Abstract
Heat exchange between a conducting plate and the environment is described here by means of an unknown nonlinear function F of the temperature u. In this paper we construct a method for recovering F by means of polynomial expansion, perturbation theory and the toolbox of thermal inverse problems. We test our method on two examples: In the first one, we heat the plate (initially at ) from one side, read the temperature on the same side and identify the heat exchange law on the opposite side (active thermography); in the second example we measure the temperature of one side of the plate (initially at ) and study the heat exchange while cooling (passive thermography).
Export citation and abstract BibTeX RIS
1. Introduction
The present paper deals with heat exchange between a homogeneous conducting plate Ω and the environment. It is known that there are cases in which linear Newton's law of cooling fails to describe the physics of the problem (see [1, 13, 20]).
Moreover, classical nonlinear laws (Dulong–Petit, Newton–Stefan laws) 'can be applied with confidence over the range of conditions usually found in laboratory calorimetric experiments' [18] but there are natural and industrial circumstances in which the form of the nonlinearity is unknown and requires a specific analysis (see for example [8, 11, 12, 17, 19]).
We assume that the heat transfer is described here by an unknown nonlinear function F of the temperature u. In this paper we construct a method for recovering F by means of a polynomial expansion, perturbation theory and, finally, the typical toolbox of thermal inverse problems including Tikhonov regularization (see also [5, 9, 13, 10]). Input data consists of a sequence of temperature maps taken on an accessible subset of the boundary of Ω.
In section 2 we describe in detail the direct model and prove the stability of the temperature with respect to the size of the nonlinearities that appear in the boundary conditions.
The inverse problem and our method for identifying F are described in section 3.
In section 4 we test the method for two different physical simulations. In both cases only one face of the plate is accessible.
In the first one we simulate the heating of the accessible side of Ω by means of a controlled flux generated by a lamp (active thermography) and assume that the cooling law on the opposite inaccessible side is an unkown nonlinear perturbation of Newton's law. We identify the nonlinear term in the cooling law from a sequence of temperature maps taken on the accessible side. Temperature ranges from to .
The second example deals with cooling from high temperature (from to ) and is taken from [5]. The specimen is not heated (passive thermography). We show a regularized approximation of the unknown function F whose quality is comparable to the reconstructions proposed in [5] though our assumptions are less restrictive.
We adopt the following notation for function spaces:
is the set of real continuous functions defined on the closed set
is the set of real continuous functions defined on the open set whose partial derivatives are continuous up to the order .
2. The direct model
We limit ourselves to the 2D problem in which Ω is a rectangular section of a thin plate. More precisely, let Ω be the open strip with . For each , we define and . Suppose that Ω represents a metallic specimen with uniform conductivity κ.
If the maximum norm of u is defined as .
The temperature in Ω satisfies the heat equation
with boundary conditions for ,
and initial data
On the vertical sides of Ω we assume the adiabatic conditions
A list of details about physical parameters and mathematical notation follows:
is the diffusivity. The plate Ω is made of a metal of density ρ and specific heat at constant pressure c;
and are the coefficients of surface heat transfer corresponding to z = a and z = 0, respectively;
and are the temperatures of the surrounding media (assumed constant) while the initial temperature Tin is a smooth function defined in Ω;
accounts for the nonlinear functional relation between the surface temperature and the rate of heat exchange through the upper side of Sτ. The parameter is the scale factor of the nonlinearity. The function f belongs to . The set J is called 'the sector between upper and lower solutions' and will be defined in the next section.
is the solution of the initial boundary value problem (IBVP) (1)–(5). This notation points out the dependence of the solution on the scale factor . Hence, u0 is the 'background' temperature corresponding to the linear heat exchange for , .
Φ is a prescribed heat flux into the surface z = 0. It is generated by a controlled heat source (a lamp, a battery of lamps, a laser). Usually it takes the form where and can be either a periodic function (lock-in thermography) or a pulse (pulsed thermography) [13].
2.1. Pao's results about the direct model
The theoretical background of the direct model consists of a set of results by Pao [14] in which existence and uniqueness of solutions of parabolic equations with nonlinear boundary conditions are proven under suitable hypotheses. The main theoretical statement (Theorem 1.1, chapter 4 [14]) suggests a strategy for the numerical approximation of the solution as limit of a monotone sequence of solutions of linear problems. Stability of the solution of the direct model (1)–(5) with respect to is a corollary of theorem 1.1 chapter 4 [14].
To lighten the notation, in (2)–(3) we set and . In the introduction, the geometry of our thin plate was described by a rectangular strip of thickness . In order to apply Pao's theorem, here we assume that Ω is the convex open set in
where
with (),
The domain Ω looks like a finite thin strip with smoothed corners. Its boundary is a closed curve of class C1.
