Design of neural network based intelligent computing for neumerical treatment of unsteady 3D flow of Eyring-Powell magneto-nanofluidic model

In the presented study, a novel application of intelligent numerical computing solver based on neural networks backpropagated with the Levenberg-Marquard scheme (NN-BLMS) is presented to interpret the chemical reactions and activation energy in unsteady 3D flow of Eyring-Powell magneto-nanofluidic system for convective heat and mass flux scenarios. The original nonlinear coupled PDEs representing the Eyring-Powell magneto-nanofluidic model (EPMNM) is transformed to an equivalent nonlinear ODEs system by exploiting similarity variables. A dataset for the proposed NN-BLMS is generated for different scenarios of EPMNM by variation of radiation, temperature ratio parameter, heat generation, Brownian motion and thermophoresis parameters by using Adam numerical method. The training, testing, and validation processes of NN-BLMS are performed to determine the approximate solution of EPMNM for different cases and comparison with reference results to verify the correctness of the proposed NN-BLMS. The performance of the proposed NN-BLMS to effectively solve the EPMNM is endorsed through mean squared error, histogram studies and regression analysis. The close matching of the proposed and reference results based on error analysis form level 10(-05) to 10(-07) validates the correctness of the proposed methodology. (C) 2020 The Author(s). Published by Elsevier B.V.