Whole-building commercial HVAC system simulation for use in energy consumption fault detection.(heating, ventilation, and air conditioning)(Technical report)

INTRODUCTION

Increasing energy costs have led to the need for a simple, reliable, and accurate diagnostic tool to gauge the energy performance of commercial buildings in real time. Historically, the energy efficiency of most buildings depreciates over time, due to issues ranging from ill-advised operational changes to failed or failing components, such as chilled water (CHW) or hot water (HW) control valves (Claridge et al. 2004; Liu et al. 2002). Fault detection and diagnostic techniques have been developed, but most focus on the component level or subsystem level and detect faults such as those in air-handling units or variable air volume terminal boxes (Norford et al. 2002; Salsbury and Diamond 1999; Xu and Haves, 2002). The focus of this paper is on the development and testing of a whole-building-level fault detection concept. Whole-building-level fault detection and diagnosis is an approach using measured building energy consumption to detect and diagnosis building-level energy consumption problems (Dodier and Kreider 1999; Breekweg et al. 2000a, 2000b). The magnitude of whole-building energy consumption faults using this approach is about five percent (Claridge et al. 1999). The technique described in this paper utilizes calibrated simulations to provide a visual comparison to the measured data. An "on-line" tool that will run in conjunction with the building's EMCS system is the ultimate goal. However, this paper focuses on describing and testing the proposed fault detection approach.

Liu and Kelly (1989) describe a two-step procedure for fault detection and diagnosis. The first step is to predict the system performance under a faultless state using a model and compare these values to measured output data. Significant differences are indicative of a fault. The second step concerns the diagnostic phase of the system, in which possible causes for the faults are constructed using a reasoning logic. This paper is restricted to examination of a whole-building-level fault detection approach.

METHODOLOGY

This paper examines the use of a visual comparison of calibrated simulation results and measured consumption data to facilitate detection of significant operational faults in a building. The comparison may also be performed using mathematical criteria to detect faults. This method is compared with the use of a visual inspection of the measured data alone, a time-honored method that can be beneficial in gaining insight into building problems (Claridge et al. 1992, 1999).

To effectively identify faults while minimizing false positives and false negatives, a rigorous methodology was developed. Fault detection studies have utilized a physics-based simulation and measured data to detect performance deviations (Xu and Haves 2002; Haves 1997). The residuals of these two data sets are then subjected to a threshold that is predetermined depending on how stable a system is. Systems that are unstable will require a large threshold range to minimize false positives. Usually three sample standard deviations of the residual under normal operating conditions are used as a threshold value (Montgomery et al. 1994; Rose et al. 1993; Farnum 1992; Fasolo and Seborg 1992).

There are two types of faults: complete (or abrupt failures) and performance degradations (Kelly et al. 1996). Performance degradations are gradually evolving faults. The methodology described in this paper is able to detect both fault types.

Several steps are necessary to ensure that the performance of the fault detection sequence is accurate. This sequence contains three preliminary steps, critical to the accuracy and performance of the system because they are tailored to the specific building's parameters. These steps are described below.

Step 1: Collect Information

The first step is to collect all the critical building and site data, such as wall composition, building orientation, internal load data, occupancy schedules, equipment data and schedules (including air-handling units and exhaust fans), measured consumption data, and weather data. This step is critical for the construction of a building-specific energy consumption simulation. For the simulation construction, about a month of data is needed. The accuracy of the energy consumption meters should be verified prior to using the measured data for simulation construction.

Step 2: Calibrate Simulation

The second step is to generate a calibrated baseline energy consumption simulation using the data collected in step 1. First, the measured energy consumption data should be examined to identify erroneous or missing data to enable generation of a clean measured data set.

To achieve a calibrated simulation of the building's energy consumption, known building parameters, acquired in step 1, are input into the simulation software to acquire a first run, or initial simulation of the building, which is then plotted against the baseline screened, measured consumption with outside dry-bulb temperature as the abscissa. The methodology chosen to calibrate the simulation presented in this paper used the following approach:

a. Adjust the cooling energy consumption profile of the simulated output, with very little attention being paid to the magnitude, to closely resemble the measured consumption profile. The profile of the simulated consumption can be adjusted using simulation inputs that are affected by the outdoor environment directly. These include conduction components such as wall compositions, U-values, ratios of glass area to wall area, as well as outside air fraction, CHW temperature schedule, etc.

b. Once the simulated cooling energy consumption profile closely resembles the measured consumption profile, the magnitude of the simulated consumption can be adjusted using simulation inputs that are not affected by the outdoor environment directly. These variables include the internal gains, occupancy, fraction of interior to exterior floor area, building area, etc.

c. Once the cooling model profile and order of magnitude closely resemble the measured consumption, repeat the process for the heating consumption. Steps a and b usually require more than one iteration because the cooling and heating consumption are sometimes functions of one another.

Step 3: "Correct" the Calibrated Simulation

The third step is to adjust the heating and cooling consumption values from the calibrated simulation so the totals of the simulated cooling and heating …