Machine learning analysis based on deep learning for fatigue diagnostics in carbon fiber reinforced polymers.

Abstract

Fatigue-induced degradation in Carbon Fiber Reinforced Polymer (CFRP) structures poses a critical challenge in long-term structural health monitoring (SHM) applications. In this study, a hybrid deep learning framework is proposed for fatigue state classification of CFRP composites using sensor-based monitoring data. The framework integrates a one-dimensional Convolutional Neural Network (1D-CNN) to extract spatial degradation patterns and an extended Long Short-Term Memory (xLSTM) network to capture long-range temporal dependencies associated with fatigue evolution. The extracted spatiotemporal features are fused and refined through Mutual Information-based feature selection, followed by a Bagging-based ensemble classifier for robust fatigue state discrimination. The proposed approach is evaluated using the NASA-CFRP dataset, achieving an average classification accuracy of 99%. While the framework is generally applicable to SHM of CFRP structures, its relevance to membrane-based gas separation systems is discussed as a representative application scenario. The results demonstrate the effectiveness of the proposed method for reliable fatigue diagnosis and maintenance decision support in CFRP-based engineering systems.