Eddy Current Detection And Classification Of Aluminum Plate Defects Based On GAN

Aluminum plate is a kind of metal material widely used in industry because of its anti-fatigue and strong ductility.However,due to various factors such as raw materials,rolling equipment and processing technology,and external environment,aluminum plates are extremely prone to defects in the production process.The existence of defects not only affects the appearance of the aluminum plate material,but also severely reduces the material’s strength,yield stress and other properties.If these defects are not found in time,serious accidents may result.Eddy current testing technology,with the feature that the coil does not need to be in direct contact with the tested object during testing,and plays a very important role in the detection of aluminum plate defects.Although the current eddy current detection method has been developed and mature,the existence of defect damage can be found from the detection signal,but the accuracy of the defect detection of aluminum plate and the accuracy of defect classification still need to be improved.In this paper,in order to improve the accuracy of eddy current testing and defect classification of aluminum plates,by combining methods such as convolutional neural networks and generative adversarial networks(Generative Adversarial Networks,GAN),research on methods of aluminum plate defect eddy current detection and defect classification are carried out.The main work is as follows:(1)Deep learning requires a great amount of data,but in practice,it is difficult to obtain a large number of defect detection data for aluminum sheet materials,and the number of defect samples restricts the detection effect of defects.This paper first uses mirroring,cropping,zooming and other methods to expand the data of aluminum plate defect eddy current detection image,and then uses generative adversarial network to generate new data of aluminum plate defect eddy current detection image as a supplement to the measured data.The experiment verifies the eddy current detection images of aluminum plate defects of different defect types,and uses different evaluation indicators to evaluate the authenticity and diversity of the generated images.The experimental results show that the proposed method can expand the data set and provide more sample data for the subsequent experiments of segmentation and classification of aluminum plate defect eddy current detection images.(2)Aiming at the problem that the edges of defects in the eddy current detection images of aluminum plate defects are not easy to identify,and the eddy current detection images have background noise interference,an improved generative adversarial network based aluminum plate defect eddy current detection image segmentation method is proposed.This method is based on the image segmentation model of generated adversarial network.The generator part adopts the idea of the U-net model.The attention module is used before the fusion of high and low-level features to adjust the weights of low-level features and high-level features,which is helpful to improving the use of image feature information enhances the feature of the defect area in the aluminum plate defect eddy current detection image and suppresses the background feature.Experimental results show that the proposed method has a better segmentation effect on the defect edge region of the defect eddy current detection image of the aluminum plate compared to other segmentation methods,and has better robustness to noise interference.(3)In the research of eddy current defect classification technology,although it is possible to classify defects by analyzing and processing the signal phase,frequency,and amplitude to obtain defect information.,it is difficult to extract features artificially,and the classification accuracy needs to be improved.In order to improve the classification accuracy of aluminum plate defect eddy current detection images,this paper proposes an image classification model for aluminum plate defect eddy current detection based on generative adversarial network.This method combines the features of aluminum plate defect eddy current detection images extracted from different convolutional layers of the discriminator network model,improves the feature extraction ability of the network model for aluminum plate defect eddy current detection images,and finally the feature classification obtained with a support vector machine(SVM)When the model is trained,the loss value is updated through the back propagation algorithm to update the parameters of the model,so that the generator and discriminator network can be further trained and optimized.Compared with other classification methods,this method improves the accuracy of eddy current detection image classification for aluminum plate defects.