You are on page 1of 17

This article was downloaded by: [Indian Institute of Technology] On: 6 June 2010 Access details: Access Details:

[subscription number 742779879] Publisher Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 3741 Mortimer Street, London W1T 3JH, UK

Numerical Heat Transfer, Part B: Fundamentals

Publication details, including instructions for authors and subscription information: http://www.informaworld.com/smpp/title~content=t713723316

Object-Oriented Development of Inverse Heat Conduction Code Adaptable to Various Configurations


Sun Kyoung Kima a Department of Die and Mold Design, Seoul National University of Technology, Seoul, Korea

To cite this Article Kim, Sun Kyoung(2006) 'Object-Oriented Development of Inverse Heat Conduction Code Adaptable to

Various Configurations', Numerical Heat Transfer, Part B: Fundamentals, 50: 3, 269 284 To link to this Article: DOI: 10.1080/10407790500508878 URL: http://dx.doi.org/10.1080/10407790500508878

PLEASE SCROLL DOWN FOR ARTICLE


Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf This article may be used for research, teaching and private study purposes. Any substantial or systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

Numerical Heat Transfer, Part B, 50: 269284, 2006 Copyright # Taylor & Francis Group, LLC ISSN: 1040-7790 print=1521-0626 online DOI: 10.1080/10407790500508878

OBJECT-ORIENTED DEVELOPMENT OF INVERSE HEAT CONDUCTION CODE ADAPTABLE TO VARIOUS CONFIGURATIONS Sun Kyoung Kim
Department of Die and Mold Design, Seoul National University of Technology, Seoul, Korea
Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

This article suggests a method for developing computer code that can solve inverse heat conduction problems (IHCPs). The concept of object-oriented development is employed to implement the computer code in an efficient and flexible fashion. The software design is conducted based on the unified modeling language. Furthermore, this article also explains how to implement the deliverable computer code using existing software development tools. The flexibility and reusability of the developed code is emphasized and verified through two examples, which are adaptable regularization and the modification of existing IHCP code for another IHCP.

1. INTRODUCTION Inverse heat conduction problems (IHCPs) have been thoroughly investigated for a variety of configurations [14]. There are many classes of IHCPs, according to the specification of the corresponding direct problem. For example, a three-dimensional steady-state IHCP [5] and two-dimensional transient IHCP [2, 4] have been solved in previous studies. Recently, IHCPs in micro- or nanoscale have also been studied [6]. The scale of the problem is now considered another measure for classification of IHCPs. IHCPs can also be classified by the kinds of unknown to be determined. Such unknowns can be the material property [7, 8], boundary conditions [1, 9], and geometry [10, 11]. Other unknowns are the heat sources [12, 13] and interface conductance [14], which are not considered in this study. Many different types of IHCPs may be encountered while analyzing real-world thermal systems [2, 3]. Implementational issues, however, are not much studied. This work suggests a unified process for implementing computer software that can solve IHCPs. This work employs the methods of object-oriented programming to realize the computer software in a rapid and reliable fashion [15]. For object-oriented development, the solution scheme for the IHCP should be based on an optimization scheme. When the inversion scheme is strongly coupled with the direct problem, the benefit of the object-oriented development is not that
Received 28 March 2005; accepted 8 November 2005. Address correspondence to Sun Kyoung Kim, Department of Die and Mold Design, College of Engineering, Seoul National University of Technology, 138 Gongreung Gil, Nowon, Seoul, 139743, Republic of Korea. E-mail: sunkkim@snut.ac.kr

269

270

S. K. KIM

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

great. The methods that modify the matrices of the direct problem are not suitable for the current approach [16, 17]. Any methods that specially treat the time domain are also not suitable [1822]. Such methods are the sequential method [1, 18], the frequency-domain method [20, 21], and filtering methods [22]. Studies on object-oriented development of numerical codes are found in previous works. They are focused on implementation of specific numerical methods such as the finite-volume method [23] or the finite-element method [24]. Masters et al. have shown how to handle data structures for numerical codes in an objectoriented way [24]. This work is not about implementation of a computer code for a specific direct problem. This work provides a framework for developing IHCP computer codes, which can be interrelated with any kind of direct problem. This framework makes it possible to build up a new IHCP code simply by assembling existing codes, since the concepts of object-oriented software development allows easier reuse, safe wrapping, and ensured interoperability. More specifically, the unified modeling language (UML), which is the standard object modeling language, is utilized in this development [25, 26]. The present article does not explain the mathematical details of the IHCP [2, 4]. Readers who are not familiar with IHCPs are recommended to read [1]. This work presents several UML diagrams and explains them. In the process of realizing those diagrams, the software design is completed. UML suggests that the scenario, use case, class, and sequence diagrams be included as essential elements in the design. This work shows and explains those four elements sequentially. By doing so, what to realize and how to implement become clear.

