Strain time history prediction of bridges using a deep learning approach

Bridges are considered one of the most critical components of transportation infrastructure. Thus, an accurate response prediction of bridge structures is crucial for long-term monitoring and safety assessment. To accomplish this, the current study presents a deep learning-based approach for predicting strain responses of the I35W Bridge at three different locations. For this purpose, the five-year measured data are adopted from this concrete box girder bridge located in Minneapolis, Minnesota. In this approach, the collected datasets are utilized as inputs to the multilevel deep neural networks, namely, deep long short-term memory (D-LSTM), deep gated recurrent unit (DGRU), and modified generative adversarial networks (GANs). The performance of these networks is also assessed using the root mean square error (RMSE), the mean absolute error (MAE), and the coefficient of determination (R2) indices. The findings reveal that the deep learning-based approach is a computationally effective, promising, and reliable method for accurately predicting the future response of the bridge structure. The comparison of the final results also indicates that the proposed D-GRU model provides the best prediction accuracy and performance among the evaluated models.