Model Uncertainty in Offshore Geotechnical Reliability

ABSTRACT Model uncertainty is a dominant source of uncertainty in offshore geotechnical reliability. Caused by limitationsin data quantity and quality, model uncertainty leads to uncertainty in the failure rate for a design. [n this paper, an approach is presented for evaluating model uncertainty and incorporating it into reliability analyses. This approach provides the ability to estimate statistics of the failure rate. The failure rate is of interest because (1) it represents an objective measure of performance, (2)it should enter into design decisions since it reflects quality, and (3) it can be observed, providing the ability to reduce model uncertainty by analyzing observed successes and failures. The value of reliability analyses will be enhanced greatly if a physical meaning (i.e., the rate of failure) can be assigned to the results. INTRODUCTION Foundation design for offshore structures represents one of the earliest applications of reliability methods in geotechnical engineering. There are substantialuncertainties associated with the offshore environment concerning storm loads, site characterization and foundation behavior. In addition, costs for offshore construction are large and the consequences of failure are significant. Because of these large uncertainties and costs, considerable attention has been devoted during the past 20 years to develop reliability-based design pproaches for offshore foundations (e.g., Bea Wu et al1). The primary product of a reliability analysis is the probability of failure for a component or system. For example, Tang and Gilbert2 evaluated the reliability three pile foundation systems constructed during the early 1980's. They obtained probabilities of collapse over a 20-year design life ranging from 0.005 to 0.01. Lacasse and Nadim3 presented failure probabilities ranging from 0.00007 for bearing capacity failure of a for gravity structure to 0.02 for failure of a platform system under seismic loading. In both papers, the authors contended that the calculated probabilities were not 'absolutd' quantities but expected probabilities, at best, hich could be used for comparing different designs. 11-wsecalculated probabilities were not based purely on actuarial data, and therefore contained a large measure f subjective judgment and interpretation. In fact, eotechnical reliability analyses are somewhat unique inhat this component of judgment and interpretation due tolimited data, which we will refer to as model ncertainty, is dominant. In recent years, several efforts have been directed at calibrating estimated failure probabilities for offshore structures4,5. Which the value of a reliability analysis is greatly enhanced if a physical meaning can be assigned to the failure probabilities, a clear understanding of model uncertainty is required to achieve this object paper, we provide a definition of model uncertainty anddistinguish it from random uncertainty, identify sources of model uncertainty, and discuss how to evaluate and represent model uncertainty in reliability analyses. DEFINITION OF MODEL UNCERTAINTY Random Events Uncertainties abound in offshore foundation design; soilproperties, loading conditions and foundation behavior are not known with certainty at different points in time and space. There is a risk that a foundation will not perform as designed because of these uncertainties.