2. DELINEATION OF IHCP 2.1. Scenario Let us briefly describe conventional solution methods for IHCPs. There are many different solution methods for IHCPs, including the sequential method [1], the whole-domain method [14] and the space-marching method [1, 2]. In the sequential method and the space-marching method, the inverse estimation and the direct problem are strongly coupled, so it is very hard to develop a flexible and reusable code [1]. Flexibility and reusability are crucial issues in software development, since those properties allow shortening the development time [15]. Although the sequential method and other methods have many excellent features, they are not considered in this work for the aforementioned reason. The scenario here means the abstract description of what happens inside the IHCP software while it runs. There are decisions that should be made according to the developmental phases. Some of them are to be decided during the design phase and others during the implementational phase or run time. Let us describe those details. Generally, in an inversion process, we reverse the direction of information flow [2]. By doing so, the causes are found from the observations of the results. In a typical IHCP, the boundary condition, of Dirichlet or Neumann type, is determined based on temperature readings elsewhere. The main issue in this sort of inversion is the instability of the solution, which is attributed to the mathematical ill-posedness inherent in the problem [14]. To overcome this difficulty, treating the problem as an

OBJECT-ORIENTED DEVELOPMENT OF IHCP CODE

271

optimization problem is widely accepted and has been proven to be reliable [2, 4]. This study also employs the optimization-based approach. The optimization problem for an IHCP finds the unknowns of interest that render a statistical consistency between the calculation and measurement data. Thus, numerical or analytical calculation should be conducted to compare the results with the experimental data. Usually, the experimental data in an IHCP are temperature readings. Sometimes, as special cases, thermal strain or heat flux measurements can also comprise such data. This study considers temperature measurements only. 2.2. Use Case Use case means a collection of scenarios which can happen while running the software [25, 26]. The conventional scenarios of numerical tools are not very complicated. There is no fatal risk that can be incurred by a failure of this kind of software run. Thus, usually a simple description can explain the entire scenario successfully. The use cases for IHCPs may be as follows: Basic flow: a. The experimenter obtains a set of experimental data. b. The analyst gathers all the information required for the analysis, including experimental data, statistical criterion, geometry, boundary condition, material property, sensor locations, and information about unknowns of interest. c. The analyst starts the inverse analysis by initiating the optimization routine. d. The optimization routine repeatedly calls the direct simulation routine to calculate deviation between experimental and numerical data with a provision of the test unknowns. e. The optimization routine eventually finds the unknowns that satisfy the criteria imposed by the analyst and returns the unknowns as a final solution set. Alternative flow 1: c. The input data are inconsistent with each other, so that the analysis cannot start. Alternative flow 2: e. The optimization routine cannot find the unknowns that meet the imposed criteria. A use case diagram can explain the above scenario in a visual fashion, as shown in Figure 1. There are no strict rules for drawing this diagram. As long as the diagram can reveal what will happen in the system during run time, any form is good enough. The use case is an overview of the entire system from the outside of the system. It clearly explains the requirements of the system. Although it does not necessarily have to be connected to the classes of the system, it gives a conceptual foundation for building the classes. As shown in Figure 1, there are two actors in the diagram, who are the Experimenter and the Inverse Analyst. The diagram shows what those actors should do. Here, we assume that the experimental system and the software are under their control. Especially, the Inverse Analysts role is crucial. He or she should initiate the analysis

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

272

S. K. KIM

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

Figure 1. Use case diagram.

