Generation of in-house accelerated brake block cycle

from Real World Usage Pattern of field vehicles &

intimation of brake pad replacement to customers
Dissertation

By

Sunil Ramesh Malage

Dissertation work carried out at:

Tata Motors Passenger Vehicle Limited, Pune

Submitted in partial fulfillment of the requirements of the

MBA Business Analytics Programme

At
Tata Motors Passenger Vehicle Limited, Pune

Nav Sahyadari Charitable Trust’s

INSTITUE OF MAMAGEMENT SCIENCE, PIMPARI BK
Tal - Pimpari(BK),Tal- Khed, Dist - Pune – 410501, Maharasharta, India

( Nov 2024 )

Page 1
 Nav Sahyadari Charitable Trust’s
INSTITUE OF MAMAGEMENT SCIENCE, PIMPARI BK
Tal - Pimpari (BK),Tal- Khed, Dist - Pune – 410501, Maharasharta, India

CERTIFICATE

This is to certify that the Dissertation entitled “Generation of in-house accelerated

brake block cycle from Real World Usage Pattern of field vehicles & intimation of

brake pad replacement to customers” and submitted by Mr. Sunil Ramesh Malage in

embodies the work done at Tata Motors Ltd

Signature Name - Sunil R Malage

Place: Pune

Page 2
 Nav Sahyadari Charatible Trust’s
INSTITUE OF MAMAGEMENT SCIENCE, PIMPARI BK
Tal - Pimpari(BK),Tal- Khed, Dist - Pune – 410501, Maharasharta, India

DISSERTATION

Dissertation Title: Generation of in-house accelerated brake block cycle from Real World Usage Pattern
of field vehicles & intimation of brake pad replacement to customers

Name of Student: Sunil Ramesh Malage

Abstract

Brake system is safety critical part of vehicle to control the speed. Hence, Brake pad wear monitoring is
essential parameter to ensure the Passenger’s safety. Therefore finding the right combination of brake pad
material & its composition is always been challenging. The Brake pad & discs need to be designed in such way
that life of brake pads are acceptable in extreme driving conditions & cost of manufacturing is also need to be
lowest. Companies spend lot of time & money for testing of different brake pad materials on in-house testing
& vehicle level testing to achieve the best possible combination.

This research work enlighten the methodology to predict the brake pad wear using Machine-learning model.
The Machine-learning model uses Initial speed, final speed, deceleration, brake energy & temperature for
generating in-house accelerated brake block cycle and predicting the replacement interval (wear) of brake
pad with an accuracy sufficient for real world brake pad wear usage.

Key Words: Brake Pad, Wear, Brake energy, Brake deceleration, Temperature, Machine learning

Page 3
Please prepare the outline as a separate document with the following sections along with the above identification
information-
1) Student details: Name
2) Date
3) Dissertation Title
4) Problem statement (what is the problem being addressed)
5) Literature review
6) Research Methodology
7) Analysis & Design
8) Data Pre-processing
9) In house brake block cycle generation
10) Model configuration & evaluation
11) Summary
12) Future work
13) Detailed Plan of Work

