response surface models for cfd predictions of air diffusion performance index in a displacement...

Upload: basil-oguaka

Post on 05-Nov-2015

220 views

Category:

Documents


0 download

DESCRIPTION

Response surface models for CFD predictions of air diffusion performance index in a displacement ventilated office

TRANSCRIPT

  • Fp

    ga

    r, Lo

    r D

    . 7,

    olog

    d fo

    me

    Energy and Buildings xxx (200

    + Models

    ENB-2319; No of Pages 8Performance Index (ADPI) in a displacement-ventilated office is presented. By adopting the technique of Computational Fluid Dynamics (CFD),

    the new ADPI models developed are used to investigate the effect of simultaneous variation of three design variables in a displacement ventilation

    case, i.e. location of the displacement diffuser (Ldd), supply temperature (T) and exhaust position (Lex) on the comfort parameter ADPI. The RSM

    analyses are carried out with the aid of a statistical software package MINITAB. In the current study, the separate effect of individual design

    variable as well as the second-order interactions between these variables, are investigated. Based on the variance analyses of both the first- and

    second-order RSM models, the most influential design variable is the supply temperature. In addition, it is found that the interactions of supply

    temperature with other design variables are insignificant, as deduced from the second-order RSM model. The optimised ADPI value is

    subsequently obtained from the model equations.

    # 2007 Elsevier B.V. All rights reserved.

    Keywords: Response Surface Methodology (RSM); Computational Fluid Dynamics (CFD); Air Diffusion Performance Index (ADPI); Thermal comfort; Air

    ventilation

    1. Introduction

    The cooling of occupied spaces, which is generally

    accomplished by mechanical ventilation, consumes a huge

    amount of non-renewable fossil energy in the world that leads

    to the pollution of atmospheric environment. Therefore, in

    order to minimise the energy usage while enabling good

    thermal comfort condition to be achieved, effective distribution

    of fresh air within an occupied space is of practical importance.

    For a long time, the heating, ventilating and air conditioning

    (HVAC) engineers and researchers have been realising that in

    order to optimise the comfort condition in an occupied space,

    efficient quantitative models that establish the relationship

    between a large group of independent parameters (design

    variables) and output variable (response) are highly desirable.

    This can be accomplished by both the experimental and

    numerical approaches.

    In order to study the relationship between the response and

    independent design variables, a large number of experiments

    are undoubtedly required. This has reflected on the increased

    total cost of the study, which is particularly true in the case of

    employing physical experimentations. Therefore, numerical

    experiments such as those accomplished by CFD have been

    gaining immense popularity within the HVAC industry since

    the past few decades. Despite the fact that it is not totally free

    from errors, it serves as a practical design tool for building

    engineers nowadays. For example, by using pure numerical

    approach, Haghighat et al. [1] have investigated the relationship

    between the concentration level in a partitioned room and

    various positions of door, supply and exhaust. Lee and Awbi [2]

    have studied the effect of partition on ventilation effectiveness

    due to its location and gap underneath. Lim et al. [3] have

    determined the optimum position of an air-conditioned unit for

    achieving good thermal comfort condition (based on the

    * Corresponding author. Tel.: +60 389286227; fax: +60 389212116.

    E-mail addresses: [email protected] (K.C. Ng),

    [email protected] (K. Kadirgama), [email protected] (E.Y.K. Ng).1 Tel.: +60 389287255, fax: +60 389212116.2 Tel.: +65 67904455, fax: +65 67911859.