and check the integrity of the entire analysis. What should be done inside the software is now quite clear. The next step is to build the classes to implement the system. 3. DIAGRAMS 3.1. Basic Class Diagram Classes are central to objected-oriented modeling. The classes explain the objects of the system and the static relationships between them. There are three different views in classes. Each of them focuses on concept, specification, and implementation. This work basically considers the conceptual class diagram, since the implementational details are not the purpose of this article. Three important aspects in solving IHCPs are how to conduct the optimization, how to simulate the phenomena numerically, and how to handle the experimental, numerical, and user-provided data. The core of the classes is the optimization class, where all the data should be processed. Thus, the class diagram in Figure 2 shows the essence of the IHCP software. The optimization software seeks the desired unknown condition that meets the rule of discrepancy between the simulated and experimental data. The implementation-level class diagram should be extended from the basic diagram shown in Figure 2. Basically, the optimization class seeks the unknown condition of interest that meets the discrepancy rule of the IHCP. Consider an IHCP system with N measurement data. The discrepancy rule is expressed as [1] J
N X Ti Yi 2 i 1

r2 i

OBJECT-ORIENTED DEVELOPMENT OF IHCP CODE

273

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

Figure 2. Basic class diagram.

where Ti , Yi , and ri are the simulated and measured temperatures and the standard deviation at the ith sensor location, respectively. Because seeking a solution that renders J % 0 returns an unstable solution [2], the optimization problem usually becomes one to find the unknown such that J % N . However, this kind of equality is quite hard p to satisfy. Thus, p the expression is replaced by an inequality condition such as N 2N < J < N 2N [1, 2]. This discrepancy rule becomes the feasibility condition of the optimization problem. In most cases, we have to employ some sort of regularization to obtain a sufficiently smooth and dependable solution [14]. Depending on the way the regularization is handled, the optimization can be constrained or unconstrained. Basically, when the inversion involves regularization, the optimization problem should be a constrained one. However, introduction of the regularization parameter relieves the burden of the constrained optimization. Consider the regularization term, R, and the Lagrange multiplier, k. In this case, the optimization problem should minimize R subject to the feasibility condition. The Lagrangian of the optimization in this case is written as L R kJ . At the optimum point, we have rL rR krJ 0. Introducing the regularization parameter, a, which is equal to 1=k, we have rJ arR 0. We can convert the expression into a positive-definite one by multiplying by rJ T, which turns out to rJ T rJ arR 0 2

For a given a, the above expression becomes a nonlinear system of equations. It should be noted that an engineer should determine a that satisfies the feasibility condition. The level of smoothing in the resulting solution varies greatly depending on the selection of the regularization parameter. Therefore, in a sense, the constrained optimization is a process that automatically determines the regularization parameter. However, the computational cost also increases dramatically in the constrained optimization [27]. The IHCP software can run in either way according to the requirement.

274

S. K. KIM

3.2. Optimization Class In order to provide flexibility in the analysis, the computer codes should be adaptable to various optimization methods. Thus, the optimization class should be an interface class to realize such adaptability and utilize many existing optimization codes. We can wrap any existing optimization code with an interface to make it adaptable to the IHCP code. This kind of wrapping is done by employing the adaptor pattern, which is shown in Figure 3 [28]. The class diagram shown in Figure 3 is not the most general or unique one. It is simply a plausible one. This class diagram shows that a variety of optimization codes can be utilized to implement the optimization class in the IHCP code. The generalization arrow in the diagram is realized by inheritance toward the direction reverse to the arrow. For example, in C the Levenberg-Marquardt method can be declared as [24]
Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

class LevenbergMarquardtMethod : public GradientBasedOptimization{ 3.3. Whole Class Diagram

};

The whole class diagram, which neglects the detail around the optimization class, is shown in Figure 4. This diagram reveals several important features of the IHCP code. The ultimate purpose of any IHCP code is to determine the unknown of interest. Most IHCPs, as in this diagram, regard geometry, boundary condition, and material

Figure 3. Optimization classes.

OBJECT-ORIENTED DEVELOPMENT OF IHCP CODE

275

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

Figure 4. Overall class diagram.

