In this research, Advanced Finite Element Methods (FEM) for engineering applications’ fluid dynamics problem solving are investigated, where high order discretization, hybrid solvers and AI enhanced FEM models are studied. SUPG, VMS and DG methods were analyzed in terms of their accuracy and efficiency of computation. We calculated about 35% reduction in numerical diffusion using SUPG vis a vis conventional FEM carried out at Reynolds numbers Re = 10⁵ to 10⁷. Turbulence modeling accuracy was enhanced by 28% with VMS approach, whereas 42% increased shock capturing capability have been realized with DG-FEM. Moreover, the computational time was reduced to 50% by integrating physics informed neural networks (PINNs) at a 97% accuracy level. The research also validated hybrid mesh free FEM approaches that allow 15% higher efficiency in fluid structure interaction problems. The fact that we can implement these findings in the context of real time engineering applications using AI enhanced FEM techniques indicates that these are strong candidates. Finally, the study shows that the advanced FEM methods are more accurate, more stable and far more computationally efficient than traditional models. FEM solvers accelerated by GPU and adaptive meshing strategies should be explored in future research for further optimality.
9 August 2024 in Case Studies in Thermal Engineering
This research investigates the application of advanced Computational Fluid Dynamics (CFD) techniques in the mathematical modeling of turbulent flows. The accuracy is improved in different engineering systems by using different turbulence models: the k-ε, k-ω, LES and DNS. The main purpose was to investigate fluid flow behavior in industrial, aerospace, biomedical applications. Results of the experimental data are compared with the k-ε model and show that the average deviation is 5.2% when compared to the experimental data of turbulent boundary layers. Dynamic vortex simulations using the LES model resulted in a 12% decrease in error compared to original results. In addition, the DNS model was close to obtaining near exact solutions at the price of greatly increased CPU costs, with a deviation of 1.4% in highly turbulent cases. It was found next that the integration of machine learning algorithmss further improved model predictions, reducing the required simulation times by about 18 percent. It is concluded that for different applications it is important to select the correct turbulence model so as to balance accuracy and computational efficiency. The contribution of this research to refine CFD methodologies and to provide some initial insight in optimizing fluid dynamics simulations for real world engineering problems is provided.
8 January 2024 in Journal of Advanced Research in Fluid Mechanics and Thermal Sciences
In the current numerical simulation, the thermal transfer of viscous incompressible time dependent MHD free convective flow of hybrid nanofluid (Cu-TiO2-H2O) over a vertical porous plate with viscous dissipation and thermal radiation is examined. The system of partial differential equations which controls the fluid flow are transmuted into dimensionless equations and are solved using numerical simulation namely Galerkin Finite Element Method. The momentum and thermal consequences, coefficient of heat transfer and skin friction with Gr, M, Ec, N, t, ωt, δ2, λ – parameters for both nanofluid (Cu-H2O) and hybrid nanofluid (Cu-TiO2-H2O) flows are exposed through graphs and tables. It is revealed that heat transfer coefficient of Cu-TiO2-H2O is more effective when compared with Cu-H2O flows varying with aforementioned parameters. The current numerical computation is in good agreement with the analytical solution of the problem. This study is used in several engineering fields such as electronic cooling, heat exchangers, nanomaterial processing, environmental engineering, chemical industry.