Knowledge discovery of data-driven-based fault diagnostics for building energy systems: A case study of the building variable refrigerant flow system

The data-driven-based methods, which rely on history data, are the most common methods used in the fault diagnostics of building energy system because of their simplicity. However, a major problem with the application of data-driven methods is its interpretability due to the complicated algorithm theory and structure. This paper therefore proposes a methodology which is able to conduct both fault diagnosis and diagnostic knowledge discovery for building energy systems. A case study is implemented in an experimental variable refrigerant flow (VRF) system. The clustering of variable around latent variables (CLV) method is used for variable selection. Then, a classification-based-on-associations (CBA) classifier is set up for fault diagnosis based on the mined association rules. It achieves an overall diagnosis accuracy of 95.33%. In addition, the class association rules (CARs) of the classifier are visualized by grouped matrix-based method and graph-based method, respectively. Further, the CARs with high confidences and supports are interpreted by domain knowledge in the individual fault level. Results show that the diagnostic outcomes comply well with the expert knowledge. The underlying system operational characteristics at faulty conditions could be mined and understood. Moreover, the diagnostic outcomes provide a reasonable and reliable reference for further FDD researches. •We proposed a method for the fault diagnostic knowledge discovery in building energy systems.•The CBA classifier is set up for fault diagnosis based on the mined association rules.•The association rules are visualized in individual fault level and analyzed by domain knowledge.•The diagnostic outcome can provide reference for further FDD researches.•Case study was conducted under various faults of the VRF system.