Offshore pile system model biases and reliability

ABSTRACT

Recent hurricanes between 2004 and 2008 in the Gulf of Mexico provide valuable new information about design model biases and uncertainties because multiple offshore platforms were loaded beyond the predicted capacity of their pile systems and because there was a failure of a pile system. A Bayesian calibration of model bias factors based on predicted versus observed performance of pile systems in hurricanes indicates that the conventional American Petroleum Institute design method for pile capacity is slightly conservative by about 10% for base shear (i.e. lateral) failures of pile systems in clay, unbiased for overturning (i.e. axial) failures of pile systems in clay, and conservative by more than 50% for overturning failures of piles systems in sand. The epistemic uncertainty in the updated bias factors is represented by coefficient of variation values of about 0.25 for base shear and overturning failures of pile systems in clay and 0.35 for overturning failures of pile systems in sand. A reliability assessment with the calibrated model bias factors shows that the current design practice produces lower reliability for a pile system with three piles versus one with eight piles and lower reliability for a pile system failing in overturning versus one failing in base shear. Therefore, the current design practice could potentially be improved by taking into account the mode of failure and the redundancy in the pile system to provide for a more uniform level of reliability.