隨著技術的發展與模型元件的複雜度日益提升，模流分析(Mold flow analysis)已普遍用於射出成形產業，利用CAD模型轉換成實體網格，讓模流分析的求解器進行分析。以不同搭建方式的實體網格，會影響到求解器分析的最終結果。其中，四面體網格雖然適用於各種形狀的模型，並能自動搭建而成，但生成的網格數量與展旋比等數據相較於六面體網格等結構化網格而言，數量有倍數上的差異與不均勻的展弦比等不良因素，致使分析結果不佳。因此結構化網格更適用於模流分析，但結構化網格需要較為規則化的區塊方能建立，使得搭建方式顯得複雜且繁瑣。本研究主要目的為開發CAD模型上的特徵辨識演算法，用於後續的特徵分解以輔助結構化網格的搭建。本研究於特徵辨識中主要著重於混接面(Blend face)、虛擬環(Virtual loop)與孔洞特徵發展自動化特徵辨識技術，並改善過去演算法中無法辨識、錯誤分類與過度辨識等問題。接著，藉由測試更多種類的模型，將辨識技術繼續向下發展，使演算法能精確判斷更多元案例的特徵，最後提供完整的辨識資料給與後續相關的程序使用。;With the development of technology and the increasing complexity of model components, mold flow analysis (Mold flow analysis) has been widely used in the injection molding industry. The CAD model is converted into a solid mesh for analysis by the mold flow analysis solver. Solid meshes constructed in different ways will affect the final result of the solver analysis. Among them, although the tetrahedral mesh is suitable for models of various shapes and can be built automatically, the data such as the number of meshes and aspect ratios generated are more expensive than those of structured meshes such as hexahedral meshes. There are unfavorable factors such as difference in multiples and uneven aspect ratio, resulting in poor analysis results. Therefore, structured meshes are more suitable for mold flow analysis, but they require relatively regular blocks to be established, which makes the construction method complex and cumbersome. The main purpose of this research is to develop a feature recognition algorithm on CAD models for subsequent feature decomposition to assist in the construction of structured meshes. In feature recognition, this research mainly focuses on the development of automatic feature recogntion technology for blend faces, virtual loops and hole features.With the purposes of improving the problems of non-recognizd, misclassified and over-recognized in the past algorithms. Then, by testing more types of models, the identification technology will continue to develop downward, so that the algorithm can accurately determine the features of more multi-cases, and finally provide complete recognition data for subsequent related programs.
