1、International Journal of Heating,Ventilating, Air-conditioning and Refrigerating Research HVAC nor may any pan of this book be reproduced, stored in a retrieval system, or ansmitted in any form or by any means-=lecimnic. photocopying, recording, or other-without permission in writing from ASHRAE. Ab
2、stracs-Abstracted and indexed by ASHRAE Abstract Center; Ei (Engineering Information, Inc.) Ei Compendex and Engineering Index; IS1 (Institute for Scientific Information) Web Science and Research Alert; and BSRIA (Building Services Research (2) lumped parameter models (ordinary differential equation
3、s); and (3) filter models, in which dynamics are added to steady-state models via a first-order Hammerstein paradigm. The focal point of all these models is the thermal response, with the flow dynamics being considered only as a time delay. The review points out that the models surveyed generally ne
4、glect the fluid thermal capacitance and thus are not suitable for liquid flows, assume a fixed fluid mass flow rate, and do not treat the issue of time delay in a realistic manner. A starting point for several model implementations was the classical transfer function approach (Tobias 1973). From the
5、 partial differential equations representing energy balances on the fluid and conduit material, a transfer function, giving the temperature response at the outlet of the duct, was obtained. The form of this function made it difficult to transpose to the time domain, so a simplified function was reco
6、mmended for practical use. Tobias model does not Vie I. Hanby is assistant director and head of research at the Institute of Energy and Sustainable Development, De Montfort University, U.K. Jonathan A. Wright is senior lecturer with the Department of Civil and Building Engineering, Loughbor- ough Un
7、iversity, Loughborough, Leicestershire, U.K. David W. Fletcher is head of thermofluid engineering and psearch and D. Neil T. Jones is climatic wind tunnel supervisor with the Motor Industry Research Association, Nuneaton, U.K. 1 HVAC the third term (containing ) is the diffusive correction factor. T
8、his F dia- gram is shown in Figure 2 for n = 7 and = 0.0 l. A similar expression was derived by Bosworth (1949); experimental results for water flow in a pipe showed that Equation (1 1) tended to over- predict the extent of longitudinal mixing. It should be noted that the length-to-diameter ratio of
9、 the duct only affects the diffusivity correction term. Well-Mixed Nodal Model The approach described in this paper is to approximate the benchmark F diagram taken as that given by Equation (1 i) by defining a model consisting of a number of well-mixed nodes in series. For a step change in input, th
10、e response of a single node is an exponential rise: (12) -T F(T) = 1-e For a number of such nodes (i) in series such that the volume of each node is V/i, then the resulting F distribution is given by As the number of nodes is increased, the order of the response rises, and as the number of nodes app
11、roaches infinity the response approaches that of plug flow. Figure 3 shows a plot of Equation (13) for 20 and 80 nodes in series, together with the analytical result given by Equation (1 1). 6 HVAC a comparison with the TRNSYS Type 3 1 is not shown because this model, with air as the working fluid,
12、has effectively no thermal transients. The nodal model satisfactorily predicted the exit temperature of the air, including the “pla- teau” at around 140 seconds caused by a delay in the control system. The maximum error observed in this time sequence was 0.3”C. The time delay/first-order model of Cl
13、ark had effec- tively too much thermal inertia at this time scale, whereas the transfer function model of Tobias overpredicted the swings in temperature. CONCLUSION A dynamic model for the thermal response of ductlpipe systems is described, based on discret- ization of the duct into well-mixed nodes
14、. The model can be used to study the effects of variations in flow rate, input temperature, and pollutant concentration on the outlet conditions of the conduit. Consideration of the residence time distribution in a conduit with fully developed turbulent flow, calculated from the radial velocity prof
15、ile and eddy diffsivity, indicates an optimal level of dis- cretization of 46 nodes, although satisfactory performance was obtained with approximately 20 nodes. This result is substantially independent of the length-to-diameter ratio of the conduit. The method generates a response that includes a ch
16、aracteristic time delay but that does not need explicit access to the system time in a dynamic simulation, making it applicable within a wide range of simulation environments. If the simulation program internally generates the time step, the resulting dynamics should appear automatically. If the tim
17、e step is user-specified, then it must be set to a rather lower value than the mean residence time of the fluid in a conduit section. A comparison of the short-term (flow-dominated) response has been made with three pub- lished dynamic duct models. The nodal model shows the expected steep initial re
