近幾年來,由於傳統產業生產型態不斷提升與改變,加上基本勞資之調升,迫使許多產業不得不找尋新方法,以降低生產成本,而製鞋產業又為其中之典型案例。傳統製鞋流程中需先由設計師手繪草圖原稿,交由鞋體打樣人員依據草稿之外觀,輔以過去之經驗,繪製出相對應之2D半面版。接著依據版上之線段組合,使用套裝軟體轉換為皮料版塊,再由工廠以此造型試產。如此由設計至加工的流程中無論在任一階段發生錯誤,都將造成無效經濟成本之提高。 本研究中,將以逆向工程方式,先將現有之楦頭、鞋體、半面版等資料彙整,並將鞋體與楦頭3D掃描至數位環境中,先令鞋體輪廓以3D線條方式重現,再將其線條之點資料匯出,並利用類神經網路學習因而建構出與2D半面版之關聯,並將訓練好之對應關係導入未來設計過程中,加速製程之設計週期。 實作中將以兩款鞋體作為研究範例,探討類神經網路之轉換品質並調整之,並與現有其他展開技術做一比較。
Shoes pattern development has been an important process for bridging design intentions and production steps. Most designers would prefer hand sketching draft than prototyping a 3D shoe model, and thus making pattern development a time consuming bottleneck. Current shoe development process requires three steps, transferring design sketches into a 2D forme, rearranging curve segments on the forme into pieces of shoe patterns, and revising forme segments by test makings. All of the steps are performed by hands with experiences. However on the production side, all processes can be driven and carried out by computer data. It will be profitable to create an automatic procedure that transfers design intentions into usable 2D patterns, for it will save lots of human efforts. This research will apply reverse engineering to scan a 3D shoe example, and acquire a 3D shoe model. An artificial neural network is then created to learn the relationships between the 3D shoe segments and the 2D corresponding pattern segments. A well trained neural network should be able to map the 3D and 2D based on one shoe last size, and will deliver more mappings for the designs of the same shoe last. The results show promising mapping ability on two pairs of shoe examples.