Discussion on “Data-driven Methods to Predict the Burst Strength of Corroded Line Pipelines Subjected to Internal Pressure https://doi.org/10.1007/s11804-022-00263-0

Mokhtari Mojtaba Melchers Robert E.

Mojtaba Mokhtari, Robert E. Melchers (2025). Discussion on “Data-driven Methods to Predict the Burst Strength of Corroded Line Pipelines Subjected to Internal Pressure https://doi.org/10.1007/s11804-022-00263-0”. Journal of Marine Science and Application, 24(1): 249-251. https://doi.org/10.1007/s11804-024-00565-5
Citation: Mojtaba Mokhtari, Robert E. Melchers (2025). Discussion on “Data-driven Methods to Predict the Burst Strength of Corroded Line Pipelines Subjected to Internal Pressure https://doi.org/10.1007/s11804-022-00263-0”. Journal of Marine Science and Application, 24(1): 249-251. https://doi.org/10.1007/s11804-024-00565-5

Discussion on “Data-driven Methods to Predict the Burst Strength of Corroded Line Pipelines Subjected to Internal Pressure https://doi.org/10.1007/s11804-022-00263-0

https://doi.org/10.1007/s11804-024-00565-5
    Corresponding author:

    Mojtaba Mokhtari mojtaba.mokhtari@ntnu.no

  • The authors (Cai et al., 2022) claim that their proposed machine learning (ML) models, which are based on three typical ML algorithms and are trained to predict the burst capacity of pitting corroded pipelines, perform better than the existing semi-empirical formulas recommended by the international engineering code developers, DNV and ASME. The authors’assessments of the semi-empirical burst capacity formulas in Figure 10(a) and Table 8 (Cai et al., 2022) incorrectly indicate that DNVGL-RP-F101, ASME B31G, and modified ASME B31G are dangerously unsafe due to significantly overestimating burst pressures in several cases. In contrast to the results and conclusions in Cai et al. (2022), these semi-empirical formulas have been consistently found to be conservative (Benjamin et al., 2000; Benjamin et al., 2002; Chena et al., 2015; Cronin and Pick, 2000; Kiefner and Vieth, 1989; Mokhtari and Melchers, 2016; 2018; 2019; Netto et al., 2005; Teixeira et al., 2008; Yeom et al., 2015; Zhou and Huang, 2012), and have been safely implemented in the oil/gas industry for many years. The main cause of the contradiction between the authors’ claims and well-established practical applications and results lies in the data upon which the study (Cai et al., 2022) is based, along with the authors’ misinterpretation of the semi-empirical formulas. These points are elaborated below.

    To train their ML models, the authors (Cai et al., 2022) collected pipe and burst pressure data from a series of experimental studies reported in the literature, the same data that were used to assess the performance of the semi-empirical burst capacity formulas recommended by the DNV and ASME codes. However, some of the collected data presented in Cai et al. (2022) do not appear to match those in the reference/source studies. Other data appear to be misinterpreted. Two examples that caused the significantly unsafe predictions (i.e. overestimations) by the semi-empirical burst capacity formulas are given below.

    Table 8 (Cai et al., 2022) shows that DNVGL-RP-F101 overestimates the burst pressure by 26.31% for S. N. 61. This drastic overestimation is because the authors incorrectly substituted the ‘true’ ultimate tensile strength (UTS) instead of the ‘engineering’ UTS for σu in Eq. (13) (Cai et al., 2022). In DNVGL (2017a), σu is the ‘engineering’ UTS, which is often replaced by the specified minimum tensile strength (SMTS) because, in practice, only the material grade is available which provides SMTS and not the UTS (Abdelghani et al., 2018; Callister, 1997; DNVGL, 2017a; DNVGL, 2017b; Gao et al., 2019; Mondal and Dhar, 2019; Mustaffa and van Gelder, 2010). The ‘engineering’ UTS for S. N. 61 is about 570 MPa in the reference study by Choi et al. (2003), while the 675 MPa listed in Table 1 (Cai et al., 2022) for S.N. 61 is the ‘true’ UTS of the pipe material. Note that the authors (Cai et al., 2022) have given the wrong reference, Freire et al. (2006), for S.N. 61, which was actually taken from (Choi et al., 2003).

