Published: 2008-02-15
ISBN:
ISSN: 1650-3686 (print), 1650-3740 (online)
Automotive design involves very complex surfaces that are constructed under rigorous constraints; a shape must not only appeal to a customer from an aesthetic point of view; but it must also give the car good aerodynamics or even serve as a structural element; and perhaps most important of all; it must have such properties that it can be mass-produced and easily handled in assembly for a low cost. The automotive industry has driven the development of information systems to support these processes; and there is no shortage of systems today – whether it is in CAD; CAM; CAE or PLM; there are very advanced tools for analysis available on the market. However; one area still remains relatively uncharted; and that is the analysis of emotional content in products. This has become a very important part of designing products that stand out; since mass-production and fierce competition has brought customers products of similar level of functionality and quality; they are now in many cases quantitatively satisfied and will choose the product which appeals to them on an emotional level. On-going research in Kansei Engineering (KE) or Design for Emotion has resulted in a number of methods to measure the emotional impact of design features in various products on human subjects; and use of this data to construct systems with mathematical models to map emotions to design has been presented. The main idea in KE is ultimately to construct systems that are more ‘sensible’ to human emotion.
However; KE methodologies are difficult to implement correctly; the inherent difficulties of extracting and transforming emotions into parameters suitable for analysis can lead to results with many sources of error. A typical task in KE is to define parameters of an artifact; which are consequently used as vectors to span a design space. This is done explicitly without any clear guidelines; and I argue that this approach is unsuitable for complex artifacts. I will refer to this as ‘explicit parameterization of design features’; or abbreviated as EPDF. EPDF is very much a ‘common sense’-approach where an artifact is examined; usually visually; to identify the obvious main design features. The features are then represented by a parameterization so that an analysis can take place. EPDF therefore requires two error-prone steps to transform a real artifact into a set of numbers – that is; a designer must know which features are important; and how to describe them numerically without loss of the emotional content the artifact contains. Naturally; it is very difficult to identify which particular features are most important with any degree of confidence.
In my previous work (Nordgren; 2005) I have used the silhouette line of cars as a descriptor of their unique character; but I came to realize that modern cars exhibit very small deviations in this descriptor as they are all subjected to the same spatial and aerodynamic constraints. The global shape of modern cars does not have a large degree of variance between cars of the same type (compact; sedan; wagon etc); instead the brand identity is expressed as small deviations from a mean; with characteristic proportions and lines defining details in the shape. For example; slightly bigger wheels or a lowered body can signify a much more “sporty” car; and a descriptor that does not consider the whole car with all its proportions; lines and details; will thus not be able to sufficiently express the car’s character. This poses a challenging problem. It is possible to describe the shape of a car in all its minute details; but this will lead to a design space with thousands of dimensions. A highdimensional space would not allow for efficient measurement of Kansei in a survey; since each vector or parameter must be accounted for. It would also be very timeconsuming to extract and parameterize all those subtle details in a design; thus rendering this approach useless in practical applications. Therefore an accurate; low dimensional descriptor for complex shapes is needed. Furthermore; this descriptor should be defined implicitly by the properties of the shapes; and not by a designer’s guesses. In this paper I will present a method to define such a descriptor by performing eigenmode analysis on pointclouds from a set of 3D-shapes spanning a design space. I will restrict my discussions to parts in the automotive domain; exemplified by the front bumper below; but I believe that my methodology is valid for most designs involving complex 3D-shapes; as long as the shapes belong to the same topology.
In this paper I use the terms ‘surface’; ‘shape’ and ‘design concept’ rather loosely; depending on the context of presentation; but I think it should be clear that I in all cases refer to the front bumper that was analyzed with the method.
