What does this research mean for the field?

A 1D Convolutional Neural Network (CNN) deep learning model significantly outperforms traditional empirical models, such as FDOT, HEC-18, and Coleman's equation, in accurately predicting maximum scour depth around complex bridge piers in steady flow environments. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

June 30, 2025Advances in Bridge Engineering13 citationsOpen Access

A CNN deep learning approach to scour depth estimation around complex bridge piers in steady flow environments

NHNgoc Thi HuynhVietnam National University Ho Chi Minh City APAnh Thu Thi PhanVietnam National University Ho Chi Minh City TTTan Tai TrieuVietnam National University Ho Chi Minh City

Key Points

Key points are not available for this paper at this time.

Abstract

Abstract Scour around complex bridge piers (CBP) caused by sediment erosion due to steady flow is a critical challenge in hydraulic engineering, often leading to structural instabilities and failures. The accurate estimation of maximum scour depth is crucial for ensuring bridge safety and optimizing design. Traditional empirical methods and physics-based models, while widely utilized, often struggle to capture the complex interactions between hydrodynamic forces, sediment transport, and varying pier geometries, leading to conservative or inaccurate predictions. This study presents a one-dimensional convolutional neural network (1D CNN) and long short-term memory (LSTM) deep learning models for predicting the maximum scour depth around CBP under steady current conditions in a clear-water environment. The proposed model leverages the ability of 1D CNNs to process high-dimensional input dataset, capturing intricate non-linear relationships between influential parameters, such as flow velocity, pier configurations, sediment properties, and water depth. The dataset was transformed into non-dimensional forms using the Buckingham Pi theorem to enhance model generalization. The 1D CNN model was trained and validated using an extensive dataset, and its performance was benchmarked against established empirical models, including FDOT, HEC-18, and Coleman’s equation. Results show that the proposed 1D CNN model significantly outperforms traditional approaches, achieving higher coefficient of determination ( R 2 = 0.85) values and lower root mean squared error ( RMSE = 0.1125), mean absolute error ( MAE = 0.1078), and scatter index ( SI = 0.1149). Moreover, the model's bias ( B = -0.0194) and standard error ( SE = 0.1147) remain minimal across unseen datasets, demonstrating robust predictive capability. This research highlights the potential of deep learning as a reliable and precise tool for scour depth prediction, contributing to improved risk assessment and sustainable bridge design under steady flow environments.

问 AI

Bookmark

View Full Paper