1、2011 ASHRAE 777This paper is based on findings resulting from ASHRAE Research Project RP-1328.ABSTRACTRecently, an ASHRAE research project (RP-1328) wascompleted in which an algorithm for a hybrid model was devel-oped (Kashef and Hadjisophocleous 2010). The hybrid modelcomprised two integrated model
2、s: a zone and a network model.The two-zone model was developed to simulate fire and smokemovement inside the room of fire origin and neighboringrooms. The network model capable of predicting both mass andenergy flow is used to simulate smoke movement into the restof the rooms that are far away from
3、the fire-origin room. Thetwo models were combined to produce a hybrid model thatallows an accurate simulation of fire dynamics in both the nearand far field. The output of the zone model provided a directinput into the network model that included the energy equation.The different steps involved in t
4、he development of thehybrid model were included in Kashef and Hadjisophocleous(2011). The application of this model permits a reasonablenumerical simulation (time and accuracy) of the fire process,which determines both mass flow and energy transfer over anentire high-rise building using a standard p
5、ersonal computer.This paper presents the implementation of a hybrid modelto simulate fires in different building geometries. The hybridmodel combined two independent models: a zone model and anetwork model. The solution procedure consisted of two parts:simulation of two-zone model, which dealt with
6、the room offire-origin and neighboring rooms, and simulation of thenetwork model, which included rooms far away from the fire.The two-zone and network models were first tested individu-ally; then the performance of the integrated model was inves-tigated in different types of applications.The two-zon
7、e model was used to simulate a fire in a roomof a simple two-story, four-room building. A comparison wasmade between the two-zone model and CFAST (Jones et al.2005), a well-known two-zone model for fire simulation.The network model is appropriate for rooms that are faraway from the fire origin. A nu
8、mber of tests were performedusing examples with different building geometries. A three-story, twelve-room building was used to simulate temperatureand pressure changes inside compartment rooms. The massflow rate comparison was made between the network modeland CONTAM (Walton 1997), an existing netwo
9、rk model formass flow rate simulation. A comparison between adaptivetime steps and fixed time steps was also included, showingbetter efficiency made by adaptive time steps.Finally, the models were integrated, where the solutions(temperature and mass flow rate) of the two-zone modelbecome input sourc
10、e for the network model. INTRODUCTIONThe studies of predicting smoke movement caused byfires are very important research areas of fire and smokesafety. A hybrid of a zone and network fire model has beendeveloped (Zhu 2009), which simulates smoke and heat move-ment induced by fires in multicompartmen
11、t buildings. A two-zone model was used to simulate fire and smoke movementinside the room where fire originated and the surroundingcompartment rooms. A network model, which predicts bothmass and energy flow, was used to simulate smoke movementfor the remaining compartment rooms that were far away fr
12、omthe fire room.Algorithm for Smoke Modeling in Large, Multi-Compartmented BuildingsImplementation of the Hybrid ModelG. Hadjisophocleous, PhD, PEng A. Kashef, PhD, PEngMember ASHRAE Member ASHRAEX. Zhu D.E. Amundsen, PhDG. Hadjisophocleous is industrial chair on Fire Safety Engineering in the Depar
13、tment of Civil and Environmental Engineering, X. Zhu is amasters student, and D.E. Amundsen is an assistant professor in the School of Mathematics and Statistics, Carleton University, Ottawa,Ontario, Canada. A. Kashef is a senior research officer at the Institute for Research in Construction, Nation
14、al Research Council Canada,Ottawa, Ontario, Canada.LV-11-011 (RP-1328)2011. American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (www.ashrae.org). Published in ASHRAE Transactions, Volume 117, Part 1. For personal use only. Additional reproduction, distribution, or transmi
15、ssion in either print or digital form is not permitted without ASHRAES prior written permission.778 ASHRAE TransactionsThe work presented in this paper is focused on the imple-mentation and simulation of the hybrid model. Compared tothe two-zone model, the network model was, in general, morecomputat
16、ionally demanding because it normally dealt with alarge number of rooms that were remote from the fire origin.In this paper, several techniques were introduced for a moreefficient implementation of the network model. They includeddecoupling of pressure and temperature and adaptive timesteps. Decoupl
