A method to predict stiffness evolution of 3D woven composites based on acoustic emission and deep learning

The deployment of 3D woven composites is ubiquitous in the realm of engineering, with their fatigue properties being of paramount importance for both the functionality and safety of composite structures. Traditional stiffness modeling approaches require extensive experimental datasets for parameterization, limiting their applicability and accuracy. To address this challenge, we propose a deep learning strategy for precise prediction of stiffness degradation in composite materials using Acoustic Emission (AE) data. The approach includes a sophisticated data aggregation technique that efficiently condenses the dimensions of AE data while substantially preserving the core features of the dataset. The Pearson correlation coefficient and the Random Forest (RF) algorithm are employed to identify features that are highly sensitive to variations in stiffness. The deep learning model introduced herein is capable of predicting stiffness degradation across various stress ratios with a relative error of less than 8.89%. It is crucial to acknowledge that the accuracy of the time series prediction is significantly influenced by the series length. Therefore, to ensure an accurate prediction of stiffness degradation, it is imperative to meticulously consider adjustments for varying materials and loading conditions.