Deviation Assessment of Complex Curved Surface Steel Structures Based on 3D Laser Scanning

• A millimeter-level TLS bias evaluation method integrating voxel downsampling + SOR+ROR + MLS denoising and smoothing has been integrated. • Taking the free form steel structure of the shell-shaped conference center as an example, two substructures are compared to reveal the mechanism of boundary progressive drift and undulation. • Provide practical construction control strategies to form a closed loop of 'scan—diagnose—correct—retest'. • Processes and parameters are reusable, facilitating rapid deployment and quality review in complex curtain wall/irregular steel structure projects. To support the installation accuracy control of complex steel structure frames for curved curtain walls, this study proposes an end-to-end deviation assessment workflow based on terrestrial 3D laser scanning for millimetre-level precision control during the installation of steel structures. The workflow integrates multi-station acquisition, registration, a parameterised point cloud denoising pipeline in MATLAB, including hollow ROI clipping, voxel down-sampling, statistical outlier removal, radius outlier removal, and moving least squares smoothing, and model-to-cloud deviation analysis. A shell-shaped conference centre is used as a validation case involving two steel structures. The workflow reduces point clouds by 88–94%, with registration errors controlled within 1.2–1.6 mm. Deviation analysis shows that Steel Structure No. 2 achieves a mean deviation of 1.93 mm and an RMS of 7.85 mm, while Steel Structure No. 1 shows larger deviations (mean 8.51 mm, RMS 25.97 mm). The workflow improves installation efficiency by approximately 30% while ensuring quality. The proposed workflow for deviation assessment of steel structures based on TLS is referential.