Defect Detection in Steel Castings Using Deep Learning with X-ray Images

For product reliability and workplace safety, flaw detection in steel castings is very important. The study presents a novel deep learning approach that can employ X-ray images to detect flaws in steel castings. With the help of a custom convolutional neural network called SteelNet, cracks, porosity, and inclusions can be accurately located and classified. In this approach, modern model characteristics such as attention processes and multi-scale detection are combined with domainspecific preprocessing methods, including noise reduction and contrast enhancement. Using transfer learning and data augmentation reduces the issue of data shortage and increases the generalizability of the model. The proposed method surpasses the state-of-the-art in defect localization using a varied dataset of X-ray images from steel castings, with a classification accuracy of 98.7 percent, a mean Average Precision (mAP) of 95.7 percent, and an Intersection over Union (IoU) of 92.0 percent. This technology enhances quality assurance in industrial operations by automating the process of defect identification and reduces the requirement for human inspection.