Session F09: Computational Fluid Dynamics: General (3:55pm - 4:40pm CST)

Effect of Magnetic Field on the Nanofluid Flow in a Circular Pipe with a Return Bend

In this article, numerical modeling is carried out to investigate the convective heat transfer of AL2O3-water nanofluid in a circular pipe with a return bend under the effect of a magnetic field. The impacts of Reynolds number, Hartmann number, and the volume concentration of nanoparticles on the nanofluid flow and the convective heat transfer behavior are investigated. The results show that the maximum heat transfer occurs in the cross-section with the return bend in a volume concentration of 15. However, by increasing the volume concentration to 20, the Nusselt number decreases. The results indicate that the effect of the magnetic field on the Nusselt number for nanofluid flows with a volume concentration of less than 5 is negligible and sometimes useless. While for volume concentrations 10 and 20 with Hartmann number 20, the ratio of the amount of the local Nusselt number under the magnetic field to the Nusselt number of simple fluid flow is 46.1 and 89.1, respectively. In other words, the heat transfer has increased by 46{\%} and 89{\%}, respectively compared to the simple fluid with der Hartmann number 20.   

Accelerating Time Averaging

Frequently, the goal of numerical modeling of a dynamical system is to obtain the long-time average of a certain parameter, such as, for example, the lift force. Nonlinear systems often exhibit a chaotic behaviour. When the fluctuations are large, obtaining time-averaged quantities with sufficient accuracy requires expensive numerical calculations. We replace the quantity being averaged with another quantity having the same average but fluctuating less. This is achieved using the ideas of the method of bounding time averages (Chernyshenko et al., 2014, Phyl.Trans.Roy.Soc.A, 372). The key idea is that for any differentiable function $V(x)$, where $x$ is the state of the dynamical system, the infinite time average of $dV(x(t))/dt$ is zero provided that $x(t)$ is bounded, which is always the case when infinite time averaging is meaningful. Hence, rather than numerically averaging the quantity of interest, which we will denote $F$, one can average $F+dV/dt$, for any $V$. The function $V(x)$ can be optimized so as to accelerate the averaging. So far, this approach was tested on the Lorenz attractor and a two-dimensional flow past a square cylinder, reducing the time required to achieve reasonable accuracy of the average by 10 to 20 \%   

Online Simulator of a Cylinder Wake for Undergraduate Fluid Mechanics Labs

Wind tunnels are commonly used in undergraduate fluid mechanics courses to help students understand fluid flow around objects, particularly wake dynamics. Large engineering and physics programs may only have one wind tunnel accessible to undergraduate students, limiting their experience to only a few minutes each. While many recent advances have been made in virtual reality (VR) fluids labs, VR systems are costly, require high computational resources, and can only be accessed by students in a dedicated computer lab. The goal of this project was to develop an online simulator of a wind tunnel which students could access at any time. A simulation of the wake behind a circular cylinder was developed, with the local velocity being measured by a virtual pitot-static probe. The probe can be moved to any location downstream of the cylinder, allowing students to extract span-wise and axial velocity profiles. The simulated local velocity is calculated from the Schlichting self-similar wake profile, with a small random perturbation that is a function of the Struhaul-Reynolds number correlation. The simulator was calibrated using actual wind tunnel measurements. A controlled educational study of the impact on students using the simulator is being planned.   

Towards Building Robust Neural Network Models for Fluid Simulations

The rise of Machine Learning (ML) for modeling complex problems in fluid physics has brought with it challenges in developing tools that are robust and explainable. Scientific data being high-dimensional, multimodal, and complex makes it difficult to choose the appropriate hyper-parameters for such networks. In this work, we explore the possibility to fully automate the network design process for a data-driven fluid physics emulator - in this case modeling a key turbulence prognostic critical for closure in a Large Eddy Simulation (LES) compressible flow code; including neural architecture search as well as optimizing the hyper-parameters using Bayesian principles. Further we investigate the network learnings by analyzing the gradient flows and conduct variance-based sensitivity analysis to understand the trained network predictions thereby improving explainability of these so called black-box models   

Breakdown of von Neumann analysis: a generalized approach for multi-level schemes

The von Neumann analysis has been extensively used to assess the stability of numerical schemes. It requires bounding the amplification factor(s) by unity to ensure stability. However, through numerical experiments using two well-known multi-level schemes, leapfrog and Du-Fort and Frankel, we note that the amplification computed from these simulations (i) is not equal to that obtained from the von Neumann analysis (ii) exhibits an oscillatory behavior in time and (iii) takes instantaneous values larger than unity despite stability constraints being satisfied. These disagreements are observed because the von Neumann analysis implicitly assumes amplification to be independent of time level. In the generalized von Neumann analysis, we relax this assumption to obtain a time varying amplification factor that agrees exactly with the corresponding numerical value at all times. We re-define stability to account for the variability of amplification with time. Furthermore, we express this new amplification factor as a continued fraction to determine the exact conditions when the standard analysis, if at all, is applicable. Analysis of asynchrony-tolerant schemes and the effect of temporal discretization on spectral accuracy of spatial schemes will also be discussed.   

