Abstract: Monitoring the quality of crack welding repair for robotic arms is crucial for ensuring industrial safety. However, traditional manual detection faces challenges such as strong subjectivity, poor real-time performance, and susceptibility to complex welding environments such as metal reflections, smoke interference, and uneven lighting; The existing artificial intelligence based detection methods face challenges such as high welding image noise, complex and variable features leading to low segmentation accuracy, and insufficient model generalization ability, making it difficult to accurately identify defect areas. Therefore, this article proposes an intelligent monitoring method for the quality of crack welding repair of robotic arms. This method first performs wavelet decomposition on image blocks, and reconstructs high-frequency information using compressed sampling and orthogonal matching pursuit (OMP) algorithm; At the same time, an adaptive clustering strategy based on regional standard deviation is used to classify image blocks, and reconstructed images are generated through non local mean weighted fusion, effectively suppressing noise and enhancing detail edge information. Subsequently, the improved watershed algorithm was applied to adapt complex features for precise image segmentation. Finally, the joint coarse screening (based on grayscale similarity) and fine screening (based on maximum inter class variance method) strategies are used to extract defect areas, eliminate misjudgments, and achieve intelligent quality assessment based on defect geometric parameters (such as area, length) and quantity. The experiment shows that the proposed method significantly improves the image structure similarity (SSIM) and crack structure fidelity, with a confidence level of over 0.95 for hidden crack detection, effectively overcoming the difficulties of manual and existing AI detection, and achieving high-precision welding repair quality monitoring..