Going from in-situ nondestructive testing to a probabilistic MAOP

摘要

This paper discussed a probabilistic approach to validating the facility MAOP in response to the IVP requirement under PHMSA's NPRM. Conforming to the pipe design formula, two factors need to be addressed: pipe grade, i.e. SMYS and pipe NWT. On account of the complexity of pipe grade dependence on various attributes, machine learning techniques were explored to infer the intended grade purchased by Operators or sold by pipe manufacturers. The naive Bayes classifier was selected to enhance convergence and to avoid over-fitting. The algorithm itself directly produced the probability associated with each possible pipe grade. An iterative procedure developed for inferential analysis demonstrated a confidence of 99% for indicating Grade B or stronger where actual pipe present is Grade B or stronger and a confidence of 80% for conservatively predicting pipe grade. It is envisioned that multiple sets of measurements on the same batch of pipe and review of results by a subject matter expert would further improve the accuracy. Regarding pipe NWT, some probabilistic approach was devised to draw inference using AWT measurements. The algorithm was built on the comparison of AWT variation between what can exist in a single joint of pipe and allowed by API 5L/5LX manufacturing specification. The overall accuracy reached 91% and the robustness of algorithm was confirmed by a sensitivity analysis. The NDE in-situ technology is capable of providing all the necessary inputs to the probabilistic design pressure calculation. Pipe grade prediction counts on mechanical strength, which can be obtained from IIT measurements or comparable technique, and chemical composition, which can be determined by optical emission spectroscopy (OES) or spark emission spectroscopy of filings. Pipe NWT prediction can be derived from conventional straight beam UT inspection at each clock positions around the pipe circumference. Subsequent to possible grade and NWT determined by using NDE data, the design pressure exhibits a discrete probability distribution, which serves as the basis of validating and evaluating a current MAOP. The results of an analysis of the type described above can be used with expected flaw distributions based on condition assessment and operational data to estimate probable fatigue life.