Research On Automobile Engine Condition Recognition Based On Audio Feature

Automobile engine is the heart of automobile,which determines whether the automobile can run normally.Vibration signals are usually detected in engine condition recognition,but acoustic signals are rarely used.Starting from the actual situation of manual engine maintenance,it is often used to judge whether the engine is abnormal by listening to whether the sound is normal.With the rapid development of acoustic technology and artificial intelligence technology in recent years,it has laid a certain foundation for using audio signals to analyze and identify engine conditions.Therefore,this paper will identify engine working conditions and abnormal sound by analyzing automobile audio signals.First of all,this paper proposes an audio signal acquisition scheme based on Android smartphones,designs and implements car audio signal acquisition software and car condition labeling software,and uses a dualthreshold endpoint detection method to detect endpoints after preprocessing the collected audio signals.Secondly,by comparing the car audio signal and the speech signal,a complete set of adaptive noise empirical mode decomposition algorithms is proposed to decompose the car audio signal,and the eigenmode components with stationary characteristics are obtained,and the Mel cepstral coefficients are extracted from the main component signals.,the Mel cepstral coefficients of all components are dimensionally reduced as improved audio features.The experimental results show that in the dataset composed of non-stationary audio signals,the recognition accuracy of the improved audio features is 2.68% higher than the Mel cepstral coefficient.Thirdly,the fourth-order Gamma Tone filter is used to simulate the human auditory filter,the filter bank is improved based on the instantaneous frequency of the eigenmode component,and the audio characteristics of the signal are calculated with the improved filter bank.The experimental results show that the recognition accuracy of the improved Gamma Tone cepstral coefficient is 4.19% higher than that of the Mel cepstral coefficient,and1.99% higher than that of the unimproved Gamma Tone cepstral coefficient.Finally,this paper uses an engine condition identification method that first identifies the vehicle operating conditions,and then further identifies the engine abnormality.The support vector machine model is used to identify three working conditions including engine idle speed.Under the engine idle speed condition,the improved audio features of the signal are extracted,and the threshold cycle unit and convolutional neural network are used to build recognition models to identify the most common abnormal noise of the engine.The experimental results show that the threshold recurrent unit model and the convolutional neural network model can recognize 94.23%and 96.15% of the abnormality of the automobile engine on the test set,respectively.