Virtual-to-real mapping: an AE missing signal generation and assisted localization method for defects in CFRP plates

Monitoring the structural health of carbon fiber reinforced polymer (CFRP) is crucial, with acoustic emission (AE) often used for defect localization. However, locating undetected signals from defects remains a significant challenge, especially when balancing cost and accuracy due to CFRP's anisotropic propagation properties. This paper presents an innovative method named virtual-to-real mapping for generating undetected signals using only two AE sensors, which realizes a cost-effective solution for data collection. First, simulated signals are generated based on velocities obtained from two distinct fiber arrangements using two AE sensors. Then, one AE sensor is used to collect the experimental time-domain waveforms. An Auto-encoder maps the experimental and simulated signals to predict missing waveforms, which are validated through both quantitative and qualitative analyses. Incorporating these generated signals into a multilayer perceptron model results in an average improvement of 21.1% in multiple missing region classification accuracy and an average of 43.9% in independent missing region classification accuracy, providing valuable insights into data-driven methods for predicting and localizing missing defects in CFRP structures. Furthermore, the proposed method is validated not only on 6061 aluminum alloy plates, demonstrating its potential applicability across different material systems, but also in the context of different sensor placements, confirming that sensor positioning has limited influence on localization performance.