Tool Wear Prediction Based On Multi-sensor Information Fusion And Transfer Learning Under Multiple Operating Conditions

The cutting tool is an important component of the machining process.The surface processing quality of the workpiece and the machining accuracy of the CNC machine tool are closely related to the state of the cutting tool.When the cutting tool is severely worn and not changed in time,the processing efficiency,quality and safety will be seriously affected,so it is particularly important for the research of tool wear condition monitoring during processing.Existing research is basically aimed at the prediction of tool wear under a single working condition,but the actual working conditions are complex and changeable,so it is of great significance to study the tool wear prediction under different working conditions.This paper proposed a tool wear prediction model based on deep residual network with multi-sensor information,and proposed a wear prediction model migration method under multiple working conditions based on transfer learning,so that the model can be applied to different actual working conditions,which plays an important role in monitoring the tool wear status in actual machining.The main works of this paper are summarized as follows:1.This paper analyzes the various sensor signals and wear value trends in the tool wear monitoring experiment.The abnormal values and invalid values in the collected sensor signals are pre-processed and the signals are optimized by wavelet denoising.180-dimensional wear features are extracted in the time domain,frequency domin and time-frequency domain,and the correlation between different features and tool wear values is compared and analyzed.Finally,the 48-dimensional signal wear features most relevant to tool wear are filtered based on the mutual information method and applied to tool wear prediction model.2.The traditional machine learning methods(Random Forest,Support Vector Machine)and common deep learning methods(Convolutional neural network,Recurrent neural network)are used to establish the tool wear prediction model under the single working condition.And a tool wear prediction model based on deep residual network is proposed.The comparison of the prediction accuracy of different models proves the effectiveness and advantages of applying the deep residual network to the tool wear prediction model.3.The tool wear prediction under different machining conditions is studied.The transfer learning method is used to carry out the migration research on the deep residual network tool wear prediction model under multiple working conditions.It is verified by multi-working condition tool wear experiments.The experimental results show that the tool wear prediction model based on transfer learning can be well applied to tool wear condition monitoring under multiple working conditions.And the feasibility of applying transfer learning method to tool wear monitoring is proved.Finally,the importance of multi-sensor signal characteristics is analyzed based on the XGBoost ensemble learning method.At the same time,an actual tool wear monitoring scheme combining offline and online is proposed.