基于有限元和深度学习的掺氢天然气缺陷管道失效压力预测

doi:10.3969/j.issn.2096-1693.2026.02.052

摘要/Abstract

摘要：

掺氢天然气输送是氢能规模化利用的重要路径，但氢环境可能导致材料性能劣化，增加含缺陷管道的失效风险，亟需高效的评估手段。为快速准确预测掺氢天然气缺陷管道失效压力，本文提出了融合有限元模拟与深度学习的方法。首先，基于ABAQUS建立了含外腐蚀缺陷的X80管线钢三维有限元模型，通过引入不同掺氢比的应力—应变曲线表征氢环境对材料性能的影响。模型经过网格无关性验证和实验对比，确保了其可靠性与准确性。随后，批量计算了401组不同掺氢比、缺陷尺寸及管道几何特征的工况，构建了失效压力数据库，为深度学习模型提供样本支撑。结果表明，缺陷深度和长度对失效压力有显著负相关影响，缺陷宽度影响较小且呈非线性，而掺氢比对失效压力影响不明显。通过深度学习模型参数优化试算，确定网络结构为3隐层、350个神经元，并采用ReLU激活函数与0.0001学习率。经过6次随机验证，预测结果与有限元计算高度吻合，平均相对误差约0.765%，最大误差低于4.02%，显示出良好的泛化能力。与有限元方法相比，深度学习模型的计算效率显著提高，单组样本的预测耗时约为0.01 s，能够有效缩短缺陷管道失效压力的评估周期。研究表明，所建立的有限元—深度学习融合框架为掺氢天然气缺陷管道失效压力的高效精确预测提供了可靠工具。

关键词: 掺氢天然气, 缺陷管道, 失效压力, 有限元模拟, 深度学习, 高效计算

Abstract:

Hydrogen-blended natural gas transportation is a key pathway for the large-scale utilization of hydrogen energy. However, the hydrogen environment may induce material performance degradation, increasing the failure risk of defective pipelines, necessitating the development of efficient and accurate assessment methods. To rapidly and accurately predict the failure pressure of defective hydrogen-blended natural gas pipelines, this study proposes a method that integrates finite element simulation and deep learning. First, a three-dimensional finite element model of X80 pipeline steel with external corrosion defects was established using ABAQUS, and the effect of the hydrogen environment on material properties was represented by introducing stress-strain curves under different hydrogen blending ratios. The model was validated for mesh independence and experimental comparison to ensure its reliability and accuracy. Subsequently, 401 cases with different hydrogen blending ratios, defect sizes, and pipeline geometries were simulated to construct a failure pressure database, providing sample support for the deep learning model. The results show that defect depth and length have a significant negative correlation with failure pressure, defect width has a relatively weak and nonlinear effect, and the hydrogen blending ratio has a negligible impact on failure pressure. The deep learning model was optimized through hyperparameter tuning, determining the network structure with 3 hidden layers, 350 neurons, and the ReLU activation function with a learning rate of 0.0001. After 6 random validation tests, the predicted results highly matched the finite element calculations, with an average relative error of approximately 0.765% and a maximum error of less than 4.02%, demonstrating good generalization capability. In terms of computational efficiency, deep learning significantly outperforms finite element analysis, requiring only about 0.01 s to predict a single sample, greatly reducing the evaluation cycle. The study shows that the established finite element-deep learning integration framework provides a reliable tool for the efficient and accurate prediction of failure pressures in hydrogen-blended natural gas defective pipelines.

Key words: hydrogen-blended natural gas, defective pipelines, failure pressure, finite element simulation, deep learning, efficient computation

中图分类号:

谭智鹏, 刘志斌, 王智琨, 张寓淇, 黄敏珂, 王博弘, 李敬法, 苏越. 基于有限元和深度学习的掺氢天然气缺陷管道失效压力预测[J]. 石油科学通报, 2026, 11(5): 1544-1556.

