Enhancement of a Modelica Model of a Desiccant Wheel

Enhancement of a Modelica Model of a Desiccant Wheel

Andreas Joos Gerhard Schmitz Wilson Casas Hamburg University of Technology Institute of Thermo-Fluid Dynamics Applied Thermodynamics , 21071 Hamburg, Germany

Abstract
This paper presents a M ODELICA model for a desiccant wheel. Desiccant wheels are used in new concepts for air conditioning systems, which can save primary energy in contrast to conventional systems. This model is based on a model, which was presented at the M ODELICA Conference 2005 [2], however in this study the model is improved with a new modeling approach to represent the wheels rotation. This struc- Figure 1: Example of an air-conditioning using a destural change made the model faster and able to pro- iccant wheel duce continuous output in contrast to the one of Casas et al., [1, 2]. This was an essential step to enhance long term simulations of desiccant systems and control strategies. These simulations are necessary to optimize such systems and to evaluate their primary energy consumption. Keywords: Modelica; Simulation; Desiccant Wheel; Air Conditioning; Sorption

Introduction

In desiccant air conditioning systems, moist air is dehumidied by means of a desiccant wheel, see gure 1. Water vapor is absorbed by desiccant material as humid air passes through the wheel. Using this technology, considerable energy savings can be obtained compared to conventional air conditioning systems. In [1] a model library has been developed to evaluate the performance of the desiccant assisted air conditioning process, so that different congurations and system concepts can be easily realized. Because it is necessary to simulate a period of a year to evaluate an air-conditioning concept, fast, dynamic models with a good accuracy are required. All these requirements argue for M ODELICA as modeling language. The main and most complex component of this library is the
email: email:

Figure 2: Schema of the old modeling approach [2] model of the desiccant wheel. Early approaches for numerical models can be found in [3, 7, 8]. These model formulations have the disadvantage in that they can not handle desiccant materials with discontinuities in their sorption isotherm (e.g. lithium chloride, LiCl). In [2] a M ODELICA model is introduced to overcome those limitations, see gure 2 for an overview. This model is discretized in such a way, that a system of ordinary differential and algebraic equations is generated, which can be easily re-congured for different set-ups. Also new relations for further sorption
Modelica 2008, March 3rd 4th , 2008

[email protected], Tel:+49 40 42878 3079 [email protected], Tel:+49 40 42878 3144 www.tt.tu-harburg.de

The Modelica Association

701

A. Joos, G. Schmitz, W. Casas

isotherms can be provided without much effort. But due to the modeling approach of the rotation movement of the desiccant wheel through the two airows, Air ow I every half revolution time a state event in the numerical solver is caused. Thereby the maximal step size of the solver is restricted. This approach is also only valid, if the change of the boundary conditions of the desiccant wheel is insignicant in half a revolution cycle. Considering the eld of application, this is a rather academic aspect. Although the user should be aware of this fact. Another disadvantage is that the model produces discrete output variables from a continuous process. To overcome these handicaps a new model approach of the coated wheels movement through the airow was developed, implemented and tested. This approach was developed during the work on [4].

Wheel Part I

Air ow II Wheel Part II

Airow I connected to Wheel Part I and Airow II connected to Wheel Part II Computing (t) =
t ti

t = ti

dt tti

M ODELICA Model of Casas et al.

t = ti + 1 T 2

The model of Casas et al. is described in detail in [2]. This section will only give a short overview of the model and highlight the structural criteria that were changed in this work. As shown in gure 2 the implementation in M ODELICA is based on control volumes for air (AirCV) and for the desiccant material (wall_A/B), which can exchange heat and moisture. M ODELICA can only handle ordinary differential equations with respect to time. Therefore the basic idea of the rst approach was a variable transformation to express the position of the rotating wheel with respect to the airows in terms of time instead of angular position. Among the assumptions made in [2], three are elementary in this approach:

OUT = (ti + 1 T ) , 2 Airow I connected to Wheel Part II and Airow II connected to Wheel Part I

Figure 3: Function principle of the model from Casas et al.

to the time t is introduced, which leads to equation (1b).