Table of Contents

1) Chapter -1: Introduction 10

1.1 Problem statement 10

2) Chapter-2: Literature review 11

2.1 Introduction 11

2.2 Definitions 11

3) Chapter-3: Research Methodology 13

3.1 Introduction 13

3.2 Research Approach 13

3.2.1 Data collection 13

3.2.2 Data understanding & description 13

3.2.3 Data Pre-processing techniques 14

3.2.4 Data visualization 16

4) Chapter-4: Analysis & Design 17

4.1 Introduction 17

4.2 Dataset Description 17

4.3 Data Cleaning 17

4.3.1 Removal of Irrelevant column 18

4.3.2 Correction of column data types 18

4.3.3 Exploratory data analysis 18

Page 4
 5) Chapter-5: Data Preprocessing 22

6) Chapter-6: In house Brake block cycle generation 22

7) Chapter-7: Model configuration & Evaluation 23

7.1 Model evaluation 25

7.2 conclusion 26

8) Summary 27

9) Future work 27

List of Figures

2.1 Brake wear/life prediction 11

3.3.1.1 Missing data in dataset 14

3.3.1.2 Drift data in dataset 15

3.3.1.3 Spike data in dataset 15

3.4 Correlation between temperature & brake deceleration (g) 16

4.2 Acquired data from data logger (Pune City) 17

4.3.2 Correlation of column data type 18

4.3.3 Exploratory data analysis 19

4.3.3.1 Brake energy kill estimation 19

4.3.3.2 Brake deceleration calculation 20

4.3.3.3 Brake deceleration brake count 20

4.3.3.4 Brake temperature band & count 21

4.3.3.5 Brake deceleration count distribution plot 21

6.1 Brake deceleration comparison of field & dyno test 22

7.1 Brake deceleration, energy, wear & life dataset 23

7.2 Random forest algorithm code 24

7.3 Logistic regression algorithm code 25

7.4 Random forest algorithm evaluation & accuracy 26

7.5 Logistic regression algorithm evaluation & accuracy 26

Page 5
 List of Table

3.2.2 Acquired parameter description 14

6.1 Brake deceleration comparison of field & dyno test 22

6.2 Brake duty cycle comparison of field & dyno test and proposed brake duty cycle 23

List of Abbreviations

1) EDA- Exploratory data analysis

2) ML- Machine Learning

Page 6
 Chapter-1 Introduction

1.1 Problem Statement

Now a day’s Vehicles are becoming more powerful and resulting high speed. But due to road
limitation braking performance is playing more vital role in Real World Usage Pattern condition.
Brake pad replacement is happening as per vehicle service-schedule interval for field vehicle
(service at each 15k kms) despite without Brake pad fully wear out, which leads to higher brake
pad consumption in field & cost to customer before actual life of brake pad on account of brake
pad under-utilized. In other way round if brake pad is not replaced then there is highly chances
that it will worn out in between 15-30k kms before next service, which leads to adjacent part
failure like brake disc.

To address this scenario in field of brake pad replacement (ensuring optimum utilization of brae
pad and avoiding pre mature failure of adjacent part), this research work carried out. The
current research project focus in the following application areas:

 Creating the accelerated in house brake block cycle using real world usage data &
drive pattern for robust validation and intimation of brake pad replacement to customer.
 Verifying, fine-tuning& implementation of the models in Machine learning for brake
pad replacement prediction
 Modifying existing block cycle for in house testing for robust validation

Page 7
 Chapter-2 Literature Review

2.1 Introduction

Image-2.1 Brake wear/life prediction

Currently in Automotive industry brake pad service life prediction is based on rig level
(Block cycle- SAE Std: J2707_202106) test & vehicle level testing. Evolution of modern
braking technologies has significantly changed the way how automobiles are made and
controlled.

Currently there is no methodology to predict life of wear & tear items like brake pad based
on consumer usage pattern. Having real world usage brake block cycle for in house test will
be useful for reducing new product development cycle. Prediction of brake pad replacement
intimation through digital will become a differentiator factor in making buying decision of
vehicle to customers.

2.2 Definitions

1) Brake Energy (mj): Brake energy generated during each braking event.

Brake Energy = ½ *{m*((Vi)^2-(Vf)^2)}

Where, m= weight of vehicle

Vi= Vehicle speed at the time of brake press
Vf= Vehicle speed after brake release.

2) Brake Deceleration (g): Brake deceleration is nothing but vehicle deceleration

during braking event & it’s classification based on intensity (g).

Brake Deceleration = {(Vi-Vf)/(Ti - Tf)}

Page 8
 Where, Vi= Vehicle speed at the time of brake press
Vf= Vehicle speed after brake release.
Ti= Time at the time of brake press
Tf= Time at the time of brake release.

3) Cumulative Brake energy kill (mj): Cumulative brake energy kill is summation of
brake energy kill in each braking event.

