Probabilistic Assessment of Structural Condition Incorporating Uncertainty of Measured Data through Multi-model Identification

This paper proposes a novel procedure of probabilistic assessment of structural condition by incorporating uncertainty of measured data. Multiple FE models are updated through the successive optimizations which use probability distribution functions representing uncertainty of the measured data. In addition, twomachine-learningalgorithms are implemented for probabilistic description and classification of the updated multiple FE models. Principal Component Analysis (PCA) method transforms the updated FE models onto the principal subspace. Accordingly, distribution feature is better represented. Next, Gaussian mixture model is used to identify probabilistic features of the updated FE models. Finally, the probabilistic information about structural condition is obtained by using the updated FE models and the identified probability models. The proposed procedure is demonstrated by applying to assessment of Yeondae Bridge, a 4-span continuous steel-box girder bridge in South Korea. Through the proposed procedure, distribution of rating factors, which represent the probabilistic information about structural condition, is estimated. Especially, clustering-based model selection procedure enables more reliable assessment of structural condition even when higher uncertainty of measured data is associated.