efficient hydraulic and thermal analysis

Upload: mohamed-elshahat-ouda

Post on 07-Jul-2018

228 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    1/29

     begell

    Begell House Inc. Publishers 

     Journal Production50 North Street

    Danbury, CT 06810

    Phone: 1-203-456-6161

    Fax: 1-203-456-6167

    Begell House Production Contact :  [email protected]

    Dear Corresponding Author,

    Attached is the corresponding author pdf file of your article that has been published.

    Please note that the pdf file provided is for your own personal use and is not to be posted on

    any websites or distributed in any manner (electronic or print). Please follow all guidelines

     provided in the copyright agreement that was signed and included with your originalmanuscript files.

    Any questions or concerns pertaining to this matter should be addressed to [email protected]

    Thank you for your contribution to our journal and we look forward to working with you again

    in the future.

    Sincerely,

    Michelle Amoroso

    Michelle Amoroso

    Production Department

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    2/29

     Multiphase Science and Technology, 25 (2–4): 311–338 (2013)

    EFFICIENT HYDRAULIC AND THERMAL

    ANALYSIS OF HEAT SINKS USING VOLUME

    AVERAGING THEORY AND GALERKINMETHODS

     Krsto Sbutega∗ & Ivan Catton

    University of California, Los Angeles, UCLA MAE, Box 951597, 48-121E4, Los Angeles, California 90095-1597, USA

    ∗Address all correspondence to Krsto Sbutega E-mail: [email protected]

     Air- and water-cooled heat sinks are still the most common heat rejection devices in electronics,making their geometric optimization a key issue in thermal management. Because of the complex

     geometry, the use of finite-difference, finite-volume, or finite-element methods for the solution of the governing equations becomes computationally expensive. In this work, volume averaging theory isapplied to a general heat sink with periodic geometry to obtain a physically accurate, but geometri-cally simplified, system model. The governing energy and momentum equations are averaged overa representative elementary volume, and the result is a set of integro-partial differential equations.Closure coefficients are introduced, and their values are obtained from data available in the literature.The result of this process is a system of closed partial differential equations, defined on a simple ge-ometry, which can be solved to obtain average velocities and temperatures in the system. The intrinsicsmoothness of the solution and the simplified geometry allow the use of a modified Fourier–Galerkin Method for efficient solutions to the set of differential equations. Modified Fourier series are chosen asthe basis functions because they satisfy the boundary conditions a priori and lead to a sparse systemof linear equations for the coefficients. The validity of the method is tested by applying it to modelthe hydraulic and thermal behavior of an air-cooled pin-fin and a water-cooled micro-channel heatsink. The convergence was found to be O(N −3.443), while the runtime was ∼0.25 s for N = 56. The

    numerical results were validated against the experimental results, and the agreement was excellentwith an average error of ∼4% and a maximum error of ∼5%.

    KEY WORDS: Galerkin method, volume averaging theory, pin-fin heat sink,micro-channel heat sink 

    1. INTRODUCTION

    The omnipresence of electronic equipment in today’s world, although at times daunting,

    is a fact. Electronic components pervade our entertainment and communication systems,

    existing in their most apparent forms as cellular phones, smart televisions, and comput-ers. However, their various roles in life support, military defense, economic prediction,

    etc., make their reliability a vital concern for our society. Our dependency on electronic

    0276–1459/13/$35.00   c 2013 by Begell House, Inc.   311

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    3/29

    312   Sbutega & Catton

    NOMENCLATURE

    a   arbitrary constant

    A1, A2   non-dimensional aspectratios

    Afs   fluid/solid interface area

    AT    heat input area

    Av   inlet area

    B   boundary operator

    cd   overall drag coefficient

    cdp   pressure drag coefficient

    cf    friction coefficient

    c p   specific heat

    D   pin fin diameter

    dh   volume averagingtheory–defined hydraulic

    diameter

    f f    Fanning friction factor

    f, g   general forcing function

    G   non-dimensional parameters

    H    micro-channel height

    h   heat transfer coefficient

    H c   heat sink height

    H,K,B,I   boundary expansion

    coefficients

    k   thermal conductivityk   effective thermal conductivity

    tensor

    L   heat sink length/linear

    partial differential equation

    operator

    lmfp   mean-free-path length

    l p   pore length scale

    M    non-dimensional parameter

    N    number of basis functions

    n   normal vector

     p   pressure, pitch (micro-channel

    heat sink)

    q    heat flux

    R   residualRh   porosity-weighted thermal

    conductivity ratio

    S,F,P,E   expansion coefficients

    S L   longitudinal pitch (pin-fin

    heat sink)

    S T    transversal pitch (pin-fin

    heat sink)

    S w   area per unit volume in

    representative elementary

    volume

    S wp   representative elementaryvolume frontal area

    T    temperature

    u x-component of velocity

    U avg   average velocity

    V     volume

    v   velocity

    w   fin thickness (micro-channel

    heat sink)

    x   position vector

    y   vector with respect to the

    representative elementaryvolume centroid

    Greek Symbols

    α    thermal diffusivity

     γ,φ   constants inside trial

    functions

    δ   Kronecker delta

    ε   porosity

     ν   dynamic viscosity

    σ   filter function

    ϕ   trial function

     ψ   arbitrary function of interest

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    4/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   313

    NOMENCLATURE (Continued)

    Subscripts

    dh   with respect to length

    scale  dhf fluid phase

    in inlet

    n,  m   coefficient numbers

    s solid phase

    w wall

    Superscripts and Symbols

    disp dispersion

    stag stagnation

    ˆ dimensional quantity

    ∼   fluctuation/deviation fromintrinsic average

      superficial average

    f  intrinsic average

    devices continues to increase exponentially, which also amplifies the potential risks as-

    sociated with the degradation or failure of these devices. Inter-diffusion, corrosion, and

    electro-migration are the leading causes of reliability degradation in electronic compo-

    nents, and each has a thermal-activation component. It is widely understood that reliabil-

    ity increases exponentially with a decrease in temperature (Schmidt, 2004). Furthermore,

    computational speed increases with a decrease in operating temperature. Therefore, an

    improved heat removal system leads to an improvement in both performance and relia-

    bility. However, the contemporary reduction in size of the devices leads to a strong need

    for more efficient heat rejection devices. Because of the ease of fabrication and applica-

    tion, reliability, and low cost (Chu, 2004), air-cooled heat sinks are still the most com-

    mon cooling solutions for electronics. First proposed by Tuckerman and Pease (1981),

    micro-channel heat sinks have gained popularity in the recent decade as an alternative to

    air-cooled heat sinks. Because of their high volume-to-surface area, they seem to be the

    most viable solution for the next generation of compact cooling devices.

    Heat sinks are complex, multi-scale, heterogeneous geometrical structures, which

    make them difficult to analyze. The most commonly used numerical methods, such as

    the finite-difference (FD), finite-element (FE), and finite-volume (FV) methods, require

    domain discretization. The complex and multi-scale nature of the geometry makes such

    discretization challenging and requires a large number of elements. The recent advances

    in computational speed and memory storage now allow for obtaining solutions to the

    governing equations on such large meshes. Large computational fluid dynamics (CFD)

    software packages have now become the standard for analysis of heat sinks (Chein and

    Chen, 2009; Park et al., 2004). Solutions using CFD are less expensive than obtaining

    experimental data, allow great flexibility, and provide a very large amount of informa-

    tion about the flow and heat transfer. Nevertheless, the computational time required to

    evaluate the performance of each heat sink is on the order of hours.The ultimate goal in most applications is to design a heat sink that can dissipate the

    given heat load with the minimum amount of pumping power, while keeping the heat

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    5/29

    314   Sbutega & Catton

    generating device under a certain temperature. This is a multi-parameter optimization

    problem. The description of the complex geometrical structure of heat sinks requires a

    large number of parameters (about 10); therefore, the optimization study would require

    a very large number of system evaluations (on the order of  103). Because of the com-

    putational time required, CFD cannot be used as a tool in such studies. Furthermore, if only overall temperature differences, pressure drop, and heat flux are of interest, it is

    inefficient to first carry out detailed local flow and energy calculations, and then average

    them in post-processing. Thus, it would be highly desirable to trade the detailed local

    information of the solution for a significant decrease in computational time.

