Bridge pressure scour occurs when the bridge deck is inundated during a flood event, causing significant threats to the stability of local streambeds and bridge foundations. In the present study, new flume experiments were conducted to investigate the influence of the flow intensity, bridge deck clearance height, bridge deck thickness, and bridge deck longitudinal length on pressure scour depth. A comprehensive pressure scour dataset was assembled, using the new experimental data and existing data from the literature, to examine detailed parameter dependencies on scour depth and derive new empirical predictive equations. During this process, XGBoost, a machine learning (ML) model, was also developed using the dataset to assist parameter analysis, primarily as a parameter analyzer to corroborate variable selection and functional dependencies, and secondarily to provide an alternative, data-driven approach to scour prediction. The results show that, by using the newly proposed equations, 47% data are within the ±20% accuracy range, while the rate for over 20% underestimation was below 12%. This performance is significantly better than the existing methods and is beneficial in its theoretical framework. While the ML model attains strong predictive accuracy, we recommend treating it as a diagnostic parameter-analysis tool at the current stage rather than a reliable design method; the lack of a physical basis limits its generalizability and makes it important to exercise extra caution in practice. Finally, future research directions are discussed in this paper, particularly the future fusion of conventional empirical methods and ML models.
Tang et al. (Fri,) studied this question.