Research areas
The Risk Consortium has expertise in a number of quantitative-based analysis techniques. Broadly its research is driven by its close relationship with industrial partners and as such, aims to address the problems facing current decision makers. Over the recent years, members of the Risk Consortium have focused on the following areas.
Reliability Growth Modelling
Modelling the growth in reliability performance can inform decisions taken by those managing the design, development and operation of systems. For example, growth modelling can provide a means of identifying and prioritising key drivers of reliability early in the design process and a mechanism for updating estimates of reliability as new information becomes available through analysis and test. Focusing upon the key reliability drivers allows managers to allocate resources to activities which will have most impact on reliability, while updating the growth model allows the effectiveness of activities to be assessed.
Lesley Walls and John Quigley have been involved in developing methods and models to support reliability growth management. Our role has been to support the theoretical development of methods and to work with engineers to investigate the impact of modelling on practice. We have developed a simple stochastic process model for growth that has been grounded in engineering practice where the underlying failure process is structured upon inherent failure modes and their potential realisations. Data classification taxonomies have been defined to support common input data structures. Methods have been developed for eliciting structured subjective judgement to support qualitative structuring of the model and quantitative instantiation. Techniques for analysing lifetime data from heritage systems using, for example, bootstrapping have been developed. The model has been formulated under both classical and Bayesian philosophies, including combined approaches that support model validation. Action research has been conducted with several aerospace companies to assess the impact of modelling on the role of reliability in the design process.
Ongoing research in this area includes development of methods for efficient reliability task allocation, trading reliability within a supply chain, empirical Bayes methods for data combination and elicitation of structured expert judgement as well as applications of growth modelling to systems in other industrial sectors.
Competing Risks
It costs a lot of money to maintain large equipment. For example, when a power station fails unexpectedly, it may cost £20,000 per hour for the first few hours because the electricity that should have been produced has to be bought from the electricity spot market. By collecting data about the time taken until equipment fails it is possible to determine the chance the equipment will work past a certain length of time. Unfortunately, often the equipment doesn't actually fail, but is removed from service for some other reason. Maybe something else failed and all the equipment had to be maintained or maybe someone thought that the equipment may fail and decided to inspect it "just in case". The "just in case" preventive maintenance poses a problem for us because we cannot tell how long the equipment would have gone on to work for. Some preventive maintenance may be very good, getting the equipment just before it fails. Other preventive maintenance may be quite poor, occurring at times that have little to do with the actual failure time. To make it even more complicated, equipment can often fail in more than one way (usually called failure modes). This sort of situation is called "competing risks" because there are different mechanisms (for example failure types or preventive maintenance) that are competing to be the first to take the equipment out of service.
In this project we developed models that take account of different possible ways of performing preventive maintenance. We were able to take account of the preventive maintenance when analysing the failure data. We also looked at the way different failure modes interacted so that we could model what would happen if we tried to delay one failure mode. Often improvements to the system and the method of preventive maintenance do not get made because not enough people are convinced that the change would be for the better. By modelling the situation we want to build a decision support tool that would enable us to predict the effect of change without doing it in practice. That way we can choose the best course of action at a low cost. In order to ensure that the solutions we come up with are really relevant to practical needs this project is carried out with help from Scottish Power who provide both staff time and data.
Common Cause Failure
- Decision Analysis
- Structured Expert Elicitation
- Bayesian Analysis
- Risk Strategy Development
- Consequence Modelling
- Dependency Modelling
- Maintenance Modelling
- Human Reliability
- Cost-Benefit Analysis
