1、OR-05-1 3-2 Application of Fault Detection and Diagnosis Techniques to Automated Functional Testing Richard M. Kelso, PhD, PE Fellow ASHRAE ABSTRACT Extensive research in the field offault detection and diag- nosis hasproduced useful tools and techniques that have been applied to continuously operat
2、ing building HVAC systems. A few researchers have applied some of these to commissioning of new buildings. This paper reports on aproject that adapted or developed models of air-handling unit components and controls and combined them into an automatedfunctional test- ing tool. Operation of the tool
3、is demonstrated in testing a real air-handling unit. FAULT DETECTION AND DIAGNOSIS METHODS Faults in the sensors of a control system can be detected by the addition of redundant sensors, often called hardware redundancy. This approach utilizes several sensors to measure the same variable. A fault in
4、 one sensor is detected when its reading varies sufficiently from the mean readings of the other sensors. Such redundancy is expensive and can become quite complicated. For some time researchers have been investigat- ing the concept of automated fault detection and diagnosis (FDD) of HVAC systems (H
5、yvarinen and Karki 1996). In this concept, HVAC systems that have direct digital control (DDC) systems can be programmed to search for, detect, and diagnose problems in the control system itself or in the HVAC system. In FDD, models of the process provide analytical redundancy (Patton et al. 1989; B
6、raun and Rossi in Hyvarinen and Karki 1996). Analytical redundancy replaces hardware redundancy with dissimilar sensors measuring different variables, but which are functionally related by the state of the system. The application of FDD to HVAC systems has been studied exten- sively under the sponso
7、rship of the International Energy Agency (IEA) Annex 25 (Hyvarinen and Karki 1996) and Jonathan A. Wright, PhD Member ASHRAE Annex 34 (Dexter and Pakanen 2001). Norford et al. (2000) tested two types of FDD schemes with both known and unknown faults using the systems at the Iowa Energy Centers Energ
8、y Resource Station (IEC ERS), the location of the auto- mated functional tests described in this work. The fault detection and diagnosis work has been focused on identiQing changes in a system as it operates over extended time periods. Thus it can be considered a form of commission- ing on a continu
9、ous basis. The goal of finding faulty operation is the same, although the types of faults and the methods may differ. Some researchers (Dexter et al. 1993; Haves et al. 1993) have applied portions of the FDD theories to the commission- ing process. Salsbury and Diamond (1999) presented the results o
10、f an automated commissioning test on a simulated dual-duct air-handling unit. This paper describes another application of the use of models based on first principles to the functional testing of an air-handling unit. Faults in a system can be detected and diagnosed by comparing the values of output
11、variables against a set of rules that establish the values expected under various combinations of input variables for both correct and faulty operation. This method is relatively easy to develop and operate, but it has the distinct disadvantage that it cannot deal with unexpected conditions or fault
12、s that are not anticipated in the rules. Model- based fault detection and diagnosis uses reference models of the system or components to provide analytical redundancy. Values of output variables read from the system are compared with reference values predicted by the models. Differences between the
13、two, termed innovations in FDD work but labeled deviations in commissioning work, are indicators for detec- tion of faulty operation (Figure 1). Two broad approaches to model-based FDD have emerged from the research. One uses “black box” models such Richard M. Kelso is a professor at the University
14、of Tennessee, Knoxville, Tenn. Jonathan A. Wright is a senior lecturer at Loughborough University, Loughborough, Leicestershire, UK. 964 02005 ASHRAE. Control Deviations Figure 2 Detection of faults using first principles models and design intent parameters (Salsbury 1996). Figure 1 lnformation flow
15、 diagram for reference models. Reference parametem i Parameter deviations Figure 3 Diagnosis of faults by parameter re-estimation (Salsbury 1996). as neural networks. These models do not require prior knowl- edge of the physical relationships of the system but do require inputs from a correctly oper
16、ating system to “train“ the model so that subsequent incorrect operation is detectable. The models are only valid over the range of training data and cannot extrapolate outside this region. The second approach uses mathematical models derived from known physical relationships, or first principles. P
17、aram- eters for these models, if identified from design information, enable the model to represent the engineering design intent as the correct operation standard. Differences between values of model output variables and system output variables indicate incorrect or faulty operation. Figure 2 is an
