A posteriori identification of dependencies between continuous variables for the engineering change management
Résumé
The objective of this document is to contribute to the modelling of engineering changes and their propagation. Usable in preliminary redesign activities, a new approach is suggested that allows greater efficiency. Given a product model, the idea is to use the experiments to calculate in advance the consequences of potential changes in continuous variables. These consequences are collected, analysed and structured in a dependency model, noted [Math Processing Error], composed of a dependency graph [Math Processing Error] and its associated set of influence functions [Math Processing Error]. Bilateral influence functions, associated with arcs, quantify the dependencies between node pairs. Multilateral influence functions, identified by the application of the total differential theorem, define the dependencies between a node and all influential nodes on which it depends. Finally, the relative error is calculated by following the infinitesimal assumption of the total differential for each variables. Our findings show that such a dependency model is informative and allows effective prediction and evaluation of changes. The approach is illustrated by a geometric bicycle model. The results are discussed and future areas of research are finally presented.