Merging Controls Technology with CFD For Responsive Computing

CFD was availed to study an industrial scale multiphase reactor used in the production of low-density polyethylene (LDPE). The reactor is divided up into four zones separated by baffles designed to control mixing and meter the communication among the reactants. Five proportional integral derivative (PID) controllers were designed to independently automate five catalyst feeds throughout the reactor to match thermocouple readings to plant data. The objective of the model is to quickly respond to upset conditions in the plant that can lead to thermal runaway due to the exothermic nature of the radical chemistry within the reactor. Controller settings were tuned independently in each zone based on process art, reaction rates, convective time scales, and the proximity of thermocouples to catalyst ports. This allowed the temperature controller errors to be driven towards zero without user involvement. In addition to the reactor model, a canonical problem was used to tune a turbulence model constant. We, therefore, demonstrate the efficacy of implementing algorithmic PID control as 1) a means of automating dynamic CFD models to match plant-scale behavior and 2) a preamble to using PID methods to facilitate machine learning.   

Structurally complete Riemann solvers for flows with non-ideal thermodynamics

Approximate Riemann solvers, such as the HLL and HLLC Riemann solvers, have been at been at the cornerstone of compressible fluid dynamics. The intercell flux is approximated based on the understanding of the wave configuration of the information transfer. We propose an extension to these approximate Riemann solvers to account for the spatially varying rarefaction wave solution which becomes important under non-ideal thermodynamics. At high pressures and temperatures, the rarefaction head and tail can move in opposite directions, thus enclosing the cell interface and determining the intercell flux. These conditions are typical of trans- and supercritical flows. Using a recently found analytical solution to the non-ideal gas rarefaction wave (Wang and Hickey, Phys. Fluid. 2020), we propose a Structurally Complete Riemann Solver (SCRS) which shows great accuracy benefits to the Riemann problem solution. SCRS shows notable improvements over traditional Riemann solvers in flows where the rarefaction and star-state regions of the Riemann solution are locally supersonic.   

Combustion LES with tabulated chemistry in the framework of a novel compressible flow formulation

This study introduces a new methodology of integrating tabulated chemistry into the compressible flow formulation. In the classical method, fully conservation Navier-Stokes equations are solved and the pressure is obtained through a linearization relation. The novelty of the new approach lies in the treatment that the energy equation is replaced by a pressure evolution equation, with which the pressure is directly resolved. The proposed formulation is assessed with a number of test cases covering different flame configurations. The convergence study shows that the new approach is able to accurately reproduce the flame speed and flame profile. Our assessment is further extended to the predictions of turbulent flames, for which a classical Bunsen slot flame and the Sandia Flame D are considered. The LES calculations are performed, and the simulation results are compared against the experimental data. The accuracy of the proposed formulation will be discussed in detail.   

Optimal control of wave energy harvesting devices using the reinforcement learning technique

It is well accepted that the known sources of fossil fuels in the world are depleting very fast and there is an increasing need to harvest energy from renewable resources. In the form of waves and currents, ocean is a tremendous source of renewable energy.Harvesting this energy from these resources has been a subject of interest from many centuries. A wave energy converter (WEC) is defined as a device that converts the kinetic and potential energy associated with a water surface wave into useful mechanical and electrical energy. One promising technique is to employ control methods for the wave energy converters to increase their energy capturing efficiency. Toward this, in this study, we employ the reinforcement learning technique to learn the optimal geometry and modify the parameters of a WEC device based on the incoming waves. A multi-body, time-domain simulation tool created for the simulation of WEC devices is used to evaluate the WEC performance. The WEC device dimensions and Power Take Off (PTO) parameters are used as controlling parameters within the reinforcement learning technique to determine which parameters maximize the constructive interaction between the environment and the device considering the device constraints.   

An efficient technique for importing complex geometry into block-structured adaptive mesh refinement codes.

