Artificial Neural Network-Based Method for Identifying Under-Inflated Tire in Indirect TPMS

Abstract

Tire pressure monitoring is essential to driving safety. Indirect tire pressure monitoring system (TPMS) is a cost-effective alternative to direct tire pressure monitoring system. Its performance depends on the algorithm for data pre-processing and analysis which is normally complicated, sensitive to initial calibration with limited working range. In this work, four tests were carried out with Baojun 530 with a different deflated tire in each case. Speed data read through ABS CAN bus was analyzed and traditional frequency based method was employed to identify the deflated tire. To simplify the data pre-processing and improve response speed and working range, a new artificial neural network (ANN) based method was also proposed to identify deflated tire based on speed data point collected through antilock brake system (ABS) sensors in tests. A long short-term memory (LSTM) network was developed to locate the deflated tire with an accuracy of 0.83 after training for individual data points. And performance of this method can be further improved by employing a soft voting mechanism with 3 LSTM networks. In proposed ANN based method, benchmark data from properly inflated tire is not required, which makes it a promising solution for multiple deflated tires cases which is challenging for traditional frequency-based method.