    0378-7788/$ see front matter # 2007 Elsevier B.V. All rights reserved.doi:10.1016/j.enbuild.2007.04.024Response surface models for C

    performance index in a dis

    K.C. Ng a,*, K. Kadira Department of Research & Applications, O.Y.L. R&D Cente

    Sungai Buloh, Selangob Department of Mechanical Engineering, Universiti Tenaga Nasional, Km

    c School of Mechanical and Aerospace Engineering, Nanyang Techn

    Received 29 January 2007; received in revise

    Abstract

    Based on the Response Surface Methodology (RSM), the developPlease cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016D predictions of air diffusion

    lacement ventilated office

    ma b,1, E.Y.K. Ng c,2

    t 4739, Jalan BRP 8/2, Taman Bukit Rahman Putra, 47000,

    arul Ehsan, Malaysia

    Jalan Kajang-Puchong, 43009 Kajang, Selangor Darul Ehsan, Malaysia

    ical University, 50 Nanyang Avenue, Singapore 639798, Singapore

    rm 22 March 2007; accepted 13 April 2007

    nt of first- and second-order models for predicting the Air Diffusion

    www.elsevier.com/locate/enbuild

    7) xxxxxxodels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

  • Predicted Mean Vote) in a typical studio-type apartment. Very

    recently, by employing the commercial CFD code (FLUENT),

    Ooi et al. [4] have studied the temperature and velocity

    distributions in an air-conditioned room for various positions of

    the air conditioner blower. Based on the results of three blower

    positions simulated, the best position is then selected for

    maximum comfort of an occupant. In addition, some

    recommendations have been given by Bojic et al. [5], based

    on the pure CFD analyses (FLOVENT), the optimum

    placement of a window-type air conditioner in a residential

    bedroom in order to achieve minimum draft subjected to the

    calculation of Air Diffusion Performance Index (ADPI). It can

    be noted in general, however, most of the numerical studies

    focus on the one-factor-at-a-time design, without having any

    idea on the behaviour of response variable when two or more

    design variables are varied at the same time. The current paper

    intends to consider this particular issue that involves making

    design decision based on several design variables, which is

    practically desirable.

    In order to demonstrate the method, the authors have

    considered the effect of simultaneous variations of three design

    variables in a displacement-ventilated office (refer to Fig. 1),

    i.e. location of the displacement diffuser (Ldd), supply

    temperature (T) and exhaust position (Lex) on the behaviour

    of response variable (ADPI). The case considered here is taken

    from He et al. [6], in which detailed numerical and

    experimental studies have been performed to investigate the

    efficiency of contaminant removal for several ventilation

    systems. Here, the CFD model developed is firstly validated

    with the experimental data provided by He et al. [6], prior to

    ion

    be

    K.C. Ng et al. / Energy and Buildings xxx (2007) xxxxxx2

    + Models

    ENB-2319; No of Pages 8Fig. 1. Configuration of the mockup office equipped with a displacement ventilat

    4A. Exact dimensions and locations of the obstacles and measurement points canorigin O, (a) isometric view and (b) top view.

    Please cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016system investigated by He et al. [6]. The measurement points are 1A, 2A, 3A and

    found in He et al. [6]. The design variables (Ldd and Lex) are measured from theodels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

  • Bui

    + Models

    ENB-2319; No of Pages 8running the response surface analyses. It has been recently

    noted by Abou-El-Hossein et al. [7] that RSM is one of the

    statistical techniques that saves cost and time in conducting

    experiments by reducing the total number of required tests.

    Furthermore, RSM helps to identify, with great accuracy, the

    effect of the interactions of different design variables on the

    response when they are varied simultaneously. In spite of this,

    within the community of building engineering, only a few

    research works based on RSM are reported, such as those by

    Klemm et al. [8] for multicriteria optimisation and Valencia

    et al. [9] for model development to predict asphalt pavement

    properties.

    In this study, based on RSM, the first- and second-order

    models for the comfort parameter ADPI are developed for a

    displacement ventilation case illustrated in Fig. 1. Based on the

    received RSM models from the CFD simulation, the most

    influential design variable is determined and the corresponding

    interactions between the design variables are subsequently

    identified. Also, based on the model equations obtained, one

    can easily identify the optimum design combination to achieve

    good thermal comfort condition based on the ADPI value,

    which is frequently used as a reference value for indoor airflow

    studies [10].

