Entropy-Based Feature Extraction for Structural Health Monitoring: A Case Study on the Z24 Bridge in Switzerland

Structural damage alters a structure’s dynamic properties, complicating health monitoring due to the need for accurate extraction and analysis of damage-sensitive features. The diversity and complexity of potential damage types further increase system unpredictability. To address this, entropy-based methods are employed to quantify and compare the complexity of structural states, highlighting unique damage characteristics. While extensive research has focused on feature extraction, there is a notable lack of comprehensive reviews of entropy methods, especially advanced multiscale and composite multiscale models, and their effectiveness in noise-prone environments. This study addresses these gaps by analyzing data from Switzerland’s Z24 Bridge, specifically evaluating the impact of noise on damage detection accuracy. Findings indicate that composite multiscale and multiscale entropies significantly outperform simpler entropy measures, with Shannon’s composite multiscale entropy showing superior resilience to noise and effectiveness in detecting both the presence and severity of damage.