property as the unknowns. There are several other unknowns, such as heat source, initial condition, and time constant of the thermal wave, in the hyperbolic heat equation. These are not considered here. The unknown condition is necessary for the renewal of the simulated data. The diagram conceptualizes the polymorphism of the unknown condition using a typical pattern. The realization of the unknown condition depends greatly on the type of the unknown. Thus, we need to unify the access method to the UnknownCondition class. Suppose the unknown is the boundary condition. The difficulty that arises here is that the unknown condition should be treated as a known boundary condition when solving the direct problem. The conventional way of handling this kind of situation is to allot related variables to a global array, but this has a negative impact on the flexibility and reusability of the code. The UnknownCondition class should hold the pointer to the data in the BoundaryCondition class instead of holding the boundary condition data values. The UnknownCondition class should update the corresponding boundary condition when it is updated. The BoundaryCondition class does not have to hold any information, whether it has any unknown condition or not. In this fashion, the UnknownCondition class and classes related directly to the DirectProblem class are isolated. Three classes derived from an abstract class, UnknownCondtion, are employed to resolve the different

276

S. K. KIM

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

Figure 5. Classes around UnknownCondition class.

natures of three different kinds of unknown conditions. Complete separation of the Geometry, BoundaryCondition, and MaterialProperty classes from the UnknownCondition derived-classes allows great flexibility in designing and implementing the code for the direct problem. A sample design of the UnknownCondition class is shown in Figure 5. The class should have some members characterizing the unknown. The one important thing is that the class should involve metadata that explain the data themselves in the class.

3.4. Sequences and Collaboration Diagrams The sequence diagram presents the procedures to be done by the software. Figure 6 shows the sequences diagram that performs the described inverse analysis. Understanding the diagram is quite straightforward. The diagram clarifies the tasks to be done by the software. The diagram does not have to include every detail of the procedures that the software should implement. As shown in the diagram, it reveals the essence of the process that should be done in an optimization-based IHCP code. The solid arrows mean actions and the dashed arrows mean the returns. The main routine, IHCPMain, initiates the procedures. Then, the optimization routine repeatedly calls the direct problem and compares the simulated data with the experimental data. By doing so, the optimization routine finds the inverse solution. The idea that was described in the use case diagram is further concretized in the sequence diagram. The role of this diagram is to declare procedurally what is supposed to happen inside the software. The collaboration diagram is similar to the sequence diagram. It further clarifies the tasks to be executed. Each operation is numbered according to the order of execution. As shown in Figure 7, this diagram depicts how each class is collaborating with other classes.

OBJECT-ORIENTED DEVELOPMENT OF IHCP CODE

277

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

Figure 6. Sequence diagram.

3.5. Implementational Issues Implementation here means realizing the executable computer software based on the abstraction presented in the diagrams. Recently, several development tools that are capable of automatically converting the diagrams into the computer codes have become available [29]. Even without such automation tools, the diagram greatly helps in the writing of codes. An automation tool or a developer creates a skeleton based on the diagrams. Then, the developer fills out the routines in the skeleton. The developer might be able to utilize many existing codes in order to complete the routines. For example, the developer can reuse any existing code that can solve the direct heat conduction problem. Furthermore, commercial analysis software can be interlinked to the IHCP software via a proper interface. Studies on object-oriented development of numerical codes for the direct problem are found elsewhere [23, 24]. Theses studies present various implementation issues involving the data structures, mesh generation, and matrix buildup scheme.

278

S. K. KIM

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

Figure 7. Collaboration diagram.

This article does not present any actual computer codes. Such realization is a result of the process but not the main issue in this article. The actual computer code, which is implemented based on the proposed UML design, will vary greatly depending on the coding convention, developing tools situation, and other restrictions. Figure 8 shows a screen view of the implemented IHCP software.

Figure 8. Implemented computer software.

OBJECT-ORIENTED DEVELOPMENT OF IHCP CODE

279

4. APPLICATION TO TWO-DIMENSIONAL BOUNDARY IHCP 4.1. Objective and Constraint We have applied different regularization methods to different IHCPs. It should be emphasized that Eq. (1) is not the objective function of the optimization problem. It is the constraint of the optimization problem. The objective function should be the regularization term. The Optimization class is associated with the Regularization class, which actually determines the objective function to be optimized. Let us apply this model to the Tikhonov regularization and the maximum entropy regularization to a two-dimensional transient IHCP.

