Machine learning based classification of guided wave signals in the context of inter-specimen variabilities
Résumé
In the context of Guided Wave-based Structural Health Monitoring (GW-SHM), ultrasonic elastic waves are used to detect damages in structures by comparing the acquired signals with those from a defectfree structure. However, the high sensitivity of GWs to environmental and operational conditions limits the validity of such references. Notably, variabilities between multiple specimens are often significant from the GWs perspective. These variabilities are particularly important in composites and are due to sensor positioning, sensor coupling and material variability. This communication presents a baseline-free approach using physics-enhanced Machine Learning (ML) for enhanced robustness. To ensure the coverage of these variabilities the approach is validated on multiple Carbon-fiber-reinforced polymers (CFRP) panels. The methodology relies on feature extraction from raw GW signals and training classification algorithms (e.g., kernel machines, neural networks). To make the classifier learn inter-specimen variabilities, an experimental database of 45 impacted composite panels is used. Half of them are used to provide pristine data, and the rest to provide damaged data so that the same sample is either in the training or test set, but never in both. Good classification performance is obtained, demonstrating that the classifier has successfully learnt to recognize defect signatures despite the variability linked to the multiple specimens and instrumentations.
Mots clés
Guided Waves based SHM
Baseline-free approach
Machine learning
Material and instrumentation variabilities
CFRP
Artificial intelligence
Structural Health Monitoring
ultrasound
instrumentation
ultrasonic elastic waves
damage detection
defect
sensor
sensor positioning
sensor coupling
material variability
multiple Carbon-fiber-reinforced polymers
signal processing
classification
online learning
kernel machine
neural network
defect signature
Origine | Fichiers produits par l'(les) auteur(s) |
---|