Stanford Center for Turbulence Research 2020 Annual Briefs

Center for Turbulence Research Annual Briefs

Machine Learning for Fluid Dynamics: Patterns

 

The data set he mentions is Johns Hopkins Turbulence Databases. Many hundreds of Terabytes of direct numerical simulations with different governing equations and boundary conditions.

Machine Learning for CFD Turbulence Closures

I wrote a couple previous posts on some interesting work using deep learning to accelerate topology optimization, and a couple neural network methods for accelerating computational fluid dynamics (with source). This post is about a use of machine learning in computational fluid dynamics (CFD) with a slightly different goal: to improve the quality of solutions. Rather than a focus on getting to solutions more quickly, this post covers work focused on getting better solutions. A better solution is one that has more predictive capability. There is usually a trade-off between predictive capability, and how long it takes to get a solution. The most well-known area for improvement in predictive capability of state-of-the-practice, industrial CFD is in our turbulence and transition modeling. There are a proliferation of approaches to tackling that problem, but the overall strategy that seems to be paying off is for CFD'ers to follow the enormous investment being made by the large tech companies in techniques, open source libraries, and services for machine learning. How can those free / low-cost tools and techniques be applied to our problems?

The authors of Machine Learning Models of Errors in Large Eddy Simulation Predictions of Surface Pressure Fluctuations used machine learning techniques to model the error in their LES solutions. See an illustration of the instantaneous density gradient magnitude of the developing boundary layer from that paper shown to the right. Here's the abstract,

We investigate a novel application of deep neural networks to modeling of errors in prediction of surface pressure fluctuations beneath a compressible, turbulent flow. In this context, the truth solution is given by Direct Numerical Simulation (DNS) data, while the predictive model is a wall-modeled Large Eddy Simulation (LES
). The neural network provides a means to map relevant statistical flow-features within the LES solution to errors in prediction of wall pressure spectra. We simulate a number of flat plate turbulent boundary layers using both DNS and wall-modeled LES to build up a database with which to train the neural network. We then apply machine learning techniques to develop an optimized neural network model for the error in terms of relevant flow features

Deep Learning to Accelerate Computational Fluid Dynamics

Lat-Net: Compressing Lattice Boltzmann Flow Simulations using Deep Neural Networks

I posted about a surprising application of deep learning to accelerate topology optimization. The thing I like about that approach is it's a strategy that could be applied to accelerate many different solvers that we use to simulate all sorts of continuum mechanics based on partial differential equations (i.e. computational fluid dynamics, structural mechanics, electrodynamics, etc.). With a bit of help from Google I found a neat paper and project on github doing exactly that for a Lattice-Boltzmann fluid solver.

Hybrid Parallelism Approaches for CFD

This previous post, Plenty of Room at Exascale, focuses on one specific commercial approach to scaling CFD to large problems on heterogeneous hardware (CPU & GPU) clusters. Here's some more references I found interesting reading on this sort of approach.

Strategies

Recent progress and challenges in exploiting graphics processors in computational fluid dynamics provides some general strategies for using multiple levels of parallelism accross GPUs, CPU cores and cluster nodes based on that review of the literature:

• Global memory should be arranged to coalesce read/write requests, which can improve performance by an order of magnitude (theoretically, up to 32 times: the number of threads in a warp)
• Shared memory should be used for global reduction operations (e.g., summing up residual values, finding maximum values) such that only one value per block needs to be returned
  
• Use asynchronous memory transfer, as shown by Phillips et al. and DeLeon et al. when parallelizing solvers across multiple GPUs, to limit the idle time of either the CPU or GPU.
• Minimize slow CPU-GPU communication during a simulation by performing all possible calculations on the GPU.

Plenty of Room at Exascale

The folks at Envenio have posted an interesting marketing video on their solver.

Hypersonics Basic and Applied Lectures

A One-Equation Local Correlation-Based Transition Model

This article is available for free download until 15 Oct 2015, h/t ANSYS blog.

Here's the Abstract:

A model for the prediction of laminar-turbulent transition processes was formulated. It is based on the LCTM (‘Local Correlation-based Transition Modelling’) concept, where experimental correlations are being integrated into standard convection-diffusion transport equations using local variables. The starting point for the model was the γ-Re θ model already widely used in aerodynamics and turbomachinery CFD applications. Some of the deficiencies of the γ-Re θ model, like the lack of Galilean invariance were removed. Furthermore, the Re θ equation was avoided and the correlations for transition onset prediction have been significantly simplified. The model has been calibrated against a wide range of Falkner-Skan flows and has been applied to a variety of test cases.

