A knowledge-guided and data-driven method for building HVAC systems fault diagnosis. (July 2021)

• Record Type: Journal Article
• Authors: Li, Tingting; Zhao, Yang; Zhang, Chaobo; Luo, Jing; Zhang, Xuejun
• Is Part Of: Building and environment. Volume 198(2021)
• Publication Date: 2021-07
• Subjects: Diagnostic bayesian networks -- Genetic algorithm -- Fault diagnosis -- Fault interpretation -- HVAC systems -- Energy conservation; Buildings -- Environmental engineering -- Periodicals; Building -- Research -- Periodicals; Constructions -- Technique de l'environnement -- Périodiques; Electronic journals; 696
• DOI: 10.1016/j.buildenv.2021.107850
• Languages: English
• ISSNs: 0360-1323

Abstract

Abstract: Fault diagnosis is crucial for energy conversation of building HVAC systems. Generally, knowledge-driven fault diagnosis methods have good interpretability, whereas data-driven fault diagnosis methods have high diagnosis accuracy. With the aim of integrating the advantages of both types of methods, this paper proposes a knowledge-guided and data-driven fault diagnosis method. The proposed method develops a diagnostic Bayesian network (DBN) based on both expert knowledge and operational data. A probabilistic framework is developed for determining the prior DBN structures based on expert knowledge. An improved genetic algorithm-based approach is raised for further optimizing the DBN structures based on the operational data. Local casual graphs are generated from the DBN for visually interpreting the fault action mechanisms. Experts can evaluate the reliability of the diagnosis results using the local casual graphs, and then make reliable decisions. The proposed method is evaluated using the experimental data from the ASHARE Project 1312-RP. The results show that the performance of the proposed method is promising. Six typical faults are interpreted by the local casual graphs. It is demonstrated that the local casual graphs can effectively reveal the action mechanisms behind the six faults. Highlights: A knowledge-guided and data-driven fault diagnosis method is proposed. A probabilistic framework is used to generate prior diagnostic Bayesian networks. An improved genetic algorithm is raised to optimize diagnostic Bayesian networks. Fault action mechanisms are interpreted based on local causal graphs.