Enhanced fault detection in energy systems using individual contextual forgetting factors in recursive principal component analysis

Improving maintenance strategies is an important component in reducing wastes of energy in industrial systems and especially energy systems, as they represent a large proportion of our total energy use. The use of fault detection methods can help transition to proactive maintenance methods and automate the detection of faults using data from the systems themselves. However, the presence of normal operational changes must be accounted for to avoid false alarms, as static fault detection methods cannot handle them. A recursive fault detection method, specifically recursive principal component analysis (RPCA), is applied to real building data in this paper to aid in addressing this challenge. Additionally, improvements to the aspect of updating forgetting factors, which are integral to RPCA's functionality, is also proposed which allows for enhanced adaptability. The results show that the improvement increases adaptability when setpoints are changed and provides similar performance otherwise. Lastly, the application of this was shown to significantly reduce false alarms in the building application, while still detecting the known faults.