Abstract:Despite rapid improvements in the performance of central processing unit (CPU), the calculation cost of simulating chemically reacting flow using CFD remains infeasible in many cases. The application of the convolutional neural networks (CNNs) specialized in image processing in flow field prediction has been studied, but the need to develop a neural netweork design fitted for CFD is recently emerged. In this study, a neural network model introducing the finite volume method (FVM) with a unique network architecture and physics-informed loss function was developed to accelerate CFD simulations. The developed network model, considering the nature of the CFD flow field where the identical governing equations are applied to all grids, can predict the future fields with only two previous fields unlike the CNNs requiring many field images (>10,000). The performance of this baseline model was evaluated using CFD time series data from non-reacting flow and reacting flow simulation; counterflow and hydrogen flame with 20 detailed chemistries. Consequently, we demonstrated that (1) the FVM-based network architecture provided improved accuracy of multistep time series prediction compared to the previous MLP model (2) the physic-informed loss function prevented non-physical overfitting problem and ultimately reduced the error in time series prediction (3) observing the calculated residuals in an unsupervised manner could indirectly estimate the network accuracy. Additionally, under the reacting flow dataset, the computational speed of this network model was measured to be about 10 times faster than that of the CFD solver.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: although the performance of central processing units (CPUs) has increased rapidly, the computational cost of using computational fluid dynamics (CFD) to simulate chemical reaction flows (especially complex reaction flows such as hydrogen combustion) is still too high, making it difficult to achieve in many cases. In order to accelerate CFD simulations and reduce computational costs, the author proposes a neural network model based on the finite volume method (FVM), namely FVMN, aiming to improve existing methods and increase prediction accuracy and speed. Specifically, the paper mainly focuses on the following points: 1. **Developing a new network architecture**: The basic principles of FVM are introduced, and a unique neural network architecture is designed, which can handle CFD problems more effectively. Compared with traditional convolutional neural networks (CNNs), this architecture only requires data from two previous time steps for prediction, instead of a large amount of image data (> 10,000 images). 2. **Physics - informed loss function**: A physics - informed loss function is introduced to prevent non - physical over - fitting problems and ultimately reduce errors in time - series prediction. 3. **Accuracy of multi - step time - series prediction**: By applying FVM principles to network design, the accuracy of multi - step time - series prediction is improved. Compared with the previous multi - layer perceptron (MLP) model, FVMN shows higher accuracy when predicting future fields. 4. **Computational speed improvement**: Tests on reaction - flow datasets show that the computational speed of the proposed network model is approximately 10 times faster than that of traditional CFD solvers. ### Key innovation points of the paper - **Combining FVM and ML**: The core idea of FVM is incorporated into the design of the neural network, enabling the model to better capture the physical characteristics in fluid dynamics. - **Reducing the need for training data**: By considering the interactions between adjacent grids, FVMN can work effectively with only a small amount of training data, thereby reducing training costs. - **Improving prediction stability**: By introducing a physics - informed loss function, the physical rationality of the model's prediction results is ensured, and the over - fitting problem is avoided. ### Summary This research aims to develop an efficient neural network model by combining FVM and machine - learning techniques to accelerate CFD simulations and reduce computational costs. The experimental results show that the proposed FVMN performs excellently on both non - reaction - flow and reaction - flow datasets, not only improving prediction accuracy but also significantly increasing computational speed.

Finite volume method network for acceleration of unsteady computational fluid dynamics: non-reacting and reacting flows

Finite Volume Graph Network(FVGN): Predicting unsteady incompressible fluid dynamics with finite volume informed neural network

Finite Volume Features, Global Geometry Representations, and Residual Training for Deep Learning-based CFD Simulation

Predicting unsteady incompressible fluid dynamics with finite volume informed neural network

A scalable convolutional neural network approach to fluid flow prediction in complex environments

Flow Completion Network: Inferring the Fluid Dynamics from Incomplete Flow Information using Graph Neural Networks

DeepCFD: Efficient Steady-State Laminar Flow Approximation with Deep Convolutional Neural Networks

Finite-difference-informed graph network for solving steady-state incompressible flows on block-structured grids

Developing an advanced neural network and physics solver coupled framework for accelerating flow field simulations

ML-LBM: Predicting and Accelerating Steady State Flow Simulation in Porous Media with Convolutional Neural Networks

A deep-learning model for predicting spatiotemporal evolution in reactive fluidized bed reactor

Adaptive Neural Network-Based Approximation to Accelerate Eulerian Fluid Simulation

Computationally Efficient Multiscale Neural Networks Applied to Fluid Flow in Complex 3D Porous Media

Accuracy Improvement Technique of DNN for Accelerating CFD Simulator

Performance and accuracy assessments of an incompressible fluid solver coupled with a deep Convolutional Neural Network

Learning-Based High-Performance Algorithm for Long-Term Motion Prediction of Fluid Flows

MH-DCNet: an Improved Flow Field Prediction Framework Coupling Neural Network with Physics Solver

Residual-based physics-informed transfer learning: A hybrid method for accelerating long-term CFD simulations via deep learning

Nonlinear constitutive calculation method of rarefied flow based on deep convolution neural networks

A Flow Feature Detection Method for Modeling Pressure Distribution Around a Cylinder in Non-Uniform Flows by Using a Convolutional Neural Network

Physics-constrained and flow-field-message-informed graph neural network for solving unsteady compressible flows