Advancing Damage Assessment of CFRP-Composite through BILSTM and Hilbert Upper Envelope Analysis

• Published in: Russian journal of nondestructive testing Vol. 59; no. 12; pp. 1241 - 1258
• Main Authors: Frik, M., Benkedjouh, T., Essaidi, A. Bouzar, Boumediene, F.
• Format: Journal Article
• Language: English
• Published: Moscow Pleiades Publishing 01.12.2023; Springer Nature B.V
• Subjects: Acoustic emission; Acoustic Methods; Aerospace industry; Automobile industry; Carbon fiber reinforced plastics; Carbon fiber reinforcement; Characterization and Evaluation of Materials; Chemistry and Materials Science; Composite structures; Damage assessment; Defects; Fatigue tests; Hilbert transformation; Laminates; Materials fatigue; Materials Science; Separation; Structural health monitoring; Structural Materials; deep learning; Hilbert transform; BiLSTM; failure analysis; composite materials
• ISSN: 1061-8309; 1608-3385
• DOI: 10.1134/S106183092360082X

The aerospace and automotive sectors widely use carbon fiber reinforced plastic because of its exceptional properties, including its high specific modulus, strength, and resistance to fatigue. However, defects such as cracks in the matrix, separation of layers, and separation from bonding can occur during manufacturing and low-velocity impacts, often remaining undetected. As these defects worsen over time, they can significantly weaken the material. To reduce the risk of major failures, regular assessments of carbon fiber reinforced plastic structures are crucial. This study introduces a structural health monitoring technique that minimizes human involvement while effectively tracking the growth of damage in carbon fiber reinforced plastic structures. The approach employs the acoustic emission method and the hilbert transform technique to identify and quantify the progression of damage in carbon fiber reinforced plastic materials. Experimental outcomes from a fatigue test conducted on cross-ply laminates are presented. To precisely predict damage and evaluate the condition of the composite specimen, researchers use the bidirectional long short-term memory model alongside envelope analysis for forecasting. The suggested method achieves a root mean square error of less than 0.03, proving its capability to precisely predict damage and evaluate the condition of the Composite structure. This novel deep learning-driven method adeptly captures the deterioration in performance of carbon fiber reinforced plastic, enhancing predictive accuracy.