Development of a rule-based approach for the dynamic behavior classification of soils during earthquakes

Date of Publication

1-15-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Civil Engineering

Subject Categories

Civil Engineering | Engineering | Structural Engineering

College

Gokongwei College of Engineering

Department/Unit

Civil Engineering

Thesis Advisor

Jonathan R. Dungca

Defense Panel Chair

Mary Ann Q. Adajar

Defense Panel Member

Erica Elice S. Uy

Andres Winston C. Oreta

Lessandro Estelito O. Garciano

Raymond Girard R. Tan

Abstract/Summary

Soil liquefaction during earthquakes poses significant risks to urban infrastructure, leading to ground instability, structural damage, and economic losses. Traditional liquefaction prediction methods fail to capture the complex, non-linear interactions between seismic and soil parameters, limiting their accuracy and practical application. This study aimed to develop rule-based models for assessing soil liquefaction potential using historical case histories to address these challenges.

The research employed rough set machine learning (RSML) as a viable tool to analyze case history datasets and experimental data, generating interpretable rules for six liquefaction RSML models. Model validation involved sensitivity analysis, theory-driven interpretation of rules, and comparisons with traditional liquefaction models, confirming the robustness of the proposed framework. A decision support tool was developed to translate the findings into actionable insights, featuring user-friendly interfaces for practical application. The results demonstrated high prediction accuracies ranging from 70.9% to 96.5%, validating the RSML models’ reliability and interpretability. Scenario maps and parameter interaction charts enhanced the practical understanding of soil behavior under seismic loading. Key insights included the discovery of threshold effects among grain size distribution parameters and the influence of fines content on cyclic softening, challenging traditional assumptions about soil liquefaction resistance.

The study concludes that RSML provides a robust and interpretable framework for liquefaction assessment, outperforming conventional models in both accuracy and interpretability. The decision support tool developed in this study bridges the gap between complex machine learning models and practical engineering applications, providing engineers and decision-makers a supplementary tool in seismic risk mitigation efforts.

Abstract Format

html

Language

English

Format

Electronic

Keywords

Soil liquefaction; Earthquake engineering

Upload Full Text

wf_no

Embargo Period

4-3-2025

This document is currently not available here.

Share

COinS