POD‐RBF Based Non‐Intrusive Reduced Order Model for Multi‐Physics Problems

ABSTRACT This paper essentially depicts a methodology to address computational frugality of multi‐physics problem. Reduced‐order modelling (ROM) is popular in engineering as it has the potential to gain CPU cost for simulations with different material parameters and/or different initial or boundary conditions. The basic idea in this article is to provide a nonintrusive setup for the ROM conducive for industrial problems using commercial software. The problem dealt here is a classical multiphase thermo‐fluid problem, although the philosophy can be extended for other multi‐physics problems. The first step is to decouple the involved physics, and thereafter a nonintrusive ROM is used through response surface method like neural networks to replace the full‐order simulation of the most expensive physics. This essentially provides reduction in computational cost for the problem studied. The reduced order bases thus generated are also used to compute separate problems with altered parameters or loading conditions thereby eliminating the necessity of computations of full‐order problems.