Rating Driver - BEST Project Rahul Mundke Master’s Student, KReSIT IITB.

Rating Driver - BEST Project

Rating Driver - BEST Project

Rahul MundkeRahul MundkeMaster’s Student, KReSIT IITB

RoadmapRoadmap•Motivation•EDR Overview•Overall System•Project Targets•Skills/Knowledge Areas to pick up•References

MotivationMotivation

• BEST( Brihan Mumbai Electric Supply and BEST( Brihan Mumbai Electric Supply and Transportation ) wants to monitor and Transportation ) wants to monitor and record per driver Km/Liter statistic. BEST record per driver Km/Liter statistic. BEST wants a mechanism to perform the fair wants a mechanism to perform the fair rating of drivers based on driving rating of drivers based on driving quality.quality.

About BEST About BEST

• Formally set up in 1873 as Bombay Tramway Formally set up in 1873 as Bombay Tramway CompanyCompany

• Mumbai saw it’s first bus run in 1926Mumbai saw it’s first bus run in 1926

• At present has 3380+ buses carrying 45 lakhs At present has 3380+ buses carrying 45 lakhs passengers daily on 335+ routespassengers daily on 335+ routes

• There are 22000+ employees.There are 22000+ employees.

• On an average 2.4 drivers operate on a Bus On an average 2.4 drivers operate on a Bus dailydaily

• Mileage calculations are done at the end of Mileage calculations are done at the end of day. There is no way to estimate the individual day. There is no way to estimate the individual driver performance by this reading.driver performance by this reading.

• Current average mileage of bus is 3 Km/Liter, Current average mileage of bus is 3 Km/Liter, BEST aims to raise it to 3.5 Km/Liter by various BEST aims to raise it to 3.5 Km/Liter by various initiativesinitiatives

Source : BEST website (http://www.bestundertaking.com/ )Source : BEST website (http://www.bestundertaking.com/ )

Event Data Recorder Overview Event Data Recorder Overview

• What is EDR ?•A physical connection that could have major damage but is still “probable” at some time after the event. Perhaps needs to be extracted from the vehicle• A physical connection that could have major damage but is still capable of being “plugged to” by some diagnostic device. Perhaps still connected to the vehicle.• A wireless device that could transmit the data at the scene to medical personnel.• A wireless device that could transmit the data automatically at the time of the event.

• Uses• Opportunity to look “inside” the crash event• Remove one level of uncertainty in collision analysis• Allows injury causes and injury mechanisms to be better understood

Source : Source : Part of Part of Draft Specification. (IEEE P1616 : MVEDR Specification))

EDR Pre-Crash DataEDR Pre-Crash Data

ON0102434-1

ON0115241-2

OFF2128044-3

OFF16134444-4

OFF9134444-5

Brake Switch Circuit Status

Percent Throttle

Engine Speed (rpm)

Vehicle Speed (mph)

Seconds before AB

Event

Deployment

Non-DataPre-Crash

Source : Source : Chalmers University of Technology. (Summary of Swedish Research Using Crash Recorder Data))

Overall System Overall System

Development PlanDevelopment Plan

System would be developed in multiple versions System would be developed in multiple versions

For First Version:For First Version:

• System will record and display raw data so as to help in System will record and display raw data so as to help in understanding/discovering new principles/concepts.understanding/discovering new principles/concepts.

• Cost of the EDR is not a major concern, time to develop is!!Cost of the EDR is not a major concern, time to develop is!!

• Expected to be complete in 3 monthsExpected to be complete in 3 months

Sub-Sequent Versions:Sub-Sequent Versions:

• System should be superior which also meets auxillay goals of System should be superior which also meets auxillay goals of First version like cost , form factor etc.First version like cost , form factor etc.

Block DiagramBlock Diagram

Tentative Application LookTentative Application Look

Project TargetsProject Targets

•Development of an algorithm to rate the driver• Approach

• Study of the real data to see the patterns ( Data mining would help!!)• Sensor data mining papers would help in this regard (Neural

networks?)• Literature survey to find the various rules/relations with parameter(s)

to determine quality of driving• Study of the research data which Automotive Research Group,India,

did on BEST buses• Trying to get the Indian driving cycle condition data from the Bajaj

Auto Pvt. Ltd.

•Research Publication

• Possible topics• The study of various classifiers used for data mining approach and

comparison of their performance• If acoustic spectrum analysis works, we could have some papers on

it?• Data Compaction Techniques ?• A publication on findings of the system?

Approach towards meeting TargetsApproach towards meeting Targets

Raw Data Files Raw Data Files containing sensor containing sensor readingsreadings

Rating Application Rating Application

ApproachApproach•Start with Simple•Don’t wait for real data•Data Mining Study•Neural Network/Genetic Algorithm ?

Proposed Prototype for AlgorithmProposed Prototype for Algorithm

•We would work on 2 algorithms • Expert Algorithm

•Would be a greedy algorithm giving good/bad rating based on current data like

- Brake, Accelerometer, Clutch position- Speed, acceleration of vehicle Vs. Gear Position

• Classifier Algorithm•Based on the collected/simulated(?) data,

this would based on a history as well and would run over a longer duration of data.

ChallengesChallenges•We will need to quantify some the parameters

which decide the quality of driving• Ability to read road ahead• Attention to driving task• Careful to operate on recommended settings

•Getting accurately tagged real data•Accounting for the variations of data based on

route, day, time, vehicle (work on 1 month data ? )

• Individual Differences among drivers• Physical and Cognitive ability• Experience• Attitude

Approach towards meeting TargetsApproach towards meeting Targets

Operating System LayerOperating System Layer

Utility LayerUtility Layer

Rating Application Rating Application

Sensor 1Sensor 1 Sensor 2Sensor 2 Sensor NSensor N

Plugin Plugin FunctionalityFunctionality

Something About SoundSomething About Sound

Engine Sound generated at different speeds Engine Sound generated at different speeds generate generate

Different sound perceivable to human ear !!Different sound perceivable to human ear !!

Skills RequiredSkills Required

Data Mining:• Different Classifiers, Feature Extraction• Various Papers on Sensor data mining• Various tools like SAS,SPSS• New method ?

Digital Signal Processing• Digital Filtering of noise• Adaptive Filter• Short Time Fourier Transform• Compaction of data ?

Domain Knowledge on Automotive• What differentiates a good (not to be confused with safe) driver from another driver• What sampling rate is sufficient to give good estimate of the result• Field experience?

SummarySummary

What is to be delivered as system and as M. Tech project is well identified

How it can be done is also identified for the current requirements

Links ,Places and People who would help in system development are identified for current requirements

ReferencesReferences• Few papers from journal on Vehicle Safety • Energy Conservation websites giving the

techniques for mileage improvement• Papers from the Technical Conference on the

Enhanced Safety of Vehicles,2002,2003• White papers from the National Highway Safety

Traffic Administration Symposium• Autobait Daily Journals for various statistics• www.ece.wisc.edu/~secsit : some papers on

the feature extraction and classification