Enhancing fault diagnosis in photovoltaic plants: Managing the simultaneity of faulty bypass diodes, series resistance increases, and partial shading effects

Modern photovoltaic (PV) monitoring systems can collect high-quality data; however, existing analysis tools need to evolve not only to identify yield issues but also to discern the nature of faults and degradation factors contributing to yield loss. Current AI-based Operation & Maintenance (O&M) tools primarily focus on detecting single faults, with their accuracy in fault recognizing, significantly diminishing in the presence of multiple concurrent issues. Consequently, diagnostics become unreliable when faults occur alongside other component failures or overlap with environmental effects. The lack of appropriately labelled data for training and validating AI-based tools is a critical weakness that impedes the rapid development of reliable diagnostic solutions. This study introduces a methodology designed to enhance the capability of machine learning (ML) algorithms in identifying and classifying faults, particularly under conditions where multiple failures or environmental effects occur simultaneously. Leveraging the RSE Fault Test Facility, which can replicate a variety of failure modes in real PV systems, we were able to generate over 1.6 million fault-labelled records across more than four years of diverse failure conditions. This study focuses specifically on combined faults scenarios involving short-circuited diodes, increased series resistance and shading effects, with a total of 32,806 labelled records of combined faults. The findings of the study provide a methodology foundation for the development of advanced fault detection and diagnosis tools, aiming to improve their reliability and performance when dealing with the simultaneous occurrence of the aforementioned faults.