Prediction Of Flow Field Research Articles

Flow-Field Prediction in Periodic Domains Using a Convolution Neural Network with Hypernetwork Parametrization

This paper deals with flow field prediction in a blade cascade using the convolution neural network. The convolutional neural network (CNN) predicts density, pressure and velocity fields based on the given geometry. The blade cascade is modeled as a single interblade channel with periodic boundary conditions. In this paper, an algorithm that enforces periodic boundary conditions onto the CNN is presented. The main target of this study is to parametrize the CNN model depending on the Reynolds number. A new parametrization approach based on a so-called hypernetwork is employed for this purpose. The idea of this approach is that when the Reynolds number is modified, the hypernetwork modifies the weights of the CNN in such a way that it produces flow fields corresponding to that particular Reynolds number. The concept of the hypernetwork-based parametrization is tested on the problem of a compressible fluid flow through a blade cascade with variable blade profiles and Reynolds numbers.

A Digital Twin Framework Embedded with POD and Neural Network for Flow Field Monitoring of Push-Plate Kiln

The push-plate kiln is a kind of kiln equipment widely used in the oxygen-free sintering of high-temperature alloy materials. Its flow field monitoring has an important application value for the manufacturing industry. However, traditional simulation methods cannot meet the requirements of real-time applications due to the high computational cost and being time-consuming. The rapid development of artificial intelligence technology will empower the traditional manufacturing industry. In this paper, we propose a data-driven digital twin framework for real-time flow field prediction by combining the CFD modeling simulation, IoT, and deep learning technologies. The framework integrates geometric, rule, physical, and neural network models to achieve the real-time simulation of physical and twin objects. The proper orthogonal decomposition (POD) and multiscale convolutional neural network (MCNN) are innovatively embedded into the framework. The POD is used to map high-dimensional data to low-dimensional features, and the MCNN is used to construct models predicting low-dimensional features for fast flow field prediction. The effectiveness of the proposed model is verified by the push-plate kiln case. The results show that the digital twin can quickly predict multi-physics fields based on the perceptual data to achieve the real-time evaluation of the operating state of the push-plate kiln.

Turbulence Modeling for Physics-Informed Neural Networks: Comparison of Different RANS Models for the Backward-Facing Step Flow

Physics-informed neural networks (PINN) can be used to predict flow fields with a minimum of simulated or measured training data. As most technical flows are turbulent, PINNs based on the Reynolds-averaged Navier–Stokes (RANS) equations incorporating a turbulence model are needed. Several studies demonstrated the capability of PINNs to solve the Naver–Stokes equations for laminar flows. However, little work has been published concerning the application of PINNs to solve the RANS equations for turbulent flows. This study applied a RANS-based PINN approach to a backward-facing step flow at a Reynolds number of 5100. The standard k-ω model, the mixing length model, an equation-free νt and an equation-free pseudo-Reynolds stress model were applied. The results compared favorably to DNS data when provided with three vertical lines of labeled training data. For five lines of training data, all models predicted the separated shear layer and the associated vortex more accurately.

Research on the collision model of high-temperature alumina droplets with cold wall for solid rocket motors

The study on the adhesion and rebound behavior of high-temperature alumina droplets impacting on the cold wall is very meaningful for the accurate prediction of slag deposition and internal flow field in solid rocket motors. In this paper, based on high-temperature alumina droplet impact experimental system, the experiment of alumina droplet impacting on the cold wall was carried out, and the temperature response of alumina droplet during the impact is obtained by the colorimetric thermometry. The study shows that the high-temperature alumina droplets impacting on the cold wall in a certain range of Oh number and Re number exhibit two kinds of behaviors: rebound and adhesion; based on experiments, the distribution laws of rebound and adhesion behaviors of alumina droplets impacting the cold wall are obtained; based on the energy conservation model, and introducing the heat transfer model between the high-temperature droplet and the cold wall, the maximum spreading coefficient model of alumina droplet impact on the cold wall and the prediction model of rebound and adhesion results were established, and the maximum spreading coefficient model prediction results were in good agreement with the experimental data.

