ABSTRACT Risk‐based seismic assessments using demand curves that provide mean annual exceedance rates for various system demand levels are increasingly being adopted. Such assessments combine the seismic hazard curve with fragility functions for each demand level, serving as a basis to quantify the average annual losses from system damage. Constructing fragility functions using maximum likelihood estimation with independent single damage states from multiple‐stripe analyses can produce unrealistic upward trends in demand annual exceedance curves with increasing demand levels, resulting in unreliable rate estimates. Alternatively, computing exceedance probabilities directly from fitted lognormal demand distributions (using first and second moments at each intensity level) can provide more stable annual exceedance rate estimates. This paper assesses the sensitivity of the demand annual rate of exceedance curves to the choice of analytical fragility fitting approach by comparing both methods. The convergence behaviour of both methods is assessed as a function of the hazard intensity levels/stripes and the number of analyses performed for each intensity level. The results reveal the inadequacy of the maximum likelihood approach for developing stable demand annual exceedance rate estimates, particularly at higher return periods, with slow convergence with the increase in both intensity levels/stripes and analyses per level. Conversely, computing exceedance probabilities directly from fitted lognormal demand distributions is found to be significantly more robust and efficient. However, when analysts are limited to using maximum likelihood approaches, prioritising increased hazard intensity levels/stripes over the number of analyses per intensity level would provide more accurate results given limited computational resources.
Khalil et al. (Wed,) studied this question.