1 Air,out =

( ,t,BC(t) ) d
0

(1a)

1. The states are not a function of the wheels radius: with , x = f(R) . 2. The variation of boundary conditions during half BC a rotation is negligible: t T 0.
2

t= leads to

T = and dt = d 2

3. The angular velocity of the wheel is constant during half a period: T 0. t

2

2 Air,out = T

t0 + T 2

(t,BC(t) ) dt
t0

(1b)

Equation (1a) gives the average outlet temperature of one airstream. To calculate the integral, the tangential outlet temperature distribution must be known. This leads to a tangential discretization of the wheel and a modeling of the the motion of the discrete pieces through one airow into the other. To display this movement the variable transformation from the angle
The Modelica Association

The advantage of this formulation is, that the wheel has only to be split in two halves; one for each air ow. At the end of half a rotation period, the boundary conditions of the two pieces are switched; the wheel has performed half a revolution. Figure 3 illustrate this behavior.
Modelica 2008, March 3rd 4th , 2008

702

Enhancement of a Modelica Model of a Desiccant Wheel

Another more process engineering oriented point of view is that the continuous process of the turning wheel is represented by batch processes, which each last half a revolution period. Among the application restrictions mentioned, this model has two other drawbacks. First it produces discrete output variables from continuous input values in contrast to the physical process. And second, as will be pointed out in section 4.2 the models computing time is quite large, because it causes every T a state 2 event.

ential velocity (2a), the conservation of mass (2b) and the radial passage area (2c). The area is obtained by dividing the longitudinal half section R L by the axial discretization n. vDF = with m = A v and A= leads to mDF,i = L R2 DF 2n (2d) LR n (2c) (2b) R 2

(2a)

Structure of the new M ODELICA Model

Based on the restriction of applicability and the large computing times mentioned in section 2, a new modeling structure has been developed. The basic idea in this approach is not to perform the variable transformation, but to use equation (1a). To reach this aim, another way to describe the motion of the wheel through the air ows had to be introduced. The construct of the air and desiccant material control volumes is no longer virtually moved through the air ow by switching the air connectors every half rotation period. Instead the control volume are locally xed and a desiccant uid was introduced, which ows through the desiccant CVs in cross ow to air ow direction. Therefore the existing control volumes for air and the desiccant material were used to build a wheel model with a discretization in axial and tangential direction. This modeling approach is sketched in gure 4(a). The black lines on the wheel should hint to the discretization. In contrast to gure 3 the air connectors of the Air CV are attached to the in- and outlet connectors of the wheel model. Also the desiccant material models are connected in series to model the rotation by keeping the desiccant uid in a continuously circulating ow. Figure 4(b) illustrates how the air and the desiccant CVs interact by exchanging heat and moisture for modeling the (de-)humidifying the air by the sorbens. These ows are sketched by the double headed arrows between the two CVs. Air ows along the cylinder axis while the solid passes its CVs in tangential direction. These two streams are indicated in gure 4(b) by the arrows near the two CVs. For these purposes a desiccant uid ow had to be introduced in the desiccant CV and set in relation to the revolution speed. This mass ow is computed by equation (2d), which results of the wheels mean circumferThe Modelica Association

The density DF represents the mass of the carrier material for the sorbent divided by the volume of the wheel, thus including its porosity. This denition was chosen, because the phase equilibrium calculation uses the loading of the carrier material with the sorbens and the loading of the sorbens with water. As mentioned before, the new wheel model was constructed by control volumes for air and desiccant material. Casas et al. used for their model Air CVs with an axial discretization n in ow direction. To include as much of the existing code as possible m of the air CVs are put side by side to get a control volume, which is discretized in two dimensions. Each stream tube in this construct can not directly interact with its neighbors. Listing 1 gives some code snippets to illustrate the implementation in M ODELICA. There are m instances of the model AirCV, which are n times discretized Air CV. Each AirCV has an HeatConnector and HumidityConnector in order to couple it with the Desiccant CV. These connectors are united in the two models Heat1Dto2D and Humidity1Dto2D. The function of these two models is to provide a n m matrix of heat and humidity ows respectively, so that an Air2D model can easily be connected to a two dimensional desiccant material model to form half a desiccant wheel. Figure 5 shows the two Air CVs connected to two Desiccant CVs, which form a closed loop with a circular ow of the desiccant uid. The Desiccant CV is constructed using an analog method. Its model name in the library is SMCV_2D.
Modelica 2008, March 3rd 4th , 2008

703

A. Joos, G. Schmitz, W. Casas

. . .

Air ow I Des. ow Air ow II

(a) Complete Wheel

Air CV Air ow Moisture Flow Heat Flow Desiccant CV

); MeanValues Mean(m=m); Heat1Dto2D HeatMatrix(n=n,m=m); Humidity1Dto2D HumidityMatrix(n=n,m=m ); AirSplit Split(m=m); equation for j in 1:m loop connect(Split.Outlet[j], Air[j] .Inlet); connect(Air[j].Outlet, Mittel.Inlet [j]); connect(Air[j].HeatConnector, HeatMatrix.Heat1D[j]); connect(Air[j].HumidityConnector, HumidityMatrix.Hum1D[j]); end for; . . . end Air2D;

Desiccant ow
(b) Detail scheme of one element from the wheel in g. 4(a)

