邢海燕, 王松弘泽, 弋鸣, 杨健平, 朱孔阳, 刘超. 基于IPSO-GRU深度学习算法的海底管道缺陷尺寸磁记忆定量反演模型[J]. 工程科学学报, 2022, 44(5): 911-919. DOI: 10.13374/j.issn2095-9389.2020.11.06.001
引用本文: 邢海燕, 王松弘泽, 弋鸣, 杨健平, 朱孔阳, 刘超. 基于IPSO-GRU深度学习算法的海底管道缺陷尺寸磁记忆定量反演模型[J]. 工程科学学报, 2022, 44(5): 911-919. DOI: 10.13374/j.issn2095-9389.2020.11.06.001
XING Hai-yan, WANG Song-hong-ze, YI Ming, YANG Jian-ping, ZHU Kong-yang, LIU Chao. Metal magnetic memory quantitative inversion model based on IPSO-GRU algorithm for detecting submarine pipeline defect[J]. Chinese Journal of Engineering, 2022, 44(5): 911-919. DOI: 10.13374/j.issn2095-9389.2020.11.06.001
Citation: XING Hai-yan, WANG Song-hong-ze, YI Ming, YANG Jian-ping, ZHU Kong-yang, LIU Chao. Metal magnetic memory quantitative inversion model based on IPSO-GRU algorithm for detecting submarine pipeline defect[J]. Chinese Journal of Engineering, 2022, 44(5): 911-919. DOI: 10.13374/j.issn2095-9389.2020.11.06.001

基于IPSO-GRU深度学习算法的海底管道缺陷尺寸磁记忆定量反演模型

Metal magnetic memory quantitative inversion model based on IPSO-GRU algorithm for detecting submarine pipeline defect

  • 摘要: 针对海底管道缺陷磁记忆定量反演的难题,提出一种基于改进粒子群优化的门控循环神经网络模型,即IPSO-GRU模型。以两端焊有盲板的X52管道作为实验材料,其上预制有不同直径、深度的缺陷,采用TSC-5M-32磁记忆检测仪,外接11-6W非接触探头,进行水下磁记忆检测试验,提取不同缺陷尺寸的磁记忆信号特征值。考虑到磁记忆信号特征值随缺陷尺寸呈复杂的非线性变化,引入门控循环神经网络,利用其双门结构能够记忆缺陷处的信号特征,非线性回归拟合能力强的特点,构建海底管道缺陷定量反演模型,进一步考虑到模型超参数选择的随机性,采用改进粒子群算法进行超参数寻优。验证结果表明:该模型对缺陷深度反演平均精度达96%;对缺陷直径反演平均精度达93%,为海底管道缺陷的磁记忆定量化识别与反演提供了新的思路和方法。

     

    Abstract: As submarine oil and gas are exploited further, the safety of submarine pipelines is receiving increasing attention. Due to the complex operating environment and harsh working environment, submarine pipelines are vulnerable to damage; this leads to accidents. Once an accident occurs in the submarine pipeline, it not only causes massive economic losses but also adversely affects marine ecology. The metal magnetic memory (MMM) technology was proposed in the 20th century to detect macro defects and hidden defects early. To overcome the difficulties of the MMM quantitative inversion of submarine pipeline defects, this study proposed a gated recurrent unit (GRU) neural network model based on improved particle swarm optimization (IPSO). The X52 pipe specimens with blind plates that were welded at both ends were used, pipes had prefabricated defects of different diameters and depths. An 11-6W noncontact probe was used for underwater testing; the host was the TSC-5M-32 MMM Instrument. After conducting simulated submarine tests to obtain the MMM signals of pipe defects, the characteristic parameters of MMM signals with different defect sizes were extracted. It is found that the MMM characteristic parameters exhibit a complex nonlinear variation for different defect dimensions. Exploiting the GRU’s dual-gate structure that can remember the signal characteristics of defects and its superior nonlinear regression fitting ability, a quantitative MMM GRU inversion model was established for detecting submarine pipeline defects. Furthermore, considering the randomness of the hyper-parameter selection in the model, the IPSO algorithm was used to optimize the hyper-parameters. Validation results show that the model has an average accuracy of up to 96% and 93% for defect depth inversion and defect diameter inversion, respectively. Using the MMM method, this study provides a new idea and method for the quantitative identification and defect inversion of submarine pipeline defects.

     

/

返回文章
返回