Experimentally validated ANN-based sustainable fault diagnosis framework for solar PV systems

Abstract

Faults in solar photovoltaic (PV) arrays significantly degrade energy yield, reliability, and operational safety, particularly on the DC side where continuous energization increases fault severity. This paper presents an integrated Artificial Neural Network (ANN)-based framework for fault detection, classification, and location (FDCL) in grid-connected solar PV systems. An ANN model is trained using irradiance and ambient temperature to predict the expected array current and voltage under normal operating conditions. Fault detection is achieved by comparing ANN-predicted and measured power, enabling robust identification of abnormal operating states under varying environmental conditions. Following detection, a rule-based classification algorithm distinguishes among major DC-side faults, including open-circuit, short-circuit, degradation, and partial shading faults, using normalized electrical indices and adaptive threshold parameters. A low-complexity voltage-sensor-based fault localization strategy is further developed to accurately identify the faulty module or string with minimal sensing requirements. The proposed FDCL framework is validated through detailed MATLAB/Simulink simulations and extensive experimental testing on a real grid-connected solar PV plant under field operating conditions. Experimental results confirm high fault detection accuracy, reliable fault classification, and precise fault localization, demonstrating the practical applicability of the proposed approach for real-time monitoring and maintenance of solar PV systems.