Speaker
Description
The structural integrity of aeroengine blades is paramount to flight safety, yet the inspection of these critical components remains a significant challenge due to their complex geometries and exposure to extreme operating environments. While Visual Inspection (VI) serves as the standard industrial baseline for identifying surface anomalies, it is inherently limited in detecting subsurface defects such as micro-cracks, delaminations, or fatigue that do not manifest externally. Infrared Thermography (IRT) offers a complementary solution by revealing subsurface thermal abnormalities. This study proposes an automated, bimodal deep learning framework to bridge this gap, specifically investigating the quantitative performance gains achieved by fusing RGB visual data with Active Infrared Thermography. The research focuses on a comparative analysis between unimodal (vision-only) and dual-modal (Vision + IRT) defect detection models applied to a dataset of paired, defective aeroengine blades. The proposed methodology utilizes a Convolutional Neural Network (CNN) architecture for the automated classification of defects. To validate the efficacy of this approach, the model is trained and tested on a standardized dataset of blades containing induced defects common to high-stress aerospace components, including surface cracks and impact damage. The dual-modal dataset is collected in a synchronized sense, where visual and thermal images are paired and matched together. This will allow for the direct comparison of inspection feasibility between the two Non-Destructive Testing (NDT) domains. The performance is benchmarked against a state-of-the-art unimodal ResNet-50 baseline operating solely on visual data. Preliminary results hope to show that the bimodal framework achieves a measurable improvement in classification accuracy and sensitivity. Specifically, the inclusion of quantitative thermographic data significantly reduces false negatives for subtle, near-surface defects that are visually ambiguous, demonstrating that thermal signatures provide critical discriminative features that visual data alone cannot capture. This work demonstrates that integrating infrared thermography with deep learning-based visual inspection creates a robust, synergistic diagnostic tool. By moving beyond unimodal reliance, the proposed bimodal framework offers a pathway toward more reliable, automated, and non-destructive evaluation workflows for the aerospace maintenance sector. The study highlights the indispensable role of thermal imaging not just as a standalone tool, but as a vital component of multi-sensor AI systems.