    S.N. 33 is another case for which all the semi-empirical burst capacity models, DNVGL-RP-F101, ASME B31G, and modified ASME B31G, are shown in the author’s Table 8 (Cai et al., 2022) to markedly overestimate the burst pressure. These overestimations are the direct result of the authors using an incorrect defect length in the semi-empirical formulas. The circumferential defects in Mok et al. (1991), from which the S. N. 33 data in Cai et al. (2022) were taken, are ring grooves and run around the entire circumference of the pipe. In Mok et al. (1991), the extent of metal loss along the axial direction of the pipe caused by the ring grooves is 100.235 mm for two of the four burst tests with circumferential defects and 203.2 mm for the other two. However, the authors (Cai et al., 2022) have substituted a very short length of 10 mm for these circumferential defects (e. g., S. N. 33 in Table 1 (Cai et al., 2022)). This error has caused the DNVGL-RP-F101, ASME B31G, and modified ASME B31G equations to overestimate the burst pressure of S.N. 33 by 35%, 21%, and 21%, respectively. Much larger false overestimations (e.g., around 70% for DNVGL-RP-F101) can be observed for S.N. 34 in Figure 10 of Cai et al. (2022) due to substituting 10 mm for the defect length that is actually 203.2 mm.

    To conclude, the DNV and ASME burst capacity models are misjudged in Cai et al. (2022) due to using poor quality data, as showcased by the examples given above. In turn, this invalidates the ML models since they have been trained with erroneous data. While there are many accurate semi-empirical models in the engineering standards and literature (e.g. PCORRC (Leis and Stephens, 1997; Stephens and Leis, 1997; Stephens et al., 2000) and CSA model (CSA, 2007)), the authors (Cai et al., 2022) chose instead to employ an outdated model, ASME B31G (ASME, 1984; 1991). The B31G model (ASME, 1984; 1991) has long been known to produce excessively conservative results and consequently, it was modified to a less conservative model, the modified B31G (Kiefner and Vieth, 1989; 1990), more than three decades ago. It is therefore difficult to understand why the authors chose to use the overly conservative B31G model in their comparative study instead of other more accurate models. This comparative study, which includes only three semi-empirical models from two engineering standards (i.e. ASME B31G and DNVGL-RP-F101), cannot justify their universal conclusion that their proposed data-driven models perform better than the semi-empirical models in the existing engineering standards.

    Regardless of the errors in the study (Cai et al., 2022), the concept of estimating the burst pressure of pitting corroded pipelines using ML models is not meaningful in the light of recently developed and powerful automated finite element analysis (FEA) tools, employed in finite element based digital twins. Further, despite the authors’ statement that FEA is too time-consuming (Cai et al., 2022), the recent automation of FEA tools has led to the rapid, accurate physics-based prediction of burst pressures (see Cabral et al. (2007), Motta et al. (2010), Silva et al. (2008)) for initial versions of such tools with rectangular defects and Pimentel et al. (2020) for a more recent academic version with complex-shaped defects). In the latest versions of fully automated FEA tools with cloud computing options, developed and implemented in consulting firms, there is almost no human intervention in the pre- and post-processing as the input data are automatically read from a text file and the required results are produced in a report file. The input data and the finite element models are automatically updated based on the field data collected by sensors. Then, the models are solved automatically with negligible computational cost due to the combination of cloud computing and efficient finite element models. Finally, the computed results/data such as the burst pressure are fed back to the operator to close the loop. These tools do not have the ML models’ generalization issue as they can automatically develop multiple internal and/or external defects with different shapes (not only rectangular) and can account for the operational/environmental loads, effects of temperature on material properties, different boundary conditions, etc. In contrast to these modern powerful tools, the ML models in Cai et al. (2022) are trained only for pipes subject to a single, external rectangular pit (not common in practice) and internal pressure. Even for such a simple case, the authors encountered a data shortage in the literature during data collection and trained their models with a small dataset. In addition, they made (human) errors in interpreting and applying the available data.

    If the data used to train an ML model is significantly erroneous such as that in Cai et al. (2022), there is always the possibility of producing highly inaccurate predictions. However, this is not an issue in the automated FEA tools because the finite element models are physics-based and validated against many experimental tests during their development phase. Even one overprediction of the burst pressure could be catastrophic given that steel pipelines transfer hazardous and flammable fluids under high pressure. Ultimately, it seems entirely unreasonable to undergo the immensely labour-intensive, time-consuming and expensive process of data collection/preparation to develop an ML model free of human errors that could account for various corrosion defect types/shapes/numbers and operational and boundary conditions when, at best, such an effort can only reach the same capability as that already available in the rapidly improving FEA tools used in digital twins.

