Research On Condition Monitoring Of Cavity Milling Process Based On Deep Residual Network

Compared with riveted structure,parts with integrated design technology have the advantages of high strength and high reliability,and are widely used in the field of aerospace.However,because their structure often presents the characteristics of multi cavity and thin wall,cutting chatter is easy to occur in the process of cavity milling.At the same time,because their materials are mostly difficult-to-machine metals,it is very easy to cause rapid degradation of tool performance.If the machining condition is not monitored,the surface quality of the workpiece will be reduced and the cutting tools and machining equipment will be damaged.Therefore,it is of great significance to carry out effective machining condition monitoring in the cavity milling process of this kind of parts.Aiming at the failure of traditional methods in monitoring the milling process of cavity with complex tool path,a tool wear prediction method based on short-time Fourier transform and improved deep residual network is proposed in this paper.Firstly,the signal is transformed into a time-frequency diagram using short-time Fourier transform.Then,in order to solve the problem that the deep residual network describes the machining state from a single angle,a feature fusion layer is added to the original model to improve it.Finally,the signal and time-frequency diagram are input into the improved deep residual network at the same time,the tool wear sensitive features screened by Pearson correlation coefficient in time domain and frequency domain are extracted from the signal,the time-frequency diagram is extracted from the residual block structure,and the tool wear sensitive features and time-frequency-domain deep features are fused in the feature fusion layer to complete the model training.The experimental results show that the average prediction deviation of the regression model established in this paper is 0.76%,which is lower than the original deep residual network,shallow convolution neural network and machine learning model based on artificial features.Aiming at the problem that a large amount of data needs to be accumulated and marked again for model reconstruction after working conditions change,which is costly,this paper proposes a cutting chatter state classification method based on short-time Fourier transform,improved deep residual network and parameter transfer learning.First,the signal and the time-frequency diagram are input into the improved deep residual network at the same time,and the cutting chatter mechanism features are extracted from the signal,which are fused with the deep features in the time-frequency domain and trained to obtain a cutting chatter state classification model under lowspeed conditions.Then,using the model and small sample data under high-speed conditions,a new model is quickly established through parameter transfer learning.The test results show that the classification accuracy of the above two models reaches 99.06%and 99.38%,respectively,which is better than the threshold model based on artificial features.Aiming at the online application of the theoretical model in this paper,a set of cavity milling process condition monitoring system is developed,with built-in model database storage,data acquisition and online monitoring modules,providing a standardized and complete monitoring scheme from establishment,storage,call to online application.