Physics-informed Machine Learning for Full Reynolds Stress Modeling of Three-Dimensional Wall Bounded Flows and Stratified Wake Flows
Abstract
-Project Abstract -We propose a basic research program wherein physics-informed machine learning is developed toaugment the formulation and calibration of Full Reynolds Stress turbulence models. Ourhypothesis is that Full Reynolds Stress Models (FRSMs) possess the requisite physics todramatically improve the predictive capability of Reynolds Averaged Navier Stokes (RANS)methods for high Reynolds number Navy relevant flows where stress anisotropy and transporteffects render eddy viscosity models (EVMs) inherently insufficient. Our emphases in thisproposed project, and a companion project recently started by Robert Kunz, are on two such flowsof interest to the US Navy s undersea research enterprise: high Reynolds number wall boundedflows exhibiting 3D separation, and high Reynolds number, high Froude number stratified wakes.The objectives of the currently funded companion work (Kunz PI) are to implement, apply andassess the performance of FRSMs for these two classes of flows, in the context of their potentiallyfar superior accuracy over EVMs, and the intractability of LES/DNS for these flows. Accordingly,as these models are evaluated for these two applications (using experimental and computationaldata developed under ONR Code-331 sponsorship), a need for improved accuracy will inevitablyarise. The present research offering recommends a modern machine-learning based methodologyaimed at this need. Specifically, we will use machine-learning to 1) reformulate conventionalmodeling assumptions, 2) calibrate modeling coefficients. We envision the outcome of thiscompanion project to be vastly improved FSRMs for important USN undersea applications, to beimplemented in research and US Navy application codes.The proposed research will be a collaborative work between the PI, Dr. Xiang Yang, and hiscollaborator Dr. Robert Kunz, both in the Mechanical Engineering Department of the PennsylvaniaState University. This research will complement the on-going FRSM project in the program. Ourteam will adapt the machine-learning based approach detailed below to develop improved FRSMs.The team will use the data from ONR Code-331 performers, open literature, and Dr. Kunz s project,to evaluate the augmented modeling, and compare results with the baseline implementation. Asidefrom the FRS model, which will directly benefit USN RANS modeling of important underseaflows, the proposed research will also demonstrate how available high-fidelity data can be used todiscover new flow physics and inform/improve low-fidelity computational tools, a researchoutcome that will benefit future developments of modeling tools.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 29, 2020
- Source ID
- N000142012315
Entities
People
- Xiang I. A. Yang
Organizations
- Office of Naval Research
- Pennsylvania State University
- United States Navy