Experimental Study of Unsteady Drag Coefficient of Droplets in a Liquid–Liquid System

In liquid–liquid contact process, the motion of droplets relative to the surrounding fluid always involves accelerating and decelerating, which affects the mass, heat, and momentum transfer. The lack of experimental data of the unsteady drag coefficient has been one of the limitations on the prediction of the unsteady flow field. In this study, the accelerated and decelerated water droplets in an organic phase were measured by a high-speed camera. The results show that with a decrease in droplet diameter, the acceleration becomes more significant, while the absolute relative velocity decreases, causing a lower Reynolds number. The Basset force and add mass force were solved numerically and compared with drag force. The unsteady drag coefficient is always smaller than the corresponding steady drag coefficient in the case of accelerating relative flow and larger than that in the case of decelerating relative flow. A new unsteady drag coefficient model has been established, which has an acceptable agreement with the experimental data.

An efficient two-stage hybrid framework to evaluate vortex-induced vibration for bridge deck based on divergent vibration

High-fidelity CFD (computational fluid dynamics) numerical modeling offers an attractive alternative of VIV investigation for bridge deck. Conventional strategy to perform VIV prediction using CFD is to directly characterize the two-way coupled free vibration of structure and flow which could become dramatically expensive in computational cost. To bridge the gap, this study presents an efficient two-stage hybrid approach that combines a divergent vibration simulation for fast aerodynamic damping extraction and a nonlinear model to predict the VIV. A negative structural damping is imposed in the motion equation to intentionally create a divergent vibration which significantly accelerates the growth of motion and takes less than 1/3 of the elapsed time compared with a conventional VIV simulation whereas maintains reasonable accuracy in aerodynamic identification. A Π-shaped bridge deck is employed as a demonstration of the proposed approach. The modeled steady-state VIV performance is well validated against wind tunnel tests. The effect of negative damping on instantaneous flow field, surface pressure and amplitude prediction are discussed and the optimal damping ratio achieving balanced accuracy and efficiency is suggested. The proposed framework shows favorable efficiency while simultaneously retains satisfactory accuracy and is a potential numerical alternative of VIV investigation for bridge deck.

Numerical visualization of extensional flows in injection molding of polymer melts

Abstract Extensional flows generally take place in a channel with varied cross-sectional areas. During polymer processing with a variety of complex geometric features, it is difficult to separate extensional rates from shear rates in state-of-the-art predictive engineering tools of computational fluid dynamics. The recently proposed method of Tseng [Tseng, H.-C., “A Revisitation of Generalized Newtonian Fluids,” J Rheol 64 493–504 (2020)] decomposed the generalized strain rate as the characteristic shear and extensional rates via the rate-of-deformation tensor rotated along streamline coordinates. As validation for an isothermal center-gated disk flow, the predicted flow field profiles fairly matched the analytical solution for Newtonian fluid. Under injection molding simulations, an objective indicator is defined to visualize the colorful contours of extensional flows encountered in the gate-vicinity and the mid-plane of the cavity’s thickness direction, as well as contraction-expansion channels.

A Hybrid Experimental-Computational Study: Prediction of Flow Fields and Full-Field Pressure Distributions on Measured Shapes of Three-Tab Asphalt Roofing Shingles Subjected to Hurricane Velocity Winds

A Hybrid Experimental-Computational Study: Prediction of Flow Fields and Full-Field Pressure Distributions on Measured Shapes of Three-Tab Asphalt Roofing Shingles Subjected to Hurricane Velocity Winds

Fast Prediction of Two-Dimensional Flowfields with Fuel Injection into Supersonic Crossflow via Deep Learning

Fast Prediction of Two-Dimensional Flowfields with Fuel Injection into Supersonic Crossflow via Deep Learning

Unsteady-state turbulent flow field predictions with a convolutional autoencoder architecture

<abstract> <p>Traditional numerical methods, such as computational fluid dynamics (CFD), demand large computational resources and memory for modeling fluid dynamic systems. Hence, deep learning (DL) and, specifically Convolutional Neural Networks (CNN) autoencoders have resulted in accurate tools to obtain approximations of the streamwise and vertical velocities and pressure fields, when stationary flows are considered. The novelty of this paper consists of predicting the future instants from an initial one with a CNN autoencoder architecture when an unsteady flow is considered. Two neural models are proposed: The former predicts the future instants on the basis of an initial sample and the latter approximates the initial sample. The inputs of the CNNs take the signed distance function (SDF) and the flow region channel (FRC), and, for the representation of the temporal evolution, the previous CFD sample is added. To increment the amount of training data of the second neural model, a data augmentation technique based on the similarity principle for fluid dynamics is implemented. As a result, low absolute error rates are obtained in the prediction of the first samples near the shapes surfaces. Even in the most advanced time instants, the prediction of the vortices zone is quite reliable. 62.12 and 9000 speed-up ratios are achieved by the predictions of the first and second neural models, respectively, compared to the computational cost regarded by the CFD simulations.</p> </abstract>

