Bayesian networks (BN)
BN's are basically a framework for reasoning under uncertainty. In a broad sense they're a set of methods for probabilistic calculation and graphical representation that can be used for most problems with uncertainty.
BN's have their background in statistics and artificial intelligence. They emerged in the 1980's when there was found to be a need for a formalism that adequately dealt with uncertainty in knowledge-based systems.
BN’s are a very useful tool in developing risk models. Their graphical representation facilitates model structuring with domain expertise but not necessarily requiring in-depth understanding of uncertainty modelling.
Our recent work
The following are provide some examples of our work in this area.
Asset management & the Forth Road Bridge
More than 11,600 steel wires grouped into 37 strands make up each main cable on the iconic Forth Road Bridge.
The cables are secured to the north and south shores of the river by pairs of anchorages, the key components of which are buried deep in rock with no direct means of access.
Civil and structural consultants W. A. Fairhurst & Partners, along with Glasgow and Strathclyde researchers Professors John Quigley and Lesley Walls are working to identify the best option for assessing the state of the Forth Road Bridge.
Although there's no external signs of a problem, work is ongoing to determine the best way of inspecting the anchorages to monitor their condition for the future.
For practical and logistical reasons only a limited sample of the tendons which form the anchorages can be inspected. The challenge is to extrapolate the information obtained by each inspection to assess the overall anchorage capacity and condition.
Professors Quigley and Walls use an analysis technique, known as Bayesian theorem, which combines engineers' expert judgement with statistical observations. Combining judgement with observations results in a more accurate depiction than can be achieved separately.
Professor Quigley said:
"We first work with engineers to identify the factors that impact on the condition of the anchorages and put these into a model. Then we assess the experts understanding of the strength of the impact of these factors, which is ultimately refined in light of statistical observations."
"Once you've got agreement on the structure of the model and expert assessment of the strength of the impact of the factors, you can combine it with data from tests.
"Say they've taken a sample of 10 tendons from the anchorage and maybe 8 out of 10 had a good grout condition, Bayes theorem and our mathematical models can give a probability of the grout on the remaining wires being in good condition."
"The technique will help Fairhurst to make an informed judgement on the best way to estimate the condition and capacity of the anchorages."
"Different types of tests will help us learn about the different impacts on the anchorages. We can then think about strategies on the best way to plan tests and inspections."
"For example Fairhurst could do a simple test and then given the results opt to do another more elaborate test if necessary."
The project complements the work of the Strathclyde Risk Consortium - an industry and academic partnership to develop tools for effective decision making in business.
Supply Chain Risk Management
Management Science academics are part of a UK-wide group that's won a prestigious major grant of almost £1 million to develop tools to better manage risks in manufacturing supply networks.
Funded by the Engineering and Physical Sciences Research Council Strathclyde won a bid in partnership with the universities of Bristol, Coventry and Nottingham. Each academic institution focuses on a different element of the project, building on their distinctive knowledge and experience of management, modelling and risk. Each element will be brought together through collaborative working over this three-year project to develop useful methods and tools.
Industry partners including Nautricity, Rolls-Royce, Dynex, Tricorn Group, PA Consulting and DSTL. They'll be involved in this research from the outset. An initial launch meeting of the research and industry team took place at Nottingham in November 2014.
Professor Lesley Walls explains:
"The idea is to develop ways of managing resilient supply networks: supply networks that can continue to deliver if there is a natural hazard or disaster. Loss of supply flow due to such hazards can lead to major disruption in global supply chains. While such risks cannot be predicted with certainty, it is possible to use our understanding of hazard types and the behaviour of networks to design supply chains that are more resilient. For example, companies can think through alternative supply providers or routes to plan mitigations and so maintain operations."
Professor John Quigley further clarifies:
"As well as natural hazards, strategic uncertainties also affect supply networks. Manufacturing is a competitive business and so there are risks associated with international partnerships. Take an example where a company is doing business in China, then it might be useful to offer decision support for the managers there so that companies can ensure contracts are watertight or appropriate incentives are used in different situations.
"Obviously, different companies will have different uncertainties that have to be addressed. However, it is possible to conceive a general framework that encompasses the types of decisions that companies might expect to make to manage a resilient supply chain."
A general approach can be applied to specific scenarios. Take an aerospace company, which by its very nature will be reliant on supplies of high-specification parts. Some of the parts may have a limited number of suppliers meaning that there is a risk that a competitor buys up required components leading to loss of supply. "Someone buying up parts critical to the supply chain would be extremely problematic," says Professor Quigley.
"In response, a company might take production in-house but that could be expensive, and so there are risks involved either way. Our work involves looking at possibilities for risks to occur and to rule out those deemed more unlikely, concentrating instead on the more likely scenarios."
Dynamic Strategic Bayesian Networks
Professors Quigley and Walls will head the Strathclyde research that will focus specifically on a novel modelling approach which they call Dynamic Strategic Bayesian Networks. This is a new idea that will extend risk modelling methodology and should lead to a reduction in supply risks associated with partnership forming and investment strategies.
"Our intention is to come up with a general model for dealing with risk. The companies will be giving their time, information and data to help us do that, and through dialogue industry will make us aware of their particular challenges. While we won’t be providing specific solutions during the research project, by the end of the research period we will provide a general approach that can be adapted by each company to its particular circumstances."