Wireless Sensor Networks for Industrial Machine Monitoring

o Introductiono Design Requirementso System Architectureo UC-TDMA MAC Protocolo Implementationso Conclusionso Future Work

By Ankit Tiwari, December 2004

Protocols and MAC

Diagnostician

Prognostician

Scheduler

An intelligent system capable of performing distributed sensing and processing, along with collaborative processing and decision making for carrying out a particular task.

Wireless Sensor Networks –Gathering, Analyzing, & Reacting

Generic Node Architecture…

Power SupplyModule

S ME ON DS UO LR E

ProcessorModule

RADIO

Memory

Converts the quantity to be sensed into signal that can be directly measured and processed.

ATmega @ 4 MHz – 128KB prog memo, 4 KB RAM

SA1110 @ 200 MHz – 4MB Ext flash, 1MB SRAM

Short-range single IC Tx-Rx –ISM band

Stores the data points, routing tables, TDMA table etc required for communication and data processing. 2MB – 4MB

Uses step-up DC-DC converter for constant supply voltage

Condition Based Maintenance

Scheduling

Diagnosis Prognosis

Process

Induced faults, if not taken care, propagates to failure

Performs distributed sensing Obtain measurements for all critical

components of the equipment.

Assesses current state of critical machine components

Performs fault classification Triggers the prognostic module

Maintenance needs? RUL calculations ? Dynamic

time-to-failure updates

Type of maintenance required? Time required to perform?

Total Time available? Schedules the Maintenance

Data Acquisition

Dynamic maintenance scheduling based on instantaneous operating conditions of the machine, so as to have minimum impact on overall functioning of system.

Diagnostics

machines

Math models

),,(),,(

ππ

uxhyuxfx

==&

Math models

),,(),,(

ππ

uxhyuxfx

==&

System Identification-Kalman filterNN system ID

RLS, LSE

Dig. Signal Processing

PhysicalParameterestimates &Aero. coeff.estimates

π̂

Sensoroutputs

VibrationMoments, FFT

FeatureVectors-

Sufficientstatistics

)(tφFault ClassificationFeature patterns for faultsDecision fusion could use:

Fuzzy LogicExpert SystemsNN classifier

Feature extraction -determine inputs for Fault Classification

Physics of failureSystem dynamicsPhysical params.

Identify Faults/Failures

Set Decision ThresholdsManuf. variability dataUsage variabilityMission historyMinimize Pr{false alarm}Baseline perf. requirements

More info needed?

Inject probe test signals for refined diagnosisInformpilotyes

π

Serious?

Informpilot

yes

SensingFault Feature Extraction

Reasoning& Diagnosis

Systems, DSP& Data Fusion

SensorFusion

Featurevectors

Featurefusion

StoredFault Pattern

Library

Model-BasedDiagnosis

no

Request Maintenance

Stored Legacy Failure dataStatistics analysis

Background Study Identify Fault Pattern Library

Physics Based Fatigue ModelingFault Mode Analysis

Failure mode classification Failure Event <~>Root cause

Best Feature VectorsMeasured Outputs needed for FV

Failure & fault ModesMeans for detecting incipient faults

Using Legacy DataSystem Identification

Digital Signal Processing

Online Phase

Offline Phase

[F. L. Lewis]

Prognostics

Prescription Libraryfailure modestrendsside effects

Rulebase expert systemFuzzy/Neural SystemPrescription decision treeBayesianDempster-Shafer

Maint. Request

Maint. Planning & Schedulingweight maint. Requests

Computer machine plannersHTN, etc.

Performance Priority Measuresearliest mission dateleast slack repair timedue date

RULEstimated time of failure

Mission criticality and due date requirements

Maintenance Requirements Planning

Maintenance PrioritiesMission Due Dates

safetyriskcost

opportunityconvenience

Automatically generated work orders.Maintenance plan with maint. Rankings

Resource assignmentand dispatching

priority dispatchingmaximum % utilizationminimize bottlenecks

resources

PrioritizedWork Ordersassigned toMaint. Units

Guaranteed QoS

User interfaces forDecision assistanceDecision Support

Adaptiveintegrationof newprescriptions

StoredPrescription

Library

Medical HealthPrescriptions Manufacturing MRP

Communications SystemScheduling & Routing

ManufacturingOn-Line ResourceDispatching

Generate:optimized maint. tasks(c.f. PMS cards)

