Research On Tool Wear Prediction Of Machine Tool Based On Transfer Learning

In actual production,due to the complex internal environment and active physical response of CNC machine tools,the data machine tool has a great loss of cutting tools in production activities.However,the traditional tool condition monitoring can not effectively use the real life of cutting tools,resulting in a certain degree of waste of cutting tools cost.In this paper,the prediction model of tool wear monitoring is established by using the data of spindle current signal of CNC machine tools,and the method of deep learning and transfer learning.It is expected to improve the utilization of tool life,so as to bring economic benefits to the actual production.(1)The collection and feature extraction of the spindle current signal of CNC machine tools.In view of the low cost and high real-time of the current signal acquisition method,this paper selects the current signal as the parameter of tool wear prediction.By using the open-loop Hall sensor,it is convenient to acquire the current signal of the machine tool spindle.At the same time,for the high-dimensional current signal affected by noise,this paper uses the methods of time domain,frequency domain and wavelet packet to extract the characteristics of current signal and cut the signal.Finally,through the correlation between the characteristics and tool wear state,the time domain mean value,variance,kurtosis and skewness are selected as the input characteristics of tool wear prediction model.(2)Research on the prediction method of cutting edge collapse.In view of the status of tool breakage in the actual machining scene,this paper uses the algorithm of deep learning to divide the prediction of tool breakage into a two classification problem.By modeling the current signal in the machining state of the tool,the relationship between the current signal and the normal state of the tool and the state of the tool breakage is mapped,and the prediction accuracy is 97.36%.At the same time,aiming at the problems of few data samples and unbalanced data types,a prediction model of tool edge collapse using anomaly detection method is proposed,which improves the recognition accuracy of tool edge collapse state.(3)Research on tool wear prediction method based on transfer learning.In the actual scene,the cost of data acquisition is high,so learning and modeling with fewer samples is a more practical method.In this paper,based on the migration learning method,the machining data in different scenes from the target domain data set is applied to the modeling of the target tool wear prediction,so as to realize the modeling and prediction with fewer samples for the tool wear prediction,and the final basis The tool wear prediction model based on migration learning achieves 74.26% prediction accuracy,which is 13% higher than that without migration learning.The tool wear prediction method based on deep learning and migration learning can realize the monitoring of tool wear status in CNC machine tools in real time and effectively.By combining the monitoring of chipping edge with the prediction of tool wear,it is more suitable for the actual machining modeling scene,which is conducive to improving the life utilization rate of the tool in the logarithmic control machine tools and the automation degree of the CNC machine tools,and reducing the manufacturing and production activities The production cost in will bring economic benefits to industrial manufacturing.