Solar Photovoltaic Fault Diagnosis: A Comprehensive Survey of Conventional and Machine Learning Techniques

Faults are the principal impediment to achieving elevated levels of efficiency, reliability, and longevity in photovoltaic (PV) systems. In this article, we reviewed solar PV fault diagnostics techniques and compared different fault signature interpretations based on three primary data sources: electrical measurements, environmental conditions, and thermal inspection images. Five overlapping diagnostic perspectives are discussed: measurement modalities (electrical and thermal), statistical methods, optimization-based methods, machine learning techniques. The different approaches are compared regarding: the extent to which faults can be detected; quality of the diagnostic evidence that is generated; what measurements are required for diagnostics; how the topology of the system affects diagnostics; ability to localize faults; under what conditions a diagnosis is validated; and at what time interval the data needs to be acquired. Overall, our evaluation shows there is no single optimal method for identifying faults of all types, primarily because of differences in datasets, definitions of faults, validation conditions and timing measures which make it difficult to make direct comparisons. Finally, we identify areas for future research; Hybrid Physical/Data-Driven Models, Digital Twins, Embedded Inference, Standardized Datasets, Learning with Limited Labels, Longitudinal Field Evaluations.