Using convolution neural network methods for the ultrasound characterization of porosity across carbon fiber reinforced polymer layers

This study investigates the use of Convolutional Neural Network (CNN) with ultrasound imaging for the characterization of porosity across Carbon Fiber Reinforced Polymer (CFRP) layers using both simulated and experimental dataset. CFRPs are widely used in aerospace and other engineering fields due to their exceptional mechanical properties. However, porosity remains a critical defect that can significantly impair their performance. Traditional non-destructive testing (NDT) methods face some challenges in accurately detecting and characterizing porosity. The present work aims to overcome these challenges by developing a CNN-based approach to improve the detection and assessment of porosity across CFRP layers. The study relies on the development of a numerical model and the acquisition of real data from fabricated CFRP samples to successfully apply CNN techniques to evaluate porosity. The CNN model demonstrated fairly good accuracy and reliability, particularly with an increased number of dataset. The results suggest valuable opportunities for improving quality control in CFRP manufacturing processes. The study presents the potential of applying machine learning techniques for the non-destructive testing of CFRP, with a relative good amount of datasets. The present work contributes to the larger project of enhancing the reliability of CFRP structures and improving the composite materials' manufacturing processes.