Home | Business News | Browse by Publication | H | HVAC & R Research

Estimating the size of incipient faults in HVAC equipment.

Article Excerpt
INTRODUCTION

A fault is defined as an unacceptable deviation of the system behavior. A fault or malfunction can disturb the normal operation of a system, causing an unacceptable deterioration of the performance of the system. The presence of faults must be diagnosed as early as possible to prevent such situations from occurring.

A fault diagnosis scheme usually performs the following tasks:

* Fault Detection--the existence of a fault, which leads to undesirable or abnormal behavior, is determined.

* Fault Isolation--the location of the fault (i.e., which component, sensor, or actuator has become faulty) is determined.

* Fault Identification--the magnitude, type, and cause of the fault are estimated.

Fault identification is necessary when a fault develops slowly and the associated loss of performance can go unnoticed. For example, the gradual fouling of a coil has been shown to result in increased energy consumption in air-conditioning systems (Montgomery and Baker 2006).

Interest in applying fault detection and diagnosis techniques to HVAC systems has grown rapidly over the last 15 years (Hyvarinen 1996; Dexter and Pakanen 2001), and many schemes have been proposed and applied successfully in practice (Katipamula and Brambley 2005). Most of these schemes estimate the magnitude of an incipient or degradation fault by detecting the occurrence of the fault when it has reached one of a finite number of predefined sizes (Dexter and Ngo 2001; Li and Braun 2007). In other words, each level of severity of the degradation fault is effectively treated as a different type of fault. The main disadvantage of this approach is that the number of possible faults may become unacceptably large if a precise estimate of the size of the incipient fault is required. There are, however, some schemes that determine the size of an incipient fault by estimating the values of the parameters of a system model, which correspond to particular faults, using on-line parameter estimation (Lee et al. 1996; Wang and Wang 2002; Buswell et al. 2003). Early fault identification schemes used ARMAX (auto-regressive and moving average with exogenous variables) models to describe the system and a recursive least-squares algorithm to estimate its parameters (Lee et al. 1996). The use of a linear model simplifies the problem of estimating the parameters on-line but restricts the applicability of the scheme to relatively simple faults such as measurement bias. Later schemes have used nonlinear system models based on first principles (Wang and Wang 2002; Buswell et al. 2003). The main disadvantages of this approach are the difficulty of estimating the parameters of a model that is nonlinear in its parameters and the high computational demands of the more complex parameter estimation schemes that must be used.

This paper describes an observer-based method of estimating the magnitude of an incipient fault in HVAC equipment. The magnitudes of two types of incipient faults, which are typical of slowly developing degradation faults in a cooling coil subsystem, are estimated using a fault identification scheme based on a fuzzy relational proportional-integral observer (FRPIO). Computer simulation is used to assess the performance of the scheme because of the length of time it takes for these types of faults to develop in a real system (Lee et al. 2004). The main causes of inaccuracy in the fault estimates are examined and long-term simulation results are presented, which demonstrate that an adaptive version of the scheme can generate satisfactory estimates of the sizes of the faults.

FAULT ESTIMATION USING A PROPORTIONAL-INTEGRAL OBSERVER

A proportional-integral observer (PIO), which is an extension of the traditional Luenberger observer, can be used for fault identification in linear dynamic systems (Linder et al. 1998). A similar approach can be used to identify the size of slowly evolving faults in nonlinear dynamic systems if the observer is based on a nonlinear model...

View this article FREE - Now for a Limited Time, try Goliath Business News
Free for 3 Days!