Convolutional LSTM Networks for Vibration-Based Defect Identification of the Composite Structure

Abstract

The increasing need for reliable and sensitive detection methods have been brought about by the design of lighter, more flexible or complicated composite structures. In the field of vibration-based testing, many studies about the assessment of composite structures have been proposed. Artificial neural networks(ANNs) have offered a novel route to explain the relationship between vibration feature and defect status, and an increasing number of ANN frameworks have been designed for adhesive defect identification in the composite structure. This paper proposed a novel ANN framework based on Convolutional Long Short-Term Memory (ConvLSTM) in order to identify composite structure with adhesive defect. The proposed framework allowed pattern classification of the defect accordingly. To verify the feasibility of the proposed approach, two specimens with different levels of debonding were tested. The obtained results indicate that the classification accuracy strongly depended on the proper adaption of ConvLSTM