Convolutional Neural Network-Based Tire Pressure Monitoring System

Abstract

Tire pressure has a significant influence on the driving safety of road vehicles; therefore, it is mandatory in many countries to equip all new road vehicles with a tire pressure monitoring system (TPMS). There are two types of TPMSs in use: the direct TPMS (dTPMS) and the indirect TPMS (iTPMS), both of which have made significant improvement in the last decade. The most accurate iTPMS methods used in commercial vehicles apply the Fourier transform on wheel speed sensor (WSS) signals and extract the pressure-dependent eigenfrequency by utilizing center of gravity (CoG) or peak search (PS) methods, the research focus is shifting towards model-based and artificial intelligence-based methods. In this paper we propose a novel advanced iTPMS method based on modern signal processing and a convolutional neural network (CNN) for eigenfrequency detection. The proposed iTPMS method uses the hybrid wavelet-Fourier transform in combination with a CNN trained for pattern recognition-based eigenfrequency detection, and according to experimental results, it outperforms the commercially most frequently used Fourier transform and CoG method combination both in terms of computational requirement and accuracy.

Keywords

• Convolutional neural networks
• tire pressure monitoring system (TPMS)
• eigenfrequency
• fourier transform
• cosine transform
• hybrid wavelet-Fourier transform