Advanced prediction of soil shear strength parameters using index properties and artificial neural network approach

Abstract

This study embarks on developing predictive models for soil shear strength parameters, cohesion (c) and angle of internal friction (ϕ), in Bishoftu town, employing Artificial Neural Networks (ANN). It aims at offering a cost-effective and time-saving alternative to traditional, often expensive, and labor-intensive laboratory methods. The research utilizes soil index properties such as Sand %, Fines %, Liquid Limit, Plastic Limit, and Plasticity Index to construct separate ANN models for c and ϕ. These models use a multi-layer perceptron network with feed-forward back propagation, varying the number of hidden layers to optimize performance. The study's dataset comprises 316 soil test results, encompassing both primary and secondary data, conforming to ASTM Standards. Soil cohesion and internal friction angle were determined using the direct shear box method. The models demonstrated remarkable success in predicting shear strength parameters, evidenced by correlation values of approximately 0.99 for cohesion and 0.98 for internal friction angle, surpassing the capabilities of existing empirical methods. Further examination of the models included comparison with existing correlation techniques and cross-validation using primary soil test data. This validation process confirmed the ANN method's superior accuracy and fit for predicting shear strength parameters over selected empirical methods. This research substantiates the efficiency of ANN in geotechnical engineering, particularly for areas with limited resources for extensive soil testing. It establishes ANN as a powerful, efficient tool for estimating soil shear strength parameters, with significant implications for future planning, design, and construction projects in similar environments.