Structural Damage Identification Using Physics-Informed Neural Network and Transfer Learning

In structural health monitoring (SHM), finite element models (FEMs) have been used to simulate diverse damage scenarios for training deep learning-based structural damage identification models; however, discrepancies between FEMs and actual structures persist. To bridge this gap, we propose a hierarchical physics-informed domain adaptation (HierPhyDA) approach that synthesizes features from FEMs to mirror those from actual structures and emulates authentic vibration signatures. The proposed solution employs an initial phase of unsupervised anomaly detection using a deep autoencoder approach, followed by a novel physics-informed domain adaptation method that serves as a digital twin of the physical structure. This approach is rigorously evaluated through numerical studies using the ASCE benchmark structure. The results show that the proposed approach outperforms state-of-the-art methods when damage cases from the actual structure are excluded from training and are mutually exclusive from those generated by the FEM.