Soil–conduit interaction: An artificial intelligence application for reinforced concrete and corrugated steel conduits
Document Type
Journal Article
Publication Title
Neural Computing and Applications
Publisher
Springer
School
School of Engineering
RAS ID
35985
Funders
Higher Education Commission, Government of the Islamic Republic of Pakistan
Abstract
Marston’s load theory is commonly used for understanding the soil–conduit interaction. However, there are no practical methods available which can estimate the Marston’s soil prism (MSP) width ratio. Moreover, the advent of soft computing methods has made many traditional approaches antiquated. The main purpose of this work is to compare and evaluate the predictive abilities of several machine learning-based models in predicting the MSP width ratio for the reinforced concrete (RC) and corrugated steel (CS) conduits. By utilizing the finite element modelling, a large-scale dataset was generated for the width of the soil prism for both types of conduit material, when buried under sandy soils of varying stiffness. After preparing the required dataset, feature validity technique based on correlation-based feature selection was employed to find the most influential parameters affecting the MSP width. Thereafter, five regression-based data driven models namely artificial neural networks (ANN), least-square support vector regression, extreme learning machine, Gaussian process regression, and multiple linear regression were developed to forecast the MSP width ratio. The results showed that the ANN outperforms the other predictive models for both the conduit types. In addition, due to the excellent overall performance of the ANN, it was translated into functional relationship for predicting the MSP width ratio for RC and CS conduits.
DOI
10.1007/s00521-021-06125-0
Access Rights
subscription content
Comments
Khan, M. U. A., Shukla, S. K., & Raja, M. N. A. (2021). Soil–conduit interaction: An artificial intelligence application for reinforced concrete and corrugated steel conduits. Neural Computing and Applications, 33(21), 14861-14885.
https://doi.org/10.1007/s00521-021-06125-0