    In this study, the volume averaging theory (VAT) is applied to a general heat sink to

    obtain a set of partial differential equations (PDEs) in the average temperatures, fluxes,

    and velocity. The result is an accurate model that can be used to evaluate heat sink per-

    formance very quickly, therefore enabling a multi-parameter optimization study. The

    VAT was developed for analysis of transport phenomena in porous media, which are

    inherently multi-scale and complex geometrically. The VAT is a rigorous mathematical

    tool that spatially smooths the governing conservation equations and produces a set of 

    PDEs that are defined at every point in the heat sink, essentially transforming the com-

    plex structure of the heat sink into a homogenous medium. The basics of the VAT are

    discussed in a short review in the next section. Every averaging process involves the

    loss of some information at the lower scale. This loss of information requires a closure

    scheme to correlate the lower-scale phenomena to the averaged quantities. In this work,

    the schemes developed by Kuwahara et al. (2001) and Travkin and Catton (2001) will

    be used to obtain a set of closed PDEs.

    The application of the VAT leads to a set of PDEs, which is defined in the entire

    domain and whose solution is inherently smooth because of the averaging process. These

    features suggest that a Galerkin method (GM) solution, with a modified Fourier series

    as the basis, can be used to efficiently obtain a solution. The GM is a subset of spectral

    methods. Spectral methods have become increasingly popular recently (especially for

    direct numerical simulations of turbulence) because of better convergence and memory

    management. The most popular spectral method is the Chebyshev collocation method.

    Horvat and Catton (2002, 2003) used a Fourier–GM as a basis in one direction to reduce

    non-dimensionalized VAT equations to a system of ordinary differential equations that

    was solved analytically with a matrix exponential. However, this is possible only for

    constant temperature boundary conditions because in this case the system reduces to an

    eigenvalue problem. In this work, the GM will be used in both directions to solve the

    VAT equation for any temperature or heat flux boundary condition.

    2. THEORETICAL FUNDAMENTALS

    The VAT is a rigorous mathematical tool that allows the study of heat transfer and fluid

    flow in hierarchical, geometrically complex systems. It was developed in the 1960s by

     Multiphase Science and Technology

    https://www.researchgate.net/publication/267584310_Modeling_of_Forced_Convection_in_an_Electronic_Device_Heat_Sink_as_Porous_Media_Flow?el=1_x_8&enrichId=rgreq-49d533c5-df74-4139-bd6e-e9c887efddec&enrichSource=Y292ZXJQYWdlOzI2NDM4Mjc4MDtBUzoyMTEyOTE1NTQ3NTA0NjVAMTQyNzM4NzI0MDM5OA==https://www.researchgate.net/publication/222646713_A_Numerical_Study_of_Interfacial_Convective_Heat_Transfer_Coefficient_in_Two-Energy_Equation_Model_of_Porous_Media?el=1_x_8&enrichId=rgreq-49d533c5-df74-4139-bd6e-e9c887efddec&enrichSource=Y292ZXJQYWdlOzI2NDM4Mjc4MDtBUzoyMTEyOTE1NTQ3NTA0NjVAMTQyNzM4NzI0MDM5OA==

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    6/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   315

    Whitaker (1967, 1969), Gray (1975), and Gray and O’Neill (1976) to deal with the em-

    piricism and lack of rigor involved in modeling of flow through porous media. The fun-

    damentals can be found in Whitaker (1999). In its goal and mathematical approach, VAT

    is analogous to the continuum approach in mechanics. Theoretically, to model transport

    phenomena in any system, Newton’s second law could be applied to each molecule toobtain its motion, and from it any physical quantity of interest. However, this is impos-

    sible computationally because of the prohibitive amount of molecules contained in most

    systems of interest. Thus, after averaging the discrete space over a loosely defined parti-

    cle, a continuum is defined in which quantities of interest can be defined at every point

    of the domain, independently of whether a molecule is present or not at the location

    of interest. Mathematically, the governing equations at the molecular scale, defined by

    the mean-free path,  lmfp, are averaged over an intermediate length scale, defined by the

    length scale of the vague concept of a particle. This process leads to continuum equa-

    tions at the length of scale of interest, i.e., the system length scale,  L. The validity of the

    procedure is dictated by the Knudsen number, Kn  =   lmfp/L. If the Knudsen number issmall, the averaging is statistically valid and the continuum approach can be used. In the

    averaging process, some of the information at the scale of  lmfp  is lost and some closure

    coefficients are required. The effect of lower-scale information on the higher scale is

    incorporated in the closure coefficients, which can include viscosity, thermal conductiv-

    ity, diffusivity, etc. These closure coefficients have to be determined and provided either

    analytically (i.e., kinetic theory of gases), experimentally, or numerically (i.e., molecu-

    lar dynamics). In a very similar manner, the VAT replaces a complex discrete geometric

    structure (e.g., porous medium) with a fictitious continuum. Theoretically, the point-wise

    governing equations can be solved over each domain (e.g., solid and fluid) to obtain any

    of the quantities of interest. However, this can be very computationally expensive or

    impossible because of the complicated geometry. Therefore, the point-wise governing

    equations are averaged and the discrete complex structure is substituted with a fictitious

    medium in which the quantities of interest are defined at every point, independently of whether it is in the solid or fluid phase. Mathematically, the governing equations at the

    pore length scale,  l p (e.g., Navier–Stokes) are averaged over a representative elementary

    volume (REV) to obtain a set of governing equations at the length scale of interest, i.e.,

    system length scale  L. The validity of the approach is, once again, given by the length

    scale disparity,   l p/L. If this number is small, the averaging is statistically valid and theapproach is justified. Again, in the averaging process, some information of phenomena

    at the lower scale (l p) is lost and some closure is required to model the effect of these

    phenomena on the quantities of interest. In our case, these closure coefficients will be

    the REV heat transfer coefficient, thermal conductivity, and friction factor. These closure

    coefficients can be determined analytically, experimentally, or numerically (CFD).

    The derivation of the VAT energy, mass, and momentum conservation equationsstarts from the incompressible, laminar, steady, constant property; point-wise mass; mo-

    mentum; and energy conservation equations. Furthermore, it is assumed that the walls

    Volume 25, Number 2–4, 2013

    https://www.researchgate.net/publication/229599842_Diffusion_and_Dispersion_in_Porous_Media?el=1_x_8&enrichId=rgreq-49d533c5-df74-4139-bd6e-e9c887efddec&enrichSource=Y292ZXJQYWdlOzI2NDM4Mjc4MDtBUzoyMTEyOTE1NTQ3NTA0NjVAMTQyNzM4NzI0MDM5OA==https://www.researchgate.net/publication/239154863_A_Derivation_of_the_Equations_for_Multi-Phase_Transport?el=1_x_8&enrichId=rgreq-49d533c5-df74-4139-bd6e-e9c887efddec&enrichSource=Y292ZXJQYWdlOzI2NDM4Mjc4MDtBUzoyMTEyOTE1NTQ3NTA0NjVAMTQyNzM4NzI0MDM5OA==

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    7/29

    316   Sbutega & Catton

    are impermeable, the effects of gravity are negligible, and no viscous dissipation or heat

    generation is present. With these assumptions, the point-wise governing equations are

    ∇ · v = 0   (1)

    ∇ · (vv) = − 1ρ

    ∇ p + νf ∇2v   (2)

    ∇ · (vT f ) = α f ∇2T f    (3)

    0 = α s∇2T s   (4)

