1、OR-05-1 3-4 Model-Based Automated Functional Testing-Methodology and Application to Air-Handling Units Peng Xu, PhD, PE Member ASHRAE ABSTRACT The paper describes a model-based approach to auto- mated functional testing at the component level and presents results from preliminary jeld testing of a p
2、rototype soffware tool that implements the method. The method is based on an integrated life-cycle approach to HVAC commissioning and performance monitoring. The tool uses component-level HVAC equipment models implemented in an equation-based simulation environment. When used for commissioning, each
3、 model is configured using design information and component manufacturers data. Once an acceptable functional test has been performed, the model is fine-tuned to match the actual performance of the equipment by using data measured during the functional test. Thefine-tuned model is then used in routi
4、ne operation for on-line monitoring and fault detection. The paper describes the method and reports test results from NVAC secondaly systems in a commercial building and an experi- mental faciliq. INTRODUCTION There is a growing interest in developing automated func- tional test methods for building
5、 HVAC systems. Functional tests can detect operation faults in HVAC systems and so save energy, reduce maintenance costs, and improve comfort. Vari- ous functional test guidelines and libraries of procedures have been developed over the last few years to promote the practice of commissioning (Sellar
6、s et al. 2003). However, currently, functional tests are mostly conducted manually by commis- sioning agents, which is relatively costly and does not take full advantage of the capabilities of energy management and control system (EMCS). This indicates a need for an auto- mated functional tests tool
7、 that can be embedded in, or coupled Philip Haves, PhD Fellow ASHRAE Moosung Kim to, the EMCS to conduct the tests automatically. Automated functional testing has a number of potential advantages over conventional manual testing. It is expected to be easier to perform and more cost-effective, and it
8、 can be performed more frequently to detect faults earlier. In addition, the format of the data generated by automated tests is easier to standardize for data analysis. One approach to automating both commissioning and performance monitoring is to use computer-based methods for fault detection and d
9、iagnosis (FDD). Component-level FDD, which is the basis of the approach presented here, uses a bottom up methodology to detect individual faults by analyz- ing the performance of each component in the HVAC system (Hyvrinen and Krki 1997; LBNL 1999; Haves and Khalsa 2000; Ngo and Dexter 1998). In thi
10、s study, an automated fault detection tool has been developed, based on an integrated life- cycle approach to commissioning and performance monitor- ing. The tool uses component-level HVAC equipment models implemented in the SPARK equation-based simulation envi- ronment (SPARK 2004). When used for c
11、ommissioning, each model is configured using design information and component manufacturers data. Next, the behavior of the equipment measured during functional testing is compared to the predic- tions of the model; significant differences indicate the pres- ence of one or more faults. Once the faul
12、ts have been fixed, the model is fined-tuned to match the actual performance observed during the functional tests performed to confirm correct operation. The fine-tuned model is then used as part of a diagnostic tool to monitor performance and detect faults during routine operation. In each case, th
13、e model is used to predict the performance that would be expected in the absence of faults. A comparator is used to determine the significance Peng Xu is a mechanical engineer and Philip Haves is a senior staff scientist at the Lawrence Berkeley National Laboratory, Berkeley, Calif. Moosung Kim is a
14、 graduate student in the Department of Mechanical Engineering, University of California, Berkeley, Calif. 02005 ASHRAE. 979 of any differences between the predicted and measured perfor- mance and, hence, the level of confidence that a fault has been detected. A comprehensive review of model-based di
15、agnos- tics techniques is given by Simami et al. (2003) and a discus- sion of their application to HVAC is given by Benouarets et al. ( 1 994). In contrast to other fictional test procedures, which emphasize start-up and performance under design conditions, the automated functional tests described h
16、ere are designed to cover the full range of the system operation. The approach involves the use of both closed loop and open loop tests. Open- loop tests check whether the mechanical system works prop- erly over the full range of operation. Closed-loop tests check the coupled behavior of the mechani
17、cal equipment and the controller, identifiing problems relating to control sequences and their implementation, including loop tuning. In open-loop tests, controllers are overridden and the mechanical equip- ment forced to the desired operating points. In closed-loop tests, different operating points
