Surface Field

Modified on Fri, 16 Dec 2022 at 03:06 PM

Description

Surface Field is an AI model that learns and predicts fields e.g., pressure or stress from 3D datasets.


Application

A lot of engineering R&D is done using 3D simulations, such as Finite Element Analysis (FEA) or Computational Fluid Dynamics (CFD). Most of the time these simulations involve 3D fields attached to a geometry such as a stress (FEA) or pressure field (CFD). Depending on the complexity of the problem and accuracy required, these fields might be costly to compute. 

The surface field model allows engineers to leverage historical 3D simulation data and learn the relationship between 3D geometries and their associated surface fields via a surrogate surface field model. 

Once the surface field model has been trained, it can be used to predict field data for new 3D geometries. A surface field model trained on historical simulation data can approximate simulations and provider a faster way to generate approximate results without needing to perform costly simulations.


How to use

Prerequisites: You must have some 3D data containing field data loaded into the notebook before you can use the surface field button.

3D data
3D data containing surface field to train the model with.
Inputs
Inputs to the model (by default the point coordinates of the mesh will be populated by default. Other fields can also be used as input). You can also attach scalar values to a 3D geometry using the Attach Operating Conditions step. These values (e.g. wind speed, ...) can then be selected as inputs to train the model.
Outputs
Outputs to the model (must be one or multiple fields).
Number of training steps
Parameter that controls the length of the model training phase. More steps means a longer training time, but the model will be more likely to converge and have higher accuracy.
Augment Data
Check box that decides whether to include data augmentation as part of the training. Data augmentation uses jitter, rotation and scaling to to allow the model to learn from a greater variety of input data. It can be used to improve the model if the fields predicted do not depend on orientation etc.
Surface Model Name
Choose the name of the resulting surface field model.

Once a model is trained, you can use the step View Surface Field Prediction to predict surface fields for new geometries, or to evaluate the surface field model on a test set. Read here to learn more about how you could evaluate surface fields.


Examples

Example 1 

The screenshot above was applied on wind turbine data. The figure below shows:

  • On the left, a new geometry that has not been seen by the model (with no field provided)
  • On the right, the same geometry, along with the model prediction of the corresponding stress field

Example 2

In this example, a similar model was trained, using the Windspeed and TipSpeedRatio as additional scalar inputs. As a result, when making predictions, these parameters can be changed (see sliders at the top of figure below), and the pressure field will be re-calculated accordingly.


Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select atleast one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article