Multi-Step Prediction Of Tool Wear Based On Attention Mechanism And Reasearch On Few Samples Learning Algorithm Under Multiple Working Conditions

In the field of machining,accurate monitoring of tool wear status can make full use of tool life.On the one hand,it can prevent tool change before its service life,which will increase the cost.On the other hand,it can avoid tool change after serious wear,which will affect the machining accuracy of the workpiece.Therefore,it is necessary to monitor the tool wear status in the cutting process.In addition,on the basis of monitoring the current tool wear status,predicting the future wear value can make the machine tool make corresponding early warning before tool wear damage and replace the tool in time,so as to ensure the processing quality and improve the production efficiency at the same time.Therefore,this paper studies the tool wear condition monitoring and prediction algorithm.The main research contents are summarized as follows.(1)Based on the theory of tool wear,the trend of signal value and tool wear value of various sensors in the experimental data is analyzed.The signal data of the sensor is preprocessed,including removing invalid value,modifying abnormal value.And the signals are optimized by wavelet threshold noise reduction,which builds the data foundation for the subsequent tool wear model.(2)In order to realize the multi-step prediction of tool wear,a comprehensive model is established.Firstly,the wear value of the current time point is monitored by multi-sensor information in the monitoring model based on Dense Net.Then considering the tool and the processing state of two near time points are closer in actual processing,that is,the short-term information of the near time point has more influence on the prediction point than the long-term information from the far distance prediction point.A pre-test model based on attention mechanism is proposed,in which the coding decoding structure can realize multi-step prediction of tool wear value,so as to obtain the multi-step prediction of tool wear value More time is required for tool control.Finally,the effectiveness and advantages of the integrated model are verified by experiments.(3)In order to solve the problem that the model can not be used in practical machining under time-varying conditions.Firstly,the applicability of the small sample learning method in multi condition tool wear problem is analyzed,and the machining parameters are introduced into the model and coded by combining meta learning and transfer learning.Then,a meta learning method with model-agnostic is used in the monitoring model to get the meta learner to solve the problem of new working conditions.At the same time,the transfer learning method based on fine-tuning is adopted for the prediction model.By freezing several layers and encoding the processing parameters,the model input is added to the full connection layer.Finally,the feasibility and accuracy of the proposed method are verified on NASA dataset.(4)Based on the above algorithm,an intelligent software system for tool wear prediction is built.The system can realize the real-time monitoring of the current tool wear and the prediction of the future tool wear.The report can be saved and outputted.