|
碩士論文公告區 |
||
| 年度： | 92 | |
| 姓名： | 林大偉(Ta-Wei Lin) | |
| 論文題目(中)： | 混凝土火害後強度折減之徑向基類神經網路分析 | |
| 論文題目(英)： | Prediction of Residual Strength of Heated Concrete Based on RBF Neural Networks | |
| 摘要(中)： | ||
| 混凝土在受高溫延時作用後會造成材料性質的改變，其中以強度折減為工程上重要評估因素之一。本研究主要針對混凝土受高溫延時作用後強度以及脈波波速折減程度之行為，整理出溫度-延時-強度折減三者間之關係，並將結果導入類神經網路。並對高溫延時實驗結果導入類神經網路觀念，建立以溫度、延時、水灰比、殘餘脈波波速預測混凝土殘餘強度的網路架構。除普通混凝土外，另採用自充填混凝土來做類神經網路的試驗，驗證是否可利用類神經網路做預測。此研究所選用之類神經網路為Radial Basis Function演算法，研究主要目的為探求類神經網路預測之值，是否與實際實驗所得之混凝土殘餘抗壓強度是否吻合，期望預測良好之網路模式的權重值及偏權值可以再利用在其他火害的案例上，也可預測到更高溫度火害下混凝土殘餘強度的變化。結果發現火害溫度在240~400℃時，殘餘強度比有些許上升，500℃之後殘餘強度比才會有明顯下降。網路訓練測試數據筆數會隨著訓練筆數增加，測試筆數減少而有較好的結果，網路模式預測受火害的混凝土殘餘強度比，驗證經訓練後徑向基網路可以產生良好的預測結果。 | ||
| 摘要(英)： | ||
| The strength reduction of reinforced concrete is an important issue after fire exposure. This study is motivated by the correlation between the residual strength and the ultrasonic pulse velocity in concrete subjected to high temperature for a given exposure period of time. Four different mixtures of concrete are used to cast 409 specimens. They are tested in an electric oven to simulate fire exposure. Artificial neural network analysis is applied to the experimental results. The input to the network includes the maximum exposure temperature, the exposure time at the maximum temperature, the cement-to-water ratio, and the ratio of residual to reference pulse velocity. The network is constructed based on the radial basis function. A systematic approach for assigning training and validating sets of data is used to train the network. The input data are presented to the network in a batch-training mode. It is found that the average residual error of predicted residual strength of concrete can be reduced to 0.05 or smaller using the trained artificial neural network. The best result achieved by the train network represents a two-fold improvement over the prediction made by the regression analysis. | ||
| 檔案： | 沒有相關檔案 | |
| 相關連結： | http://ethesys.lib.cyut.edu.tw/ETD-db/ETD-search-c/view_etd?URN=etd-0823104-101729 | |