    2. Introduction to response surface methodology

    Response Surface Methodology (RSM) is a collection of

    mathematical and statistical techniques for empirical model

    building. By careful design of experiments, the objective is to

    optimise a response variable (output variable), which is

    influenced by several independent design variables (input

    variables). An experiment is a series of tests, called runs, in

    which changes are made in the input variables in order to

    identify the reasons for changes in the output response.

    Originally, RSM has been developed to model experimental

    responses and then migrated into the modelling of numerical

    experiments. The difference is in the type of error generated by

    the response. In physical experiments, inaccuracy can be due to

    measurement errors whereas in numerical experiments, errors

    may due to incomplete convergence of the iterative process,

    round-off errors and the discrete representation of continuous

    physical phenomena. In RSM, the errors are assumed to be

    random.

    RSM is a methodology of constructing approximations of

    the system behavior using results of the response analyses

    calculated at a series of points in the design variable space.

    Optimisation of RSM can be solved in the following three

    stages:

    Design of experiment. Building the response surface model. Solution of minimization/maximisation problem accordingto the criterion selected.

    The concept of a response surface involves a dependent

    variable y called the response variable and several independent

    K.C. Ng et al. / Energy anddesign variables x1, x2, . . ., xk. If all of these variables are

    Please cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016assumed to be measurable, the response surface can be

    expressed mathematically as:

    y f x1; x2; . . . ; xk (1)For practical design purpose, the goal is to optimise the

    response variable y, subjected to certain combination of design

    variables. In what next, the ADPI model in the form of Eq. (1)

    will be expressed.

    3. Model of air diffusion performance index

    Draft is a frequent concern when designing indoor

    environments [11]. In order to account for the presence of

    draft, which is defined as any localised feeling of coolness or

    warmth of any position of the body due to both air movement

    and air temperature, the ADPI parameter is used in the current

    study. ADPI presents the percentage of locations where values

    are taken that meet specifications for effective draft temperature

    (1.7 K < u< 1.1 K) and air speed (WS < 0.35 m/s). If ADPIreaches its maximum value, i.e. 100%, the most desirable

    condition is thereby achieved [5]. The effective draft

    temperature is expressed as:

    u Tx Tc aWS b (2)where Tx is the local dry bulb temperature for air (8C), Tc theaveraged room dry bulb temperature (8C) and WS is the airspeed (m/s). The constants a and b are taken as 8 K s/m and

    0.15 m/s, respectively.

    With reference to RSM, where the response variable is ADPI

    in the current study, the relationship between the investigated

    three design variables and the response variable can be

    represented by the linear Eq. (3):

    y1 b0x0 b1x1 b2x2 b3x3 (3)Here, y(1) is the first-order prediction model for ADPI and b is

    the model parameter. x0 is dummy variable (x0 = 1) and b0 is anarbitrary constant. The design variables such as x1, x2 and x3 are

    the location of the displacement diffuser (Ldd), supply tem-

    perature (T) and exhaust position (Lex), respectively.

    In most of the practical cases, the response surface

    demonstrates some curvature effects in most ranges of the

    design variables. Therefore, it would be more useful for a

    designer to consider the second-order model. The practical

    importance of second-order model is to help one to understand

    the second-order effect of each design variable separately and

    the two-way interaction amongst these design variables. This

    second-order model y(2) can be represented by Eq. (4), in

    general, for three design variables:

    y2 b0x0 b1x1 b2x2 b3x3 b11x21 b22x22 b33x23 b12x1x2 b13x1x3 b23x2x3 (4)

    Here the two indices (subscripts) of variable b represent the

    interaction between the corresponding variables. For example,

    b12 represents the significance of interaction between design

    ldings xxx (2007) xxxxxx 3variable 1 and 2.

    odels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

  • 4. Research methodology

    4.1. CFD simulation

    The ADPI models, as discussed in the previous section, are

    determined numerically in the current work. The CFD package

    used has been constantly verified with the available experi-

    mental measurement and reference solution (see [12,13]). Here,

    prior to performing the sensitivity study based on RSM, the

    flow model is validated with the experimental data given by He

    et al. [6]. The ADPI values for various design combinations

    (obtained from the BoxBehnken method to be discussed later)

    are then determined by the validated CFD model.