18、sponse (caused by the fluid flow characteristics), followed by a more gradual temperature rise due to the thermal dynamics of the conduit material. The response of the model in the thermal regime was compared with experimental test results where rapid changes were made in the inlet temperature of a
19、steel-lined, insulated duct of 12 HVAC Chen and Braun 2001; Dexter and Benouarets 1996; Dexter and Ngo 2001; Haves et al. 1996a; House et al. 2001; Hyvarinen 1996; Lee et al. 1996a, 1996b; Li et al. 1996; Peitsman and Bakker 1996; Salsbury 1996; Stylianou and Nikanpour 1996; Tsutsui and Kamimura 199
20、6; Yoshida et al. 1996). Steven R. Shaw is an assistant professor at Montana State University, Leslie K. Norford and Steven B. heb are asso- ciate professors with the Massachusetts Institute of Technology, and Dong Lu0 is a senior engineer with United Tech- nologies Corporation. 13 14 HVAC as implem
21、ented in the test building, abrupt faults were introduced as such and degradation faults were introduced over one- to three-day periods. Electrical power FDD methods are no different from others in their ability to find abrupt faults more easily than degradations. Test results presented in this pape
22、r are limited to the faults introduced in the test building and as such are demonstrations of the methods rather than comprehensive assessments of their effi- cacy. A final report and the companion paper (Norford et al. 2000,2002) summarize the results of the blind tests conducted as part of 1020-RP
23、. This paper, in effect, lays a foundation for the summary paper. While the presentation focuses on a small number of artificially introduced faults, the pre- sented FDD methods can in principle be extended to cover additional AHU faults and faults in other systems. The obvious prerequisite is that
24、any fault to be detected by these methods must Table 2. List of Air-Handling Unit Faults Detected and Diagnosed with Electrical Power Data Fault Type AU Mixing Section Stuck-closed recirculation damper Abrupt Leaking recirculation damper Filter-Coil Section Degradation Leaking cooling coil valve Deg
25、radation Reduced coil capacity (water-side) Fan Degradation Drifting pressure sensor Degradation Unstable supply fan controller Abrupt Slipping supply fan belt Degradation Note: Fault implementation is described in Norford et ai. (2002). 16 HVAC constant power Nearly constant power if piping is bala
26、nced Nearly constant power if piping Off Constant at minimum value Energized; minimal Increasing as a power to meet thermal polynomial function loads in piping Increasing with Tout of flow or speed Increasing as a polynomial function of flow or speed Increasing as a polynomial function Increasing wi
27、th To, outside airflow value is balanced of flow or speed aFor a VAV system with variable-speed supply and return fans, constant supply air temperature, and a constant-speed secondary chilled water pump. Table 4. Detectability of Faults with Electrical Power Data Stuck-Closed Leaky Leaky Reduced Pre
28、ssure Unstable Slipping Temperature Recirculation Recirculation Cooling Cooling Sensor Pressure Supply Fan Region Damper Damper Coil Valve Coil Capacity Drift Controller Belt 1 Yes No No No Yes Yes Maybe 2 No No Yes No Yes Yes Maybe 3 No Yes No Maybe Yes Yes Maybe 4 Yes No No Yes Yes Yes Yes models
29、fitted to these data. Outdoor dry-bulb temperature is a reasonable predictor because the sensible fraction of envelope loads (latent heat transfer excluded) influences the total amount of air delivered to occupied spaces. However, this predictor does not account for variations in air- flow, and henc
30、e fan power, due to changes in internal or solar loads. Correlations with measured airflow provide a more precise estimate of fan power, as established in an earlier study of appli- cations of electrical load monitoring to fault detection in ventilation systems (Norford and Little 1993). Such correl
31、ations have also been used to estimate fan energy consumption before and after variable-speed drive retrofits (Englander and Norford 1992). Figure 1 shows a third-order polynomial correlation between fan power and airflow for a VAV supply fan with a variable-speed drive. The use of a third-order pol
32、ynomial correlation is based on the fan laws, which show that power varies as a cubic function of speed for a variable- speed centrifugal fan. A similar correlation has also provided a good fit in practice for data col- lected from VAV fans equipped with inlet vanes. Ninety percent confidence interv
33、als were established from uncertainties in the polynomial coefficients and a t-statistic. The confidence intervals express the confidence that a single new measurement point will lie between the upper and lower intervals, if the measurement is subject to the same conditions as occurred during the tr