4.2. Problem Statement


Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

The IHCP considered in this test case is as follows. The governing equation for temperature T x; y; t, where x (0 x a) and y (0 y b) are the Cartesian coordinate variables over the solid domain R, and t is time (0 t tf ), is described as C qT r k rT 0 qt in R 3a

Here, C and k are the volumetric heat capacity and the thermal conductivity, respectively. The corresponding conditions for the top, left, bottom, and right boundaries A1 and A2 , respectively, are specified as follows: k qT f x; t qy qT 0 qx on A1 3b

on A2

3c

In addition, initially the domain is maintained at a uniform temperature, T0 , that is, T x; y; 0 T0 in R 3d

A direct solution of the problem gives temperature as a function of location and time for given initial and boundary conditions assuming the surface heat flux f x; t known. Thus, temperature for a particular function f x; t can be denoted as T x; y; t; f . Through inverse analysis, the unknown condition which is consistent with temperature observation can be estimated. The temperature measured on the lower surface can be expressed as a continuous function Y x; t. If no error is included in the measured temperature, the inverse solution can be obtained from Find f x; t such that T x; 0; f Y x; t 4

However, we have uncertainty of ri for each temperature reading, Yi , which is obtained at a specified sensor location and at a given time. Thus, the solution should

280

S. K. KIM

satisfy Eq. (1), which is the constraint of the optimization problem. Such a solution is called a feasible solution. Thus the problem can be stated as: Find a solution that minimizes the regularization functional among those satisfying the feasibility condition, Eq. (1).

4.3. Reconstruction of Unknown Boundary Condition In this test case, the unknown boundary condition class in Figures 4 and 5 is utilized. The inverse analysis is conducted using the simulated measurement data embedded with random errors with known standard deviation, which are artificially generated with the same computer code that implements the DirectProblem class in Figure 2. The unknown boundary condition to be estimated is the boundary condition that was imposed while the simulated measurement data were generated.
Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

4.4. Adaptable Regularization Regularization classes are derived from the Objective class associated with Optimization class as shown in Figure 4. Consider two derived classes, MaxiumEntropy and Tikhonov, which are supposed to implement the maximum entropy and Tikhonov regularizations, respectively [30, 31]. The adaptable regularization is realized in this way. The regularization functional in the maximum entropy method and the zerothorder Tikhonov regularization are given as R f R f
W X i 1 W X i 1

fi ln fi2

fi F

where f ff1 ; f2 ; . . . ; fW g and F

W X i 1

fi

5a 5b

It is known that the maximum entropy method has more resolving power than the conventional Tikhonov regularization method. However, it takes a lot of computational resources, owing to the strong nonlinearity induced from the logarithm in Eq. (5a) [30]. Thus, we need to change the regularization method on the demand of the resolution. We have implemented both the classes and solved the specified IHCP. In this study, Optimization class has been realized based on the successive quadratic programming [27]. Figure 9a shows the original heat flux imposed as the boundary condition while generating the simulated measurement data. Figure 9b shows the result obtained by the zeroth-order Tikihonov regularization. Figure 9c shows the result by the maximum entropy method. As found in the literature, the maximum entropy method shows better resolution and accuracy [31]. The numerical scheme itself is not the main scope of this study. Thus, we do not present any more about the maximum entropy method. However, it should be noted that the calculation time for the maximum entropy regularization is approximately 10 times longer than that for the Tikhonov regularization. The regularization method should be properly selected

OBJECT-ORIENTED DEVELOPMENT OF IHCP CODE

281

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

Figure 9. Adaptable regularization: (a) original heat flux; (b) conjugate gradient method with zeroth-order Tikhonov regularization; (c) maximum entropy regularization.

