Application of an Artificial Neural Network Model Based on Mineral Composition to the Prediction of Physical and Mechanical Properties

The physical and mechanical properties of soil are crucial in engineering construction, but conducting extensive experimental tests can be time-consuming, laborious, and subject to uncertainties due to the heterogeneity of the soil and variations in experimental conditions. Soil is composed of various minerals, and the mineral composition is the fundamental determinant of various physical and mechanical properties of the soil. The purpose of this study is to establish a convenient and reliable soil property prediction model based on mineral composition. To achieve this end, a dataset comprising the percentage content of different minerals in the soil, as well as the soil's mechanical and physical properties, was collected. Using artificial neural network methods, prediction models for liquid limit, plastic limit, internal friction angle, and cohesion of the soil were developed based on mineral composition. Each model's performance was evaluated through deviation analysis, and models with poor prediction accuracy were optimized using a genetic algorithm. The results demonstrate that the neural network model based on mineral composition can accurately predict soil properties with high applicability. This research provides a method for accurately predicting the majority of engineering properties of soil using experimental data on mineral composition, which is significant for cost savings and improving work efficiency in engineering projects.