Tool Wear State Recognition And Prediction Based On Acoustic Emission

Nowadays,from the international perspective,the manufacturing position has become increasingly prominent.The technological transformation,represented by the intelligent manufacturing,is rebuilding the global manufacturing industry and forming a new industrial Internet world and becoming an international competitive strategic high ground.All world's manufacturing industries take great effort to develop in the direction of intelligence,which make the CNC machining technology become more intelligent.Tool wear state monitoring constitutes an important part of intelligent manufacturing technology.As the direct executor of the cutting process,it will inevitably wear in the cutting process.Thus,accessing to the exact tool wear state realtime has important significance for improving the part accuracy and surface quality,realizing the personalized manufacturing,improving the level of intelligent machine tools,and the system error compensation technology.As the above problem,the main works are summarized as follows:1.The characteristics of acoustic emission are studied.An indirect online monitor system based on acoustic emission is built.The tool wear feature,mechanism and influencing factor of milling tool are briefly discussed.The tool wear standard is determined and the tool wear state is classified into initial wear,normal wear and sharp wear.Based on the orthogonal experiment,the variation law of the acoustic emission signal with the cutting parameter and the tool wear is studied.2.The characteristic values of tool wear are extracted by different signal processing methods.Firstly,based on time domain analysis,mean,root mean square(RMS),variance and root amplitude are extracted.Then,based on wavelet packet transform analysis method,the wavelet entropy features of signals are extracted.Empirical mode decomposition(EMD)is carried on the signals to extract RMS of each intrinsic mode function(IMF)as feature values.The time feature values and timefrequency feature values are optimized to reduce data redundancy and to remove the irrelevance.3.The optimized feature values are input into the LS-SVM algorithm for learning,and the tool wear state recognition model is established.Considering the penalty factors and the kernel parameters of LS-SVM have a great impact on the model's accuracy,then the particle swarm optimization algorithm is applied to LS-SVM with the advantage of its globalization and convergence.The tool wear state recognition model and wear prediction model based on PSO-LS-SVM are established,and the model based on LS-SVM algorithm and BP neural network are also established.The prediction effect of the three models was tested with the same test samples.The result shows that PSO-LS-SVM model is better than those two models.Finally,the residual stress and roughness of the machined surface are measured off-line,and the results are used to indirectly evaluate tool wear and verify the prediction of tool wear.