Speaker
Description
Active infrared thermography (IRT) is a well-established non-destructive testing technique for detecting subsurface defects in materials by analyzing thermal responses to external excitation. However, defect detectability is often limited by noise, spatial non-uniformities, and the complex nature of heat diffusion in heterogeneous materials. Recent research has demonstrated that combining multiple post-processing methods or excitation sequences can significantly enhance the signal-to-noise ratio (SNR) and robustness of defect detection. Building on these advances, this work introduces an adaptive multi-domain fusion framework that integrates information from multiple temporal and frequency-domain representations to improve defect visibility under variable inspection conditions.
The proposed approach extends previous efforts on multi-sequence fusion and robust SNR modeling by incorporating both statistical and structural features of the thermographic response. Instead of relying on a single processing technique, the framework adaptively combines complementary results from several post-processing algorithms—such as Pulsed Phase Thermography (PPT), Principal Component Thermography (PCT), and Thermographic Signal Reconstruction (TSR)—along with different excitation sequences or acquisition settings. Each input contributes differently to defect highlighting depending on defect depth, material properties, and noise characteristics. By using a data-driven weighting strategy based on local SNR estimation and inter-method correlation, the system optimizes the contribution of each modality to produce a final defect-enhanced map.
A key contribution of this work is the integration of a robust SNR evaluation model that considers both the thermal signal evolution and the spatial context. This model provides a quantitative measure of the detectability improvement achieved through fusion, enabling objective comparison across experiments. Moreover, the method accounts for variations in emissivity, surface reflections, and environmental noise, which often degrade the reliability of thermographic measurements. By analyzing local signal fluctuations and temporal coherence, the algorithm effectively discriminates true defect-induced contrasts from noise-related artifacts.
Experimental validation was conducted on composite and metallic samples with artificial and natural defects using pulsed and modulated excitation. The results demonstrate that the proposed adaptive fusion strategy significantly increases SNR and contrast-to-noise ratio (CNR) compared to individual processing methods. Particularly for deep or low-contrast defects, the fused results reveal enhanced defect boundaries and improved consistency across repeated measurements. Quantitative analysis confirms up to 40% improvement in detectability metrics, highlighting the effectiveness of the multi-domain integration.
The proposed framework offers a general and flexible methodology for thermographic data enhancement, independent of the excitation scheme or sensor configuration. It can be easily adapted to different inspection scenarios, including lock-in, step-heating, and pulsed thermography. Future work will focus on extending the method with deep-learning-based fusion strategies and uncertainty quantification to further automate defect characterization and improve interpretability.
