Publications Details
Application of optimization to the inverse problem of finding the worst-case heating configuration in a fire
Thermal optimization procedures have been applied to determine the worst-case heating boundary conditions that a safety device can be credibly subjected to. There are many interesting aspects of this work in the areas of thermal transport, optimization, discrete modeling, and computing. The forward problem involves transient simulations with a nonlinear 3-D finite element model solving a coupled conduction/radiation problem. Coupling to the optimizer requires that boundary conditions in the thermal model be parameterized in terms of the optimization variables. The optimization is carried out over a diverse multi-dimensional parameter space where the forward evaluations are computationally expensive and of unknown duration a priori. The optimization problem is complicated by numerical artifacts resulting from discrete approximation and finite computer precision, as well as theoretical difficulties associated with navigating to a global minimum on a nonconvex objective function having a fold and several local minima. In this paper we report on the solution of the optimization problem, discuss implications of some of the features of this problem on selection of a suitable and efficient optimization algorithm, and share lessons learned, fixes implemented, and research issues identified along the way.