Immersed boundary methods (IBM) are becoming viable alternatives to traditional unstructured meshes for embed complex geometry on structured meshes. We present an efficient method to import geometry from a STL file and generate a signed distance function.~The method uses a ray-tracing algorithm to find the intersection of the STL triangles along grid lines in all three directions. Usually, each ray requires searching through the entire list of triangles, which is prohibitively expensive for large STL files.~The efficiency of triangle search is increased dividing the domain into prisms using a quad-tree data structure. The ray-tracing algorithm is then performed over a reduced triangle list in each prism.~The resulting triangle-ray intersection points are used to directly form a signed distance function. Results show that the proposed algorithm is up to 20 times faster. Adaptive mesh refinement introduces additional challenges where the grid can dynamically change on the embedded surface. We present a technique that reforms the signed distance function only on AMR boxes that contain new data during regridding. The results show a computational saving of up to 2.3 over forcing grid refinement on the entire surface.   

A Universal Parametric Study of Shark Denticles' Anti-Flow-Reversal Mechanism

The purpose of this study was to use computational methods to investigate the ability of shark scales to mitigate flow separation. To enable a broad, parametric analysis, we simplified a single denticle as a thin wall within a 2D Couette flow, wherein the upper moving wall represents a turbulent streak atop the viscous sublayer. We observed the effects of varying geometric parameters and characteristic numbers -- including one novel parameter -- on several metrics. Near-wall flow reversal is a precursor to flow separation, so we considered metrics such as mass flow through our domain and peak backflow velocity along a vertical midline. Our chief result was that blockage ratio -- the proportion of the channel's height blocked by the denticle -- is the primary factor correcting backflow, but we also found that smaller denticle angles more efficiently prevent backflow at a given blockage ratio. Our findings offer universal implications about the ability of sharkskin to impede separation.   

A Hardware Accelerated Unstructured Overset Method to Simulate Turbulent Flows

A numerical method is presented which can effectively use graphic processing units (GPUs) to simulate moving bodies in incompressible turbulent fluid flow. The method is an unstructured overset method where unstructured overset meshes are attached to individual bodies and connected throughout the flow domain to produce a single domain solution through an overset assembly process. Efficient algorithms from real-time ray-tracing and collision detection are used to accelerate the overset assembly process, producing O(100x) speed-up for core assembly operations. A novel method to simulate turbulent fluid flow is presented which uses domain over-decomposition to allow asynchronous calculation of the steps of the method while simultaneously overlapping GPU data transfer and calculations. A pressure regularization based on artificial compressibility is used with mixed precision linear solvers to provide optimal performance while maintaining desired accuracy. Timings and results are shown for canonical cases demonstrating the method’s accuracy and effectiveness when using GPUs. A maximum of 80x speed-up is found when compared to a benchmark overset solver for large cases.   

Novel Use of a Common Respiratory Treatment: Diminishing COVID-19 Transmission

This paper demonstrates the use of a 130-million cell hex-dominant mesh to study the spread of aerosolized particles. Our model consists of a full hospital room including two patients and four caregivers. All six airways were accurately modeled using medical imagery and resolved with 0.5 mm mesh elements. Independent breathing curves were represented using sophisticated time-varying boundary conditions that capture key characteristics such as unique sinusoidal inspiratory and exponential expiratory curves, as well as randomly varying tidal volumes. We present results from multiple sets of numerical methods to demonstrate numerical accuracy. Based on these results, we offer general findings on the spread of contagion within a hospital room and demonstrate the effectiveness of a novel respiratory apparatus at reducing aerosol emission. This apparatus consists of a high velocity nasal insufflation cannula in conjunction with a PVC face mask connected to suction.   

A flexible framework of high-order shock-capturing schemes for convection-dominated problem

In this work, we further extend the TENO framework proposed by Fu [Fu et al., Journal of Computational Physics 305 (2016): 333-359] with flexibility to control the nonlinear dissipation property of TENO schemes in nonsmooth regions while maintaining the performance of TENO in smooth regions. While a set of candidate stencils of incremental width is constructed, each one is indicated as smooth or nonsmooth by the ENO-like stencil selection procedure proposed in TENO scheme. Rather than being discarded directly in TENO schemes, the nonsmooth candidates are filtered by an extra nonlinear limiter, e.g. monotonicity-preserving (MP) limiter. Consequently, the high-order reconstruction is achieved by assembling the candidate fluxes with the optimal linear weights since they are either smooth reconstructions or filtered ones which feature good non-oscillation property. Based on the proposed framework, several new six- and eight-points TENO schemes with controllable dissipation are developed. A set of critical benchmark cases reveal that the proposed new TENO schemes capture the discontinuities sharply and resolve the high-wavenumber fluctuations with low dissipation, while maintaining the desired accuracy order in smooth regions.   

