Intelligent Recognition Of End-milling Tool Wear State Based On Multi-Source Information Fusion

The tool is the main execution part of the milling process,and its wear state during the machining process is not only directly related to the surface quality of the machined workpiece,but also affects the machining cost and safety.Therefore,it is of great value in practical engineering applications to realize intelligent monitoring and identification of the tool wear state during the machining process.In this paper,an intelligent monitoring method of tool wear state based on multi-source information fusion is proposed.This method monitors the spindle radial vibration acceleration signal,the sound pressure signal of the machining cabin and the spindle motor current signal of the CNC milling machine in the actual production process,and uses the deep learning algorithm to realize the self-adaptive feature extraction and wear state identification of the multi-source fusion signal.The main research contents are as follows:(1)The research of tool wear state monitoring at home and abroad is fully analyzed and discussed.The current tool wear state monitoring methods and the main methods of data feature extraction are discussed.The commonly used deep learning monitoring models are briefly introduced and analyzed.The wear mechanism of the end-milling tool is studied.Through theoretical analysis,it is demonstrated that the rationality and feasibility of multi-source information fusion as a monitoring signal of tool wear state.Finally,the experimental scheme for data acquisition is designed.(2)Experiments are performed and the data sets obtained in the experiments are divided into samples.Combined with the requirements of the surface quality of the workpiece,the wear state is divided according to the wear amount of the flank face,and the samples are split.Since the actual processing method is trochoidal milling,the original data contains the signal of the empty tool part.In order to reduce the data preprocessing workload of cutting out the empty tool signal and the impact on the integrity of the original signal,the data are made into180*320*3 three-dimensional picture samples,so that each sample contains processing data.(3)Variational auto-encoder and extreme learning machine are used to monitor the wear state of tools.The superimposed fusion of the three original time-domain signals are used as the input of the variational autoencoder,and the characteristic information reflecting the tool wear state is extracted.The hidden feature information of the original signal is extracted through unsupervised pre-training.Then the data features extracted by the Variational auto-encoder are input into the extreme learning machine for supervised classification and identification,so as to realize the identification of different tool wear states.(4)In order to solve the problem that it may be difficult to obtain complete tool life data in practical engineering applications,a residual learning network is introduced on the basis of the original model.The encoding part of the Variational auto-encoder is improved through the residual network,so that the model has the ability of transfer learning.The pre-trained model is fine-tuned by inputting a small amount of data from the new data set,so that it still has good identify performance for the new data set.(5)Based on the Lab VIEW platform,a tool wear state monitoring system is built.In the system,various practical functions such as data acquisition,signal display,data storage,analysis and processing and status alarm can be realized.By selecting different working condition parameters,the trained deep learning model is used to monitor the tool wear state,and the real-time monitoring results of the tool wear state can be directly obtained.When the tool wear reaches the warning value,it will alarm to remind the operator to replace the tool.