Updating piping reliability with field performance observations

• Incorporating inequality information from field observations to reduce piping-related uncertainty.
• Field observations can increase or decrease reliability ("good vs. bad performance").
• The more unexpected the observation, the higher the impact on the probability of failure.
• We show how to distinguish between reducible and irreducible uncertainty.
• The information improves safety assessments and retrofitting design.
• Method with potential for application in flood event management.

Flood defenses are crucial elements in flood risk mitigation in developed countries, especially in deltaic areas. In the Netherlands, the VNK2 project is currently analyzing the reliability of all primary flood defenses as part of a nationwide flood risk analysis. In this project, as in most other reliability analyses of flood defenses, prior probabilities of relevant parameters such as ground conditions use to be based on sparse site investigation data and/or expert judgment. What is largely neglected is the observed performance during extreme events such as excessive seepage or sand boils. Using this information and thereby reducing uncertainties contributes to identifying weak spots or to increasing reliability where positive signs of performance are observed. Ultimately, this contributes to focusing investments in flood defenses where they are needed the most. This paper proposes a method based on Bayesian Inference for updating uncertainties and focuses on the failure mechanisms uplift and piping. Attention is paid to the system reliability effects in this failure mode, too. The methodology is applied to a case study in the Netherlands, the prior probabilities stem from the VNK2 project. The results suggest that depending on the observation, the probability of failure can either increase or decrease by about a factor 10. The findings clearly contradict the common perception, at least in the Netherlands, that if a structure survives an extreme (load) event its reliability always increases. That is only true unless bad performance-related observations have been made.

Graphical AbstractFigure optionsDownload as PowerPoint slideThe reliability of levees with respect to uplift, heave and piping can be updated by Bayesian posterior analysis (or Bayesian updating) using the information provided by field observations during flood events such as seepage or sand boils.