System Identification and Prediction in Radiative Heat Transfer Using Dynamic Mode Decomposition

Abstract

Transient heat transfer processes that are abundant in industry, can be highly non-linear, and complex. Although numerical models can be developed for the processes, the models can be slow, and computationally expensive. Most often, such models are developed in commercial software, rendering the final assembled equations inaccessible to the user. These models may need to be simplified and reduced in order to be utilized in real time applications such as model based controls. Here, a non-solver intrusive data-driven system identification and model order reduction method for practical radiative heat transfer problems is presented for real-time predictions. The method uses a novel modification to the technique of dynamic mode decomposition with controls. The underlying data for system identification could be from either experiments or numerical simulations. In this demonstrative study, the data was generated from a numerical simulation, which represented heat transfer in a lumped thermal mass network. The typical dynamic mode decomposition formulation was augmented with polynomial terms to better identify the non-linear form of the equations governing radiative heat transfer. The extracted system was then used to predict results with different initial and boundary conditions. The predicted results were compared with data generated from the numerical simulation.