Skip to Main content Skip to Navigation
Theses

Formalizing and qualifying models for constraint-based embodiment design

Yoann Vernat
Abstract : Embodiment design involves: (i) decision making while the problem's data are poorly defined and quite imprecise, (ii) design space exploration that must be kept as large as possible, (iii) choices involving both discrete and continuous variables, and (iv) multidisciplinary optimization. Consequently, classical simulation models involved in the preliminary stages of the design process are usually improper to efficient decision-making. They are either too simple or specialized to make design choices: either the choices are distorted by modelling mistakes, or some possible solutions remain undetected. An approach is proposed to model formalization that is more consistent with decision-making in embodiment design. The methodology aims at obtaining models both parsimonious and accurate. Therefore, these models are easier to handle in preliminary design, and consistent with the designer expectations, in order to make the model treatment more efficient. The developed method involves: (i) a global approach based on functional decomposition to keep consistency between the models of the different components, (ii) the use of four criteria to qualify the models and to ensure their adequacy with the preliminary design goals, (iii) the use of model adaptation techniques to make design choices with the help of Constraint Satisfaction Problems (CSP) solvers. The four qualification criteria are: (i) the model parsimony, related to the minimum variables and equations number that represent the system behaviour, (ii) the model accuracy, estimating the adequacy between the model results and those provided by a reference model, (iii) the model precision, evaluating the domain extent of each variable, related to a lack of knowledge or uncertainty, and (iv) the model specialization which is an estimate of the restriction of the model applicability, in relation to the amount of information used in the model. The four adopted criteria appear relevant for modelling in preliminary design since: parsimony ensures model simplicity, specialization contributes to define the model application range and thus, the design space limits, and finally, accuracy and precision provide a measurement of model fidelity. Throughout the methodology, they provide a way to control the models until they reach the expected form. Finally, the methodology is illustrated by an example of electric vehicle battery and two models are compared for two different systemic levels.
Document type :
Theses
Complete list of metadatas

Cited literature [105 references]  Display  Hide  Download

https://pastel.archives-ouvertes.fr/pastel-00002003
Contributor : Ecole Arts Et Métiers Paristech <>
Submitted on : Friday, December 1, 2006 - 8:00:00 AM
Last modification on : Monday, October 19, 2020 - 10:54:39 AM
Long-term archiving on: : Thursday, September 30, 2010 - 7:48:50 PM

Identifiers

  • HAL Id : pastel-00002003, version 1

Collections

Citation

Yoann Vernat. Formalizing and qualifying models for constraint-based embodiment design. Engineering Sciences [physics]. Arts et Métiers ParisTech, 2004. English. ⟨NNT : 2004ENAM0027⟩. ⟨pastel-00002003⟩

Share

Metrics

Record views

315

Files downloads

1113