Inverse design and control of thermal systems

This study considers the design of thermal processing systems where the goal is to design a system with the correct geometry and materials, so that together with the necessary energy input from the engineering devices, it would satisfy the needs of the process to be carried out. In some systems, conditions at a specified steady state are of interest, while in some others the goal is to follow a specific thermal history. Some of the common applications for such systems include rapid thermal processing of semiconductor wafers, curing, annealing, chemical vapor deposition applications, industrial baking or certain biomedical applications. Solution of a coupled boundary condition estimation problem together with geometry and property estimation problems is necessary. This study focuses on boundary condition design so that the challenges of the problem can be investigated and tackled in isolation from the other two problems. The traditional method of solution for such a problem is by trial-and-error methods, which consider the solution through a series of forward problems, where the effects are calculated for prescribed causes. Trial-and-error solution methods are computationally very expensive and it is often hard to achieve reasonable solutions. An alternative approach, inverse design, is based on formulating the design problem as an inverse problem and it is used here so that a direct solution is possible. Here the cause for a certain effect is sought directly. However, the use of an inverse formulation leads to an ill-posed problem where solutions become unstable and even unphysical. Therefore, the use of regularization techniques is essential to achieve reasonable and accurate solutions. Generic design methodologies are developed and presented to solve steady and transient boundary condition design problems making use of an inverse formulation and regularization. The developed solution techniques are experimentally validated and their applications are demonstrated through solution of sample thermal design problems in radiating enclosures. Moreover, control algorithms based on artificial neural networks are developed for control of distributed transient systems.