Prediction Of Milling Cutter Wear Based On Depth Transfer Learning

The proposal of industry 4.0 has triggered a global industrial revolution with intelligent manufacturing as the core.Intelligent maintenance is an important part of intelligent manufacturing,and tool health monitoring is the key research object of intelligent maintenance.If the tool wear exceeds the corresponding index but is not replaced in time,it will cause the loss and waste of time and economy.Therefore,it is of great significance and value to predict the tool wear in the machining process.At present,more and more data can be obtained in the workshop,and the data is heterogeneous.However,it is difficult to make full use of the shallow machine learning used in the traditional tool health monitoring technology.In addition,in the actual processing process,the working conditions are diverse,different tools or processing parameters,tool wear law is not completely consistent,each working condition is likely to lack a sufficient number of labels.Therefore,in view of the two problems of multi-source heterogeneous data and lack of label samples in the workshop,this paper proposes the method of deep learning combined with transfer learning to process the multi-source heterogeneous data collected in the milling process,so as to realize the prediction of milling cutter wear under the condition of few samples.Firstly,this paper designs the milling cutter wear experiment,selects the voltage,current,three-axis vibration,acoustic emission data as the monitoring signal,carries on the experiment and collects the data,according to the change law of the morphology and width of the wear belt on the flank,analyzes the data and removes the abnormal value.Secondly,according to the collected data of voltage,current,three-axis vibration and acoustic emission,this paper carries out feature extraction and analysis.According to the principle of three-phase two table method,the voltage and current signals are synthesized into power signals.For each signal,the extraction includes square root value,peak value factor,waveform factor,waveform factor and so on Kurtosis factor and other seven time-domain characteristics and short-time Fourier transform spectrum time-frequency characteristic index.In view of the small sample label data set,a learning model based on convolution neural network and transfer learning is proposed to predict the wear of milling cutter,which achieves good results with small relative error.It proves that the deep transfer learning network model established in this paper can effectively predict the wear of small sample label data set with high accuracy.Finally,based on the development background of green cutting and dry cutting,the synchronous acquisition system of structured data and unstructured thermal image data is realized.The milling cutter wear experiment in dry cutting environment is designed and the data acquisition is completed.The state prediction of milling cutter wear is completed based on convolution neural network and thermal image data,which has a certain value and significance.At the same time,the availability of thermal image data in dry cutting is verified.The results show that under the dry cutting condition,adding heating image data as input can improve the prediction effect of milling cutter wear.