Saip2014 522
Saip2014 522
Saip2014 522
Abstract. Thermal analysis and solution of heat problems most often utilizes known thermal
conductivity material data which is typically experimentally determined from heat flux
measurements through the application of Fourier’s law. The challenge posed by this approach
is the need for known thermal conductivity reference materials which may be inhomogeneous
and have large associated uncertainties in industrial physics applications. In this paper we
investigate the feasibility of developing a thermal conductivity measurement system that utilizes
known radiometric input sources and temperature output measurements which may have smaller
relative uncertainties by formulating the system as an inverse heat conduction problem utilizing
recently reported research results from the fields of geophysics and mathematical optimization.
1. Introduction
Thermal physics analysis and solution of heat problems often utilizes known thermal conductivity
material data which is typically experimentally determined from heat flux measurements through
the application of Fourier’s law q 00 = −k∇T , where q 00 /[W m−2 ] is the heat flux, k/[W m−1 K−1 ]
is the thermal conductivity, and T /[K] is the absolute temperature following the nomenclature
in [1]. The use of Fourier’s law presents a straight forward mechanism to define and infer the
thermal conductivity through the ratio of the heat flux measured using standard techniques
as discussed in [2], and the temperature gradient for a specified direction of heat flow where
the unknown thermal conductivity may be expressed in terms of known reference quantities
such as thermal conductivity, wall thickness and wall temperature amongst other experimental
measurements for homogeneous isotropic materials as illustrated in figure 1.
Unfortunately materials such as aluminium/steel thermal properties may be inhomogeneous
and/or non-isotropic either from physical effects in the smelting process or manufacturing
effects from the fabrication of structures requiring knowledge of thermal conductivity such as
reflectors/concentrators in for example solar plants [6] where in addition the thermal conductivity
may vary with temperature [7, 8] as illustrated in figure 2 for a selection of materials.
As a result it is desirable in applied industrial research contexts to infer thermal conductivity
information directly by an inverse problem formulation using direct temperature measurements
which are more practical and experimentally convenient, either when reference thermal
conductivity materials are unavailable for prior testing, impractical due to prevailing operating
conditions, or when in situ process or condition monitoring measurements are necessary.
In general accurate heat sources such as calibrated lasers or electrical resistance heating
elements, and temperature measurement devices such as thermocouples or resistance
thermometers, are simple and straightforward to procure and utilize. As a result in this paper
we opt to analyze a system in which the thermal conductivity may be inferred through an Inverse
Heat Conduction Problem (IHCP) modelling approach utilizing just radiometric/electrical
heating sources and temperature measurements which are easily accessible and which avoids
the need for specific reference thermal conductivity material components, and specialist heat
flux instruments and blackbody equipment [9].
2. Literature Review
Within the statistical literature simulation problems may typically be classified as either
direct/forward or indirect/inverse where in general terms the former are cases in which PDE
parameters are known and one utilizes this information to solve for the PDE solution, whilst in
the latter one utilizes the PDE solution to infer the underlying PDE parameters.
In the context of thermal physics a direct/forward problem would correspond to using known
thermophysical properties such as thermal conductivity and/or specific heat capacity as inputs
with suitable boundary conditions for a boundary Γ = ∂Ω to solve the heat diffusion equation
∇ · (k∇T ) + q̇ = ρcp ∂T
∂t for some problem domain Ω whilst an indirect/inverse problem would
in a certain sense “work backwards” to use the PDE temperature field solution to infer the
corresponding thermophysical parameters as discussed in [10]. Traditionally inverse problem
studies have been common in the geophysical sciences to determine rock densities, gravitional
field strengths, and in oil and gas exploration as discussed in [11, 12, 13] with newer insights into
Monte Carlo based inverse parameter uncertainties reported in [14], and which has generally been
formulated and solved using the Monte Carlo and Levenberg-Marquardt techniques as outlined
in [15, 14].
For IHCP studies methods of solution have mainly to date traditionally consisted of the
Levenberg-Marquardt and conjugate gradient techniques for mixtures of problems in laboratory
thermophysical testing and space vehicle atmospheric reentry design and research studies after
the development of techniques such as Tikhonov regularization which were able to modify and
reformulate the original ill-posed/unstable inverse problems as approximate well-posed problems.
More recently investigations in inverse theory problems across different fields has involved
applications of newly developed techniques such as the variational iteration method or (VIM)
[16], the method of fundamental solutions or MFS [17, 18, 19, 20, 21, 22, 23, 24, 25], the lattice
Boltzmann method or LBM [26], heuristic approaches drawing from a mixture of techniques
[27, 28], and meshfree approaches [29]. Finite difference methods for IHCP studies are further
discussed in [30, 31] who explore how the Tikhonov regularization method may be used to
regularize the ill-conditioned linear system of equations for a non-steady two dimensional heat
conduction problem, and a iteration approach to determine the regularization parameter was
explored in [32]. A measurement methodology for infrared thermography of inverse models that
used a maximum entropy principle was reported in [33] for a die forging application in order to
deduce the unknown boundary condition of a heat flux using a finite difference discritization.
3. Mathematical Development
Based on the literature review we opt for simplicity since the underlying problem is nonlinear to
use a Levenberg-Marquardt optimization in our formulation to avoid a Tikhonov regularization.
