摘要: |
While various data-driven models are proposed in the literature to forecast bridge deterioration, these models either suffer from low accuracy or are too complex to be applicable in practice. With the research team's prior work, they have demonstrated that deep learning (DL) can significantly outperform other analytical modeling methodologies in bridge deterioration forecasting. However, such models solely rely on data, and unlike physics-based models, cannot benefit from the vast knowledge and experience of bridge engineers encoded in existing physics-based models. As a result, accuracy and efficiency of these models are suboptimal. With this proposal, the team intends to develop hybrid physics-based DL models that can benefit from both effectiveness of DL and the prior knowledge encoded in physics-based bridge models. Such hybrid models are expected to outperform the DL-only models in terms of accuracy and efficiency; hence, enabling further enhanced bridge management. |