Research On Wear Status Recognition And Prediction Method Of Milling Cutter Based On Deep Learning

The current generation of information technology and the traditional manufacturing industry are gradually merging to form a new manufacturing model based on information physics systems.In the future intelligent manufacturing environment,the data collected by various sensors in the factory will be transmitted to the cloud in the form of continuous data flow,and the hidden state information of the equipment will be extracted by the server analysis to optimize and control the production system.Different from the traditional shallow machine learning method,the deep learning model has obvious advantages in terms of data volume processing scale,nonlinear ability and convergence.In this paper,for the problem of tool wear during milling,the deep learning method is used to establish the tool wear state identification and wear amount prediction model.The main research work is as follows:(1)Comparing the advantages and disadvantages of four common monitoring signals and the applicable range,and selecting the vibration signal as the monitoring signal.The wavelet denoising method is used to perform signal preprocessing,and feature extraction is carried out from three aspects: time domain,frequency domain and time-frequency domain.Comparing the four wavelet scale maps,it is found that the complex Morlet wavelet scale map works best.(2)Two models of tool wear state recognition based on deep learning are established.The first model is a convolutional neural network and transfer learning model.By comparing the modeling effects on different data sets,the superiority of transfer learning on small sample sets is proved.The second model is the stack sparse auto-encoder network model.The feature vector set is filtered for dimensionality reduction and then incorporated into the Softmax layer for classification.Finally,two traditional neural network models are compared,which proves the efficiency and precision of the deep learning model.(3)Establishing a prediction model of tool wear based on deep learning.The feature vector reduction and feature post-processing are performed on the filtered feature vector set.Then the least squares support vector machine with adaptive step cuckoo algorithm is used to predict the tool wear.The accuracy and effectiveness of the proposed model are verified by comprehensive comparison of features such as dimension reduction without deep learning,uncharacteristic post-processing,and two traditional neural network models.(4)Combined with the theoretical research of this paper,the tool wear monitoring system interface is preliminarily designed based on the MATLAB software GUI module,and two functions of state recognition and wear quantity prediction are included,which realizes simple operation and visualization.