Axially Loaded Drilled Shafts: LRFD Resistance Factor Calibration and Predictive Techniques for Settlement
Loading...
Authors
Stanton, Kevin
Issue Date
2016
Type
Dissertation
Language
Keywords
Artificial Neural Network , Deep Foundations , Drilled Shafts , LRFD , Machine Learning , Reliability
Alternative Title
Abstract
In areas where local geology consists of unique or uncommon subsurface materials, uncertainty in deep foundation design may lead to excessive construction costs and inadequate reliability. While the calibration of area-specific LRFD resistance factors for the strength limit state offers a potential solution to this issue, current calibration methods require that traditional or local design procedures be capable of predicting nominal capacity with a reasonable and consistent level of accuracy. However, if local geomaterials are characterized by unusually high or low strength, this condition may not hold true. Evaluation of the service limit state is also complicated in regions where such materials exist. Thus, the goal of this study is to investigate more robust LRFD resistance factor calibration techniques and to evaluate new methods for analyzing the response of drilled shafts subjected to axial loads. Three LRFD calibration methodologies are considered: Level 1, Level 2, and Level 3. Level 1 represents the most basic and traditional approach. Level 2 introduces a data ranking system as well as a framework to improve design procedures involving problematic geomaterials. Level 3 incorporates the same features as Level 2 but also addresses uncertainty arising from the interpretation of geomaterial properties from subsurface investigation data. Each level of calibration was tested using the results of 41 full scale load tests conducted on drilled shafts in the Las Vegas Valley. The calibrations are conducted with the Monte Carlo simulation method and were validated with the First Order Reliability Method for the target reliability index of 3. Nearly all of the site investigations indicate the presence of calcium carbonate cemented sandy soils colloquially referred to as caliche. While caliche is known to have the potential to contribute unusually high frictional resistance to deep foundation systems, there are no clearly established guidelines for modeling it in design. Hence, four potential approaches for treating cemented materials are investigated. These include the adaptation of codified models for dense sand (current practice), cohesive IGM, and rock from AASHTO (2014) and AASHTO (2012) as well as a new technique developed specifically for this study. Overall, it is found that the proposed technique provides the highest accuracy and yields the most conservative resistance factors at the target level of reliability. Regarding the service limit state, three methods are evaluated by comparing outputs to measured load test data. The first two are t-z style analyses while the third employs an artificial neural network which relies on foundation geometry, material index properties, and standard penetration data as inputs. On average, the artificial neural network produces the most accurate predictions of load-settlement behavior.
Description
Citation
Publisher
License
In Copyright(All Rights Reserved)