Reduced Order Modeling for 2D Static Fields

In this blog article we explore the capabilities of image-based Reduced Order Models, or 2D Static Field models as they are known, inside Simcenter Reduced Order Modeling (ROM), and go through the necessary steps to export 2D field data from Design Manager. Field ROMs like these are one way to bring CFD results into a running twin, as our guide to reduced order models in a digital twin explains.

Simcenter Reduced Order Modeling is a software tool which creates mathematical models based on results from either simulation or testing. These simplified models replicate the behavior of the more complex simulation models with significantly reduced computational resources. There is however a caveat, the result-space needs to be sufficiently large for the ROM to be able to replicate the desired behavior.

In previous posts we have shown how Simcenter ROM can be leveraged to capture the time dependent behavior of scalar results, e.g. temperatures and pressures, to create fast running models using multiple input variables. Time Dependent Neural Networks with Simcenter Amesim’s ROM Builder provides an example on this. In this article we instead focus our attention on configuring simulation models in Simcenter STAR-CCM+ to export 2D field data which we will later use in Simcenter ROM to train models.

The intention with the approach outlined here is not to run multiple CFD studies with the sole purpose of creating image-based Reduced Order Models. Rather, as a design study is carried out it is possible to add to the value of the analysis by creating 2D Static Field models. The Reduced Order Model offers quick access to scalar images with the additional capability of predicting behavior between known results through interpolation.

The video below goes through the necessary steps and setup in Simcenter STAR-CCM+ required for image export to Simcenter Reduced Order Modeling. The workflow shown here starts with an existing steady-state model of a graphics card mounted in a computer case and air-cooled by two fans. The model used two physics continua, one for the air enclosed within the case, and one for the multi-part solid comprising of the card, heat sinks, GPU, capacitors and so on. The fans’ rotational speed, and the solid component’s heat rejection are used as input parameters during the design sweep to assess surface temperatures on the board.

We hope you have found this article interesting. If you have any questions or comments, please feel free to reach out to us on support@volupe.com