Brake Energy = Σ ½ *{m*((Vi)^2-(Vf)^2)}

4) Brake wear (mm): It’s nothing but your brake pad thickness difference between
initial pad thickness & current pad thickness.

Brake wear = Initial pad thickness – Current pad thickness

5) Brake Life (km): Brake life is nothing but brake pad replacement interval. Brake life is
replacement once brake pad reached to Brake wear indicator thickness (BWI) on
pads.

Brake Life = Current kms * {(Initial thickness – unusable thickness) / (Initial thickness
– Current thickness)}

Page 9
 Chapter-3 Research Methodology

3.1 Introduction
This section addresses possible systematic approaches to achieving research goals. First,
let's take a closer look at all the columns present and the data set we chose for this study.
We also discuss the sources of data collection and why this dataset was chosen for this
work. We then detail all relevant text preprocessing techniques performed on this dataset,
followed by a comprehensive exploratory data analysis. In the next section, we then discuss
the techniques for brake energy, brake temperature & brake deceleration effect on brake
wear (life) & any other additional functionality we want to derive from the existing
functionality. The modeling section that follows describes the model types chosen for the
correlation of braking energy, temperature, and deceleration under wear, the details of the
model architecture, and the required input representations. Having explained the
advantages of this model for our task, let's move on to creating other variants of this model.
After the model building phase, we perform model evaluation by computing the relevant
classification metrics and compare all the models considered in this section in detail with
the computed metrics. This chapter concludes with a description of the various hardware
and software requirements required as part of this overall research methodology exercise.

3.2 Research Approach

3.2.1 Data collection:-

To work on problem statement stated in chapter-1, we need to acquire various parameter
vehicle CAN/telematics data in different cities on India to build precise model for prediction
of brake wear (life) based on customer usage pattern. Currently vehicle CAN data acquired
for Pune city for understanding Brake wear in metro city. In Pune City various parameter
data is acquired like brake temp, Clutch Press state, Eng Torque, Acceleration pedal
position, Brake press state, Engine Speed, Coolant temperature, Vehicle Speed, Vehicle
Deceleration & Acceleration & odometer reading.

3.2.2 Dataset understanding & description:-

Once data is acquired for Pune city, next step is understanding of data and its description.
Refer below table (3.2.2) for understanding of description of acquired parameters.

Sr No Parameter Name Data Description

1 FRH Outer Front RH Outer brake pad temp
2 FRH Inner Front RH Inner brake pad temp
3 FLH In Front LH Inner brake pad temp
4 FLH Outer Front LH Outer brake pad temp
5 RLH_Liner Rear LH brake liner temp
6 RRH Liner Rear RH brake liner temp
7 FRH Shim Out Front RH Outer brake shim temp
8 FRH Shim In Front RH Inner brake shim temp
9 FLH Shim In Front LH Inner brake shim temp
10 FLH Shim Out Front LH Outer brake shim temp

Page 10
 11 EngTrqMe Engine Torque
12 ClutPressStateEMS Clutch Press state
13 AccPedPosn Acceleration pedal Position
14 BrkPressState Brake Press state
15 EngSpd Engine Speed
16 ClntTemp Coolant Temp
17 VehSpdGS Vehicle Speed
Vehicle Acceleration & Vehicle Acceleration &
18
Deceleration Deceleration
19 Odo Reading Odometer reading value
Table 3.2.2 Acquired parameter description

3.3 Data pre-processing technique:-

Raw data obtained tends to contain many data quality issues, such as missing data, spikes,
and data drift, so input for the next steps in the channel, such as parameter derivation,
correlation, and model building not suitable for feeding as below are various preprocessing
techniques performed on the data. Please note that depending on the quality and relevance
of the data available, some of the steps below may be skipped or additional steps not
included in the list below may be performed.

3.3.1.1 Missing Data- The acquired data will have some missing values due to logger not
working, Telematics unit failure, human error (driver not switched on logger),
vehicle hardware error (OBD not working). Refer below image for data missing.
Refer below image for data missing-

Image 3.3.1.1 Missing data in dataset.

