Pushing the boundaries of green composites: a novel robust inspection system for damage identification and classification in NFRPs

Natural fiber composites have gained attention as sustainable alternatives to their synthetic counterparts due to their biodegradability and renewable origins. However, their heterogeneous properties often lead to higher failure rates, making reliable quality assessment crucial. Non-destructive evaluation (NDE) therefore plays a key role in assessing these materials. While significant progress has been made in applying machine learning to NDE, further research is needed to assess its effectiveness with natural fiber-reinforced polymers (NFRPs). To address this gap, this study investigates the use of machine learning, particularly convolutional neural networks (CNNs), to improve the defect detection process in NFRPs. For this purpose, flax/epoxy composite laminates subject to diverse damage scenarios were manufactured and scanned using phased array ultrasonic testing (PAUT). Three distinct datasets were created: the raw data, the raw data processed using the Hilbert transform, and reconstructed images derived from the raw data using Principal Component Analysis (PCA). These datasets were used to fine-tune separate pre-trained ResNet50 models to evaluate and compare their performance in classifying the images and distinguishing between those containing visible defects and those without. Experimental results showed that the proposed imaging system can accurately detect and classify a significant range of material defects in NFRPs of diverse dimensions, size and in-depth location through the laminate. Furthermore, the results highlight also the potential of CNN-based methods in automating and enhancing defect detection in NFRPs, offering a pathway to more reliable and efficient inspection of these materials.