34、aining phase. Increasing the confidence interval would make the method less sensitive to faults and less likely to generate false alarms. In the test building, the use of 90% Confidence intervals did not generate false alarms, but in practice the confidence interval could be enlarged, reducing the n
35、umber of both detected faults and false alarms. To tighten the correlation and improve the sensitivity of the method, only training data with duct static pressure within 25 Pa (0.1 in. of water) of the 300 Pa (1.2 in. of water) set point were accepted. This tolerance on duct static pressure was arbi
36、trarily selected after the data were examined by eye and was intended to elimi- 18 HVAC the discharge air temperature will tend to rise above its set point; the valve controller will open the valve to send more flow through the coil and less to the bypass loop; the overall flow resistance will incre
37、ase and the pump will ride up the pump curve to a lower total flow and reduced pump power. Correlating pump power with valve control signal is therefore sufficient to detect flow block- ages under cooling loads sufficiently high that a substantial fraction of the total flow is directed through the c
38、oil. However, it is not likely to detect coil fouling, where a very thin coating of cal- cium carbonate can drastically reduce heat transfer across the coil but can have a small impact on flow resistance. The change in pump power as correlated with valve position is illustrated in Figure 5. The trai
39、ning period did not include system loads large enough to cause the cooling coil valve to open more than 70%. Pump power was assumed to remain nearly constant under higher flows through the cooling coil. Chiller power can provide another indication of faults in chilled water piping. For low to modera
40、te cooling loads, the valve controller will compensate for a flow restriction in the cooling coil by directing more water to the coil and less to the bypass piping. Under high cooling loads, the valve controller will saturate, flow through the coil will be less than needed to maintain the discharge-
41、air temperature, the building will be undercooled, and chiller power will therefore drop. Monitoring to determine a reduction in chiller power at high load also offers the advantage of detecting not only flow-restriction faults but also reduced thermal conductivity due to depos- its on the water sid
42、e of the cooling coil. Whole-building energy studies have correlated building electricity consumption with outside temperature as a means of analyzing the buildings energy requirements for cooling, typically with linear change-point models (Ruch and Claridge 1992, Ruch et al. 1993). More detailed st
43、udies of chiller power have established that chiller power is primarily a function of load on the chiller and the temperature difference between leaving condenser water and chilled water flows, and that a biquadratic functional form is a reasonable model (Braun et al. 1987). For an air- cooled chill
44、er as is used in the test building, outside air temperature directly affects condenser performance. In this study, an HVACSIM+ simulation of a building modeled for a controls simulation test bed (Haves et al. 1996b, 1998) was used to correlate cooling load, as measured by heat transfer across the co
45、oling coil, against outdoor temperature. It was possible to detect a fouled coil. This method was not applied in the test building to detect the coil capacity fault because the chiller is a two-stage reciprocating unit and power levels are discrete, rather than continuously varying. Cycling periods
46、between states were not regular and not easily discerned at high loads. At low cooling loads, however, chiller cycling was both regular and revealing of certain faults, as will be discussed. 22 3 i 400 350 -. a 300 250 3 HVAC THE AHU-A COOLING COIL VALVE WAS LEAKING 09:27-21:56 6000 3 4000 $ 3000 e
47、2 2000 1 O00 O O II 120 240 360 480 600 720 840 960 1080 1200 1320 1440 TIME, min. Figure 6. Detection of Leaky Cooling Coil Valve by Analysis of Cycling Period of Reciprocating Chiller The power-flow correlation for the fan that was readily made with submetered power data could not be generated wit
48、h electrical power measurements from the NILM installed on the motor control center in the test building. The NILM yielded one data point per day, when the fan was turned off in the evening. The startup point was not valuable because the fan motor has a variable-speed drive that has a slow and compl
49、ex startup pattern that is very different from an abrupt change most easily seen by the change-of-mean detection algorithm applied to data col- lected by the NILM. Further, there was little variation in flow at the time of fan shutdown and no opportunity to generate a polynomial relationship between power and flow. The limited range of data also made it impossible to correlate fan power with the motor-speed-control signal, a corre- lation established with submetered data and used to detect and diagnose the slipping fan belt. The fact that airflow, fan speed and fan power showed li
