Abstract
Electrical power substations are visually complex and safety-critical environments with restricted access and highly variable lighting; a digital twin (DT) framework provides a controlled and repeatable context for developing and validating vision-based inspections. This paper presents a novel sensor-centric DT framework that combines accurate 3D substation geometry with physically based lighting dynamics (realistic diurnal variation, interactive sun-pose control) and representative optical imperfections. A Render-In-The-Loop (RITL) pipeline generates synthetic datasets with configurable sensor models, variable lighting, and time-dependent material responses, including dynamic object properties. A representative case study evaluates how well the framework reproduces the typical perceptual challenges of substation inspection, and the results indicate strong potential to support the development, testing, and benchmarking of robotic perception algorithms in large-scale, complex environments. This research is useful to utility operators and asset management teams, robotics/computer vision researchers, and inspection and sensor platform vendors by enabling the generation of reproducible datasets, benchmarking, and pre-deployment testing.