Figure 4: Scheme of the new modeling approach

Air CV Desiccant CV

It has n stream tubes, which are discretized m times. The mass ow of the virtual uid is computed in both SMCV_2D models by equation (2d). The connectors Desiccant CV between the Desiccant CV only contain the temperature and the water loading of the desiccant uid. The Air CV reason for this is to avoid initialization problems with a circular incompressible ow. The disadvantage of this approach is, that the wheels rotation direction is xed. But this restriction is also true for the real desicFigure 5: Dymola representation of the new modeling cant wheel, which was used for the experimental part. approach Another simplication of this model is that the entrainment of air from one air ow to the other is not modeled. So the simulation of fast rotating wheels will lead to errors. But in their application the wheels circum- 4 Comparison of the two Models ferential speed is small in comparison to the velocity of the air ows. 4.1 Results Listing 1: Excerpt from the Air CV 2D Matrix model
model Air2D parameter Integer n = 1 "Axial Discretization"; parameter Integer m = 1 "Tangential Discretization"; . . . AirCV[m] Air( each n=n,

The model of Casas et al. contains sorption isotherms in the medium model for LiCl, which where validated with measured values. The steady state results of the whole wheel were checked against the manufacturers data and the transient simulations against the model of Rau et al. [5]. These isotherms were also implemented in the desiccant material CV of the new model. To validate the implementation the test model shown in gure 6 was used. This model consists of a desiccant material CV which is connected to an Air CV. The desModelica 2008, March 3rd 4th , 2008

The Modelica Association

704

Enhancement of a Modelica Model of a Desiccant Wheel

Des. CV

Air CV

Figure 6: Test model for the comparison with the Figure 8: Test model for comparisons of the whole isotherm data of Rau et al. wheel model, here with the model form this work
temperature in C 80 This Work 60 40 20 0 Casas et al. Rau et al.

1.2 standardized temperature / water content X 1.15 1.1 1.05 1 0.95 0.9 0.85 0 water content air flow I water content air flow II temperature air flow I temperature air flow II

50

100

150

200 250 time in s Casas et al.

300

350

400

water content x in kg/kg

0.02 This Work 0.015 0.01 0.005 0 0 50 100 150 200 250 time in s 300 350 400 Rau et al.

10 15 20 tangetial Discretization m

25

30

Figure 7: Comparison of a single element of the model Figure 9: Inuence of the tangential discretization to from Casas et al. and the new one the accuracy iccant uid ow is set to zero in this model, so that the results can be compared to the single blow simulations done by Rau et al. [5] and Casas et al. [1, 2]. Figure 7 shows the results of the three models. It is obvious that the sorption isotherms implementation in the new model is equivalent to the model of Casas et al.. In this aspect it is adequate, because the mathematical implementation of those isotherms was already adapted to the use in M ODELICA. For all results, which are discussed below, models were used, which contain an instance of a whole desiccant wheel model and sources and sinks for the air ows. One of those test congurations is presented in gure 8, in this case with the new model. The accuracy of the new model depends on the tangential discretization m. The behavior of the desiccant material CV approaches that of the real wheel as the computational grid is rened, viz. the more disThe Modelica Association

crete control volumes are instantiated in ow direction of the desiccant uid. Figure 9 shows standardized outlet temperatures and water contents X of the two air ow through the desiccant wheel model. They are plotted against the tangential discretization m. The reference value is the corresponding simulation result from a calculation with m = 50. The values were computed from the steady state results from a step response after 5000 s. It can be seen, that at m = 8 the relative error is smaller 2%. Later the consequences on the CPU time will be discussed, but it can be seem from table 1, that depending on the required accuracy m should be chosen as 5 or 8. Figure 10 shows a detail view on the step response of an outlet temperature of three test models, one with the desiccant wheel form Casas et al. and two with wheel models from this work with different numbers of tangential CVs. The complete simulation time was
Modelica 2008, March 3rd 4th , 2008

705

A. Joos, G. Schmitz, W. Casas

49.5 Casas et al. this work m = 8 this work m = 50

70 60 50 Casas et al. this work m = 5 this work m = 8 this work m = 10

49 temperature in C

48.5

CPU time in s
60 70 time in s 80 90 100

40 30 20

48

47.5

10
47 50

0 0

50

100

150

200 250 time in s

300

350

400

Figure 10: Comparison of complete the model from Figure 11: Impact of tangential discretization m on Casas et al. and the new one computing time, compared with the model from [2] 5000 s. The new model produces continuous output variables in respect to the solver step-size in contrast to the discrete output of the old one. The model with the tangential discretization of m = 8 has a slightly larger deviation from the one with m = 50 than the model of Casas et al.. But as will be shown in the next section, it has a remarkable advantage concerning computing times. And compared with measurement errors, the accuracy is sufcient. is high, but due to the large time steps in regions with small gradients, the overall computing time is lower for the new model with a tangential discretization of m = 5,8 and 10 for simulation times larger than 300 seconds in this example. This behavior is expected to lead to a large decrease in computing time, especially at long time simulations of air-conditioning systems. Table 1 gives an overview of the equations, which are created by the different test models, so the number of equations is a little larger than in the stand alone wheel model. The table also shows the computational time of the test model for a simulation time of 5000 s. Due to the above mentioned effect of the state events during the simulations of Casas et al.s model, the new model with m = 5 is nearly 20 times faster even though it consists of about 10 times the number of equations. In this case with the old wheels test model 4166 state events occurred during simulation time.

