基于无监督学习的连铸铸坯缺陷检测方法

Defect detection method for continuous casting billets based on unsupervised learning

铸坯在连铸过程中受温度分布不均、流动速度不稳等多种因素的影响,容易出现各种表面和内部缺陷,如疏松、偏析、缩孔、裂纹、气泡、夹杂等。这些缺陷不仅影响产品的外观和性能,还可能对工程结构的安全性产生潜在威胁。针对这一问题,提出了一种基于无监督学习的连铸铸坯缺陷检测方法,该方法利用图像频域处理技术和深度学习算法学习连铸铸坯正常样本的图像特征,通过图像重建方式自动检测铸坯图像中的缺陷。首先,通过频率解耦模块对铸坯图像进行图像频率分离,得到铸坯的低频图像与高频图像。然后,采用带有自监督预测卷积注意模块的生成器网络集合,分别重建低频图像和高频图像。最后,通过判别器网络对铸坯的原始图像与重建图像进行判定,以确定铸坯图像中是否包含缺陷。实验结果表明,该方法能够有效检测连铸铸坯的缺陷,具有较高的准确性和可靠性,可为提高连铸产品质量和生产效率提供有力支持。

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.