Novel method for identifying and diagnosing electric vehicle shock absorber squeak noise based on a DNN

Abstract

Abstract Suspension shock absorber squeak noise is becoming increasingly apparent within the overall noise level of all-electric vehicles (EVs) due to the extensive reduction in power system noise. Although the early identification of shock absorber squeak noise via bench tests can save costs and time, such identification remains a challenge for the industry. In this paper, a novel method for identifying and predicting EV shock absorber squeak noise is proposed. In contrast to other studies on shock absorber noise that focus on highly complex designs and feature extraction, this study uses the original time signals and frequency spectra to predict the shock absorber squeak noise based on deep neural networks (DNNs). To implement this method, an EV road test is conducted on five different pavements, and the grade evaluation method (GEM) is applied in a subjective evaluation of the annoyance of the shock absorber squeak noise. The vibration signals of the shock absorber piston rod are collected and preprocessed via a bench test. Then, a DNN is developed to automatically extract the shock absorber squeak noise feature and intelligently identify the subjective annoyance (SA) grade of the squeak noise. This novel identification method effectively solves the problem in which the annoyance level cannot be evaluated via the GEM by relying on the rich auditory experience of the evaluation subject. In the validation analysis, the DNN outperforms two other intelligent methods, the genetic algorithm-back propagation neural network (GA-BPNN) and the genetic algorithm-support vector machine (GA-SVM), based on a confusion matrix and an error analysis.