Predictive assessment of delamination characteristics in E-glass/epoxy composites using sequential back-propagation neural networks

• The effect of the length and thickness positions of delamination is investigated. • The coupling impact of delamination is evaluated in E-glass/epoxy composite. • A BP neural network is provided to predict the important output variables. • The proposed method can adequately identify the presence of damage. This study focuses on the detection and quantification of delaminations in E-glass/epoxy composite samples. A comprehensive investigation involving 603 distinct numerical tests was conducted, each varying in terms of the intensity, number, and spatial arrangement of damage traversing the sample's length and thickness. The primary objective was to extract the first five natural frequencies from these samples. To enable damage prediction, we devised a sequential back-propagation artificial neural network. This network was trained utilizing the initial five natural frequencies as input data. Importantly, the output of each network in the sequence was fed as input to the subsequent network. The results underscored the network's efficacy and robustness in predicting damage severity, count, and precise locations along both the sample's length and thickness. Furthermore, an exploration of the influence of delaminations on the natural frequency values revealed a coherent and meaningful correlation. Notably, variations in the natural frequency values demonstrated a consistent relationship with damage attributes, encompassing both intensity and spatial distribution (across the length and thickness of the sample). This study thus establishes a sound foundation for employing sequential neural networks in the accurate assessment of delamination characteristics within composite structures, while also shedding light on the interconnectedness of damage features with alterations in natural frequency behavior.