUEFBU MPRns /.).*..(

2, ………….(1)

where,

..FB : Bandwidth factor between nodes s and MPR ( 1n )

SA BBFB /.. ………………….(2)

AB : available incoming bandwidth at the MPR ( 1n )

SB : expected outgoing bandwidth at the source node (s)

..UE : Energy utilization between MPR ( 1n ) and 2

n

211 /.. nnTX

nA EEUE ………..(3)

1nAE : available node energy at the MPR ( 1

n ) in Joules 21 nn

TXE

: transmission energy per byte from 1n to 2

n in Joules/byte

D : end to end delay from node s to node 2n in secs

S N2N1B.F. E.U.

Page 30: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

30

The Routing Algorithm1. Neighbor Sensing2. MPR selection3. Topology information declaration4. Routing Table Calculation

1. Cost of a route is given by:

k

k

k

kk

ndn

nnn

nnn

nnsds CCCCC ,,

2,1

1,, 1

1

232,,......,,

…(4)

where,

MPRnsC 2, : Cost metric of the link from node s to node

2n through the chosen MPR ( 1

n ) and )(2

2 sNn ; is given by:

MPRns

MPRns UC

22 ,, /1 ………………(5)

Page 31: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

31

Multiple Channels over a Link

• A node may be equipped with radios capable of receiving data over multiple non-interfering channels

• A receiving node may be already receiving data from a different source through a channel and hence, the bandwidth available among the receiving channels may be different. So the data must be sent be in a balanced mode among the various channels for optimal performance

• Based on the characterization of optimal routing given in Bertsekas and Gallager*, the optimal condition is achieved when:

1

1 21

1

2

1

)()(

k

jii k

i

k

imm

jj

j

BbbBi

Bi

bB

B

*D. Bertsekas and R. Gallger, Data Networks, New Jersey: Prentice Hall, Inc., 1987, pp. 374-380

Page 32: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

32

Optimality Analysis

• Assumption 1: If the one-hop neighbor of a node S, has no direct link to at least one of the two-hop neighbors of S, then it is not on the optimal path from S to its two-hop neighbors. However, in order to reach a two-hop neighbor from S through such a node, the path has to go through another one-hop neighbor which has a direct link to the two-hop neighbor

• Corollary 1: “The MPR selection based on the utilization metric – based MPR selection provides the optimal route from a node to its two-hop neighbor”

Page 33: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

33

Optimality Analysis

• Corollary 2: “The set of MPRs selected for its two-hop neighbors is optimal”

• Corollary 3: “The intermediate nodes on the optimal path are selected as multipoint relays by the previous nodes on the path.”

• Theorem 1: The routing protocol selects the optimal route based on the cost metric between any source-destination pair.

Page 34: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

34

Results

• Ns2 2.30 extended with multi-channel and multi-interface capabilities

• Nodes have multiple interfaces, one of them dedicated for transmission and the rest for receiving with independent channels assigned to each

• IEEE 802.11 standard was used

Page 35: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

35

OLSR Multi-channel vs Porposed multi-channel protocol

• At 10s, source S starts traffic sending data to destination node D at a rate of 0.542 Mbps over a channel

• At 12s, a new flow is introduced from node A to node B with a rate of 0.542 Mbps.

• To illustrate, the performance advantage of the proposed routing protocol, the new route as chosen by the proposed routing protocol is activated at 15s.

Topology 1

Page 36: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

36

OLSR Multi-channel vs Porposed multi-channel protocol

OLSR (multi-channel)

Proposed Protocol (multi-channel)

OLSR (multi-channel)

Proposed Protocol (multi-channel)

OLSR (multi-channel)

Proposed Protocol (multi-channel)

Throughput of receiving data Throughput of dropping data

• If the route had been maintained as would have been done by OLSR considering only the number of hops, packets are dropped between 12s and 15s as the receiving throughput goes down

• Once the new routing protocol chooses an alternate route, even though the number of hops is increased from 2 to 3, the throughput is guaranteed and the packets are no longer dropped by the source nodes

