Energy Systems Condition Monitoring: Dynamic Principal Component Analysis Application

Abstract

Faults are estimated to cause 30 percent and 40 percent of energy consumption in building energy systems and district heating systems, respectively. It is therefore critical to detect these faults as early as possible to decrease this unnecessary waste of energy. Faults can also lead to lowered comfort of the customers. To detect these faults a data-driven methodology is applied, which utilizes dynamic principal component analysis (DPCA) for a generalized representation of the data, by projecting it into a subspace. In conjunction with DPCA, two multivariate statistical methods are applied for process condition monitoring: Hotelling’s T2 statistics and Q statistics. For fault diagnosis sensor contribution plots are utilized. The methodology has been applied to two cases: A district heating substation and a study space in a building in Denmark, with accompanying results and discussions. The methodology has proven to be easy to implement for both cases, showing that is exceptionally generalized and scalable. Furthermore, it has been able to detect known faults and identify the sensors responsible for the faults, in the data from the two cases. It has the potential to be adopted in real-time, however, more testing is necessary with other known faults.