Automated Monitoring of Bridge Structures Based on Artificial Intelligence

Abstract. Problem. Bridges are a critical component of transport infrastructure, requiring continuous monitoring to detect defects and prevent failures. Traditional visual inspections and field tests are being supplanted by non-destructive methods – ground‑penetrating radar, infrared thermography, laser scanning, and video analysis – combined with IoT sensor networks. Integrating these data with artificial intelligence and machine learning algorithms enables automated anomaly detection, defect classification, and predictive maintenance, greatly enhancing diagnostic accuracy and responsiveness. The review of current approaches highlights the use of CNN, R‑CNN, YOLO, MiniRocket, and hybrid physics-informed models, and identifies key challenges: data quality and diversity, model interpretability, computational constraints in field deployment, and robustness to environmental disturbances. Future research directions are proposed to develop transparent, adaptive, and energy-efficient bridge monitoring systems. Results. The integration of non‑destructive testing methods – ground‑penetrating radar, infrared thermography, and video analysis – with continuous IoT sensor networks and artificial intelligence algorithms significantly enhances the responsiveness and objectivity of bridge monitoring, enabling the detection of microscopic defects at their earliest stages. Employing state‑of‑the‑art deep learning architectures such as YOLO, R‑CNN, and MiniRocket, alongside hybrid physics‑informed models, allows for precise damage classification, prediction of defect progression, and optimization of maintenance schedules based on risk assessment and remaining service life. However, real‑world deployment faces several critical challenges: ensuring sufficient quality and diversity of training data, increasing the interpretability of “black‑box” deep networks, and optimizing computational algorithms for edge‑device operation under field conditions. Further research should focus on developing explainable AI models, standardizing data exchange formats, designing energy‑efficient edge‑AI solutions, and achieving deeper integration of engineering physics with machine learning methods to deliver safer, more resilient, and cost‑effective bridge management strategies. Originality. This review is the first to comprehensively survey AI‑driven bridge monitoring research from 2015 to 2025, systematically classifying sensing modalities, data‑processing strategies, defect‑detection algorithms, and hybrid physics‑AI architectures. By consolidating disparate studies into a unified framework, it highlights underexplored areas, such as transparent model interpretability and field‑optimized edge inference, and outlines clear directions for future innovation. Practical value. By mapping the state of the art and pinpointing critical challenges (data quality, real‑time constraints, multimodal fusion), this work equips bridge engineers, infrastructure managers, and researchers with a structured guide to select appropriate technologies, prioritize investments, and design more reliable, cost‑effective maintenance programs – thereby accelerating safe, long‑lived bridge operations.