Deep learning has revolutionized structural health monitoring (SHM), yet data scarcity remains a critical bottleneck limiting its deployment in real-world engineering applications. Meta-learning—an emerging paradigm enabling models to learn from limited examples—offers a compelling solution to this challenge. Herein, we systematically investigate meta-learning’s efficacy across three key SHM applications: surface damage detection, structural response prediction, and data-driven damage identification. Our experiments demonstrate that meta-learning achieves comparable performance with substantially reduced data requirements. For surface damage detection, meta-learning maintains detection accuracy while modestly decreasing sample dependency. In response prediction tasks, although the number of prediction errors increases marginally, the data efficiency gains are substantial. Similarly, damage identification shows slight accuracy trade-offs but dramatic reductions in required training samples. These findings establish meta-learning as a practical pathway for deploying deep learning in data-constrained SHM scenarios, potentially accelerating the adoption of intelligent monitoring systems in critical infrastructure. Our results suggest that the traditional data-hungry nature of deep learning need not be a barrier to advancing automated structural health assessment.
Yu et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: