Research On Tool Wear State Recognition Method Based On Deep Learning

The cutting tool of computer numerical control machine tool is the core parts of computer numerical control machine tool.In the process of computer numerical control machine tool cutting,the workpiece is processed by extruding the cutting layer on the surface of the workpiece to be machined.Its wear state directly affects the machining accuracy of the workpiece surface and the use efficiency of the computer numerical control machine tool.Therefore,how to accurately identify the tool wear state of computer numerical control machine tools,reasonable replacement of tools,and on the premise of ensuring the surface quality of the workpiece to be machined to improve the processing efficiency,is an urgent problem to be solved.When deep learning is applied in the field of tool wear state recognition,it has advantages such as feature self-extraction ability,simple operation and independent of expert knowledge.Therefore,this paper takes computer numerical control machine tools as the research object and carries out research on tool wear state recognition method based on deep learning.The main contents are as follows:(1)Data acquisition platform of computer numerical control lathe tool wear state.By analyzing the cutting characteristics of computer numerical control lathe and combining with the related tool wear theory,the paper summarizes the influence of various signal characteristics generated in the cutting process and the sensor installation position on the acquired signal quality.Finally,vibration signals were selected to build a data acquisition platform for tool wear state,and the whole-life wear state data of computer numerical control lathe tools were obtained,providing data sources for accurate identification of tool wear state.(2)The identification method of tool wear state between different tools is studied.Aiming at the problem of poor tool wear state recognition effect caused by the fact that the tool used in the training recognition model is not the same as the tool to be identified when the tool wear state is recognized in practical engineering.A method of tool wear state recognition based on joint matching of depth features is studied.Considering the advantages of deep learning feature self-extraction,SDAE network was used to self-extract the features of the sample spectrum to obtain the depth features representing the wear state.Combined with the characteristics of Transfer learning to reduce the difference of data distribution between the two domains,the depth features obtained are combined with TJM to reduce the difference of depth features of different tool wear states.The problem of accurate identification of wear state between different tools with the same type and wear state is solved.(3)The method of tool wear state recognition with Data Augmentation is studied.In order to solve the problem of low accuracy of tool wear state recognition due to the lack of label sample data of computer numerical control machine tool wear state in practical engineering,a tool wear state recognition method based on data enhancement was studied.Firstly,a small number of labeled data sample spectrum graphs are taken as input of DCGAN,and the distribution characteristics of the original data sample spectrum graphs are learned through continuous game training between generator and discriminator,so as to obtain new samples that are approximately consistent with the distribution of the original data sample spectrum graphs to expand the number of samples;Then,the expanded training sample set and test set are input into DANN together,and the differences between the training set and test set are narrowed through the continuous game between the feature extractor and the discriminant;Finally,Softmax classifier is used to output the recognition results.Through experiments,it is proved that the proposed method can identify the tool wear state effectively and accurately when the training samples are insufficient,and it has good generalization ability.