Automated Identification of Open Circuits in Photovoltaic Arrays via Mask-RCNN, Thermographic Signal Processing and Hybridized Region-Growth Algorithms

We present a hybrid workflow for the automated detection of open circuits (open strings) in utility-scale photovoltaic (PV) plants using aerial infrared thermography. The approach combines three stages: (i) module localization via instance segmentation to delineate each panel, (ii) temperature profiling to compute per-module statistics and identify gradient breaks along strings, and (iii) a region-propagation algorithm that expands from anomalous seeds to recover entire strings, even when occlusions or partial detection errors occur. This design ensures robustness under heterogeneous environmental conditions and does not rely exclusively on large training datasets. The system outputs both quantitative indicators (per-module temperature profiles and string-level labels) and qualitative evidence (binary masks and annotated thermal images), enabling scalable fault localization across thousands of modules. Validation on multiple PV plants demonstrates consistent performance, with accurate detection of open strings, reduced false positives compared to purely deep-learning approaches, and processing times compatible with large-scale aerial inspections.