Steel stud deformations in construction manufacturing cause screwing defects when fixed-angle fastening mechanisms encounter bent surfaces. Traditional 3D measurement requires extensive camera calibration (30–60 minutes) impractical for manufacturing environments. We present a smartphone-based stereo vision system achieving 0.20mm thickness measurement accuracy without formal calibration. Our uncalibrated approach combines RANSAC-robust fundamental matrix estimation, Fusiello’s rectification algorithm, and adaptive neural-stereo depth fusion optimized for reflective, textureless metallic surfaces. A differential geometry-based bend detection algorithm quantifies surface normal deviations from reference planes, enabling screwing operation optimization. Validation using vernier caliper measurements on industry-standard steel studs (24” × 3.75” × 1.75”) demonstrates 0.50mm position accuracy and 5.05 FPS processing rate. Laboratory testing with deliberately bent specimens (2°–25° angular deviation) shows 96% bend classification accuracy with 0.3° angular precision. The system reduces setup time by 98% while maintaining construction-grade accuracy.