Fractured fragments with jagged and irregular surfaces, discovered at crime scenes, are identified as “being a match” using comparative microscopy and physical pattern analysis. The surface topography of said fragmented pieces, as measured by 3D microscopy, is utilized to establish a quantitative basis for declaring a match, complete with quantified probability and error rates. The comparison scale is configured to capture the transition of fracture surface topography from self-affine to non-self-affine (surface roughness that is independent of the observation window). At this transitional scale, fracture surfaces display distinctive roughness characteristics, determined by intrinsic material properties, microstructure, and exposure history to environmental and external forces.
In the case of the examined class of hardened alloys, which are common in cutlery and tool steel, the identified scale is approximately two times the grain diameter. This scale closely resembles the characteristic distance necessary for the initiation of cleavage fractures in semi-brittle and hardened metallic alloys. Consequently, the imaging scale required is approximately 20 times the grain diameter. For each pair of fractures, six overlapping images were recorded, with an overlap ratio of 50%. The acquisition of spectral representations for various wavelengths and critical features on the fracture surface was accomplished using the mathematical Fourier Transform. Subsequently, quantitative topological descriptions were devised for the image pairs by performing correlation comparisons on two spectral bands encompassing the transitional fracture scale. These frequency bands are bounded by frequencies corresponding to 2-4 and 4-8 grain diameters. Consequently, each set of fracture pairs under examination yields a total of 12 correlation values. A statistical learning tool was then formulated, employing multivariate statistical analysis methods to classify the fracture pairs based on this collection of 12 topological descriptors. This classification offers a foundation for establishing the uniqueness of forensic comparisons.
The efficacy of the proposed statistical learning methodology was assessed using a robust training dataset and validated with a set of 38 distinct broken pairs, encompassing knives fractured in bending and stainless-steel rods with comparable grain sizes, broken under either tension or bending. The versatility of this framework was also examined across various loading conditions and on ceramic fragments. Remarkably, all broken pairs were accurately classified. This framework establishes the groundwork for forensic applications involving quantitative statistical comparisons across a wide spectrum of fractured materials, characterized by diverse textures and mechanical properties.