Nonlinear estimation and fault detection in large-scale industrial hvac systems

Reliable and timely fault detection (FD) is a significant and still open practical problem in the Heating, Ventilation, and Air-Conditioning (HVAC) industry. In this chapter, we present a nonlinear estimation and FD method, developed recently by us, which is useful for systems with many nearly identical units (or subsystems) operating in a shared environment, as is the case for HVAC systems. Real world issues in HVAC which are problematic for other FD schemes are addressed, such as mode switching, parameter drift, preexisting faults, and incipient failure onset. The HVAC problemdomain is briefly introduced and a new class of hybrid system models is proposed to describe such multi-unit systems. A general algorithm for estimation and change detection is developed based on estimating a common Gaussian-mixture distribution for unit parameters which incorporates information from all units. Results from performance studies based on Monte Carlo simulation and illustration by real data from three operational HVAC systems are presented.