In this video, we will introduce and demonstrate the use of surrogate models in Tecplot Chorus.
Surrogate Models approximate expected results for cases you do not have actual results for, by using the correlations between inputs and results of cases that have been produced. This is especially useful for highly dimensional CFD projects, which would otherwise require a large quantity of time consuming simulations in order to fully explore the design space.
A surrogate model provides a faster way to:
- Confirm your results
- Identify trends
- Establish correlations
- Locate areas of interest from an initially limited number of cases
This video uses the example project #1 in your Tecplot Chorus installation. In this project the same geometry was used in a series of wind conditions where Alpha (the angle of attack), Beta (the yaw angle), and the Mach number of the vehicle were tested. We will be investigating how lift and drag are affected with these three input variables.
We begin with a line plot using symbols of Alpha versus Drag Coefficient for our existing cases. We can further refine our plot by grouping the data by Mach Number and sorting the data by Alpha.
To generate a surrogate model, select Training to define which variables to use. At the top of the dialog, select the type of surrogate model created. For now, keep the default of quadratic response surface. Note that only the independent variables for the project show up here. If you do not see any variables for your project, first ensure that your independent and dependent variables are set up correctly from Project > Change Variable Nature. To display the surrogate model, toggle on Show. If the toggle for show doesn’t appear, ensure that only one or two independent variables are being used in the plot (for help, see the Surrogate Model Section of the Tecplot Chorus User’s Manual).
Once the model has been trained, we can set the location where the surrogate model is evaluated by selecting Evaluation Point to adjust the appropriate variable values. The same can be done by right-clicking a data point in any view and selecting Set Surrogate Model Evaluation Point to have the surrogate model evaluated with independent variable values of that specific case.
Notice how for this plot, adjusting Alpha or Mach No. has little to no effect on the surrogate model, as we are already plotting our data points against Alpha, and grouping them by Mach Number. Increasing Beta, however, increases the steepness of the surrogate model curve.
Adjusting variable filters in the Filters sidebar will only affect the cases plotted, and will leave the surrogate model curves untouched.
To adjust the range and number of the surrogate model curves we wish to plot, select Range and Sampling. From here we can define region the surrogate model curves are plotted within by setting a range of values and number of points, or by using a list of specific points.
If we bring up another line plot, this time with lift coefficient versus Alpha, we can see how the Range and Sampling is unique to each plot, whereas the surrogate model and the evaluation point itself is a global value for the entire project.
Surrogate models can also be viewed in 3D scatter plots, however instead of a curve, a 2D surface is plotted.
To change the type of surrogate model used, and how the curves and surfaces are drawn, select Training and choose from the options at the top of the dialog.
Currently, we have been using a quadratic response surface, which is a least-squares approximation using a second-degree polynomial and is good for creating a smoother curve or surface for projects with a large number of cases.
Kriging is a statistical interpolation that yields the best linear unbiased prediction of values between cases using a covariance model. It is more computationally intensive and is best used for a small number of cases, but it produces curves and surfaces with greater precision.
This concludes the tutorial on surrogate models.
Thank you for watching.
Try Tecplot 360 for Free