Research On On-line Detection Method Of Surface Quality Based On Vibration Analysis And Support Vector Machine

The on-line detection of the workpiece surface quality can actively detect thequality condition of workpiece processing, it combine statistic analysis performedaccording to the result of detection with the grading index of the product quality,thenit obtain the assessment data of the individual or batch products so as to supply adecision basis for production, quality and production department. It has importantpractical significance to ensure workpiece processing quality and improve machiningefficiency. Based on the vibration analysis and support vector machine (SVM)relatedresearch of surface quality on-line detection in the mechanical processing isresearched. Firstly, using kernel independent component analysis, the paper carry outthe bind source separation of the collected vibration signals, and then the vibrationseparated signals are analyzed by the wavelet packet analysis and the characteristicparameters of the vibration signals are extracted, finally the surface quality isclassified and distinguished by means of support vector machine (SVM). This paper isstudied mainly from the following three aspects of the research content.Firstly, the acquisition of state information, The vibration signals contain muchinformation of operation state within the mechanical equipment, in this paper, basedon the vibration signals it study the inner connection between the vibration signalsand the machined surface quality and the condition recognition of the machinedsurface quality. On the basis of the analysis to cutting vibration theory, vibrationsignal acquisition and analysis processing theory, surface roughness and other relatedtheories, this paper research to identify the relevant experimental conditions and setup experimental platform so as to complete the measurement and analysis of surfaceroughness, vibration signal acquisition.Secondly, the feature extraction of state information. After the vibration signalsare determined to be state information, we hope that it can find out characteristicquantity, which is sensitive to the surface quality status, by vibration signals analysisand processing, through these characteristic quantity can effectively express oridentify different surface quality status it realize effective detection of the surfacequality. It studies related theory and carries on the simulation experiment to kernelindependent component analysis and wavelet packet analysis shows that the kernelindependent component analysis is a kind of nonlinear independent componentanalysis algorithm, for nonlinear and unstable vibration signals its blind sourceseparation is more accurate, more flexible. the vibration component is not related andrespectively independent separated by kernel independent component analysis andcan be more effective for expressing mechanical operation state. The wavelet packet has more accurate partial analysis ability to non-stationary signal, the energy featurein each frequency band from the decomposition of the vibration signal can reflectessential characteristic of the different mechanical operation state. In view of thevibration signal is usually a nonlinear and non-stationary signal series in themechanical processing, based on kernel independent component analysis and waveletpacket energy this paper realize the feature extraction of vibration signal.Thirdly, this paper realize classification identification of the surface roughnessby means of the support vector machine (SVM), compared with the traditionalstatistical pattern recognition method, the support vector machine exhibit manyunique advantages in solving small sample, nonlinear and high dimensional patternrecognition. On the basis of research and analysis to theory and algorithm related tosupport vector machine, the energy features of vibration signals is extracted by thekernel independent component analysis combined with wavelet packet energy assample input of the support vector machine, the intelligent pattern recognition ofsurface quality is realized and achieved good results.Theoretical studies and experimental results show that it can effectively detectthe mechanical processing surface quality on the basis of kernel independentcomponent analysis, wavelet packet analysis and support vector machine, it lays theexperimental and theoretical foundation for the realization of the possibility researchthat is on-line detection of the surface processing quality.