Pedagogical notes
A set of pedagogical notes has been and is currently written, only in french unfortunatly, to introduce, as much educational as possible, the usual approaches of the Verification, Validation and Uncertainty Quantification (VVUQ). They are available below.
These notes are now available and gathered all at once as a HAL document which can be found and cited by following this link.
These notes are now available and gathered all at once as a HAL document which can be found and cited by following this link.
This note introduces the concept of design-of-experiments along with the their interest in the frame of numerical simulation and experimental planification. These technics are particularly useful for sensitivity analysis and surrogate model generation. Several methods are introduced (depending on the underlying hypothesis) and are applied to dedicated illustrative cases to show their pros and cons.
This note introduces the concept of surrogate model by proposing a generic methodology to use them, along with discussing the advantages of the different kind of models. From a very simple physical problem, few surrogate solutions are tested following the Occam razor, i.e. goign from the simplest solution to the more refined ones.
This note introduces the sensitivity analysis procedure discussing its main impact depending on the use case considered. A broad description of the methodology is performed, introducing the most well-known principles, but also using more advances concepts (for which this note provides mostly references). Using a dedicated toy model, several methods are applied, allowing for various analysis levels.
This note introduces the concept of deterministic numerical optimisation for numerical simulation and experimental problems. These methods, useful for themselves in decision making approach, are needed to be able to build surrogate models (to estimate the best parameters values). This note is not exhaustive but introduces the notions commonly used along with some of the classical technics.
This note introduces the parameters calibration concepts. Several technics are presented, depending on the underlying hypothesis on the physical model and on the uncertainty nature, going from the simple least square approach, to the Bayesian methods.