Jonathan Sadeghi

Slider 1



I gained an MPhys in Physics with Theoretical Physics at the University of Manchester, before joining the Next Generation Nuclear Center for Doctoral Training to study for a PhD in Engineering. After one year taught course in Nuclear Science and Engineering I started my PhD research, which intersects the disciplines of Machine Learning. Nuclear Engineering, Software Engineering and Uncertainty Quantification.

Research Interests: Machine learning, Bad data, Imprecise probability, Safety Analysis

Research project title: Robust Probabilistic Risk/Safety Analysis for dealing with Scarce and Limited Data

Project description: The safety of nuclear systems and plants relies on the reliability and availability of different safety systems. Under realistic conditions, these systems and components are affected by uncertainties, caused by lack of sufficient knowledge and/or by natural unpredictable external events. Uncertainty analysis is an essential tool to obtain a robust representation of model predictions consistent with the state-of knowledge. Generally speaking, there are two kinds of uncertainty involved in nuclear safety evaluation, which are aleatory uncertainty and epistemic uncertainty. For instance, the calculation uncertainty of best estimate event analysis must be quantified by considering both model uncertainty and plant status uncertainty, and generally, a conservative surrogate sequence is assumed for a particular event safety evaluation. For example, in the large break loss of coolant accident analysis, the loss of off-site power and single failure are conservatively assumed in the evaluation surrogate sequence. In fact, owing to the random failure of mitigation systems or components, there are many possible event sequences which exist with different occurrence probabilities.

Probabilistic Safety Analysis has a vital role to play in safety analysis of complex nuclear plants. These techniques can be used to provide a quantitative analysis of the identified risk. However, such analyses depend on the assumptions and the failure distributions for each component. Often not enough data is available to define such distributions and some expert judgement is necessary.

The overall aim of the project is to develop a robust methodology for probabilistic safety (risk) assessment. It involves the development of an efficient general purpose computational tool and strategies able to deal with scarce, vague and imprecise information and a methodology that will provide a robust support tool for risk-informed decision making.

Supervisory team: Edoardo Patelli; Marco de Angelis

ORCID: 0000-0003-4106-2374