Introducing a Physics-informed Deep Learning Framework for Bridge Scour Prediction

• DOI: 10.48550/arxiv.2407.01258
• Publication Date: 2024-07-01
• Authors: Negin Yousefpour; Bo Wang
• Keywords: Bridge (graph theory)

This paper introduces scour physics-informed neural networks (SPINNs), a hybrid physics-data-driven framework for bridge prediction using deep learning. SPINNs are developed based on historical monitoring data and integrate physics-based empirical equations into as supplementary loss components. We incorporated three architectures: LSTM, CNN, NLinear the base data-driven model. Despite varying performance across different models bridges, overall outperformed pure models. In some cases, SPINN reduced forecasting errors by up to 50 percent. this study, we also explored general clusters, trained aggregating datasets multiple bridges in region. The mostly benefited from approach, particular with limited data. However, bridge-specific provided more accurate predictions than almost all case studies. Also, time-dependent derived showed reasonable accuracy estimating maximum depth, providing compared HEC-18. Comparing both learning traditional HEC-18 equation indicates substantial improvements accuracy. study can pave way physics-machine methodologies be implemented design maintenance.