top of page


Mr. John Smith

Job title



Wear of brake friction materials and coefficient of friction depends on many factors such as Pad Temperature, Applied Load, Disc Size, Weight of the car, Velocity, Braking Characteristics and Proportion of Friction Materials. It has therefore been very difficult to predict wear and coefficient of friction with a certain degree of accuracy.

We have developed an application which predicts the Brake Wear, Disc Wear and Coefficient of Friction virtually and dynamically, for any Car, any Disc Size, any Pad Size, any Pad Raw materials, any route and any kind of driver behavior, very quickly without having to go through elaborate Dynamo-meter Tests and Test Track Tests and without relying on the brake sensors.

This application is very helpful for Braking System and Friction materials suppliers as they can virtually test a new design of a brake disc and pad on an existing car or a new car, for a particular kind of driving characteristics over a particular route to predict the wear.

They can also virtually use existing discs and pad materials on a new design of a car to find out whether the wear is acceptable, or a new disc and pads have to be designed. Currently, a new disc design is either tested on a dynamo-meter or on a test track driven for thousands of kilometres to determine the wear of the brake pads. Once the wear is measured, either the design is frozen or goes through the next iteration. The entire process of designing a new pad and disc might take as long as three to four months. The design time can be significantly shortened by using our application.

This application was developed using Machine Learning and Artificial Intelligence. Dynamo-meter Tests were carried out for many combinations of Raw Materials (Ingredients of Pad), Initial Vehicle Speed, Final Vehicle Speed, De-acceleration, Initial Brake Temperature, Disc Sizes, Pad Dimensions, Car Inertia and Roll Radius.

The resulting Brake Wear, Disc Wear and Co-efficient of Friction are noted.

This data was used to train ML algorithms with multiple layers. The trained ML algorithm was then used to predict any Brake Wear, Disc Wear and Co-efficient of Friction based on new input data like Raw Materials %, Initial Vehicle Speed, Final Vehicle Speed etc. The accuracy of the prediction of wear was found to be around 85%, when compared to the actual measured brake pad wear values.

Additionally, we have developed our own hardware which goes into the On-Board Diagnostic (OBD) ports of 100 vehicles and the Travel Conditions, Braking Characteristics (severity and number of braking instances) are continuously tracked.

Doing a Statistical Analysis of the data helps us to map a specific route and braking characteristics in terms of the Travel Conditions, Braking Characteristics - severity and instances, for a particular type of driver behavior.

Mapping this data on the Brake Dynamo-meter Simulation helps us to predict Brake Pad Wear, Brake Disc Wear and Coefficient of Friction for vehicles driven over different routes with different vehicle weights and varying ambient temperatures without using any brake sensors.

This application is being used by car or fleet owners to know how much wear has taken place on the brake pads for an entire trip or life of the brakes which is then used to publish when the pad needs to be changed as part of predictive maintenance calendar.

Kaushik Choudhuri, Domain Specialist, iGloble; Dr. Amit Shekhar, iGloble

Predicting Brake Pad Wear Using Machine Learning

EB2020-STP-015 • Full Paper • EuroBrake 2020 • Fundamentals of Braking Technology


Sign up or login to the ICC to download this item and access the entire FISITA library.

Upgrade your ICC subscription to access all Library items.

Congratulations! Your ICC subscription gives you complete access to the FISITA Library.


Retrieving info...

Available for purchase on the FISITA Store


bottom of page