User tests for the convergence of a car silhouettes design model using an Interactive Genetic Algorithm

François Cluzel (1) , Bernard Yannou (1) , Markus Dihlmann (2)

1 LGI - Laboratoire Génie Industriel - EA 2606
2 US - University of Stuttgart

We propose in this paper a car silhouettes design model based on an Interactive Genetic Algorithm. The design solutions are indirectly encoded by a Fourier analysis approach. Thus after creating an initial population, the designer is able to evaluate himself generations of car profiles, in order to converge towards an individual which meets its requirements. An automatic test has already been conducted, but it is not sufficient to validate the model; we propose here 6 new tests. The two first ones are based on the user satisfaction towards the created design solution. The data were collected during a short subjective evaluation workshop. The 4 other tests use a similarity index (that quantify the perceived distance between 2 individuals) to process the same data and validate the model. These results reveal a really satisfactory behaviour and open perspectives on the understanding of the user perception.

Mots clés: Interactive Genetic Algorithm; Evolutionary Design; car profile; user test; subjective evaluation

Domaines: Mécanique [physics.med-ph]; Intelligence artificielle [cs.AI]

François Cluzel, Bernard Yannou, Markus Dihlmann. User tests for the convergence of a car silhouettes design model using an Interactive Genetic Algorithm. IDMME - Virtual Concept 2008, Oct 2008, Beijing, China. Virtual Concept_P104. ⟨hal-00796963⟩