Research And Implementation Of Tool Wear Prediction Platform Based On Sequential Convolutional Networ

In the process of machine processing,the tool will wear with the progress of machine cutting process,and serious wear of the tool may lead to avalanche,cause accidents and affect production safety.If the tool wear cannot be accurately predicted,the tool will be replaced before it is completely worn out,which will result in waste of the tool.Effectively predicting tool wear is an indispensable part of milling process.Therefore,it is important to design and implement a tool wear prediction platform for wear monitoring and wear value prediction alarm.In this thesis,a tool wear prediction algorithm based on residual time series convolution network is designed and implemented by studying tool wear during machine processing.During the processing of the machine,cutting force signals and vibration signals along X,Y and Z axes,as well as signals from acoustic emission sensors,are collected in seven dimensions.Firstly,the invalid values of the data are eliminated,and the data set is reduced by pre-processing such as data downsampling.Then,the time domain,frequency domain and time-frequency domain features of the original signal data are extracted and analyzed respectively,totaling 168 features.Finally,through the correlation coefficient method,the first 30 features with high correlation are selected to form a new feature matrix for training tool wear prediction.In the tool wear prediction model,the existing residual network adjusts the original step to learn more feature information.At the same time,in order to improve the prediction accuracy,a residual time series convolution network is constructed by combining attention mechanism with time series convolution network.The effectiveness and accuracy of this prediction algorithm are illustrated by comparing with the main algorithms.Compared with LSTM,the average absolute error is reduced by 14%,and the running speed is increased by 4 times.Finally,this thesis designs and implements a tool wear prediction platform,uses the tool wear prediction algorithm to display the tool wear prediction results in real-time visualization,and constructs modules such as online service and offline service.Online service provides user management,tool management,process management and tool wear prediction functions.Offline service mainly provides model training services.The system is validated through function and performance tests,and the mobile client and Web side are adapted to meet the usability requirements,ensuring the good application and engineering significance of the algorithm.