CFD-Informed Reduced-Order Modeling of \\Extreme-Speed Turbochargers

In order to improve their efficiency and performance, aircraft compression-ignition engines often incorporate turbochargers originally designed for ground-based applications. To sufficiently power the aircraft, these turbochargers must operate outside of their standard operating envelopes and consequently encounter high-cycle fatigue brought on by aerodynamically-induced blade resonances. The onset of fluid-structural interactions in turbochargers at these conditions has not been extensively studied. In this talk, we investigate the behavior of the turbine-side of the turbocharger utilizing computational fluid dynamics (CFD) and computational structural dynamics (CSD) methods to understand the mechanisms responsible for turbine blade resonance. A reduced-order model is constructed utilizing the Euler-Lagrange equation. The structural response is described utilizing a method of assumed modes approach, informed by CSD, and the unsteady fluid response is informed by CFD. We specifically investigate the unsteady fluid dynamics model that links blade deformations to the induced surface pressure fluctuations.   

A parallel Local-Adaptive-Mesh-Refinement-enabled Immersed Boundary Method for biological flows

A fast parallel implementation of projection method for the time-dependent incompressible Navier-Stokes equation is presented on a patch-based local refined mesh on a hierarchy of rectangular blocks. A sharp immersed boundary method compatible with the coarse-fine interface interpolation was implemented on the non-conforming Cartesian grids for flows with immersed bodies. A two-dimensional Taylor-Green vortex example shows that a multi-dimensional at least second-order Lagrange interpolation is critical to achieving second-order accuracy in space. An intra-layer communication was identified for overlapping multi-blocks which enables a flexible refinement strategy and better load balance among computing nodes. The newly developed algorithm was benchmarked using flow passing fixed sphere and the results matched well with literature. The numerical simulation shows the parallel code is comparable in execution time to serial execution with the same dense mesh. It decreases slightly with an increased level of refinement. The newly developed algorithm is efficient for the bio-inspired flows where the refined region is predeterminable. Fish swimming and free-flying dragonfly cases are computed by the multi-block and moving block strategies, respectively. Applications to biomedical flow problems including human snoring and animal phonation are also demonstrated in this presentation.   

Towards a hypersonic strand/Cartesian adaptive mesh refinement solver

\\ High resolution in the boundary layer and shock region is required to obtain accurate heating results from Computational Fluid Dynamic simulations of hypersonic vehicles. Manually creating a suitable mesh often becomes a bottle neck, especially if the shape of the vehicle changes, due to ablation, flexible heat shields, or moving components. In this work, a hypersonic “strand/Cartesian AMR” solver has been developed to enable automated mesh generation around hypersonic vehicles. In a strand/Cartesian AMR solver, the strand mesh technique is used to create a high-quality mesh in the near-body region, which adequately resolves the boundary layer. Cartesian AMR techniques are used in the off-body region, highly resolving off-body shock structures, and the two regions are joined by overset algorithms. \\ An existing Cartesian AMR solver has been extended to incorporate the two-temperature model, and to enable mapped body-fitted simulations. Strand mesh and overset algorithms have been developed to create the mapped mesh and join the domains, respectively. Verification and validation results from the two individual solvers and the combined solver indicate that the strand/Cartesian AMR method could alleviate the meshing bottlenecks encountered when modelling hypersonic vehicles.   

An Interface Capturing Procedure for Simulating Incompressible Two-Phase Flows on Adaptive Unstructured Grids

To design many industrial systems, accurate and efficient simulations of complex two-phase flows are required. In this context, handling complex geometries becomes necessary. The use of unstructured grids fulfills this requirement, and with Adaptive Mesh Refinement (AMR) computational resources can be allocated according to need. This work presents an Accurate Conservative Level-Set/Ghost-Fluid algorithm for unstructured grids, implemented in the YALES2 incompressible finite-volume flow solver. In the ACLS framework, the interface is defined as the iso-contour of a hyperbolic tangent function, which is advected by the fluid, and reshaped using a reinitialization equation. A new form of this equation, that better preserves the interface shape, has been recently proposed by Chiodi et al, and we extend it to unstructured grids in this study. To compute interface normals and curvature, the signed-distance function is reconstructed in a narrow band around the interface using a geometric-projection marker method. Isotropic AMR is automatically triggered based on interface displacement. Interface transport and two-phase flow tests are firstly simulated to validate the procedure. We then perform LES of a water jet in quiescent air from a low-pressure compound nozzle.