A STUDY ON THE ACCURACY OF SOIL LIQUEFACTION POTENTIAL ASSESSMENT METHODS
| Autor: | Chin-Hua Huang, 黃金華 |
|---|---|
| Rok vydání: | 2016 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 104 Since Seed & Idriss (1971) proposed the SPT-N liquefaction potential evaluation method, most researches only focused on the study of influence factors thereafter. The accuracy of liquefaction potential assessment has fewer studies for a while. Although those traditional SPT-N liquefaction potential evaluation methods are very handy, the prediction accuracy obtained is no more than 79 %. It means 21 mistakes exist in one hundred estimation events. Ten traditional methods are therefore used in this article to examine more influence factors and the estimation accuracy of neural network method. Ninety-six domestic liquefaction data and 305 oversea data were employed in order to include more extensive resources in this research. It is found that Fine Particle Index and Non-liquefiable Overburden Soil Thickness are two important influence factors. When a back-propagation neural network model for liquefaction potential evaluation trained with more than 300 data, it will obtain better accuracy than traditional methods. But, the traditional method still presents better estimation in an area with less data. In the probability analysis, the coefficient of variation for liquefaction analysis is strongly affected by the local soil characteristics. Monte Carlo simulation together with Seed et al. method (1985) was used to perform the liquefaction potential probability analysis. It is found that the coefficients of variation are adopted from oversea resource and from domestic one will cause the difference of analysis results up to 15.78 %. It is then recommended to build domestic coefficient of variation parameters before processing a liquefaction probability analysis. Finally, ten traditional liquefaction analysis methods were examined. Eight influence factors, such as the Maximum acceleration (amax), SPT-N value (N), SPT-N depth (ds), Ground water table depth (dW), Fine particle content (Fc), Mean grain size value (d50), Non-liquefiable Overburden Soil Thickness (NLOST) and Fine Particle Index (FPI), were recommended as the most important influence factors. They would help in promoting the best liquefaction potential prediction accuracy up to 94 %. The accuracy of rest other methods are also improved to a value better than 90 %. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|