Advanced intelligence frameworks for predicting maximum pitting corrosion depth in oil and gas pipelines

https://doi.org/10.1016/j.psep.2021.01.008Get rights and content

Abstract

The main objective of this paper is to develop accurate novel frameworks for the estimation of the maximum pitting corrosion depth in oil and gas pipelines based on data-driven techniques. Thus, different advanced approaches using Artificial Intelligence (AI) models were applied, including Artificial Neural Network (ANN), M5 Tree (M5Tree), Multivariate Adaptive Regression Splines (MARS), Locally Weighted Polynomials (LWP), Kriging (KR), and Extreme Learning Machines (ELM). Additionally, a total of 259 measurement samples of maximum pitting corrosion depth for pipelines located in different environments were extracted from the literature and used for developing the AI-models in terms of training and testing.Furthermore, an investigation was carried out on the relationship between the maximum pitting depths and several combinations of probable factors that induce the pitting growth process such as the pipeline age, and the surrounding environmental properties. The results of the proposed AI-frameworks were compared using various criteria. Thus, statistical, uncertainty and external validation analyses were utilized to compare the efficiency and accuracy of the proposed AI-models and to investigate the main contributing factors for accurate predictions of the maximum pitting depth in the oil and gas pipeline.

Introduction

Corrosion is a natural process affecting metallic structures such as petroleum infrastructures, especially oil and gas pipelines (Arzaghi et al., 2020). The failure of corroded pipelines has great potential to bring about large negative environmental issues, including climate change caused by the soil’s and sea’s pollution, toxic releases due to explosion, and fires related to the failure of pipelines (Wang et al., 2019a). With intensive fossil fuel exploitation, the integrity and safety of operating corroded pipelines play a significant economical role in the oil and gas industry (Lam and Zhou, 2016). In terms of reliability, safety, and integrity management of oil and gas pipelines, access to related-data from the corrosion on the pipe surface is the most important factor (Keshtegar et al., 2019). These data are employed in the reliability analysis and prediction performance of the structures (Guillal et al., 2020). The state of corrosion in operating oil and gas pipeline can be evaluated by using inspection operations by utilizing in-line inspection tools and validated by in-site excavation measurements (Wang et al., 2019b). The costs associated with the inspection operation are relatively high, especially for large pipeline systems. Moreover, these operations are not possible for the entire available pipeline systems, as some of them are not equipped with the necessary inspection structures (El-Abbasy et al., 2016). While on-site excavations are used to reduce and calibrate the inspection error, this process consists of choosing several corrosion points from the inspection report and re-measuring by the pipeline operators, using specific tools. Even though this process is more accurate for improving the precision of the measurements, it has several risks as damaging the operated pipeline while excavating, can cause serious problems (Li et al., 2019). On the other hand, at each pipeline location, the growth of the corrosion is different as this phenomenon is highly dependent on the environment as local properties in contact with the pipe surfaces change the corrosion state by time and location. Thus, the high cost of measuring equipment and the requirements of specific and periodic calibration operations conclude that the state of corrosion defect measurements is not easily available.

For field engineering and owners of oil and gas pipelines, as well as for safety and environment professionals, it is very important to have accurate approaches for the estimation of maximum pitting depth in oil and gas pipelines based on the available inspection/field reports (El Amine Ben Seghier et al., 2018). The most commonly used parameter in several empirical models, which have been used to calculate the maximum pitting depth on the external surfaces of the pipelines is soil properties (Balekelayi and Tesfamariam, 2020). The soil texture contains several parameters that help corrosion initiation and growth on the pipe-walls (Bazán and Beck, 2013). Other parameters such as air, temperature, and humidity can be utilized (Nešić, 2007). The first model used for the prediction of corrosion depth in metallic structure was proposed early by Romanoff (1957), Rousum (1969) and Sim et al. (2014), in which a nonlinear model was derived to explain the relationship between real experiments and time of exposure. Velázquez et al. (2009) modified the Rousum formula by introducing the initiation time of corrosion and improving the formula by replacing the constant variables with ones that depend on the soil properties, in which their extraction implied the use of the real, large database. This improved model has been the most widely used correlation for the calculation of pitting corrosion depth in oil and gas pipelines (Seghier M el et al., 2018; Amine BSM el et al., 2017). Later, other models appeared which are based on the nonlinear regression utilizing different corrosion databases such as the Alamilla et al. (2009) model. The problem with the above models is the development basis where simple techniques such as linear and non-linear regression methods were used, and which have been proven to be inaccurate approaches for handling high dimensional databases with large complexity as in the case in pitting corrosion problem in pipelines.

More recently, Artificial Intelligence (AI) models have been used to overcome the problems of analytical, classical linear, and nonlinear approaches for modeling complex and chaotic regression problems based on experimental or real-field databases. AI-techniques are already applied when dealing with various civil engineering fields such as developing new structural reliability methods (Chojaczyk et al., 2015; Bagheri et al., 2020), safety and risk assessment (Mamudu et al., 2020; Osarogiagbon et al., 2020; Guo et al., 2020), and modeling and prediction problems (Almheiri et al., 2020; Shokry et al., 2020; Ben Seghier et al., 2020). Different AI-approaches can be found in the literature such as the most well known AI-technique, which is the Artificial Neural Network (ANN) (El-Abbasy et al., 2014; Din et al., 2015; Nazari et al., 2015). Regarding the modeling of the corrosion parameters in general cases and pitting depth for pipelines as a special case, only few works can be found where the authors applied or refer to the use of AI-techniques. Wen et al. (2009) are one of the first researchers who conducted a study using AI-approaches for the modeling of corrosion rate in 3C steels, whereas this latter was submerged in five seawater environments. Two AI-models were proposed as Support Vector Regression (SVR) and back-propagation neural network (BPNN), in which the SVR shows more accurate results compared to the BPNN. The study was based on very small databases that were considered as a limitation in showing the performance of theAI-models. Later, Chou et al. (2017) applied four different AI-models using the same database as Artificial Neural Networks (ANNs), Classification, and Regression tree (CART), Support Vector Regression/Machines (SVR/SVM), and linear regression (LR). Results indicate the performance of the AI-models for the modeling of the pitting corrosion risk and the marine corrosion rate. Recently El Amine Ben Seghier et al. (2020) developed three models based on SVR technique as a first attempt for modeling the maximum depth of corrosion in oil and gas pipelines. In their work, the SVR-model was optimized using the Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Firefly Algorithm (FFA), and its efficiency was compared to the Velázquez et al. (2009) and Alamilla et al. (2009) correlations. The work results revealed that those correlations are inaccurate compared to the precision and efficiency of the AI-techniques. Despite the previous studies, the applicability of several AI-techniques is yet to be explored for modeling of the maximum pitting depth in oil and gas pipelines in terms of accuracy, robustness and efficiency.

