Abnormal Noise Identification of Engines Based on Wavelet Transform and Bispectrum Analysis

Abstract

Abnormal noise is the most prominent problem for motorcycle, which affect consumers' purchasing desire and driving experience. The abnormal noise engines are always detected by manual auscultation(MA) after assembly. While MA is also affected by subjective and objective factors, its accuracy fluctuates greatly and has a large labor intensity. The most important is that it can not be applied in a corporation with mass production and high quality. To overcome those problems, this paper proposed an online engine abnormal noise detection method based on wavelet transform(WT) and bispectrum analysis(BA), which improves the accuracy and stability of engine abnormal noise identification and reduces the cost of the check. Firstly, the engine's acoustic signal is acquired by using a free-field microphone. Secondly, eliminated the background noise of signals by using the wavelet correlation coefficient(WCC) theory and extracted the signal features by applying power spectrum estimation and bispectrum analysis respectively. Thirdly, feature vectors are normalized before being used as support vector machine(SVM) samples. Fourthly, the appropriate kernel function and parameters are selected to train the vector machine using the training sample. Finally, testing samples are used to inspect the accuracy of vector machines. The result shows that the training accuracy and testing accuracy of the samples is high by using the method of WT-BA-SVM (wavelet transform - bispectrum analysis - support vector machines). WT-BA-SVM identified engine fault types effectively and provided the theoretical foundation for the establishment of an engine abnormal noise online detection system.