18、information flow Reference M odcl Installed System Measured Figure 4 Fault diagnosis by expert rules. diagram showing the fault detection process used in this report. Faults detected by the presence of these differences, termed deviations herein (because they are initial differences and not changes
19、from initial agreements), can be diagnosed by comparison with a set of expert rules or by optimization of the parameters. Optimization is accomplished by altering the values of the parameters until the modeled outputs match the measured outputs or until the parameter changes are mini- mized. Differe
20、nces in the parameter values serve as indicators of the magnitude of faults, Figure 3 shows the method of diag- nosis using parameter re-estimation. Commissioning has the advantage over operational fault detection and diagnosis in that each component can be excited by a series of control inputs sele
21、cted to expose faults ifpresent. The fault can be isolated to the selected component by testing each component in series while progressing downstream along the air path of the AHU. The expert rules can be less complicated if each component can be tested in turn. Figure 4 diagrams the concept. ASHRAE
22、 Transactions: Symposia 965 Intake Preheat Louver Coil L Heating Water Cooling Water Figure 5 Diagram of air-handling unit and system. A building HVAC system must be tested when the construction schedule indicates, not when the thermal condi- tions are optimal. The models, then, must be reliable and
23、 accu- rate over a range of operating conditions, not just at fll load, and they must be able to extrapolate from the test conditions to design conditions. First principles models are suitable for such extrapolation. Simple but reasonably accurate models that incorporate parameters for control chara
24、cteristics such as leakage, nonlinearity, and hysteresis are required. Simplicity is desirable for ease of understanding and computer coding as well as for efficient use of computer memory. These principles of FDD will be applied in functional testing of the assembly of coils, fans, filters, and mix
25、ing box commonly identified as an air-handling unit (AHU) (Figure 5). This is one of the most important and common parts of an HVAC system. This assembly is the interface between the water conditioned by the primary plant and the air delivery system. In addition to the previously listed mixing box,
26、filters, heating and cooling coils, and fan, in some cases a return or relief fan is included in the AHU system and is coupled via pressure and flow to the supply fan. Because of its pivotal role in the HVAC system and its widespread use and because it is complicated enough to afford a challenging a
27、ppli- cation of the techniques to be investigated here, the AHU was selected as the first system of the overall HVAC system to be investigated. SIMULATING PERFORMANCE WITH FIRST-PRINCIPLES COMPONENT MODELS The approach to automated functional testing in this work is based on the proposition that a m
28、odel-based scheme devel- oped from FDD research can be used to commission an air- handling unit. An important part of this approach is the ability to use design information to establish values of parameters 7 + Deviation Figure 6 Overall plan of commissioning testingprocedure. that will enable the m
29、odels to accurately reproduce the intended performance of the system. Associated with this is the need to develop and test models that accurately portray the performance of the components over their range of operation. Still another significant task is to develop tests that facilitate the detection
30、of likely faults. Figure 6 illustrates the testing procedure developed for automated functional testing. The starting point is the development of component models and identification of the model parameters from construction document design data. In addition to steady-state thermal models, simple fir
31、st-order dynamic models and pres- sure models of the air system have been developed and eval- uated. The advantages of these models are in improving detection of faults and reducing the time required for the fnc- tional testing process to be evaluated. Figure 7 pictures the flow of information leadi
32、ng to the system model. A compan- ion paper describes the component models in greater detail. EXPRESSING DESIGN INTENT WITH MODEL PARAMETERS Design intent can be interpreted on various levels. Engi- neering design intent is defined as the construction docu- ments schedules, manufacturers schedules,
33、and published performance data. The designer interprets the owners design intent and writes or approves this information. This is the stan- dard selected as the required performance for the building and specifically for the air-handling unit and system under consid- eration here. Air-handling units
34、are custom-built from modular designs that allow a given size of unit to have many options of mixing box, filters, coils, fans, and arrangements. Once the designer has estimated the heating, cooling, and ventilation 966 ASHRAE Transactions: Symposia Table 1. Heating Coil Schedule Airflow 1 kgls 1 E!
