摘要: |
This report presents a novel framework for promptly assessing the probability of barge-bridge collision damage of piers based on probabilistic-based classification through machine learning. The main idea of the presented framework is to divide the potential damage region of a pier into multiple discrete sub-regions, and define a classifier for each sub-region based on a probabilistic logistic regression. Several different classification models are considered for each classifier due to the uncertainties associated with the logistic regression model, and trained through Bayesian inference by using simulation data before damage events occur. Bayesian model selection is adopted to select the best classification model and its corresponding effective features extracted from measurements for enhancing the prediction accuracy. The best trained classification models then are used to expeditiously predict the probabilities of damage occurring in all sub-regions. The presented framework can be implemented in a recursive manner by dividing a structure into several hierarchical levels of different divisions of sub-regions, and can be extended to identify other types of structural damage. The effectiveness and applicability of the presented framework are demonstrated through the numerical simulation of identifying barge-bridge collision damage locations of one pier of a prototype bridge. Finally, limitations and future research directions are also discussed. |