Resistance Factors for Driven Piling Developed from Load-Test Databases

Abstract

A methodology is developed to determine resistance factors for driven piling. Several pile load test databases are used to identify simple statistical parameters necessary to describe the uncertainties associated with predicting pile capacities. These statistics are used to determine the uncertainty with which axial pile capacities are predicted, and use these estimates of uncertainty to determine resistance factors, ~, for and LRFD approach. Introduction There have been several approaches used to develop and apply LRFD to pile foundations (Goble, 1999). Such approaches have included the following: 1) back-calculating resistance factors based on calibration with existing FS methods, 2) determination of resistance factors based on site-specific load tests and soil-information (Yoon and O'Neill, 1997), and 3) determination of resistance factors based on the predicted and measured capacity for a collection of load tests on driven piling. This study employs the third approach to develop resistance factors. Statistics are employed herein to determine the uncertainty with which axial pile loads are predicted, to use these estimates of uncertainty to determine the reliability of pile foundations, and to justify the resistance factor, if, to achieve a desired reliability. Load test databases and statistics are used as tools to quantify the uncertainty with which axial capacity can be predicted. The ratio of predicted axial capacity to measured axial capacity (Rv/R~) is used as the measure for a method's ability to predict capacity. A value of unity for Rp/R~ corresponds to perfect agreement between predicted and measured. A ratio greater than unity occurs when the method over-predicts capacity and a value less than unity occurs when the predicted capacity is less than measured. When several load tests are available, the ratio (R~/Rm) can be determined for each load test, and statistics can be used to assess the precision and bias for a predictive method. This paper is in two parts. The first part presents a discussion of simple statistics used to quantify the bias and precision associated with a predictive method. The lognormal distribution is discussed and used to fit data associated with R~/Rm. The second part associates load and resistance factors, ~, and ~ with a desired degree of reliability. The factor, ~, is determined for cases in which the load is known, and for cases in which there is uncertainty in the load. The factors are quantified using the standard deviation of the predictive method as the measure of uncertainty in capacity.