35、 1 :; 1 1 E 1 y$ AHU-A 1.814 37.8 82.2 71.1 60.9 Symbol (cfm) (mW (3200) (100) (180) (160) (208) Owners design intent G Parameter values for the project ?- AHU system model Figure 7 Information flow in model development. requirements at design conditions, the next step is to select the components to
36、 deliver these design flow rates. The designer or a representative of the manufacturer, using catalogs with capacity tables or, more commonly now, selection software provided by the manufacturer, makes the selection. The designer communicates the engineering design intent to other participants by in
37、cluding on the drawings a schedule of the principal design performance variables of each air-handling unit. The schedule sets forth the design conditions and char- acteristic quantities at these conditions. Physical parameters, such as number of rows of tubes in a coil, may be explicitly stated or l
38、eft to the option of the manufacturer. Competitors bidding on the equipment make their own selections based on the schedule. There is no universal definition for the design quantities in the schedule. One designer may write into the schedule an estimation of the design loads and interpret them as mi
39、nimums and require equipment suppliers to meet or exceed the design values. Another may select equipment based on hisher esti- mate, then write the selected capacities in the schedule and interpret them as approximate targets for other manufacturers. In either case, some uncertainty is built into th
40、e selection and conservative selection with excess capacity is the usual result. Part-load performance of the component is not usually expressed in the schedule. Rather, the designer describes a sequence of actions the control system is to make to regulate the output of the component at less than de
41、sign conditions. (ft wg) 14.95 (0.25) Frequently the designer includes selection of a specific product. Final modifications of the design intent are some- times made when submittals for alternative products or mate- rials are approved. The form and content of these various expressions of design inte
42、nt are critical to the development and application of the commissioning models. The first principles models have parameters that reflect some aspect of the requirements from the previous paragraph and that calibrate the commissioning tool for the specific system under test. Thus, the models must be
43、developed to incorporate this infor- mation and to have parameters whose values can be deter- mined from these sources. A discussion of this factor, component by component, follows. A typical heating coil schedule indicating engineering design intent is given in Table 1. In this schedule, EDBT repre
44、sents entering dry-bulb temperature of the airstream and LDBT the leaving tempera- ture. EWT and LWT are entering and leaving water tempera- tures, duty (or capacity) is total heat transferred, and PD indicates pressure drop. The airflow rate is based on standard air at 1.2 kg/m3 (0.075 lb/ft3). Man
45、ufacturers are now publish- ing coil performance data in the form of computer programs that enable a designer to input some of the performance data from the schedules and to receive as output several choices of coils. Submittal data consist of certified performance tables and drawings of the chosen
46、coil. The models consist of equations developed from first principles of physics using standard methods such as heat balances. Among the variables in the equations, some are inputs from the test data and are called state variables because they describe the state of the system, some are inputs fixed
47、for the duration of the test and are calledparameters, and the rest are outputs. The parameters describe physical aspects of the components and are selected so that values can be determined from engineering design data presented in the construction documents or from manufacturers performance data. T
48、his is a key aspect ofthe premise of this study. If data are insufficient or the models cannot reliably predict system performance, the premise will not hold. The parameters for a heating coil model are width, height, number of rows, number of circuits, tube inside diameter, tube outside diameter, a
49、ir-side resistance, water-side resistance, metal resistance, UA scale, maximum duty, convergence toler- ance, water maximum flow rate, and supply air maximum flow rate. Of these parameters, maximum duty and maximum air and water flow rates are taken directly from the drawing schedule. Width, height, rows, circuits, and tube dimensions ASHRAE Transactions: Symposia 967 can be found indirectly from the drawings by referring to submittal data and other manufacturers literature. Air, water, and metal resistance can be found in the technical literature. For a control valve, the parameters
