Physics-inspired deep learning for bridge scour prediction: site-specific and transferable models
Negin Yousefpour, Bo Wang
Abstract
We introduce Scour Physics-Inspired Neural Networks (SPINNs), a hybrid physics-data-driven framework for bridge scour prediction using deep learning. SPINNs integrate physics-based, empirical equations into deep neural networks and are trained using historical bridge scour monitoring data. Long-short Term Memory Network (LSTM) and Convolutional Neural Network (CNN) are considered as the base deep learning (DL) models. Despite variation in performance, SPINNs outperformed pure data-driven models in the majority of cases. In some bridge cases, SPINN reduced forecasting errors by up to 50 percent. We also explore transferable/general models, trained by aggregating datasets from a cluster of bridges, versus the site/bridge-specific models. Both pure data-driven and hybrid (SPINN) models showed transferability across bridges in Alaska. The general DL models particularly proved effective for bridges with limited data. In addition, the calibrated time-dependent empirical equations derived from SPINNs showed great potential for maximum scour depth estimation. The proposed time-dependent SPINNs and the corresponding calibrated empirical equations demonstrate substantial improvement in scour prediction accuracy compared with the traditional HEC-18 empirical model. This study can pave the way for further exploration of physics-inspired machine learning methods for scour prediction.