    The appropriate boundary conditions can be divided in the system and interface bound-

    ary conditions. The interface boundary conditions are internal; they determine the energy

    exchange within the system and are applied on surfaces of scale  l p. The system bound-

    ary conditions are applied on the external boundaries, determine energy inputs into the

    system, and are applied to surfaces of scale  L. The interface boundary conditions are

    given by

    v = 0T f   = T s,   on   Afs

    −kf ∂T f 

    ∂ x

    · n  =

    −ks

    ∂T s

    ∂ x

    · n

    (5)

    where Afs  is the fluid/solid interface area inside the system and is usually very hard to

    characterize. The system boundary conditions are usually more dependent on the prob-

    lem, but some examples are

    v =  vin,   on   Av

    T f  =  T w   or

    −kf 

    ∂T f 

    ∂ x

    · n  =  q w   on   AT 

    (6)

    where Av  and AT   are the inlet area and heat input area, respectively. As discussed pre-

    viously, Eqs. (1)–(4) are defined only within their domains. The mass and momentum

    equations are defined only in the fluid domain, while the solid energy equation is de-

    fined in the solid domain. Thus, the challenge in these equations comes mostly from

    the determination of the interfaces in the intricate geometry. However, after the VAT is

    applied, the internal boundary conditions are absorbed by the governing equations and

    the difficulties related to internal boundaries are bypassed.

    The averaging of the governing equations starts by associating to every point  x   a

    REV  V   , of which  x  is the centroid (see Fig. 1). The superficial and intrinsic averaging

    operators are defined, respectively, as

     ψ|x =  1

    V    (x)

     V  f (x)

     ψ (x + yf ) dV     (7)

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    8/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   317

    FIG. 1: Definition of the vectors in REV (Whitaker, 1999).

     ψf 

    x=   1V  f  (x)

     V  f (x)

     ψ (x + yf ) dV     (8)

    where ψ  is a function of interest (tensor of any rank). The two averages are related by

     ψ = V  f 

    V     ψf  = εf   ψ

    f  (9)

    where  εf   is the fluid volume fraction. In order to be able to average Eqs. (1)–(4), a

    relationship between the average of a gradient and the gradient of the average is required.

    An extension of 

    Leibniz’s rule, known as the spatial averaging theorem (SAT), gives the necessary

    relationship:∇ ψ =  ∇  ψ +

      1

    V  

     

    Afs

    n ·  ψdS    (10)

    where  Afs   is the interface area and   n   is the unit vector normal to Afs   pointing from

    the fluid toward the solid. Slattery (1972) gives a detailed derivation of the theorem.

    With averaging operators (7) and (8) and the SAT, it is possible now to move on to the

    development of the VAT equations.

    The superficial averaging operator is applied to Eqs. (1)–(4) and they are averaged

    over the REV to obtain

    ∇ ·  v = 0   (11)

    ∇ · (vv) =  − 1ρ ∇ p +  ν

    ∇2v

      (12)

    ∇ · (vT f ) =  α f 

    ∇2T f 

      (13)

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    9/29

    318   Sbutega & Catton

    0 =  α s

    ∇2T s

      (14)

    The SAT can be applied to Eq. (11), along with the no-penetration boundary condition,

    to obtain

    ∇ · v = 0   (15)

    which shows that the superficial velocity is solenoidal. This is an expression of the VAT

    continuity equation. An alternative form can be found using the relationship given in

    Eq. (9):

    ∇ · vf  = −∇ε

    ε

    vf  (16)

    This equation shows that for constant porosity, the intrinsic velocity is also solenoidal;

    however, this is not the case when the porosity is changing. In the validation process, the

    results will be calculated for a constant porosity heat sink and the homogeneous form of 

    Eq. (16) will be used. Next, the SAT is applied to the pressure term of Eq. (12):

    ∇ p =  ∇  p +  1

    V   Afs

     pndS    (17)

    Applying the SAT to the diffusive terms, they can be rewritten as

    ∇ · ∇v =  ∇ · ∇v +  1

    V  

     

    Afs

    n · ∇vdS    (18)

    ∇ · ∇T i =  ∇ · ∇T i +  1

    V  

     

    Afs

    n · ∇T idS    (19)

    where i =  {f , s}. Using the SAT again, and the no-penetration boundary condition to the

    first term in Eq. (18), the momentum diffusion term reduces to

    ∇ · ∇v =  ∇2 v +  1

    V  

     

    Afs

    n · ∇vdS    (20)

    Similarly, Eq. (19) can be rewritten as

    ∇ · ∇T i =  ∇2 T i + ∇ ·

    1

    V  i

     

    Afs

    nT idS 

    +   1

    V  i

     

    Afs

    n · ∇T idS    (21)

    The averaging process of the convective terms is more elaborated. First, the SAT andno-flow-through conditions are used to reduce the terms to

    ∇ · (vT f ) =  ∇ · vT f    (22)

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    10/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   319

    ∇ · (vv) =  ∇ · vv   (23)

    These equations are still incomplete because the terms are expressed as the average of the

    product while a closed form of the VAT equations requires the product of the averages.

    The two terms are separated by decomposing the function into the average and a spatial

    deviation: ψ =   ψi + ˜ ψ   (24)

    A similar decomposition is performed in time averaging of turbulent Navier–Stokes

    equations. However, in Eq. (24) function  ψ   is averaged spatially and the fluctuations

    are the differences from the spatial average. If the averaged function is smooth enough

    (for a detailed length scale analysis, see Carbonell and Whitaker, 1984), the right-hand

    sides in Eqs. (22) and (23) can be decomposed to obtain

    ∇ · vv =  ∇ ·εf  vf  v

    f  + εf  ṽṽf 

      (25)

    ∇ · vT f  =  ∇ ·εf  v

    T f 

    + εf 

    ṽ˜T f 

      (26)

    Following this analysis and with some manipulation of the equation, a first form of the

    VAT mass, momentum, and solid and fluid energy equations can be written as follows:

    ∇ · vf  = 0   (27)

    εf  vf  · ∇ vf  = −

    εf 

    ρf ∇  pf  + εf  νf ∇

    2 vf  + νf 

    V  

     

    Afs

    n · ∇vdA +  1

    ρf V  

     

    Afs

     pndA

    (28)

    − εf ∇ · ṽṽf 

    εf  vf  · ∇ T f 

    f  = εf α f ∇2 T f 

    f  + α f 

    V  

     

    Afs

    n · ∇T f dS  + ∇ ·

     

    Afs

    T f ndS 

    (29)

    − εf ∇ ·

    ṽT̃ f 

    0 = εsα s∇2 T s

    s + ∇ ·

    α s

    V  

     

    Afs

    nT sdS 

    +  1

    V  

     

    Afs

    n · α s∇T idS    (30)

    The goal is to obtain a set of equations for the average velocity and temperatures. The set

    given above still contains certain point-wise terms in the integrals and some fluctuation

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    11/29

    320   Sbutega & Catton

    terms. Therefore, this is not a closed form of the governing equations because it contains

    information about the lower-scale phenomena that is not known a priori. In order to close

    these equations, it is necessary to develop a scheme that will relate these fluctuation and

    integral terms to averaged quantities. Travkin and Catton (2001) defined the following

    closure coefficients for the VAT momentum equation:

    cd  =  cdpS wp

    S w+ cf  −

    ∇ ·εβ ṽβṽβ

    β

    1

    2ρβ v̄β

    2  (31)

    cdp = 2

     Afs

    n pdA

    ρf  v2 S wp

    (32)

    cf  = 2

     Af 

    nτ wdA

    ρf  v2S w

    (33)

    where S w  is the interface surface per unit volume in the REV:

    S w = Afs

    V    (34)

    When substituted into Eq. (28), a deceivingly simple closed form of the VAT momentum

    equation for constant porosity is obtained

    εf  vf  · ∇ vf  = −

    εf 

    ρf ∇  pf  + εf  νf ∇

    2 vf  − 1

    2cdS w

    vf 

    2(35)

    Similarly, Kuwahara et al. (2001) defined a VAT heat transfer coefficient and an effective

    thermal conductivity as

    h =

    (kf /V   )  Afs n · ∇T f dAS w

    T s

    s − T f f 

      (36)

    ki,eff  · ∇ T ii = ki,stag · ∇ T i

    i + δf iki,disp · ∇ T ii (37)

    where i  =  {s, f }  and the Kronecker delta is used simply to point out that dispersion iszero in the solid phase. Dispersion thermal conductivity is a tensor and it is defined as

    ki,disp · ∇ T ii = −ρic pi

    εi

    T̃ iṽi

    i  (38)

    while the stagnation thermal conductivity is the sum of the diffusive term and the tortu-

    osity

    ki,stag · ∇ T ii = εiki∇ T i

    i +   1V  

     Afs

    nT idA   (39)

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    12/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   321

    As it can be inferred from the name, the stagnation thermal conductivity is defined as

    the effective thermal conductivity that the system would have if the velocity was zero.

