雙面非球面鏡片射出成形之多目標最佳化研究

本研究主要探討射出成形參數對於雙面非球面鏡片形狀精度的影響，以類神經網路建構出品質預測器再搭配基因演算法找出射出成形參數的最佳條件組合。實驗過程中，先以田口實驗方法設計實驗參數，以模具溫度、塑料溫度、保壓壓力為主要製程參數，並以這些參數組合作為類神經網路的訓練組。然而由實驗結果所得出兩個特徵面的最佳參數並非同一個組合，故再導入多目標田口方法，找出一組對於兩個特徵I及II面都是較佳的參數組合，再施以類神經網路建構出品質預測器。將測試結果與預測結果作回歸分析，可以得出品質預測器的效果，最後再搭配基因演算法演算出一組最佳的成形參數，以提升雙面非球面透鏡的形狀精度。實驗結果得知，多目標田口方法求得的最佳化參數組合為模具溫度110℃，塑料溫度250℃及保壓壓力65MPa，其實際成形得到雙面非球面透鏡之特徵I面形狀精度為1.874μm，特徵II面形狀精度為2.181μm；品質預測器搭配基因演算法求得之最佳化得出的結果為模具溫度100.6℃，塑料溫度251.3℃及保壓壓力57.2MPa，其實際得到雙面非球面透鏡之特徵I面形狀精度為1.221μm，特徵II面形狀精度為0.968μm，相較於多目標田口方法，品質預測器搭配基因演算的實驗方法使雙面非球面透鏡的兩個特徵面獲得更佳的形狀精度。

關鍵字

非球面鏡片；射出成形；類神經網路；基因演算法

並列摘要

This study focuses on obtaining the injection molding parameter settings for producing high precision double-sided aspherical lenses. A neural network approach is used to construct a quality predictor, and a genetic algorithm is followed to search for the optimal parameters settings of the predictor. Mold temperature, plastic temperature and holding pressure are selected as the major process parameters in experiments. At first, Taguchi method is carried out to find the optimal parameter combination for each side of the lens. However, the optimμm parameter combinations obtained for two feature surfaces are not the same. Therefore, multi-objective Taguchi method is used to identify a preferred parameter combination for the two surfaces. The obtained parameter combination is then used as the training set for a neural network quality predictor. Regression analysis is conducted by using predicted results and test results to realize the performance of the quality predictor. Finally, a genetic algorithm is applied to obtain a set of optimal combination of parameters which can improve the accuracy of double-sided aspherical lenses. Results of optimization parameters obtained by multi-objective Taguchi method are mold temperature 110 ℃, plastic temperature 250 ℃, and holding pressure 65MPa. The surface shape accuracy of double-sided aspherical lens for feature surface I is 1.874μm, and 2.181μm for feature surface II. The optimization parameters obtained by quality predictor and genetic algorithms are mold temperature 100.6 ℃, plastic temperature 251.3 ℃, and holding pressure 57.2MPa. The surface shape accuracy of double-sided aspherical lens for feature surface I is 1.221μm, and 0.968μm for feature surface II. Comparing the obtained results, parameter settings obtained by using the optimization method combining quality predictor with genetic algorithm can improve the surface accuracy of the double-sided aspherical lens.