Document Type : Research Paper
Authors
1 University of Maragheh
2 Civil Engineering Department, Engineering Faculty, Malekan Branch, Islamic Azad University, Malekan, Iran
Abstract
Keywords
Main Subjects
Scour, as a phenomenon induced by water flow that results in the erosion and removal of riverbed materials around submerged pipelines, can lead to structural instability, damage to infrastructure, and significant economic and environmental losses. Traditional scour prediction methods rely on empirical and hydraulic Methods, which often lack sufficient accuracy under complex flow conditions. With advancements in artificial intelligence (AI), Methods such as Artificial Neural Networks (ANN), Support Vector Machines (SVM), and QNET have been developed for more precise scour depth forecasting. This study is based on experimental data from Ataeian (2011) and aims to compare the performance of these three AI methods in predicting scour depth around pipelines crossing riverbeds. Key parameters investigated include flow velocity, water depth, pipe diameter, pipe clearance from the bed (e/D), and bed material characteristics. Previous research by Chao and Hennessy (1972), Kjeldsen et al. (1973), Bijker and Leeuwestein (1984), Ibrahim and Nalluri (1986), Maza (1987), Chiu (1991), De et al. (2006), Bateni et al. (2007), Choi et al. (2006), Azamatullah et al. (2012), Oliveira et al. (2013), and Khan et al. (2014) highlights the superiority of AI-based approaches over traditional models. Additionally, studies on QNET by Dey (2008, 2010), Kazilos et al. (2011), Rahimi et al. (2015), Heydari et al. (2016), Sharifi et al. (2017), Abbasi and Kazemi (2018), Mattioli et al. (2019), and Ferdesou (2020) emphasize its accuracy across diverse hydraulic conditions.
This study utilized 36 experimental datasets from Ataeian (2011), conducted in the hydraulic laboratory at Urmia University. Experiments were performed in a rectangular flume with transparent Perspex walls, measuring 8 m long, 0.6 m wide, and 0.42 m deep, with a longitudinal slope of 0.0001. The flume featured a uniform sandy bed with a median grain size (d50) of 0.56 mm. Eight pipes (diameters: 2.2–11.6 cm) were tested to evaluate scour depth. Water was supplied from an underground reservoir, regulated via a head tank, and entered the flume through a guiding channel. The test reach, a 1.8-m-long movable bed section (20 cm deep), was located 4 m from the flume inlet. A downstream movable weir (0.6 m wide, 0.36 m high) controlled water levels. Three AI Methods Artificial Neural Networks (ANN), Support Vector Machines (SVM), and QNET were assessed for scour depth prediction using 11 input combinations (Table 1). These included Froude number (Fr), pipe clearance ratio (), flow depth to pipe diameter ratio (), and bed material size to pipe diameter ratio (). ANN employs multilayer structures for learning nonlinear patterns, SVM excels in classifying complex data, and QNET leverages deep learning for optimized predictions. Model performance was evaluated using Root Mean Square Error (RMSE), coefficient of determination (R²), and Nash-Sutcliffe Efficiency (DC). Sensitivity analysis identified influential parameters by recalculating metrics after parameter removal.
The investigation revealed that the first input combination, incorporating maximum scour depth ratio (), pipe clearance ratio (), flow depth to pipe diameter ratio (), Froude number (Fr), and pipe length to diameter ratio (), yielded optimal performance across all three artificial intelligence Methods : Artificial Neural Networks (ANN), Support Vector Machines (SVM), and QNET. For ANN, the training phase produced a Root Mean Square Error (RMSE) of 0.0272, a coefficient of determination (R²) of 0.9932, and a Nash-Sutcliffe Efficiency (DC) of 0.9925, while testing results were 0.1180, 0.8959, and 0.8935, respectively. SVM exhibited training values of 0.0519 (RMSE), 0.9724 (R²), and 0.9814 (DC), with testing values of 0.1819, 0.8322, and 0.8350. QNET recorded training metrics of 0.0611, 0.9081, and 0.9080, and testing metrics of 0.1275, 0.7015, and 0.4924. Analysis of scatter plots indicated strong correlations between predicted and observed data, though slight deviations occurred at extreme values. Sensitivity analysis identified as the most influential parameter; its exclusion significantly reduced model accuracy and increased errors across all Methods . Comparative evaluation demonstrated that ANN outperformed SVM and QNET, achieving the lowest RMSE and highest R², indicating superior predictive precision. These AI-based approaches offer robust alternatives to conventional empirical Methods, enhancing scour depth prediction accuracy, reducing risks associated with pipeline failures, and supporting optimized design and management of hydraulic infrastructure.
This study demonstrates that Artificial Neural Networks (ANN) outperform Support Vector Machines (SVM) and QNET in predicting scour depth around pipelines crossing riverbeds, achieving superior accuracy with lower Root Mean Square Error (RMSE) and higher coefficient of determination (R²). The optimal input combination, incorporating maximum scour depth ratio (), pipe clearance ratio , flow depth to pipe diameter ratio (), Froude number (Fr), and pipe length to diameter ratio (), was identified as the most effective, with e/D being the most critical parameter influencing scour depth. Sensitivity analysis revealed that excluding significantly reduced model accuracy, underscoring its pivotal role in scour dynamics. These AI Methods provide robust alternatives to traditional empirical approaches, offering enhanced precision in predicting scour depth, which is vital for mitigating risks of pipeline instability and environmental damage.
However, limitations exist, including minor discrepancies in predicting extreme scour values and reliance on controlled laboratory data from Ataeian (2011). Complex real-world flow regimes, such as turbulent or unsteady conditions, may challenge model generalizability. The practical implications include improved design and maintenance of hydraulic infrastructure, reducing economic and environmental risks associated with scour-induced failures. Theoretically, this work advances the application of AI in hydraulic engineering, highlighting the efficacy of ANN in capturing nonlinear scour patterns. Future research should incorporate diverse real-world datasets and explore complex flow conditions to enhance model robustness and applicability, ultimately supporting sustainable water infrastructure management.
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
Mahdi Majedi Asl: Writing - original draft preparation
Arash Faraji: Resources, Software, Manuscript editing
Tohid Omidpour Alavian: Formal analysis and investigation
Mahdi Majedi Asl, Arash Faraji: Visualization, Supervision
Tohid Omidpour Alavian: Conceptualization, methodology
Statement: During the preparation of this work the author(s) used [ChatGpt] in order to [Text Editing]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.
This declaration does not apply to the use of basic tools for checking grammar, spelling, references, etc. If there is nothing to disclose, there is no need to add a statement.
The data supporting the findings of this study are available from the corresponding author upon reasonable request.
We express our sincere thanks and appreciation to the esteemed staff University of Maragheh.
This study did not involve human participants or animals and therefore did not require ethical approval. The authors confirm that all ethical standards of research, including avoidance of data fabrication, falsification, and plagiarism, were fully observed.
The authors declare that they have no conflict of interest.
)