    The flow solver is based on the finite-volume formulation

    on structured meshes using the cell-centered approach. It uses

    a non-staggered variable storage technique, which is more

    robust as compared to the traditional staggered arrangement

    [14]. Therefore, in order to avoid the pressure oscillations

    steady when the percentage difference of the successive

    change between the variables at current and previous time

    steps is less than 0.01%.

    The configuration of the displacement ventilation flow case

    has been illustrated in Fig. 1. The design variables in this

    particular design combination (based on that of [6]) are:

    Ldd = 1.5550 m, T = 15.9 8C and Lex = 2.33 m. Figs. 2 and 3compare the predicted speed and temperature profiles with the

    available experimental data at four locations (see 1A, 2A, 3A

    and 4A in Fig. 1). In general, the predicted speed and

    temperature variations match the measurements and, by

    considering the coarseness of the mesh system employed

    (30 25 16), the agreements can be considered satisfacto-rily. The discrepancies between the predicted and measured

    speed profiles may due to, partly, the low speed values

    associated in most of the space in which the hot-sphere

    anemometers may fail to give accurate results [6]. For the

    temperature profiles, all the predictions follow the similar

    in a

    K.C. Ng et al. / Energy and Buildings xxx (2007) xxxxxx4

    + Models

    ENB-2319; No of Pages 8arisen due to the non-staggered arrangement, the pressure

    interpolation technique similar to the one proposed by Rhie

    and Chow [15] is adopted here. The issue of pressurevelocity

    decoupling associated with the current incompressible flow

    equations is resolved via the SIMPLE algorithm of Patankar

    [16]; more recent details of SIMPLE algorithm can be found

    in Jasak [17]. The Bi-Conjugate Gradient (Bi-CGSTAB)

    method proposed by Van der Vorst [18] has been used to solve

    the sparse matrix system arisen from the discretised flow

    equations. In the current study, the first-order upwind

    differencing scheme for convective discretisation is adopted

    for robustness purpose. This is acceptable in the current

    context due to the fact that trend analysis deduced from the

    simulation results of various designs is more important here.

    In order to model the flow turbulence, the RNG ke equationsare adopted. Buoyancy is modelled via the Boussinesq

    approximation. In order to promote numerical stability of the

    buoyant flow simulation, a transient approach has been used

    with a time step size of 0.1 s. The results are assumed to be

    Fig. 2. Comparison of speed profiles on mesh 30 25 16 at four locations

    prediction.

    Please cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016trends of those measured. Here, the predicted and measured

    temperature profiles have shown clear stratification associated

    with the displacement ventilation system.

    4.2. Experimental design for RSM

    With the validation of the current CFD model, the ADPI

    models, i.e. Eqs. (3) and (4), are now readily to be determined.

    The model parameter b is calculated from the least square

    method, in which the calculation is performed by adopting the

    commercial statistical software, MINITAB. In order to reduce

    the total number of numerical tests and allow simultaneous

    variation of the three independent design variables, the

    numerical procedure has to be well designed.

