A deep learning method for damage prognostics of fiber-reinforced composite laminates using acoustic emission

Abstract

Damage prognostics of fiber-reinforced composites using advanced nondestructive test techniques is of great significance due to their complicated damage mechanisms. This paper employs the acoustic emission (AE) technique to monitor the performance degradation process and to estimate the residual load-bearing abilities of glass fiber/epoxy composite laminates with the damage evolution of various failure modes. Based on the prior knowledge of AE signals, a prognostic model by combining the feature evaluation algorithms and deep learning methods is developed. First, the model conducts the feature evaluation on twenty-four AE features and filters out the degradation-insensitive features from the multiple perspectives of different AE sensors. Second, a convolutional neural network model is built and trained on five informative features for the degradation estimation. The estimation accuracy is validated to be generally high that depends on the degradation stage. Third, the effect of the number of AE signals in an input sequence on the estimation is further investigated. Results show that such a prognostic model provides a feasible path to quantify the degradation process and damage tolerance of composite materials.