Experimental and Numerical Modeling of Sustainable Railway Ballast

Abstract

A railway network is a modern, energy-efficient transportation system that connects business, industry, and residential areas for both freight and passenger movement. The construction of durable railway infrastructure relies on high-quality ballast aggregates, traditionally sourced from natural volcanic rock such as basalt or gabbro. In the United Arab Emirates (UAE), gabbro is primarily quarried from Fujairah. However, the continued reliance on these natural quarries raises concerns about resource depletion. Concurrently, the accumulation of industrial byproducts, such as Electric Arc Furnace (EAF) slag, presents an environmental challenge that necessitates sustainable reuse strategies. This study investigates the feasibility of using EAF slag as a sustainable alternative to gabbro for railway ballast in the UAE through a comprehensive program of laboratory testing, constitutive model development, and machine learning applications. A series of rigorous experimental investigations, including physical and mechanical tests, are conducted to compare the performance of EAF slag with gabbro. Monotonic, cyclic, and post-cyclic triaxial tests are performed to evaluate key parameters such as shear strength, stiffness, permanent deformation, and degradation characteristics under field-representative stress conditions.
Beyond experimental analysis, this research develops and validates a methodology for simulating ballast behavior under different loading conditions. The NorSand model is first employed to simulate the monotonic response of ballast and analyze uncertainties in the critical state line, addressing challenges in capturing critical state behavior through laboratory experiments. The impact of parameter uncertainty on model predictions is carefully examined through parameter covariance analysis. The conditional covariance analysis reveals the most important influential parameters across strain regimes. Novel and detailed parameter calibration strategies were presented that increased practical applicability of NorSand model to model drained monotonic behavior of ballast aggregates. Additionally, a new constitutive model is proposed based on the existing SANISAND04 model framework to model cyclic densification of ballast aggregates under high cycle, large amplitude cyclic loading. The proposed model introduces the concept of terminal state parameter and its corresponding limiting maximum friction angle, along with an evolution of the plastic modulus that accounts for accumulated compressive and dilative plastic volumetric strain related to fabric reinforcement and destruction in to bounding surface model. These modifications enable the model to capture stress history, cyclic densification, and hysteresis effects under long term repeated cyclic loading. The proposed model is validated for different ballast aggregates and results demonstrate improved performance over the original SANISAND04 model. Furthermore, both the NorSand and SANISAND04 are extended to account for particle breakage when degradation becomes significant, though laboratory tests indicate minimal breakage in the studied materials in this study.
In parallel, a deep learning–based (DL) constitutive model is proposed as a flexible alternative to traditional constitutive model approaches. A recurrent encoder-decoder framework using long short-term memory (LSTM) networks is developed to learn stress strain behavior directly from triaxial test data. The model leverages soil-specific features such as void ratio, relative density, and particle size, eliminating the need for manual calibration of constitutive parameters and enhancing adaptability to site-specific conditions. Additionally, it offers faster deployment, reduced reliance on expert judgment, and improved scalability across diverse soil types. Comparisons of the proposed DL model with conventional models such as NorSand model, reveals that the DL approach offers faster, more accurate predictions suitable for smart infrastructure monitoring.
This study has two important implications for rail infrastructure and geotechnical engineering. First, by demonstrating that electric arc furnace (EAF) slag can be a sustainable alternative to traditional ballast, the study promotes cost savings, the reuse of industrial waste, and progress toward a circular economy. Second, the development of advanced constitutive and deep learning (DL) models gives railway engineers powerful tools for predicting ballast behavior. These tools can improve track design, lower maintenance costs, and promote digital innovation in geotechnical modeling.

Date of Award	2025
Original language	American English
Supervisor	Tadahiro Kishida (Supervisor)