New Optical Imaging and Analysis Techniques for the Early Detection and Quantification of Corrosion on Special Nuclear Materials Packages

This PhD research will focus on designing novel signal and image processing techniques using a combination of state-of-the art Deep Neural Networks and Traditional methods to segment, analyse and quantify corrosion products on materials used in special nuclear materials packages including stainless steel and aluminium extrusions. The work will be conducted in collaboration with University of Manchester (funded by Sellafield and FIND CDT) who will corrode a series of stainless steel and aluminium coupons over time with chlorides of interest for which the chloride index will be known in each experiment.

Corrosion coupons will be imaged at regular intervals throughout their lifetime using a range of optical sensor technologies including Stereomicroscopy, Optical Microscopy, Laser Confocal Microscopy, LIBS and RAMAN (University of Manchester) through to RGB and Hyperspectral Imaging (University of Strathclyde). Novel algorithms will be designed at the University of Strathclyde (UoS) to analyse images of all modalities which will result in the creation of a set of multiscale image analysis techniques to detect and quantify corrosion from the microscopic to macroscopic scale using spatial and spectral information.

The acquisition of the described dataset, combined with the proposed analysis routines, will be a first-of-their-kind suite of tools to detect - with confidence - the earliest point at which corrosion products can be robustly identified in all imaging technologies. It is anticipated that the detection ability and sensitivity of the sensors will be enhanced by the design of our proposed image analysis routines which will also fuse data from an optimal set of sensors to be identified in this study.

Fundamentally, this research will be made possible through the analysis of microscopic imaging (to provide reliable ground truth of the earliest presence of corrosion) and spectroscopic analysis techniques combined with full knowledge of all chlorides used to induce corrosion and sophisticated monitoring and analysis of the entire corrosion lifecycle on a range of metals of interest.

From an academic perspective, the proposed research aims to make the following key novel contributions:

-Introduction of new techniques for processing and analyzing corrosion products at the macroscopic scale for inspection purposes using data from a diverse set of advanced optical sensor technology.

-Design, implementation and testing of new corrosion detection tools by advancing state-of-the-art Machine Learning techniques for data fusion and classification of a range of complementary, multimodal, optical sensors.

-Creation of novel corrosion chronology analysis algorithms capable of modelling the progression of corrosion over time. Once developed using data gathered for training, this will be applied to unseen inspection footage and used to predict corrosion progression and help identify any necessary intervention to prevent adverse outcomes.

-Robust validation of the proposed array of optical sensor technology when combined with our proposed techniques using expert microscopy and materials science (UoM).

-Creation of a unique, information-rich dataset gathered using a suite of state-of-the-art optical sensor technology. To the best of the authors' knowledge, this will be the first dataset of its kind to allow the time-lapse analysis of corrosion product lifecycle from the microscopic to macroscopic scale alongside the underlying chemical properties of the materials understudy.