Optimal design of experiments for the modelling of food processes
Résumé
of unknown coefficients in dynamic models. This approach is validated over 4 applications, 2 of which being taken
from literature: (A) Reproduction of the pharmacokinetic results from [1]: diffusion of a drug between 2 compartments;
(B) Reproduction of the fermentation results from [2], non sequential optimal design of 6 fermentations
trials. (C) Order 1 reaction of ascorbic acid in stewed apples; (D) Rice drying, most complex case with no analytical
solution: optimal use of experimental device to identify unknown heat and mass transfer coefficients
The objective is to diminish the experimental effort needed to make this identification within acceptable confidence
ranges. After each experiment, the next experiment is A-, D- or E-optimally designed. Our design of
experiments can take into account all experimental constraints and experimental results of al previous experiments
to calculate best experimental conditions and to obtain smallest uncertainties on estimated parameters. In all contexts
(A-D), this methodology is applied to a simulated noised experiment, and its stability and convergence is
shown to be effective. In the last context (D), this methodology is also applied to a drying pilot plant; the identification
made with only three real experiments with non-constant drying conditions are shown to be as effective as
an identification based on two-factor three-level grid of nine experiments at constant conditions.
Domaines
Sciences du Vivant [q-bio]
Fichier principal
Full_6p_ICEF11_Goujot_{B74AF9CA-8CFB-4AB8-9907-54D487091883}.pdf (203.1 Ko)
Télécharger le fichier
presentation_ICEF11_print_{CC484F50-7E76-460F-A159-089C659EB030}.pdf (216.88 Ko)
Télécharger le fichier
Origine : Accord explicite pour ce dépôt
Origine : Accord explicite pour ce dépôt