TAN Zhipeng, LIU Zhibin, WANG Zhikun, ZHANG Yuqi, HUANG Minke, WANG Bohong, LI Jingfa, SU Yue. Prediction of failure pressure for defective hydrogen-blended natural gas pipelines using finite element analysis and deep learning[J]. Petroleum Science Bulletin, 2026, 11(5): 1544-1556.

https://sykxtb.cup.edu.cn/CN/Y2026/V11/I5/1544

图/表 14

图1 管道示意图

Fig. 1 Schematic diagram of the pipeline

表1 钢材力学性能

Table 1 Mechanical properties of steel

密度(g/cm³)	弹性模量(GPa)	泊松比
7.85	206	0.3

图2 压力载荷与网格划分示意图

Fig. 2 Pressure load and mesh generation schematic diagram

表2 网格无关解的计算

Table 2 Grid-independent solutions

序号	网格数	失效压力/MPa
1	6140	19.589
2	9198	19.617
3	14 994	19.557
4	17 612	19.320
5	21 312	19.205
6	25 212	19.206

图3 实验验证

Fig. 3 Experimental verification

图4 缺陷几何尺寸对失效压力的影响

Fig. 4 The influence of defect geometric dimensions on failure pressure

图5 不同掺氢比下的应力—应变曲线

Fig. 5 Stress-strain curves under different hydrogen blending ratios

图6 不同应力—应变曲线失效压力计算结果

Fig. 6 Calculation results of failure pressures on different stress-strain curves

图7 单层感知机与深度学习模型结构示意图

Fig. 7 Schematic diagram of single-layer perceptron and deep learning model structure

表3 常用激活函数

Table 3 Commonly used activation functions

函数名称	数学表达式	导数表达式	特性简述
Sigmoid	$\sigma(z)=\frac{1}{1+e^{-z}}$	$\sigma^{\prime}(z)=\sigma(z)[1-\sigma(z)]$	将输出压缩至(0，1)易导致梯度饱和
Hyperbolic Tangent(Tanh)	$\tanh (z)=\frac{e^{z}-e^{-z}}{e^{z}+e^{-z}}$	$\tanh ^{\prime}(z)=1-\tanh ^{2}(z)$	输出以0为中心，范围(-1，1)
Rectified Linear Unit(ReLU)	$\operatorname{ReLU}(z)=\max (0, z)$	$\operatorname{ReLU}^{\prime}(z)=\left\{\begin{array}{ll} 1, & \text { if } z>0 \\ 0, & \text { otherwise } \end{array}\right.$	计算高效，缓解梯度消失，但可能导致神经元死亡
Leaky ReLU	$\operatorname{LReLU}(z)=\max (\alpha z, z)$	$\operatorname{LReLU}^{\prime}(z)=\left\{\begin{array}{ll} 1, & \text { if } \quad z>0 \\ \alpha, & \text { otherwise } \end{array}\right.$	ReLU的改进，为负输入保留一小部分梯度(α为常数)

