Study On Identification Method Of Tool Wear Based On Support Vector Machine

Tool wear states of tools used in metal working not only directly affect the machining accuracy and workpiece surface quality, but also affect the parts of the processing cost and production efficiency and so on. How to achieve real-time monitoring of tool wear has become one of the key technologies in machining, so the tool wear states monitoring technology has important theoretical significance and practical value.Acoustic emission is very suitable as an effective monitoring signal to the tool states monitoring system, because it is associated with tool cutting states in high degree of correlation. As the different acoustic emission signals corresponded to different wear states characteristics, acoustic emission signal was analyzed and processed to extract the main features which effectively reflected the different states.In this paper, acoustic emission signals under different cutting conditions were collected and analyzed by using the acoustic emission signal monitoring system. Signal analysis results show that acoustic emission signal is difficult to extract features from only purely time domain or frequency domain processing. Using fourier analysis and singular spectrum transform combined to extract signals as the main features to reflect tool wear states. Meanwhile, the tool states were affected by cutting parameters to some extent, the three factors of cutting (cutting speed, cutting depth, feed rate) as additional features.The principal component analysis was used to process the feature vectors which constitute the main features and auxiliary features, then the principal components were seen as support vector machine input vector. The PCA was used to process the feature vectors, data processing results show that PCA not only realizes the feature vectors dimension reduction, but also eliminates the correlation between feature vectors. The principal components can be used as support vector machine input data. At the same time,the network model that used to implement the states monitoring of tool wear was established by using the self-learning, adaptive, fault tolerance and non-linear mapping capability of support vector machine. The results show that the training error, test error of support vector machine is less than BP neural network, and the learning times that requests the same error of support vector machine is significantly lower than BP network. The system is established under support vector machine has a good stability, quick recognition speed and can be more accurate for tool states monitoring.