In the DOE mode, it takes ~15s to change from normal to Gaussian, for example.
To make any change, it just takes a lot of time.
It does not take any CPU power when it is "thinking." When switching to Monte Carlo simulation, it is all very fast and normal.
What is the cause of this?
|
In a DOE, we generate a matrix of design points. When we are finished solving each of these design points, we fit the solution using a quadratic response surface. Monte-Carlo does not generate a matrix of design points up front, nor does it do any fitting to approximate the results. Instead, with Monte-Carlo, each sample point in optimization is an FEA solve. When you have more than 10 input parameters, the DOE matrix generation in DX at v10.0 is very slow. This is because of the new DOE matrix types we added at v10.0. Some very time consuming calculations are done to find the `optimal` DOE for certain criteria; as such it takes a long time to regenerate this matrix when you change the properties of the input parameters (such as type of distribution, upper bound, lower bound, etc.) We plan to address this issue in the 11.0 version. Until then, the customer can open the options panel for DX and under `Automatic Design Points` they can change the Central Composite Design `Design Type` to `Rotatable` or `Face Centered.` They should then be able to update the input parameters without any slowness. When they are ready to solve the design points, they should go back to the options panel and change the `Design Type` back to `Auto Defined (Default).` |
||
|