Reducing the Training Samples for Damage Detection of Existing Buildings through Self-Space Approximation Techniques.

Area of Science:

• Structural Engineering
• Data Science
• Heritage Conservation

Background:

• Heritage structures often lack physical models and historical data, necessitating data-driven approaches for damage detection.
• Traditional structural health monitoring (SHM) often overlooks the critical selection of damage-sensitive features and data sampling strategies.
• Existing methodologies struggle with the unique challenges posed by complex, aging structures.

Purpose of the Study:

• To propose a novel multistep strategy for selecting meaningful, correlated features for damage detection.
• To address the gap in attention given to feature selection and data sampling in SHM for heritage buildings.
• To enable damage detection as a one-class classification problem using selected features.

Main Methods:

• Development of a multistep strategy for identifying and selecting correlated, damage-sensitive features.
• Implementation of numerical methods for reducing data acquisition needs.
• Validation using a dense dataset from long-term monitoring of a heritage structure (Church of 'Santa Maria de Belém', Lisbon).

Main Results:

• The proposed strategy effectively identifies key features for damage detection in complex structures.
• Numerical methods demonstrated efficiency in reducing data sampling requirements.
• Successful validation on a real-world heritage structure confirms the strategy's applicability and performance.

Conclusions:

• The developed strategy enhances the reliability of data-driven damage detection for heritage structures.
• Optimized feature selection and data sampling are crucial for effective structural health monitoring.
• This approach provides a robust framework for assessing the condition of historical buildings.