A waveform-based clustering and machine learning method for damage mode identification in CFRP laminates

Abstract

To gain an insight into the damage mechanism in carbon fiber reinforced polymer, a real-time analytical approach for damage mode identification of composite based on machine learning and acoustic emission is proposed. Firstly, waveform features are extracted from the acoustic emission signals with low information entropy through wavelet packet transform, where the high-dimensional feature vectors represent the main features of the reconstructed signals in the frequency domain. Combined with the autoencoder and k-means ++ algorithm, a waveform-based clustering model is constructed to reveal the relevance between acoustic emission signals and damage modes. Finally, the damage mode recognition of different types of composite laminates is achieved by the developed softmax layer classifier. The identification and the quantitative analysis of damage modes for prefabricated defects specimens demonstrate the robustness of the method. The method is effective and feasible for real-time monitoring of the damage evolution process of carbon fiber reinforced composite components.