Research On Multi-condition Drill Wear State Monitoring Based On Deep Learning

Cutting tool is an indispensable element in cutting,and its state change has an important influence on the quality and precision of machining.Meanwhile,the cutting tool is also the most active factor in the cutting process,and the time variability of its state puts forward higher requirements for monitoring accuracy.Therefore,accurate tool condition monitoring is the key to ensure efficient production and machining.With the development of intelligent manufacturing mode,the role of data in manufacturing industry has been paid more and more attention.Data-driven methods represented by deep learning are gaining prominence in the field of tool condition monitoring.Different from traditional physical modeling methods,data-driven models have obvious advantages in processing large-scale,multi-dimensional and multi-scene data.In this paper,aiming at the problem of monitoring drill wear state in the process of drilling,a variety of monitoring models of drill wear state based on deep learning method are established,and finally the monitoring task of drill wear state under multiple working conditions is realized.The main research work of this paper is as follows:Taking the exchangeable drill as the research object,the processing characteristics and structure of drill are analyzed,and the failure forms and causes of drills are described,and the blunting standards are formulated based on drilling experiments.The signals generated during drilling are analyzed,and a drilling wear signal acquisition platform was built.The force,torque,vibration and acoustic emission signals of drilling process under various working conditions were collected,and the drill wear amount during drilling process was recorded.Finally,the multi-condition drilling process data set is obtained,which makes preparation for the subsequent signal processing and the establishment of wear monitoring model.A drilling signals processing system was set up,which includes signal preprocessing,signal noise reduction,signal feature extraction and selection.Firstly,invalid values and outliers in the signals were removed to solve the problem of poor acquisition quality.Then,wavelet denoising was used to further optimize the signals,and the features of time domain,frequency domain and time-frequency domain were extracted.The variation trend of each feature with drill wear was analyzed,and the feature with strong correlation with wear was selected by correlation coefficient method.A tool wear monitoring framework based on deep learning network was proposed.A variety of deep learning methods including convolutional neural network,residual network and long short term memory network are introduced,and a variety of monitoring models of drill wear state under constant working conditions based on deep learning are established by using the above methods,realizing the purpose of monitoring drill wear state under constant working conditions.Aiming at the problem that the prediction accuracy of single type of deep learning network is low and the monitoring process is cumbersome,an integrated network is proposed to monitor the drill wear state.Compared with the single deep learning network model,the integrated model has higher accuracy.The monitoring methods of drill wear state under multiple working conditions is studied.The drill wear data set under multiple working conditions was expanded by using generative adversarial network,and then a drill wear monitoring model under multiple working conditions was established based on transfer learning method.The effectiveness and accuracy of data expansion technology and transfer learning method in monitoring drill wear state in multi-condition are proved by comparing with other comparison methods.