Scientific Article

Improving defect detection of nuclear welds using adaptive imaging based on uncertain material parameters optimization

Université Paris-Saclay, CEA, List, F-91120 Palaiseau, France

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Received: 31 March 2026
Accepted: 29 June 2026

Abstract

Weld inspection in the nuclear sector is crucial for ensuring the structural integrity and safety of reactor cooling piping systems. Given the strict safety requirements, reliable and accurate Non-Destructive Evaluation (NDE) techniques are essential for the early detection and characterisation of defects. In this context, phased-array ultrasonic imaging has emerged as a powerful tool for advanced NDE. Among these techniques, the Total Focusing Method (TFM) with Full Matrix Capture (FMC) is widely recognised for its superior signal-to-noise ratio and imaging performance. Weld inspection remains challenging due to the anisotropic and heterogeneous nature of weld microstructures, which distort wavefronts, increase structural noise, and are often only imprecisely characterised. This work addresses the latter issue, as TFM images become distorted when the Time of Flight (ToF) used for reconstruction deviates from the physical ToF. Correcting the ToF is therefore necessary to compensate for these aberrations. This study enhances an existing adaptive imaging approach based on TFM, the use of a complex weld model, and the optimisation of an imaging criterion within the space of the weld model parameters. Unlike the initial version, which required prior knowledge of defect locations, the enhanced framework automatically identifies subzones with a high likelihood of containing defects. An improved global normalised criterion is then applied to enhance image quality. A more robust hybrid optimiser, combining global exploration and local refinement, is employed, improving both stability and computational efficiency. The approach was validated using experimental and simulated FMC datasets, demonstrating improved defect detectability and image quality, and was implemented as a CIVA plugin.