Application-oriented multi-level framework for dimensional quality inspection of prefabricated laminated slabs

Abstract

Dimensional quality inspection (DQI) of prefabricated laminated slabs (PLS) is critical for on-site construction efficiency, yet existing techniques suffer from delayed inspection timelines and limited inspection scopes. To address these issues, an application-oriented multi-level DQI framework for PLS is proposed in this paper, namely, Identification-Segmentation-Measurement based on YOLOv8 and the Segment Anything Model (YS-ISM). YS-ISM enables the online DQI of PLS in production. YS-ISM consists of four steps: (1) pre-training of the benchmark identification model, (2) targeted improvement of the benchmark identification model, (3) pixel-level target segmentation, and (4) multi-target dimensional measurement. Numerical experiments demonstrate that the improved model, featuring the proposed Context-Enhanced Feature Fusion Pyramid Network (CEFFPN), enables more effective feature aggregation and contextual information propagation. By leveraging the channel partition selection mechanism of Dimension-Aware Selective Integration (DASI) and Shape-IoU, which balances focus on bounding box shape and scale, it overcomes inaccurate small-target localization caused by nesting and occlusion in PLS. The improved model achieves 82.4 % and 70.2 % mAP0.5_0.95 for holes and PVC boxes, representing 15.5 % and 10.9 % improvements over the benchmark model. The effectiveness of YS - ISM has been verified in practical applications. For the contour dimensions of the laminated panel and the coordinates of the PVC box, the maximum recognition errors are 1.4 % and 34.3 mm, respectively. For the hole, the maximum diameter error is 10.6 mm and the maximum coordinate error is 25.4 mm. which has a positive impact on prefabricated concrete components quality control. The outcomes of the application demonstrate the considerable potential of YS-ISM in the DQI of PLS, which has a positive impact on the development of prefabricated buildings.