3.3.1.2 Data Drift- During data acquisition some parameter may show drift in data due

Page 11
 to various issues like (permanent set of temperature or hardware module). Refer
below image for data drift. Refer below image for data drift-

Image 3.3.1.2 Drift data in dataset

3.3.1.3 Spike detection- Essentially, it determines the average difference between data
points. Spikes are more likely to occur when data exceeds multiples of this value
in succession. Refer below image for spike detection

Image 3.3.1.3 Spike data in dataset

Page 12
3.4 Data visualization:-
Data visualization not only helps you get a feel for our data, but it also helps us to identify
hidden patterns and trends in our data. This helps provide useful insights that support the
feature engineering process.

Correlation Analysis between various parameter. E.g

Image 3.4 correlation between temperature and brake deceleration (g)

Page 13
 Chapter-4 Analysis & Design

4.1 Introduction
This section presents all the steps involved in model design, starting with data preparation
and preprocessing, through model configuration and experiment setup. First, let's
understand the data by looking at some examples, the number of rows present, features (all
splits - training, validation and testing). These are readily available. Then, see if our data has
missing entries/rows, spikes, or deviations, visualized in the form of graphs. This is followed
by a detailed exploratory data analysis phase where correlation analysis (using bar charts) is
performed.
This is followed by a section, which includes distribution plot to understand the how much
brake energy kill and higher brake deceleration band in overall data. Next, we discuss all the
most commonly used preprocessing and data cleansing steps for our data and how they
relate to our data.

After that, we will discuss different models to implement on the dataset. This section ends
with a flow chart of the complete research methodology that provides an overview of all
the steps carried out in the framework of this paper.

4.2 Dataset description

The datasets available to work with are in .csv format. Below is an image of the first few
rows of the dataset:

Image 4.2 Acquired data from logger (Pune city)

1st & 2nd column is Time (sec) & other 22 columns in dataset having different parameter data
in it. Data acquired for parameters are mentioned in section 3.2.2

4.3 Data Cleaning

Collected data can contain many data quality issues, such as: missing rows/entries,
unnecessary/irrelevant columns, presence of invalid characters or nulls, data drift, spike
detection, etc. All of this needs to be addressed before feeding the model for training

Page 14
4.3.1 Removal of irrelevant column:-

The dataset contains “Acc pedal position, clutch press state, engtrq, clnt temp, which is
basically not required during brake analysis. This is just extra data, so it's not a useful
feature to add to model training. Therefore, this column is omitted for subsequent steps
and analyses.

4.3.2 Correction of column data type:-

To make sure our input data is clean and proper, it's important to check that all
columns/characteristics are of the correct data type. Data with inappropriate data types can
indicate poor data quality and affect the performance of our trained model. Below are the
observed data types for each feature present:

Image 4.3.2 Correction of column data type

It turns out that no datatype casting is necessary as all the features/columns in the data are
already of the appropriate type. It's fine to use further.

4.3.3 Exploratory data analysis:-

Based on the acquired data, we need to derive some parameter like brake energy kill in
each event (mj).

Page 15
 Image 4.3.3 Exploratory data analysis

Cumulative brake energy kill in (mj)= 485.8359500385803

Image 4.3.3.1 Brake energy kill estimation

After Brake energy kill, brake deceleration band need to be check. Like

Page 16
 Image 4.3.3.2 Brake deceleration calculation

4.3.3.1 X-Y Plot (Correlation analysis)

1) Brake deceleration count in each band in data set

Image 4.3.3.3 Brake deceleration level counts

2) Temp count in each band in data set

Page 17
 Image 4.3.3.4 Brake temperature band and count

3) Brake deceleration count distribution plot-

Image 4.3.3.5 Brake deceleration count distribution plot

Following are the observations that can be made from Above Exploratory data analysis-

1) Front brake pad Inner temperature higher than Outer pad.

2) Inner Pad temperature consistently lies in 50-150° C.
3) Also Brake deceleration is higher in lower “g” levels, Lies in between (0 to 0.2 g)

Page 18
 Chapter-5 Data Pre-Processing

Below are some of the data preprocessing steps performed before the data is accepted into
the EDA-

a. Removal of irrelevant column

b. Derive additional parameter from available parameter.

