The Research Of Tool Condition Monitoring Based On Deep Learning

Driven by the innovation of Made in China 2025,the modern manufacturing industry is experiencing a historical leap from automation to intelligence.The deep integration of new generation information technology and manufacturing industry has become the key to realize the transformation of intelligent manufacturing in China.CNC machine center as the "industrial master machine",is the core production base of high-end equipment manufacturing.As the core component of the machine tool,accurate condition monitoring of the tool could help ensure the machining efficiency and product quality of the machine tool.Therefore,how to design an accurate tool condition monitoring method has become an important research topic for intelligent manufacturing.As an important technology,deep learning can effectively mine the information of tool wear features from the data to achieve accurate identification of tool status,and thus has received widespread attention at domestic and overseas.However,the existing deep learning-based tool condition monitoring methods still have the following problems:without considering the variability of multi-dimensional signals on the expression of wear status,and the quality of feature fusion of each dimensional data is weak,which leads to the low recognition accuracy of tool wear status;in the actual industrial production site,the number of tool failure samples is relatively limited compared with normal samples,and it is difficult to obtain a sample set with balanced data volume,which seriously restricts the generalization ability of the model.To address the above problems,the specific research of this paper is as follows.(1)For the problem of insufficient multidimensional data fusion,this paper proposes a tool condition monitoring method based on the Improved Multiscale-Channel Attention Network(IMS-CA Net).Firstly,based on the idea of data level fusion,the vibration signals in X,Y and Z directions are concatenated as three channels of the model input feature map respectively.After that,the different levels of features of the vibration signals are extracted through the multibranch structure of the Improved Multiscale network.On this basis,the Channel Attention mechanism is introduced in the model,and using channel feature learning,the vibration signals in the three directions are adaptively mapped into the feature map with different degrees of influence on the tool wear classification task,so as to effectively utilize the complementary information between the data in each dimension,enhance the expression of data features,and improve the recognition accuracy of tool status.This paper collects real production machining data to validate the proposed method and conducts comparative analysis with other algorithms.The experimental results show the effectiveness of the proposed method.(2)To address the data imbalance problem in industrial production sites,this paper proposes a tool condition monitoring method based on Deep Convolutional Generative Adversarial Network(DCGAN).First,DCGAN is used to learn the intrinsic distribution of a few classes of sample data in the vibration signal sample set.After that,pseudo-samples similar to but not identical to the real sample data are generated,which are used to expand the abnormal wear samples in the training set.Finally,the tool condition monitoring model is used to modeling the training set with balanced amount of data.The experimental results show that the proposed method can effectively enhance the generalization ability of the model,improve the accuracy of tool condition monitoring,and avoid the situation that the model classification results are heavily biased towards normal wear samples,and avoid the situation that the model classification results are heavily biased towards normal wear samples.(3)Based on the above proposed method,this paper designs a corresponding tool status monitoring system,which includes operator registration and login function,real-time data collection and display function,and deep learning-based tool status monitoring function,so as to realize real-time analysis of tool monitoring signal data and real-time monitoring of tool status during cutting process,and thus assist workshop operators in tool maintenance decision making.