Machine learning-assisted fatigue performance optimization for cutout geometry of orthotropic steel bridge decks

Diaphragm cutout is a typical fatigue detail of orthotropic steel decks (OSDs). The cutout ge-ometry could alter the structural responses and fatigue performance, significantly challenging the design process. This study proposes an efficient computational framework addressing fatigue performance optimization for the cutout geometry design. Sensitivity analysis is first carried out to identify the significant parameters, and the datasets are established according to the random sampling technique and the finite element (FE) model. Then, a comparison of prediction per-formance between the back-propagation neural networks (BPNNs) and radial basis function neural networks (RBFNNs) is employed to present the applicability of two types of artificial neural networks (ANNs) in the prediction of structural responses. Finally, the cutout geometry optimization is performed by integrating the prediction model and the multi-objective particle swarm optimization (MOPSO) algorithm. The validity and applicability of the framework are demonstrated with a real-world application. The optimization results show that the fatigue life is increased from 119.8 to 150.2 years for cutout detail 1 and from 37.4 to 70.2 years for cutout detail 2. The proposed framework can significantly reduce the computational burden and deliver an optimized scheme for the cutout design.

Automated summary

This study proposes a computational framework for fatigue performance optimization of diaphragm cutout geometry design in orthotropic steel decks. Sensitivity analysis and predictive modeling using artificial neural networks are used to identify the significant parameters and optimize the cutout design. The framework is demonstrated to be effective and applicable in a real-world application.