Application of Machine Learning Models to Enable Virtual Development of High Performance Brake Systems

The once rarified field of Artificial Intelligence, and its subset field of Machine Learning have very much permeated most major areas of engineering as well as everyday life. It is already likely that few if any days go by for the average person without some form of interaction with Artificial Intelligence. Inexpensive, fast computers, vast collections of data, and powerful, versatile software tools have transitioned AI and ML models from the exotic to the mainstream for solving a wide variety of engineering problems. In the field of braking, one particularly challenging problem is how to represent tribological behavior of the brake, such as friction and wear, and a closely related behavior, fluid consumption (or piston travel in the case of mechatronic brakes), in a model. This problem has been put in the forefront by the sharply crescendo-ing push for fast vehicle development times, doing high quality system integration work early on, and the starring role of analysis-based tools in enabling this strategy. Focusing even further, brake corner systems under duress – such as high temperatures, and high braking power, can exhibit highly non-linear and in-stop varying behavior that can be exceedingly difficult to model accurately. The present work chronicles efforts by the author and colleagues to develop machine learning models that capture this complex behavior and generalize sufficiently well to continue representing the performance of the brake under high energy driving conditions, even as the models are presented with new braking conditions that were not part of the training of the models. The utility of the models in the prediction of system-level performance is demonstrated through a case study application to calibrating a fade warning feature. The present work is shown from the perspective of a practicing engineer, not a data scientist, with some details that may prove mundane to the latter – but a strong motivation behind this work is to share the experience of getting started and some practical lessons learned towards the use of these powerful machine learning tools to solving practical problems in the field of brake engineering.