Advanced detection and reconstruction of welding defects in irregular geometries using eddy current pulsed thermography

Abstract

The global pursuit of net-zero goals has accelerated the growth of solar energy, positioning photovoltaic (P.V.) systems at the forefront of renewable energy due to their efficient and sustainable electricity conversion. Detecting and classifying welding defects in P.V. systems—especially those with intricate surfaces and varying defect scales—is critical yet complex, as irregular geometries challenge traditional nondestructive testing (NDT) methods. To address this, we present an advanced framework integrating Graph Signal Processing (GSP) into Eddy Current Pulsed Thermography (ECPT). In this method, temporal thermographic sequences are mapped onto a graph topology, where nodes correspond to discrete time points and edges encode temporal dependencies through an adjacency matrix. The graph Laplacian operator, constructed based on the temporal adjacency relationships, is eigen-decomposed to project thermal response dynamics into the graph spectral domain. This transformation enables frequency-resolved analysis of time-evolving thermal waves, inherently isolating defect-induced transient signatures from steady-state thermal backgrounds. In our experimental study, Helmholtz coils generate a uniform current density combined with the novel framework, facilitating effective inspection of intricate surfaces by integrating 3D surface measurements with 3D thermography. Furthermore, we compare this method with other state-of-art algorithms. This multidimensional feature analysis framework robustly separates defect profiles from their backgrounds, addressing the unique challenges posed by the irregular geometries in P.V. systems.