17、ing the pressure and temperature equationsaimed at avoiding “stiff equations” due to the widely differingconvergence speeds for the solutions of these variables. Theuse of adaptive time steps not only improved the efficiency ofthe simulation, but also maintained solution accuracy. ANewton-GMRES solv
18、er with Krylov subspace methods (Saad1996; Trefethen and Bau 1997) was used to solve the pressureequations, and DLSODE (Hindmarch 1983; Radhakrishnanand Hindmarch 1993), a well known ODE solver package, wasused to solve the temperature equations.Simulations using different types of buildings wereinc
19、luded in this paper. In the two-zone model, temperature andmass flow rate solutions were used as input source for thenetwork model. The network model produces output of thetemperature and pressure in each compartment room of thebuilding. An example of the implementation of the hybridmodel is present
20、ed in the following sections.IMPLEMENTATION OF THE HYBRID MODELThe implementation of the hybrid model (Zhu 2009) isbased on the assumption that there is a one-way couplingbetween the zone and the network models. The solution proce-dure consisted of two parts: (1) simulation of two-zone model,which p
21、redicts the conditions in the room of fire origin andneighboring rooms, and (2) simulation of the network modelfor rooms away from the fire. The flowchart of the hybridmodel is shown in Figure 1.Decoupling for Network ModelIn the preliminary simulations of the network model itwas observed that conve
22、rgence speed for pressure was muchfaster than for temperature, which made it difficult to choosethe desirable time scale when trying to solve pressure andtemperature simultaneously. In order to address this issue, thepressure and temperature terms were decoupled. As such, ateach time of the simulati
23、on the pressure part was solved first,assuming that the temperature remains unchanged. After thepressure solution was obtained, the program then solved thetemperature part. Given the different time scales, the errorarising from this was negligible and acceptable for thesubstantial gains in computati
24、onal efficiency.Adaptive Time StepSince the fire was assumed localized, the two-zone modelconsisted of a small number of rooms. On the other hand, thenetwork model dealt with a much larger number of rooms. Asa result, this often caused slow convergence and longerrunning time. A number of aspects hav
25、e been considered inorder to improve the efficiency of the network modelanimportant one was the introduction of an adaptive time step.Unlike a fixed time step, in which the time step was given bythe user at the beginning of the simulation, an adaptive timestep allows the program itself to adjust the
26、 time step based onthe performance of the solution. The basic idea behind theadaptivity algorithms is that, at each time step, the programcompares the solution of pressure at the current time step withthe previous time step solution. If the change in the solution isgreater than a certain tolerance,
27、the time step is then decreasedfor a more accurate solution. If the change of the solution issmaller than a certain tolerance, then the time step is increasedto speed up the simulation time without sacrificing too muchits accuracy. By associating the size of the time step to a certaintolerance, the
28、adaptivity improved the efficiency of the simu-lation while maintaining the solution accuracy.Newtons Method with GMRESBased on the fact that the associated matrix for the pres-sure equations is sparse, a Newton-GMRES solver withKrylov subspace methods (Saad 1996; Trefethen and Bau1997) was used for
29、 solving the pressure equations.Figure 1 Flowchart of the hybrid model.2011 ASHRAE 779DLSODE Solver PackageAn ODE solver package, called DLSODE (Hindmarch1983; Radhakrishnan and Hindmarch 1993) was then used tosolve the system of nonlinear differential equations for thetemperature.Limitations of Imp
30、lementation:Cannot handle isolated rooms in the input buildingNeeds sufficient connection to ambientInner room openings need to be sufficiently largeNeglects floor thickness in calculation of vertical flowOne-way coupling between the two-zone and the net-work modelsNote that each of these limitation
31、s can be addressed, butfor the purpose of the present implementation, the focus wason the primary aspects of the hybrid model.TWO-ZONE MODEL IMPLEMENTATIONA simple two-story, four-room building was used in thesimulation of the two-zone model. The fire was located inroom 1; three windows/doors connec
32、t rooms 1, 2, 3, and ambi-ent, and two ceiling vents connect rooms 3, 4, and ambient.Table 1 shows the dimensions of the four rooms. Figure 2shows the front view and the three-dimensional view of thebuilding with fire origin. The purpose of this simulation is to compare the two-zonemodel with CFAST.
