Research On Prediction Technology Of Milling Cutter Wear Status Of CNC Machine Tools

As our country's manufacturing industry moves towards more and more intelligent,CNC machine tools have become indispensable mechanical equipment in automated production.According to statistics,the downtime of CNC machine tools due to tool wear accounts for a large proportion of all downtime.Therefore,in recent years,tool wear status detection has gradually become a research hotspot in the field of mechanical equipment fault diagnosis.Most tool wear state prediction models rely on traditional feature extraction methods and traditional machine learning algorithms.Establishing a prediction model of tool wear status not only requires a large amount of signal processing theory,but also the generalization ability of prediction models based on traditional machine learning is usually poor.In addition,there are few methods for predicting tool wear under multiple working conditions in existing research results.The prediction method under single working condition is difficult to apply to the actual processing environment.Against the above background,this article takes milling cutters as the research object and designs a milling cutter wear status prediction scheme that uses vibration signals as monitoring signals.This paper combines the deep residual shrinkage network and the long-short-term memory network to build the wear prediction model;further integrates the transfer learning theory to improve and train the established model.This model can realize the prediction of tool wear under multiple working conditions.The milling cutter wear status prediction scheme designed in this paper is of great significance for improving the processing efficiency of CNC machine tools.The specific research content of the paper is as follows:(1)This paper first preprocesses the monitoring signals and extracts multi-dimensional feature vectors from the signals through feature extraction methods,and then uses the recursive feature elimination method based on support vector machine to select the feature vector that can best characterize the wear state of the milling cutter.This paper chooses PSO-optimized least squares support vector machine(PSO-LS-SVM)and long short-term memory network(LSTM)as the milling cutter wear prediction model,and applies the preprocessed signals and feature vector training model respectively.Experiments show that the LSTM model is more suitable for the prediction of milling cutter wear,but the feature extraction ability is insufficient.(2)Aiming at the problem of insufficient model feature extraction ability,this paper chooses convolutional neural network(CNN)as the feature extraction module and builds the CNN-LSTM network model,converts the preprocessed signals into grayscale images and use them as the network model input to train the model.Compared with LSTM,the accuracy of prediction results is greatly improved.Aiming at the problem that different wavelet denoising parameters have a greater impact on model prediction accuracy,this paper uses a deep residual shrinkage network(DRSN)with adaptive denoising performance to replace CNN to build the DRSN-LSTM model.The grayscale images of the signals without noise reduction are used as the network model input to train the model,and the final model has a better prediction effect.(3)Aiming at the prediction of wear under multi-conditions,this paper proposes a model improvement method with additional multi-condition processing parameters,and combines the transfer learning theory to train DRSN-LSTM model.Finally,the improved model is trained and tested by monitoring signals under multiple operating conditions,and a better prediction effect can be achieved.The experiment verifies the feasibility of the milling cutter wear status prediction scheme proposed in this article.