Harnessing neural networks for precise damage localization in photovoltaic solar via impedance-based structural health monitoring

Abstract

Accurate fault detection and monitoring are crucial for maintaining photovoltaic (PV) system performance. While previous studies mainly focused on PV system faults, they often lack a comprehensive approach to integrating advanced diagnostic techniques, leading to duplicated research efforts and insufficient exploration of novel methodologies. This paper investigates the use of the finite element method to simulate the electromechanical impedance technique for fault detection and classification in PV systems. A 3D finite element model of a photovoltaic panel was created using ANSYS software to understand the basics of this technique. Studies on different locations of structural cracks were conducted to assess their impact on PV system output. For model verification, various fault and normal state simulation datasets were collected, normalized using data from piezoelectric sensors, and preprocessed. These datasets were then fed into an extreme learning machine (ELM) algorithm designed to predict and classify damage locations. The results highlight the superior efficacy of the ELM algorithm in defect detection, boasting an impressive overall accuracy rate of 85%.