Utilizing the heat diffusion equation ∇ · (k∇T ) + q̇ = ρcp ∂T ∂t which reduces to a generalized
∂ ∂T ∂ ∂T
two dimensional Poisson equation ∂x (k ∂x ) + ∂y (k ∂y ) = −q̇ in the special case for steady-state
conditions for a planar domain for the particular geometry of the experimental system illustrated
in figure 2 where the domain Ω is a circular region of diameter D where the boundary Γ is held
at a constant temperature Tf , where T /[K] is the temperature, k/[W m−1 K−1 ] the thermal
conductivity of the sample, and q̇ is an energy sink/source term which is positive if energy is
generated within the medium and negative if energy is being consumed.
Adopting a radial basis function discretization following the discussions in [34, 35, 29] we
approximate the thermal conductivity as k = N
P k
PNT i=1 αi ϕk (||x − xi ||, and the temperature field
as T = j=1 βj ϕT (||x − xj ||. For the temperature field let NT be the total number of points
composed of NI interior points and NB boundary points such that NT = NI + NB . In this
approach it is not necessarily the case that Nk = NT however this will simplify subsequent
calculations. The terms αi (i = 1, . . . , N ) and βj (j = 1, . . . , N ) are coefficients that are used
to build up the k(x) and T (x) fields for x ∈ Ω in terms of the RBF’s ϕk and ϕT respectively.
2 2
Utilizing Gaussian RBF’s for simplicity to illustrate the approach adopted of form ϕk = e−εk r
2 2
and ϕT = e−εT r where r = (x − xi )2 + (y − yi )2 and where εk and εT are suitable shape
p
parameters for the thermal conductivity and temperature fields respectively, and substituting
into the generalized two dimensional Poisson equation we then have using the short hand ri = r
∂k ∂T ∂2T ∂k ∂T ∂2T PNk −ε2k ri2 ],
for points in the interior that 0 = ∂x ∂x + k ∂x2 + ∂y ∂y + k ∂y 2 + q̇ where k = i=1 [αi e
−ε2k ri2 ], ∂ 2 k = 2 −ε2k ri2 ], and similar
∂k PNk 2
PNk 2 2
∂x = i=1 [−2εk (x − xi )αi e ∂x2 i=1 [−2εk αi {1 − 2εk (x − xi ) }e
expressions for the corresponding temperature field.
Let Tk be the experimental measured temperature and θk the estimated temperature for
some choice of assumed parameter P , where in our particular problem the parameter is used to
specify the thermal conductivity. Constructing an objective function S(P ) = N 2
P
k=1 [T k −θ k (P )]
it follows that we solve the IHCP by determining the choice of parameter such that S(P ) is
minimized. The objective function may be written in matrix form by specifying the measured
and estimated temperatures as a column vectors T T = [T1 , . . . , TN ] and θ T = [θ1 , . . . , θN ] and
then constructing the objective function as S(P ) = [T − θ(P )]T [T − θ(P )]. A solution will
occur when ∇S(P ) = 0 which after algebraic manipulations reduces to the following equations
4. Discussion
In this paper we have investigated how to use an IHCP formulation as a means to determine
thermal conductivity values for materials from experimental temperature measurements and
numerical simulations. Potential benefits of the method that have been investigated are that
there is no need for specialist laboratory reference thermal conductivity material standards, the
method can be applied in industrial/plant environments with standard equipment/instruments
with minimal physical complexity, and that the numerical algorithm to implement the method
may be utilized to accurately relate and estimate the thermal conductivity uncertainties in terms
of the input experimental data.
References
[1] Incropera F P and DeWitt D P 1996 Fundamentals of Heat and Mass Transfer 4th ed (John Wiley & Sons)
[2] Tsai B K, Gibson C E, Murthy A V, Early E A, DeWitt D P and Saunders R D 2004 Heat flux sensor
calibration Tech. rep. National Institute of Standards and Technology nIST Special Publication 250-65 (37
pages)
[3] JCGM/WG1 2008 Evaluation of measurement data – guide to the expression of uncertainty in measurement
(JCGM 100:2008) Tech. rep. BIPM/IEC/IFCC/ILAC/ISO/IUPAC/IUPAP/OIML
[4] JCGM/WG1 2008 Evaluation of measurement data – supplement 1 to the “guide to the expression of
uncertainty in measurement” – propagation of distributions using a Monte Carlo method (JCGM 101:2008)
Tech. rep. BIPM/IEC/IFCC/ILAC/ISO/IUPAC/IUPAP/OIML
[5] JCGM/WG1 2011 Evaluation of measurement data – supplement 2 to the “guide to the expression of
uncertainty in measurement” – extension to any number of output quantities (JCGM 102:2011) Tech.
rep. BIPM/IEC/IFCC/ILAC/ISO/IUPAC/IUPAP/OIML
[6] Hasuike H, Yoshizawa Y, Suzuki A and Tamaura Y 2006 Solar Energy 80 1255–1262
[7] Rodriguez-Sanchez M R, Soria-Verdugo A, Almendros-Ibanez J A, Acosta-Iborra A and Santana D 2014
Applied Thermal Engineering 63 428–438
[8] Powell R W, Ho C Y and Liley P E 1966 Thermal conductivity of selected materials Tech. rep. National
Institute of Standards and Technology (previously NBS) http://www.nist.gov/data/nsrds/NSRDS-NBS-
8.pdf
[9] Miklavec A, Pusnik I, Batagelj V and Drnovsek J 2013 Measurement Science and Technology 24 1–8
[10] Ozisik M N and Orlande H R B 2000 Inverse Heat Transfer – Fundamentals and Applications (Taylor and
Francis)
[11] Mosegaard K and Tarantola A 1995 Journal of Geophysical Research 100 12431–12447