Reconstructing industrial workpieces from multi-view images is challenging because these objects often have weak texture, repeated structures, and reflective surfaces. In this paper, we present IndVGGT, an end-to-end industrial reconstruction framework based on VGGT. To better preserve sharp geometric structures, we introduce a High-Frequency Enhancement Module (HFEM) and a Curvature-Aware Normal Loss. We also build an industrial benchmark that includes 45,000 training models and four evaluation sets covering synthetic, industrial-specific, out-of-distribution, and real-world cases. Experiments show that IndVGGT consistently improves reconstruction quality over the original VGGT baseline. On Ind3D-Syn, the overall error is reduced from 14.82 to 3.40, and on Ind3D-Spe from 5.31 to 1.04. On the real-world set, the Chamfer Distance decreases from 1.13276 mm to 0.91909 mm. These results show that IndVGGT improves reconstruction quality across synthetic and real-world benchmarks, while reducing the gap to line laser scanning on real industrial data.