Developing a cost-effective multispectral imaging system for real-time nuclear fuel pellet inspection

The production of sintered uranium dioxide (UO2) nuclear fuel pellets is a crucial process in the nuclear power industry, where ensuring the integrity and quality of the pellets is paramount. Quality control, traditionally reliant on manual and timeconsuming visual inspection, can benefit significantly from the application of advanced imaging technologies. Hyperspectral imaging (HSI) offers a non-destructive option to enhance inspection accuracy, but its practicality in industry is often constrained by high cost and operational complexities. In contrast, multispectral imaging (MSI), which typically captures between 3 and 20 channels compared to the hundreds of bands used in HSI, can be performed rapidly and at lower cost. The spectral limitations of MSI can be mitigated by using findings from hyperspectral analysis to inform the choice of which wavelengths to select, resulting in a cost-effective means to make the manufacturing process more responsive. HSI has been successfully used to characterise UO2 pellets, identifying useful spectral regions in the short-wave infrared (SWIR) range. In particular, prominent absorption bands at around 1100 nm and 2300 nm, as well as a series of smaller spectral features between 1300 nm and 1750 nm, have been identified as characteristic of UO2 and may be used to distinguish potential contaminants and impurities that adversely affect the chemical composition of the pellets. Based on this knowledge, we develop a multispectral camera that utilises an InGaAs sensor and a 10-filter wheel, with filters selected to target the spectral features identified in the 950 nm to 1700 nm range. The filter selection is initially validated through the generation of simulated multispectral data from existing hyperspectral imagery, ensuring that the identified wavelengths capture relevant features.
The primary motivation for choosing a multispectral camera over a hyperspectral one lies in the comparatively high cost of hyperspectral cameras, which, combined with the risk of contamination or damage in radioactive environments, makes MSI a more practical choice for widespread use. Additionally, the use of a multispectral camera allows for faster data acquisition without sacrificing the spatial resolution needed to detect physical defects such as chips, cracks, and irregularities in the geometry of the fuel pellets on a high-throughput production line.
The developed multispectral camera is designed to be a practical tool for responsive manufacturing, a process in which near real-time adjustments can be made during production to ensure the quality and consistency of the final product. In the context of nuclear fuel pellet manufacturing, the capability to respond to issues affecting fuel batches early could lead to significant improvements in production efficiency, while also reducing waste. The multispectral system can be integrated into the production line, offering continuous monitoring and quality control with minimal disruption. To enable continuous imaging of moving pellets, channels must be precisely co-registered using a least-squares-based algorithm, ensuring accurate alignment across wavelengths. Spectral features are obtained from the relative values of channels around key absorption bands, as individual reflectance values can vary depending on illumination conditions. One-point calibration is used to compensate for the differing sensor responses and filter bandwidths of the channels.
Future work will involve field-testing and validating the system’s performance in a production environment, as well as exploring potential enhancements to further improve detection capabilities for both chemical and physical defects.