| With the continuous development trend of cutting technology to high speed and high precision,as an important part of CNC machine tools,tool wear condition has become a key factor affecting the dimensional accuracy of the workpiece,surface quality,processing safety and production efficiency.With the advent of the era of intelligent manufacturing,online tool wear monitoring technology has become one of the important research directions of advanced manufacturing technology.In this paper,the wear of the milling cutter during the milling process was taken as the research object.The signals from different sensors were analyzed and processed by multi-sensor information fusion and finally the support vector regression model was established to realize accurate prediction of tool wear.The main work of this research is as follows:In terms of signal acquisition,through the comparison of different types of signals,this paper selected the three-direction vibration signals and current signal of the spindle as the monitoring objects,built a milling test platform,and completed the signal and tool wear value acquisition of the milling cutter’s entire life cycle successfully.In terms of signal processing,after the time domain,frequency domain and time-frequency domain feature extraction from original signals,the method of correlation coefficient was used for feature selection,and the optimal post-processing method was determined through comparison.In terms of model construction and optimization,by studying the influence of kernel functions on the model fitting and generalization abilities,it was determined to use linear kernel to build a support vector regression model based on genetic algorithm parameter optimization;the tool wear prediction model was constructed using linear kernel,and its generalization ability was verified on an unknown experimental data set.The comparison with the neural network model proved the superiority of this model.To solve the problem of low prediction accuracy at certain sample points of the above model,a feature selection and modeling based on the three wear stages was proposed.The optimal average relative error on the test set is 4.99%,the average absolute error is 4.47μm,the maximum error is 11.16μm,and the mean square error is 28.38μm~2.Finally,experiments were designed to verify the feasibility of real-time compensation for machining using the tool wear prediction model. |
