HiFi-TURB Project - Turbulence Modeling with AI and ML

Today, a significant challenge in applied fluid dynamics is the lack of understanding of turbulence-dependent features. Leaving us with sparse industrial confidence in applying CFD to applications such as flow detachment over an aircraft wing, shock-boundary layer interactions, etc. Improving the capabilities of complex fluid flow models can reduce energy consumption, greenhouse gas (GHG) emissions, and noise radiations from aircraft, cars, and ships. These complex models can favorably impact the economy and industrial leadership in a highly competitive manner. Hence, understanding, modeling, and predicting turbulence phenomena is key for efficient and environmentally safe design. To make this a reality, the HiFi-TURB project sets out on a highly ambitious and innovative program to address some deficiencies in advanced turbulence models.

The HiFi-TURB project rests on the following pillars of excellence : 

• The exploitation of high-fidelity LES/DNS data for a range of reference flows that contain key flow features of interest.
• The application of novel artificial intelligence and machine-learning algorithms to identify significant correlations between representative turbulent quantities.
• The guidance of the research towards improved models by world-renowned industrial and academic experts in turbulence modeling.

This consortium is a collaboration between major aeronautical companies and software editors, with Cadence Design Systems acting as a coordinator. Well-known research centers and academic groups, including ERCOFTAC, act as a source of turbulence expertise. 

---

Case Study: Studying Turbulence with High-Fidelity Simulation and Machine Learning

The large-scale availability of high-performance computing (HPC) opens the door to a truly novel approach to turbulence model development. This study applies artificial intelligence (AI) and machine learning (ML) techniques to a database of high-fidelity, scale-resolving test case simulations containing features of separated flow regions or complex 3D flows. Figure 1 shows an example of a flow field used as the basis for the turbulence modeling task.

Figure 1. Flow structure of T161 cascade (typical flow field used as input for the turbulence model improvement).

The huge amount of data generated in these simulations requires a new data mining approach. This is where neural concept brings in its toolchain based on deep learning to analyze large amounts of data provided by 3D scale-resolving simulations. Using geometry-based variational auto-encoders (VAE), Cadence CFD was able to gain insights into correlations between averaged flow variables. The VAE compresses the data physically meaningfully into ‘embeddings’ and then reconstructs the original input from the compressed data. This is done with a high accuracy, which allows the use of the ML models as a replacement (surrogates) for the original data. The advantages are easier data handling, possible exploitation of data mining, and analysis techniques that can provide insights into the physics involved in the study.

Figure 2. Investigated simulation domain (left) and statistical analysis of quantities (right).

Figure 2 is an example of the possible analysis. The symbol colors on the 2D plot correspond to the ‘embedding’ value and are the same in the 3D view (left) and in the 2D plot (right). Points of the same color have the same value for all the considered physical quantities. The 3D view colored by the embedding value is a global statistical representation for several physical quantities over the investigated domain. Both plots provide a new perspective on the flow behavior using a machine learning model. 

---

HiFi-TURB Work Program

Work Package 1: Management 

Task 1: General Coordination/project steering              
Task 2: Knowledge exchange (inter-communication / website) 
Task 3: Dissemination/exploitation

Work Package 2: Further improvement of HOM towards reduced CPU cost and curved grid generation

Task 1: Reducing CPU cost for high-fidelity LES/DNS
Task 2: Towards industrial curved-grid generation techniques

Work Package 3: Generation of high-fidelity LES/DNS data sets for identified physical phenomena

Task 1: Selection of basic and industrial relevant test cases with identified physical properties
Task 2: Generating new h - High-fidelity LES/DNS data sets
Task 3: Evaluation, reliability, and quality of LES/DNS data

Work Package 4: Feature detection and advanced analysis of LES/DNS data

Task 1: Analysis of basic turbulence- averaged data via data-driving methodologies
Task 2: Analysis of time turbulence data via AI and deep learning methodologies connected to HRLM and WMLES

Work Package 5: Turbulence modeling assessments and improvements – monitored by WP5 Task Group

Task 1: Develop, improve, and assess EARSM turbulence models
Task 2: Develop, improve, and assess DRSM turbulence models
Task 3: Develop, improve, and assess wall models for WMLES and Hybrid RANS-LES

Work Package 6: Validation of new turbulence models applied to representative and industrial-relevant test cases

Task 1: Validation on external flow configurations (High-Lift and Drag Pred.  WS cases) - baseline to new
Task 2: Validation of internal flow configurations for fixed (diffusor) and rotating cases - baseline to new
Task 3: Assessment and recommendations

Work Package 7: Management of the LES/DNS databases for open accessibility (ERCOFTAC)

Task 1: Definition of database criteria and implementation rules
Task 2: Creation and management of the LES/DNS database
Task 3: Integrating results of WP6 in the ERCOFTAC Wiki Knowledge Base

---

HiFi-TURB Consortium Members

• Cadence Design Systems Belgium (coordinator)
• Dassault Aviation
• SAFRAN S.A.
• Imperial College London
• ANSYS Germany
• Cineca Consorzio Interuniversitario
• Barcelona Supercomputing Center – CENTRO National de Supercomputacion
• Centre de Recherche en Aéronautique ASBL - CENAERO
• Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique - CERFACS
• Office National d’Etudes et de Recherches Aerospatiales - ONERA
• Deutsches Zentrum für Luft und Raumfahrt - DLR
• Università degli Studi di Bergamo
• Universite Catholique de Louvain
• European Research Community on Flow Turbulence and Combustion – ERCOFTAC
• Central Aerohydrodynamic Institute -TsAgi

---

If you would like to try scale-resolving simulations using Fidelity CFD for your CFD applications, request a demo today!