Prescription

Scheduling

Priority Costs

Dispatching

DiagnosticFaultcondition

Background StudyFault Mode Time Analysis Remaining Useful Lifetime Analysis

Best Feature Combinations Best Decision Schemes FC MTTF & MTBF from Legacy DataFailure Time Pattern Library

IdentifyPredict

Remaining Useful Lifetime Time-to-Failure

Account For -

Fault Propagation & Progression Dynamic Time-to-Failure

Offl

ine

Phas

e

Onl

ine

Phas

e

Maintenance prescription & Scheduling Procedures [F. L. Lewis]

Data Acquisition

Force

Velocity

Position

TemperatureVibration

Flow

Pressure

Small-Size Induction Motors

Current

Current Voltage

10KHz – 20KHz

Up to 20KHz

Up to 20KHz

Vibration

Velocity

Quantities Measured

Sampling Rates

Modular I/O NetworkEmbedded PC

4 V / I – Inputs/Outputs

Configurable per channel

Data rate – 100Mbps

Matchbox PC, MAX-PC, etc

Processor – 300 MHz, Pentium Class

SDRAM – 256 MB Hard-drive – 14 GB

Ethernet, CAN, RS-232NI Fieldpoint, MAX-IO, etc

12-bit resolution

Plug and Play operation

Hardware

Design Requirements

Continuous SensingPeriodic Data TransmissionUser-Prompted Data QueryingEmergency Addressing & AlarmsReal-Time TransmissionAdaptabilityNetwork Re-configurabilityScalabilityEnergy EfficiencyFeed Back Control

System Architecture

Base Station

Base Station

SN

SN

SN

Feed Back

Feed Back

DataBase

DataBase

DisplayDisplay

AnalysisAnalysis

Battery Operated

Sensing Nodes

Central Control

Data Interpretation & Decision making algorithms

with high Computational requirements

Fault Pattern

Libraries

Prescription

Libraries

Conclusions &

Decisions

Minor Control or

M/C Resetting

Efficient & Energy Aware MAC protocol

Physical Layer FunctionalityKnow-How of Application Layer

along with Physical Layer

How to incorporate physical layer functionality as well ?

Some approaches:Hardware Energy Models: Shih, Chandrakasan, et al. (MIT) ACM SIGMOBILE’02Systematic power analysis : Raghunathan, Srivastava, et al. (UCLA) IEEE Sig. Proc. Mag’02

For direct applicability, obtain energy consumptions in mA-Hr.

The Battery Consumption

Equation

Hardware Determined Parameters

Irx, Itx, Trx-tx, Trx-s

Trx , Ttx , TsProtocol Determined

Parameters

Battery Consumed in mA-Hr

SYMBOLIC RADIO MODEL

Receiver Electronics

Transmitter Electronics

Irx

ItxItxm

RF_IN

RF_OUT

Other Elex

Is

Gets Activated in transmit mode

Total current drawn in transmit modeActual Modulation Current which

produces RF_OUT

Antenna

Gets Activated in Receive modeTotal current drawn in Receive mode

Electronics for driving radio in either modes

Sleep mode Current

The Battery Consumption Equation

( )[ ] ( )[ ]( )[ ] ( )[ ]( )[ ] ( )[ ] onturnrx/txsstxsstxssrxssrx

rxrxtxrxrxtxrxrxsrxrxs

txtxmtxtxrxtxtxrxtxtxmtxtxstxtxs

TITTINTTINTTINTTIN

TITTINTITTINAmpHrs/Hr

−−−−−

−−−−

−−−−

+++++++++

+++++=

Number of times per hour, radio switches totransmit mode from sleep/receive mode

Time taken by radio to switch to Transmit mode from sleep/receive mode

Actual time for which radio transmit,each time it is in transmit mode

Actual time for which radio receives, each time it is in receive mode

Time taken by radio to switch to receive mode from sleep/transmit mode

Number of times per hour, radio switches toreceive mode from sleep/transmit mode

Itx> Irx > Is

Topology – Energy ConsiderationsMany-to-One Communication Paradigm

Single – Hop TopologyMulti – Hop Topology (d2

AB+d2

BC< d2

AC)

RFM Datasheet

RF Output Power – 1.5dBm

Tot. current in Tx sec. – 12mA

Curr. contri. to RF O/P – 0.45mA

12.7 meters

Current Drawn – 18.1mA

5.3 meters

5.3 meter

s

Current Drawn – 13.7mACurre

nt Drawn –

13.7m

A

Total Current Drawn 27.4mA

Min. 11mA currentconsumed by transmit section of each node for every transmission

