Research On The Recogniton And Prediction Of Tool Wear State

Metal cutting processing is the most common way used in mechanical design and manufacturing. And that, the tool is a crucial factor of production in the manufacturing process of metal cutting. Therefore, the performance, quality and management of tool can directly affect the stability of machining process, the product reliability, processing time and production efficiency. So, the tool wear condition should be monitored, the useful information should be extracted, and the tool wear status should be analysed, which can be implemented to reduce the cost of production, reduce production failures, improve production efficiency. Based on the tool wear condition monitoring, the value of tool wear in future time can be predicted on the known value of tool wear. Measures could be taken in time before the tool blunt. Therefore, the prediction of tool wear is very important.Vibration signals and cutting force signals of every cutting process of a cutting tool which varied from new to broken were acquired. Also the tool wear of the cutting tool after the completion of each cutting were recorded. The analysis of cutting force signals in time and frequency domain gave the features related to tool wear. The features that ware sensitive to the variation of wear loss of the cutting tool were selected. The analysis of vibration signals showed that the characteristics extracted from time domain or frequency domain could not distinguish different conditions of the cutting tool well. Hence the analysis of vibration signals using wavelet pack is necessary. The energy of frequency band decomposed by wavelet pack was extracted as features, and the sensitived features were selected.All the features selected after normalization regarded as the input vector of neural network. A three layer BP neural network were established. Used training samples to train the network, and then used the test samples to test. The verification of the trained BP neural network with testing samples showed that the wear loss of the cutting tool could be recognised properly.The ARMA model, BP Neural Network model and combination of them are applied to predict the tool wear. The ARMA model and BP Neural Network model with the appropriate parameters can acceptably predict the tool wear of some moment in the future separately. The combined model combined the advantages of two models.Although the effect of every prediction value with the combination of the two models is not better than each method, it was better on the whole.