A hybrid optimization algorithm for the thermal design of radiant paint cure ovens

Continuous paint cure ovens have many important industrial applications. In particular, convection ovens are extensively used in auto industries. Radiation paint cure ovens have attractive features as well and attempts have been made to design the oven and the radiation panels such that the moving loads experience desirable, nearly uniform, heating process. Due to the motion of the load and the variation of the radiation exchange factors during the curing process, the solution of this design problem corresponds to the solution of a dynamic optimization problem. This is computationally demanding in a realistic three-dimensional case and the computational cost needs to be minimized. Two-dimensional test problems provide opportunities for algorithm development and quick evaluation. This paper focuses on the convergence acceleration of this thermal optimization algorithm for a 2D test problem. By combining the features of an optimization algorithm with the capabilities of the neural network method, a hybrid design algorithm is obtained which is considerably faster than the original algorithm. It is shown that by employing a neural network trained by a simplified physical model, the computational cost can be reduced close to an order of magnitude without significant loss of accuracy.