    In the current study, the BoxBehnken design method,

    which is based on the combination of the factorial with

    incomplete block design, has been adopted. The attractive part

    of this method is that it does not require a large number of tests

    as it considers only three levels (lowest 1, middle 0 and

    displacement ventilated room. H = 2.26 m. ^: Experiment [6], : currentodels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

  • ns i

    erim

    K.C. Ng et al. / Energy and Buildings xxx (2007) xxxxxx 5

    + Models

    ENB-2319; No of Pages 8Fig. 3. Comparison of temperature profiles on mesh 30 26 16 at four locatiosupply temperature (15.9 8C), Te is the exhaust temperature (24.8 8C). ^: Exp

    Table 1

    Levels of design variables

    Design variable Coding of levels

    1 (lowest) 0 (middle) 1 (highest)Location of the displacement 0.700 1.905 3.110highest 1) of each design variable. The maximum and

    minimum levels (constraints) of each design variable are

    normally determined based on the recommendations given by

    the manufacturer as well as users preferences. The levels of the

    three design variables are given in Table 1. The BoxBehnken

    design is normally used for non-sequential experimentation,

    where a test is conducted only once, which in turn allows

    efficient evaluation of the model parameters in the first- and

    diffuser, Ldd [m]

    Supply temperature, T [8C] 13 16 19Exhaust position, Lex [m] 0.00 2.36 4.72

    Table 2

    CFD simulation conditions according to BoxBehnken design and the predicted A

    Test number Location of the displacement

    diffuser, Ldd [m]

    Supply temperature,

    T [8C]

    1 3.110 16

    2 3.110 16

    3 1.905 13

    4 0.700 19

    5 1.905 19

    6 1.905 13

    7 3.110 13

    8 0.700 16

    9 3.110 19

    10 0.700 16

    11 0.700 13

    12 1.905 16

    13 1.905 16

    14 1.905 16

    15 1.905 19

    Please cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016second-order models. By using MINITAB, the simulation

    conditions of 15 tests are generated, as shown in Table 2. Based

    on these testing conditions, the comfort parameter (response

    variable), ADPI is then computed from the in-house CFD

    package as described earlier. The CFD-predicted ADPI values

    are plotted in Fig. 4 for different test numbers, on top of those

    predictions based on RSM, which will be discussed in the next

    section.

    5. Results and discussions

    5.1. Development of first-order ADPI model

    After performing the 15 numerical tests using CFD, the

    ADPI simulated is used to find the model parameters appearing

    in the postulated first-order model (see Eq. (3)). In order to

    n a displacement ventilated room. H = 2.26 m. T* = (T Ts)/(Te Ts). Ts is theent [6], : current prediction.perform the calculation of these parameters, the least square

    method is used with the aid of MINITAB. The first-order linear

    DPI models based on CFD and RSM

    Exhaust position,

    Lex [m]

    ADPI (%)

    CFD 1st-order RSM 2nd-order RSM

    4.720 36.23 34.27 35.42

    0.000 35.70 33.72 35.14

    4.720 22.87 24.99 23.33

    2.360 40.96 43.13 40.62

    4.720 42.21 43.45 42.00

    0.000 22.77 24.44 22.99

    2.360 21.92 24.76 22.26

    4.720 35.02 34.17 35.58

    2.360 42.59 43.23 43.62

    0.000 33.97 33.62 34.78

    2.360 26.08 24.66 25.06

    2.360 35.04 33.95 35.71

    2.360 35.04 33.95 35.71

    2.360 35.04 33.95 35.71

    0.000 41.72 42.90 41.25

    odels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

  • locations of the displacement diffuser (Ldd) and the exhaust

    (Lex) do not contribute much to the variation of the ADPI. In

    general, the increase of all the design variables will cause

    the ADPI to become larger, which is desirable from the

    design point of view. Chung and Lee [10] have performed a

    similar trend analysis on ADPI based on different values of

    inlet air temperature. It is worth to mention here that, as

    illustrated in Fig. 5, the current predicted model agree

    K.C. Ng et al. / Energy and Buildings xxx (2007) xxxxxx6

    + Models

    ENB-2319; No of Pages 8Fig. 4. Comparison of ADPI models against CFD predictions.model for predicting the ADPI can be expressed as:

    ADPI1 15:6377 0:0241Ldd 3:0770T 0:1153Lex(5)

    From this linear expression, by examining the values of

    the coefficients, one can easily deduce that the response

    variable ADPI is significantly affected by the supply

    temperature (T). Also, it is interesting to note that the

    Fig. 5. Comparison of ADPI model based on supply temperatures. For RSM,

    Ldd is 3.11 m and Lex is 4.72 m.

