Statistical inference and performance evaluation for failure assessment models of pipeline with external axial surface cracks
Lingyun Guo,Markus Niffenegger,Zhou Jing
DOI: https://doi.org/10.1016/j.ijpvp.2021.104480
IF: 2.56
2021-01-01
International Journal of Pressure Vessels and Piping
Abstract:It is important to investigate the uncertainty of failure assessment models for safety management and reliability calculations of pipelines containing cracks. In this paper, firstly, we predict the burst pressure of 250 experiments with 9 different failure assessment models and obtain the prediction accuracy (PA) based on the measured test data. Secondly, the PA of different models is analyzed according to the new model prediction performance evaluation system, which includes distributional location characteristics (rough accuracy, risk, robustness and conservativeness) and stability (dispersion, correlation and multi-peakedness). Finally, the critical safety factors are determined for different models to control reliability based on the best-fit distribution. The results show that for distribution location characteristics, CorLAS is the most accurate; the R6-2A-I (R6-Option 2A, global solution based on the Tresca criteria, 2016), API-Ⅱ (API RP 579-Level 2, 2016), R6-2A-III (R6-Option2A, local solution, 2016) and SINTAP-I (SINTAP-Level 1, 1999) are robust; CorLAS and R6-2A-II (R6-Option 2A, global solution based on the von Mises criteria, 2016) are risky, and GB-Ⅱ (GB/T-Conventional Assessment, 2019) and BS7910-Ⅰ (BS 7910-Option 1, 2019) are very conservative. The order of their stability is: ① R6-2A-III; ② R6-2A-I, R6-2A-II, API-Ⅱ and SINTAP-I; ③ GB-Ⅱ and BS7910-Ⅰ; ④ Battelle; ⑤ CorLAS. GB-Ⅱ, BS7910-Ⅰ, Battelle and CorLAS are dispersed. The correlation between PA and pc is weak and negligible for all models. The best-fit distributions are lognormal for R6 and SINTAP and GMM for the others. Finally, the use of all models is recommended by considering the critical safety factors.