Deep-learning accelerated calculation of real-fluid properties in numerical simulation of complex flowfields

Milan, Petro Junior; Hickey, Jean-Pierre; Wang, Xingjian; Yang, Vigor

doi:10.1016/j.jcp.2021.110567

Cited by 20 publications

(4 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Machine learning techniques are a promising tool to develop computationally efficient surrogate thermodynamic models. Milan et al (2021) performed a first comprehensive study on the use of a deep feed-forward neural network with appropriate boundary information to calculate the thermophysical properties of a multicomponent mixture for the simulation of supercritical fluid injection. Finally, another emerging technique is based on approximating fluid thermodynamic properties with Chebyshev expansions (Bell & Alpert 2021).…”

Section: Accuracy Of Ceosmentioning

confidence: 99%

Nonideal Compressible Fluid Dynamics of Dense Vapors and Supercritical Fluids

Guardone,

Colonna,

Pini

et al. 2024

Annu. Rev. Fluid Mech.

View full text Add to dashboard Cite

The gas dynamics of single-phase nonreacting fluids whose thermodynamic states are close to vapor-liquid saturation, close to the vapor-liquid critical point, or in supercritical conditions differs quantitatively and qualitatively from the textbook gas dynamics of dilute, ideal gases. Due to nonideal fluid thermodynamic properties, unconventional gas dynamic effects are possible, including nonclassical rarefaction shock waves and the nonmonotonic variation of the Mach number along steady isentropic expansions. This review provides a comprehensive theoretical framework of the fundamentals of nonideal compressible fluid dynamics (NICFD). The relation between nonideal gas dynamics and the complexity of the fluid molecules is clarified. The theoretical, numerical, and experimental tools currently employed to investigate NICFD flows and related applications are reviewed, followed by an overview of industrial processes involving NICFD, ranging from organic Rankine and supercritical CO2 cycle power systems to supercritical processes. The future challenges facing researchers in the field are briefly outlined.

show abstract

Section: Accuracy Of Ceosmentioning

confidence: 99%

Nonideal Compressible Fluid Dynamics of Dense Vapors and Supercritical Fluids

Guardone,

Colonna,

Pini

et al. 2024

Annu. Rev. Fluid Mech.

View full text Add to dashboard Cite

show abstract

“…The neural network approach allows for the approximation of any continuous function, making it a versatile tool. Advanced neural networks like physics-informed machine learning (PIML) and deep forwarded neural networks-boundary conditions (DFNN-BC) [9] can provide new perspectives on fluid flow through complex geometries [10]. Raissi et al [11] utilized PIML to solve various computational science problems, including Burgers' equation and Navier-Stokes.…”

Section: Developing Physics-informed Machine Learning (Piml) For Turb...mentioning

confidence: 99%

Developing physics-informed machine learning (PIML) for turbulent flow based on transient training data set: A case study on flow passing through the pore-scale porous media (PSPM)

Mesgarpour,

Mahian,

Shadloo

et al. 2024

The 12TH INTERNATIONAL CONFERENCE ON MECHANICAL ENGINEERING (TSME-ICoME 2022)

View full text Add to dashboard Cite

We propose a novel technique that combines physics-informed machine learning (PIML) with the wall-adapting local eddy viscosity model for predicting flow patterns over time. In order to generate loss functions for turbulence flow and calculate dissipation rates within the domain, a new form of machine learning-based WALE-LES model includes updating the boundary matrix and tensor of component velocity and pressure tensors over time. By utilizing boundary conditions, Navier-Stokes-based equations will be discretized via automatic differentiation in PIML. We present a novel point cloud-based method for pore-scale porous media (PSPM) for the first time. This method transforms complex geometry into a matrix of data based on a point cloud. As a result of this method, boundary conditions are included in transient calculations. We modified the standard form of momentum equations to obtain the loss function for flows passing through a PSPM. According to the findings, high-fidelity numerical simulation and PIML prediction results are very similar in pressure and velocity. Additionally, this method can reduce calculation costs by up to 41% compared to the standard method.

show abstract

“…In response to these challenges, a deep-learning based approach, termed deep feedforward neural network (NN) with boundary conditions, has been developed for the efficient evaluation of thermophysical properties in complex real-fluid flows [14]. This approach replaces the direct calculation of the EoS with a NN trained with appropriate boundary information, significantly improving computational efficiency.…”

Section: Introductionmentioning

confidence: 99%

Computation of Real-Fluid Thermophysical Properties Using a Neural Network Approach Implemented in OpenFOAM

Sahranavardfard,

Aubagnac-Karkar,

Costante

et al. 2024

Fluids

View full text Add to dashboard Cite

Machine learning based on neural networks facilitates data-driven techniques for handling large amounts of data, either obtained through experiments or simulations at multiple spatio-temporal scales, thereby finding the hidden patterns underlying these data and promoting efficient research methods. The main purpose of this paper is to extend the capabilities of a new solver called realFluidReactingNNFoam, under development at the University of Perugia, in OpenFOAM with a neural network algorithm for replacing complex real-fluid thermophysical property evaluations, using the approach of coupling OpenFOAM and Python-trained neural network models. Currently, neural network models are trained against data generated using the Peng–Robinson equation of state assuming a mixture’s frozen temperature. The OpenFOAM solver, where needed, calls the neural network models in each grid cell with appropriate inputs, and the returned results are used and stored in suitable OpenFOAM data structures. Such inference for thermophysical properties is achieved via the “Neural Network Inference in C made Easy (NNICE)” library, which proved to be very efficient and robust. The overall model is validated considering a liquid-rocket benchmark comprised of liquid-oxygen (LOX) and gaseous-hydrogen (GH2) streams. The model accounts for real-fluid thermodynamics and transport properties, making use of the Peng–Robinson equation of state and the Chung transport model. First, the development of a real-fluid model with an artificial neural network is described in detail. Then, the numerical results of the transcritical mixing layer (LOX/GH2) benchmark are presented and analyzed in terms of accuracy and computational efficiency. The results of the overall implementation indicate that the combined OpenFOAM and machine learning approach provides a speed-up factor higher than seven, while preserving the original solver accuracy.

show abstract

Deep-learning accelerated calculation of real-fluid properties in numerical simulation of complex flowfields

Cited by 20 publications

References 44 publications

Nonideal Compressible Fluid Dynamics of Dense Vapors and Supercritical Fluids

Nonideal Compressible Fluid Dynamics of Dense Vapors and Supercritical Fluids

Developing physics-informed machine learning (PIML) for turbulent flow based on transient training data set: A case study on flow passing through the pore-scale porous media (PSPM)

Computation of Real-Fluid Thermophysical Properties Using a Neural Network Approach Implemented in OpenFOAM

Contact Info

Product

Resources

About