Session: Research Posters

Paper Number: 113333

113333 - A Solution to an Inverse Heat Transfer Problem With Phase Change by Means of Meta-Heuristics and Artificial Neural Networks: A Comparative Study 

Many engineering problems involve heat and mass transfer. A typical solution procedure for such a problem represents the determination of thermal behaviour and thermal response of a system under specific conditions, including thermal conditions (initial and boundary conditions) and material-related parameters (thermo-physical properties). Such a problem is referred to as a direct problem. In certain cases, however, some of these conditions or parameters are not known. Instead, information about the thermal behaviour is available. This is the case in which an inverse heat transfer problem has to be solved. Such a problem is actually a data-fitting and optimization task since conditions and parameters, which minimize the error between actual and prescribed data, are searched indirectly. A crucial distinction between direct and inverse problems is their posedness: direct problems are well-posed, meaning that there exists a unique solution and a small perturbation in input data causes a small perturbation in output data. On the other hand, inverse problems are ill-posed, which means that no existence nor uniqueness of the solution is guaranteed, and a small change in input data leads to significant changes in output data. A number of gradient-based methods to inverse problems have been developed in the past, including the Levenberg-Marquardt method, the conjugate gradient method, and the well-known Beck’s sequential function specification method. However, they suffer from some disadvantages, including proneness to get trapped in local optima or a low performance in large-scale problems. This is the reason why so-called soft computing methods have experienced great development in recent years.  In contrast to hard computing including deterministic gradient-based algorithms, a soft computing approach means that the algorithm does not require gradient evaluation, seeks a sufficiently good solution instead of solely a global solution, and randomness involved in a soft computing algorithm plays a crucial role, allowing e.g. to escape from local minima. In this paper, seven meta-heuristic algorithms and one algorithm based on an artificial neural network (ANN), referred to as an LSADE algorithm, were applied to the solution of an inverse heat transfer problem with phase change. The problem involved an inverse identification of parameters of an effective heat capacity function, which is a common technique used in phase change modelling. An air-PCM heat exchanger for latent heat thermal energy storage and solar air heating was used as a study case. Results obtained within the scope of the study indicate that the ANN-based LSADE algorithm significantly outperformed other meta-heuristic algorithms, which makes it a very promising tool for the solution of similar kinds of problems.

Presenting Author: Lubomir Klimes Brno University of Technology

Presenting Author Biography: Associate professor at the Department of Thermodynamics and Environmental Engineering, Faculty of Mechanical Engineering, Brno University of Technology, Brno, Czech Republic. Areas of interest: computational heat transfer, thermal energy storage, renewable energy. Further information available at https://www.vut.cz/en/people/lubomir-klimes-77488.

Authors:

Lubomir Klimes Brno University of Technology
Jakub Kudela Brno University of Technology
Martin Zalesak Brno University of Technology
Pavel Charvat Brno University of Technology