近年來穿戴型電子裝置逐漸普及於日常生活中，其中一種新型電子皮膚技術在該領域中陸續受到市場關注，如智慧型敷料可以透過監測傷口狀況達到釋放藥物以防止傷口感染或加速癒合的敷料。然而使用三維生物列印技術製作出符合傷口外型之客製化敷料，目前仍有許多缺點需要克服，以改善敷料對於慢性傷口的吻合度與包覆軟性陶瓷基板的平整度等。 本研究的主要目的是提升生醫材料所印製的產品表面品質。經實驗發現，產品形貌受到噴頭擠出壓力、噴頭與平台間隙和噴頭進給速度所影響。為了找出最佳製程參數，本研究提出實驗結合多層感知器與反應曲面法的方法，以實際應用案例驗證最佳化結果。第一階段使用單層單向單線的製造方式印製產品並量測形貌數值，建立品質預測模型。第二階段執行形貌列印最佳化的實際應用案例。最後，本研究以包含架構為3-9-6-3的多層感知器模型和雙因子交互模型，證明使用中央複合設計或全因子設計作為訓練資料集，均可取得預測誤差值為6.60 %以下的預測模型；以及證明使用實驗設計方法確實比隨機取用訓練資料集，更有效使用實驗數據且能減少生醫材料損耗。在模擬智慧光療感測型敷料的印製情境下，透過品質預測模型產生的最佳製程參數，控制產品形貌高度使印製表面平整化，驗證本研究提出的方法可以優化生醫材料所印製的產品表面品質。 ;In recent years, wearable healthcare electronic devices are becoming popular in daily life, and a new electronic skin technology has received widespread attention in this field. As intelligent wound dressing can detect the injured condition and release drugs timely to prevent the wound from infection and accelerate wound healing. However, using 3D bioprinting technology to produce customized dressings that fit the wound profile, there still exists amounts of weak points. For example, the anastomosis of the dressing for chronic wounds and the flatness of the encapsulated soft ceramic substrate. This study proposes a method to improve the surface quality of product printed with bio-material. The experiments show that the product shape is affected by three process parameters: the extrusion pressure of the material, the feed rate of the nozzle, and the gap between the nozzle and the platform. This study build models using Multilayer Perceptron and Response Surface methodology with the experimental data to optimize process parameters. Firstly, single-layer single-line samples are manufactured to collect the data from the product shape and the process parameters to train the models. Next, both of the Multilayer Perceptron model with 3-9-6-3 framework and the Two-Factor Interactions model, trained using the datasets of either Central Composite Design or Full Factorial Design, perform with the predict error of 6.60 % or less. This approach can reduce the consumption of bio-material compared to randomized datasets. Finally, in practical applications, the proposed method is verified to optimize the product surface quality printed with bio-material.
