Bayesian Damage Prediction on Berm Breakwaters in the Arctic
Abstract
A model for the computation of failure probabilities on partly reshaping mass-armored berm breakwaters in Arctic is presented. The model consists a reliable tool for the design of port structures in the rapidly changing Arctic environment and takes into account the simultaneous effects of wave and ice forces.
The presented probabilistic approach is based on Bayesian inference. Hydrodynamic and ice historical data from Prudhoe Bay (Alaska) are collected and analyzed to supply Bayesian network with a large pool of information for the analysis. The model performs afterwards real-time predictions based on historical data and user s prior knowledge and assigns relevancies to load and resistance parameters. The predictive skill of Bayesian network is validated with log-likelihood tests, confusion matrices, and score skills. Finally, the main outputs (overtopping discharge, wave height, ice thickness, recession) are applied for a Level III (fully probabilistic) reliability assessment of the structure.
The study proves that a well-treated Bayesian Network can be a powerful tool in the design process and reliability analysis of coastal structures in highly unpredictable environments, such as Arctic. Compared to first order reliability models that simplify the joint probability density functions, the Bayesian model marks a step forward in the forecasting of structural response to loads. A key finding is that the developed model can predict damage intensities on a berm breakwater, given offshore wave and ice conditions.
It is very important that the model can identify dependencies between wave and ice loads and relate them with breakwater s characteristics and response, as numerical models have not been sufficiently developed yet in this field. The evaluation of network s performance showed sufficient skills in predictions hydrodynamic induced forces but a poor forecasting ability in ice ride-up loads. This could be improved with the incorporation of additional field data, available in the literature and further development of the related analytical solutions in the future.
The probabilistic Bayesian model provides a novel tool for failure prediction of port defense structures that lie on the Arctic region, such as the berm breakwater. It offers a time efficient tool to the existing body of research on probabilistic design of coastal structures and confronts the deficient predictive skills of numerical models in structural analysis.