Research On Fault Diagnosis Of Speaker Abnormal Sound Based On LMD And LSSVM

Loudspeaker is a elementary device as necessity for human-computer interaction in the application of new technologies such as 5G,AI,robot,speech recognition and VR,so it is very necessary to make automatic diagnosis of its abnormal sound fault in the production process.The traditional detection method is to judge the loudspeaker's abnormal sound by the method of human ear auditory perception with a pre-given excitation signal to the loudspeaker.The accuracy and consistency of the artificial diagnosis method are not high due to the reason that the different sensitivity of each human individual to sound plus the long working hours.The current essay takes the normal speakers as well as loudspeaker with three failure types of tympanic membrane rupture,flat iron powder impurities and small sound as the research object.Through the structuring of the experimental platform,4 types of the loudspeaker excitation signal is given,and the characteristics extraction is done to the corresponding response signal of the loudspeaker,an classification is done by the utilization of the pattern recognition method.On the basis of Local Mean Decomposition(LMD)algorithm,a method combining LMD and energy entropy is proposed to extract characteristic values of loudspeaker response signals,which lays a foundation for the next step of fault diagnosis classification.The speaker response signals is decomposed by LMD Decomposition method to obtain a series of Product Function(PF)components,and calculate their energy entropy to constitute the characteristic values of speaker response signals.The experimental results show that the LMD and energy entropy methods can effectively extract the characteristic values of the speaker response signals.On the basis of the Least Squares Support Vector Machine(LSSVM)classification algorithm,the radial basis function is selected to establish the LSSVM model for the pattern recognition of speaker mispronunciation fault diagnosis.The characteristic values of loudspeaker energy entropy extracted from LMD decomposition were input into the LSSVM algorithm for training model and test recognition and classification.Under different nuclear parameters and penalty factors,the results of LSSVM speaker difference recognition were 85.00%,88.75%,91.25% and 92.50%.On the basis of particle swarm optimization(PSO)algorithm,the parameters of LSSVM are optimized to improve the accuracy of pattern recognition.Aiming at the problem that PSO algorithm is easy to fail to realize the global Optimization,so the Simulated Annealing algorithm(Simulated Annealing,SA)is introduced to make up the lack of PSO algorithm,establish SAPSO-LSSVM loudspeaker sound recognition model,through the SA-PSO algorithm to optimize the nuclear parameters of LSSVM and penalty factor.The experimental results show that the differential recognition rate of LSSVM loudspeaker is 96.25%.In this paper,LMD decomposition method is used to analyze and process the response signal of the loudspeaker and extract the characteristic value of the loudspeaker response signal,and then use the LSSVM classification algorithm optimized by SA-PSO to carry out fault recognition and classification,so as to provide a reference for the study of automatic differential recognition of the loudspeaker.