Abstract
Magnetic Particle Testing (MT) is a method for determining the presence or absence of defects in a test object by using a ferromagnetic test object that is expected to have defects as a magnet. This method is used to determine the presence or absence of defects by having a skilled technician evaluate the magnetic powder. It allows inspection without destroying the test object. However, there are some problems with Magnetic Particle testing, such as the possibility of overlooking defects. To solve these problems, this paper develops a deep learning based defect image classification method for automated Magnetic Particle Testing. The proposed method first performs segmentation based on the structure of Eff-UNet, which can perform segmentation with high accuracy even for small defects. Then, an algorithm that combines the result of segmentation with the last part of the encoder of Eff-UNet is used to determine whether a defect is present or not. Using this method, defects were classified from the images obtained from Magnetic Particle Testing. The results showed that Accuracy of 92.4%, TPR of 89.2%, and FPR of 7.62%.
