Bayesian inverse transient analysis for pipeline condition assessment: parameter estimation and uncertainty quantification

Abstract

Strategic pipeline asset management requires accurate and up-to-date information on pipeline condition. As a tool for pipeline condition assessment, inverse transient analysis (ITA - a pipeline model calibration approach) is typically formulated as a deterministic problem, and optimization methods are used for searching a single best solution. The uncertainty associated with the single best solution is rarely assessed. In this paper, the pipeline model calibration problem is formulated as a Bayesian inverse problem, and a Markov Chain Monte Carlo (MCMC) based method is used to construct the estimated posterior probability density function (PDF) of the calibration parameters. The MCMC based method is able to achieve parameter estimation and uncertainty assessment in a single run, which is confirmed by numerical experiments. The proposed technique is also validated using measured hydraulic transient response data from an experimental laboratory pipeline system. Two thinner-walled pipe sections (simulating extended deterioration) are successfully identified with an assessment of the parameter uncertainty. The results also suggest that proper sensor placement can reduce parameter uncertainty and significantly enhance system identifiability.