    Table 3

    Analysis of variance (ANOVA) for first-order equation (from MINITAB)

    Source of variation Degree of freedom (d.f.) Sum of sq

    Regression 3 682.312

    Linear 3 682.312

    Residual error 11 45.991

    Lack-of-fit 9 43.324

    Pure error 2 2.667

    Total 14 728.303

    Please cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016qualitatively well with that of Chung and Lee [10], in which

    the ADPI values increase as the supply temperature

    increases.

    As seen from Fig. 4, the predicted ADPI values obtained

    from the first-order model agree well with the CFD values. The

    adequacy of the first-order model is verified by using the

    analysis of variance (ANOVA). At a level of confidence of 95%,

    the model is checked for its adequacy. As shown in Table 3, the

    P-value of 0.236 (> 0.05) is not significant with the lack-of fitand F-ratio is 3.61. This implies that the model can fit and it is

    adequate [19].

    5.2. Development of second-order ADPI model

    Here, the second-order model is formulated to describe the

    effect of the three design variables investigated on the ADPI,

    given by MINITAB:

    ADPI2 82:2641 6:2873Ldd 12:3392T 0:3862Lex 0:0074L2dd 0:3143T2 0:0875L2ex 0:4004LddT 0:0452LddLex 0:0143TLex (6)

    Similar to the first-order model, by examining the

    coefficients of the first-order terms, the supply temperature

    (T) has the most dominant effect on the ADPI. The contribution

    of exhaust location (Lex) is the least significant here. Also,

    owing to the P-value of interaction is 0.248 (>0.05), one caneasily deduce that the interactions of distinct design variables

    are not significant here. In other words, the most dominant

    design variable T has minimum interaction with others in the

    current context.

    As seen from Fig. 4, the predicted ADPI using the second-

    order RSM model is able to produce values close to those

    computed using CFD, and, as it should be the case, it exhibits

    better agreement as compared to those from the first-order RSM

    model. The ANOVA shown in Table 4 indicates that the model

    is adequate as the P-value of the lack-of-fit is not significant

    (>0.05).

    uares Mean squares F-ratio P-value

    227.440 54.400 0.000

    227.440 54.400 0.000

    4.181

    4.814 3.610 0.236

    1.333odels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

  • Table 4

    Analysis of variance (ANOVA) for second-order equation (from MINITAB)

    Source of variation Degree of freedom (d.f.) Sum of sq

    Regression 9 720.851

    Linear 3 682.312

    Square 3 30.050

    Interaction 3 8.489

    Residual error 5 7.452

    Lack-of-fit 3 4.785

    Pure error 2 2.667

    3

    K.C. Ng et al. / Energy and Bui

    + Models

    ENB-2319; No of Pages 85.3. Design optimisation

    With the ADPI models obtained, the optimised response

    variable (ADPI) can then be determined. Here, the goal is to

    maximise the ADPI from the correct combination of the design

    variables.

    For optimisation purpose, the response variable is trans-

    formed using a specific desirability function shown in Fig. 6.

    The weight defines the shape of the desirability function for the

    response, which can be selected from 0.1 to 10.0 to emphasise

    or de-emphasize the target value (set to 100%). Aweight can be

    Less than one (minimum is 0.1) places less emphasis on thetarget, or

    Equal to one places equal importance on the target and thebounds, or

    Greater than one (maximum is 10.0) places more emphasis onthe target, which is the main concern of the current work.

    Therefore, the weight is set to 10.0.

