Moment and shear estimations in steel girder highway bridges using committees of artificial neural networks and finite element models

The moment and shear demands generated by the applied rating loads are required for girder bridge load-rating evaluation. Load demands are typically based on line-girder (1D) analyses. Bridge design and evaluation specifications such as AASHTO allow for less conservative and more rigorous methods, but they have an uncertain return on extraneous efforts employed. This research shows that a small set of refined analyses combined with ANNs can be used to predict the likely outcome for similar methods and bridges in a population. For moment and shear, two ANN-based prediction models were considered: (1) single-best-network and (2) committee networks (CN). The accuracies of moment and shear prediction were investigated on a hybrid subset of bridges that included both hypothetical and real bridges representative of a steel bridge inventory. The governing inputs (structural and geometric bridge characteristics) were mapped to moment and shear values obtained from 3D FE analyses using ANN-based prediction models. Refined moment and shear demands can be reliably estimated (about 5 and 4% mean absolute errors, respectively) using a properly trained network model with an optimized model complexity, according to prediction accuracy for bridges outside the training subset. The CN model outperformed the single-best-network approach in terms of shear prediction accuracy and confidence levels.