Assumptions12-bit encoded data at 19.2 kbps using OOK modulation No threshold at the receiver. A 3 dB filter bandwidth of 14.4 kHz is used (noise BW = 1.25 * 3 dB BW).A receiver noise figure of 7.5 dB is assumed.Antennas with 1 dB of gain are used. A 20 dB fade margin is chosen (99% Rayleigh probability).Packets are 38 bytes long (excluding preamble), or 456 bits.System goal is to achieve 90% packet reads on the first try.The operating frequency is 868 MHz. Assuming 20 dB fade margin and 1 dB transmitter/receiver antenna gain:

Po+ 80.9 dB = -27.6dB + 20log(F) +40log(D), F in MHz.For F= 868 MHz, Po + 49.73dB = 40log(D)

Allowed Path Loss = 80.9 dB +PO (in dBm)

Remaining Node circuitry & Protocol Overhead not Considered

Chipcon Transceiver

Topology

Developed Single-Hop Topology

Latency Requirements

Energy Constraints

Considering The

Simplicity @ Nodes

To Avoid

Loss Of Data

Hot Spots Routing

Min Control Overhead

Centralized Control

Max Throughput

To Provide

Sleep-Listen duty-cycle – Revelation

In just switching 0.05952 mA-hrIn transmitting data for 17.7 Seconds 0.059 mA-hr

Transmitting @ 1000 sweeps/sec, 17.7 Sec transmission 17700 Data Points from each channel

RFM Radio, In 1 Hr Reason – High Switching time (Ts-rx)

L LS

30 S

LLL SSSS

30 S 30 S30 S

Battery Consumption

Schedule sleep times such that nodes sleep for maximum possible duration with minimum possible switching frequency.

MAC Protocol – Relevant Work

Contention based protocol.Sleep/Listen duty cycle for Energy Saving.Inspired from PAMAS, avoids overhearing In-channel-signaling. Sacrifices the latency requirementsIgnores the throughput considerations.Necessitates sleep schedules at all the nodes.

S-MAC – Ye, Estrin, et al. (UCLA)

Contention based protocol.Adaptive Sleep/Listen duty cycle.Adaptively ends the active part of DC. Activation Events. Messages arriving immediately after time TA suffers high latency.

T-MAC – Dam & Langendeon (Delft Univ.)

TDMA based protocol.Balance the energy consumption of network.Different Sleep/Listen duty cycle for nodes.More energy-critical nodes Sleeps longer. Nodes sleep only in its own time-slot.Each node maintain two-tuple receive table. Overhearing & Protocol Overhead Problem.

ER-MAC – Kannan, et al.

TDMA based protocol.No TDMA table at nodes.Runs Real-Time scheduling algorithm.Trades-Off Memory with Computation. Interference between Network and Application Computational tasks.

Contention free scheduler MAC – Carley, et al.

MAC Protocol

Negligible Protocol

Processing @ nodes

Scheduling & Contention

Energy SavingsSleep/Listen DC Minim

al Contention

Overhead @

nodes

Low Protocol Overhead

Developed UC-TDMA MAC Protocol

Overhearing Reduction

Collision Avoidance

Deterministic

Slot Allocation

UC-TDMA Frame

Rate of change of Current > Rate of change of Vibration > Rate of change of Temperature

How to allow application to take advantage of established relationship between measurements ?

Let Application/User configure the Time Frame

Frequency ExtractionLength of Vibra. data > Length of Press. data

If 60 < OilPress <= 84 and EvaporatorPress <= 2.65, then Refrigerant is contaminated (73% confidence from the data)

If Chilled Water Inlet Temp. is above 53 degrees and Outlet Temp. is above 44 degrees then Switch to Overload Mode

For AC Plant

Frame 1

N….3121N….3121 …..

Frame 2

Time

Nodes Might access Channel more than Once in given frame

Length of slots for different nodes can be differentTrades Off Fairness at each node

UC-TDMA MAC Protocol

Explicitly Define Data Collection Sequence.Establish Relationship between two measurements & draw Conclusions.TDMA base offers collision avoidance & energy preservation.

