MB3:  Numerical Methods and Models for High Performance Fluid/Structure Computations

Room: Old Main Academic Center 3070

Webex Link

Organizer: Shanti Bhushan, Mississippi State University

William England, ERDC-ITL, USA

Time: 10:00 am - 10:25 am (CST)

Title: Multiphysics Coupling Strategies for Hypersonic Aerothermoelasticity

Abstract: Effective design and analysis of vehicles subject to hypersonic flows, or flows where Ma > 5, continue to be of tremendous impact to defense and academic institutions.However, analysis of hypersonic flows is plagued by complex interactions between many physical domains, including fluid dynamics, structural mechanics, and heat transfer. To this end, there exist two major approaches to model the interaction of aerothermoelastic effects, namely partitioned and monolithic approaches. Partitioned approaches model interactions through transfer of boundary data between single-physics solvers and enjoy wide use in many physical simulation domains due largely to their relative ease of implementation.However, partitioned methods suffer physical instabilities due to the inherent leading and lagging of the constituent physics as well as physical inconsistencies from interpolation schemes between meshes. Monolithic approaches remedy these problems by solving a single equation system comprised of all physical equations on a single mesh. Due to this, monolithic methods can exhibit time-accurate results and facilitate more rigorous analysis of discretizations of the physical equations. Still, the implementation of monolithic methods is often more difficult than that of partitioned methods. In this work, we provide an overview of partitioned and monolithic coupling schemes for accurate simulation of hypersonic aerothermoelasticity necessary to advance the state-of-the-art indesign and analysis of hypersonic vehicles.

Seshendra Palakurthy, CAVS, Mississippi State University

Time: 10:25 am - 10:50 am (CST)

Title: Analysis of Panel Flutter Dynamics due to Turbulent Shock Wave Boundary Layer Interactions

Abstract: Thin skin panels of high speed aero-vehicles are subjected to in-tense loading due to shock wave-boundary layer interaction. The flow unsteadiness coupled with the panel’s non-linear dynamic response produces flutter that makes the skin panels susceptible to fatigue failure. In this work, we investigate the panel flutter at Mach 2.0, pressure ratio of 2.3, mass ratio 0.1, and non-dimensional dynamic pressure ranging from 1800 to 4200. The flow is assumed turbulent, and the shock wave impinges on the panel at mid-point. The coupled fluid-structure interaction simulations are performed using flow Psi for high fidelity fluid solution and MAST for high fidelity structural dynamic response. We employ various modal analysis techniques such as fastFourier transforms (FFT), proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD) to understand the effect of the changing non-dimensional dynamic pressure on the panel’s dynamic response.

Joint work with Anup Zope, Eric Collins, and Shanti Bhushan.

Xiaoling Tong, Mississippi State University

Time: 10:50 am - 11:15 am (CST)

Title: Algorithm Development of Physics Based Fluid and Dust Simulations on Cartesian Grids for Moving Vehicles

Abstract: Algorithm development of a prototype flow and dust simulation model is conducted using artificial compressibility approach (ACA) on Cartesian meshes to simulate low Mach complex flow with more computational efficiency. A fourth-order low dissipation scheme with a third order Roe dissipation term is constructed to support large eddy simulation (LES) model. To account for the impact of boundary effect, the immersed boundary method (IBM) based on Dirac Delta function (an impulse function) or direct surface integral is employed to construct forcing terms that reflects the presence of immersed boundary. Equilibrium particle model is adopted for dust simulation. The prototype is excised on testing flow over various shapes including a simplified truck, and it shows that the code is capable of predicting turbulent fluid motion and particle transport around complex objects with robustness and computational efficiency.

Joint work with E. A. Luke and E. Collins.

Wesley Brewer, DoD HPCMP PET/GDIT, USA

Time: 11:15 am - 11:40 am (CST)

Title: Scalable Integration of Computational Physics Simulations with Machine Learning

Abstract: The integration of machine learning with simulation is a growing trend, however, the augmentation of codes in a highly-performant, distributed manner poses a software development challenge. In this work, we explore the question of how to easily augment a legacy simulation code on a high-performance computer (HPC) with a machine-learned surrogate model incurring minimal slow down and allowing for scalability. Initial naïve augmentation attempts required significant code modifications, while also incurring significant slowdown. This led us to explore using inference serving techniques, which allow for inference through drop-in functions. In this work, we investigated TensorFlow Serving with gRPC and RedisAI with SmartRedis for server-client inference implementations, where the deep learning platform is run as a persistent process on the GPUs ofHPC compute nodes and the simulation makes client calls while running on the CPUs. We evaluated inference performance for several use cases at scale on Summit, including rotorcraft aerodynamics, real gas equations of state, and super-resolution techniques. Additionally, a machine-learned boundary condition was implemented in a CFD solver where the unsteady wake of a rotor modeled by a physics-informed neural network is injected as an inflow boundary condition. We will discuss key findings on performance across Python, C++, and Fortran APIs for inference clients and compare the results to ideal performance (using TensorFlow C-API in-process inference). The lessons learned may provide useful advice for researchers to augment their simulation codes in an optimal manner.

Joint work with Mathew Boyer, Daniel Martinez, and Ian Dettwiller.

Mohammed Elmellouki, CAVS, Mississippi State University

Time: 11:40 am - 12:05 pm (CST)

Title: Numerical Investigation of heat Transfer Characteristics in Liquid Metal Flows

Abstract: Liquid metals such as Lead (Pb), Sodium-Potassium (Na-K) are often used as coolants in nuclear reactors since those liquids are characterized by their high thermal conductivity and low heat capacity compared to other fluids such as water and air, thus involve low-Prandtl number (Pr). The predictions of turbulent heat transfer in low-Pr flows remain an open challenge for the engineering community, since those liquids tend to conduct the heat more rapidly than momentum via molecular diffusion mechanisms resulting in a thermal boundary layer that extends beyond the momentum layer. Nowadays, the turbulence models are developed and validated for high Pr flows, which rises a need to validate the available turbulence models for lower Pr flows. The aim of this study is the assessment and validation of the predictive capability of high and low fidelity turbulence models for low Pr flows for plane channel flow with unequal wall temperature and gravity effect, a flow through a backward facing step to investigate the predictive capabilities of the turbulence models for separated and reattached flows with heat transfer and gravity effect, and forced convective flow through a rod bundle assembly and a vertical pipe with heat transfer.

Joint work with Christopher Pilmaier and Shanti Bhushan.