Chapter-6 In house Brake block cycle generation

Based on acquired vehicle data & standard available for dyno brake test need to be
compare for each brake deceleration band.
No of brake application
Deceleration
Field Dyno Test
Range (g)
vehicle standard
0-0.1 16709 0
0.1-0.3 529 455
0.3-0.5 4 682
>0.5 2 45
Data acquired kms 3784 1100
Table 6.1 brake deceleration comparison of field and dyno test

Image 6.1 brake deceleration comparison of field and dyno test

From comparison of field vehicle & existing block cycles, It’s clearly visible that-
1) No of brake application in lower “g” levels are not considered in dyno test cycle &
higher “g” levels are much higher in dyno test compared to actual field vehicle.
2) Lower “g” level braking count mostly rises brake temp gradually and results into
higher brake wear
Based on above comparison, we are recommending to
1) Increase the no of brake application in lower “g” levels for more duration, which
lead to gradual increase in temperature.

Page 19
 2) This gradual increase in temperature affects the brake pad wear & it easier to
correlate with field brake pad life.
With proposed brake block cycle wear observed on dyno-level is nearly same as field vehicle
after testing one set of the brake pads-

No of brake application
Deceleration
Range (g) Dyno Test standard Dyno Test standard
Field vehicle
(old cycle) (Proposed cycle)
0-0.1 4281 0 4281
0.1-0.3 656 455 656
0.3-0.5 6 682 5
>0.5 1 45 1
Wear (mm) 1.05 0.75 1.04
Life (kms) 39556 28254 39179
Table 6.2 brake duty cycle comparison of field and dyno test & proposed brake duty cycle

Based on dyno-test result new brake block cycle correlation with field is well accepted &
implementation on rig level is in progress.

Chapter-7 Model Configuration

For the given dataset, two different models – Random forest & logistic regression are
planned to be built.
Sample dataset for model prediction based on acquired data for Pune city & dummy data-

Image 7.1 brake deceleration, energy, wear & life dataset

Total 200 field vehicle data is used for model prediction (dummy data).
1) Random Forest- Random Forest is a supervised machine learning algorithm
commonly used in regression and classification problems. Build decision trees for
different samples and for regression take majority votes for classification and
averaging

Page 20
One of the most important features of the random forest algorithm is its ability to
handle datasets containing continuous variables, as in regression, and categorical
variables, as in classification. gives better results in classification problem

Image 7.2 Random forest algorithm code

Page 21
 2) Logistic regression- Logistic regression is a statistical analysis technique used to
predict binary outcomes. Yes or no based on previous observations of the dataset.

Logistic regression models predict dependent data variables by analyzing the

relationships between one or more existing independent variables.

Image 7.3 Logistic regression algorithm code

7.1 Model Evaluation- Model scoring is the process of using various scoring metrics to
understand the performance of a machine learning model and its strengths and
weaknesses. Model evaluation is important for assessing model validity in the early stages
of research and plays a role in model oversight.
1) Random Forest-

Page 22
 Image 7.4 Random forest algorithm evaluation & accuracy

2) Logistic regression-

Image 7.5 Logistic regression algorithm evaluation & accuracy

7.2 Conclusion:-
The random forest algorithms have given decent accuracies on test data. The Random
forest model has given an accuracy of 90% on test data, and the logistic regression model
has given 48% accuracy on test data. From the results its evident that accuracy for brake
pad replacement with random forest algorithm is higher & this helps customer to plan

Page 23
service action as per his convenience based on early intimation and saves other
components like brake disc from worn out. The random forest model performs well with
optimize estimator (n_estimator) = 140 wherein in logistic regression max iteration kept is
100 still it is showing less accuracy compared to random forest algorithm.

