New Hybrid Wavelet and CNN-Based Indirect Tire-Pressure Monitoring System for Autonomous Vehicles

Abstract

Since the tire pressure has a significant influence on driving safety, even self-driving vehicles need to be aware of their current tire pressures. Two major types of methods for estimating tire pressures exist: direct and indirect methods. In spite of recent advancements in direct Tire Pressure Monitoring Systems (TPMSs), indirect pressure monitoring systems still play a significant role due to their low costs. Indirect systems rely on the processing of signals from wheel speed sensors. In most cases, a transformation is applied to generate a frequency spectrum from which the tire pressure-dependent eigenfrequency can be extracted. The most accurate methods apply the Fourier transform, but these require the highest computational power. After the spectrum of signals from the wheel speed sensor is created, the eigenfrequency must be extracted. Several methods are available to extract significant frequency components. One of the easiest methods is peak searching, however, it is susceptible to noise. On the other hand, more accurate methods that are less sensitive to noise require more computational power. If a transform that consumes less computational power can be applied, then the freed resources can be used by a better eigenfrequency identification method. In this paper, a Hybrid Wavelet-Fourier Transform and Convolutional Neural Network-based method is presented, which exhibits a promising level of noise tolerance.