A photovoltaic performance dataset for the benchmarking of failure detection and diagnosis algorithms

The race to reduce the Levelized-Cost-Of-Energy (LCOE) is driving the introduction of higher efficiency PV technologies with insufficient on-field testing, and the manufacturing of cheaper PV modules with lower quality control. Consequently, many new PV plants show degradation rates higher than expected in less than 5-10 years. Additionally, some severe failures that have been studied for more than two decades, like module electrical circuit corrosion, solar cell cracks, encapsulant yellowing or potential induced degradation (PID), can still be detected in these new PV facilities. Thus, the detection and diagnosis of these failure modes becomes essential to achieve the expected LCOE and PV system lifespan. Different strategies are proposed from the PV operation and maintenance (O&M) sector to address this challenge. Among them, the development of algorithms for the remote failure detection and diagnosis (FDD) from SCADA data is one of the most interesting. This approach promises a reduction of the on-field O&M costs thanks to an optimized predictive maintenance. Unfortunately, the development of FDD algorithms requires high-quality datasets with known degradation modes. Although there are PV performance datasets publicly available, they often present a lack of information of existing failures. To fill this gap some researchers have developed synthetic datasets with simulated degradation. However, these synthetic data do not reflect the complexity observed in real PV systems. This study overcomes this limitation through the continuous monitoring of a set of seven PV modules with well-characterized failure modes at the TECNALIA facilities. The LPVO-MS1X16 monitoring system developed by the University of Ljubljana was used for this purpose. Individual maximum power point (MPP) trackers recorded the MPP voltage and current of each PV module every minute. An IV tracer measured the IV curve of each module every five minutes. Module and air temperature, plane-of-array (POA) and global horizontal irradiance (GHI) were synchronously measured among other environmental parameters. The PV module set was composed by four defective devices from dismantled Spanish PV plants, two new modules in which failures were artificially generated, and one new module used as reference. The combination of visual inspection and electroluminescence revealed electrical circuit corrosion, interconnect ribbon breaks, solar cell cracks and encapsulant yellowing in the four defective modules. Additionally, failure modes were simulated in two new modules by artificially increasing the series resistance (RSERIE) and decreasing the shunt resistance (RSHUNT) respectively.