Fault diagnosis of automobile power seat with acoustic analysis and retrained SVM based on smartphone

Abstract

Automobile power seats (APS) fault diagnosis method research is crucial for manufacturers to adjust production strategies and improve economic efficiency. Although acoustic analysis has been used in production process to detect APS faults, current methods are inefficient and cannot deal with sharp noise interference. Thus, this paper presents a novel method for fault diagnosis of APS based on smartphone. Firstly, we propose an Adaptive Filter Banks Method (AFBM) to extract spectral features and reduce the subsequent computation, which solves the fixed spectrum resolution problem of short-time Fourier transform. Secondly, we use an outlier detection algorithm based on K-Nearest Neighbor and statistical theory to exclude noise-interfered samples. Finally, we propose a retrained support vector machine (SVM) model to identify the fault according to the clustering combination of different faults. To verify the effectiveness of the proposed method, we measured the actual acoustic signal of a typical APS 20DL80X9D-7. We use smartphone instead of professional microphone and processor to adapt to different application scenes. The accuracy and immediacy of fault identification results verify the superiority of our method.