Open Access

Compared with conventional optical excitation thermography techniques, such as pulsed thermography (PT), lock-in thermography (LIT), step-pulse thermography (SPT) and linear-chirp thermography, chirped-pulse thermal-wave radar thermography has been demonstrated to provide a higher signal-to-noise ratio (SNR) and more effective depth-profiling capability. This enhanced performance originates from the use of short chirped pulses, which maintain a nearly flat power spectrum over a broad bandwidth, even in the presence of diffusive attenuation. However, conventional laser-based signal-modulated heating schemes require both spatial and temporal modulation – typically generating a rectangular heating area via projection or lensing, followed by digital waveform modulation. This leads to localized heating areas and prolonged heating durations, which significantly limit their applicability in practical industrial inspections. In addition, industrial samples, such as cast components, often exhibit large-scale and complex geometries. To perform infrared thermography on such samples, they must typically be divided into multiple sub-areas for sequential heating and recording, introducing pronounced boundary effects and complex data concatenation issue. In this work, we propose a spatially modulated heating strategy combined with a robotic arm moving at a constant speed. In details, the heating source consists of multiple linear heating strips arranged with different spatial intervals, while the heat source and infrared camera remain fixed. The robotic arm moves the sample at a constant speed across the heating and recording area. Using a dynamic-to-static reconstruction algorithm, the chirped-pulse thermal-wave radar signal can be generated for each pixel. This approach enables continuous chirp-pulse thermal-wave radar thermography by replacing temporal modulation with spatial modulation, thereby not only simplifying the system and reducing overall cost but also allowing for uniform heating of large and complex samples. Experiments and simulations are conducted to optimize key parameter, and various image processing algorithms - including principal component thermography (PCT), pulsed phase thermography (PPT), and partial least square regression (PLSR) - are applied to further enhance detectability. Image quality is quantitatively evaluated using signal-to-noise ratio metrics. The results demonstrate that this method holds strong potential for real-world industrial inspection applications. Moreover, the strategy of replacing temporal modulation with spatial modulation is expected to be extended to other emerging applications, such as random-coded and adaptive thermography, broadening its versatility in non-destructive testing.

Pseudo-noise–based pulse-compression thermography has become an increasingly attractive alternative to conventional pulsed and lock-in approaches, as coded excitations enable high deposited energy while retaining broadband frequency content. However, practical implementation requires careful consideration of the inherent DC component introduced by the unipolar nature of the applied heating. While explicit DC-removal procedures are typically required, this study shows that Legendre sequences can be modified such that they provide intrinsic DC-offset auto-compensation during correlation-based reconstruction. The natural offset cancellation achieved during matched filtering removes the need for polynomial DC-fitting or measurements near thermal steady-state conditions, thereby simplifying the processing chain while maintaining the full benefits of pulse-compression. These findings highlight a practically relevant advantage of pseudo-noise thermography: the ability to simultaneously access high SNR, long-duration coded heating, and robust DC-offset suppression of the thermal excitation without additional modelling steps.

Heat propagation, governed by phonon interactions, is described by partial differential equations (PDEs) linking thermal transport to material properties. Conventional thermography relies on surface emissions, limiting subsurface resolution, while existing tomographic methods capture only single-layer frames. Physics-informed neural networks (PINNs) integrate data with physics but mainly fit external temperatures, lacking direct access to internal fields. Here we propose a Helmholtz-informed neural network (HINN) to predict internal temperature distributions without interior measurements. By converting the time-domain diffusion equation into a pseudo-Helmholtz form, HINN incorporates both real and imaginary components of the thermal field, followed by inverse Fourier transform to reconstruct 3D thermal maps with defects. A truncated operation and conjugate symmetry repair further improve efficiency and accuracy. Results show HINN surpasses PINNs and inverse solvers, enabling non-invasive thermography for materials, biomedicine, and nondestructive evaluation.

Accurate crack detection and segmentation in complex components is critical for assessing the structural health and integrity of systems exposed to extreme thermal and mechanical loading. While pixel-wise convolutional neural networks provide a strong benchmark, they often require dense annotation and local information of defect geometry. This study presents a superpixel graph attention network (GAT) for crack segmentation based on crack classification and regression using infrared (IR) and visual (RGB) images acquired using automated laser thermographic inspection. Each image is segmented into superpixels to form region-adjacency graphs (RAG), which consist of nodes and edges that represent local features and spatial relationships, respectively. Node features include intensity, geometric and positional data from IR and RGB images. The GAT network consists of two heads with focal cross-entropy for crack classification and mean squared error for crack intensity regression, which provides robust learning for the dataset with class imbalance. The integration of IR and RGB images acquired from robot-assisted laser-line thermography improved the detection of fine cracks and reduced false positives, which indicated cooling hole edges as cracks and heat maps that correlated with crack intensity. Compared to the conventional U-Net segmentation model, the proposed approach provides generalized crack segmentation results with meaningful graph representations for explainable defect characterization.

The GUWEM (Guided Ultrasonic Waves: Emerging Methods) Workshop 2026 is held at Mont Sainte-Odile, France, from 29 June to 2 July 2026, organized by the Institut Langevin (ESPCI Paris | PSL, CNRS) and the Bundesanstalt für Materialforschung und -prüfung (BAM). The workshop brings together researchers from across Europe working on guided ultrasonic waves, covering the theory and application of elastic waveguides with topics ranging from semianalytical and numerical modeling methods to material characterization, damage detection, zero-group-velocity phenomena, laser ultrasonics, and industrial inspection applications including pipeline integrity and additive manufacturing.