Tool state image classification method based on edge mark graph neural network

Abstract

The invention discloses a cutter state image classification method based on an edge mark graph neural network. The method comprises the steps that S1, a high-definition electronic measurement microscope is used for shooting a cutter abrasion image when a machine tool is shut down during machining; s2, normalization and other processing are carried out on the cutter abrasion image, the image is input into an edge mark graph neural network to extract features, then adjacent edges are established according to the existing features and initialized, a full connection graph is formed, and each edgerefers to the relation type of two nodes connected with the edge; and S3, after the graph initialization is completed, updating the node features and the edge features, further obtaining the final node category prediction from the finally updated edge features, and obtaining the category to which the test set sample belongs by using a weighted voting method in combination with the sample labels inthe training set and the prediction values of the connecting edges of the sample labels and the test sample. The method has the advantage that the wear state of the cutter can be well recognized under the condition of small sample images.

Description

technical field [0001] The invention relates to a method for classifying tool state images, in particular to a method for classifying tool state images based on an edge label graph neural network. Background technique [0002] With the advent of Industry 4.0, manufacturing enterprises have higher and higher demands for intelligent production processes, and the automation of CNC machine tools is an important part of the intelligentization of most manufacturing processes. As the most easily damaged part of CNC machine tools in machining, cutting tools are very important for real-time and effective fault identification and status monitoring. The main reasons are: (1) According to statistics, in cutting processing, tool failures usually account for about 20% of the time, and frequent shutdowns and tool changes seriously affect the production efficiency of the enterprise; (2) If the tool breaks down and is not found in time, it will directly affect the surface quality and dimensi...

AI Technical Summary

Problems solved by technology

However, these methods all need certain preconditions to achieve ideal results. For example, the fast Fourier transform (FFT) requires the signal to be stationary, while the tool signal is non-stationary in the process of machine tool processing; although the wavelet transform (WT) It can deal with non-stationary signals, but it needs to construct and select a wavelet basis function that matches the fault characteristic waveform and has excellent properties

For the tool damage process of machine tool with little prior knowledge, it is very difficult to choose the appropriate wavelet basis function; the artificial neural network (ANN) algorithm needs a large number of sample data for training, which is very difficult for tool condition monitoring

In the case of limited samples, the artificial neural network (ANN) is prone to overfitting problems, that is, the generalization ability is poor; the hidden Markov chain (HMM) method assumes that the state duration obeys an exponential distribution, and the tool damage process is difficult to satisfy This condition; Support Vector Machine (SVM) shows good classification performance in the case of small samples, but it can only identify according to the signal characteristics at the current moment, and cannot make full use of the state information before and after the signal. Time-varying tool damage processes have limitations

Images