Impact energy assessment of sandwich composites using an ensemble approach boosted by deep learning and electromechanical impedance

Abstract

Sandwich composites are prone to delamination and fracture during service when exposed to external low-velocity impact. One hindrance to overcome before a broader deployment of sandwich composites is the issue of impact energy assessment (IEA). To promote the solution to this issue, an ensemble deep learning approach is proposed in this study. The approach comprises data expansion, series-to-image conversion, and convolutional neural networks (CNN). The data expansion is implemented using vertical average interpolation. The enhanced data are transformed into images via the Gramian angular summation field to build an image dataset for the CNN model. To validate the developed ensemble approach, hammer-dropping impact experiments on the honeycomb sandwich composites are carried out based on the piezoelectric wafer active sensor network and electromechanical impedance measurement. Accuracy, precision, recall, and F1-score indicators are introduced to evaluate the ensemble approach performance. The above indicator values are all above 0.9600, demonstrating the effectiveness of the proposed ensemble approach in settling the IEA issue.