282

S. K. KIM

according to the demand for computational accuracy. As in this case, it is convenient to employ multiple regularization methods and to select the best one among them considering the trade-off between calculation time and resolution. The objectoriented development allows easier implementation of IHCP code with selectable regularization through the adaptor pattern. 4.5. Modification to Different IHCPs This code can be easily reused for various IHCPs. For example, we can apply the method to the three-dimensional IHCP simply by modifying the DirectProblem class. Moreover, for the multidimensional case we can use commercial software as the direct solution method. In this case, the DirectProblem, BoundaryCondition, and MaterialProperty classes should be interfaced to the commercial software, which should allow such interfacing. Another example is changing the unknown of interest from the boundary condition to the geometry or material property. Object-oriented development allows this modification, simply and robustly. The material property case is easier than the geometry case. Usually, the material property case results in a parameter estimation problem [1, 7]. This renders the problem easier because the number of unknowns is fewer. To do this, the ptrCoefficents member in the MaterialProperty class should be appropriately defined. In the geometry case, the difficulty arises from the complexity of gradient evaluation and geometry updating. In a geometry reconstruction problem, the number of unknowns usually becomes very large. The conventional methods of gradient evaluation, such as the finite-difference method, which allows easier encapsulation of the class, demands unacceptably long computational time. Thus, an analytical scheme which makes the implementation of Gradient class involving and intriguing is inevitably necessary to reduce the computational time. However, even in this case, the Gradient class can be encapsulated by using another DirectProblem class solely for the gradient evaluation. In the unknown geometry case, a certain portion of boundary is supposed to deform during the calculation. Thus, the numerical mesh should be regenerated during the run time. This deformable geometry system is supposed to be implemented solely by the UnknownGeometry class (see Figure 5). A method of the geometry class, updateNodes( ), updates positions of nodes by changing the nodal coordinate values through a property of the Geometry class, ptrNode. As shown here, within this framework any existing code can be easily reused and modified to solve a given IHCP. 5. CONCLUSION This article has presented a method to develop a computer code for the inverse heat conduction problem (IHCP) in an object-oriented fashion. The development process includes the scenario description, use case, class, and sequence diagrams. Through completing the diagrams, the concept and idea inherent in the original problem become concretized, getting close to an implementational level. There are meaningful implications in introducing the object-oriented method to numerical tool developers. First, while realizing the UML diagrams, the developer

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

OBJECT-ORIENTED DEVELOPMENT OF IHCP CODE

283

becomes aware of every important feature to be implemented as well as underlying concepts. Second, revising and reusing the software becomes convenient and efficient. The current study has shown how to develop an involved numerical code that solves IHCPs. The IHCP considered is solved with the optimization-based method for the flexibility and generality of the code. The completed diagrams clearly and systematically explain what should be done by the IHCP code and how to realize it. Flexibility and reusability have been examined through two examples. First, we have shown that different regularization methods can be easily adapted to the existing IHCP code using the adaptor pattern. Second, we have also shown that the object-oriented development allows easier reuse of the existing IHCP. The boundary IHCP code can be modified to other IHCPs without great difficulty.
Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

REFERENCES
1. J. V. Beck, B. Blackwell, and C. R. St. Clair, Jr., Inverse Heat Conduction, Wiley, New York, 1985. 2. O. M. Alifanov, Inverse Heat Transfer Problem, Springer-Verlag, Berlin, 1994. 3. K. Kurpisz and A. J. Nowak, Inverse Thermal Problems, Computational Mechanics Publications, Swansea, UK, 1995. 4. M. N. Ozisik and H. R. B. Orlande, Inverse Heat Transfer: Fundamentals and Applications, Taylor & Francis, 2000. 5. S. K. Kim and W. I. Lee, Inverse Estimation of Steady-State Surface Temperature on a ThreeDimensional Body, Int. J. Numer. Meth. Heat Fluid Flow, vol. 12, no. 8, pp. 10321050, 2002. 6. S. K. Kim and I. M. Daniel, Solution to Inverse Heat Conduction Problem in Nanoscale Using Sequential Method, Numer. Heat Transfer B, vol. 44, pp. 439456, 2003. zis 7. G. P. Flach and M. N. O ik, Inverse Heat Conduction Problem of Simultaneously Estimating Spatially Varying Thermal Conductivity and Heat Capacity per Unit Volume, Numer. Heat Transfer A, vol. 16, pp. 249266, 1989. 8. S. Kim, B. Chung, M. C. Kim, and K. Y. Kim, Inverse Estimation of TemperatureDependent Thermal Conductivity and Heat Capacity per Unit Volume with the Direct Integration Approach, Numer. Heat Transfer A, vol. 44, no. 5, pp. 521535, 2003. 9. J. Su and G. F. Hewitt, Inverse Heat Conduction Problem of Estimating Time-Varying Heat Transfer Coefficient, Numer. Heat Transfer A, vol. 45, no. 8, pp. 777789, 2004. 10. C. K. Hsieh and A. J. Kassab, A General Method for the Solution of Inverse Heat Conduction Problems with Partially Unknown Geometries, Int. J. Heat Mass Transfer, vol. 29, no. 1, pp. 4758, 1986. 11. S. K. Kim and W. I. Lee, Implementation of Inverse Method for Estimating Undetermined Boundary in a Two-Dimensional Slab Based on Temperature Measurement, Numer. Heat Transfer A, vol. 46, pp. 515523, 2004. 12. C. Huang and S. Cheng, A Three-Dimensional Inverse Problem of Estimating the Volumetric Heat Generation for a Composite Material, Numer. Heat Transfer A, vol. 39, no. 4, pp. 383403, 2001. 13. G. Blanc, J. V. Beck, and M. Raynaud, Solution of the Inverse Heat Conduction Problem with a Time-Variable Number of Future Temperatures, Numer. Heat Transfer B, vol. 32, no. 4, pp. 437451, 1997. 14. T.-C. Chen and P.-C. Tuan, Inverse Problem of Estimating Interface Conductance between Periodically Contacting Surfaces Using the Weighting Input Estimation Method, Numer. Heat Transfer B, vol. 41, no. 5, pp. 477492, 2002.