Keywords: Laminar-turbulent transition, Correlation, Local variables
Authors: Florian R. Menter, Pavel E. Smirnov , Tao Liu, Ravikanth Avancha

Transition location, and subsequent turbulence modeling remain the largest source of uncertainty for most engineering flows. Even for chemically reacting flows the source of uncertainty is often less the parameters and reactions for the chemistry, and more the uncertainty in the fluid state driven by shortcomings in turbulence and transition modeling.

HiFiLES v0.1 Release

The folks at the Stanford Aerospace Computing Lab have recently released version 0.1 of HiFiLES. "HiFiLES is a high-order Flux Reconstruction solver for the Euler and Navier Stokes equations, capable of simulating high Reynolds number turbulent flows and transonic/supersonic regimes on unstructured grids."

From the release notes:

High-order numerical methods for flow simulations capture complex phenomena like vortices and separation regions using fewer degrees of freedom than their low-order counterparts. The High Fidelity (HiFi) provided by the schemes, combined with turbulence models for small scales and wall interactions, gives rise to a powerful Large Eddy Simulation (LES) software package. HiFiLES is an open-source, high-order, compressible flow solver for unstructured grids built from the ground up to take full advantage of parallel computing architectures. It is specially well-suited for Graphical Processing Unit (GPU) architectures. HiFiLES is written in C++. The code uses the MPI protocol to run on multiple processors, and CUDA to harness GPU performance.

The main reference for the code right now is this V&V paper.[1] The code uses an Energy Stable Flux Reconstruction (ESFR) scheme. Here are a couple papers on that approach.[2, 3].

References

[1]   López-Morales, M. R., Bull, J., Crabill, J., Economon, T. D., Manosalvas, D., Romero, J., Sheshadri, A., Watkins II, J. E., Williams, D., Palacios, F., et al., “Verification and Validation of HiFiLES: a High-Order LES unstructured solver on multi-GPU platforms,” .

[2]   Vincent, P. E., Castonguay, P., and Jameson, A., “A new class of high-order energy stable flux reconstruction schemes,” Journal of Scientific Computing, Vol. 47, No. 1, 2011, pp. 50–72.

[3]   Castonguay, P., Vincent, P. E., and Jameson, A., “A new class of high-order energy stable flux reconstruction schemes for triangular elements,” Journal of Scientific Computing, Vol. 51, No. 1, 2012, pp. 224–256.

SU2 Community Verification Studies

I think there is quite a bit of excitement and community involvement building around the SU2 code. Other than all of the updates and improvements in the recently released version 3 and SU2_EDU release, I am excited to see the wider community start to do some serious verification studies. The advecting vortex case linked in that discussion thread would be a good one to add to the Test Case collection.

The core SU2 devs have a recent AIAA paper on verification/validation cases that they have successfully run with SU2 and compared favorably to other codes. One thing that is conspicuously absent is grid convergence studies to verify order of accuracy. This is an ideal place for the community to contribute because you don't have to have hugely in-depth knowledge of the source code base to run a grid convergence study or contribute a tutorial or test case (though you do have to be a fairly competent user). Much to their credit, the SU2 team is soliciting just this kind of contribution (my emphasis):

Expanded tutorials: we would like additional tutorials that complement the existing set found in the web-based documentation. The tutorials can either detail and explain the functionality of SU2 (shape optimization, parallel computing, mesh deformation, etc.) or demonstrate general CFD knowledge (highlighting good meshes vs. bad meshes, the importance of CFL number, etc.). Tutorials are intended to be tools for teaching and learning, and they should follow the same style as the existing tutorials. They must provide any mesh or config files that are necessary for their completion. New or unique verification and validation cases would be of particular interest here.

Exciting times in open source CFD!

SU2 v3 Released

The folks at Stanford Aerospace Design Lab have released a new major version of Stanford University Unstructured (SU2). Here's the announcement:

Dear Colleague,

Since its introduction in January 2012, SU2, The Open-Source CFD Code, has been downloaded thousands of times by users and developers in academia, government, and industry, including many leading companies and universities. As an open-source project, the growth of active user and developer communities is a crucial goal for SU2. Given the incredibly positive response, we are pleased to announce a new version of the code with major improvements and a entirely new package for educational purposes.

