Browsing by Author "Samuel Hailemariam"

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Ground Penetrating Radar Simulation for Estimating Track Bed Thickness and Material Characterization.
    (Addis Ababa University, 2024-06) Samuel Hailemariam; Fiseha Nega (PhD)
    One of the most important substructures is the railway, which is a vital component of a country and demands significant investment. A railway track bed's surface layer is a multilayered structure. Three sublayers are often present: the top surface layer, known as ballast, the intermediate surface layer, sometimes known as sub-ballast, and the subsequent surface layer (Subgrade). An essential instrument for evaluating the status of railroad track beds is ground penetrating radar (GPR), which makes it possible to estimate the thickness of the track bed and classify the materials. . However, accurate interpretation of GPR data is challenged by the resolution limitations of GPR and the similar permittivity of track material sublayers. This study aims to verify and optimize GPR simulations using GprMax to improve the accuracy of determining track bed thickness and characterizing materials within railway infrastructure. The research methodology involves simulating various parametric conditions such as ballast fouling, variable track bed layer thicknesses, and different moisture content scenarios (wet and dry conditions). A structured approach is employed, starting with the establishment of study area characteristics, followed by configuring the geometry and materials in GprMax. The appropriate GPR antenna and frequency settings are then defined, and simulation settings and boundary conditions are established to ensure numerical stability and accuracy. Simulations are conducted, and the results are analyzed through post-processing techniques to examine the impact of parameter changes on GPR responses. Visualization capabilities of GprMax are utilized to compare simulated GPR scans under different conditions. The simulated results are validated against known field data or theoretical expectations to verify the simulation setup and parameters. The study concludes that GPR simulations in GprMax can effectively model the impact of ballast fouling, layer thickness variations, and moisture content on GPR signals. These simulations provide valuable insights into improving GPR data interpretation, promoting cost-effective maintenance strategies by reducing the need for extensive physical testing. This research contributes to enhancing the reliability and efficiency of GPR in railway infrastructure maintenance.
  • No Thumbnail Available
    Item
    Modular Federated Learning for Non-IID Data
    (Addis Ababa University, 2025-06) Samuel Hailemariam; Beakal Gizachew (PhD)
    Federated Learning (FL) promises privacy-preserving collaboration across distributed clients but is hampered by three key challenges: severe accuracy degradation under non-IID data, high communication and computational demands on edge devices, and a lack of built-in explainability for debugging, user trust, and regulatory compliance. To bridge this gap, we propose two modular FL pipelines—SPATL-XL and SPATL-XLC—that integrate SHAP-driven pruning with, in the latter, dynamic client clustering. SPATL-XL applies SHAP-based pruning to the largest layers, removing low-impact parameters to both reduce model size and sharpen interpretability, whereas SPATL-XLC further groups clients via lightweight clustering to reduce communication overhead and smooth convergence in low-bandwidth, high-client settings. In experiments on CIFAR-10 and Fashion-MNIT over 200 communication rounds under IID and Dirichlet non-IID splits, our pipelines lower per-round communication to 13.26 MB, speed up end-to-end training by 1.13×, raise explanation fidelity from 30–50% to 89%, match or closely approach SCAFFOLD’s 70.64% top-1 accuracy (SPATL-XL: 70.36%), and maintain stable clustering quality (Silhouette, CHI, DBI) even when only 40–70% of clients participate. These results demonstrate that combining explainability-driven pruning with adaptive clustering yields practical, communication-efficient, and regulation-ready FL pipelines that simultaneously address non-IID bias, resource constraints, and transparency requirements.