Efficient prediction of compressive strength in geotechnical engineering using artificial neural networks

Abstract

In recent years, artificial neural networks (ANNs) have emerged as highly effective tools for addressing the intricate challenges encountered in geotechnical engineering. ANNs find application in a variety of geotechnical problems, showcasing promising outcomes. This study aims to improve the efficiency of predicting intermediate values from unconfined compressive strength (UCS) data obtained from laboratory tests through the use of ANNs. The modelling of artificial neural networks was carried out using the Regression Learner program, integrated with the Matlab 2023a software package, offering a user-friendly graphical interface for AI model development without the need for coding. The ANNs' validation and training were based on UCS test data obtained from the Geotechnical Laboratory of Iowa State University, USA. These laboratory tests focused on engineering properties, specifically the UCS of soils treated with biofuel co-products (BCPs). The dataset, organized in a matrix of size 216 × 5, features columns providing information on soil type (Soil 1; Soil 2; Soil 3; Soil 4), sample type (pure soil-untreated; 12% BCP- treated soil; 3% cement; 6% cement; 12% cement treated soil), time (1, 7, and 28 days), moisture content (OMC-4%, OMC%, and OMC+4%), and corresponding UCS peak stress (psi) values. The AI predictions for the test data output achieved an outstanding R2 score of 0.93, showcasing the potential of employing ANNs to efficiently acquire a substantial amount of data with fewer experiments and in less time. This approach holds promise for applications in geotechnical engineering.