Artificial, convolutional, and deep neural networks for structural health monitoring: A review

Abstract

The reliability and durability of civil engineering structures rely heavily on advanced Structural Health Monitoring (SHM) systems based on in-depth knowledge. Conventional approaches, often based on manual inspections suffer from high costs, long inspection times, and subjective judgment. Neural Network-based methods have recently emerged as powerful tools, enabling not only the automation of monitoring processes but also the structuring and formalization of engineering knowledge and decision-making mechanisms.This review provides three main contributions: (a) a systematic comparison of Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), and Deep Neural Networks (DNNs) in SHM, including accuracy, computational cost, data requirements, and interpretability; (b) a synthesis of hybrid models that combine neural networks with optimisation algorithms such as, GA, PSO, AOA, and others, as an emerging trend; and (c) a critical discussion of knowledge formalisation and explainable AI in SHM, an aspect often overlooked in previous reviews. Furthermore, different learning strategies supervised, unsupervised, and reinforcement learning summarizing their strengths and limitations. Challenges related to data availability, model complexity, interpretability, and the integration of domain knowledge are also discussed. By synthesizing recent advances and identifying future research directions, this review provides researchers and practitioners with a comprehensive reference for selecting and developing neural network-based SHM systems.