284

S. K. KIM

15. P. Coad and J. Nicola, Object-Oriented Programming, Pearson Education, 1993. 16. A. Behbahani-nia and F. Kowsary, A Direct Transformation Matrices Method for Solution of Inverse Heat Conduction Problems, Numer. Heat Transfer B, vol. 46, no. 4, pp. 371386, 2004. 17. N. Al-Khalidy, A General Space Marching Algorithm for the Solution of Two-Dimensional Boundary Inverse Heat Conduction Problems, Numer. Heat Transfer B, vol. 34, no. 3, pp. 339360, 1998. 18. T. S. Kumar, A Serial Solution for the 2-D Inverse Heat Conduction Problem for Estimating Multiple Heat Flux Components, Numer. Heat Transfer B, vol. 45, no. 6, pp. 541563, 2004. 19. A. L. Koay, S. H. Pulko, and A. J. Wilkinson, Reverse Time TLM Modeling of Thermal Problems Described by the Hyperbolic Heat Conduction Equation, Numer. Heat Transfer B, vol. 44, no. 4, pp. 347363, 2003. 20. J. Blum and W. Marquardt, An Optimal Solution to Inverse Heat Conduction Problems Based on Frequency Domain Interpretation and Observers, Numer. Heat Transfer B, vol. 32, no. 4, pp. 453478, 1997. 21. T. Lu ttich, A. Mhamdi, and W. Marquardt, Design, Formulation, and Solution of Multidimensional Inverse Heat Conduction Problems, Numer. Heat Transfer B, vol. 47, no. 2, pp. 111113, 2005. 22. N. Daouas and M.-S. Radhouani, A New Approach of the Kalman Filter Using Future Temperature Measurements for Nonlinear Inverse Heat Conduction Problems, Numer. Heat Transfer B, vol. 45, no. 6, pp. 565585, 2004. 23. I. Masters, I. Rees, and R. W. Lewis, Advanced Programming Methods and Data Structures for Computational Modeling Using Edge-Based Finite Volume Methods, Int. J. Numer. Anal. Meth. Geomech., vol. 28, pp. 15211532, 2004. 24. I. Masters, A. S. Usmani, J. T. Cross, and R. W. Lewis, Finite Element Analysis of Solidification Using Object-Oriented and Parallel Techniques, Int. J. Numer. Meth. Eng., vol. 40, pp. 28912909, 1997. 25. M. Folwer, UML Distilled: A Brief Guide to the Standard Object Modeling Language, 3rd ed., Addison-Wesley, 2003. 26. G. Booch, J. Rumbaugh, and I. Jacobson, The Unified Modeling Language User Guide, Addison-Wesley, 1998. 27. J. S. Arora, Introduction to Optimum Design, McGraw-Hill, New York, 1989. 28. E. Gamma, R. Helm, R. Johnson, and J. Vlissides, Design Patterns, Addison-Wesley, 1995. 29. T. Quatrani, Visual Modeling with Rational Rose 2002 and UML, Addison-Wesley, 2002. 30. A. N. Tikhonov and V. Y. Arsenin, Solution of Ill-Posed Problems, Winston and Sons, Washington, DC, 1977. 31. S. K. Kim and W. L. Lee, Solution of Inverse Heat Conduction Problems Using Maximum Entropy Method, Int. J. Heat Mass Transfer, vol. 45, no. 2, pp. 381391, 2002.

Downloaded By: [Indian Institute of Technology] At: 12:04 6 June 2010

You might also like