18、 are achieved by manipulating the controller setpoint. There are two aspects of functional tests that can be auto- mated: the exercising of the system under test and the analysis Mixing Box Leakage of outside air damper Incorrect minimum position of out- door air damper Outside or exhaust dampers st
19、uck closed or partially closed Leakage of return air damper of the results. Tools that automate only one of these two aspects will be referred to as semi-automated. Automation of each aspect is discussed below. This paper describes simple open loop tests for mixing boxes, variable-air-volume fan sys
20、tems, and cooling coil subsystems and reports results of field tests designed to test the models and the data analysis procedures implemented in a prototype automated functional testing tool. Further results are presented by Xu et al. (2004a). SupplyIReturn Fan Range error in variable-frequency driv
21、e HeatingICooling Coil a significant difference in the output of the system indicates hysteresis. If the models used to analyze the results of the test are steady-state models, only measurements taken when the system is close to steady state can be used. At each step, a steady-state detector verifie
22、s that the system is in steady state before the data are recorded and the test moves on to the next step. Table 2 lists the minimum sequence of operating points for an open-loop mixing box test. The control points required for the test are: Measured Points Return air temperature ( Tret) Mixed air te
23、mperature (Tmk), if present and considered reliable Supply air temperature (Tsup), used when mixed air tem- perature sensor is missing or unreliable; subtract assumedkalculated temperature rise across supply fan to estimate mixed air temperature Outside air temperature (T,J Damper position (control
24、signal) Calculated Point (1) mix - rei out - ret OAF = Figure 1 illustrates the identification of the different fault groups from the measured outside air fiaction (OAF). The system is exercised by means of a series of step tests in which the damper position is increased from 0% to 100% and then dec
25、reased to 50%. At each step, the outside air fraction is calculated in order to identi the presence of one or more faults. The identification can either be performed by a direct comparison of the measured outside air fraction at different operating points or by comparing the deviations of the ASHRAE
26、 Transactions: Symposia 981 Step Number Demanded Valve Position (YO) 1 O L I I I - I Fractional Airflow (YO) Fault to be Detected Minimum Leakage 2 10 3 25 I41 1 O0 I 1 O0 I CaDacity I Minimum Valve/actuator mismatch 1 O0 Nonlinearity - - 5 1 O0 6 90 171 25 I 1 O0 I Hysteresis I Minimum Valve/actuat
27、or mismatch Minimum Valve/actuator mismatch Table 4. Open-Loop Test Sequence for VAV Fan Subsystem Step Number 1 2 3 Demanded Fan Speed/Capacity (YO) Terminal Box Damper Openings Fault to be Detected O Minimum Range error in VFD/inlet guide vane, sensor offset 50 Minimum Hysteresis in VFD/inlet guid
28、e vane 90 Maximum Range error in VFD 4 5 measured outside air fractions from those predicted by a refer- ence model (see below). Table 3 shows the minimum sequence of operating points for an open-loop heating or cooling coil test. Minimum airflow is used to detect control valve leakage since this ma
29、ximizes the air-side temperature change across the coil. Minimum airflow also minimizes the temperature rise across the supply fan and the corresponding uncertainty in that temperature rise, minimiz- ing the error in the inference of off-coil temperature from the measurement of supply air temperatur
30、e (see below). Minimum airflow is also used for Steps 2,5, and 6 for the same reason. A control signal value of 25% is used for Steps 3 and 7 since control valves are typically oversized, with the result that the temperature rise or drop across the air-side of the coil changes more rapidly in the fi
31、rst half of the operating range than in the second half. It is then easier to characterize any nonlinearity and hysteresis at a control signal value that is significantly less than 50%. In situations in which it is not possible to vary the airflow rate or there is pressure of time, the test can be s
32、implified by using the same airflow rate for all steps. Table 4 shows the minimum sequence of operating points for an open-loop test of a VAV fan subsystem consisting of either a supply fan and the associated supply system flow resistances or a return fan and the associated return system flow resist
33、ances. The supply and return fans should be tested in parallel to limit the variation in building pressure. After passing the open loop tests, the system is then subjected to the closed loop tests. The procedures for the closed loop tests are similar to those for the open loop tests, except that it
