Bayesian inspection planning for large industrial systems

Abstract

The implementation of consistent and repeatable methods for inspection planning is a problem faced by a wide range of industries. The theory of Bayesian design problems provides a well established method for the treatment of inspection planning problems, but is often difficult to implement for large systems due to its associated computational burden. We develop a tractable Bayesian method for inspection planning. The use of Bayes linear methods in the place of traditional Bayesian techniques allows us to assess properties of proposed inspection designs with greater computational efficiency. This improvement in efficiency allows a greater range of designs to be assessed and the design space to be searched more effectively. We propose a utility based criterion for the identification of designs that offer improved prediction for future system properties. Designs with good typical performance are identified through utility maximisation. The viability of the method is demonstrated by application to an example based on data from a real industrial system.