Modified UNet with attention gate and dense skip connection for flow field information prediction with porous media

Porous media flow phenomena are commonly found in nature and industry. Understanding the patterns can improve industrial production and efficiency. Traditional numerical methods can predict flow field information within porous media exactly, even on pore scale. However, once the pore structures of porous media are changed, the traditional numerical methods have to start over again, which implies poor extension and low efficiency for realistic application. To address the challenge, we propose a new deep learning method for image segmentation based on convolutional neural networks that can quickly predict the saturation variation and flow velocity profile of multiphase flow directly from the structure of porous media. In the present method, the UNet network structure is improved by replacing parallel splicing with different levels of dense skip connection to extract flow information at different depths. When processing flow topological information, the introduction of attention mechanism allows effective focus on adjusting the flow field edges and front shapes to improve accuracy. Depending on the pore structures information, the saturation of the porous media also can be evaluated. This result shows that the flow field information (saturation and velocity) of porous media can be quickly predicted by deep learning methods. Compared with the computational fluid dynamics (CFD) method, this approach has a better prediction efficiency for porous media. The machine learning method is a good prediction tool for porous media with different porosities and pore structures.

Machine Learning-Enabled Prediction of Transient Injection Map in Automotive Injectors With Uncertainty Quantification

AbstractAccurate prediction of injection profiles is a critical aspect of linking injector operation with engine performance and emissions. However, highly resolved injector simulations can take one to two weeks of wall-clock time, which is incompatible with engine design cycles with desired turnaround times of less than a day. Hence, it is important to reduce the time-to-solution of the internal flow simulations by several orders of magnitude to make it compatible with engine simulations. This work demonstrates a data-driven approach for tackling the computational overhead of injector simulations, whereby the transient injection profiles are emulated for a side-oriented, single-hole diesel injector using a Bayesian machine-learning framework. First, an interpretable Bayesian learning strategy was employed to understand the effect of design parameters on the total void fraction field. Then, autoencoders are utilized for efficient dimensionality reduction of the flowfields. Gaussian process models are finally used to predict the spatiotemporal void fraction field at the injector exit for unknown operating conditions. The Gaussian process models produce principled uncertainty estimates associated with the emulated flowfields, which provide the engine designer with valuable information of where the data-driven predictions can be trusted in the design space. The Bayesian flowfield predictions are compared with the corresponding predictions from a deep neural network, which has been transfer-learned from static needle simulations from a previous work by the authors. The emulation framework can predict the void fraction field at the exit of the orifice within a few seconds, thus achieving a speed-up factor of up to 38 × 106 over the traditional simulation-based approach of generating transient injection maps.

Multi-fidelity model and reduced-order method for comprehensive hydrodynamic performance optimization and prediction of JBC ship

The multi-fidelity Co-Kriging surrogate model can be applied to combine the accuracy advantage of high-fidelity sample evaluation with the efficiency advantage of low-fidelity sample evaluation. In this paper, the hull form optimization process for resistance and wake performance of a Japan bulk carrier (JBC) hull at design speed is given in detail, where the evaluation results from medium and coarse grids are regarded as the high- and low-fidelity data, respectively. 60 high-fidelity hydrodynamic evaluations have been done to construct the Kriging model, while the Co-Kriging model uses 30 high-fidelity and 60 low-fidelity evaluations with a 25% reduction of the total computation time. The optimization results show that, for the total drag, the Kriging-based optimal hull has a 4.57% reduction, while the Co-Kriging-based optimal hull has 5.67%; for the axial wake fraction reduction at the propeller disk, the Kriging-based optimal hull has a 7.20% reduction, while the Co-Kriging-based optimal hull has 10.37%. Furthermore, in the latter stage of hull form optimization, dimensionality reduction field learning can be performed to fully use the viscous-flow-based calculation results. An accurate and efficient viscous-flow-based wake field learning method is proposed based on the Kriging model and the Proper Orthogonal Decomposition (POD) method with qualitative and quantitative error analysis, which can guide the sensitivity analysis of the design variables, the selection of the design variables and spaces, and new flow field prediction for comprehensive hull form performance optimization.

