Surfing Virtual Waves to Thermal Tomography: From model- to deep learning-based reconstructions

Thermographic imaging is a fast and contactless way of inspecting material parts. Usually, with model-driven evaluation procedures, lateral heat flow is ignored, and, thus, 1D reconstruction is applied to detect defects. However, to correctly size defects, the lateral heat flow must be considered, which requires a full 3D reconstruction. The 3D thermal defect imaging is a major challenge because heat propagation is an irreversible process. The virtual wave concept (VWC) is a recently developed method that considers both lateral and axial heat flows and, therefore, allows multidimensional reconstruction at improved spatial resolution. This approach decomposes the problem into two steps; can be used for 1D, 2D, and 3D heat conduction problems; and provides new alternatives to using physical priors (e.g., nonnegativity and/or sparsity), all of which improve reconstruction accuracy at a relatively low computational cost. We present two applications of the VWC in nondestructive material testing using thermal tomography in 2D and 3D. We demonstrate the performance of this approach in a comparative study including physics-based forward modeling, regularization, and artificial intelligence methods as well as introduce state-of-the-art hybrid approaches that combine model- and deep learning-based reconstructions.