設計早期階段構想發展的認知行為及電腦模擬之研究

Abstract

設計的行為是人們在複雜心智下，所產生有構思、有計畫的創造性活動，同時也包含了許多理性且合乎邏輯的思考活動。在設計過程中，形的認知與轉換，以及之間所蘊含的知識，是設計思考中最重要的關鍵。以目前的專家系統(expert systems)或知識基礎系統，全然亳無能力來處理人類在設計行為中的視覺認知的問題，電腦在複雜運算的速度和精確度遠勝於人腦，但人腦在對於不完整或不正確的輸入，仍舊能進行處理，給予最佳的建議。 近年來許多研究都從認知實驗中，來探討設計師在設計草圖階段中的設計行為。在了解一些視覺認知的理論後，進而用神經網路初步模擬一些圖形的辨識、轉換及聯想(Coyne and Newton, 1989; Liu, 1993,1995, 1996a, 1996b; Rueckl et al., 1989)；另一些研究也嘗試著從設計準則方面找出一些在設計過程中的知識，進而用電腦來模擬(Stiny, 1980; Akiner, 1986; Liu, 1991)。在設計師發展草圖階段中，這二方面的現象都是快速並且反覆地在大腦中運作著，因此這二方面的研究會使人們對於設計的行為有進一步的了解。 本論文第一階段的目的是在探討設計構想草圖階段，設計師對單一物件產品造形的聯想以及設計條件之間的現象，同時將此現象試著用神經網路來模擬，試著電腦架構出設計師對於圖形的思考過程，以檢驗設計思考是否也可以像人一樣，可能會有一些設計的行為產生，使電腦能更進一步來模擬人們初步的設計認知行為。 在第二階段的目的則是，在探討設計師如何對多物件產品產生構想的過程，將此現象用神經網路來模擬，試著電腦架構出設計師對於多物件產品產生構想的過程，同時可以解釋設計師的設計行為。最後，試著將隱藏層(hidden layer)及輸出層(output layer)的結點加以分組，探討將結點分組後，對於神經網路訓練速度的影響，進而提出增進神經網路訓練速度的結構。

Design, which was generated by people with complex mental behavior, is a well-thought and deliberate activity, including more rational and logical thought. In the design process, the key of design thinking is the knowledge within and/or between shape recognition and transformation. Recently, there are many researches, which study designers’ activities during sketching stage of design process, by cognitive experiment. On the other hand, in the area of visual cognitive behavior simulated by computer, symbolic systems can’t simulate how people recognize the shapes and associate from one shape to another. However, neural networks can manipulate this kind of simulations. Moreover, neural network also can recognize incomplete shapes and retrieve some attributions of shapes from long-term memory (LTM). In the first part of this thesis, the objective is to study how designers generate one-element product in terms of shape associations and evaluations. Moreover, simulating this activity, which occurred during idea generation stage, by neural networks evaluates that can computers think as well as designers. Therefore, computer systems not only simulate the visual behavior of designers, but also generate new ideas which can be sued in the design. The objective of second part is to investigate how designers generate ideas of multi-component product, and to simulate this phenomenon by neural networks. At the same time, this part tries to inspect the design cognitive behavior of the idea-generating stage, and to explore the designers’ cognitive phenomenon.