Inter-model comparison of CFD and neural network analysis of natural convection heat transfer in a partitioned enclosure

Abstract

CFD analysis of heat and mass flow due to natural convection in partitioned enclosures has recently been the focus of many CFD researchers. In some cases, it was reported in the literature that different CFD solutions (due to different numerical stability characteristics) were obtained for different mesh quality, time step, and discretization order. The objective of this paper is to investigate the feasibility of using neural networks (NNs) as a means lending support to the authenticity of steady-state CFD solutions for such ill-posed problems through inter-model comparisons. Attention is focused on using NNs trained on a database generated by numerically-stable CFD analysis to predict flow variables for the aforementioned ill-posed cases, thereby giving confidence in steady-state CFD results for these cases. Three types of NNs were evaluated and parametric studies were performed to optimize network designs for best predictions of the flow variables. A validated CFD code was used to generate a database that covered the range of Rayleigh number Ra from 10 4 to 4.7 × 10 6 . The CFD analysis was used to calculate the normalized temperature θ and stream function ψ throughout a partitioned enclosure. The results of the CFD were used to train and test the neural networks. The robustness of the trained NNs was tested by applying them to a number of “production” data sets that the networks have never “seen” before. The trained NNs were used to verify CFD solutions in cases of higher Ra ranges for which CFD simulations produce different solutions for different discretization schemes. The results showed that it is possible to design and train certain architectures of NNs to accurately predict θ and ψ distribution within the enclosure, and hence impart confidence in the legitimacy of CFD solutions for new cases.