Advanced monitoring of photovoltaic (PV) systems can ensure efficient operations, but extensive monitoring of large quantities of data can be cumbersome for the individual analyst. To address this challenge, Sandia National Laboratories‘ Fault Detection Tool project uses machine-learning algorithms embedded into data collection devices on-site or at a central server to detect performance issues and failures automatically. The algorithms, based on machine learning techniques such as Gaussian Process (GP), Laterally Primed Adaptive Resonance Theory (LAPART), and Support Vector Machines, can be used to detect and classify faults. In this work, a programmable data storage device, Raspberry Pi (RPI), was placed in-situ with an actual PV array (pictured)The data was stored in a local database. Using a GP algorithm, researchers accurately estimated performance. Based on the residuals between the computed estimate and the actual values, the fault detection tool classified the behavior as normal or in a fault condition. This machine-learning approach not only identified fault behavior, but also defined the lost energy revenue caused by the fault condition.