This project will primarily be based at The University of Liverpool.
The hypothesis underpinning the research project is that luid flows can be represented, or characterised, by feature vectors using decomposition and that this will allow quantitative comparison of predicted and measured flows for the purposes of validation and updating of computational fluid dynamics models.
Thus, the aim of the project is to generalise a validation metric based on relative errors to allow integration with orthogonal decomposition and, or dynamic mode decomposition of predicted and measured data, thereby unifying quantitative validation approaches in solid and fluid mechanics, for the first time. Progress towards this aim will be made via the following objectives: (i) to acquire and evaluate suitable sets of measured and predicted data from fluid flows using the literature, on-going research and performing experiments and simulations; (ii) to establish algorithms for the application orthogonal decomposition and dynamic mode decomposition to fluid flows; (iii) to demonstrate the application of the algorithms for decomposition of fluid flow data from both measurements and predictions; (iii) to develop a quantitative validation method based using a probabilistic assessment of the congruence of the measurements and predictions; (iv) to apply the algorithm and method to a number of case studies relevant to the nuclear industry and consider their use in support of model updating.
Establishing credibility in engineering simulations through a process of model validation is an essential part of any engineering analysis [Patterson, 2015]. It is well-established that validation of computational models should include a comparison of predictions from the simulation with measurements from a physical experiment or prototype that closely resembles the conditions in service, i.e. real-world, everyday circumstances. Advances in digital sensor technology allow information-rich data fields to be acquire in real-time leading to large quantities of measured data, sometimes referred to as ‘big data’, which presents challenges in making quantitative and meaningful comparisons with predictions from simulations. In solid mechanics, Patterson and his co-workers have used orthogonal decomposition to reduce the dimensionality of both measured and predicted data to feature vectors that can readily be compared [Dvurecenska et al, 2018]. This process is becoming routine for two-dimensional fields of data and is being developed for volumes of data. In both cases, feature vectors representing measurements and predictions can be compared to establish the probability that they belong to the same population. In this project, these techniques will be extended to computational fluid dynamics in collaboration with Professor Rob Poole. It is expected that the outcome will be a significant advance in the rigour with which simulations based on computational fluid dynamics models can be validated through detailed comparison to measurements and that this will enable updating of simulations to increase their fidelity.
The project is sponsored by EPSRC through the GREEN CDT and the National Nuclear Laboratory. Dr Steve Graham from NNL will be the industrial supervisor and will provide guidance on the industrial usage of the technology. There will cohort-based training in year 1 provided by the CDT that will include taught modules on nuclear science and engineering and specialised training tailored to this PhD project which might include a placement at NNL.
More detail about the ways in which we positively promote equality and diversity may be found in our Diversity and Equality of Opportunity Policy.