Strain-based seismic damage identification in brick masonry panels through domain adversarial neural networks: a numerical study

Historic masonry structures are particularly vulnerable to seismic events, posing critical challenges for the preservation of built heritage. Structural Health Monitoring (SHM) techniques provide a promising approach for the continuous assessment of structural integrity, enabling timely and targeted retrofit interventions. However, the scarcity of real-world labelled data representing diverse damage scenarios limits the development of robust predictive models for damage identification. To partly address this challenge, this study presents a methodology based on finite element micromechanical modelling and Domain Adversarial Neural Networks (DANN) for strain-based damage identification in masonry panels. The application focuses on knowledge transfer between different panel configurations, specifically designed to investigate the dependency of the diagnostic performance on sensor characteristics. Multiple damage types are considered at varying severity levels to capture the progressive nature of structural degradation. Monitoring data are generated numerically assuming the use of smart bricks, innovative brick-like strain sensors designed for SHM of masonry structures. A core aspect of this research is the examination of how sensor distribution and quantity influence DANN performance and damage identification. By employing a domain adaptation strategy, the DANN is trained to transfer information across different structural settings, mitigating the challenges posed by limited data and varying sensor configurations. Results highlight the effectiveness of the proposed approach in identifying structural anomalies across a wide spectrum of damage types and severity levels. Overall, the methodology demonstrates significant potential for real-world SHM applications, providing preliminary insights into optimal sensor placement and system robustness for the continuous monitoring of historic masonry structures.