Solving the Non-Linear ILI Sizing Challenge of Crack-Like Flaws With Bayesian Logic

Abstract The sizing performance of in-line inspection (ILI) tools are commonly illustrated on unity plots relating the ILI-reported size to a validation measurement, typically a non-destructive evaluation (NDE) field-measured height obtained using techniques expected to be more accurate than the ILI. The comparison of ILI and NDE field measurements is simplified as a linear relationship and tolerance interval. While the slope for a perfectly performing system is 1, it is rarely 1 in practice. Pipeline operators have continuously strived for better ILI sizing performance to optimize maintenance cost and prioritize repairs. API 1163 has comprehensively addressed sizing performance with the incorporation of Level 3 (Bayesian Approach) that enables pipeline operators to bring new analysis techniques to validate performance, but the linearity of the model still presents challenges because it assumes the performance is consistent across all flaw sizes. The challenges associated with a linear model are especially problematic for thin-wall pipelines due to height sizing errors. Height tends to be the most influential parameter for the fitness-for-service acceptance criteria of crack-like flaws. In this paper, a non linear sizing model that uses Bayesian logic is described with the intention of enhancing ILI validation analysis and for future inclusion and interpretation by the API 1163 technical committee. A case study of a population of crack-like flaws is used to illustrate the non-linear sizing performance in ILI results and explain the causes for common sizing errors. While this paper focused on crack-like flaws, the non-linearity of ILI sizing performance is also applicable to metal loss ILI results.