Mobile-focused spatial inspection of industrial parts using 2D image processing and LiDAR

• Mobile-based system for the spatial inspection of industrial parts. • Optimized deep learning with D LiDAR enables effective quality assurance. • Core inspection logic successfully moved to mobile for on-device processing. • Optimized pipeline enables near-interactive mobile performance with over 10 × speedup. Fixed optical benches and desktop 3-D scanners remain costly and inflexible for inspecting small manufactured parts in the field. This study introduces a fully mobile, on-device spatial-inspection pipeline that runs on an iPhone 15 Pro, combining Core ML-optimised YOLOv11 instance segmentation of 1920 × 1440 RGB frames with real-time ARKit LiDAR depth maps and calibrated camera poses. Detected 2-D masks are un-projected to 3-D, yielding metric contours used to compute diameters, slot lengths, and thickness. Bench tests on representative sheet-metal coupons show sub-millimetre mean-absolute errors for features below 10 mm (0.06 mm for 3 mm holes; 0.14 mm for 8 mm slots) and 2–3 mm error for 55 mm holes, while the complete analysis cycle averages 0.6 s on-device, enabling near-interactive feedback. A calibration mode alerts users when the handset drifts more than 2 cm from its reference pose, limiting accumulated depth error. These results demonstrate that consumer-grade LiDAR, when coupled with hardware-aware deep-learning compression, can deliver factory-relevant dimensional accuracy at a fraction of traditional metrology costs, though further work is required to improve large-feature precision and achieve sub-100 ms frame rates.