-
Notifications
You must be signed in to change notification settings - Fork 93
A template of User's Day talk
In this page, we suggest a template of table of contents for a talk at OpenTURNS's User's Day. This template is illustrated in:
"Construction de courbes numériques POD. Application en qualification de procédés CND." Anne Dutfoy & Loïc Le Gratiet & Bertrand Iooss. Journées Utilisateurs OpenTurns. 12 juin 2015.
This talk is available at:
https://raw.githubusercontent.com/openturns/presentation/master/userday2015/OT2015_LeGratiet.pdf
Introduce briefly the engineering application and its goals.
Introduce the physical model or the data at hand.
- What computer code ?
- Source of the data : measures, etc.
- Accuracy of the code / measures?
- Input variables : number of variables, description of the physical meaning.
- Output variable of interest : number of variables, description of the physical meaning.
- Computer cost of evaluation : in seconds, hours, days, months. Is parallel computing required ? What number of cores ? CPU or GPU ?
- Quantity of interest :
- Step A : description of the study
- Step B : inference of the input distribution, dependency (copula ?)
- Step C : central dispersion, probability of exceeding a threshold, metamodel,
- Step C' : sensitivity analysis, calibration, inversion,
- robust optimization, etc.
Usually, the description of the input variables is provided as a table.
Variable | Description | Distribution | Parameters |
---|---|---|---|
E | Young's modulus | Uniform | a=1, b=3 (MPa) |
F | The force applied | Normal | mu=2, sigma=3 |
L | The length of the beam | Normal | mu=2.5, sigma=2.6 |
I | Moment of intertia | Beta | alpha=2.5, beta=3 |
If the input variables have a dependency, the copula should be presented. Presenting the tools in terms of the ABC method is not required, but highly recommended if the steps are appropriate.
The results should be introduced for an audience who does not necessarily have a knowledge in the physical field. The management of uncertainty should be detailed a little further that usual, given that this is the topic of the User's Day.
- Number of model evaluations?
- Validation of metamodel?
- Validation of input distribution?
The OpenTURNS classes which have been used may be presented, with their strengths and weaknesses.