Faults in the actual outdoor performance of Building Integrated Photovoltaic (BIPV) systems can go unnoticed for several months since the energy productions are subject to significant variations that could mask faulty behaviors (Leloux et al., 2014). Even large BIPV energy deficits can be hard to detect (Drews et al., 2007). BIPV systems could be manufactured in a wide range of sizes and installed in the most remote locations. In addition, the energy output of these systems is typically the only accurate information about them (Leloux et al., 2014). These characteristics of BIPV systems make cost-effective and automatic fault detection of BIPV systems challenging (Leloux et al., 2014). These faults could lead to safety issues and fire hazards (Zhao et al., 2015). This study assesses the feasibility of using the historical energy productions of BIPV systems as the only source of information to detect faults. This research introduces a new fault detection method that only uses the historical information about the BIPV energy productions. The new approach is cost-effective as the proposed method is capable of detecting the BIPV faults using only one source of information, the time series of historical energy productions. This innovative approach makes the automatic fault detection applicable to all sizes of photovoltaic systems integrated with any buildings, even those that are in most remote areas. The hypothesis of this research is that the rich information contained in the history of BIPV energy production provides invaluable information to detect these faults. The real-world performances of the PV systems contain valuable information that has not been fully utilized to evaluate the faulty PV systems (Leloux et al., 2015). This research capitalizes on this opportunity by providing a cost-effective approach for fault detection that works with minimal information that is typically available. Haeberlin and Beutler (1995) were probably among the first researchers that pointed out the possibility of online error detection if PV power and losses are normalized. Since 1995, several approaches have been proposed for the PV fault detection using normalized performance indicators. Existing approaches for fault detection are typically based on the calculation of a normalized performance indicator, such as performance ratio (PR), and operating condition data, such as solar irradiation (Woyte et al., 2013; Drews et al., 2007; Stettler et al., 2006). For example, Drews et al. (2007) developed a fault detection approach working based on the difference between the simulated and actual energy yields. This approach uses the satellite-derived irradiance and a PV simulation model. Chouder and Silvestre (2010) also developed an automatic fault detection procedure for PV systems based on the comparison of the simulated and measured energy yields. It takes into account the environmental irradiance and module temperature evolution. This fault detection procedure was further extended by Silvestre et al. (2013) for grid connected PV systems. Firth et al. (2010) used performance data of 27 PV systems over a one- or two-year period to construct simple empirical models of the performance of PV systems during normal operation. This performance during normal condition was used as a baseline to identify faults. Bonsignore et al. (2014) developed a neuro-fuzzy fault detection method that identifies faulty behavior by comparing the value of six parameters along with the I-V curves in normal and faulty conditions. Hachana et al. (2015) developed a PV emulator for both normal and abnormal operating conditions. The amount of power losses along with the information extracted from this emulator was used to detect defects. Platon et al. (2015) developed a fault detection method based on the comparison between the measured and modeled AC power productions. The model predicts the AC power production using solar irradiance and PV panel temperature. Ghasempourabadi et al. (2016) combined real-time shading simulations with BIPV performance monitoring to detect faults. Dhimish et al. (2017) created a fault detection algorithm based on the analysis of the theoretical curves that describe the behavior of an existing PV system considering a given set of working conditions.