    It should be noted that the above argument refers to ML model development efforts that focus on the specific topic studied in Cai et al. (2022), which is “time-independent burst capacity prediction of corroded pipelines based on experimental data”. Otherwise, there are many challenges in the pipeline industry that could benefit from ML methods (e.g., see Chen et al. (2022)) and more practical ways to obtain sample data. For example, sample data could be efficiently generated by validated and automated finite element models with negligible errors to save a tremendous amount of time and effort, avoid misinterpretation of data, and eliminate the generalization issue of the ML models.

    Competing interest The authors have no competing interests to declare that are relevant to the content of this article.
  • Abdelghani M, Tewfik G, Djahida D, Sid Ahmed S (2018) Prediction of the rupture pressure of transmission pipelines with corrosion defects. Journal of Pressure Vessel Technology 140(4): 041701. https://doi.org/10.1115/1.4039698
    ASME (1984) Manual for determining the remaining strength of corroded pipelines. American Society of Mechanical Engineers, ANSI/ASME B31G-1984
    ASME (1991) Manual for determining the remaining strength of corroded pipelines. American Society of Mechanical Engineers, ASME B31G-1991
    Benjamin AC, Freire J, Vieira RD, de Castro JT (2002) Burst tests on pipeline with nonuniform depth corrosion defects. 21st International Conference on Offshore Mechanics and Arctic Engineering, Oslo, 51–57. https://doi.org/10.1115/OMAE2002-28065
    Benjamin AC, Vieira RD, Freire JLF, de Castro JT (2000) Burst tests on pipeline with long external corrosion. 3rd International Pipeline Conference, Calgary, V002T06A013. DOI: 10.1115/IPC2000-193
    Cabral HL, Willmersdorf RB, Afonso SM, Lyra PR, De Andrade EQ (2007) Development of computational tools for automatic modeling and FE analysis of corroded pipelines. International Journal of Modeling and Simulation for the Petroleum Industry 1(1): 9–22
    Cai J, Jiang X, Yang Y, Lodewijks G, Wang M (2022) Data-driven methods to predict the burst strength of corroded line pipelines subjected to internal pressure. Journal of Marine Science and Application 21(2): 115–132. https://doi.org/10.1007/s11804-022-00263-0
    Callister WD (1997) Materials science and engineering. Wiley
    Chen BQ, Videiro PM, Guedes Soares C (2022) Opportunities and challenges to develop digital twins for subsea pipelines. Journal of Marine Science and Engineering 10(6): 739. https://doi.org/10.3390/jmse10060739
    Chena Y, Zhang H, Zhang J, Li X, Zhou J (2015) Failure analysis of high strength pipeline with single and multiple corrosions. Materials and Design 67: 552–557. https://doi.org/10.1016/j.matdes.2014.10.088
    Choi J, Goo B, Kim J, Kim Y, Kim W (2003) Development of limit load solutions for corroded gas pipelines. International Journal of Pressure Vessels and Piping 80(2): 121–128. https://doi.org/10.1016/S0308-0161(03)00005-X
    Cronin D, Pick R (2000) Experimental database for corroded pipe: evaluation of RSTRENG and B31G. Proceedings of the 3rd International Pipeline Conference, Calgary, V002T06A010. https://doi.org/10.1115/IPC2000-190
    CSA (2007) Gas pipeline systems. CSA standard, Z662-07
    DNVGL (2017a) Corroded pipelines. DNVGL, DNVGL-RP-F101
    DNVGL (2017b) Submarine pipeline systems. DNVGL, DNVGL-ST-F101
    Freire JLF, Vieira RD, Castro JTP, Benjamin AC (2006) Part 3: Burst tests of pipeline with extensive longitudinal metal loss. Experimental Techniques 30(6): 60–65. https://doi.org/10.1111/j.1747-1567.2006.00109.x
    Gao J, Yang P, Li X, Zhou J, Liu J (2019) Analytical prediction of failure pressure for pipeline with long corrosion defect. Ocean Engineering 191: 106497. https://doi.org/10.1016/j.oceaneng.2019.106497
    Kiefner JF, Vieth PH (1989) A modified criterion for evaluating the remaining strength of corroded pipe. Technical report, Battelle Columbus Div., OH (USA), PR-3-805
