Effectiveness Assessment of TMDs in Bridges under Strong Winds Incorporating Machine-Learning Techniques

Abstract

Tuned mass dampers (TMDs) are widely used to control excessive wind-induced vibration in the box girders of long-span bridges. Although the optimal design of TMDs has been investigated abundantly in the last few years, the effectiveness of TMDs in use has not been thoroughly studied. In this paper, a method combining a machine learning (ML)–based approach is developed to evaluate the TMD effectiveness. The theoretical formulation and the flowchart of the method are firstly presented, which utilizes characteristics of TMD vibration amplitude, phase shift between TMDs and the bridge, and the mode resonant frequency component. Seven commonly used ML techniques, i.e., artificial neural network (ANN), decision tree (DT), k-nearest neighbors (KNN), random forest (RF), adaptive boosting (AdaBoost), gradient boosting regression tree (GBRT), and extreme gradient boosting (XGB), were adopted to generate the predictive models, and the structural health monitoring (SHM) data of the bridge were used as the training data. The wind properties and temperature were set as the input, and the TMD accelerations are set as the output. Meanwhile, the Shapley Additive Explanations (SHAP) was adopted to identify the influences of the input variables on the TMD’s performance. The result indicated that the proposed method is reliable to evaluate the effectiveness of the TMDs, and it was shown that wind velocity is the most important parameter. BecauseTMDs are often widely used to control vibration in bridges, the proposed ML-based method can be used as an effective tool to assess and/or cross-check the effectiveness of TMDs.