A fault diagnosis method for HVAC Air Handling Units considering fault propagation

Abstract

As a major component of Heating, Ventilation and Air Conditioning systems (HVACs), an Air Handling Unit (AHU) is used to condition and circulate air to rooms to satisfy human comfort requirements. In an AHU, cooling coil and supply air fan are two key components, where a cooling coil reduces air temperature and a fan delivers air to rooms. The fan and the cooling coil are interconnected, and a fault in one component may cause faults in the other. To save energy and satisfy human comfort requirements, diagnosing faults of the fan and the cooling coil is critical. In this paper, a method based on Coupled Hidden Markov Models (CHMMs) and Statistical Process Control (SPC) is developed to consider dependent component states and state transitions. Three major difficulties are overcome: (1) couplings among components are captured by the CHMM, and properties of CHMM states are considered to obtain reasonable distribution assumptions of state estimates; (2) a new initialization scheme based on both physical knowledge and observations is developed to identify appropriate initial values for CHMM parameters to avoid trapping at local optimums; and (3) coupling among components are also captured in SPC by considering Markov properties of state transitions to detect dependent faults. Experimental results show that the method can accurately diagnose faults with a low false alarm rate.