Multi-Objective Interactive Evolution Of Fashion Accessories For 3D Printer
Authors: Halimah Manja and Jason Teo Tze Wi
Although Evolutionary Computing has already expanded their realms into solving real life problems, there has been significantly less emphasis on their applications towards design of everyday aesthetic objects. Fashion accessories cannot be defined using a single objective fitness function but rather it requires a multi objective fitness function in order to solve this kind of problem. Prior research has used Genetic Algorithms to evolve the design. However, the typical Genetic Algorithm requires a genetic representation of the solution domain and a fitness function to evaluate the solution domain. Therefore, an Interactive Evolution Computing approach or other words as known as aesthetic selection in general is one of the methods of Evolutionary Computing that uses human evaluation. With the addition of multi-objective optimization, the challenge is which optimal decision is to be taken. By implementing a Pareto optimal solution method, the multi-objective approach will not be dominated by any other solution. In the experiments, the tuning parameters are based on the mutation rate and scale factor. Mutation rate had a low value if it was added to the gene, therefore enhancement of the shape mutation was more successful when being multiplied with the scale factor. The results of the experiment produced different shapes and many evolved shapes were successful in the optimization of their shapes. Subsequently, each successfully evolved object from the respective runs was chosen to be fabricated using 3D printing. The results from the human evaluation show that the rating of the physical object dropped compared with the nonphysical object in silico. The shape that was printed was not as smooth as how it appeared in the system. Therefore, the conclusion was that the user tended to give higher ratings to higher number of faces. However because of the low resolution of the available 3D printer, the object that was being printed was not as smooth as in silico. As the result, the human evaluated printed objects’ ratings were not as high as the in silico evaluations.