Finite volume method network for the acceleration of unsteady computational fluid dynamics: Non?reacting and reacting flows
نویسندگان
چکیده
Despite rapid improvements in the performance of central processing unit (CPU), calculation cost simulating chemically reacting flow using CFD remains infeasible many cases. The application convolutional neural networks (CNNs) specialized image field prediction has been studied, but need to develop a network design fitted for recently emerged. In this study, model introducing finite volume method (FVM) with unique architecture and physics-informed loss function was developed accelerate simulations. model, considering nature where identical governing equations are applied all grids, can predict future fields only two previous unlike CNNs requiring images (>10 000). baseline evaluated time series data from non-reacting simulation; counterflow hydrogen flame 20 detailed chemistries. Consequently, we demonstrated that (a) FVM-based provided significantly improved accuracy multistep compared MLP (b) physic-informed prevented non-physical overfitting problem ultimately reduced error (c) observing calculated residuals an unsupervised manner could monitor accuracy. Additionally, under dataset, computational speed measured be about 10 times faster than solver.
منابع مشابه
Hydrodynamic Improvement of underwater glider by Computational Fluid Dynamics method
Gliders are new marine vehicles which have research and military uses and they move by sequent diving and climbing. Suitable design of its main body and wings are important for the most advance velocity. hydrodynamic design variables are main body form, wings (cross section, dimensions, shape, longitudinal and vertical position) and hydrostatic parameters (static trim angle, amount of added for...
متن کاملbuckling of viscoelastic composite plates using the finite strip method
در سال های اخیر، تقاضای استفاده از تئوری خطی ویسکوالاستیسیته بیشتر شده است. با افزایش استفاده از کامپوزیت های پیشرفته در صنایع هوایی و همچنین استفاده روزافزون از مواد پلیمری، اهمیت روش های دقیق طراحی و تحلیل چنین ساختارهایی بیشتر شده است. این مواد جدید از خودشان رفتارهای مکانیکی ارائه می دهند که با تئوری های الاستیسیته و ویسکوزیته، نمی توان آن ها را توصیف کرد. این مواد، خواص ویسکوالاستیک دارند....
Unsteady-state Computational Fluid Dynamics Modeling of Hydrogen Separation from H2/N2 Mixture
3D modeling of Pd/α-Al2O3 hollow fiber membrane by using computational fluid dynamic for hydrogen separation from H2/N2 mixture was considered in steady and unsteady states by using the concept of characteristic time. Characteristic time concept could help us to design and calculate surface to volume ratio and membrane thickness, and adjust the feed conditions. The contribution of resistance be...
متن کاملMULTI PHASE COMPUTATIONAL FLUID DYNAMICS MODELING OF CAVITATING FLOWS OVER AXISYMMETRIC HEAD-FORMS
In the present paper, partial cavitation over various head-forms was studied numerically to predict the shape of the cavity. Navier-Stokes equations in addition to an advection equation for vapor volume fraction were solved. Mass transfer between the phases was modeled by a sink term in vapor equation in the numerical analysis for different geometries in wide range of cavitation numbers. The r...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: International Journal of Energy Research
سال: 2022
ISSN: ['0363-907X', '1099-114X']
DOI: https://doi.org/10.1002/er.7879