Identifying Outliers, Defective Areas and Performance Evaluation of a PV Module using EL Image Processing: A Variational Autoencoder Approach

Enhancing the commercial viability and efficacy of PV solar cell usage necessitates a comprehensive investigation into fault identification. This study introduces the development of a defect detection system tailored for identifying solar cell defects while approximating the subsequent power and efficiency losses. The methodology proposed encompasses two primary phases: Firstly, an anomaly detection model utilizing the Variational Auto Encoder identifies defects, reconstructs defect-free solar cells or modules, and generates a heatmap illustrating defect intensities. Subsequently, in the second phase, the defective area is quantified to compute total power and efficiency losses of a PV module. The method's outcomes, utilizing Electroluminescence (EL) images, machine learning, and defect simulation, demonstrate its capability to operate with minimal available data and estimate overall power and efficiency loss. This research establishes a correlation between defects and power generation loss, offering an inclusive estimation of PV module profitability, potentially mitigating the soft costs of PV systems and enhancing fault diagnosis methods.