Hybridization of Finite-Element Method and Support Vector Machine to Determine Scour Bridge Safety Level

Abstract

Bridge failure as a result of scour may cause a serious disaster because it tends to occur suddenly and without prior warning. Although the scour depth around a bridge pier is an important index to evaluate the condition of the bridge, it is difficult to measure directly because the bridge foundation is under water. In addition, the safety level of the bridge could be arbitrary, depending on the environment and bridge. In this study, an approach that combines a finite-element simulation and support vector machine (SVM) is used to determine the scour safety level of a bridge through the variation in the natural frequency of the bridge structure. A series of finite-element simulations is implemented with different environmental scenarios to obtain bridge natural frequencies. Moreover, SVM is used to classify the data and provide a safety level for a bridge scour prewarning. At the same time, a five-fold cross-validation and two-step grid search technique are introduced to optimize the SVM parameters. Finally, the examination results are presented that show the high performance of the SVM classification.