Advancing product surface defect detection and quality inspection with Gaussian Splatting: insights and preliminary experimental evidence

Scheduled inspection is a key element of industrial maintenance, enabling wear and deterioration to be detected before leading to severe failures in machinery or manufactured products. However, inspection records are typically constrained to viewpoints captured during acquisition, limiting the possibility of retrospectively investigating defects that were not adequately documented. This paper proposes the use of a ‘scene’, a high-fidelity digital representation of an object at a specific point in time, reconstructed through Gaussian Splatting (GS). Generated from a set of images, the scene acts as a digital archive of the object’s physical condition and can be interactively explored from arbitrary viewpoints, including perspectives not originally acquired. This capability enhances inspection flexibility and enables more effective retrospective assessment of defect occurrence, contributing to Industry 5.0 goals, including waste reduction and improved maintenance. Successive inspection stages are preserved as navigable scenes and systematically compared to identify surface anomalies and monitor their progression.