Furthermore, we write down our boundary conditions in the form
where and ν is the outward unit normal to the boundary . The function g is continuous and it is piecewise defined by
and
In the rest of the boundary we have .
To describe the essentials of this result, we must introduce the definition of upper (lower) solution: a function is called an upper solution of (1), homogeneous (4) and (6) if it satisfies the inequalities
and the initial condition
As for the lower solution the definition is the same, only changing with . The sector is defined as
Assume that and that g is Lipschitz in J.
The existence of a unique solution u of (1), homogeneous (4) and (6) is proved by Pao in theorem 4.1.1 [14] under the assumption that there exist a lower and an upper solution and of it. The proof is based on the iterative construction of two sequences, and , that converge monotonically to the solution u. The decreasing sequence starts with the upper solution and approximates the solution u from above while starts with the lower solution and converges to u monotonically from below. Numerical implementation of and seems to be very expensive though each is determined by solving a linear BVP whose boundary conditions involve . Details about the definition of sequences and are in [14] section 4.1. This construction furnishes the main tool for providing the stability estimate for u given in subsection 2.2.
Finally, we show that the sector J is not empty. Actually, consider the linear function
with and . Straightforward calculations show that is an upper solution and is a lower solution of (1), homogeneous (4) and (6).
2.2. Stability of the direct model with respect to
Theorem. Let be the solution of (1), homogeneous (4) and (6). We have
Proof. We recall that converges monotonically to . It means that
for . The IBVP solved by is linear. If is an upper solution of (1), homogeneous (4) and (6) when , it is well known (stability of linear IBVP w.r.t. parameters, see for example [16] page 507) that
On the other hand, we have
for . Finally, (10) turns out to be true for and .
3. The inverse problem
The IBVP (1)–(5) is the frame (direct model) in which we define the following inverse problem:
IP Identify the nonlinear term from the knowledge of a finite sequence of temperature measurements taken on a portion of the boundary of Ω.
A similar problem, posed in the stationary frame of Laplace's equation, has been studied in [6, 7]. A stability estimates for the solution of IP is given in [15].
3.1. Perturbative analysis of the direct model
The nonlinear term in (2) is unknown. We will use the equations (1)–(5)of the direct model and the knowledge of thermal contrast at z = 0 to recover .
First, we plug the formal expansion in the IBVP (1)–(5) and transform it in a perturbative hierarchy of linear problems. We consider only order zero and order one of the scheme.
The term u0 satisfies the heat equation
with boundary-initial conditions
The solution u0 is just the background temperature u0. Observe that (14)–(34) is well-posed and, in particular, once the physical parameters Tin, α, γ, κ and Φ are known, u0 is uniquely determined. On the other hand, the function fulfills the heat equation
with linear boundary conditions
with
and initial condition
The thermal contrast (measured in real cases by means of an infrared camera)
gives us the following noisy additional boundary condition that will be used in section 3.2 to recover :
3.2. The expansion of is not merely formal
Actually, we prove that . We set
We have
with boundary and initial conditions
and
Since is supposed Lipschitz in J, we have from (10) that
It comes from classical estimates (see [16] page 507) that
3.3. Discretization and approximation of
Let be a sequence of linearly independent functions that span so that . Furthermore, let solve the linear IBVP
with boundary conditions
and initial data
It comes from linearity that the solution of (19)–(23) takes the form of
Since (or, equivalently, the vector β) is not known, we can try to approximate it from the knowledge of the thermal contrast . To do this, we set the following minimum problem in finite dimension
Since the are linearly independent, any finite Gram matrix is expected to be nonsingular, even if it could be severely ill-conditioned. For this reason, regularization is required to handle truncation errors and the effects of noise affecting our data.
As for the penalty , it must be chosen using a priori information if available. In fact, we assume that is smooth and increasing. This assumption is supported by a number of known examples of nonlinear heat transfer coefficients ([1, 4, 5, 12] and many others) and by private discussions. The idea is that the higher the temperature, the greater the rate at which heat passes through an interface. Moreover, sudden jumps related to the temperature increase are not expected. A good choice of B could be the L2 norm of the first derivative of .
Hence, we assume that () is a non decreasing function which can be approximated by means of low-order polynomials. At this stage, we assume so that we will work with the finite expansion
In our tests, we take N = 4. This value of N comes from a compromise between the accuracy of the approximation and the stability of the solutions. We write the L2 norm of as the quadratic form in β
Finally, the Euler equations of our minimum problem are
for .
