The paper Bayes linear learning about large industrial systems has been accepted to RSS 2010. The abstract for the paper is given below:

Bayes linear learning about large industrial systems [Abstract]

Objectives

Careful inspection and maintenance of large industrial systems is essential to assure system efficiency and integrity. For example manufacturing systems, susceptible to corrosion, require careful monitoring and maintenance to avoid economic and environmental costs. Modelling these complex systems can improve the effectiveness of inspection and maintenance activities. Typically, system components are modelled independently despite their inter-dependence, to make analysis tractable. Here, we adopt a Bayes linear framework to facilitate multivariate modelling of the full system. Model forecasts provide a basis for future inspection design.

Methods/Models

A Bayes Linear approach is adopted, simplifying parameter estimation and avoiding often-unrealistic distributional assumptions. Exchangeability assumptions are made to facilitate analysis of sparse inspection time-series. We estimate system evolution in time, together with key model variances. As a result, future system measurement campaigns can be designed with respect to an utility-based expected loss criterion, to aid decision making.

Results and Conclusions

Evolution of wall thickness and corrosion rate and their dependence for multiple corroding components in a large industrial system are modelled. Exploiting dependence structure improves forecast quality, allowing optimal design of future inspection campaigns.