Study On Tool Wear Monitoring And Prediction Technology Based On Multi-Parameter Information Fusion

The research of this thesis comes from National Key Basic Research Program of China (973):"Research on large power equipment manufacturing"(2007CB707703-4). Through the review and analysis of the present research situation of tool condition monitoring, a series of studies have been conducted aiming at existing problems. First, the experimental program was scientifically designed. Cutting force, vibration, acoustic emission and cutting temperature signals were collected in real-time, under the different CNC cutting conditions. The research was carry out to tool wear state signal feature extraction and pattern recognition, using the method of the Joint Approximate Diagonalization of Eigenmatrices based Ensemble Empirical Mode Decomposition (J-EEMD), Artificial Neural Network(ANN) and Support Vector Machine(SVM) decision fusion technology. In particular, the tool wear states have been scientifically predicted, applying gray-hidden Markov model.This paper carried out the following research work:(1) The experimental platform was set up using a dynamometer, ceramic accelerometer, infrared cameras, acoustic emission sensors and digital acquisition systems, in order to study cutting tool wear state monitoring that based on the integration of multi-parameter information. The monitoring system can timely monitor the signals of cutting force, vibration, cutting heat and acoustic emission in CNC turning process. After monitoring the entire life-cycle of tool wear states, the scientific basis is provided for signal feature extraction, pattern recognition, and tool state prediction.(2) Observed signals were processed using the method of J-EEMD. This method is based on the characteristics of the signal itself decomposed into several Intrinsic Mode Functions (IMF), and then transform the energy ratio between the IMF, the original vibration signals and acoustic emission adaptive tool state characteristics under different wearing can be extracted. These experiments show that the method can identify the different states of tool wear based on the measured data. Tool wear state can be recognized by BP network and Elman network training.(3) Decision fusion method based on support vector machine was proposed for the limitations of commonly used Bayesian algorithms and D-S evidence theory. The decision fusion can be achieved based on the recognition results of BP and Elman network. Experimental results show that decision fusion method based on support vector machine has a good recognition rate and robustness. At the same time, this approach saves time than single neural network.(4) Gray-hidden Markov model is established based on the tool wear characteristics. A tool wear state of the next time is predicted through the prediction of the eigenvalues of the follow-up state of the tool. Experimental results show that the present method is feasible and effective, can accurately predict the next moment state of the tool. Real-time detection prediction system based on this model can reduce the downtime of CNC machine, and achieve the maximum economic benefits.