Fast Prediction of Flow Field around Airfoils Based on Deep Convolutional Neural Network

We propose a steady-state aerodynamic data-driven method to predict the incompressible flow around airfoils of NACA (National Advisory Committee for Aeronautics) 0012-series. Using the Signed Distance Function (SDF) to parameterize the geometric and flow condition setups, the prediction core of the method is constructed essentially by a consecutive framework of a convolutional neural network (CNN) and a deconvolutional neural network (DCNN). Impact of training parameters on the behavior of the proposed CNN-DCNN model is studied, so that appropriate learning rate, mini-batch size, and random deactivation rate are specified. Tested by “unseen” airfoil geometries and far-field velocities, it is found that the prediction process is three orders of magnitudes faster than a corresponding Computational Fluid Dynamics (CFD) simulation, while relative errors are maintained lower than 1% on most of the sample points. The proposed model manages to capture the essential dynamics of the flow field, as its predictions correspond reasonably with the reconstructed field by proper orthogonal decomposition (POD). The performance and accuracy of the proposed model indicate that the deep learning-based approach has great potential as a robust predictive tool for aerodynamic design and optimization.

A general deep transfer learning framework for predicting the flow field of airfoils with small data

The flow field under different flow conditions contains abundant structure information and is of great significance for aerodynamic analysis and aircraft design. Deep learning (DL) techniques have received great interest for flow field prediction due to their high capability of capturing structure characteristics in recent years. However, successful training of DL models usually requires a large amount of data, whose acquisition is prohibitively expensive and learning from insufficient flow field data still remains underexplored for DL models. This paper proposes a general deep transfer learning framework to predict the flow field of airfoils with insufficient data by transferring knowledge learned from other conditions with a large amount of data. A novel and robust flow field prediction model based on the idea of generative adversarial networks, BiFlowAN, is first proposed to precisely predict the flow field over an airfoil under a condition with sufficient data. Transfer learning technique is then introduced to transfer the learned knowledge of BiFlowAN to a new model, BiFlowAN-TL, for improving the generalization of the model under another condition with a small-scale dataset. Compared with other alternatives, the flow field predictions of supercritical airfoils show that BiFlowAN achieves higher accuracy and generalization in predicting the flow fields with sufficient data and BiFlowAN-TL significantly improves the prediction accuracy than BiFlowAN on the small-scale dataset, exhibiting the potential applicability of the proposed methods to solve the problem of rapidly and accurately evaluating aerodynamic performance with insufficient data.

Transient flow performance and heat transfer characteristic in the cylinder of hydraulic driving piston hydrogen compressor during compression stroke

With the advantages of large flow capacity and high pressure, the use of hydraulic driving piston compressors in hydrogen refueling stations is becoming the development trend. Understanding transient flow and heat transfer characteristic is the key issue for the design and application of hydrogen compressors. The transient model of the hydraulic driving piston compressor is constructed by dynamic mesh and the National Institute of Standards and Technology (NIST) real hydrogen model, which accurately predicts flow field and heat transfer. Moreover, the effect of piston reciprocating cycle frequency on hydrogen parameters variation and heat transfer characteristic is investigated. Adiabatic compression theory is commonly applied in the design of reciprocating compressors. The results show that due to the heat transfer, the exhaust temperature predicted by the adiabatic compression theory is 6.29 K higher than the actual value. This study provides beneficial references for the design optimization and reliable operation of hydraulic driving piston hydrogen compressors.

