Data-driven Predictive Maintenance Alerting Routine for utility-scale Photovoltaic Systems

Predictive maintenance is crucial for optimizing the performance of photovoltaic (PV) systems by proactively addressing potential issues before they escalate. Even though many fault diagnosis methods have been proposed, a major challenge remains the lack of accurate predictive maintenance routines for utility-scale PV systems. The scope of this work is to address this fundamental challenge by proposing a data-driven predictive maintenance alerting routine (that leverages digital twin technologies and decision rules) for detecting and predicting PV system faults. The results demonstrated that the developed predictive maintenance routine yielded a sensitivity of 93.9% for PV utility-scale failure detection over a yearly test period. The findings of the predictive maintenance alerting model proved its capability to anticipate faults up to 10 days in advance, achieving a sensitivity of 83.3%. Finally, the predictive maintenance alerting system can be used by operation and maintenance teams for scheduling corrective actions and optimizing field activities.