%0 Conference Paper
%K Bayesian network
%K Dynamic Bayesian network
%K Fault diagnosis
%A Ojas Pradhan
%A Jin Wen
%A Yimin Chen
%A Xing Lu
%A Mengyuan Chu
%A Yangyang Fu
%A Zheng O'Neill
%A Teresa Wu
%A K. Selcuk Candan
%B The 1st ACM International Workshop on Big Data and Machine Learning  for Smart Buildings and Cities (ACM BALANCES)
%D 2021
%G eng
%R doi.org/10.1145/3486611.3491124
%T Dynamic Bayesian Network-Based Fault Diagnosis for ASHRAE Guideline 36: High Performance Sequence of Operation for HVAC Systems
%8 11/2021
%X <p dir="ltr">A dynamic Bayesian Network (DBN) is proposed in this study to diagnose faults for building heating, ventilating, and air-conditioning (HVAC) systems that are controlled based on American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE)’s Guideline 36: High Performance Sequence of Operation for HVAC (hereinafter Guideline 36). Guideline 36 provides recommendations on supervisory-level control. HVAC systems that adopt these strategies have more comprehensive setpoint reset schedules and more advanced control logics than typical HVAC systems. It is hence of interest to understand how faults might affect the performance of HVAC systems that are controlled based on Guideline 36 and whether we can develop strategies to diagnose and isolate faults even for systems with such comprehensive control sequences. Contrarily to a Bayesian Network (BN), DBN method incorporates the temporal dependencies of fault nodes between time steps using temporal conditional probabilities. This allows fault beliefs to accumulate over time and thus improves diagnosis accuracy. In this study, the accuracy and scalability of the proposed method is evaluated using the data from a Modelica-based simulated testbed. Overall, the developed DBN shows good potential in diagnosing and isolating the root fault causes for HVAC systems that are controlled based on the Guideline 36 control sequence.</p>