33、 The size of the fire for the two-zonemodel, with maximum heat release rate of 1 MW (948 Btu/s),was calculated using the quadratic equationwhere HRR is the heat release rate, is a constant (in this case = 42.19 kW/s240 Btu/s2), and t is time in s. The simulationwas conducted for 300 s.Table 2 shows
34、the upper-layer temperature comparisonsfor the two models at the end of the simulation (at 300 s). Itindicates that the two models have the same pattern of temper-ature changes, i.e., temperature in the fire room rose the quick-est, while temperature in room 4 at the lower floor remainedunchanged be
35、cause hot air only went up. CFAST predicts ahigher upper-layer temperature for the room of fire origin thanthe developed two zone model. The temperatures in the otherrooms are comparable. It should be noted that due to the differ-Table 1. Rooms Dimensions of the Two-Story, Four-Room BuildingRoom #Wi
36、dth,m (ft)Depth,m (ft)Ceiling Height,m (ft)Floor Elevation,m (ft)1 7.8 (25.5) 8.9 (29.2) 3.3 (10.8) 3.3 (10.8)2 2.3 (7.5) 1.1 (3.6) 3.3 (10.8) 3.3 (10.8)3 5.0 (16.4) 2.5 (8.1) 3.3 (10.8) 3.3 (10.8)4 5.0 (16.4) 2.5 (8.1) 3.3 (10.8) 0.0 (0.0)HRR t2=(b)Figure 2 Geometry of two-story, four-room building
37、 in (a)two-dimensional and (b) three-dimensionalviews.(a)780 ASHRAE Transactionsence of treating conduction and radiation rates in the twomodels, defining and predicting exactly the same fires for thetwo models is not trivial, giving rise to temperature discrep-ancies. NETWORK MODEL IMPLEMENTATIONA
38、12-room building was designed for simulation of thenetwork model. This building consisted of three floors, andeach floor has four rooms. A total of 22 vents were opened onexternal and internal walls; 2 ceiling vents, which were locatedin room 4 and room 8, connect all three floors together. Table 3a
39、nd Figure 3 show the geometry of the building.A non-isothermal simulation was performed to test thenetwork model. In this case, all rooms in the building startedwith ambient pressure and temperature; a constant masssource of 1.5 kg/s (3.3 lb/s) with a temperature of 77C(171F) was placed into room 11
40、. The simulation wasTable 2. Temperature Comparison between Two-Zone Model and CFASTRoomTemperature, C (F)Two-Zone CFAST1 258 (496) 299 (570)2 179 (354) 208 (406)3 92 (198) 91 (196)4 20 (68) 20 (68)Table 3. Room Dimensions of the 12-Room, 3-Story BuildingRoomNo.Width,m (ft)Depth,m (ft)Ceiling Height
41、,m (ft)Floor Elevation,m (ft)1 6 (20) 5 (16) 3 (10) 6 (20)2 6 (20) 5 (16) 3 (10) 6 (20)3 12 (39) 5 (16) 3 (10) 6 (20)4 12 (39) 3 (10) 3 (10) 6 (20)5 6 (20) 5 (16) 3 (10) 3 (10)6 6 (20) 5 (16) 3 (10) 3 (10)7 12 (39) 5 (16) 3 (10) 3 (10)8 12 (39) 3 (10) 3 (10) 3 (10)9 6 (20) 5 (16) 3 (10) 0 (0)10 6 (2
42、0) 5 (16) 3 (10) 0 (0)11 12 (39) 5 (16) 3 (10) 0 (0)12 12 (39) 3 (10) 3 (10) 0 (0)Figure 3 Geometry of three-story, twelve-room building in(a) two-dimensional and (b) three-dimensionalviews.(b)(a)2011 ASHRAE 781in room 11 rose first and reached 77C (171F) in the shortesttime because the mass source
43、with high temperature waslocated in this room. While room 4, being the farthest from thesource, the temperature inside the room started to raise theslowest. On the other hand, pressure in each room changedrapidly at the beginning of the simulation based on the amountof mass each room obtained, but t
44、hen remains firmly constant.In Figure 4, the delta pressure is defined as the pressure differ-ence relative to an ambient reference pressure.In order to test the accuracy of results of the networkmodel, a mass flow rate comparison between the networkmodel and CONTAM is listed in Table 4, correspondi
45、ng to anisothermal simulation of the 12-room building shown inFigure 3. Both models assumed a constant mass source of1.5kg/s (3.3 lb/s) with ambient temperature into room 11 atthe ground floor. Shown in Table 4, both models indicated thatthe majority of the mass entering room 11 at first floor wenti
46、nto ambient through rooms 9 and 10, and the results matchedquite well for mass flow rates at the first floor. At the secondand third floors, there were differences in mass flow ratesbetween the two models. These small discrepancies wereacceptable, and could be caused by rounding errors or thediffere
47、nt tolerance used in the two models.As mentioned earlier in the paper, for buildings thatcontain a large number of rooms, adaptive time steps can beintroduced to improve efficiency of the network model. Anexample of a 1-story, 61-room building case is used to demon-strate the improved efficiency of
48、adaptive time steps over fixedtime steps. Figure 5 shows the layout of the building. Using the same initial time step (0.001 s) and the samesimulation time (1000 s), a constant mass source of 1.5 kg/s(3.3 lb/s) with high temperature 77C (171F) was placed intoroom 61. Computer running time showed tha
49、t adaptive timestep performed 11 times faster than fixed time step. The massflow rates and temperatures values shown in Tables 5 and 6demonstrate the better efficiency of the adaptive time stepprocedure over a fixed time step while maintaining the samesolution accuracy.INTEGRATION OF TWO-ZONE AND NETWORK MODELSThis example is used to evaluate the performance of theintegrated two-zone and network models. The results obtainedfrom the two-story, four-room building (Figure 2) exampleusing the two-zone model was used as input for the networkmodel. Upper layer temperatures and