    As discussed previously, these closure coefficients are the equivalent of transport coef-

    ficients such as thermal conductivity, viscosity, etc., and need to be determined before

    moving on to the solution of the VAT equations.Using these closure coefficients, Eqs. (29) and (30) can be rewritten as

    εf ρf c pf  vf  · ∇ T f 

    f = ∇

    kf,eff  · ∇ T f 

    f  + hS w

    T s

    s − T f f 

      (40)

    ks,stag · ∇ T ss =  hS w

    T s

    s − T f f 

      (41)

    Now, a full set of closed equations, given by Eqs. (16), (35), (40), and (41) has been

    developed for the averaged quantities of interest, namely vf f , T s

    s, and T f f . A key

    point to emphasize is that these equations, and therefore their solutions, are defined at

    every point in the domain. Therefore, an average velocity and solid and fluid temper-

    atures are assigned to each point, without consideration of whether the point is actuallywithin the solid or fluid phase. The discrete heat sink geometry has now been turned

    into a continuous medium. The determination of the closure coefficients is obviously a

    key component in the accuracy of the solution. Vadnjal (2009) showed that the Fanning

    friction factor, f f , closely approximates the VAT-defined drag coefficient,  cd; therefore,

    the Fanning friction factor will be used in the governing equations. The heat transfer

    coefficient and the friction factor can then be taken from available experimental data for

    common geometries.

    This study analyzes the behavior of a periodic heat sink in two dimensions. The

    extension to a three-dimensional domain is not difficult. However, the implementation

    can be quite tricky and will be more computationally expensive, thus it is not consid-

    ered here. Because of the periodicity of the geometry, the fluctuation terms will die out

    quickly away from the boundaries and as a result can be ignored. Furthermore, because

    of the small local Péclet number and the high thermal conductivity ratio, the tortuosity

    and dispersion effects can be ignored. It is further assumed that the flow is fully devel-

    oped; however, axial conduction is not assumed to be negligible because of the small

    length scale considered. A graphical description of the domain and boundary condition

    is given in Fig. 2. With the given assumptions, the VAT governing equations, Eqs. (35),

    (40), and (41), become

    0 ≤  ẑ ≤  Ĥ c

    −εf ̂ νf ∂ 2 ûf 

    ∂z2  +

     1

    2

    f f  Ŝ w ûf f 2

    = −εf 

    ρ̂f 

    ∂ ˆ pf f 

    ∂ ̂x

    (42)

    0 ≤  x̂ ≤  L̂,   0 ≤  ẑ  ≤  Ĥ c

    Volume 25, Number 2–4, 2013

    https://www.researchgate.net/publication/234368820_Modeling_of_a_heat_sink_and_high_heat_flux_vapor_chamber?el=1_x_8&enrichId=rgreq-49d533c5-df74-4139-bd6e-e9c887efddec&enrichSource=Y292ZXJQYWdlOzI2NDM4Mjc4MDtBUzoyMTEyOTE1NTQ3NTA0NjVAMTQyNzM4NzI0MDM5OA==

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    13/29

    322   Sbutega & Catton

    FIG. 2: Schematic of the geometry and boundary conditions.

    ĉ pf ρ̂f εf ̂uf f ∂ T̂ f 

    f ∂ ̂x

      =  εf k̂f 

    ∂ 2T̂ f 

    f ∂ ̂z2

      + εf k̂f 

    ∂ 2T̂ f 

    f ∂ ̂x2

      + ĥŜ w

    T̂ s

    s−T̂ f 

    f    (43)

    0 ≤  x̂ ≤  L̂,   0 ≤  ẑ  ≤  Ĥ c

    εsk̂s

    ∂ 2T̂ s

    s∂ ̂x2

      + εsk̂s

    ∂ 2T̂ s

    s∂ ̂z2

      = ĥŜ w

    T̂ s

    s−T̂ f 

    f  (44)

    where ˆ denotes dimensional quantities.

    As discussed previously, the internal boundary conditions have been absorbed in the

    equations. Also, the continuity equation [Eq. (16)] was not listed but has been used todetermine that the  z-velocity component is zero. The system boundary conditions for

    the momentum equation are given by the no-slip condition (assuming no by-pass) as

    follows:

    ûf f z=0

    =  ûf f z= Ĥ c

    = 0   (45)

    The energy equation system inputs are given by

    T̂ f 

    f x=0

    =  T̂ in

    ∂ T̂ f f 

    ∂ ̂x

    x=L̂

    =

    ∂ T̂ f f 

    ∂ ̂z

    z= Ĥ c

    = 0

    (46)

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    14/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   323

    ∂ T̂ s

    s

    ∂ ̂x

    x=0

    =∂ T̂ s

    s

    ∂ ̂x

    x=L̂

    =∂ T̂ s

    s

    ∂ ̂z

    z= Ĥ c

    = 0   (47)

    Because there is no heat source or sink at the outlet, it is assumed that any heat transport

    at this location is simply convected away by the fluid, therefore resulting in an adiabaticboundary condition at the outlet. The boundary condition at the bottom of the channel

    can vary depending on whether temperature or heat flux is specified. For the case of 

    specified temperature  T̂ w, the boundary condition is straightforward:

    T̂ f 

    f z=0

    =T̂ s

    sz=0

    =  T̂ w   (48)

    For the case of a heat flux boundary condition, the situation is more complicated. Be-

    cause of a scale discontinuity, the determination of the heat transfer from the base to

    the porous channel is still an active area of research (Imani et al., 2012; Ouyang et al.,

    2013; Yang and Vafai, 2011). In this work, the base is assumed to be thin relative to its

    length, and conjugate effects are ignored. It is also assumed that the input heat flux willbe distributed to the two phases as follows (Imani et al., 2012):

    −εf ̂kf 

    ∂ T̂ f 

    ∂ ̂z

    z=0

    =  −εsk̂s∂ T̂ s

    s

    ∂ ̂z

    z=0

    = q̂ w   (49)

    In the next section, a method will be developed to efficiently solve Eqs. (42)–(44) with

    the boundary conditions given by Eqs. (45)–(49). The result will be a fast running code

    for modeling of thermal and hydraulic behavior of heat sinks.

    3. SOLUTION METHOD

    A solution to the VAT energy equations will be obtained using a GM with a modi-

    fied Fourier series as the basis functions. The GM gets its name from Boris Galerkin

    (Galerkin, 1915), a Russian engineer that developed and used the method to solve dif-

    ferential equations resulting from problems in statics. The semi-analytical nature of the

    method and its rapid convergence made it very popular when computers were not widely

    available (Finlayson, 1972). The GM, in its global formulation, is advantageous for sim-

    ple geometries and smooth solutions. In recent years, the wide availability of computers

    has made the FD and FV methods more popular because of the ease of implementation

    and versatility. Local GMs have been used quite often in FE methods. The application

    of the VAT to the governing equation ensures that the solution is smooth and simplifies

    the geometry, effectively bypassing the shortcomings of the GM.