34、is the control setpoints that are changed instead Maximum Capacity 1 O0 50 Minimum Hysteresis in VFD/inlet guide vane of the control signal. The setpoints are increased in steps from their minimum to their maximum allowed values. As in the open loop tests, a steady-state detector analyzes the trende
35、d data to determine whether the system is in steady state before moving on to the next step. The test procedures for the mixing box, fan, and coil are described below. The measurement points required for the tests are listed, and rules for faults diagnosis are presented also. The diagnostic procedur
36、e for mixing boxes is illustrated in Figure 1, as an example. The diagnostic procedures for fan and coil subsystems are essentially similar. Execution For each functional test, there are a number of conditions that must be satisfied before the test can be conducted. For example, for the mixing box t
37、est, the fans must be running and the difference between the outside air temperature and the return air temperature must exceed a minimum value (ideally -20F 1 1 OC) to avoid the estimation of the outside air frac- tion being dominated by sensor errors. If the supply air temperature, corrected for t
38、he temperature rise across the fan, is used as a proxy for the mixed air temperature, the heating and cooling coil valves must be closed. Since these valves may leak, it is better to turn off the circulation pumps as well. These and other checks need to be incorporated in a fully auto- mated tool. A
39、 test signal generator was developed to facilitate the automated functional tests. The tool automatically gener- ates the test sequence and signal described above. There is a trend module to record the data and a steady-state detector module to determine whether the system is in steady state. The 98
40、2 ASHRAE Transactions: Symposia I , Start * 1 _+ 4 set-points Disable feedback wntroiier Closedloop Readinminandmax Step upfdown set point or controi signai Readin set-point and trending data point Read in control Signal and trending data point I t 4 yes I -1 Endofsequence? I Figure 2 Automated test
41、 signal generator step tests will move on to the next step after the system reaches steady state. Figure 2 is a flow chart of the active test tool. On start-up, the tool requires the user to choose between the closed loop test and the open loop test. If the open loop test is selected, the feedback c
42、ontrol loop is then disabled. If closed loop is selected, the maximum and minimum values of the setpoint are needed as inputs. After that, the program requires the user to input the addresses or names ofthe control and sensor points. The step test generator will then override the control signal valu
43、e automatically, based on predefined sequence, as described above. The new value is then uploaded into the controller. The trended data are analyzed in real time to determine whether the system is in steady state. When the system reaches steady state, the tool will move to the next step, until the e
44、nd of the test sequence. The software structure is generic, with only the data transfer between control system and the sofware being vendor-specific. AUTOMATED ANALYSIS In the approach adopted here, the analysis of the test results is divided into two stages: fault detection and fault diagnosis. Fau
45、lt detection is performed by comparing the measured behavior to that predicted by a model configured using design information and manufacturers data. The models may be based on first principles or may be empirical or may be a hybrid, depending on the component and the nature of the information that
46、specifies expected performance that is available during commissioning. Models of the three air-handling unit subsystems discussed above are described below. Fault diagnosis is performed by using rules to analyze the variation across the operating range of the deviation between the expected and measu
47、red performance. The fault diagnosis method is not addressed further in this paper. DESCRIPTION OF MODELS AND THEIR DATA REQUIREMENTS The field tests are focused on three component models in the model library described by Xu and Haves (2004): the mixing box, the VAV fan subsystem, and the cooling co
48、il subsystem. A brief overview of these models is presented here; more detailed information is given in Xu and Haves (2004a). Table 5 list the inputs, outputs, and parameters of the models. In general, measured values of all the inputs to a model are required to drive the model, and a measured value
49、 of at least one of the model outputs is required in order to compare the performance of the real system to that predicted by the model. One issue in modeling for fault detection is that some degree of imperfect operation may be tolerated in prac- tice (e.g., leakage of valves or dampers) and so must be included in models that ostensibly represent correct operation. Mixing Box Prediction of the mixed air temperature and humidity in an air-handling unit involves estimating the outside and return air fractions and then performing heat and moisture balances on the mixed ai
