CFD Aerospace: Scaling and Parallelizing CFD Runs
Aerospace manufacturers operate at the leading edge of technology in materials, solid mechanics, fluid dynamics, electronics, and several other engineering disciplines. In an industry where 100% product reliability is expected and achieved on a regular basis, the product development process is at the core of company performance.
Across the aerospace industry, whether in spacecraft, aircraft, jet engines, or other systems, advanced engineering simulations allow engineers to cycle through various design iterations early in the design process, without investing in hardware. However, the complexity of simulation models is quickly increasing. For example, a 3D model of a high-pressure turbine rotor can contain millions of elements for a finite element analysis (FEA). Running these simulations can take hours or days, depending on the speed and age of the available high-performance computing (HPC) hardware. In addition, running design iterations in parallel to fully explore the design space remains extremely difficult. Engineers quickly run into HPC capacity constraints and have to wait in a virtual queue to run their jobs. This can lead to sub-optimal designs as engineers don’t have the tools or the time to fully optimize their designs.
To address these concerns on a new development program, an aerospace OEM turned to Rescale. With Rescale, this customer gained access to a comprehensive suite of simulation software tools, along with the capabilities of a fully secure, large commercial cluster on demand and at a fraction of the cost. Our customer used Rescale to perform multiple-parameter sweeps and designs of experiments (DOEs) to improve the designs of various sub-systems and critical individual components. The customer ran analyses on several models and used commercially available aerospace computational fluid dynamics (CFD) solvers.
To compute large-scale parallel simulations in a short time requires hundreds of clustered processors, resulting in significant upfront and maintenance costs for IT infrastructure . As a result, this type of simulation has largely remained out of reach to all but the largest of aerospace companies. Using the Rescale platform, the user was able to set up a full parameter sweep in a matter of minutes. For a typical job, upon submission by the user, the Rescale platform performed as follows:
- Hundreds of processors were dynamically provisioned within five minutes of job submission.
- The user-chosen solver was invoked across the entire multiple-processor cluster, simulating thousands of time steps
- Results were delivered to local servers for post-processing and analysis.
- All computing instances across the cluster were deleted upon completion.
Security concerns were fully addressed: Rescale jobs are run on SOC 2 and ITAR-certified infrastructure; all customer data was transferred in an end-to-end encrypted environment. Dedicated cluster instances were provisioned to ensure that data mingling did not occur. Finally, data was purged upon completion of the case.
In this case, the Rescale customer estimated that running these analyses on Rescale reduced runtime by >95% and saved hundreds of engineer-hours when compared to the alternative of running locally. Additionally, the ability to run much broader simulations revealed critical insights to the engineers that would have remained hidden due to time constraints had they run these jobs on local machines.
To learn more, contact us at email@example.com