Bayes linear Bayes graphical models in the design of optimal test strategies

Test and analysis plays a vital role in reducing uncertainty about the true performance of an engineering system. However tests can be expensive and designing an optimal test strategy can be challenging. We propose a Bayesian modelling process, which takes the form of a Bayesian Network, to determine anticipated test efficacy. Such a model supports engineering managers in assessing trade-offs between test resources and uncertainty reduction. Inference based on a full Bayesian model can be computationally demanding to the extent that it can limit practical application. To overcome this constraint, we develop a Bayes linear approximation for inference. This approach is known as a Bayes linear Bayes graphical model. After explaining the key principals of the method, we provide an application to a real industrial test to establish the condition of an ageing engineering system.