Numerical assessment of hemodynamic perspectives of a left ventricular assist device and subsequent proposal for improvisation

Due to the unavailability of donors, the use of left ventricular assist devices has emerged to be a reliable line of alternative treatment for heart failure. However, ventricular assist devices (VAD) have been associated with several postoperative complications such as thrombosis, hemolysis, etc. Despite considerable improvements in technology, blood trauma due to high shear stress generation has been a major concern that is largely related to the geometrical feature of the VAD. This study aims to establish the design process of a centrifugal pump by considering several variations in the geometrical feature of a base design using the commercial solver ANSYS-CFX. To capture the uncertain behavior of blood as fluid, Newtonian, as well as non-Newtonian (Bird-Carreau model), models are used for flow field prediction. To assess the possibility of blood damage maximum wall shear stress and hemolysis index have been estimated for each operating point. The results of the simulations yield an optimized design of the pump based on parameters like pressure head generation, maximum shear stress, hydraulic efficiency, and hemolysis index. Further, the design methodology and the steps of development discussed in the paper can serve as a guideline for developing small centrifugal pumps handling blood.

A transfer learning-physics informed neural network (TL-PINN) for vortex-induced vibration

Vortex-induced vibration (VIV) is a typical nonlinear fluid-structure interaction (FSI) phenomenon, which widely exists in practical engineering (such as flexible risers, bridges and aircraft wings). Conventional numerical simulation and data-driven approaches for VIV analysis often suffer from the challenges of computational cost and dataset acquisition. This paper proposed a physics-informed neural network (PINN) enhanced by transfer learning (TL) to study a VIV system (2D). The TL-PINN only used 1/2, 1/4 and 1/8 of the training set (for PINN model) to reconstruct the information of flow field and structure, but with the same prediction accuracy as PINN model. In addition, a stepwise iterative training strategy was proposed to train PINN model. The strategy can effectively reduce the dependence of neural networks on data sets, so as to reduce the training cost of PINN model. The results show that PINN with the stepwise iterative training strategy and TL-PINN can enhance learning efficiency and keep predictability without requiring a huge quantity of datasets. Based on the proposed method, limited and scattered label data from monitoring, numerical and experimental can be fused to realize the reconstruction and prediction of flow field and structure information. It can break the limitation of monitoring equipment and methods in practical projects, and promote the in-depth study of VIV.

Towards high-accuracy deep learning inference of compressible flows over aerofoils

The present study investigates the accurate inference of Reynolds-averaged Navier–Stokes solutions for the two-dimensional compressible flow over aerofoils with a deep neural network. Our approach yields networks that learn to generate precise flow fields for varying body-fitted, structured grids by providing them with an encoding of the corresponding mapping to a canonical space for the solutions. We apply the deep neural network model to a benchmark case of incompressible flow at randomly given angles of attack and Reynolds numbers and achieve an improvement of more than an order of magnitude compared to previous work. Further, for transonic flow cases, the deep neural network model accurately predicts complex flow behavior at high Reynolds numbers, such as shock wave/boundary layer interaction, and quantitative distributions like pressure coefficient, skin friction coefficient as well as wake total pressure profiles downstream of aerofoils. The proposed deep learning method significantly speeds up the predictions of flow fields and shows promise for enabling fast aerodynamic designs.

Learning time-aware multi-phase flow fields in coal-supercritical water fluidized bed reactor with deep learning

The supercritical water fluidized bed (SCWFB) reactor has been utilized to achieve clean and efficient conversion of coal as well as the generation of hydrogen. The complicated time-aware multi-phase flow fields inside the reactor however are usually obtained via time-consuming experiments or transient numerical simulation. To this end, we propose a data-driven deep spatio-temporal sequence model, named ReactorNet, for intelligent modeling and efficient prediction of the complex time-aware multi-phase flow fields inside the SCWFB reactor. Particularly, the model employs the U-Net architecture wherein the coupled convolution module and BiConvLSTM module in a single block are stacked to simultaneously extract the multi-scale spatial and temporal features of multi-phase flow fields in the reactor. The numerical results showcase that the ReactorNet could successfully and simultaneously learn the evolution of multi-phase flow fields in the reactor under different temperature conditions, thus performing accurate multi-step-ahead prediction. The flow fields predicted by ReactorNet are not only highly consistent with the simulated results, but also hundreds of times faster than the traditional numerical simulation. In addition, the ReactorNet can carry out both reasonable spatial and temporal extrapolation and even long-term rollout prediction according to the learned evolution principle of flow fields, thus showcasing remarkable generalization ability. It is the first attempt to apply deep learning to the modeling and predicting of time-aware multi-phase flow fields inside the challenging SCWFB reactor.