Quantitative Damage Monitoring of Filament Wound Composites by Using Machine Learning-Based Techniques

Abstract

Composite structures in transportation industries have gained significant attention due to their unique characteristics, including high energy absorption. Non-destructive testing methods coupled with machine learning techniques offer valuable insights into failure mechanisms by analyzing basic parameters. In this study, damage monitoring technologies for composite tubes experiencing progressive damage were investigated. The challenges associated with quantitative failure monitoring were addressed, and the Genetic K-means algorithm, hierarchical clustering, and artificial neural network (ANN) methods were employed along with other three alternative methods. The impact characteristics and damage mechanisms of composite tubes under axial compressive load were assessed using Acoustic Emission (AE) monitoring and machine learning.Various failure modes such as matrix cracking, delamination, debonding, and fiber breakage were induced by layer bending. An increase in fibers/matrix separation and fiber breakage was observed with altered failure modes, while matrix cracking decreased Signal classification was achieved using hierarchical and K-means genetic clustering methods, providing insights into failure mode frequency ranges and corresponding amplitude ranges. The ANN model, trained with labeled data, demonstrated high accuracy in classifying data and identifying specific failure mechanisms. Comparative analysis revealed that the Random Forest model consistently outperformed the ANN and Support Vector Machine (SVM) models, exhibiting superior predictive accuracy and classification using ACC, MCC and F1-Score metrics. Moreover, our evaluation emphasized the Random Forest model's higher true positive rates and lower false positive rates. Overall, this study contributes to the understanding of model selection, performance assessment in machine learning, and failure detection in composite structures.