Defect Detection from X-Ray Images Using A Three-Stage Deep Learning Algorithm

Defect detection is a crucial step in the process of manufacturing auto parts such as engines. Air bubbles are common defects in the engine which may result in engine failure leading to the breakdown of the car or even catastrophic accidents. Currently, X-ray images are used for air bubbles detection which adds complexity to the detection task due to the overlay of defects with complex engine 3D structures in 2D X-ray images. In this paper, we propose a three-stage deep learning algorithm to learn various patterns of the bubbles in engines. We then test the algorithm using normal and defected images. The results show that the proposed deep learning method can accurately identify bubbles in the X-ray engine images. This deep learning technique can also be extended to detect other surface level defects such scratches, missing components and physical damage. In this paper, we report that the accuracy of our defect detection method is above 90%.