Impression-Based Automatic Pattern Generation for Product Design Support

This research proposed a method for automatically creating patterns that match the impression desired by users in product design support. An impression rating experiment clarified the relationship between pattern physical aspects and impressions. Impression evaluation models that estimate the impression of patterns are implemented by training convolutional neural networks using experimental scores. Patterns are generated by evolutionary computation with impression evaluation models serving as evaluation functions. The validity of the automatically generated patterns is examined by the generation of evolutionary computation and the final generation patterns.