    The GM is a subset of the larger class of spectral methods (Canuto et al., 2006) anda brief review is given here. A linear PDE can be formulated as

    L (u) = f    on   Ω   (50)

    Volume 25, Number 2–4, 2013

    https://www.researchgate.net/publication/276982720_Estimation_of_heat_flux_bifurcation_at_the_heated_boundary_of_a_porous_medium_using_a_pore-scale_numerical_simulation?el=1_x_8&enrichId=rgreq-49d533c5-df74-4139-bd6e-e9c887efddec&enrichSource=Y292ZXJQYWdlOzI2NDM4Mjc4MDtBUzoyMTEyOTE1NTQ3NTA0NjVAMTQyNzM4NzI0MDM5OA==

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    15/29

    324   Sbutega & Catton

    where L is a linear operator; Ω is the domain of the solution; and f  is a forcing function.The extension to a non-linear equation is very similar but will not be addressed in detail

    here. The boundary conditions can be expressed as

    B (u) = g   on   ∂ Ω   (51)

    Spectral methods assume that the solution can be expanded in a truncated series:

    uN  (x) =N n=0

    anϕn   (52)

    with undetermined coefficients an. The substitution of this assumed solution in Eq. (50)

    will not satisfy the solution exactly, and it will result in a residual R:

    L (uN ) − f  = R   on   Ω   (53)

    The idea behind spectral methods is to multiply both sides of Eq. (53) with weight func-

    tions wm (x). The resulting expression is integrated over the domain, and the coefficients

    are chosen such that they drive the residual to zero Ω

    L (uN )wmdΩ −

     Ω

    fwmdΩ = 0   (54)

    Various spectral methods differ in the choice of the weight functions. Pseudo-spectral

    methods, such as the collocation methods, force the residual to zero at given points in

    the domain. The GM uses trial functions that are equal to the basis functions,  ϕ  =  w.The result is a set of algebraic equations for coefficientsan. The boundary conditions can

    be treated in several ways; however, choosing basis functions that satisfy the boundary

    conditions a priori often gives the best convergence (Finlayson, 1972). The method can

    be expanded to two-dimensional problems by using a tensor product of basis functions:

    uN  (x, y) =N n=0

    N m=0

    anmϕn (x)φm (y)   (55)

    Following the same procedure described for the one-dimensional cases, a set of  N 2

    algebraic equations for coefficients  anm   is obtained. In this work, a modified Fourier

    series will be used as the basis functions for the problem described in Fig. 2 because of 

    its ability to satisfy the boundary conditions a priori.

    It is always good practice to non-dimensionalize the governing equation. The domain

    and the quantities of interest are non-dimensionalized as follows:

    T ii =

    T̂ ii

    −  T̂ in

    T̂ in,   uf 

    f  =  ûf 

    Û avg,   Û avg =

       dh

    2ρf f f 

    ∆ p

    L

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    16/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   325

    x =  x̂

    L̂,

    ...   z  =  ẑ

    Ĥ c, dh  =

      4εf 

    S w(56)

    where i   =  {s, f }, and the superscript ˆ denotes dimensional quantities. By substitutingthese parameters into the governing equations [Eqs. (42)–(44)], the following equations

    are obtained:0 ≤  z  ≤  1

    −M 1∂ 2 uf 

    ∂z2  +

    uf 

    2

    = 1

    (57)

    0 ≤  x  ≤  1,   0 ≤  z  ≤  1

    G1 uf f   ∂ T f 

    ∂x  − G2

    ∂ 2 T f f 

    ∂x2  −

     ∂ 2 T f f 

    ∂z2  − G3

    T s

    s − T f f 

     = 0

    (58)

    0 ≤  x  ≤  1,   0 ≤  z  ≤  1

    G2

    ∂ 2 T ss

    ∂x2   +

     ∂ 2 T ss

    ∂z2   − G3Rh

    T ss

    − T sf 

     = 0

    (59)

    The non-dimensional parameters are defined as

    M 1   =  ˆ νf 

    Ĥ 2c

      ρ̂f L̂d̂h

    2f f ∆ˆ p, G1  = Prf RedhA1A2, G2  = A

    2

    2, G3   = 4NudhA

    2

    1

    A1  =Ĥ c

    d̂h, A2   =

    Ĥ c

    L̂, Rh   =

      εf k̂f 

    εsk̂s

    (60)

    The non-dimensional number M 1  represents the ratio of diffusive effects to friction and

    pressure forces. Parameter  G1   is a Péclet number multiplied by two geometrical non-

    dimensional ratios; A1   is the ratio of the  z-direction diffusion length scale  ˆH c   to the

    convection length scale  d̂h, while A2  is the ratio of  z - and x-direction diffusion length

    scales. The boundary conditions [Eqs. (45)–(49)] can be rewritten in non-dimensional

    form as

    uf f z=0

    =   uf f z=1

    = 0   (61)

    T f f x=0

    = 0

    ∂ T f f 

    ∂x

    x=1

    = 0,  ∂ T f 

    ∂z

    z=1

    = 0   (62)

    ∂ T ss

    ∂x

    x=0

    =  ∂ T ss

    ∂x

    x=1

    ,  ∂ T ss

    ∂z

    z=1

    = 0   (63)

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    17/29

    326   Sbutega & Catton

    ∂ T f f 

    ∂z

    z=1

    = − q̂ w Ĥ c

    ε̂f ̂kf  T̂ in= q w,

      ∂ T ss

    ∂z

    z=1

    = −  q̂ w Ĥ c

    ε̂sk̂s T̂ in= q wRh   (64)

    T f f  = T w  =

    T̂ w −  T̂ in

    T̂ in

    = T ss (65)

    The non-linearity of the momentum equation makes a FD solution advantageous. The

    diffusive term is approximated using a second-order centered difference on a non-uniform

    grid. Chebyshev nodes are chosen for the grid points to capture the steep gradients at the

    boundaries. The non-linear term is linearized around point i. The discretized form of the

    momentum equation is given by

    −  2M 1

    (zi+1 − zi) (zi+1 − zi−1)  uf 

    k+1

    i+1+

      2M 1(zi+1 − zi) (zi − zi−1)

      uf 

    k+1

    i(66)

    −  2M 1

    (zi − zi−1) (zi+1 − zi−1)  uf 

    k+1

    i−1+ uf 

    k+1

    iuf 

    k

    i= 1

    The resulting system is solved using a Gauss–Seidel iterative method. Further iteration

    is required because the friction factor, which is hidden in  M 1, depends on the Reynolds

    number. The tolerance was set to be ε = 10−6 for the relative error. The grid was refined

    until the solution reached the tolerance condition, and it was found that 100 points were

    enough to capture the gradients at the boundaries. Following this procedure, the VAT

    momentum solution can be obtained in about 0.04 s.

