Genetic Fuzzy Prediction of Mass Perception in Non-Functional 3D Shapes

When designers create new forms they integrate both quantitative objective elements and qualitative subjective elements. However, users will generally react to these forms without knowing the intended Kansei integrated into them by the designer. Human beings are doted with a complex brain structure and it is argued that human attributes originate from three different levels of the brain: the visceral level; the behavioral level and the reflective level. This paper focuses upon the visceral level of reaction by automatically building a link between geometric properties of non-functional 3D shapes and their perception by observers. The link between geometry and human perception is created using a genetic learning algorithm combined with a fuzzy logic decision support system. Human evaluations of the non-functional 3D shapes against two contrary perception adjectives (massive versus lightweight) are used as the learning data set. The non-functional 3D shapes were designed by engineering design students from the Technical University of Denmark who were asked to design non-functional 3D shapes evoking either the adjective massive or light. Eight fuzzy models were developed: three (3) models constructed manually by the author and five (5) genetically generated. The fuzzy models were constructed using different sets of inputs of quantitative geometric properties. Combination of the different inputs resulted in different sets of fuzzy rules that can eventually be used as design guidelines for designers. The results obtained and presented in this paper are very promising. Correlations as high as 99% between fuzzy and human perception were obtained along with errors as low as 0.14 on a scale ranging from -3 to 3.