Simulated hybrid framework identifies and localizes module-level faults in 3 × 5 photovoltaic array

A hybrid deep-learning and circuit-analysis framework classified the topology of a simulated 3 × 5 photovoltaic array with 99.76% accuracy and localized module-level faults at reported accuracies of 100% for variable-resistance line-to-ground faults and 100% for zero-resistance line-to-line faults in solid intra-string bypass configurations. These findings are summarized in the abstract of an accepted Scientific Reports article that was published online on 12 October 2026 and remains subject to further edits.

The proposed pipeline separates topological classification from physics-based localization. Its first stage uses a hybrid CNN-BiLSTM-Attention architecture to classify the array’s topology, addressing ambiguities in electrical profiles at boundaries between healthy operation and partial-shading conditions. The identified topology is then supplied to a deterministic circuit-analysis framework, whose non-iterative mathematical core maps terminal metrics—calibrated module voltages, module currents and end-of-string current—to a physical fault location.

The study evaluated the approach under non-ideal operating conditions, including realistic cable and busbar resistances. Simulation results indicated that diagnostic performance remained high in the presence of measurement noise and parameter uncertainty. The localization method was also reported to be invariant to fluctuations in irradiance and temperature when those environmental parameters were uniform across the array. That condition limits the scope of the invariance result and does not establish performance under spatially varying environmental conditions.

This is a simulation study rather than a field deployment or laboratory evaluation of an operating photovoltaic system, and the supplied evidence comes from the article abstract rather than the full study. The reported performance concerns a 3 × 5 array and specific fault configurations, including solid intra-string bypass arrangements. The abstract presents the framework as a promising diagnostic method, but it does not provide broader validation across array sizes, sites, fault types or real-world operating histories.

Subject of Research: Deep-learning and circuit-based fault diagnosis and localization in simulated photovoltaic arrays

Article Title: A hybrid deep learning and circuit-theoretic framework for module-level fault diagnosis and localization in pv arrays

Article References: Abdullah, A. A., & Sadeh, J. (2026). A hybrid deep learning and circuit-theoretic framework for module-level fault diagnosis and localization in pv arrays. Scientific Reports. https://doi.org/10.1038/s41598-026-75608-8

Image Credits: AI Generated

DOI: 10.1038/s41598-026-75608-8

Keywords: photovoltaic arrays, module-level faults, fault localization, deep learning, CNN-BiLSTM-Attention, circuit analysis, partial shading, simulation study, line-to-ground faults, line-to-line faults, hybrid, deep