图8 不同隐藏层数与神经元个数下的计算精度

Fig. 8 Prediction accuracy for different numbers of hidden layers and neurons

图9 不同激活函数与学习率下的计算精度

Fig. 9 Accuracy under different activation functions and learning rates

图10 随机测试结果

Fig. 10 Results of random testing

表4 计算效率对比

Table 4 Comparison of computational efficiency

计算方法	计算时间/s
有限元	3080
离线深度学习	120
在线深度学习	60

参考文献 32

[1]	张军, 王财林, 张慧敏, 等. 在役天然气管道掺氢应力腐蚀开裂研究进展及展望[J]. 油气储运, 2025, 44(10): 1081-1094.
	[Zhang J, Wang C L, Zhang H M, et al. Progress and prospects of research on stress corrosion cracking in in-service natural gas pipelines transporting hydrogen-doped blends[J]. Oil & Gas Storage and Transportation, 2025, 44(10): 1081-1094.]
[2]	李敬法, 苏越, 张衡, 等. 掺氢天然气管道输送研究进展[J]. 天然气工业, 2021, 41(4): 137-152.
	[Li J F, Su Y, Zhang H, et al. Research progresses on pipeline transportation of hydrogen-blended natural gas[J]. Natural Gas Industry, 2021, 41(4): 137-152.]
[3]	李天娇, 谢飞, 张瑛, 等. 天然气管道掺氢输送技术研究进展[J]. 油气储运, 2024, 43(12): 1337-1347.
	[Li T J, Xie F, Zhang Y, et al. Advancements in hydrogen-blended natural gas pipeline transmission technology[J]. Oil & Gas Storage and Transportation, 2024, 43(12): 1337-1347.]
[4]	张博, 梁永图, 魏雪梅, 等. 不同边界条件下掺氢对天然气管网运行工况的影响分析[J]. 石油科学通报, 2024, 9(5): 808-818.
	[Zhang B, Liang Y T, Wei X M, et al. Analysis of the influence of hydrogen blending on the operation of natural gas pipeline network under different boundary conditions[J]. Petroleum Science Bulletin, 2024, 9(5): 808-818.]
[5]	刘啸奔, 张东, 武学健, 等. 掺氢天然气管道完整性评价技术的进展与挑战[J]. 力学与实践, 2023, 45(2): 245-259.
	[Liu X B, Zhang D, Wu X J, et al. Advances and challenges of hydrogen-blended natural gas pipeline integrity assessment technology[J]. Mechanics in Engineering, 2023, 45(2): 245-259.]
[6]	陈健, 穆科, 赵洪超, 等. 腐蚀管道剩余强度计算方法研究和修正[J]. 石油工程建设, 2025, 51(4): 54-59.
	[Chen J, Mu K, Zhao H C, et al. Research and correction of calculation methods for residual strength of corroded pipeline[J]. Petroleum Engineering Construction, 2025, 51(4): 54-59.]
[7]	王赤宇, 虞维超, 陈磊, 等. 考虑两层相关性的腐蚀油气管道失效概率计算方法[J]. 石油科学通报, 2024, 9(5): 819-830.
	[Wang C Y, Yu W C, Chen L, et al. A method for calculating the failure probability of corroded oil and gas pipelines considering the two layers of correlation[J]. Petroleum Science Bulletin, 2024, 9(5): 819-830.]
[8]	陈灵搏. 基于混合式模型的天然气管道剩余强度预测研究[D]. 成都: 西南石油大学, 2024.
	[Chen L B. Study on residual strength prediction of natural gas pipeline based on hybrid model[D]. Chengdu: Southwest Petroleum University, 2024.]
[9]	Su Y, Li J F, Yu B, et al. Fast and accurate prediction of failure pressure of oil and gas defective pipelines using the deep learning model[J]. Reliability Engineering and System Safety, 2021, 216: 108016
[10]	Abyani M, Bahaari M R, Zarrin M, et al. Predicting failure pressure of the corroded offshore pipelines using an efficient finite element based algorithm and machine learning techniques[J]. Ocean Engineering, 2022, 254: 111382.
[11]	姚志澜. 城市埋地油气管道剩余强度预测及可靠性研究[D]. 重庆: 重庆交通大学, 2025.
	[Yao Z L. Residual strength prediction and reliability study of urban buried oil and gas pipelines[D]. Chongqing: Chongqing Jiaotong University, 2025.]
[12]	Zhang H, Tian Z G. Failure analysis of corroded high-strength pipeline subject to hydrogen damage based on FEM and GA-BP neural network[J]. International Journal of Hydrogen Energy, 2022, 47(7): 4741-4758.
[13]	Miao X Y, Ma Y H, Sun X L, et al. Residual strength prediction of hydrogen-blended natural gas pipelines based on incremental knowledge distillation[J]. Energy, 2025, 341: 139456.
[14]	Xie M J, Wei Z Q, Zhao J L, et al. Failure analysis of corroded hydrogen-blended natural gas pipelines based on finite element analysis and genetic algorithm-back propagation neural network[J]. Reliability Engineering and System Safety, 2025, 262: 1111174.
[15]	曹宇, 陈鹏, 杨建功, 等. 掺氢环境下含腐蚀缺陷X52天然气管道的结构强度失效机理研究[J]. 中国舰船研究, 2024, 19(4): 186-192.
