- 學位論文
應用類神經網路於模擬皮克林乳液製備方法之研究
Study of Application of Artificial Neural Networks modeling for Pickering Emulsions
摘要
皮克林（Pickering）乳液為水相與油相添加固體粒子所製備而成的乳化系統，而利用固體粒子與界面活性劑協同的方法可以使乳液更加的穩定。此實驗方法中所影響乳液油滴粒徑的重要變數有Tween 80和膨潤土的含量、水溶液與油的比例及均質機的轉速，分別為兩個配方因素及兩個製程因素。為了能在此次研究中得到實驗變數的最佳化模型，我們使用了類神經網路(ANN)系統搭配回應曲面法(RSM)的結果來模擬且預測皮克林乳油滴液粒徑大小。 本次類神經網路模組採用了3層回傳遞神經網路(BPNN)，分別為輸入層、隱藏層、輸出層，各層神經元數分別為4、18、1。而輸入變數值為Tween 80含量(3.8-19.8%)、膨潤土含量(0.58-2.58%)、水油比(3.44-5.44)及均質機轉速(8000-12000rpm)，輸出變數值則為皮克林乳液油滴粒徑大小。經由神經網路的模擬訓練後，對照模組輸出值與實驗數據可獲得高度的模式判定係數，顯示了此次之類神經網路模型可以成功的預測皮克林乳液油滴粒徑大小。
並列摘要
Pickering emulsions are made up of solid particles, water phase and oil phase mixture in an emulsified system. One way to increase the stability of emulsions is combining the solid particles with surfactant. The important variables in this experiment are %Tween 80 content, %bentonite content, an aqueous dispersion-to-oil ratio (W/O), and the rotational speed of homogenizer. There are two factors for formulations variables and two factors for processing variables. In order to obtain the optimization model in this study, we applied artificial neural networks (ANN) modeling to fit the experimental data from response surface methodology (RSM) and predict the particle sizes of Pickering emulsions. We used a 3-layers back-propagation neural networks (BPNN) with 4, 18 and 1 neurons in input, hidden and output layers. The input variables are %Tween 80 (3.8-19.8%), %bentonite (0.58-2.58%),W/O (3.44-5.44), and the rotational speed of homogenizer(8000-12000rpm), while the oil droplet size of Pickering emulsions is the output variables. By training ANN system, a comparison of the model outputs with experimental data gave high goodness of fit, indicating that the model is suitable to predict the oil droplet size of Pickering emulsions.