Provides On-board memory for in-network data processing

No need to maintain table at any of the nodesCentral BS maintains TDMA table for all nodes

Saves Memory & Complexity at Nodes

Sleep Schedule Calculations

[ ]( )[ ]( )[ ] )3(3600

)2(

givenRateupdatingiffTNdiagRS

givennotRateupdatingiffTNdiagNTS

SdiagUT

Tp

Ts

Tud

Tp

Ts

Tspd

rp

×−×=

×−×=

÷=

Given, sweep rates for all node, number of data points from each node, frequency at which each node transmits (every r hours), the sleep durations for all the nodes in network is given by :

Sweep Rate Matrix (1Xn)Time Period Matrix

No. Of Data Points Matrix (1Xn)

Sleep Duration Matrix(in Sec)

Total time taken by all nodes for transmitting their data

Time taken by each node in transmitting its data

Sleep duration for any given node is Total time – its own transmission time

Updating rate – approximate sleep Duration for that particular node

If Updating rate is 1 hr for some node which transmits for 2 sec in each slot => the node will tx 2Sec, then sleep for 3600-2sec, and then again repeats..

Updating Rate Matrix

UC-TDMA – Hybrid ProtocolScheduling AloneScheduling Alone Contention AloneContention Alone

Clock Drift ProblemsEmergency addressing ProblemsRe configurability ProblemsScalability Problems

Energy InefficientSleep/Listen DC Implementation ProblemsHigh Control Packet OverheadRTS, CTS message collisionsHigh protocol processing (NAV, backoff timer)

Wise to spend some energy in Contention along with Scheduling

Fusion of Two Access TechniquesFusion of Two Access TechniquesSeeks to

Minimize major energy wastage sources viz. Collisions & Protocol overheads.Add Flexibility to the overall paradigm.

Exploits Resourceful Base Station.Point of control in the architecture

Generating Virtual RTS….

BS knows which node in N/W has access to channel at any particular instant.Instant any new node acquires the ChannelBS assures, No other node is talking to it.Itself generates Virtual RTS on behalf of that node.Node is ready to receive CTS from BS.

Mechanism…

BS

SN

1] CTS (node addr. a

ppended)

2] Data Points (predetermined)

3] Request-to

-Sleep

SN

SN

Node Perspective…Collisions /Contention

Overheads ZERO

Huge Energy Savings(Attributed to short packet transmission prevention)

RF Channel Acquisition Processing @ Node ZERO

Node Simply Sleeps, Wakes-Up according to Timer

Start

Status Quo

Set J=0

Is Node missing?

Calculate sleep schedule for each node

Is J > 0

Configure nodes with defined functions & sleep schedule

Set i=1, J=1, S=n+1

Is J > =10

Is S > n

Retrieve data from node i

Read node type, data rate no. of data points & sequence no.

Insert node in existing slot sequence assigned

Remove failed node from TDMA slot sequence

B.S. pings for new node.

Report user about missing node & node type

Set J=1

Add new node?

Set S=S+1

Append sleep schedule command data to node i.

Set S=1

Any Data?

Node i sleep

Is i=n?

Set i=i+1

Stop?

Stop

Set i=1, J=J+1

No

No

YesNo

Yes

No

Yes

Yes

User Interface

Functionality definition for each node

B.S. Checks availability of all defined nodes in N/W

Yes

Node Parameters like – Sweep Rate, No. of Sweeps, Node No., Sequence No., Active Channels

New Node Addition

Failure of Existing node

Emergency Addressing

Nodes keep sensing & comparing the sensed value to set threshold value, while radio is asleep.

Anytime sensed value exceeds the set threshold, wakes up its radio and transmit its node address to BS until responded.

Channel Occupancy Causes Collisions, resulting in continuous checksum error at BSBS interprets these continuous collisions as an indicator of emergency. Hangs up the ongoing operation and receives the address of the node in emergency.

UC-TDMA – Modes Of Operation

Continuous Mode Useful for newly deployed networks.Try to Answer - How frequently data should be collected from various sensor nodes ?Collect data continuously and sequentially from all nodes.Keeps base station busy all the time – Achieving Max ThroughputSleep durations given by equation (2).

Non-Continuous Mode Useful for previously operational networks.Updating Rates can be obtained Using C-Mode data analysis .Generates lesser Data Traffic.Different Nodes can have different Updating rates.Nodes Sleep most of the time => longer System life time.Sleep durations given by equation (3).