Chapter-8 Summary

This section begins with a detailed description of the dataset, followed by a section on data
cleansing. We then detailed the various EDAs, or correlation analyses, performed on the
collected data. Next, we examined the various data preprocessing steps performed on the
data to transform it into a format suitable for model training. Based on EDA, brake pad
temperature lies highly in between 50-150° C and Brake deceleration lies in lower “g” levels.
Duty cycle of vehicle level data and block cycle data compared and the gap is identified in
the existing block cycle. New brake block cycle developed and tested on rig level. Based on
dyno-test result new brake block cycle correlation with field is well accepted &
implementation on rig level is in progress.
For brake pad life estimation, Random forest algorithm is used due to higher accuracy (90%)
& for Pune city life prediction is 39k kms which replicating the real life scenario to customer
than replacement at 15k kms at service schedule. This helps customer to plan service action
as per his convenience based on early intimation and saves other components like brake
disc from worn out

Chapter-9 Future work

The results from the work show promising results in Random forest algorithm and logistic
regression algorithm shows lesser accuracy, hence these could be explored deeper.
Therefore, based on accuracy result, Random forest model is considered in this thesis for
one location (Pune city) & as soon as more data (different cities) gathered accuracy would
again go above 90% & it predict accurate brake pad life to customer before replacing the
brake pad.

In addition to this, integration of brake pad replacement using ML-model (random forest)
with Connected Vehicle/Car technology (IRA technology) is in progress and software testing
for brake pad life intimation to customer is in progress. Data acquisition for various location
to improve accuracy of prediction of brake pad life is in progress.

This model predicts the brake pad life for Indian terrain conditions, which need to explore
for global application. Therefore, it will be acceptable globally.

Page 24
 Detailed plan of work

Expected
Sr
Task date of Names of Deliverables Status
No
completion
Submission of 26th–30th Literature Review and prepare Completed
Dissertation May 2022 Dissertation Outline
1 Outline/
Abstract
report
Design & 31st May – Acquire customer data from Completed
Development 12th Jul Company server, Verification of
2022 required parameter, Data
2 cleaning, Derive required
parameter, Generation of
Accelerated brake block cycle
for in house testing
Mid-Semester Before Mid-Semester evaluation and Completed
with 15th July completion the Mid-Semester
3
Organization 2022 Evaluation Sheet
Mentor
Submission of 15th - 19th Mid-Semester review Completed
4 Mid-Semester July 2022
report
Testing 20th–22nd Design & Development of Completed
August Prediction model & Prediction
5 2022 model accuracy check,
Software testing,
Implementation & Conclusion
Final Before End-Semester evaluation and Completed
dissertation 27th completion the End-Semester
6 review - with August Evaluation Sheet
Organization 2022
Mentor
Final 27th - 31st Final Review and submission of completed
7 Dissertation August Dissertation
Report 2022

Page 25
Literature References

1) Hoffmann A, Lehner S, Weissenbacher C, Zechlin H 2018 Audi A3 e-tron Boxberg

Driving Test Technical report Robert Bosch GmbH, CC-AS/EYB3, 74232 Abstatt,
Germany.

2) Jegadeeshwaran R, Sugumaran V 2015 Brake fault diagnosis using clonal

selection classification algorithm (csca) | a statistical learning approach
Engineering Science and Technology, an International Journal 181

3) Chetan C Harlapur, Priyatamkumar Kadiyala, and Ramakrishna S 2019 Brake

pad wear detection using machine learning International Journal of Advance
Research, Ideas and Innovations in Technology

4) Burnaev E 2019 Algorithmic foundations of predictive analytics in industrial

engineering design Journal of communications technology and electronics

5) Burnaev E 2019 On construction of early warning systems for predictive

maintenance in aerospace industry Journal of communications technology and
electronics

6) Burnaev E 2019 Rare failure prediction via event matching for aerospace
applications In Proceedings of the 3rd International Conference on Circuits,
System and Simulation (ICCSS-2019)

7) SAE std-J2707_202106 Wear test procedure on inertia dynamometer for brake

friction materials (Stabilized Oct 2012).

8) TML Testing standards of vehicle-level brake test & in-house test rig

Page 26