An automatic welding defect detection method based on deep learning for super 8-bit high grayscale X-ray films of solid rocket motor shells

Abstract

The solid rocket motor has a wide range of applications in military weapons and model rockets. The shell is the main component, which is the welding and long-term operation, some defects will inevitably appear, which directly affect the performance of the solid rocket motor. This paper aims to solve the visual enhancement and defect detection of X-ray film of solid engine shells with unbalanced brightness and contrast and indistinct details in dark parts. To solve the problem that high grayscale RAW images cannot be displayed normally on low-bit monitors, an adaptive enhancement algorithm based on the high grayscale image is proposed. Further, to improve the observability of detailed information, a pseudo-color enhancement algorithm based on multi-chromatographic space fusion and controllable brightness is proposed. In addition, we constructed a new small sample dataset for super-8-bit welding defect detection and an object detection model that can be used to identify super-8-bit welding defects. The experimental results show that the method designed in this paper can effectively improve defect recognition in high grayscale RAW images, and can better detect defect types. In addition, we try to implement a texture mapping based 3D surface image rendering method and apply the 2D defect detection method to the 3D rendering image, which has a good detection performance and provides an effective idea for the 3D rendering of welding defects and surface defects detection.