Detection Of The Blunt State Of Acoustic Emission Of Grinding Wheel Based On HDP-HSMM

In the grinding process,the different blunt state of the grinding wheel significantly affects the processing efficiency and quality.A seriously blunted grinding wheel would even lead to the occurrence of waste products.Therefore,attention has been aroused on how to monitor the blunt state of the grinding wheel in the grinding process.Traditionally,abrasive tools are monitored by experienced workers.In order to realize the automation of production,various methods have been proposed for intelligent monitoring.In this paper,an online monitoring method based on acoustic emission signal is proposed.From the mechanism of grinding acoustic emission,it is found that the grinding acoustic emission signal is mainly derived from the plastic deformation of the material during the grinding process,and the frequency band of the elastic wave is between 100KHz and 300KHZ.Therefore,the signal collected by the acoustic emission sensor is firstly de-noised by the wavelet soft threshold denoising method,following by the segmented analysis for dividing the denoised acoustic emission signal into multiple overlapping segments.In the second step,by setting a threshold voltage,the acoustic emission hits are intercepted for each frame of acoustic emission signal and 8 statistical features of each acoustic emission hit are extracted.The average value of 8 dimensional features of the acoustic emission hits in the frame is calculated to form the acoustic emission vector instead of the frame acoustic emission signal.In this way,the acoustic emission vectors of all frame acoustic emission signals are acquired to constitute the acoustic emission data set.Finally,the hierarchical Dirichlet processs-implicit semi Markov model is employed to build a nonlinear relationship between the acoustic emission data set and different grinding wheel blunt level.As a non parametric Bayesian inference method,hierarchical Dirichlet processs does not need to specify the number of state categories,and the number of the state of the grinding wheel is obtained by clustering the acoustic emission data set adaptively.The hidden semi Markov model(HSMM)is an improvement of the hidden Markov model(HMM),which avoids the exponential decay of the state duration of the HMM model,and is often used in the problems of fault diagnosis,mechanical state recognition and so on.Combining the characteristics that HSMM can mark the real time data in time series and HDP's adaptive clustering based on the data itself,HDP-HSMM is proposed.HDP-HSMM is an unsupervised learning method,that is to say,the method does not need to have a marked data sample,and the obtained acoustic emission data set is just difficult to mark it clearly.Good agreement is observed between the HDP-HSMM trained by the acoustic emission data set and our expectations,evidenced by the 93.7%accuracy of the trained model on the test data set.The results strongly prove that the method can effectively identify the different blunt state of grinding wheel accurately,which is of great value for industrial applications.In Chapter ?,the preface mainly introducese the significance and background of the grinding wheel blunt state detection.The application actuality of acoustic emission signals in the field of tool detection is presented,and the signal processing and model recognition methods used in tool detection are summarized.The main contents of the paper are expounded.In Chapter ?,the mechanism of acoustic emission and the reason for the blunt of the grinding wheel in grinding process are analyzed from the point of view of grinding mechanics.In Chapter ?,the identification method of blunt state of grinding wheel based on HDP-HSMM is presented.It includes wavelet threshold denoising,frame processing technology,statistical feature extraction of acoustic emission signal and HDP-HSMM theory knowledge.In Chapter ?,each component of the tool condition monitoring system is illustrated,and the research history of all aspects is returned.The grinding experiment and grinding wheel detection system are described in this chapter.Chapter ? gives the result of HDP-HSMM blunt state recognition of grinding wheel,analyzes the blunt classification of grinding wheel,and tests the trained model on the test data set.Finally,Chapter ? gives the the main conclusions of this study and the outlook for the future.