Abstract:  Aiming at the problems of low accuracy of tool wear prediction and lack of comparison of single sensor coverage features in milling process, a tool wear judgment method combining particle swarm optimization (PSO)-least squares support vector machine (LS-SVM) algorithm and multi-sensor eigenvalues was proposed. Firstly, a tool wear recognition system using vibration, cutting force and sound emission as tool wear monitoring signals was constructed. Then, the time domain eigenvalues: maximum p1, root mean square p2, standard deviation p3, absolute mean value p4 combined with the wavelet frequency band energy eigenvalues were used to analyze the milling processing signals. The recognition model of tool wear state and the prediction model of tool wear value were established by PSO-LS-SVM algorithm. Finally, through the process of time domain and wavelet analysis, the 71-dimensional signal eigenvalues were extracted from the vibration, milling force and acoustic emission signals, and the dimensionality was reduced to 24. With 24 dimensional eigenvalues as input, tool wear state and tool wear value as output, the tool wear recognition and prediction algorithm were verified. The research results show that the tool wear state recognition model based on PSO-LS-SVM algorithm has a wear identification accuracy of 99.39% on multi-sensor features, which is higher than that of single sensor features. The prediction accuracy of the tool wear prediction model reaches 99.75%, which is 8.02% higher than that of other models.
Key words: tool wear monitoring; multi-sensor eigenvalues; eigenvalues extraction; particle swarm optimization (PSO); least squares support vector machine (LS-SVM); wear recognition and prediction