Probabilistic bulk property estimation using multimodality surface non-destructive measurements for vintage pipes

Serving as energy lifelines, pipelines remain one of the most efficient and economical ways to move natural resources. The mechanical properties of pipelines installed decades ago decline with time. To maintain the safety of vintage pipelines requires accurate estimation of the strength. This paper focus on the reliability-based strength prediction using nondestructive multimodality information by the method of Bayesian model averaging (BMA). A class of models are formed from all cases of linear combinations of the surface property measurements. The models are averaged based on the posterior model probabilities. Occam's window is introduced to reduce the number of models under consideration while keeping the predictive accuracy. By not conditioning on any single model, BMA provide more reliable strength prediction by accounting for model uncertainties. In addition, the usefulness of the variables used to predict the strength are evaluated according to the frequency of appearance in the models with high posterior probabilities. The variables with paramount predictive importance can be selected by this way. Thus, BMA method shows advantages in both vintage pipe strength prediction and model selection.