Development of an ANN-based model for predicting unfrozen water content in sandy soils: Insights from experimental data and GUI implementation
Year: 2025
Type: article
Abstract: Predicting the unfrozen water content curve is essential to understanding the thermal‒hydraulic‒mechanical behavior of frozen soil. However, existing methods often require extensive experiments. When experimental data is lacking, these methods rely on other empirical models, resulting in overly complex workflows. To overcome these issues, we developed two artificial neural network (ANN) models to predict the unfrozen water content of saline sandy soils for two distinct soil types. The trained model files were then implemented in a Graphic User Interface (GUI) program for end-users. The training datasets were collected through a series of laboratory experiments that varied void ratio, pore water salinity, and soil type using a newly designed freezing chamber system. A total of 186 datasets were collected and divided into training and testing sets. The training results have a coefficient of determination (R 2 ) greater than 99% for both ANN models. Meanwhile, the R 2 of test results were over 98% demonstrating the accuracy of predicting the unfrozen water content, closely matching the measured values. In addition, the trained ANN models are embedded in a GUI program for wide application in coastal and permafrost environments.
Authors: Hee Won Kim, Dinh-Viet Le, Gyu-Hyun Go
Institution: Kumoh National Institute of Technology
Language: English
Topic: Soil and Unsaturated Flow
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,800
Award: G5763159341
OpenAlex ID: W4415931289