    The solution to the VAT energy equations will obviously depend on the boundary

    conditions. A general expansion of the temperature in the modified Fourier series can be

    given for both boundary conditions given in Eqs. (64) and (65):

    T f f  = I n sin ( γnx) + H n sin ( γnx)sin( γ0z)   (67)

    + sin ( γnx) [F nm cos (φmz) + E nm sin ( γnz)]

    T ss = Bn cos (φnx) + K n cos (φnx)sin( γ0z)   (68)

    + cos (φnx) [S nm cos (φmz) + P nm sin ( γmz)]

    where γn  = (2n + 1)π /2 and  φm  =  mπ . The coefficients K n  and  Bn  are the Fourierseries coefficients of the input functions. For analytical input functions these can be

    obtained analytically (such as in the constant case analyzed in the Results section). For

    any other input, these coefficients can be determined in  O (N  log N ) operations using adiscrete Fourier transform. The coefficients  H n   and  I n  are the modified Fourier series

    coefficients of the input functions and can also be calculated analytically for given func-

    tions. For general discrete functions they can be determined using a Filon-type quadra-ture in O(N ) operations (Adcock, 2011). For specified temperature boundary conditionsthe coefficients K n, H n, F nm, and S nm will be zero. For the case of a specified heat flux,

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    18/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   327

    Bn, I n, P nm, and E nm  will be zero. Thus, given the boundary conditions, there will be

    2N 2 + 2N   coefficients to be determined. The number of basis functions needed for

    a converged solution will be controlled by either the number of coefficients required to

    correctly represent the input functions or the number of coefficients required to represent

    the solution function. The convergence of these Fourier coefficients is determined by thesmoothness of the function; therefore, the overall convergence will be determined by the

    smoothness of either the input functions (which is known a priori) or the smoothness

    of the solution function (which is not known a priori). It is well known that Fourier ap-

    proximation of discontinuous functions leads to Gibbs phenomena and they have a slow

    rate of convergence; therefore, they are the worst-case scenario. However, filtering can

    be used to reduce Gibbs phenomena and recover exponential conversion away from the

    discontinuity. The validation procedure will require inputting constant input functions

    (constant heat flux), which are discontinuous in a sine Fourier representation; hence, a

    raised cosine filter will be employed:

    σn =

     1 + cos (2nπ /N )

    2   (69)

    The series expansions of the coefficients given in Eqs. (67) and (68) are differentiated

    and substituted in the governing equations. The result is then multiplied by weight func-

    tions that are the same as the basis functions and the equations are integrated over the

    domain. The resulting integrals can all be solved analytically, except for the integral in-

    volving the velocity, which would have to be solved numerically. However, in both cases

    considered, the non-dimensional parameter  M 1   in Eq. (57) is at least three orders of 

    magnitude smaller than the other parameters. Therefore, the velocity profile will be very

    uniform and a constant average velocity can be used in the energy equation without much

    loss in accuracy. After this process, all the integrals can be solved analytically and the

    system of linear equations for the coefficients can be obtained without any computation.

    The resulting linear system of equations will obviously depend on the boundary condi-

    tions selected. For the case of a heat flux input, the system is given in tensor notation as

    follows:

    G1

     γnJ 1,pn

    δc,qm

    2

     + G2

     γ2

    n

    δs,np

    2

    δc,qm

    2

     +

    φ2m

    δs,np

    2

    δc,qm

    2

      (70)

    +G3

    δs,np

    2

    δc,qm

    2

    F nm +

    −G3

    J 2,pn

    δc,qm

    2

    S nm

    =

    −G1 ( γnJ 1,pnJ 2,0q )−G2

     γ2

    n

    δs,np

    2  J 2,0q 

     γ20

    δs,np

    2  J 2,0q 

    −G3

    δs,np

    2  J 2,0q 

    H n + G3 (J 2,pnJ 2,0q )K n

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    19/29

    328   Sbutega & Catton

    G3Rh

    J 2,np

    δc,qm

    2

    F nm +

    φ2m

    δc,np

    2

    δc,qm

    2

    − G2

    φ2n

    δc,np

    2

    δc,qm

    2

      (71)

    − G3Rhδc,np

    2

    δc,qm

    2 S nm =

     γ20

    δc,np

    2  J 2,0q

    + G2φ2n

    δc,np

    2  J 2,0q

    + G3Rh

    δc,np

    2  J 2,0q

    K n + [−G3Rh (J 2,npJ 2,0q)] H n

    where the tensors J 1,np, J 2,np and  δs,np, δc,qm are defined as

    J 1,pn =

    1 0

    cos( γnx)sin( γ px) dx

    =

    [ p + (1/2)] − (−1)n (−1) p [n + (1/2)]

    π ( p2 − n2 + p − n)  if    n = p

    1π (2n + 1)   if    n =  p

    (72)

    J 2,pn =

    1 0

    sin( γ px)cos(φnx) dx =  4 p + 2

    π (4 p2 − 4n + 4 p + 1)  (73)

    δc,np =

    2   if    n =  p  = 0

    1   if    n =  p  = 0

    0   otherwise

    ,   δs,np =

      1   if    n =  p  = 0

    0   otherwise

      (74)

    The tensors J 1 and  J 2  are full; however, their outer product with a diagonal matrix will

    produce a block diagonal matrix. The resulting linear system can be cast in a matrixform:  

      A1   A2A3   A4

      F 

    =

      b1b2

      (75)

    Matrices A1, A2, and A3 are block diagonal (because they result from an outer product of 

    full matrices with the identity), while matrix A4 is diagonal. Figure 3 shows the structure

    of the matrix for  N  = 56. The number of non-zero elements is 3N 3 + N 2 out of thetotal 4N 4 elements, and for  N >  6 the matrix will be sparse and the system can be

    solved very efficiently even for large  N . The entire code was developed in MATLAB

    (The MathWorks, Inc.) and the built-in sparse solver was used. All calculations were

    performed using double-digit precision on a quad core Intel i7 2700k processor running

    Windows 7 with 16GB of RAM. For N  = 64, which is used during the validation process,the solution time is about 0.15 s. After the coefficients are obtained, the temperature and

    heat flux can be obtained by recombining the solution at any point or grid.

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    20/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   329

    0 1000 2000 3000 4000 5000 6000

    0

    1000

    2000

    3000

    4000

    5000

    6000

    nz = 529984

    FIG. 3: Matrix structure.

    4. RESULTS AND DISCUSSION

    The accuracy of the theory, closure, validity of assumptions, and solution method are

    determined by validating the numerical results with experimental results available in theliterature. In both experimental cases, the boundary condition at the base is a constant

    heat flux condition. As discussed previously, coefficients  H n  and  K n  can be calculated

    analytically:

    H n   =  σn2q w

     γn γ0, K n   =

    q wRh

     γ0n   = 0

    0   otherwise

    (76)

    The raised cosine filter [Eq. (69)] is applied to the modified sine Fourier series because

    it is the approximation of a step function. The filter will reduce Gibbs phenomena and

    improve the convergence rate away from the discontinuity point. However, a significant

    number of terms will still be necessary to obtain a good approximation of the constant

    function; therefore, the overall convergence will be controlled by the H n  coefficients. Itis important to note that this is a worst-case scenario, and for the case of a smooth heat

    input, the solution would converge within much fewer terms.

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    21/29

    330   Sbutega & Catton

    The code will be validated by modeling two different geometries with different work-

    ing fluids: an air-cooled staggered pin-fin heat sink and a water-cooled micro-channel

    heat sink. The Nusselt and Reynolds numbers are defined as

    Nudh  =ĥd̂

    hk̂f 

    (77)

    Redh  =Û avg d̂h

    ˆ ν(78)

    where dh is the VAT-derived hydraulic diameter and is defined in Eq. (56). The geometry,

    pressure drop, and heat input for the staggered pin-fin heat sink were taken from Rizzi

    (2002) and are shown in Table 1.

    The Rizzi (2002) experiments were conducted by placing an aluminum heat sink 

    in a wind tunnel and recording its thermal response for four different inlet velocities,

    which modeled four different Reynolds numbers in the 900–2400 range. The Nusselt

    and Reynolds numbers were defined using the same length scale used in Eqs. (77) and(78), and the heat transfer coefficient and friction factor were defined as

    ĥ  =  q̂ 

    T̂ s,max −  T̂ in(79)

    f  = 1

    2

    ∆ P̂ 

    L̂ d̂h

    ρ̂f  Û avg

    (80)

    where  T̂ s,max is the maximum temperature in the heat sink. The same definition of the

    heat transfer coefficient and friction factor is used when comparing the numerical results.

