Failure prediction and analysis techniques

Abstract

Techniques for implementing Bayesian model for refining and petrochemical wall thickness monitoring are presented. The construction of a model may be automated for each piping circuit or piece of major fixed equipment, utilizing, for example, specific component data, historical thickness measurements, inspection practices and related inspection program information. The model contains nodes describing the most significant sources of variability, namely component original thicknesses, wall thickness degradation over time, corrosion rates and thickness measurement error. Bayesian prior distributions are assigned using readily available inspection program information, including assigned damage mechanisms, inspector and industry experience regarding the expected range of corrosion rates and degree of non-uniform corrosion, thickness monitoring practices, including surface preparation and instrument calibration, thickness data recording practices and component original thicknesses based on applicable industry specifications, typical values for size/component combinations or detailed ultrasonic thickness scanning data generated specifically for this purpose.

Dependencies are inferred from the claim text.