Thermal Battery Multi-Defects Detection and Discharge Performance Analysis Based on Computed Tomography Imaging

Highlights This paper focuses on the automatic detection of multiple defects in thermal batteries. It designs image acquisition experiments based on the X-ray CT imaging system and trains an automatic defect detection model based on the Yolov5s network. Discharge experiments for typical defects are also conducted to analyze the impact of different defects on the discharge performance of thermal batteries. The highlights of this paper include: An improved Yolov5s network is utilized to achieve high-precision automatic detection of typical defects in thermal batteries, significantly enhancing identification accuracy. Through thermal battery discharge experiments, discharge performance curves for normal and three defective batteries are established, providing a deep analysis of the impacts and mechanisms of different defects on discharge performance. An automatic stitching scheme is proposed to solve the issue of interlayer information overlap caused by the increased cone angle in Digital Radiography (DR) images. To tackle the problems of low image contrast and limited defect data in thermal battery imaging, the defect dataset is expanded through designed image preprocessing steps, improving the image contrast. The introduction of a multi-head self-attention mechanism in Transformer and the use of Focal Loss instead of cross-entropy loss function improve the recognition accuracy of subtle defects while ensuring detection speed.