This release marks the third major version of the SU2 open-source code (su2.stanford.edu). SU2 is a collection of C++ software tools for performing Partial Differential Equation (PDE) analysis and for solving PDE-constrained optimization problems, with special emphasis on Computational Fluid Dynamics (CFD) and aerodynamic shape design.

We'd like to ask you to please distribute this announcement with the attached flyer to any colleagues and students in your department that might be interested.

Version 3.0 has a number of major additional capabilities:

• Adjoint-based RANS shape optimization.
• New unsteady analysis and design optimization capability.
• Upgrades to the underlying parallelization and file I/O.
• Significant improvements to the accuracy, performance, and robustness of the software suite.

Alongside Version 3.0 of SU2, we are introducing SU2 Educational (SU2_EDU): a new, educational version of the Euler/Navier-Stokes/RANS solver from the SU2 suite. The simplified structure of SU2_EDU makes it suitable for students and beginners in CFD. By focusing on a handful of key numerical methods and capabilities, SU2_EDU is ideal for use in CFD courses, for independent studies, or just to learn about a new field!

SU2_EDU is also geared toward anyone interested in high-fidelity airfoil analysis. The initial version of SU2_EDU is an intuitive, easy to use tool for computing the performance of airfoils in inviscid, laminar, or turbulent flow including non-linear effects in the transonic regime, that only requires the airfoil coordinates.

Finally, we would like to thank the open-source community for their interest, help, and support.

The SU2 team

One of the most interesting parts to me is the new SU2_EDU version. I've downloaded the code, but haven't had a chance to browse it or run any examples yet. I think this is a neat idea that will hopefully lower the barriers to entry that George pointed out previously.

CFD Vision 2030: Discretizations, Solvers, and Numerics

There are lots of interesting parts to the study that Phil Roe mentioned in his Colorful Fluid Dynamics lecture. Continuing the theme that algorithm improvements are just as important as hardware improvements here are some of the areas concerning discretizations, solvers and numerics (pp 24) that the report claims will lower the need for high levels of human expertise and intervention in running and understanding CFD analysis:

1. Incomplete or inconsistent convergence behavior: "There are many possible reasons for failure, ranging from poor grid quality to the inability of a single algorithm to handle singularities such as strong shocks, under-resolved features, or stiff chemically reacting terms. What is required is an automated capability that delivers hands-off solid convergence under all reasonable anticipated flow conditions with a high tolerance to mesh irregularities and small scale unsteadiness."
   
2. Algorithm efficiency and suitability for emerging HPC: "In order to improve simulation capability and to effectively leverage new HPC hardware, foundational mathematical research will be required in highly scalable linear and non-linear solvers not only for commonly used discretizations but also for alternative discretizations, such as higher-order techniques89. Beyond potential advantages in improved accuracy per degree of freedom, higher-order methods may more effectively utilize new HPC hardware through increased levels of computation per degree of freedom."
   

Phil Roe: Colorful Fluid Dynamics

Echos of Tufte in one of his introductory statements: "It's full of noise, it's full of color, it's spectacular, it's intended to blow your mind away, it's intended to disarm criticism." And further on the dangers of "colorful fluid dynamics":

These days it is common to see a complicated flow field, predicted with all the right general features and displayed in glorious detail that looks like the real thing. Results viewed in this way take on an air of authority out of proportion to their accuracy.
--Doug McLean

This lecture is sponsored by MConneX.

Roe wraps up the lecture by referencing a NASA sponsored study, CFD Vision 2030, that addresses whether CFD will be able to reliably predict turbulent separated flows by 2030. The conclusion is that advances in hardware capability alone will not be enough, but that significant improvements in numerical algorithms are required.

Algorithmic Improvements: just as important as Moore's Law

There were a couple interesting comments on slashdot recently about future computing technologies that might allow us to enjoy the continued price/performance improvements in computing and avoid the end of Moore's Law. Here's one that highlights some promising emerging technologies (my emphasis):

I see many emerging technologies that promise further great progress in computing. Here are some of them. I wish some industry people here could post some updates about their way to the market. They may not literally prolong the Moore's Law in regards to the number of transistors, but they promise great performance gains, which is what really matters.

