Overview
This project will deliver computational models for use in radiological incidents, both accidental and malicious, utilising artificial intelligence/machine learning algorithms to efficiently analyse large data sets and aid in the production of radiation protection advice.
About this opportunity
Radiation exposure during routine maintenance and in the event of radiological incident poses significant health risks to personal and responders, workers and the public and creates substantial issues in the wider environment. Current radiation protection methodology involves Radiation Protection Advisors (RPA’s) and other health physics specialists utilising well-established guidelines and radiation protection frameworks (such as ICRP publications, which feed into national radiation protection legislation) to generate advice on how to optimise dose whilst balancing emergency response and life-saving action. Production of this advice involves complex analysis of nuclide decay mechanisms, radiation exposure and transport pathways of material. Environmental specialists will further analyse the dispersion of a nuclide through waterways and the air, to assess the impact of radiological release to nature as well as the potential for further exposure to the public via migration of a source or contaminant. As the next innovative generation of nuclear reactors is constructed and the risk of radiological terrorism ever present, real-world exposure scenarios are often increasingly complex and rapidly changing. Monitoring of such situations can be intricate, and application of guidance nuanced.
With the advent of artificial intelligence and machine learning models, the ease with which large data sets can be analysed has greatly increased. Artificial intelligence models are capable of interpreting data sets and generating situation specific analysis and advice, with a speed and adaptability far greater than human counterparts. The application of this technology to radiological incident planning and response would allow for far more comprehensive and tailored advice generation, which has the capability of drawing from predictive models as well as real world data from previous incidents and applying relevant international safety standards. The efficiency of this pathway of data collection, through to processing and eventually the generation of radiation protection advice, combined with per calculated simulaitons, could be greatly improved using these developments.
This project aims to create and train an artificial intelligence model, intended for use in the event of malicious radiological incidents (e.g. the aftereffects of a radiological dispersal device) as well as civil nuclear incidents (such as those encountered during transport of a novel reactor, or operation of a mobile reactor e.g. a floating reactor or molten salt reactor). This model will echo the current best practice for radiation protection advice generation, but with the added ability of great computational power to utilise data from previous real-world scenarios and real-time data from detector networks, as well as elements of existing predictive computational models such as VRDose and PACE. The model could then draw from relevant legislation and frameworks to predict an optimal dose reduction solution, which RPAs could utilise to advise accordingly. Monte Carlo simulation and machine learning will form the bulk of the operation, allowing the model to expand its ‘knowledge’ and improve clarity and precision of recommendations over time as more data is provided. The accuracy, reliability and ethics of the model will also be considered throughout.
This project will draw on resources from the University of Liverpool’s collaboration with IBM Research and STFC Daresbury. Research will be undertaken on a part time, distance learning basis.
Further reading
Artificial Intelligence (AI); A Revolution in Radiation Protection in Modern Life (Refahi, Shahi, Davaridolatabadi) JBPE-14-209.pdf
Artificial Intelligence in Nuclear Safety and Radiation Protection (Romera, Nunez, Herrera, Gonzalez) 9788490525609.pdf
VRDose VRdose – IFE
PACE PACE Probabilistic Accident Consequence Evaluation Software – Home