3D fracture surface reconstruction of high-strength martensitic steel under quasi-static loading conditions

This work is devoted to a novel method for the reconstruction of three-dimensional (3D) surface models from a single two-dimensional (2D) image, in which advanced neural network algorithms are employed to address the fundamental limitations of monocular depth estimation. The proposed computational pipeline is structured around the sequential application of the state-of-the-art vision-language model Qwen3‑VL, which is combined with the OpenCV library for initial depth map generation and geometric preprocessing. The resulting output is subsequently refined through the VGGT model, by which multi-view geometric consistency constraints are enforced to transform sparse or implicit depth cues into a coherent and dense 3D surface representation. The principal novelty of the presented approach is found in the feasibility of constructing a high-fidelity volumetric surface model without reliance on stereoscopic pairs, structured light projection, or multi-view image sequences, as the reconstruction is accomplished exclusively from a single 2D photograph. For the purposes of computational validation, steel specimens of grade 55KhN2MA (55ХН2МА) were selected, the choice being motivated by their complex fracture surface morphology developed under dynamic loading conditions. It is demonstrated through experimental results that both the global macroscopic fracture profile across the entire specimen cross-section and its fine microstructural irregularities can be successfully reconstructed by the proposed modeling framework. This dual-scale reconstruction capability is considered to be of significant potential for quantitative fractography, as the precise identification and spatial delineation of brittle cleavage facets and ductile dimple rupture zones are thereby facilitated. Moreover, the geometric fidelity attained in the reconstructed models is shown to enable the automated detection and volumetric analysis of dispersed particles that are attributed to carbonitride phases, the spatial distribution and morphological characteristics of which are recognized as critical factors in the understanding of the material’s failure mechanisms.

1. Baikov Institute of Metallurgy and Materials Science, Russian Academy of Sciences. - ГЗ 075-00320-26-00
2. Kurnakov Institute of General and Inorganic Chemistry, Russian Academy of Sciences. -
3. Moscow Aviation Institute -