Seismic Probabilistic Safety Assessment (SPSA) has long relied on simplifying assumptions regarding component failure dependence, most notably, the use of bounding fragilities based on the weakest component or the adoption of fully independent/fully dependent models. While computationally convenient, these assumptions can introduce systematic bias, either masking critical risk contributors or overstating conservatism. This paper addresses a pressing gap in current practice: the rigorous, yet practical, treatment of partial correlation among non-redundant and spatially distributed components within fragility families.
Building on regulatory drivers and recent methodological guidance, this work implements and extends the Reed–McCann framework to quantify joint seismic failure probabilities by explicitly decomposing epistemic and aleatory uncertainties into independent and common components. Computational workflow, supported by Latin Hypercube Sampling and numerical integration, is developed to generate system-level fragility curves that consistently capture partial dependency effects across multiple failure modes and components. The approach is benchmarked against established NUREG formulations and demonstrated through application to safety-significant systems in a full-scale nuclear plant SPSA.
The results reveal that commonly used “weakest-link” representations can be non-conservative when component capacities are comparable, with deviations in High Confidence of Low Probability of Failure (HCLPF) values that are non-negligible for risk-informed decision-making. Sensitivity studies further show that correlation structure, not just median capacity, systematically shifts fragility curves, influencing both joint and union failure probabilities in ways that cannot be captured by traditional bounding assumptions. Importantly, the findings identify regimes where partial correlation meaningfully alters risk insights, particularly for complex, non-redundant systems with shared demand or capacity characteristics.
This work is novel in that it converts an academically established, but not widely applied, methodology into a practical and scalable framework suitable for regulatory use. It effectively bridges the gap between theoretical formulation and SPSA implementation. By enabling the treatment of partial dependency, the approach goes beyond simplified correlation assumptions and provides a more rigorous and transparent representation of dependency in fragility modeling.
For the international PSA community, these results are particularly relevant given increasing regulatory scrutiny, aging infrastructure, and the shift toward risk-informed licensing. The proposed methodology offers a defensible means to reduce epistemic bias, improve risk prioritization, and enhance the technical credibility and realism of seismic PSA evaluations.