Integrated Physics-Informed Machine Learning Framework for Structural Damage Detection, Localization, and Severity...

Area of Science:

• Civil Engineering
• Structural Health Monitoring (SHM)
• Computational Mechanics

Background:

• SHM is crucial for civil infrastructure safety, involving damage detection, localization, and quantification.
• Existing physics-based and data-driven methods have limitations in accuracy and data requirements.
• A unified framework for all three SHM tasks is needed for comprehensive assessment.

Purpose of the Study:

• To propose an integrated hierarchical physics-informed domain adaptation (I-HierPhyDA) framework for SHM.
• To address limitations of physics-based and data-driven approaches in structural damage assessment.
• To perform damage detection, localization, and severity classification in a unified, hierarchical manner.

Main Methods:

• Developed an I-HierPhyDA framework integrating reduced-order and higher-fidelity finite element models (FEM).
• Generated consistent vibration signatures across FEM domains for improved accuracy.
• Employed domain adaptation techniques for damage localization and severity classification without target domain training data.

Main Results:

• The framework accurately detects and localizes structural damage in numerical benchmark models.
• Achieved superior performance in damage severity classification, evidenced by high mean accuracy and Macro-F1 scores.
• Demonstrated robust results under structural uncertainties and measurement noise, with low metric standard deviations.

Conclusions:

• The proposed I-HierPhyDA framework offers an effective solution for comprehensive structural condition assessment.
• The approach successfully integrates damage detection, localization, and quantification within a unified system.
• Future research will focus on experimental validation using real-world structural data.