    From the second-order ADPI model, the optimized ADPI

    value is 43.41% (calculated from MINITAB), subjected to the

    following combination of the design variables:

    Ldd 3:11m; T 19 C; Lex 4:6487m: (7)In order to verify the optimised ADPI value predicted from

    RSM, the CFD simulation is performed again, by adopting the

    combination of design variables shown in Eq. (7). The ADPI

    computed is 42.68% (%difference = 1.71%), and it is worth to

    mention here that it is indeed the highest ADPI value as

    compared to those from the previous 15 CFD tests (see Table 2).

    Apparently, all the design variables are approaching their

    Total 14 728.30maximum values (level = 1) in the case of maximum ADPI

    value is desired. This condition holds true even for the first-

    Fig. 6. Desirability function for maximising the response.

    Please cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016order model, in which the linear model has recommended the

    ceiling values of those design variables in order to achieve the

    most desirable comfort condition, by maximising the ADPI

    value in the current context.

    6. Conclusion

    CFD studies have been applied extensively to the simulation

    of indoor/outdoor airflow. However, most of the numerical tests

    are based on a one-factor-at-a-time design, without having any

    idea about the behaviour of an output parameter (response)

    when two or more design variables are varied simultaneously.

    The current study focuses on the effect of simultaneous

    variations of three design variables in a displacement-ventilated

    office, i.e. location of the displacement diffuser (Ldd), supply

    temperature (T) and exhaust position (Lex) on behaviour of the

    response variable (air diffusion performance index).

    In the current work, the response surface methodology has

    been proven to be a successful technique to perform the trend

    analysis of air diffusion performance index with respect to

    various combinations of three design variables. By using the

    least square method, the first- and second-order models have

    been developed based on the test conditions in accordance with

    the BoxBehnken design method. The models have been found

    to accurately representing the ADPI values with respect to those

    simulated using CFD. The equations have been checked for

    their adequacy with a confidence interval of 95%.

    Both RSM models reveal that the supply temperature is the

    most significant design variable in determining the ADPI

    response as compared to the others. In general, within the

    working range of the supply temperatures considered here,

    ADPI increases as the supply temperature increases. Based on

    uares Mean squares F-ratio P-value

    80.095 53.740 0.000

    227.44 152.610 0.000

    10.017 6.720 0.033

    2.830 1.900 0.248

    1.490

    1.595 1.200 0.485

    1.333

    ldings xxx (2007) xxxxxx 7the second-order RSM model, the supply temperature does not

    interact much with the remaining design variables. Therefore,

    one may exclude both locations of displacement diffuser and

    exhaust for indoor comfort design purpose (based on ADPI) in

    the current design case. With the model equations obtained, a

    designer can subsequently select the best combination of design

    variables for achieving optimum comfort condition.

    Acknowledgements

    The first author would like to express his sincere

    appreciation to his former colleague, Dr. T.K. Lim (now in

    AMD, Singapore) for his recommendation. We also acknowl-

    odels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

  • edge Mr. W.M. Chin (OYL R&D, Malaysia) for showing

    consistent interest and support on the current work. The

    software facilities provided by Universiti Tenaga Nasional

    (UNITEN) are greatly appreciated. Also, special thanks to Mr.

    Anuar (UNITEN), for developing the Graphical User Interface

    of the current CFD package.

    Reference

    [1] F. Haghighat, Z. Jiang, J. Wang, A CFD analysis of ventilation effective-

    ness in a partitioned room, Indoor Air 4 (1991) 606615.

    [2] H. Lee, H.B. Awbi, Effect of partition location on the air and contaminant

    movement in a room, in: Proceedings of Indoor Air 99, vol. 1, Edinburgh,

    August 813, (1999), pp. 349354.

    [3] T.K. Lim, Y.L. Ong, M. Hamdi, Locating the Optimum Position to Place

    an Indoor Air Conditioner in Studio-Type Apartment to Achieve Good

    Thermal Comfort, FLOVENT Technical Paper, Paper V45, total number

    of pages: 7, 2005.

