Research On Milling Tool Wear Status Monitoring Based On Deep Learning

With the intelligent manufacturing technology and the rapid development of sensing technology,the cutting processing field has developed rapidly with regard to intelligence.In the cutting process,tool wear is inevitable.If the tool wear failure is not monitored in time,it will lead to abnormal cutting process and seriously affect the machining efficiency and quality.Intelligent monitoring of tool wear enables the machining system to sense the real-time status of the tool in advance and make early warnings and decisions,which is an effective way to ensure the efficient operation of the machining and manufacturing system.Therefore,identifying tool wear states accurately and predicting tool wear accurately are critical factors in achieving intelligent cutting and safe machining.This paper takes milling cutters as the research object,collects the tool status monitoring signals in milling processing,analyzes the data characteristics,extracts the data features associated with the tool wear status in the data,and the deep learning based tool wear state recognition and wear value prediction network models are constructed respectively for solving the problems in tool wear state monitoring and realizing the smart monitoring of tool wear state.The key research of this paper is as follows:(1)Pre-processing of the tool monitoring signal and construction of the data set.To address the problem of large and redundant tool monitoring signal data,this paper analyzes and processes the tool monitoring data set,including data slicing processing,conversion of 1D data to 2D image processing,etc.,so as to construct the tool wear status identification data set and tool wear value prediction data set respectively,and provide reliable data assurance for model validation.(2)Accurate recognition of tool wear status.In order to solve the problems of complexity of large amount of data and imbalance of data types in the process of machining data collection,which leads to complex construction of tool wear state recognition model,long training time and low recognition accuracy,a WGAN-GP-SE-Shuffle Net network model was constructed in this paper based on generative adversarial networks and lightweight convolutional neural network.The WGAN-GP network is used to enhance and balance the tool monitoring data and convert the one-dimensional signal data into two-dimensional grayscale images.The existing Shuffle Net network is improved by adding a channel attention mechanism to construct the whole model to identify the wear status of the tool and analyze it in comparison with the traditional identification methods.It is demonstrated that the proposed method can achieve accurate recognition of tool wear status.(3)Accurate prediction of tool wear value.In this paper,a tool wear prediction method based on domain adversarial adaptive and squeezed excitation channel attention multi-scale convolutional long and short-term memory network(SE-DAAMSCLSTM)is proposed to solve the situation that the prediction effect of tool wear value prediction model is restricted due to the distributional inconsistency of tool monitoring signal data under the same working condition and variable working condition.A feature extractor combining multiscale convolution and channel attention mechanism and introducing domain antagonism mechanism is constructed to extract domain-independent multiscale spatio-temporal features describing tool wear to accurately predict tool wear values.By validating the model on a milling dataset and comparing it with conventional prediction methods,the results show that the model is able to make accurate predictions in the presence of changes in tool monitoring signals,demonstrating the superiority of the method in predicting tool wear.