    Because the heat flux is constant, the thermocouple closest to the exit, which was placed

    TABLE 1: Pin-fin heat sink inputs

    Parameter Value

    L   113.75 mm

    W    113.75 mm

    H c   38.10 mm

    tb   8.25 mm

    D   3.18 mm

    S T    4.76 mm

    S L   4.76 mmT in   298.0 K

    Q   50.0 W

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    22/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   331

    about 15 mm from the outlet, recorded the highest temperature. Hence, the numerical

    T s,max  was calculated at the location of the last thermocouple. The geometry and def-

    inition of the REV are shown in Fig. 4. In the experimental studies, a wooden block 

    was used to eliminate flow bypass but is omitted in Fig. 4 to better illustrate the internal

    geometry. The closure coefficients h and f f  for the pin-fin heat sink were taken from theZukauskas (1987) experimental correlations. The formulas for porosity and the specific

    surface can be obtained geometrically, and are given by

    εf   = 1−  π D2

    8S T S L, S w   =

    π D

    2S T S L(81)

    The geometry, pressure drop, and heat input for the micro-channel heat sink were taken

    from the Qu and Mudawar (2002) experimental study and are shown in Table 2. The ex-

    periments were conducted for single-phase deionized water flowing over an oxygen-free

    copper heat sink fitted with a polycarbonate plastic cover plate. The Reynolds number

    was varied by changing the volumetric flow rate through the heat sink, and the tem-

    peratures were recorded. Qu and Mudawar (2002) defined the Reynolds number usingthe traditional hydraulic diameter as the ratio of four times the cross-sectional area to

    the wetted perimeter. However, it can be shown that the VAT-defined hydraulic diameter

    FIG. 4: Schematic of the pin-fin heat and REV definition (without the cover plate) (Zhou

    et al., 2011).

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    23/29

    332   Sbutega & Catton

    TABLE 2: Geometry and inputs

    for micro-channel heat sink 

    Parameter Value

    L   44.8 mm

    W    10.0 mmH c   0.713 mm

    tb   3.00 mm

     p   0.467 mm

    t   0.236 mm

    T in   288.0 K

    Q   448.0 W

    used in our definition of the Reynolds number reduces to the same expression for rect-

    angular channels; therefore, no conversion factors are necessary.

    The closure coefficients h and f f 

     were obtained from the Copeland (2000) numerical

    results for plane fins. A schematic of the geometry is given in Fig. 5 and the formulas

    for porosity and specific surface for this geometry are given by

    εf  =  p− w

     p  , S w =

     2H c + 2 ( p−w)

    Hp  (82)

    FIG. 5: Schematic of the micro-channel geometry.

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    24/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   333

    First, a basis independence study is conducted to determine the number of basis func-

    tions  N   necessary to obtain a converged solution. The solution was refined until the

    relative difference in the interface fluid midpoint temperature was less than 10−3. The

    fluid temperature was chosen because it is known that its convergence will be slower due

    to the slower convergence of coefficients H n. The results are shown in Fig. 6. The errorwas interpolated using a power fit, and the interpolation was almost perfect with R2 =

    0.9994. It was determined that the convergence rate is of ON −3.858

    , and N  = 64 will

    give a relative error below the required tolerance. For further proof of convergence, the

    energy imbalance was found to be only 0.01% for N  = 64. Therefore, this value will be

    used in the rest of the validation process.

    To compare the numerical and experimental results, the input pressure drop for the

    staggered pin-fin heat sink was varied to obtain a relationship for Nudh   and  f   versus

    Redh . The numerical and experimental results are compared in Fig. 7. The agreement

    for the Nusselt number is excellent. The average error is only 3.2% while the maximum

    error is 5.1%. The agreement for the friction factor is also excellent with an error of 

    less than 2.6% everywhere except for the last point, where it is 4.5%. This is to be

    expected because the last point is at a Reynolds number of 2560, which is already in

    the transition zone. The error bars for the experimental values are not provided, but

    the numerical results are expected to be well within experimental uncertainties for all

    Reynolds numbers.

    30 40 50 60 70 80 90 100 11010

    −4

    10−3

    10−2

    10−1

    N

       R  e   l  a   t   i  v  e   E  r  r  o  r

    FIG. 6: Relative error in the midpoint fluid temperature as a function of the number of 

    basis N .

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    25/29

    334   Sbutega & Catton

    800 1000 1200 1400 1600 1800 2000 2200 2400 26000

    20

    40

    60

    80

    100

    120

    140

    160

    180

    Red

    h

          N     u      d

          h

    800 1000 1200 1400 1600 1800 2000 2200 2400 26000

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    0.7

    0.8

    0.9

          f

     

    f experimental

    f predicted

    Nu experimental

    Nu predicted

    FIG. 7: Nusselt number and friction factor versus the Reynolds comparison for a pin-fin

    heat sink.

    To demonstrate the geometric flexibility of the code and to further validate the theory

    and numerical method, the code is used to model a water-cooled micro-channel heat sink.

    It is important to notice that the equations to be solved and their domain do not change.

    The only differences are in the calculation of the porosity, specific surface, the closure

    parameters h  and  f f , and the viscosity of the fluid. The comparison of the bottom wall

    temperature is shown in Fig. 8 and the results are once again excellent. At a Reynolds

    number of 890, the average and maximum errors are 3.8 and 7.3%, respectively. At a

    Reynolds number of 1454, the errors are 2.4 and 1.2%. The higher error at the lower

    Reynolds number can be explained by the fact that at lower Reynolds numbers the effect

    of the base will be more significant. The results for the pressure drop are shown in Fig. 9.

    Once again, the agreement between the numerical and experimental data is excellent.

    The average error is 4.1% while the maximum error is 5.2%. Both are expected to be

    within experimental uncertainties.

    Hence, it has been shown that the code can accurately model the hydraulic and ther-

    mal behavior of heat sinks with considerably different geometries and different working

    fluids. The ability to quickly compare different geometries makes the code particularlysuitable for population-based optimization algorithms such as genetic algorithms and

    particle swarm optimization. In such algorithms, the confidence in the optimality of the

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    26/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   335

    0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

    10

    20

    30

    40

    50

    60

    x/L

       T  s

       [   0    C

       ]

     

    Numerical at Re=890

    Numerical at Re=1454

    Experimental at Re=890

    Experimental at Re=1454

    FIG. 8:   Comparison of the numerical and experimental bottom wall temperatures at

    different x  locations for the water-cooled micro-channel heat sink (q  = 100 W/cm2).

    400 600 800 1000 1200 1400 16000

    10

    20

    30

    40

    50

    60

    70

    Red

    h

          ∆   P    [

       k   P  a   ]

    FIG. 9:   Comparison of the numerical and experimental pressure drop versus the

    Reynolds number for the water-cooled micro-channel heat sink.

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    27/29

    336   Sbutega & Catton

    global maximum increases with the number of heat sinks evaluated. One of the lim-

    its of the numerical method is that the number of basis functions required will depend

    on the smoothness of the solution, which often is not known a priori, and that of the

    boundary conditions. In the validation process considered, the overall convergence of 

    the solution is controlled by coefficients H 

    n. It is important to note that the solutionfor variable heat flux is already built into the numerical solution and can be obtained

    by using different coefficients H n  and K n. Reduced convergence is also expected if the

    solution itself has very steep gradients (e.g., low Prandtl fluids). In contrast, for smooth

    input heat fluxes (or temperatures), the convergence is expected to improve significantly;

    as a consequence, this implies that the number of basis functions required to meet the

    tolerance would be significantly lower, and the solution time would be further reduced

    considerably.