    Kiefner JF, Vieth PH (1990) New method corrects criterion for evaluating corroded pipe. Oil and Gas Journal 88: 32
    Leis B, Stephens D (1997) An alternative approach to assess the integrity of corroded line pipe-part Ⅰ: current status. The Seventh International Offshore and Polar Engineering Conference
    Mok D, Pick R, Glover A, Hoff R (1991) Bursting of line pipe with long external corrosion. International Journal of Pressure Vessels and Piping 46(2): 195–216. https://doi.org/10.1016/0308-0161(91)90015-T
    Mokhtari M, Melchers RE (2016) Advanced numerical method for failure assessment of corroded steel pipes. 26th International Ocean and Polar Engineering Conference
    Mokhtari M, Melchers RE (2018) A new approach to assess the remaining strength of corroded steel pipes. Engineering Failure Analysis 93: 144–156. https://doi.org/10.1016/j.engfailanal.2018.07.011
    Mokhtari M, Melchers RE (2019) Next-generation fracture prediction models for pipes with localized corrosion defects. Engineering Failure Analysis 105: 610–626. DOI: 10.1016/j.engfailanal.2019.06.094
    Mondal BC, Dhar AS (2019) Burst pressure of corroded pipelines considering combined axial forces and bending moments. Engineering Structures 186: 43–51. https://doi.org/10.1016/j.engstruct.2019.02.010
    Motta RdS, Afonso SM, Willmersdorf RB, Lyra PR, de Andrade EQ (2010) Automatic modeling and analysis of pipelines with colonies of corrosion defects. Mecánica Computacional 29(80): 7871–7890
    Mustaffa Z, van Gelder P (2010) A review and probabilistic analysis of limit state functions of corroded pipelines. Proceedings of the Twentieth International Offshore and Polar Engineering Conference, Beijing, 626–632
    Netto TA, Ferraz US, Estefen SF (2005) The effect of corrosion defects on the burst pressure of pipelines. Journal of Constructional Steel Research 61: 1185–1204 https://doi.org/10.1016/j.jcsr.2005.02.010
    Pimentel JT, Marques Ferreira AD, de Siqueira Motta R, Figueiroa da Silva Cabral MA, Maria Bastos Afonso S, Brito Willmersdorf R, Maciel Lyra PR, de Andrade EQ, da Silveira Cunha DJ (2020) New procedure of automatic modeling of pipelines with realistic shaped corrosion defects. Engineering Structures 221: 11030 https://doi.org/10.1016/j.engstruct.2020.111030
    Silva R, Guerreiro J, Drach P (2008) Automatic finite element solid modeling, burst and error analyses of corroded pipelines. International Journal of Mechanics 2(3): 77–86 http://www.naun.org/main/NAUN/mechanics/m-29.pdf
    Stephens DR, Leis BN (1997) Material and geometry factors controlling the failure of corrosion defects in piping. American Society of Mechanical Engineers, Pressure Vessels and Piping Division PVP 350: 3–11 http://www.researchgate.net/publication/292278070_Material_and_geometry_factors_controlling_the_failure_of_corrosion_defects_in_piping
    Stephens DR, Leis BN, Kurre M (2000) Development of an alternative criterion for residual strength of corrosion defects in moderate-to high-toughness pipe. 3rd International Pipeline Conference, Calgary, V002T06A012. https://doi.org/10.1115/IPC2000-192
    Teixeira AP, Guedes Soares C, Netto TA, Estefen SF (2008) Reliability of pipelines with corrosion defects. International Journal of Pressure Vessels and Piping 85(4): 228–237. https://doi.org/10.1016/j.ijpvp.2007.09.002
    Yeom KJ, Lee Y-K, Oh KH, Kim WS (2015) Integrity assessment of a corroded API X70 pipe with a single defect by burst pressure analysis. Engineering Failure Analysis 57: 553–561. https://doi.org/10.1016/j.engfailanal.2015.07.024
    Zhou W, Huang G (2012) Model error assessments of burst capacity models for corroded pipelines. International Journal of Pressure Vessels and Piping 99: 1–8. https://doi.org/10.1016/j.ijpvp.2012.06.001
WeChat click to enlarge
Publishing history
  • Received:  20 September 2022
  • Accepted:  21 April 2023

目录

    /

    Return
    Return