Research On Loudspeaker Fault Detection Based On Blind Source Separation

Fault detection based on audio signal is a new research point in the field of artificial intelligence,which has significant application value.In the practical application scenario,in the traditional TV manufacturing factory,the fault detection of the TV speaker generally uses the means of manual listening.This method has different standards,which is easy to cause the problem of misdiagnosis,leading to the problem products into the market,and it also needs a large amount of human resources,and the long-time noise working environment will also cause damage to workers' hearing.With the advent of industry 4.0,all kinds of factories are carrying out automation transformation to improve efficiency and accuracy.Therefore,it is of great significance to replace the traditional artificial fault listening method with artificial intelligence method for fault detection based on audio signal.In this paper,firstly,the research background and significance of signal fault detection are described,and the research status of blind source separation and fault detection at home and abroad is comprehensively analyzed.This paper mainly studies three types of blind source separation algorithms,including the blind source separation algorithm based on independent component analysis,the blind source separation algorithm based on joint diagonalization and the blind source separation algorithm based on deconvolution,and verifies the signal separation performance of different types of algorithms by establishing simulation signal simulation of factory environment.Secondly,the actual environment of the factory was investigated and data collection steps were designed.The appropriate equipment for collecting audio signals was selected to collect the four types of TV speaker signals in the factory environment,and analyzed the data characteristics of the acquisition signals.Complex signals in the actual plant environment were detected using the blind source separation algorithm described above,After comparing the advantages and disadvantages of different types of methods for blind separating speaker signals,For the independent component analysis algorithm,However,the deconvolution algorithm has incomplete separation signal,Considering the sparse characteristics of the audio signal and the reference sweep signal is known,the noise from background is unknown,Integrating sparse constraints and graph regularization techniques into a unified model framework,We propose a non-negative matrix decomposition algorithm based on dual-sparse graphs(Dual Sparse Graph Non-negative Matrix Factorization,DSGNMF)algorithm,blind source separation is also implemented through the matrix decomposition and reconstruction based on the DSGNMF algorithm.Separated waveforms and evaluation metrics indicate that our algorithm can achieve better results than existing blind algorithms.By performing the singular value feature extraction SVM classification experiment on the isolated signal,the importance of the blind source separation as the signal preprocessing is clarified.Our proposed DSGNMF algorithm combines the SVD's maximum and mean feature,indicating the superiority of the fault detection method based on the DSGNMF algorithm.Finally,by investigating the needs and processes of the factory production line environment,analyzing the requirements in practical application,we designed a TV speaker fault detection software based on five modules of recording,signal cutting,blind source separation,feature extraction and fault detection and the audio fault detection system combined with the hardware was completed.After passing the logic,function and other tests,the system is loaded on the production line and uses fault TV to test the audio fault detection system.The test results verified the effectiveness of the test system and the automatic TV speaker fault detection scheme is completed.