Automated fatigue damage detection and classification technique for composite structures using Lamb waves and deep autoencoder. (15th January 2022)

• Record Type: Journal Article
• Authors: Lee, Hyunseong; Lim, Hyung Jin; Skinner, Travis; Chattopadhyay, Aditi; Hall, Asha
• Is Part Of: Mechanical systems and signal processing. Volume 163(2022)
• Publication Date: 2022-01-15
• Subjects: (max 6): Lamb waves -- PZT transducers -- Composite fatigue -- Damage detection and classification -- Deep autoencoder -- Feature learning; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2021.108148
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: Deep autoencoder-based fatigue damage diagnosis technique for composite structures. Non-destructive fatigue damage evaluation using ultrasonic Lamb waves. Fatigue damage detection by analyzing statistical patterns in Lamb wave signals. Automated composite fatigue damage classification through feature learning approach. Abstract: This paper presents the development of a robust automatic damage diagnosis technique that uses ultrasonic Lamb waves and a deep autoencoder (DAE) to detect and classify fatigue damage in composite structures. Piezoelectric (PZT) transducers are installed on carbon fiber reinforced polymer (CFRP) composite plate specimens to interrogate structural integrity under uniaxial fatigue loading. Fatigue damage evolution from matrix cracking to delamination is monitored by periodically acquiring the ultrasonic wave response. A deep autoencoder (DAE) model is adopted for effective tracking of ultrasonic response variations and for diagnosing fatigue damage in the composite specimens. The ultrasonic signals collected from pristine specimens are processed and used for training the DAE model. To improve the accuracy and sensitivity of the damage diagnosis, the architecture and hyperparameters of the DAE model are optimized, and a statistical detection baseline is defined to capture damage indicators. The ultrasonic signals obtained after applying additional fatigue cycles are introduced into the trained DAE model to validate the damage detection and classification capabilities. The damage sensitive features automatically extracted from the bottleneck layer of the DAE model are used to classify the fatigue damage mode. Singular value decomposition (SVD) is used to further reduce feature dimensionality. The patterns in the reduced features are then analyzed using a density-based spatial clustering of applications with noise (DBSCAN) algorithm. The results show that the proposed technique can accurately detect and classify the fatigue damage in composite structures, while removing the need for manual or signal processing-based damage sensitive feature extraction from ultrasonic signals for damage diagnosis.