Determination of Basalt Fiber Reinforcement in Kaolin Clay: Experimental and Neural Network-Based Analysis of Liquid Limit, Plastic Limit, and Unconfined Compressive Strength

Abstract

The use of basalt fibers, which are employed in various fields, such as construction, automotive, chemical, and petrochemical industries, the sports industry, and energy engineering, is also increasingly common in soil reinforcement studies, another application area of geotechnical engineering, alongside their use in concrete. With this growing application, scientific studies on soil reinforcement with basalt fiber have also gained momentum. This study establishes the effects of basalt fiber on the liquid limit, plastic limit, and strength properties of soils, and the relationships among the liquid limit, plastic limit, and unconfined compressive strength of the soil. For this purpose, 12 mm basalt fiber was used as a reinforcement material in kaolin clay at ratios of 1.0%, 1.5%, 2.0%, 2.5%, and 3.0%. The prepared samples were subjected to liquid limit, plastic limit, and unconfined compressive strength tests. As a result of the experimental studies, the fiber ratio that provided the best improvement in the soil properties was determined, and the relationships among the liquid limit, plastic limit, and unconfined compressive strength were established. The experimental results were then used as input data for an artificial intelligence model. The used neural network (NN) was trained to obtain basalt fiber-to-kaolin ratios based on the liquid limit, plastic limit, and unconfined compressive strength. This model enabled the prediction of the fiber ratio that provides the maximum improvement in the liquid limit, plastic limit, and compressive strength without the need for experiments. The NN results were in great agreement with the experimental results, demonstrating that the fiber ratio providing the maximum improvement in the soil properties can be identified using the NN model without requiring experimental studies. Moreover, the performance and reliability of the NN model were evaluated using 5-fold cross-validation and compared with other AI methods. The ANN model demonstrated superior predictive accuracy, achieving the highest correlation coefficient (R = 0.82), outperforming the other models in terms of both accuracy and reliability.