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 p_c 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.

研究失效评估模型的不确定性对于含裂纹管道的安全管理和可靠性计算非常重要。在本文中，首先，我们使用 9 种不同的失效评估模型预测了 250 次实验的爆破压力，并根据实测测试数据获得了预测精度 ( PA )。其次，PA根据新的模型预测性能评价体系分析不同模型的分布位置特征（粗略准确度、风险、鲁棒性和保守性）和稳定性（离散性、相关性和多峰性）。最后，根据最佳拟合分布确定不同模型的关键安全系数以控制可靠性。结果表明，对于分布位置特征，CorLAS最准确；R6-2A-I（R6-Option 2A，基于 Tresca 标准的全局解决方案，2016），API-Ⅱ（API RP 579-Level 2，2016），R6-2A-III（R6-Option2A，本地解决方案， 2016) 和 SINTAP-I (SINTAP-Level 1, 1999) 是稳健的；CorLAS 和 R6-2A-II（R6-Option 2A，基于 von Mises 准则的全局解决方案，2016）有风险，GB-Ⅱ（GB/T-Conventional Assessment，2019) 和 BS7910-Ⅰ (BS 7910-Option 1, 2019) 非常保守。它们的稳定性顺序为：①R6-2A-III；② R6-2A-I、R6-2A-II、API-Ⅱ和SINTAP-I；③ GB-Ⅱ和BS7910-Ⅰ；④ 巴特尔；⑤ CorLAS。GB-Ⅱ、BS7910-Ⅰ、Battelle 和 CorLAS 分散。之间的相关性PA和p _c对于所有模型都很弱并且可以忽略不计。最佳拟合分布是 R6 的对数正态分布，其他分布是 SINTAP 和 GMM。最后，通过考虑关键安全因素，建议使用所有模型。