薄殼零件通常因為其複雜的幾何結構而採用模具製造或鑄造技術。這些零件的厚度與寬度（或長度）比通常較小，並且具有突起特徵以增強其機械性能。然而，由於薄殼結構的複雜性，使用傳統有限元素法進行模擬會面臨挑戰。儘管四面體網格元素能夠處理複雜的幾何形狀，但與六面體網格元素相比，其計算成本較高。即便如此，目前的六面體網格生成演算法仍然限於較簡單的基本形狀，且手動體積分解方法—雖然在某些情況下有效—卻可能引入錯誤。本研究旨在解決自動化薄殼結構六面體網格生成的挑戰，提出一種能夠有效處理複雜幾何結構的演算法。此過程首先是對模型進行分類，並識別薄殼零件每個面類型。薄殼模型根據幾何形狀和拓撲結構可分為規則型和不規則型。完成分類後，演算法會生成簡化幾何形狀並有助於體積分解的（cap face）。本研究還引入了突起特徵圖（Protrusion Feature Graph, PFG），以協助有效地生成這些蓋面。接著，通過利用蓋面進行模型的分割與合併，生成子體積。本研究提出了兩種六面體網格生成方案：一種是使用 CUBIT，另一種是使用 Moldex3D。CUBIT 允許完全自動化生成六面體網格，無需進行額外的幾何編輯；而 Moldex3D 提供了類似的過程，但目前僅限於單一來源和單一目標方案。為了測試演算法的效能，本研究提出了35個面類型識別的測試案例以及8個體積分解與網格生成的測試案例，涵蓋各種幾何形狀和拓撲結構。;Thin-shell parts are often produced using molding or casting techniques due to their complex geometry. These parts typically feature a small thickness-to-width (or length) ratio and include protrusions that enhance their mechanical properties. However, simulating such thin-shell structures using traditional finite element methods is challenging due to their complexity. While tetrahedral mesh elements can accommodate complex geometries, they come with a high computational cost compared to hexahedral meshes. Despite this, current hex mesh generation algorithms are limited to simpler, primitive shapes, and manual volume decomposition—while effective in some cases—can introduce errors. This research addresses the challenge of automating hexahedral mesh generation for thin-shell structures by proposing an algorithmic approach that effectively handles complex geometries. The process begins with classifying the model and identifying face types for each face of the thin-shell part. Thin-shell models are classified into regular and irregular categories based on their geometry and topology. Once classification is complete, the algorithm generates "cap faces" that simplify the geometry and aid in volume decomposition. A protrusion feature graph (PFG) is also introduced to help create these cap faces efficiently. The model is then split and merged using the cap faces to generate subvolumes. Two schemes for hex mesh generation are proposed: one using CUBIT and the other using Moldex3D. CUBIT allows for fully automated hex mesh generation without requiring additional geometry edits, while Moldex3D offers a similar process, though it is currently limited to single-source, single-target schemes. To test the algorithm′s performance, 35 test cases for face type recognition and 8 test cases for volume decomposition and mesh generation, covering a range of geometries and topologies, are presented.