Page 37: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

37

Throughput and end-to-end delay analysis

Topology 2

• At 10s,source a flow is introduced from node G to node B at a data rate of 38.125 kbps over the channel

• At 12s, source node S starts traffic to node D• OLSR-MC always chooses the route S->B->C->D with three hops in all• The proposed multi-channel protocol may choose the path S->A->F->E-

>D even though the number of hops is 4• Simulation is run for 20 secs

Page 38: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

38

Throughput Analysis

0 100 200 300 400 500 600 700 8000

100

200

300

400

500

600

Offered load (kbps)

Th

rou

gh

pu

t o

f re

ce

ivin

g d

ata

(k

bp

s)

OLSR-single channelOLSR-MCProposed Multi-channel

0 100 200 300 400 500 600 700 8000

100

200

300

400

500

600

Offered Load (kbps)

Av

era

ge

dro

pp

ed

da

ta (

kb

ps

)

OLSR-single channelOLSR-MCProposed Multi-channel

Throughput of dropping data Throughput of receiving data

• Single channel provides an optimal throughput only for very low data rates when there are no drop data

• OLSR-MC which maintains the minimum number of hops does not fully utilize the advantages of the multi-channel scenario

• At high offered loads the proposed routing protocol, chooses an alternate path that is optimized in terms of the received throughput

Page 39: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

39

End to end Delay

0 100 200 300 400 500 600 700 8000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Offered load (kbps)

En

d t

o e

nd

de

lay

(s

ec

s)

OLSR-single channelOLSR-MCProposed MC Protocol

0 100 200 300 400 500 600 700 8000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

0.2

Offered Load (kbps)

Co

nte

nti

on

tim

e (

se

cs

)

OLSR-MCProposed Protocol

• End to end delay increases drastically in single channel scenarios at very low loads

• Compared to the OLSR-MC, proposed routing protocol has a slightly higher end to end delay for loads greater than 450 kbps due to higher number of hops. However, for such offered loads, the OLSR-MC has dropped packets due to congestion at node B

• Proposed protocol has lower contention delay than OLSR-MC as the source accesses a different channel for the next hop because the route is changed for higher loads and hence lower end to end delay for the proposed protocol

Page 40: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

40

Performance Evaluation

• Performance metric (P) is always higher for the proposed protocol than that of OLSR-MC, even when the energy spent is higher – this is because there is tradeoff for higher throughput and lower end to end delay

Offered load (kbps)

Energy Spent in network, E (J)

End to end Delay, D (s)

Received Throughput, T (bps)

Performance metric, P = T / (E*D)

OLSR-MC

Proposed Protocol

OLSR-MC

Proposed Protocol

OLSR-MC

Proposed Protocol

OLSR-MC

Proposed Protocol

36.8 59 59 0.01 0.01 36.8 36.8 60.32 60.32

91.8 62 62 0.01 0.01 91.8 91.8 143.19 143.19

367 77 77 0.01 0.01 367 367 460.95 460.95

417.7 79 87 0.08 0.013 417.7 417.7 64.18 359.09

495.4 83 91 0.15 0.049 475.4 495.4 37.8 136.78

601.2 83 93 0.2 0.18 475.4 507.4 28.63 30.28

Page 41: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

41

Conclusions

• The proposed routing protocol considers throughput maximization and minimizing end to end delay for determining the route which have been verified with the presented results

• For networks involving higher data rate as in real-time applications, multiple channels and interfaces with the proposed routing protocol are far more efficient

• OLSR-MC which minimizes the number of hops for a source-destination pair does not give necessarily the best route for higher throughput and lesser end-to-end delays– The proposed routing protocol on the other hand,

using its cost-metric, is able to achieve higher throughput and lesser end-to-end delay compared to OLSR-MC for higher offered loads.

Page 42: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

42

Future Work

• Evaluation of the routing protocol:– random topologies– varying mobility– varying number of nodes– different data source, such as TCP/FTP

Page 43: 1 Reghu Anguswamy MS – Systems Engineering Missouri University of Science and Technology 15 May 2008 WIRELESS MOTE –BASED IN-PROCESS DIAGNOSTICS USING

43

Thank you!