State Machine at Nodes

Sleep State

Receive State

Setup State Transmit State

Time out

Set cmd

Emergency

Data out

Transmit cmd

Sleep cmd

Done

It takes 69.78 % more Energy to startup Node in Tx Mode than that in Rx Mode

Looks For Commands from BS

Sets up Various Node parameters

Radio Turned Off, Continuous Sensing

Transmits Data or Parameters Desired

Implementations

Hardware Used V-Link Wireless NodeG-Link wireless NodesSG-Link Wireless NodeBase StationExternal 9 V BatteriesLaptop (Intel Pentium IV – 1.99GHz, 256 MB RAM)USB2Serial Converter

Software Used MATLAB version 6.5.1LabVIEW version 6.1MS Windows XP Home/Professional Ed.

G-Link V-Link

MATLAB Implementation

Data link layer – To establish an RF communication link.A serial link between BS and Terminal PC.Terminal program to issue commands to base station for communicating with wireless node.

Acceleration along 3-Axes

Rea

l-tim

e D

ispl

ay

MATLAB - LabVIEW

MATLAB graphics are inherently slow.Creating MATLAB GUI – Not Very FlexibleLabVIEW – Intrinsic support for real-time data acquisition.LabVIEW – Flexible GUI development MATLAB tools like DSP, Fuzzy Logic, Neural Networks, Statistical Analysis, etc , Required for advanced data processing and interpretation.

Implementation Architecture

Neural N

et

Artificial Intelligent

Fuzzy Logic

Neural N

et

Artificial Intelligent

Fuzzy Logic

Path to Decision & Display

Signal DataTransition

Information

Knowledge

Wisdom

Display

Display

Fast & Efficient RTDA tools, GUI Development tools

Easy to Implement Data Processing, Analysis & Interpretation tools

Implementation Architecture

OSI Layers Addressed

Data Link

Physical

Presentation

Session

Transport

Network

Application

Provided

UC-TDMA MAC Protocol

Application GUIs in LabVIEW

Provides all the services required by Application layer

OSI Layers

Physical InstallationOptimal Locations Small Form Factor

Tight placement

LabVIEW Implementation

Two separate GUIsNetwork configuration wizard

Engineer’s interfaceTo specify various Network parametersDifferent Configuration Files for different operation-phases

Application GUI Sets up Node Parameters by using Config. File Creates UC-TDMA frame by using Config. FileSequences real-time display windows for configured nodesAcquire, Process, & Display the data in RT in corresponding display windowsCalculates & Display FFT of Time-domain data acquiredStores raw data.

Network Configuration Wizard

Useful for making minor changes to node parameters

Loads with Default Values for Parameters

On Clicking, Current/default settings for that node appears in the next screen

Try to Eliminate Node Naming Issue

Select channel nos. on the node

from which to acquire data

No. of data points to acquire

from each selected channel during each time slot of this node

Select sensor node to be configured

Data Sampling Rate (1 sweep=1Sample from all active ch)

Comm. Port connected to Base station?

Transceiver address of

selected node

Node with Sequence no. 1 is the first to

begin data acquisition cycle.

All these settings are saved in a configuration file,so user need not to configure network every time.

Application GUI

Time-Varying FFT Calculated using Data Obtained in One Time Slot

Vibration Signature at any Instant

Continuous Mode of Operation

No. of Sweeps – 2000 Sweep Rate – 829 sweeps/sec Active Channels – 3 (X, Y, Z) Data Loss per Slot – 1 % (approx.) Delay bet. two acqui. – 200 msec

Sweep rate & no. of sweeps ascertains the time duration of TDMA slot for the node.

Thus, length of the slot for any particular node can be defined.

Sequence number resolves the position of slot in the frame.

Conclusions

Application and Physical layer driven design – Surprisingly favorable results.Sensor deployment scenario Consideration – Simplifies the designSingle-hop transmission - Best topological solution for CBM networks.UC-TDMA MAC – Prototype for large-scale deployment

UC-TDMA MAC

UC-TDMA MAC Protocol

UC-TDMA MAC Protocol

UC-TDMA MAC Protocol

Clustering Algorithm

Security Protocol

Data Processing

RF-Link efficiency Metrics

Coordinated data & protocol processing

Heterogeneous Sensor Nodes RF Channel – drawbacks into properties

Multiple Modes of Operation

Utilizing Sensor deployment Scenarios

Other Key Ideas

Future Work