Non-Destructive Testing System for Weld Defects Based on Improved YOLOv8

Regular inspection and maintenance of welds are crucial for critical industrial facilities such as large storage tanks. Traditional detection and identification methods rely on manual operation by professionals, leading to low efficiency and poor accuracy. For this problem, a flaw detection module for the non-destructive testing of weld defects in large storage tanks is designed in this paper. The module employs non-contact magnetic attraction technology to fix ultrasonic probes and integrates them into a robotic inspection system to achieve automated non-destructive testing. The collected data is analyzed using an improved YOLOv8 deep neural network model. Experimental results show that the improved YOLOv8 achieves an accuracy of 91.3% in detecting stomata defects and an overall mean Average Precision (mAP) of 97.5%. Therefore, this method effectively enables automated data acquisition and intelligent identification of weld defects in non-destructive testing.