Quantification of Fatigue Damage of Structural Details in Slender Coastal Bridges Using Machine Learning Based Methods

ABSTRACTWith coupled dynamic impacts from coastal multi-hazards, such as hurricane induced wind and waves, slender coastal bridges could have complex dynamic responses and stress states in structural details, leading to fatigue damage accumulations. Simulation of the coupled vehicle-bridge-wind-wave (VBWW) system could be very complicated. Different loading scenarios especially associated with stochastic environmental loadings from wind and waves with potential correlations have even larger computational demands. Therefore, there is a significant need in modeling the coupled VBWW more effectively. High computational efforts are also needed to establish multi-scale coupled dynamics with sub-modeling techniques to capture the stress variations in the local details that might have short cracks initiated or propagated. To account for the uncertainties resulting from stochastic loads, machine learning (ML) algorithms have been applied in the probabilistic assessment of fatigue damage accumulation for bridges. Since the performance of different ML algorithms, such as the support vector machines (SVM) and Gaussian process (GP), strongly depends on the size and structure of the data, it is necessary to check their applications on the probabilistic fatigue damage assessment. In the present study, SVM and GP are implemented to quantify the fatigue damage of bridge details based on the established probabilistic fatigue damage assessment framework for coastal slender bridges. Firstly, the long-term field measurements are employed to build parametric probabilistic models of truck loads as well as correlated wind and wave loads. In addition to these input parameters, the multi-scale finite-element analysis (FEA) were carried out based on VBWW system to obtain the dynamic stress ranges and fatigue damage accumulation. With different training strategies, the fatigue life for critical local details can be obtained considering the life-cycle changes of coastal environmental conditions. The training and testing results show that GP algorithm outperforms SVM algorithm even though SVM exhibits reasonable capability of predicting the fatigue damage accumulation. Case studies for a coastal slender coastal bridge are provided.


  • English

Media Info

  • Media Type: Web
  • Features: References;
  • Pagination: pp 122-133
  • Monograph Title: Structures Congress 2019: Bridges, Nonbuilding and Special Structures, and Nonstructural Components

Subject/Index Terms

Filing Info

  • Accession Number: 01723006
  • Record Type: Publication
  • ISBN: 9780784482230
  • Files: TRIS, ASCE
  • Created Date: Nov 20 2019 9:42AM