Abrasive particle morphology database creation method based on conditional generative adversarial network

Abstract

An abrasive particle morphology database creation method based on a conditional generative adversarial network comprises: deeply mapping three-dimensional abrasive particle images to form two-dimensional images, and the two-dimensional images serving as output target images in a training sample; constructing an abrasive particle model according to the real abrasive particle image, which mainly comprises an abrasive particle contour, abrasive particle color filling and surface texture representation symbols, and taking the constructed abrasive particle model as an input image of a CGAN training sample; inputting the paired input images and the target image as samples of the CGAN into a network for training, and generating a generator for expanding the samples by the CGAN; and finally, carrying out network testing by using the input image as an input, counting an error between the generated image and the target image by using a discriminator, and finally judging whether the image is the target image or the generated image. According to the method, the effectiveness of loss function evaluation is improved, the accuracy of the network output image is also improved, and the similarity between the network output image and the real image is very high.

Description

technical field [0001] The invention belongs to the technical field of wear particle analysis in the field of machine wear state monitoring, and in particular relates to a method for creating a wear particle shape database based on a conditional generative confrontation network. On the basis of graph transformation, a three-dimensional wear particle sample database is created based on a conditional generative confrontation network method. Background technique [0002] In order to meet the requirements of intelligent, automatic, and real-time monitoring of equipment operating status, intelligent algorithms have been gradually applied to the identification of abrasive particle types, providing a new method for judging equipment wear and wear mechanism. However, when using intelligent algorithms for wear particle type identification, a large number of wear particle samples are the basis for training various intelligent algorithms or networks and optimizing the parameters of int...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the defect that the computer intelligent algorithm cannot be applied to the identification of wear particle types due to insufficient samples, the purpose of the present invention is to provide a method for creating a wear particle shape database based on conditional generative confrontation network. The conditional generative adversarial network (Conditional Generative Adversarial Networks, CGAN) is used to expand the typical wear particle samples, and then derive a large number of new typical wear particle images; the CGAN algorithm requires paired training samples, one image is the input image, and the other image is The target image, the network after training, hope that the generated image after the test has a high similarity with the real target abrasive image. The present invention transforms the three-dimensional abrasive image into a two-dimensional image after depth mapping, and uses these two-dimensional images as The output target image in the training sample; construct the abrasive grain model according to the real abrasive grain image, mainly including abrasive grain outline, abrasive grain color filling and surface texture representative symbols, and use the constructed abrasive grain model as the input image of the CGAN training sample, The paired input image and target image are used as the sample input network of CGAN for training, and then CGAN will generate a generator for expanding samples; finally, the input image is used as input for network testing, and the discriminator statistics are used to generate images and target images The error is finally used to judge whether the image is the target image or the generated image. When the generated image has a high degree of acquaintance with the target image, it proves that the network training is successful. At this time, the generated wear particle image can be used as a sample for the wear particle recognition algorithm

Images