細粒料含量影響飽和砂土應力-應變行為分析與模擬

Analysis and Simulation of Stress-Strain Behavior of Saturated Sand with Fines Content

詹鎮宇(Chen-Yu Chan)

指導教授 ： 陳怡睿

長榮大學/管理學院/土地管理與開發學系(所)/碩士(2012年)

https://doi.org/10.6833/CJCU.2012.00024

摘要

瞭解土壤的應力-應變行為，對於解決大地工程的問題，有很大的幫助。台灣西部平原在台中以南有厚層含粉土質之砂性土壤，對於土層的結構會造成很大的影響，因此有進一步探討之必要。過往一些經驗公式往往是經由觀察有限之試驗現象，並使用統計迴歸分析所獲得。然而，類神經網路不需要瞭解系統的數學模型為何，而直接以神經網路取代系統的模型，一樣可以得到輸入與輸出之間的關係。因此，若藉由類神經網路建立土壤應力-應變之行為，將可提供更快速的解決方法。
本研究針對雲林縣麥寮鄉海岸新興工業區之砂土，運用三軸剪力試驗儀器，進行等向壓密不排水單調及反覆剪力試驗，用以探討飽和砂土應力-應變行為。研究中，試體之相對密度控制在60%，細粒料含量以5%為間隔，由5%逐漸增加至30%，初始平均有效正向應力分別控制為50、100及250kpa；此外，在施作壓密不排水反覆剪力試驗時，反覆應力比則分別控制為0.1、0.15、0.2、0.25及0.3。本研究並將試驗結果，作為類神經網路進行訓練與測試之輸入及輸出值。
三軸不排水單調剪力試驗結果顯示，當平均有效正向應力固定時，相角轉換之現象會發生在細粒料含量較低之試體；但隨著細粒料含量增加，相角轉換現象則不復存在。且在相同之初始有效正向應力下，細粒料含量愈高，尖峰軸差應力比值(qpeak/p’in)愈小；而在相同細粒料含量的情況下，初始平均有效正向應力愈高，qpeak/p’in值有下降的趨勢。三軸不排水反覆剪力試驗結果顯示，不論細粒料含量多寡，試體達3%軸向單振幅應變之作用次數會隨著反覆剪應力比之增加而減少；且在相同反覆剪應力比之下，反覆作用次數隨細粒料含量增加而減少。
此外，結果顯示類神經網路模擬能將些許不穩定之試驗數據作良好的歸納，且在平均有效正向應力與軸差應力關係的模擬結果可知，其相角轉換的位置與破壞線趨勢與試驗數據趨於一致。研究中亦發現基因演算法自動演化類神經網路的模擬速度明顯比試誤式類神經網路的模擬速度快，且兩者皆可適當的模擬三軸剪力試驗之結果。

關鍵字

細粒料 ； 三軸剪力試驗 ； 類神經網路 ； 基因演算法

並列摘要

Understanding stress-strain behavior of soil is a great help to solve geotechnical problems. In Taiwan’s western plains to the south of Taichung, sandy soil with fines content (FC) dominant the structures of soil layers. Therefore, it is necessary to further explore the stress-strain behavior of soil with FC. In the past, the empirical stress-strain relations of soil were usually obtained by statistical regression analysis based on experimental observations. In general, computer simulation of stress-strain behavior of soil has some advantages. It saves money and time and can extract more information. By using artificial neural networks (ANN), it is not necessary to assume the type of stress-strain relations of soils. Furthermore, ANN can solve complex nonlinear problems with high accuracy. Therefore, ANN was used to simulate stress-strain relation of saturated sand with fines content in this study. In addition, genetic algorithm is employed to evolve the parameters used in ANN to obtain global optimal solution.
This study collected the soil sample from industrial areas in Mailiao Port. During experimental works, the initial relative density of the soil sample was set around 60%. FC increased from 5% to 30% with increment of 5%. Triaxial test apparatus was used to perform isotropic consolidated undrained monotonic and cyclic tests with the initial mean effective normal stresses (p′) of 50, 100 and 250kpa, respectively. In the loading path of cyclic undrained shear tests, the cyclic stress ratios (CSR) were set to be 0.1, 0.15, 0.2, 0.25 and 0.3, respectively. The experimental results then were used as ANN training and testing data.
In undrained monotonic tests, phase transformation occurs under constant effective normal stress, while fines content is low. However, the phenomenon of phase transformation disappears with increasing fines content. With the same initial mean normal effective stress, the higher of fines content, the smaller of peak deviator stress ratio. With the same fines content, peak deviator stress ratio tends to decrease with increasing mean normal effective stress. No matter the amount of fines content, the number of cycles to cause 3% axial strain decreases with increasing CSR or initial effective normal stress. Furthermore, under the same CSR, the number of cycles decreases with increasing the fines content.
The locations of phase transformation line are almost on the same inclination for predicted data and experimental data. The results show that both genetic adaptive neural networks (GANN) and trial-and-error type ANN can simulate the test results appropriately, while the running time for GANN is half of that for the latter.

並列關鍵字

fines content ； triaxial shear test ； ANN ； GA

參考文獻

陳嘉謙(2008)。飽和砂土等向及非等向壓密不排水三軸試驗力學特性之研究。長榮大學土地管理與開發學系碩士論文。

鄒承府(2006)。振動台大型剪力盒麥寮砂試體準備之評估。國立台灣大學土木工程學研究所碩士論文。

俞旗文(2002)。案例類比法應用於台灣區隧道支撐工設計建議。地工技術，91，21-30。

游家豪(2007)。低塑性細料對粉質砂土動態性質之影響。國立成功大學土木工程研究所碩士論文。

林智偉(2006)。無塑性細料對砂質土壤液化阻抗之研究。國立成功大學土木工程研究所碩士論文。