Applications are sought for this EPSRC Industrial CASE studentship at Aston University offered in collaboration with Rolls-Royce plc. The project will form part of the research portfolio of the Knowledge Transfer Network for Industrial Mathematics.

Industrial CASE studentship in Reliability Assessment Using Bayesian Criticality Analysis

The aim of the project is to provide new tools for design-risk analysis of complex systems allowing early design decisions to be informed by their reliability impact (and in certain cases, by implication, their safety impact). Additionally, maintenance and fault diagnosis could be based on a deep understanding of the system components and their interactions. Bayesian belief networks (BBNs) will be used to represent entire engineering systems in a probabilistic way (latest generation gas turbines will be used for proof-of-concept) in order to analyse the interactions at both module and component level and perform a design-risk analysis. We shall investigate how uncertainty in probabilities can be represented, inferred with, and used to help designers assess the impact of component and design choices. The techniques will be validated using historical information (including service histories) from existing aeroengines. A sucessful project outcome could be transferred into other safety-related domains, such as marine, civil, and automotive engineering, manufacturing and process industries, and health care. Our intention is to create a joint-venture spin-out company (involving both Rolls-Royce and Aston University) to exploit these opportunities. The project will involve a close collaboration between Rolls-Royce (providing the engineering and design expertise) and Aston University (providing the mathematical and computational research base). The main goal of the project is to develop Bayesian belief networks (BBNs) at the design stage as a risk assessment tool (thus enhancing current industrial practice) and to allow design engineers to make informed decisions over component choice etc. This project will deliver a prototype BBN-based tool that is robust enough to carry out system modelling, validated on a Rolls-Royce civil aerospace product. The innovative aspects of the project lie in the following areas: Large-scale (whole-engine) modelling with Bayesian belief networks has not been carried out successfully before, predominantly since the standard method of inference in BBNs (i.e. the junction tree algorithm) does not scale well to large dense networks. We shall develop novel modular inference algorithms based on recent methods used in other domains (such as multi-user communication) that were developed at Aston. These methods are based on techniques drawn from the field of statistical physics and are highly accurate approximations that are computationally efficient enough on dense graphs for practical use on systems of 2000 nodes. The novelty of our approach will be to combine conventional inference algorithms (such as the junction tree) with novel algorithms derived from the statistical physics framework that takes advantage of the modular structure of the belief networks we shall be studying. Integrating design, risk assessment, and maintenance into a single probabilistic model. The current industry standard (Failure Mode, Effect and Criticality Analysis (FMECA)) for risk assessment does not generally use probabilistic tools for analysis of real large-scale systems. Modelling component interactions. FMECA considers components as independent because there is no appropriate modelling and inference tool to deal with interactions. Past safety analysis in this domain has generally considered attributes either to be independent or for which the dependencies correspond to a tree structure. Modelling uncertainties in probabilities. It is important not only to represent and infer with uncertainty (whether derived by statistical analysis of historical data or by an expert) but also to represent the results in an appropriate form for the design engineer. Providing usable measures of uncertainty in the model predictions of failure probabilities will require the development of new techniques (probably based on Markov Chain Monte Carlo sampling) of a mathematical nature. Presenting this information in an informative way will require advanced visualisation and other pattern analysis techniques. Industrial CASE studentships leading to a PhD are fully funded for three and a half years, including home fees and a maintenance grant of £15,500 per annum (average). For informal enquiries, please contact Prof. Ian Nabney. To obtain an application form please contact: Vicky Bond NCRG Aston University Aston Triangle Birmingham B4 7ET e-mail: v.j.bond@aston.ac.uk Tel: 0121 204 3652

These photographs are reproduced with the permission of Rolls-Royce plc, copyright © Rolls-Royce plc 2006