Kamiar Khayambashi

Kamiar Khayambashi

PhD researcher applying machine learning and quantitative modeling to reliability, risk analysis, and critical infrastructure systems.

Applications of Graph Neural Networks in Civil Infrastructures

Overview

Led the methodological synthesis and technical structuring of a comprehensive review of Graph Neural Network (GNN) applications across civil infrastructure systems, including transportation networks, electric power grids, water distribution systems, and structural infrastructure.

The goal of the study was to move beyond fragmented, domain-specific surveys and develop a unified analytical framework that categorizes GNN architectures, application tasks, scalability challenges, and unresolved research gaps across infrastructure sectors.

Methodological Framework

A structured review methodology was developed to identify and analyze recent GNN applications in civil infrastructure research.

Key steps included:

  • Systematic keyword searches across Web of Science, Google Scholar, and ArXiv
  • Two-stage screening for novelty, methodological contribution, and domain relevance
  • Cross-domain classification of GNN architectures
  • Task-based categorization of infrastructure applications
  • Comparative analysis of scalability, interpretability, and deployment readiness

The review synthesizes several hundred recent publications spanning multiple infrastructure domains.

Technical Synthesis

GNN Architecture Landscape

The study provides a structured comparison of major graph learning architectures used in infrastructure applications, including:

  • Graph Convolutional Networks (GCN)
  • Graph Attention Networks (GAT)
  • GraphSAGE
  • Spatio-temporal GNN models
  • Recurrent and dynamic graph neural networks
  • Autoencoder-based graph architectures
  • Graph transformers and physics-informed GNN models

Mathematical formulations and unified notation were introduced to enable consistent comparison across application domains.

Infrastructure Application Domains

Applications of GNNs were organized into key operational categories across infrastructure sectors.

Transportation Systems

  • Traffic flow and congestion prediction
  • Signal control and traffic management
  • Crash prediction and safety analysis
  • Infrastructure maintenance and resilience planning

Power Systems

  • Power flow and optimal power flow analysis
  • Renewable generation forecasting
  • Load and electricity price prediction
  • State estimation and grid monitoring
  • Cascading failure and cyber-attack detection

Water and Structural Systems

  • Water distribution network monitoring
  • Infrastructure degradation prediction
  • Structural health monitoring
  • Asset reliability assessment

Key Research Gaps

The review identifies several cross-domain challenges limiting large-scale deployment of graph learning in infrastructure systems:

  • Scalability to large real-world infrastructure networks
  • Real-time adaptation to streaming operational data
  • Integration of physical constraints into graph learning models
  • Uncertainty-aware prediction and decision support
  • Model interpretability for engineering practitioners
  • Multi-layer and multi-modal infrastructure graph representations

These gaps highlight opportunities for future research at the intersection of civil infrastructure systems and graph-based machine learning.

Contribution

This work provides one of the first cross-sector syntheses of graph neural network applications in civil infrastructure research. The resulting framework offers:

  • A unified taxonomy of GNN architectures used across infrastructure domains
  • Structured comparison of application tasks and methodological approaches
  • Identification of shared challenges across infrastructure sectors
  • A research roadmap for scalable and deployable infrastructure AI systems

Relevance

Graph neural networks are increasingly important for modeling complex infrastructure systems that exhibit network structure and spatial interdependencies. By synthesizing applications across multiple infrastructure sectors, this work helps establish a common foundation for future development of graph-based infrastructure analytics.

The review also informs emerging research directions in temporal graph learning, physics-informed graph models, and reinforcement learning–graph neural network integration for infrastructure decision support.

Authors

Harsh Anand, Kamiar Khayambashi, Zanko Zandsalimi, Mehdi Taghizadeh, Md Abul Hasnat, and Negin Alemazkoor

Preprint — Under Review