Effective image and data analytics approaches for sub-surface defect detection and identification in composite materials from pulse thermography scans

The early detection of internal defects in wind turbine blades made of composite materials is crucial to prevent potential damage that could compromise their structural integrity and performance. Even minor defects, like porosity or fiber misalignment, can significantly affect the blade’s durability and strength, leading to costly maintenance or replacement. This study examines four common types of sub-surface defects in composite material specimens, such as defective fiber misalignment, inclusion, fiber breakage, and porosity. We utilize a series of thermal images obtained through long-pulse active thermography of samples with induced sub-surface defects. Our approach analyzes the thermal image sequence to capture temperature changes during heat pulse exposure. In this work we present a segmentation method that effectively identifies regions corresponding to sub-surface defects, based on k-means clustering and watershed transform. Furthermore, our approach utilizes principal components projections for the visual identification of sub-surface defects. This valuable data is then fed into a set of machine learning classifiers, enabling their automatic classification. Remarkably, the na¨ıve Bayes classifier excelled in accurately identifying the four distinct types of sub-surface defects in the specimens. By integrating the proposed segmentation and classification methods, our study enables the comprehensive analysis of sub-surface defects in composite materials, providing significant advancements for quality control in the manufacturing of wind turbine blades. Additionally, our study demonstrates the reliability of the long-pulse active thermography data for identifying and classifying subsurface defects in wind turbine blades.