![](https://csdnimg.cn/release/download_crawler_static/20251085/bg10.jpg)
14
HVAC&R
RESEARCH
A fault is detected when the observed behavior of a system differs from the expected behavior
by some threshold. The expected behavior of the system is often expressed in a model, whether
physical, statistical or fuzzy. The number of measurements required to use a given model for
fault detection is of central importance, as there is a cost associated with installation and mainte-
nance of sensors that are not required by the control system.
A fault is diagnosed when it is detected and a cause is determined. Diagnosis is significantly
more difficult than detection, because measured effects must be attributed to a particular cause
to the exclusion of all other possibilities. The difficulty of diagnosis increases with the number
of possible causes, and careful validation in the context of the target application is required for
any proposed diagnostic technique.
Current approaches to WAC FDD have been driven by the confidence researchers have in their
own modeling approach-its ease of implementation, robustness, and ability to generalize to many
different HVAC faults-and in the availability and accuracy of measured data. The approach in
this paper begins with the premise that electrical power measurements are useful in
FDD,
based on
the authors’ experience extracting device-level information from centralized power measurements.
The paper presents FDD techniques using submetered power data that, in some cases, may be
readily adapted for use with centralized measurements. Whether these techniques are economically
and technically useful needs to be explored in the field over the coming years.
There are four generic approaches to consider for use of electrical power data in HVAC FDD,
as shown in Table 1. Power signals can be obtained from meters attached to fans, pumps, chill-
ers, or other individual pieces of electrically powered
HVAC
equipment. Alternatively, power
data can be collected at a single point. Further, detection and diagnosis of faults can be based on
analysis of changes in steady-state electrical power, or analysis of the dynamic variation of
power during the typically very brief startup transient. For submetered power, steady-state is
defined as one-minute averages at times other than those encompassing a startup or shutdown
event or substantial power oscillation. In contrast, for centralized power measurements, steady-
state power is the difference in power before and after a turn-on or turn-off event. Techniques
presented in this paper cover three of the four approaches, as noted in Table 1. The detection and
diagnosis of faults from startup transients recorded centrally, the most powerful but difficult
method, has not been attempted to date.
Centralized load monitoring takes its name of Non-Intrusive Load Monitoring (NILM) from
its origins in load analysis in houses, where the normal revenue meter was replaced by a meter
capable of clustering step changes in real and reactive power and associating these clusters with
major household appliances (Hart 1992). Over
the
last decade, the early techniques have been
extended in two directions. Fust, enhanced detection of step changes in power have been shown
to screen out power spikes caused by switching electronics and to find power oscillations caused
by poorly tuned controllers (Norford and Mabey 1992, Hill 1995, Lu0 et al. 2002). Second, a
powerful detection approach has been developed
to
analyze the rapid changes in power that
occur when a motor, lamp, or computer is first turned on (Leeb 1993, Leeb et al. 1995, Norford
and Leeb 1996, Abler et al. 1998). The startup patterns are governed by the physics of the device
and are generally not masked by power electronics that reduce reactive power. In addition, anal-
Table
1.
Options for Detecting and Diagnosing HVAC Faults by Analyzing Electrical
Power Signals Explored in
this
Paper
Analysis
of
Changes in
Analysis
of
Power Dynamics During
Location
of
Meters Steady-State Power Equipment Startup
Individual loads
X
X
Centralized
X
Provided by IHS under license with ASHRAE
Licensee=Naval Aviation Depot/5918326100
Not for Resale, 01/19/2008 02:22:50 MST
No reproduction or networking permitted without license from IHS
--`,`,`,``,,`,`,,,`,``,,,`,`,,``-`-`,,`,,`,`,,`---