4. Numerical tests
We apply the perturbative scheme introduced in section 3.1 to the reconstruction of unknown nonlinear heat exchange laws in two different examples:
- (1)a metallic plate is warmed up starting from
- (2)a purely theoretical sample is allowed to cool starting from a temperature of .
4.1. Active thermography
We suppose that the heat flux Φ is constant in x and that the unknown nonlinear function is with and . Hence, it is natural to reduce ourselves to a one-dimensional problem (see also [3]), where . We provide heat to the boundary z = 0 through a flux of density while nonlinear heat exchange takes place on the opposite boundary z = a.
The temperature of our one-dimensional sample is a function that solves the IBVP
where , , , , .
We solve numerically equations (39)–(42) by means of a finite difference scheme (for the explicit and implicit-explicit numerical schemes adopted here, see [2]). The trace of the solution is assumed to play the role of real temperature data. In experiments, these values are taken by means of an infrared camera.
is the background temperature corresponding to , i.e. the situation in which heat exchange through the boundary z = a follows Newton's cooling law.
Consider the expansion . It is easy to check that is just the thermal contrast considered in the definition of the inverse problem IP. Since is actually not known, we set and . We plug this expansion in the IBVP above and carry out a perturbative analysis.
4.1.1. Order zero
The function u0 fulfills
Because of the hierarchic structure of perturbations, u0 will enter as a given quantity in the following IBVP corresponding to the order one.
4.1.2. Order one
The scaled first order solution W solves
It comes from the linearity of (46)–(49) that
where is the solution of
for .
We recall that is the thermal contrast defined in (20). To make contrast data more realistic, we add white Gaussian noise of average zero and standard deviation °C.
Since the matrix is ill-conditioned, to determine the values of we solve the linear system (38) for N = 4 with an optimal choice of the regularization parameter λ. Recall that the penalty function B is the Euclidean norm of the first derivative of the nonlinear term f.
The parameter λ is determined by means of Hansen's L-curve method [9]. In figure 1 we summarize the essentials of the numerical example:
Figure 1(a). We plot the graphs of (that plays the role of experimental temperature data at the boundary z = 0) and (the solution of order zero).
Figure 1(b). We plot the curve parametrized by λ. The corner suggests a choice for the numerical value of λ.
Figure 1(c). Finally, we compare the unknown with the polynomial reconstructions with and .
4.2. Cooling
We simulate cooling of a one-dimensional specimen represented by the interval by means of the following IBVP (see [5]):
where
Let us assume
where A = 367 500. We recall that there is no controlled heat flux here ().
The function simulates the collection of data by means of the infrared camera. Simulated temperature for z = a become more realistic by adding for all a random variable g with uniform distribution on the interval (the same noise used in [5]).
In order to apply again the perturbative scheme of the previous example we introduce a linear part so that . A natural choice for q is the surrounding temperature . We leave as a free parameter that will be identified in the next subsection.
4.2.1. Order zero. Choice of the linear part
The order zero solution (we are stressing the dependence on ) solves the IBVP
The parameter must be chosen to minimize the integral distance . The best value is .
4.2.2. Order one
We compute a finite basis solving the IBVP
for . Then, using equations (37) and (38), we estimate the vector coefficient β to approximate the nonlinear part .
Remark. We observe numerically that the temperature approximation of the first order is a very good approximation of the solution of (54)–(57) (see figure 2).
Download figure:
Standard image High-resolution imageSince the matrix is severely ill-conditioned, the values of are obtained by solving the regularized linear system (38) for N = 4. The regularization parameter λ is determined by means of Hansen's L-curve method [9].
In figure 3, we summarize the essentials of the numerical example:
Download figure:
Standard image High-resolution imageFigure 3(a). Here, we plot the functions (that plays the role of experimental temperaure data at the boundary z = 0) and (the solution of order zero).
Figure 3(b). We plot of the curve parametrized by λ. The corner suggests the choice for λ.
Figure 3(c). We compare the unknown with the polynomial reconstructions with and .
Finally, the regularized curve shown in figure 3(c) is a monotone approximation of obtained by smoothing the input data and regularizing the solutions. In [5], the additional condition
is used to obtain a monotone approximation of F. In our approximation, we don't use any assumption on the values of F.
5. Conclusions
We have used active infrared thermography and perturbation theory in order to recover an additive nonlinear term in Newton's cooling law. Theoretical background in nonlinear boundary value problems for parabolic equations gives us a stability estimate for the direct model. As for the constructive procedure, we remove the nonlinearity by means of perturbation theory and produce a regularized polynomial approximation by means of least squares minimization. At present, our algorithm is working with synthetic data.
Acknowledgments
We wish to thank Dr Paolo Bison for fruitful discussions, and the referees for their precious suggestions and comments.