This content is not included in
your SAE MOBILUS subscription, or you are not logged in.
Application of Machine Learning Technique for Development of Indirect Tire Pressure Monitoring System
Abstract:
Tire inflation pressure has a significant impact over vehicle driving dynamics, fuel consumption as well as tire life. Therefore, continuous monitoring of tire pressure becomes imperative for ride comfort, safety and optimum vehicle handling performance. Two types of tire pressure monitoring systems (TPMS) used by vehicles are - direct and indirect TPMS. Direct systems deploy pressure sensors at each wheel and directly send pressure value to the vehicle Controller Area Network (CAN). Indirect sensors on the other hand use the information from already existing sensors and some physics-based equations to predict the value of tire pressure. Direct TPMS tend to be more accurate but have higher cost of installation while indirect TPMS comes with a minimum cost but compromised accuracy.
A digital proof-of-concept study for indirect TPMS development of a non-ESP vehicle based on machine learning (ML) technique is elaborated in this paper. The study aims to propose a methodology for development of an indirect TPMS having an accuracy equivalent to that of a direct TPMS. A full vehicle model designed in Amesim software is used to extract data to train the machine-learning algorithm for different test cases. Simulation model is validated against the test data of vehicle dynamics parameters to ensure the accuracy of data extracted for ML model training. Multilayered feed forward, back-propagation artificial neural network is trained using three prediction algorithms and sensitivity of different algorithms, network parameters is analyzed against selected driving scenarios.
Proof-of-concept study suggests that the proposed tire pressure prediction algorithm has a potential to predict tire pressure accurately at par with Direct TPMS. It lays a foundation for developing ML based indirect TPMS using physical testing data by providing assistance in test plan preparation, exploring data pre-processing techniques and algorithm selection. Furthermore, the generic methodology mentioned in this paper can be referred for initial development of any ML based project.