Intelligent Surface Flatness Detection of Waffle Slab Structures Using Terrestrial Laser Scanning

Abstract

Waffle slab floors in semiconductor fabrication facilities demand millimeter-level flatness tolerances. Conventional straightedge inspection is inefficient and limited to sparse sampling, while existing point cloud-based flatness detection methods face challenges of dense near-surface noise interference and nonstructural elevation anomalies introduced by formwork cover plates in waffle slab systems. This study presents a terrestrial laser scanning–based framework that addresses two intertwined challenges specific to waffle slab flatness detection. First, a two-stage denoising pipeline combining region-of-interest extraction with side-view projection-based density filtering removes near-surface construction clutter while preserving the true concrete surface. Second, because formwork cover plates cannot be reliably distinguished from the postpouring concrete surface, a fully automatic cross-temporal registration method based on column topology graphs is developed; cover-plate geometry extracted from prepouring scans is transferred to the postpouring coordinate frame via triangle-invariant random sample consensus coarse alignment and ground-constrained point-to-plane iterative closest point refinement, enabling targeted cover-plate removal prior to flatness computation. Flatness is then evaluated at both global and local (2 m) scales through grid-based elevation extraction and deviation mapping. Field validation on three working surfaces (approximately 2,625    m 2 total) in a semiconductor project achieves cross-temporal registration fitness of 87.2% with root mean square error of 1.8 mm, local flatness mean absolute error of 1.28 mm against manual measurements, and total processing time of approximately 150 min.