DNS, LES, Or RANS Simulation For Your Next Automotive Design

The automotive industry is constantly advancing, with new innovations and developments emerging almost every day. The industry is committed to creating a more sustainable future with the growing trend towards electric-powered vehicles and mass production. In 2022, the global production of motor vehicles increased by 5.7%, producing 85.4 million vehicles. However, the industry faces the challenge of accommodating wind tunnel testing or prototype testing for multiple designs or even the smallest design changes, such as adding a new headlight, splitter, or side skirt, while meeting fuel efficiency standards. As a result, simulation-based approaches are becoming more popular since they can significantly reduce the number of required experiments by predicting flow field information and performance-related quantities of interest, such as drag coefficient.

Understanding turbulence in fluid flow and replicating the same through computational fluid dynamic (CFD) simulations involves the use of different turbulence models. The choice of the right turbulence model based on the automotive application and resource availability can help speed up the design cycle. In this blog post, we will discuss the different types of turbulence models, when to use them, and how our Cadence Millennium M1 CFD supercomputer is a disruptor when it comes to high-fidelity turbulence modeling for complex geometries.

Modeling the chaos in fluids

Turbulence is the term for an irregular flow characterized by chaotic changes in pressure and velocity. We experience it in our daily lives, and it plays a vital role in various engineering applications such as aerodynamics, combustion, mixing, heat transfer, and more. However, the Navier-Stokes equations, which govern fluid dynamics, are highly nonlinear partial differential equations, and there is no theoretical solution for turbulent flows. Modeling and simulating turbulence can be challenging since it involves a wide range of spatial and temporal scales. Usually, the large eddies carry most of the energy produced by turbulence, while the small eddies dissipate this energy into heat. This phenomenon is known as the “energy cascade.”

Different turbulence modeling approaches have been developed over the years, and the three most common ones are briefly explained below.

• Direct Numerical Simulation (DNS): In DNS, turbulence is resolved at all scales using a fine grid and very small time steps without any models or approximations. The computational cost for DNS is prohibitively high, but the results are most accurate. DNS simulations are used as “numerical experiments” to provide comprehensive information on the turbulent flow field.

• Large-Eddy Simulation (LES): As the name hints, this turbulence modeling technique solves the large eddies and models the small eddies with universal characteristics. While reducing the computational cost by skipping the smallest length scale, LES simulations still detail the fluctuating components of turbulence that evolve over time.

• Reynolds-Averaged Navier-Stokes Model (RANS): The RANS equations were derived by taking a time average of the Navier-Stokes equations. The turbulence effect is simulated by modeling the additional unknown Reynolds stresses term. RANS simulations resolve the mean flow and average out the turbulent fluctuations, thus are much more cost effective than the other two approaches.

Making the right choice – DNS, LES, or RANS

Choosing the right turbulence model is a crucial aspect of any simulation, and it primarily depends on the purpose of the simulation, the Reynolds number of the flow, the geometry, and the available computational resources.

When it comes to academic research, DNS simulations provide the best results for understanding the fundamental mechanisms and structures of turbulence. DNS is suitable for cases with low Reynolds numbers, but it’s not a practical choice for most industry applications due to the enormous time and resources required.

On the other hand, LES is a feasible option for handling industrial cases, which usually involve complex geometries with high Reynolds numbers. The high-fidelity results produced by LES enable designs with the next level of performance improvement, which is critical in the competitive automotive market.

Compared to LES, RANS simulations are less accurate due to the extent of the approximation made. However, the balance between fidelity and computation expense makes RANS a common solution for industry users with limited computational resources and time for simulation. This method is also widely used when multiple cases need to be analyzed within a short time frame.

High-fidelity LES simulation

High-fidelity LES simulation provides exceptional accuracy and stability. However, the high cost associated with LES has inhibited its practical application, with a single simulation consuming thousands of CPU cores for days. Fortunately, the Cadence Millennium M1 CFD Supercomputer presents a promising solution. It combines advanced graphic processing units (GPUs) and GPU-native Fidelity LES Solver with scalable high-performance computing (HPC) system configuration, thus ensuring the most efficient high-fidelity CFD simulations.

Such a turnkey solution enables the wide application of LES to various engineering domains, including aerospace, automotive, and turbomachinery. It is designed to efficiently tackle the toughest challenges in fluid dynamics by accurately solving complex problems in aerodynamics, aeroacoustics, combustion, etc.

Conclusion

Millennium M1 brings a paradigm shift to the automotive industry by leveraging GPUs to drastically reduce the time required for LES simulations from days to hours while maintaining high levels of accuracy. Moving forward, it is predicted that by 2032, the automotive industry will experience a surge in vehicle sales due to increasing income levels and the resolution of issues such as the shortage of semiconductor chips faced between 2020-2022. As sustainability and innovation remain top priorities for the industry, high-fidelity simulation technologies that deliver precise solutions in less time will be in demand.