Research On The Fault-Detection Method Based On The Time Frequency Analysis For The Loudspeakers

In production process of loudspeakers, the first procedure of quality detection is "pure-tone" detection. Namely, to meet the requirements of users, its task is detecting whether the response signal, which is imposed under the sweep signal in its rated power, from loudspeakers qualified. At present, the online detection of "pure-tone" mainly relies on the manual detection. This subjective detection method not only relies on the workers' hearing and experience, but also has great relations with working environment and the workers' state. It isn't beneficial for automation of production process and quality assurance of product.According to the demand for the electro-acoustic industry and scientific and technological project in Tianjin, this thesis researches on the fault detection method for the loudspeakers based on the time frequency analysis. Through acquisiting and analysising the response signal by computer, the detection is performed.In order to meet the research needs of detection method, swept LAU is used to incentive the measured loudspeakers. At the same time,this thesis has collected and intercepted a cycle of loudspeakers' response signal as samples to analysis with computer. This thesis has used short-time Fourier transform and wavelet packet transform method to analysis time frequency of the signals respectively. Then extracted the feature of the loudspeakers and recognize by using Support Vector Machine and artificial neural network.The main results of research on topics including:(1) This thesis has used the method of short-time Fourier transform for the analysis of the loudspeakers' response signal, and has proposed the effective method for the loudspeakers feature extraction.(2) This thesis has used the method of wavelet packet transform for the analysis of the loudspeakers' response signal, and has proposed the method based on the characteristics of energy extraction.(3) This thesis has established the L-M BP neural network model, which put the characteristic energy as an input vector, identify the network structure. And has completed the study and training of loudspeakers samples. This thesis has implemented the loudspeakers' fault detection and recognition.(4) Aimed at the small samples of the loudspeakers, this thesis has established the loudspeakers fault detection and discrimination system base on the Support Vector Machine, and has determined the parameters of the Support Vector Machine, and has implemented the loudspeakers' fault detection and recognition.(5) Software is modular in design. The software integrates the entire system combine the functions of fault detection design of the loudspeakers. In this thesis, both theory and experiment in the fault detection of loudspeakers has been researched. The results showed that the fault detection method of the loudspeaker based on the time frequency analysis can recognize the loudspeakers effectively.