Predictive Maintenance: Fault Detection and Diagnosis in HVAC systems

  • Timothy Mulumba

Student thesis: Master's Thesis


Non-nominal operation of HVAC equipment wastes an estimated 15-30% of energy used in US commercial buildings. In the hot-humid climate of the UAE and the gulf region in general, HVAC systems consume an estimated 40% of overall electricity usage. Therefore sub optimal operation of indoor climate control equipment has a high financial and environmental impact on the region. Although many buildings in the country were built recently and are fitted with modern building automation and control systems, the potential of such systems for implementing intelligent monitoring and Fault Detection and Isolation (FDI) of the most energy intensive building systems is almost entirely untapped. We propose to improve the energy efficiency of commercial HVAC systems by implementing a Kalman filter based fault detection and diagnosis scheme to accurately identify certain categories of abnormal conditions that are the most prevalent in this region. We study the fault diagnosis performance of the Kalman filter and we propose a scheme for recursively estimating parameters using a sliding window. We compare the performance of machine learning techniques of Decision Trees and Neural Networks when used with parameter estimates of a time series model obtained from these system identification methods. We show on data obtained during ASHRAE Project RP-1043 for a 90-ton chiller that the algorithms using the Kalman filter parameter estimates significantly improve performance in comparison to those using raw data and sliding window. To demonstrate the utility of the approach, several examples of both suddenand gradual faults are presented.
Date of AwardMay 2014
Original languageAmerican English
SupervisorAfshin Afshari (Supervisor)


  • Ventilation and Air Conditioning (HVAC) systems; Air Conditioning; Maintenance; Cooling.

Cite this