Design Of Loudspeaker Abnormal Sound Extraction Module And Research On Classification Method For Loudspeaker Abnormal Sound

Loudspeakers are widely used.As the basic sound unit,loudspeakers exist in various products,instruments and equipments.Defects in design,production,or assembly can cause speaker faults.The faulty loudspeakers would produce abnormal sound,which will affect human's listening experience.Therefore,it is of great significance to detect the faulty loudspeaker in the production stage and prevent it from flowing into the market.In this paper,aiming at the problem of fault speaker detection,an abnormal sound extraction module was designed,and then machine learning methods are used for classification.The work of this paper is as follows:A loudspeaker abnormal sound extraction module was designed,which was composed of two modules,excitation-filter module and audio acquisition module.Among the sub modules of excitation-filter module,excitation signal module generates excitation signal,band-pass filter module manages the excitation signal and amplitude adjustment module adjusts the amplitude of the excitation signal,in conjunction with other instruments which are power amplifier and microphone adapter,notch filter module makes fundamental notching for the loudspeaker response signal,the notching frequency is controled by clock signal module.The signal after fundamental notching is inputted into the audio acquisition module,Among the sub modules of audio acquisition module,audio processing module does data acquisition for the signal after fundamental notching,extended storage module stores the acquired data,and then USB high-speed communication module uploads the acquired data to upper computer.The loudspeaker abnormal sound extraction module realize the function which is fundamental notching,the influence of fundamental frequency signal is weakened,which is beneficial to feature extraction and classification for the extracted signal.The experiment platform was set up using the abnormal sound extraction module.For loudspeakers whose models are JIEFUEAB8₁0,330 sets of qualified loudspeaker signals and 330 sets of loudspeaker signals with damaged diaphragm were collected using the abnormal sound extraction module,and then feature extraction and classification experiments were conducted.Two feature extraction methods were used,the first is wavelet packet and sample entropy,the second is ensemble empirical mode decomposition(EEMD)and sample entropy.Two classification methods were used to verify the validity of the feature extraction,the first is support vector machine(SVM),and the second is XGBoost.After the classification experiments using a single feature,features representing high-frequency signal informations among the two features were selected for feature fusion,and the fusion features were used for experiments.The experiement results have shown that SVM algorithm achieved 94.6970% classification accuracy by using fusion features,and XGBoost algorithm achieved 96.2121% classification accuracy by using fusion features.Compared with one feature,higher classification accuracy were achieved by using fusion features.This paper contains 66 figures,9 tables and 58 references.