3D chips. As materials science and manufacturing precision advances, we will soon have multi-layered (starting at a few layers that Samsung already has, but up to 1000s) or even fully 3D chips with efficient heat dissipation. This would put the components closer together and streamline the close-range interconnects. Also, this increases "computation per rack unit volume", simplifying some space-related aspects of scaling.

Memristors. HP is ready to produce the first memristor chips but delays that for business reasons (how sad is that!) Others are also preparing products. Memristor technology enables a new approach to computing, combining memory and computation in one place. They are also quite fast (competitive with the current RAM) and energy-efficient, which means easier cooling and possible 3D layout.

Photonics. Optical buses are finding their ways into computers, and network hardware manufacturers are looking for ways to perform some basic switching directly with light. Some day these two trends may converge to produce an optical computer chip that would be free from the limitations of electric resistance/heat, EM interference, and could thus operate at a higher clock speed. Would be more energy efficient, too.

Spintronics. Probably further in the future, but potentially very high-density and low-power technology actively developed by IBM, Hynix and a bunch of others. This one would push our computation density and power efficiency limits to another level, as it allows performing some computation using magnetic fields, without electrons actually moving in electrical current (excuse me for my layman understanding).

Quantum computing. This could qualitatively speed up whole classes of tasks, potentially bringing AI and simulation applications to new levels of performance. The only commercial offer so far is Dwave, and it's not a classical QC, but so many labs are working on that, the results are bound to come soon.
3D chips, memristors, photonics, spintronics, QC

I think Moore's Law is a steamroller. But, like the genomics sequencing technology highlighted in that post on Nuit Blanche, there are improvements just as fast, or faster than Moore's law. The improvements from better algorithms can yield exponential speed-ups too. Here's a graph (from this report) depicting the orders of magnitude improvement in linear solver performance:

Couple these software improvements with continually improving hardware and things get pretty exciting. I'm happy to live in these interesting times!

Open Source Aeronautical Engineering Tools

* permanent page with updates: Open Source Aeronautical Engineering Tools*

I was reading through some of the papers and cites related to OpenMDAO and stumbled across a couple interesting papers[1, 2] on open source software in aeronautics.

Abstract: Open source software has become an alternative to commercial software for industrial users. Industrial users adopting to OSS and the underlying concepts need to consider changing their software development practices and organization in order to benefit from the OSS model. These changes may involve both technical and non-technical aspects. Openness and collaboration with a community are two non-technical concepts that may require such changes, while evaluating OSS products or evaluating different strategies for integrating OSS products are two technical aspects that may require adoption.

The objective of this paper is to create an awareness for the adoption of OSS in an industrial context. OSS can be used in different ways. Four categories of using OSS are presented by discussing motivation, implication and experience for each category. The main conclusion from this work is that organizations should have a realistic expectation of both the designated benefits and extension of organizational changes necessary to adopt to OSS. This conclusion is based on observations from industrial organizations using OSS, including OSS in the aeronautics industry, as well as from observations reported in the literature on open source.

BG CFD Group at TU Berlin Youtube Channel

Some great visualizations of detached eddy simulations around cylinders, airfoils and cascades on the youtube channel for BG CFD group at TU Berlin.

11th World Congress on Computational Mechanics (WCCM XI)

The 11th World Congress on Computational Mechanics (WCCM XI) is coming up in July in Barcelona. Dates: abstracts due 29 Nov 2013, and full papers due 28 Feb 2014. The deadline for mini-symposium proposals has passed. There are some interesting looking ones (ECCM, ECFD),

• Uncertainty Quantification Techniques for Fluid-flow Problems
  
• Verification & Validation
  
• Frontiers of Numerical Uncertainty Quantification in Computational Mechanics
  
• New Trends in Topology Optimization
  

CFD V&V Case Resources

This list is from a discussion on the CFD group on LinkedIn.

• CFDOnline Validation and Test Cases: wiki with problem descriptions and some calculation results but not case files / grids (on the handful that I checked)
  
• ERCOFTAC Classic Collection: "The database has undergone some re-structuring and expansion to include, amongst other things, more details of the test cases, computational results, and results and conclusions drawn from the ERCOFTAC Workshops on Refined Turbulence Modelling. At the moment, each case should contain at least a brief description, some data to download, and references to published work. Some cases contain significantly more information than this." Registration required.
  