	[Cao Y, Chen P, Yang J G, et al. Structural strength failure mechanism of X52 natural gas pipeline with corrosion defects under hydrogen-blended environment[J]. Chinese Journal of Ship Research, 2024, 19(4): 186-192.]
[16]	Bjornoy O H, Sigurdsson G, Cramer E. Residual strength of corroded pipelines, DNV test results[C]//Proceedings of the 10th international offshore and polar engineering conference. Seattle: International Society of Offshore and Polar Engineers, 2000: 189-196.
[17]	史雪婷. 复杂载荷作用管道腐蚀缺陷剩余强度非线性有限元研究[D]. 西安: 西安石油大学, 2017.
	[Shi X T. Non-linear finite element research for the corrosion defects of pipeline residual strength on complex loadings[D]. Xi’an: Xi’an Shiyou University, 2017.]
[18]	Meng B, Gu C H, Zhang L, et al. Hydrogen effects on X80 pipeline steel in high-pressure natural gas/hydrogen mixtures[J]. International Journal of Hydrogen Energy, 2017, 42(11): 7404-7412.
[19]	Yang H C, Zhang H M, Liu C W, et al. Effects of defect on the hydrogen embrittlement behavior of X80 pipeline steel in hydrogen-blended natural gas environments[J]. International Journal of Hydrogen Energy, 2024, 58: 158-173.
[20]	赵杰, 徐广旭, 张烘玮, 等. X80掺氢天然气管道的氢脆与腐蚀耦合作用研究[J]. 中国腐蚀与防护学报, 2025, 45(2): 407-415.
	[Zhao J, Xu G X, Zhang H W, et al. Coupling effect of hydrogen embrittlement and corrosion of X80 pipeline steel in hydrogen-doped natural gas[J]. Journal of Chinese Society for Corrosion and Protection, 2025, 45(2): 407-415.]
[21]	Zhang M, Guo Y B, Xie Q J, et al. Defect identification for oil and gas pipeline safety based on autonomous deep learning network[J]. Computer Communications, 2022, 195(C): 14-26.
[22]	Liu B, Ding L Y, He L Y, et al. Residual generative adversarial network-driven Data enhancement for magnetic flux leakage-based defect recognition in oil and gas pipelines[J]. Engineering Applications of Artificial Intelligence, 2025, 162: 112720.
[23]	Liu Z E, Long W N, Chen Z L, et al. A novel optimization framework for natural gas transportation pipeline networks based on deep reinforcement learning[J]. Energy and AI, 2024, 18: 100434.
[24]	Siriwardana G, Rajapakse A, Bandara T, et al. Multiclass classification of two-phase flow patterns using classical and deep learning models with automated hyperparameter optimization[J]. International Communications in Heat and Mass Transfer, 2025, 169: 109886.
[25]	Xu Q, Wang X Y, Luo X Y, et al. Machine learning identification of multiphase flow regimes in a long pipeline-riser system[J]. Flow Measurement and Instrumentation, 2022, 88: 102233.
[26]	Bayazitova G, Anastasiadou M, dos Santos V D. Oil and gas flow anomaly detection on offshore naturally flowing wells using deep neural networks[J]. Geoenergy Science and Engineering, 2024, 242: 213240.
[27]	张驰, 郭媛, 黎明. 人工神经网络模型发展及应用综述[J]. 计算机工程与应用, 2021, 57(11): 57-69.
	[Zhang C, Guo Y, Li M. Review of development and application of artificial neural network models[J]. Computer Engineering and Applications, 2021, 57(11): 57-69.]
[28]	Kiliçarslan S, Celik M. RSigELU: A nonlinear activation function for deep neural networks[J]. Expert Systems with Applications, 2021, 174: 114805.
[29]	Al-Nawi N, Madain A, Abdullah M. MishReLU: A hybrid activation function for deep neural networks[J]. Expert Systems with Applications, 2026, 319: 131953.
[30]	Yotov K, Hadzhikolev E, Hadzhikoleva S, et al. Finding the optimal topology of an approximating neural network[J]. Mathematics, 2023, 11(1): 217.
[31]	Glorot X, Bengio Y. Understanding the difficulty of training deep feedforward neural networks[C]//Proceedings of the thirteenth international conference on artificial intelligence and statistics. Sardinia: PMLR, 2010, 9: 249-256.
[32]	Smith L N. Cyclical learning rates for training neural networks[C]//2017 IEEE winter conference on applications of computer vision (WACV), Santa Rosa, CA, USA, 2017: 464-472.

选择文件类型/文献管理软件名称

选择包含的内容