Solar panel soiling detection is an important problem as soiled panels produce substantially reduced energy. This paper describes a collaborative project between Arizona State University and the University of Cyprus on fault detection. The project is part of an NSF program called International Research Experiences for Students on using machine learning for energy applications. In this study, we focus specifically on two methods for identifying soiling in residential solar installations. The first method aims to calculate a daily energy-lost-due-to-soiling value by comparing two calculated power curves: the expected best case scenario curve and a weather corrected curve, which estimates what the day's power curve would be in the absence of cloud cover. The second method compares the performance of sites in the same weather region using a multi-level k-means clustering strategy. Initial results with ground truth feedback suggest that this second method is effective. The key take-away from this study is that these methods do not require feature rich datasets, which are often unavailable, rather they operate solely on time-series power values.