標題: | 類神經網路橋梁結構延伸樁桿件之設計與評估模式之建立 Development of Artificial Neural Network Based Model for Bridge Extended Pile-Shafts Design |
作者: | 李綸桓 Lee, Lun-Huan 洪士林 Hung, Shih-Lin 土木工程學系 |
關鍵字: | 橋梁延伸樁桿件;性能目標;性能目標;Extended pile-shafts;Performance-based objective;Artificial neural network(ANN). |
公開日期: | 2011 |
摘要: | 橋梁延伸樁桿件為一種在美國加洲廣泛被使用的橋梁結構系統,它直接將橋柱延伸到地表下,其地震下的破壞方式是會在地表下的樁桿件形成一個塑性鉸。於2006年6月,Song等人發表了一篇有關橋梁延伸樁桿件初步耐震設計的文章,此篇文章提出了一套以結構耐震性能為設計目標的橋梁性能設計方法,正因此方法詳盡的考慮各個設計細節,導致其設計流程複雜且繁瑣。人工智慧(AI)中的類神經網路是AI領域中重要且具學習能力的理論模式之一,並廣泛應用於不同領域中。
因此,本研究希望利用類神經網路來建構一套橋梁延伸樁桿件輔助設計與評估系統,以模擬人類專家完成整個設計流程。本研究共建立了三個不同功能的類神經網路,其中一個用來輔助設計,另兩個則用來檢驗與評估設計後的結果。首先,本研究利用Matlab依不同設計參數來產生案例資料庫,並將其分為訓練與測試案例,再將訓練案例由多至少分成三至五組,分別使用不同訓練組別來訓練神經網路。由測試案例得到其相關測試結果,並藉以驗證本系統的可行性與精確度。驗證結果顯示網路訓練與測試之判定係數(R2)均達0.9以上。最後,藉由執行一設計案例,證實類神經網路所建構之輔助設計與評估系統是可行且正確的。 A common type of bridge foundation uses the so-called“extended pile-shafts”, where the circular column is continued below the ground level. Under the design level earthquake, however, extended pile-shafts for bridges may be expected to experience some level of damage below the ground level. A design procedure that incorporates soil properties into the process was developed by Song et al on 2006. The seismic design of extended pile-shafts requires a careful consideration of all the details, so the procedure was complicated and experience oriented. Artificial neural network (ANN), one of well-developed models in Artificial Intelligence (AI), is a learning-capability computing model and being widely used in different areas. The aim of this work is to develop an ANN-based model for bridge extended pile-shafts design, to simulate the human experts to complete the entire design process. Three ANN sub-models were established. One for aided design and the other two are intended to test and assess the results. First, a case base about one million cases was created according to different design parameters using Matlab, and the cases were divided into training and test cases. The training cases were then randomly divided into three to five groups to test the learning performance of these sub-ANN models. The training results revealed the feasibility of these models. Meanwhile, validation results revealed that the coefficients of determination(R2) were more than 0.9 for these models in training and testing. Finally, a complete design case was employed to test the feasibility and performance of the developed ANN-based bridge extended pile-shafts design model. The results confirmed that the system is feasible and the results are correct and engineering acceptable. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#GT079816515 http://hdl.handle.net/11536/47272 |
顯示於類別: | 畢業論文 |