How Meancalc Achieved 5.3X Cost Savings for Automotive Aerodynamics

Due to their superior accuracy, high-fidelity computational fluid dynamics (CFD) approaches are rapidly becoming the standard for automotive aerodynamics. These unsteady methods must be time-averaged for analysis, which can introduce significant uncertainty if done incorrectly. At the same time, the efficient use of computational resources is critical. Teams are often allocated a limited number of CPU hours to conduct complex CFD simulations, needed for the design and optimisation of  vehicle performance. 

These pressures encourage teams to optimise their simulation processes for speed on the one hand, while also fostering a keen attention to detail that enhances the accuracy of their findings. Without a thorough understanding of statistical convergence and the potential impact of biases, teams may operate with flawed data, which can have significant repercussions for their competitiveness. 

In this context, the introduction of an advanced tool capable of intelligently controlling time averaging in CFD simulations presents significant value. This tool can help teams identify and resolve errors, ensure the effective convergence of simulations, and ultimately improve the reliability of their results. Consequently, this enables informed decision-making. At the same time, significant cost savings and productivity gains can be made. 

In this article, we will explore the advantages of the Meancalc runtime control tool. We analyse how it was deployed to create a pioneering aerodynamic dataset for Machine Learning, delivering significant benefits compared to traditional, fixed-runtime approaches:

• 5.3 times reduced computational cost while maintaining equivalent accuracy

• 3.1 times more accurate simulations at equivalent cost. 

These features highlight the remarkable potential of Meancalc to optimise CFD simulation processes.

Dilemma of Time Averaging in CFD Simulations

When time-averaging CFD simulations, two primary types of errors can emerge: bias error and random error. These concepts are illustrated in the figure below for a motorsport CFD simulation.

Bias error occurs when the initial transient phase of a simulation is included in the time-averaging process, as shown by the grey box around the early portion of the data. Using the analogy of aiming shots at a target, the effect of this error is illustrated on the left: data points are consistently skewed in one direction. To eliminate bias, it is essential to start time-averaging only after the transient effects have decayed and a statistically steady state has emerged, effectively discarding the initial portion of the data. This practice significantly improves the accuracy and reliability of the results.

Once the bias is removed, the middle target shows the improvement, however there is still a significant degree of scatter. This is random error, due to insufficient data points to compute a reliable mean. To reduce this random error, the simulation must be run for longer to obtain more data, illustrated by the tighter clustering of data points in the right-hand target plot.

However, correct time averaging is challenging for two reasons. Simulating a longer duration increases costs, making an already expensive process even more burdensome. Furthermore, the correct times to start and stop the time averaging process can’t be determined in advance, as this varies significantly from one case to another. Mastering these issues is essential for enhancing the fidelity of CFD simulations while effectively managing resource demands.

How Meancalc Can Help with Time Averaging in CFD Simulations

Meancalc is a time series analysis software that utilizes innovative algorithms to enhance performance. It identifies the initial transient, helping users determine the optimal moment to begin averaging. Additionally, it quantifies random error, indicating when enough data has been collected for reliable averaged values. When integrated with simulation software, Meancalc can halt simulations once the desired accuracy level is reached. This powerful approach streamlines the computational process, minimising unnecessary run times and optimising efficiency. Simulations run no longer than necessary, leading to significant time and cost savings.

However, it’s essential to consider that some users may currently be running very short simulations to minimise cost and maximise productivity. These teams might find that some simulations run for longer when employing Meancalc. It is essential to understand the dangers of bias or random error in results, which can lead to erroneous design choices. For example, the simulation shown in the above figure appears to settle to a reliable mean value, only to suddenly correct down to a significantly (6%) lower value after around 20 convective time units. Meancalc runtime control gives peace of mind in such situations, correctly identifying cases with long  initial transient and ensuring that data is processed correctly.

🔧 Integration and Ease of Use

In addition to its analytical power, Meancalc is designed for seamless integration into existing CFD workflows to enable automatic runtime control. It can be deployed either using simple Python scripts or REST API calls, making it easy to automate and scale across different projects. Meancalc is compatible with any simulation software, with integration for OpenFOAM ready to go. Whether you're working in a bespoke setup or a standardized simulation pipeline, Upstream CFD can help you to integrate Meancalc in your own specific environment.

Insights from the DrivAerML Dataset

The DrivAerML dataset was established to promote research in machine learning applications for automotive aerodynamics. It is a collaborative effort that utilizes the baseline DrivAer Notchback geometry and wind tunnel data provided by Ford for workflow validation. Siemens Energy and BETA-CAE managed geometry parameterization and automated meshing using ANSA, creating consistent meshes with approximately 160 million cells. AWS provided the computing power and storage, while Upstream CFD implemented an automated high-fidelity simulation workflow. Meancalc runtime control was deployed to optimise HPC costs and ensure consistent statistical accuracy across a dataset that includes detached-eddy simulation (DES) solutions for 484 geometry variants, along with complete flow fields, surface data, and flow visualisation images.

How does Meancalc runtime control compare to fixed run times?

The benefits of using Meancalc runtime control hinge on the fixed runtime settings chosen for comparison, so we explore two distinct scenarios: one emphasising cost and the other focusing on error. When the fixed runtime setup is adjusted to align with Meancalc’s cost across all cases, Meancalc achieves an impressive 3.1x increase in the number of accurate simulations. On the other hand, when the fixed runtime is tuned to match Meancalc based on accuracy across all scenarios, Meancalc provides a remarkable 5.3x cost reduction.

The histogram generated for the DrivAerML dataset indicates that while most simulations converge in relatively short durations, there is a long tail of cases that require significantly extended run times to achieve the targeted accuracy level. This discrepancy highlights a significant opportunity for cost savings, particularly in cases that converge quickly. Meancalc exploits these trends, enabling CFD practitioners to allocate computational resources to cases where they will yield the most significant benefit.

How to Install Meancalc Runtime Control?

Meancalc offers the flexibility of being installed on-premises, making it particularly well-suited for enterprise clients, or accessible via a cloud API. It is compatible with all simulation software, with integration for OpenFOAM currently available and additional integrations under development or available upon request. To facilitate a successful implementation, we provide bespoke customer success projects designed to integrate Meancalc into existing simulation workflows and to identify application-specific best practices, ensuring efficient and effective deployment.

Conclusion: Make Every Simulation Count

In a resource-constrained engineering landscape, the ability to run accurate CFD simulations efficiently is more important than ever. Meancalc empowers teams to optimise simulation run times, eliminate costly errors from transient phases, and make data-driven decisions with confidence. By integrating seamlessly into your workflow and quantifying statistical reliability in real time, Meancalc ensures that every CPU hour delivers maximum value.

By choosing Meancalc, you can achieve impressive cost savings for the same level of accuracy, or enjoy significantly greater accuracy for the same cost. Discover how Meancalc can help you reduce costs, improve accuracy, and streamline your path to better engineering outcomes.