4.2

Computing Time

The computing times discussed in this section correspond to calculations on one core of an Genuine Intel(R) CPU T2300 @ 1.66 GHz on a laptop with 1 GB RAM. Figure 11 shows a comparison in computing time between the previous approach and the new model. The old model has no tangential discretization, but has to modify the connect statement between the air ows an the wheels control volumes every half period. Whereas the new one needs to be divided in at least ve to eight parts (m in g. 11) per control volume to produce good output values. This leads to the behavior presented in the plot. The old model produces at every half revolution time an event while switching the sides, which wastes computing time while reinitialization of the equation system. This leads to the nearly linear characteristic consisting of numerous small steps. The new model contains a multiple ( m times) of equations compared the the old one, so the computing time for one step is much higher, but due to the model structure time steps larger than half the revolution time are possible. In highly dynamic regions, like the beginning of the plot in gure 11, the computational effort
The Modelica Association

5 Summary and Outlook

Because of the enhancement of the desiccant wheel model from Casas et al. a M ODELICA model could be created, which combines good accuracy with acceptable computing times. It was successfully used in further work of Applied Thermodynamics, like [4, 6], as the heart of a library for desiccant systems. Several simulation of complete climate periods were performed as well as studies concerning different control strategies of those systems. For the analysis of control strategies the models were exported to Matlab/Simulink to nd and optimize control parameters.
Modelica 2008, March 3rd 4th , 2008

706

Enhancement of a Modelica Model of a Desiccant Wheel

Table 1: Number of equations and CPU time of the test model against the tangential discretization m [1] Casas, Wilson: Untersuchung und Optimierung m No. of equations CPU timea in s sorptionsgesttzter Klimatisierungsprozesse. PhD thesis, TU Hamburg-Harburg, 2005. 1 1681 1 2 3043 4 [2] Casas, Wilson, Katrin Proelss, and Gerhard 5 7129 20 Schmitz: Modeling of desiccant assisted air con8 11215 48 ditioning systems. In Proceedings of the 4th Inter10 13939 78 national Modelica Conference, volume 2, pages 15 20749 186 487496. Modelica Association, 2005. 20 27559 289 30 41179 810 [3] Casas, Wilson and Gerhard Schmitz: Nume50 68419 2154 rische Untersuchungen an einer sorptionsgesttzen Klimaanlage. In VDI Fortschrittliche Casas et al. 1184 827 Energiewandlung- und Anwendung, volume 1594 a For a simulation time of 5000 s in Dymola using Dassl of VDI-Berichte, 2001. Especially for this part it was very helpful, that the new model no longer produces discrete output. It was also possible to adapt the model parameters in such a way that data from existing air conditioning systems could be recomputed. [4] Joos, Andreas: Untersuchung und Optimierung eines solargesttzten Heiz- und Klimatisierungssystems fr ein Einfamilienhaus. Masters thesis, TU Hamburg-Harburg, Institut fr Thermouiddynamik, 2006. [5] Rau, J. J., S. A. Klein, and J. W. Mitchell: Characteristics of lithium chloride in rotary heat and mass exchangers. Int. Journal of Heat and Mass Transfer, 34(11):27032713, 1991. [6] Schmitz, Gerhard, Wilson Casas, and Andreas Joos: Entwicklung eines thermisch betriebenen Klimatisierungssystems fr Ein- und Zweifamilienhuser, December 2006. Institute of ThermoFluid Dynamics, Hamburg University of Technology. [7] Simonson, C.J. and Robert W. Besant: Heat and Moisture Transfer in Desiccant Coated Rotary Energy Exchangers: Part I. Numerical Model. HVAC&R Research, 3(4):325340, 1997. [8] Zheng, W. and W.M. Worek: Numerical simulation of combined heat and mass transfer processes in a rotary dehumier. Numerical Heat Transfer, 23:211232, 1993.

References

Nomenclature
Latin Symbols A L m m n R t T v x X Area Length Tangential discretization Mass ow Axial discretization Radius Time Period Velocity Water content Standardized water content

Greek Symbols

BC CV DF i

Temperature Standardized temperature Density Angle Angular Velocity

Abbreviations and Subscripts Boundary condition Control Volume Index for desiccant uid Index for i-th element
707 Modelica 2008, March 3rd 4th , 2008

The Modelica Association