Research On Tool Wear State Prediction Based On Vibration Signal And Audio Signal

As a necessary tool in machining,the tool wear is inevitable in milling.How to find a reliable and convenient way to predict the tool wear state in industrial production and processing is of great significance to improve productivity,reduce production cost and maintain the quality of machining workpiece.This paper takes the cutting tool of CNC machine tool as the research object,and carries out the following research work:1.Set up an online data acquisition system for tool wear state.Based on vibration sensor and audio sensor,the on-line data acquisition system of tool wear state was built.Through the system,the vibration signal and audio signal of tool milling work were collected.2.Extract tool wear characteristic values.In time domain,the vibration signal and audio signal were analyzed by time-domain statistical analysis method,and the mean square value of audio signal and the variance of vibration signal were selected as the characteristic values of tool wear.In the time-frequency domain,the audio signals were analyzed by EMD decomposition technology,and the root mean square values of IMF1 and IMF2 components were selected as the tool wear characteristic values.Wavelet packet decomposition technology was used to decompose the vibration signal,and the energy of the subband after decomposition was counted.The energy percentage of the frequency band(0～1.25 k Hz)and the frequency band(7.5～8.25 k Hz)of the vibration signal were selected as the characteristic value of the tool wear.A total of 6eigenvalues were extracted from the two signals.3.The tool wear state prediction model is established.BP neural network was used to fuse six feature vectors of vibration signal and audio signal in the feature layer.The results show that the model can predict tool wear state online,but it has the disadvantages of low accuracy and slow convergence speed.4.Genetic algorithm was used to optimize the prediction model.The optimal initial parameters of BP neural network were determined by genetic algorithm,and the prediction model of GA-BP tool wear state was established.By comparing the prediction results of the two,it is found that the prediction model optimized by genetic algorithm has higher accuracy and faster convergence speed,which is more suitable for the prediction of tool wear state in industrial production.