Online identification, quantitative evaluation and quality grading for weld surface defects in engineering structures based on laser scanning

Online inspection of surface defects in welds is critical for ensuring the quality of engineering structures manufacturing throughout the comprehensive production, from sheet cutting and groove preparation to robotic welding. Restricted by low return on investment, the quality inspection still mainly depends on manual visual operation. This subjective, inefficient and inaccurate method calls for intelligent techniques to achieve precise identification, quantitative evaluation and quality grading of surface defects. An online intelligent detection workstation based on laser scanning is designed in this paper. Following a predetermined quality inspection trajectory, a binocular single-line laser sensor realizes fully automatic, all-round scanning of the workpiece to be tested. This process enables high-fidelity depth information of the weld surface to be obtained in real time. The obtained depth map is divided into equal-resolution subframes using an equidistant partitioning strategy to meet the requirements of high-speed online detection. Subsequently, the subframes are sequentially input into an optimized lightweight VGG16-UNet network according to their spatiotemporal order to accurately identify and classify the background, porosity, spatter, and incompletely filled groove. The network reaches 0.86 precision, 0.843 recall and 0.892 mDice with an inference speed of 65.73 frames per second (FPS). Furthermore, a contour analysis strategy extracts defect spatial locations (horizontal and vertical positions) and geometries (length and width) from the segmentation masks, restricting dimensional measurement errors to within 0.1 mm. Moreover, the 3D weld model is reconstructed and optimized through point cloud mapping, cubic interpolation, and Laplacian smoothing, facilitating precise quantification of geometric characteristics (height, area, volume). Finally, the quantitative evaluation results are combined with selected grading criteria from ISO 5817:2014 to enable automated defect quality grading. Experimental results indicate that the proposed technology provides accurate and intuitive characterization of the investigated defects under the tested conditions, thereby providing a technical basis for automated post-weld quality assessment and demonstrating its potential for intelligent post-weld inspection in engineering applications.