Defect detection method for continuous casting billets based on unsupervised learning
During the continuous casting process,various surface and internal defects such as looseness,segregation,shrinkage,cracks,bubbles,and inclusions often occur in billets due to factors such as uneven temperature distribution and unstable flow velocity.These defects not only affect the appearance and performance of products but may also pose potential threats to the safety of engineering structures.To this issue,this paper proposes an unsupervised learning based defect detection method for continuous casting billets.This method utilizes image frequency domain processing technology and deep learning algorithms to learn the image features of normal samples of continuous casting billets,and automatically detects defects in the billet images through image reconstruction.Firstly,the frequency decoupling module is used to separate the billet image into low-frequency and high-frequency components.Then,a generator network set with a self supervised predictive convolutional attention module is used to reconstruct low-frequency and high-frequency images respectively.Finally,the original image and reconstructed image of the casting billet are determined through a discriminator network to judge if defects are present in the image of the billet.The test results show that the method can effectively detect defects in the billets with high accuracy and reliability,and provide strong support for improving the product quality and production efficiency.
国家科技期刊平台
NSTL
万方数据
