On the Use of Machine Learning for Damage Assessment in Composite Structures: A Review

• Published in: Applied composite materials 2024-02, Vol.31 (1), p.1-37
• Main Authors: Ribeiro Junior, Ronny Francis, Gomes, Guilherme Ferreira
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Characterization and Evaluation of Materials; Chemistry and Materials Science; Classical Mechanics; Complexity; Composite materials; Composite structures; Damage assessment; Damage detection; Damage localization; Failure modes; Heterogeneity; Industrial Chemistry/Chemical Engineering; Machine learning; Materials Science; Neural networks; Polymer Sciences; Principal components analysis
• ISSN: 0929-189X; 1573-4897
• DOI: 10.1007/s10443-023-10161-5

Composite materials are those formed by combining two or more different materials to take advantage of the best characteristics of each one. However, due to this heterogeneity, composite materials suffer from non-linear failure modes. Because of this complexity, damage to composite structures cannot be identified by visual inspection or conventional techniques. Therefore, several complex techniques are employed in this type of material, with Machine Learning being the main way of dealing with the excessive data extracted from these techniques. Given the rapidly increasing use of composite materials in real-world applications, the demand for damage assessment (detection, quantification, and localization) methods is increasingly high. This article reviews the main and most recent works on ML methods for the damage assessment of composite structures. The selected studies are then covered in detail to provide researchers with an in-depth comprehension of what is new in ML algorithms for the damage assessment of composite structures. From 2019 to now, there has been a large increase in the number of publications related to damage assessments of composite materials, with a strong predominance of Artificial Neural Networks, Convolutional Neural Networks, and Principal Component Analysis techniques. However, there is still a lack of studies with real cases in real environments. Finally, future research directions and a summary of all selected works are suggested, presenting possible improvements of the state of the art.