Tool Wear Prediction And State Recognition Based On Deep Learning Under The Concept Of Internet Of Manufacturing Things

The rise of the Internet of Things(IoT),which has gradually infiltrated into manufacturing,has created a new manufacturing model,the so-called Internet of Manufacturing Things(IMoT).In the context of IMoT,the controllers and sensors deployed in workshops will generate a large amount of real-time data flow.Such data contains information about the operating status of the system(equipment).It is necessary to analyze the raw data and extract information and knowledge from it,so as to realize real-time monitoring and precise control of the system(equipment).Traditional shallow learning methods lack the ability to characterize non-linear models,requiring expert knowledge to extract suitable features from the raw data in order to train the model.In recent years,emerging machine learning methods represented by deep learning have attracted extensive attention in the academic and industrial fields,further promoting the application of artificial intelligence in many fields.In order to solve the defects of shallow learning methods,this paper takes the tool wear of cutting process as the research object,and addresses the use of deep learning methods to achieve tool wear prediction and state recognition.Firstly,the connotation of IMoT is addressed,and a prototype platform of IMoT is built.Raspberry Pi is used as the sensor network node and the function of real-time data pushing to the monitoring system is realized by using open source software.The research status of tool Prognostic and Health Management is analyzed,and a tool monitoring system is established under the IMoT platform,which lays the foundation for the subsequent tool wear prediction and state recognition.Secondly,for dealing with the noise in the machining vibration signal,smoothing denoising and wavelet threshold denoising are compared,which shows that the latter has a better performance.The advantages and disadvantages among the Short-Time Fourier Transform,the Wavelet Transform and the Hilbert-Huang Transform to deal with nonstationary and non-linear signals are analyzed,which shows that the Hilbert-Huang Transform gets the best.Thirdly,the tool wear prediction model based on BP Neural Network and Fuzzy Neural Network is obtained by using the traditional shallow learning methods with the artificial extraction feature as the model input,which is compared with that based on deep learning methods,and it is shown that the tool wear prediction based on the deep learning methods is more accurate.Finally,by taking the advantages of Convolutional Neural Network(CNN)in image processing,CNN is used in tool state recognition.For dealing with classifiers on small data sets,transfer learning is applied to tool state recognition.The results show that the transfer learning method helps to improve the accuracy of the small sample dataset classification model.