AI-driven simulation of chemically reactive non-Darcian Casson nanomaterial flow with quadratic convection and variable thermo-solutal characteristics

Simulation of complex non-Newtonian nanofluid systems such as porous media, nonlinear convection and coupled heat-mass transport are still a daunting problem in thermal and fluid engineering. Traditional numerical methods have been known to experience challenges with strong nonlinearities, stiffness and multi-physical interactions. The recent developments in machine learning, specifically the Artificial Neural Networks (ANNs), have proven to have a high potential of predicting nonlinear transport through a cost-reduced calculation. The flow behavior in viscoplastic nanofluids is further complicated by the existence of porous structure, inertia effect and non-classical thermal and solutal diffusion mechanisms. Embarking on smart data-driven schemes together with the conventional numerical solvers would provide a promising avenue of improving the quality of predictions and the strength of models in these complicated thermo-fluid systems. This research will examine the nonlinear mixed convection flow of viscoplastic Casson nanofluids through a vertical stretching porous surface due to the effects of porous medium resistance and inertia. It aims to examine the interaction of the major physical variables such as porosity, inertia coefficient, Casson fluid parameter, Brownian motion, thermophoresis, nonlinear thermal and solutal convection, on the momentum, heat and mass transfer properties. The paper further aims to identify the effectiveness and accuracy of the Artificial Neural Networks that have been trained using the Levenberg Marquardt Scheme (ANN LMS) when forecasting complicated transport behavior as compared to the traditional numerical solutions. The nonlinear partial differential equations that govern the transport of momentum, energy, and the concentration of nanoparticles are developed based on the two component nanofluid model developed by Buongiorno, Cattaneo-Christov non-Fourier and non-Fickian non-Fourier and non-Fickian nonlinear partial differential equations. By means of suitable similarity transformations, the system is brought to a group of coupled nonlinear ordinary differential equations, which are solved numerically with the MATLAB-based bvp4c solver to provide reference datasets. These data sets are used to train, test and validate the ANN-LMS model in a high range of governing parameters. Mean square error, error histograms, and regression are used to determine the ability of the neural network to predict. The graphical and statistical analyses are also provided to demonstrate the implications of the important parameters on velocity, temperature, concentration profiles, and engineering quantities like the skin friction. The ANN-LMS framework proposed in the research shows a high level of accuracy with absolute error values of 10 − 9 to 10 − 11 , showing that the ANN-LMS framework can be used to predict complex non-Newtonian nanofluid transport phenomena with a high degree of precision.