為瞭解裂隙岩層中的溶質傳輸行為，開發三維(3D)離散裂隙網路(DFN)水流與溶質傳輸模式是項重要任務。由於裂隙與母岩水流與傳輸能力差異較大，並且兩者在三度空間交互作用涉及複雜空間幾何，計算分析困難。因此，多數水流與傳輸研究僅考慮裂隙網路。因建構3D DFN，生成裂隙非結構化網格及模擬水流與傳輸是極具挑戰的任務，過去的裂隙網路水流及傳輸分析，通常針對二維或一維問題。本研究主要目的在開發、驗證與應用三維離散裂隙網路水流與傳輸模式，模式具備建構三維DFN、有效DFN網格生成以及DFN水流與傳輸數值模擬等功能。建構3D DFN模式首先必須分析現地調查資料，建立裂隙參數統計結構。為重現調查裂隙結構，本研究開發的DFN模式考慮裂隙中心位置符合Poisson分佈，其餘裂隙統計參數先考慮均勻分佈，未來將朝增加統計結構分析精進。DFN網格生成採用Delaunay三角網格演算法，並且透過邊界回復技術修正所有節點於裂隙面共線與共點位置為唯一，網格生成可自動轉換網格資訊，作為本研究開發的水流與傳輸模式，或者轉換為既有模式如TOUGH系列或HYDROGEOCHEM模式網格設定。不同於過去以質點追蹤分析，本研究中模擬3D DFN水流與傳輸，將透過裂隙方位逆轉流程，結合有限元素法直接求解移流延散偏微分控制方程式，因模式在生成網格時已經考慮回復裂隙共線內邊界節點，可確保每個網格求解結果連續，簡化3D DFN傳輸模擬時方程式的空間轉換。開發完成的DFN水流與傳輸模式與既有TOUGH系列模式與HYDROGEOCHEM模式進行驗證，藉由單一平板裂隙模擬進行穩態水流與傳輸模式分析比較。模式應用區分為兩部份，包括DFN網格資訊應用於TOUGH系列模式，模擬CO2地質封存與核種傳輸行為。以及利用蒙地卡羅法產生多組三維裂隙網路，裂隙頻率為1至6 m-1，討論依質量守恆原理計算的裂隙岩體等效水力傳導係數，分析裂隙頻率與等效水力傳導係數關係。此外，亦由多組3D DFN，分析不同觀測策略與位置條件下，濃度穿透曲線的差異與不確定性。研究結果顯示，在開發的DFN水流與傳輸模式中，若以單一水平裂隙進行模擬可重現TOUGH2與HYDROGEOCHEM結果。DFN水流模擬推估等效水力傳導係數結果發現，相似的等效水力傳導係數對應的二維裂隙頻率(P21)變化大於三維裂隙頻率(P32)變化，裂隙岩體等效水力傳導係數約低於裂隙水力傳導係數2至3個級數(order)。根據DFN傳輸分析結果顯示，不同觀測方式濃度穿透曲線差異較劇烈時刻，常發生於濃度團前緣通過觀測位置時。經由多組傳輸模擬結果發現，不同觀測方法觀測平均最大濃度，可能達不到裂隙岩層初始設定最大濃度值。相同的觀測位置，濃度標準偏差可超過一半的初始釋放濃度，隨著觀測位置遠離釋放源，不同觀測策略觀測濃度標準偏差無明顯變化。;Development of flow and contaminant transport models for three-dimensional (3D) discrete fracture networks (DFNs) is critical to characterize flow and transport in fractured rocks. The fractures in a rock are relatively permeable as compared with the rock matrix. The difficulty in resolving complex fracture and matrix interactions in 3D domains has motivated investigators to focus mainly on fracture networks for characterizing flow and transport in fractured rocks. Because of the complex fracture geometry and connectivity, generations of fracture unstructured mesh and simulations of flow and transport in 3D DNF become challenging tasks. The objectives of this study are to develop, test, and implement numerical models for generation of 3D DFNs, generation of DFN meshes, and simulations of DFN flow and advection-dispersion transport. The developed DFN generator enables the Poisson and uniform distributions to be implemented for fracture locations and other fracture properties such as sizes, trend, and plunge. Other distributions of fracture properties can be employed in the developed DFN generation model. The DFN mesh generation model employed the Delaunay triangulation algorithm and applied a boundary recovery technique to resolve detailed fracture intersections and fracture concurrent points. This mesh generation model can automatically export mesh formats for other public domain models such as TOUGH series models and HYDROGEOCHEM model. Unlike previous investigations that focused on the particle tracking algorithm for DFN transport simulations, this study proposed a back rotation process (BRP) for fracture orientations and directly solved Eularian-based advection-dispersion equation for 3D fracture networks. The developed flow and transport model were validated with the TOUGH2 and HYDROGEOCHEM models using single porous fracture plate. The implementations of the developed models were simply divided into two different parts. The first part is implementations of DFN and DFN mesh generations for the TOUGH2 model and the associated equation of state (EOS) modules, including ECO2N and EOS7R. The second part focuses on implementations of the developed DFN flow and advection-dispersion models for issues of equivalent hydraulic conductivity in DFN upscaling and for analysis of transport uncertainty based on different monitoring strategies. The results of model validations showed that our DFN flow and transport models can reproduce identically the solutions of flow and concentration obtained from TOUGH2 and HYDROGEOCHEM models. The estimations of equivalent hydraulic conductivity based on multiple DFN realizations showed that different fracture intensities can lead to variations of equivalent hydraulic conductivity values in 2 to 3 orders of magnitude lower than the value of the fracture hydraulic conductivity. The comparisons of 2D fracture intensity (P21) and 3D fracture intensity (P32) indicate that the variations of P21 values are relatively high as compared with a specified range of P32. High variations of concentration breakthrough curves were obtained when fronts of plumes passing the observed locations. The maximal mean concentration breakthrough curves for different averaging strategies might not reach the maximal concentration value released at continuous sources. The standard deviations of concentration at different times can be higher than half of the source concentration at different monitoring locations. Additionally, the values of concentration standard deviations are similar for different monitoring points and monitoring (averaging) strategies.