    5. CONCLUSIONS

    The governing equations for mass, momentum, and energy in a heat sink were derivedby averaging the corresponding point-wise equations over a REV. The equations were

    closed using transport coefficients that model micro-scale behavior. A code was devel-

    oped to solve the resulting VAT-derived system of PDEs. The momentum equation is

    solved using a FD scheme because of the non-linearity. The solid and fluid tempera-

    tures are expanded in a modified Fourier series, and the GM is used to obtain a sparse

    linear system in the coefficients. The basis functions are chosen such that they satisfy

    the boundary conditions a priori. The system is solved using a sparse solver to model

    different heat sinks with constant heat flux. The method was applied to model thermal

    behavior of an air-cooled pin-fin and a water-cooled micro-channel heat sink for a given

    heat load and pressure drop. It was found that about 64 basis functions in each direc-

    tion were necessary to obtain convergence, and the overall runtime was about   ∼0.25

    s. The convergence is known to improve as the smoothness of the boundary condition

    increases. The thermal behavior of the heat sink obtained numerically is compared to

    experimental values, and the agreement is excellent with an average error of less than

    4% for both cases. The code also predicted the average velocity for the given pressure

    drop with very good accuracy for both geometries. For the analyzed air-cooled pin-fin

    heat sink, the agreement with experimental data is excellent with an error of less than

    2.6% for Re < 2300. For the water-cooled micro-channel heat sink, the average error is

    4.1% and the maximum error is 5.2%. Although these errors are slightly higher than in

    the pin-fin case, the prediction is still excellent and the error is expected to be within the

    experimental uncertainties. Overall, the agreement with experimental data and the com-

    putational efficiency show the validity, accuracy, and advantage of using a GM solution

    to solve VAT-based conservation equations. The code will be extended to include casesof constant temperature, non-negligible base conduction, and changing geometry in the

    flow direction.

     Multiphase Science and Technology

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    28/29

    Efficient Hydraulic and Thermal Analysis of Heat Sinks   337

    6. ACKNOWLEDGMENTS

    We would like to sincerely thank the late Dr. Novak Zuber and the Kerze-Cheyovich

    endowment, which made this research possible.

    REFERENCES

    Adcock, B., Convergence acceleration of modified Fourier series in one or more dimensions,

     Math. Comput ., vol. 80, no. 273, pp. 225–261, 2011.

    Canuto, C., Hussaini, M. Y., Quarteroni, A., and Zang, T. A., Spectral Methods, Berlin: Springer,

    2006.

    Carbonell, R. G. and Whitaker, S., Heat and mass transport in porous media,  Fundamentals of 

    Transport Phenomena in Porous Media, J. Bear and M. Y. Corapcioglu, Eds., vol.   82, pp.

    121–198, Dordrecht, The Netherlands: Martinus Nijhoff, 1984.

    Chein, R. and Chen, J., Numerical study of the inlet/outlet arrangement effect on microchannel

    heat sink performance, Int. J. Therm. Sci., vol. 48, no. 8, pp. 1627–1638, 2009.Chu, R. C., The challenges of electronic cooling: Past, current and future,   J. Electron. Packag.,

    vol. 126, no. 4, pp. 491–500, 2004.

    Copeland, D., Optimization of parallel plate heatsinks for forced convection,  Proc. of 16th IEEE 

     Annual Symposium on Semiconductor Thermal Measurement and Management , Presented Pa-

    per, 2000.

    Finlayson, B. A., The Method of Weighted Residuals and Variatonal Principles, Mathematics in

    Science and Engineering Series, vol. 87, New York: Academic, 1972.

    Galerkin, B. G., Series solution of some problems of elastic equilibrium of rods and plates,  Vestn.

     Inzh. Tekh., vol. 19, pp. 897–908, 1915.

    Gray, W. G., A derivation of the equations for multiphase transport,  Chem. Eng. Sci., vol. 30, no.

    3, pp. 229–233, 1975.

    Gray, W. G. and O’Neill, K., On the general equations for flow in porous media and their reduc-

    tion to Darcy’s law,  Water Resour. Res., vol. 12, no. 2, pp. 148–154, 1976.

    Horvat, A. and Catton, I., Modeling of forced convection in an electronic device heat sink as

    porous media flow, Proc. of ASME International Mechanical Engineering Congress and Ex-

     position, Paper No. IMEC2002-39246, pp. 207–212, 2002.

    Horvat, A. and Catton, I., Application of Galerkin method to conjugate heat transfer calculation,

     Numer. Heat Transfer, Part B, vol. 44, no. 6, pp. 509–531, 2003.

    Imani, G. R., Maerefat, M., and Hooman, K., Estimation of heat flux bifurcation at the heated

    boundary of a porous medium using a pore-scale numerical simulation,   Int. J. Therm. Sci.,

    vol. 54, pp. 109–118, 2012.

    Kuwahara, F., Shirota, M., and Nakayama, A., A numerical study of interfacial convective heat

    transfer coefficient in two-energy equation model for convection in porous media, Int. J. Heat  Mass Transfer , vol. 44, no. 6, pp. 1153–1159, 2001.

    Ouyang, X.-L., Jiang, P.-X., and Xu, R.-N., Thermal boundary conditions of local thermal non-

    Volume 25, Number 2–4, 2013

  • 8/19/2019 Efficient Hydraulic and Thermal Analysis

    29/29

    338   Sbutega & Catton

    equilibrium model for convection heat transfer in porous media,   Int. J. Heat Mass Transfer ,

    vol.  60, pp. 31–40, 2013.

    Park, K., Choi, D.-H., and Lee, K.-S., Numerical shape optimization for high performance of a

    heat sink with pin-fins,  Numer. Heat Transfer, Part A, vol.  46, no. 9, pp. 909–927, 2004.

    Qu, W. and Mudawar, I., Experimental and numerical study of pressure drop and heat transferin a single-phase micro-channel heat sink,  Int. J. Heat Mass Transfer , vol.   45, no. 12, pp.

    2549–2565, 2002.

    Rizzi, M., An experimental study of pin fin heat sinks and determination of end wall heat transfer,

    M.S. Thesis, University of California at Los Angeles, Los Angeles, CA, 2002.

    Schmidt, R., Challenges in electronic cooling—Opportunities for enhanced thermal management

    techniques—Microprocessor liquid cooled minichannel heat sink, Heat Transfer Eng., vol. 25,

    no. 3, pp. 3–12, 2004.

    Slattery, J. C.,   Momentum, Energy, and Mass Transfer in Continua, McGraw-Hill New York,

    1972.

    Travkin, V. S. and Catton, I., Transport phenomena in heterogeneous media based on volume

    averaging theory, Adv. Heat Transfer , vol.  34, pp. 1–144, 2001.

    Tuckerman, D. B. and Pease, R. F. W., High-performance heat sinking for VLSI,   Electron Device Lett., IEEE , vol.  2, no. 5, pp. 126–129, 1981.

    Vadnjal, A., Modeling of a heat sink and high heat flux vapor chamber, Ph.D. Thesis, University

    of California at Los Angeles, Los Angeles, CA, 2009.

    Whitaker, S., Diffusion and dispersion in porous media,  AIChE J., vol.  13, no. 3, pp. 420–427,

    1967.

    Whitaker, S., Advances in theory of fluid motion in porous media,  Ind. Eng. Chem., vol.  61, no.

    12, pp. 14–28, 1969.

    Whitaker, S., The Method of Volume Averaging, Dordrecht: Kluwer Academic, 1999.

    Yang, K. and Vafai, K., Analysis of heat flux bifurcation inside porous media incorporating iner-

    tial and dispersion effects—An exact solution,  Int. J. Heat Mass Transfer , vol.  54, no. 25-26,

    pp. 5286–5297, 2011.Zhou, F., Geb, D. J., Chu, J., and Catton, I., Volume averaging theory based modeling of pin fin

    heat sinks, Proc. of ASME Power Conference, Denver, CO, Presented Paper, 2011.

    Zukauskas, A., Heat transfer from tubes in crossflow,  Adv. Heat Transfer , vol.  18, pp. 87–157,

    1987.

     Multiphase Science and Technology