    [4] Y. Ooi, I.A. Bahruddin, P.A.A. Narayana, Airflow analysis in an air

    conditioning room, Building and Environment 42 (3) (2007) 1531

    1537.

    [5] M. Bojic, F. Yik, T.Y. Lo, Locating air-conditioners and furniture inside

    residential flats to obtain good thermal comfort, Energy and Buildings 34

    (2002) 745751.

    [6] G. He, Z. Yang, J. Srebric, Removal of contaminants released from room

    surfaces by displacement and mixing ventilation: modelling and valida-

    tion, Indoor Air 15 (2005) 367380.

    [7] K.A. Abou-El-Hossein, K. Kadirgama, M. Hamdi, K.Y. Benyounis,

    Prediction of cutting force in end-milling operation of modified AISI

    P20 tool steel, Journal of Materials Processing Technology 182 (2007)

    [8] K. Klemm, W. Marks, A.J. Klemm, Multicriteria optimisation of the

    building arrangement with application of numerical simulation, Building

    and Environment 35 (2000) 537544.

    [9] L.E.C. Valencia, A.M. Ramirez, G.L. Barcenas, E.A. Guzman, Modelling

    of the performance of asphalt pavement using response surface method,

    Building and Environment 40 (8) (2005) 11401149.

    [10] K.C. Chung, C.Y. Lee, Predicting air flow and thermal comfort in an

    indoor environment under different air diffusion models, Building and

    Environment 31 (1) (1996) 2126.

    [11] J. Toftum, Air movement-good or bad? Indoor Air 14 (Suppl. 7) (2004)

    4045.

    [12] K.C. Ng, Multigrid solution using high-resolution NVF differencing

    schemes for solution-adaptive unstructured meshes, Ph.D. Thesis,

    Department of Mechanical Engineering, Universiti Tenaga Nasional,

    Malaysia, 2006.

    [13] K.C. Ng, M.Z. Yusoff, E.Y.K. Ng, Higher-order bounded differencing

    schemes for compressible and incompressible flows, International Journal

    for Numerical Methods in Fluids 53 (1) (2007) 5780.

    [14] J. Zhu, On the higher-order bounded discretisation schemes for finite

    volume computations of incompressible flows, Computer Methods in

    Applied Mechanics and Engineering 98 (1992) 345360.

    [15] C.M. Rhie, W.L. Chow, Numerical study of the turbulent flow past an

    airfoil with trailing edge separation, AIAA Journal 21 (1983) 15251532.

    [16] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere,

    Washington, DC, 1981.

    [17] H. Jasak, Error Analysis and Estimation for the Finite Volume Method

    with Applications to Fluid Flow, Ph.D. Thesis, Imperial College, Uni-

    versity of London, UK, 1996.

    [18] H.A. Van der Vorst, BI-CGSTAB: a fast and smoothly converging variant

    of BI-CG for the solution of nonsymmetric linear systems, SIAM Journal

    of Scientific and Statistical Computing 13 (2) (1992) 631644.

    [19] C.R. Hicks, Fundamental Concepts in the Design of Experiments, fourth

    ed., Oxford University Press, USA, 1993.

    K.C. Ng et al. / Energy and Buildings xxx (2007) xxxxxx8

    + Models

    ENB-2319; No of Pages 8241247.Please cite this article in press as: K.C. Ng et al., Response surface m

    displacement ventilated office, Energy & Buildings (2007), doi:10.1016odels for CFD predictions of air diffusion performance index in a

    /j.enbuild.2007.04.024

    Response surface models for CFD predictions of air diffusion performance index in a displacement ventilated officeIntroductionIntroduction to response surface methodologyModel of air diffusion performance indexResearch methodologyCFD simulationExperimental design for RSM

    Results and discussionsDevelopment of first-order ADPI modelDevelopment of second-order ADPI modelDesign optimisation

    ConclusionAcknowledgementsReference