📝 SummarXiv

Abstract

Efficient and intelligent assessment of post-earthquake structural damage is critical for rapid disaster response. While data-driven approaches have shown promise, traditional supervised learning methods rely on extensive labeled datasets, which are often impractical to obtain for damaged structures. To address this limitation, we propose a physics-informed multi-source domain adaptation framework to predict post-earthquake structural damage for a target building without requiring damage labels. The multi-source domain integrates actual damage data and numerical modeling data from buildings similar to the target structure. The framework operates through three key steps. First, the similarity of key physics from each domain are analyzed to form a weight matrix, which enhances domain differentiation. Second, features from the multi-source and target domains are extracted and fed into a classifier and a discriminator. The classifier ensures that the features are damage-sensitive and accurately assign damage states, while the discriminator enforces that the features remain domain-invariant. Finally, the key parameters matrix is applied as weights during adversarial training to optimize the contribution of features from each source domain. The proposed framework provides a robust solution for assessing structural damage in scenarios where labeled data is scarce, significantly advancing the capabilities of post-earthquake damage evaluation.