• LaRC Turbulence Modeling Resource: " The objective is to provide a resource for CFD developers to obtain accurate and up-to-date information on widely-used RANS turbulence models, and verify that models are implemented correctly. This latter capability is made possible through "verification" cases. This site provides simple test cases and grids, along with sample results (including grid convergence studies) from one or more previously-verified codes for some of the turbulence models. Furthermore, by listing various published variants of models, this site establishes naming conventions in order to help avoid confusion when comparing results from different codes."
  
• CFD and Coffee: a list of resources focused on compressible RANS V&V.
  
• cfd-benchmarks: benchmarks focused on room air distribution
  
• Saturne Test Cases: a list of test cases for Code Saturne, which also links the Working Group 21 AGARD database among others
  
• NPARC Alliance Verification and Validation Archive: grids available, but not much grid convergence info
  
• Vassberg's NACA 0012 Grids: Hegedus Aero code-to-code calcs based on 2009 AIAA paper / grids for NACA 0012
  

I added the link to a Hegedus Aero page detailing his code-to-code and grid convergence results based on a previous AIAA paper and those same grids. The AIAA paper that page is based on is worth a read.

SU2 and OpenMDAO Joint Workshop

Here's an announcement for an exciting workshop combining the open source computational fluid dynamics code SU2 with the open source optimization framework OpenMDAO.

Dear Colleagues,

This is a friendly reminder about the OpenMDAO and SU2 joint workshop on Sept.30 - Oct.1 in Durand 450 (Stanford University). So far, we have had great interest for both in-person, and virtual attendance to the event and we are looking forward to a great workshop. There is still room for more participants and we encourage you to join us for as much workshop as you can. As a reminder, no experience with the software is necessary.

If you plan on attending in-person or virtually, all participants should register at the SU2 homepage:
http://su2.stanford.edu

For virtual attendance, you must additionally register for the online webinar at the following locations:
Day 1: https://attendee.gotowebinar.com/register/7114168239445654785
Day 2: https://attendee.gotowebinar.com/register/208563415407826177

OpenMDAO is an open-source framework for Multidisciplinary Design, Analysis and Optimization (MDAO). Written in Python, it functions to integrate analysis tools into a single design environment allowing for the investigation of engineering designs and optimization of designs subject to user requirements.

SU2 is an open-source numerical PDE solver that can be used to solve the equations of fluid motion. SU2 has been used to simulate a range of fluid including wind turbines, supersonic business jets, and hypersonic re-entry vehicles.

These powerful codes are suitable for general purpose applications. They have been downloaded more than ten-thousand times and have a continually growing footprint in a growing field of research. Because these codes are freely available and can be modified to suit specific needs, they can be used as part of countless research problems.

If you’re interested in multi-physics simulations applied to design, this is the workshop for you!

Each day will have a morning session to introduce the software packages and an afternoon session to give hands-on experience working with the codes. These sessions will focus on linking OpenMDAO and SU2 for specific multi-disciplinary problems. The best utilization of the two packages will be awarded an iPad!. The deadline for the competition (only on-site workshop attendees) will be on Thursday afternoon (October 3rd) and the winner will be announced on October 4th in the SU2 website.

The agenda for the workshop is attached. Food will be provided both days.

We hope you are able to join us for this exciting and engaging event!

Cheers,
The SU2 and OpenMDAO development teams

I wish I could spare the two days to participate virtually; looks like a really good event.

Update: Workshop Materials

• OpenMDAO Introduction, Installation, and Quick Start
  
• SU2 Introduction and Installation
  
• SU2 Advanced Analysis Topics (presentation)
  
• SU2 Advanced Analysis Topics (mesh and configuration example files)
  
• Shape optimization using SU2 (presentation)
  
• Shape optimization using SU2 (ONERA M6 files)
  

Update: videos and more from the workshop.

SU2 Now on Git: Build and Install Latest on Fedora

As I wrote previously the Stanford University Unstructured (SU2) code is now available on git hub. Here's the steps to get the latest code, build and install it.

• Use the git client to clone the repo:
  
git clone https://github.com/su2code/SU2

  
• Follow the 'from source' install instructions for your system. It is should look something like:
  
./configure
make
make install

  This will give you (after a little help from the kindly devs) a plain-Jane version of SU2 for serial computations, without CGNS or Metis (graph partitioning) support.
  

Happy number crunching!