Intelligent Code for Assessing Performance and Reliability of Failure Assessment Models

Abstract

Pipelines often fail due to structural degradation and environmental factors, impacting the accuracy of failure assessment models (FAMs) such as Structural Integrity Assessment Procedures for European Industry (SINTAP), R6, and BS7910. Traditional error analysis methods are limited by their complexity and lack of practical application. This paper introduces an Intelligent Code for Evaluating Models (ICEM) to accurately assess FAM prediction accuracy. ICEM uses an experimental database to compute characteristic parameters and distribution patterns of prediction accuracy. It comprises four components: feature parameter extractor (FPE), fitted distribution processor (FDP), overfitting processor (OP), and recommendation factor processor (RFP). FPE calculates moment parameters like mean, median, standard deviation, and coefficient of variation. FDP identifies the best-fit distribution using the maximum likelihood method, EM algorithms, and goodness-of-fit (GoF) tests (K-S test, NlogL, AIC, and BIC). OP prevents overfitting by removing outliers, and RFP provides a recommended factor (rf) based on the best-fit distributions and specific project risk degrees. By offering a systematic and practical approach, ICEM enhances the reliability of FAMs, thereby significantly improving prediction confidence and accuracy, ultimately contributing to better pipeline management, safety, and operational efficiency.