Keywords

Hydraulic conductivity, compacted soil, van Genuchten parameters, machine learning, deep learning

Abstract

Machine learning models (MLMs) were developed to predict the saturated hydraulic conductivity of compacted soil barriers and help to identify appropriate soils for the construction of landfill liners and covers. Three machine learning algorithms (random forest, gradient boosting decision tree, and neural network) were used to develop MLMs, and multiple linear regression was used to compare the precision of predictions with the MLMs. Results from this study showed that the random forest model provided the best prediction of the hydraulic conductivity of compacted soil barriers, with 100% of predicted hydraulic conductivity within 100-time differences to measured hydraulic conductivity and 93% within 10-time differences. Feature importance analysis showed that void ratio after compaction, fines content, specific gravity, degree of saturation after compaction, and plasticity index of soils are the top-five factors that influence the hydraulic conductivity of compacted soil barriers and are recommended for a precise prediction. Unsaturated hydraulic properties of compacted soils were studied using artificial intelligence methods as well. Except for the basic statistical method linear regression and one machine learning algorithm (random forest), two deep learning algorithms, such as multilayer perceptron (MLP)-ReLU and convolutional neural networks (CNN), were utilized to develop models to predict van Genuchten (VG) parameters (i.e. α and n) which are crucial for estimating the unsaturated hydraulic conductivity of compacted soils via soil water characteristics curve (SWCC). MLP-ReLU performed best on predicting the n parameter, which has 92.6% and 63% of the predicted n values within the range of 1.2-fold and 1.1-fold of the calculated n parameters respectively. As for predicting the α parameter, Random Forest had the smallest MSE which is 0.0048, and the smallest difference between predicted and calculated α parameters, 96.3% and 64.8% of the predicted α values are within 5-fold and 2-fold of the calculated α values respectively.

Completion Date

2024

Semester

Summer

Committee Chair

Chen, Jiannan

Degree

Master of Science in Civil Engineering (M.S.C.E.)

College

College of Engineering and Computer Science

Department

Department of Civil, Environmental and Construction Engineering

Degree Program

Civil Engineering

Format

application/pdf

Identifier

DP0028894

Language

English

Rights

In copyright

Release Date

2-15-2025

Length of Campus-only Access

None

Access Status

Masters Thesis (Open Access)

Campus Location

Orlando (Main) Campus

Accessibility Status

Meets minimum standards for ETDs/HUTs

Download

Share

COinS