Temporal Evaluation of Scour Hole Dimensions due to Plain Wall Jets in Noncohesive Sediments Using a Soft Computing Approach: White-Box versus Black-Box Modeling

Abstract

Jet scour presents a significant challenge for hydrological analysis and hydraulic design of river structures, with the temporal dynamics of scour hole dimensions posing a critical concern. This study analyzed the effectiveness of two AI-based models, extreme learning machine (ELM) and multigen genetic programming (MGGP), in predicting these fluctuations and identifying governing parameters. Both models demonstrated substantial predictive accuracy, exceeding the performance of existing empirical models. MGGP outperformed ELM in the training and testing phases, yielding four interpretable equations for practical applications. These equations enable designers to precisely predict temporal variations in scour hole dimensions based on key parameters, with nondimensional scouring time identified as the most influential factor. Surprisingly, channel width ratio and sediment standard deviation impacted model accuracy minimally. Additionally, the study emphasized the relevance of using the densiometric Froude number to capture temporal scour hole dynamics from plain wall jets. This research underscores the potential of using AI-based models to enhance scour prediction and design optimization of related structures. The proposed MGGP equations offer a practically relevant and accurate tool for managing jet scour, surpassing the limitations of previous approaches.