Home > Case Studies > Using Machine Learning to Make Deuterium Uptake Inspection Scope Recommendations

Using Machine Learning to Make Deuterium Uptake Inspection Scope Recommendations

Image of a digital head with icons in the background
Location: Toronto, ON, Canada Year: 2019-2020

Project Objective

The amount of Deuterium in CANDU pressure tubes, which carries the nuclear fuel in the reactor vessel, affects the service life of the pressure tubes. During operation, the Deuterium concentration increases as Deuterium diffuses from the coolant into the pressure tube. The empirical models built to forecast the Deuterium concentration are constantly updated using the data obtained from periodic in-service inspections. Since the cost to perform these inspections are very high, the project objective was to provide recommendations for inspection locations in an upcoming campaign where it would have the highest impact on the evolution of the empirical models. 

Scope of Work

The overall strategy was to evaluate how sets of hypothetical inspection results from different candidate inspection locations will affect the Kinectrics’ in-house subject matter expert’s (SME) opinion on the Deuterium uptake that is expressed through the empirical models. Normally, the SME may take days to manually evaluate the effect one set of inspection results will have on the empirical model. Given there are many combinations of inspection locations, it may take weeks of evaluations to obtain the scope recommendations. To facilitate this analysis, a novel Kinectrics proprietary machine learning algorithm was developed to mimic the SME’s empirical modelling preferences, as well as incorporate the state-of-the-art of Deuterium uptake, and applied to automate the numerous evaluations.


The algorithm that was developed reduced the evaluation time by about 85% and produced inspection recommendations that the SME and the client agreed with. This demonstrated how the Kinectrics proprietary machine learning algorithm:

    1.Automated a highly manual evaluation process, 
    2.Captured the Kinectrics’ in-house SME’s expertise in empirical modelling, 
    3.Incorporated the state-of-the-art understanding into the evaluation and ultimately the recommendations.
This application assisted in overcoming the initial hesitancy in adopting machine learning techniques. The client requested that the novel algorithm be used to produce inspection recommendations for an upcoming Deuterium concentration inspection campaign and assist in the development of the empirical model that will be incorporated in safety assessments. ​