Hence, the question of the current research is: what is the potential of the AI-techniques to mimic the relationship between the maximum pitting depth in oil and gas pipelines and surrounding environment properties and time of exposure? Also, and to the best knowledge of the authors, the present research is the first implementation and investigation of different input combinations to predict the maximum pitting depth. Thus, the research findings can be capitalized for the field researchers as the following perspectives (1) suggest a review to identify the works related to modeling corrosion in structures using data-driven models. (2) Compare the efficiency of various AI-techniques for the modeling of the maximum pitting depth in the oil and gas pipeline and (3) identify the relative importance of various uncertain input parameters on the prediction of maximum pitting corrosion depth. The paper is outlined as follows. A recall of the proposed AI-techniques theory and backgrounds used for modeling the maximum pitting depth are reported in Section 2. The modeling methodology including the real excavation database is described in Section 3. The comparative results and discussions are provided with details in Section 4. The paper is concluded in Section 5 with the salient points noted in the current study.

Section snippets

Artificial Neural Network (ANN)

The main effort during the modeling process using the ANN is to provide the best connection between input variables and output nonlinear/complex responses. Consisting of three layers including input, hidden, and output layers, the ANN is widely used as a machine learning tool to approximate nonlinear mathematical forms (Hornik et al., 1989). As represented in Fig. 1, the ANN nonlinear regression is utilized to predict the maximum pitting depth by using the following relations (Mai et al., 2021):

Collected database

A comprehensive database is a crucial step for the building of the above-described AI-models with high-accuracy. Generally, the database should be composed of a large number of data samples obtained from real experimental tests or field measurements. Velázquez et al. (2010) database is considered the largest database that covers a variety of important parameters included in the initiation and growth of corrosion defects. 259 sample data were measured and collected by Velázquez et al. (2010)

Results and discussion

This section covers the evaluation of the obtained results using the six established AI-models for the eight scenarios. The performance metrics are reported in Table 4. Nine statistical indicators (i.e. MBE, RMSE, MARE, RMSRE, U95, NSE, WI, and CI) were used to evaluate and compare the efficiency effectiveness and accuracy of the 48 resulting cases for dmax predictions. In accordance with Table 4, the best AI-models are the ones with the lowest MBE, RMSE, MARE, and RMSRE values and the highest U

Conclusion

During the prediction of maximum pitting depth for oil and gas pipelines, related data, including the exposure time to the surrounding environmental properties at each specific location, are essential. However, for many oil and gas pipelines, these measurements are not easily available due to the cost of inspection operation, excavation requirements and the failure risks of the vulnerable pipelines. Besides, for human-safety, economics and environmental reasons, it is essential to define

Declaration of Competing Interest

The authors report no declarations of interest.

References (57)

  • G.-B. Huang et al.

    Extreme learning machine: theory and applications

    Neurocomputing

    (2006)
  • P.D. Jamieson et al.

    A test of the computer simulation model ARCWHEAT1 on wheat crops grown in New Zealand

    Field Crops Res.

    (1991)
  • B. Keshtegar et al.

    Modified response surface method basis harmony search to predict the burst pressure of corroded pipelines

    Eng. Fail. Anal.

    (2018)
  • B. Keshtegar et al.

    Reliability analysis of corroded pipelines: novel adaptive conjugate first order reliability method

    J. Loss Prev. Process Ind.

    (2019)
  • C. Lam et al.

    Statistical analyses of incidents on onshore gas transmission pipelines based on PHMSA database

    Int. J. Press. Vessel Pip.

    (2016)
  • M.-F. Li et al.

    General models for estimating daily global solar radiation for different solar radiation zones in mainland China

    Energy Convers. Manage.

    (2013)
  • X. Li et al.

    Risk-based operation safety analysis during maintenance activities of subsea pipelines

    Process Saf. Environ. Prot.

    (2019)
  • A. Mamudu et al.

    A hybrid intelligent model for reservoir production and associated dynamic risks

    J. Nat. Gas Sci. Eng.

    (2020)
  • A. Nazari et al.

    Offshore pipeline performance evaluation by different artificial neural networks approaches

    Meas. J. Int. Meas. Confed

    (2015)
  • S. Nešić

    Key issues related to modelling of internal corrosion of oil and gas pipelines - a review

    Corros. Sci.

    (2007)
  • S. Sim et al.

    A review of the protection strategies against internal corrosion for the safe transport of supercritical CO2 via steel pipelines for CCS purposes

    Int. J. Greenh. Gas Control

    (2014)
  • Y. Wang et al.

    Non-probabilistic time-dependent reliability analysis for suspended pipeline with corrosion defects based on interval model

    Process Saf. Environ. Prot.

    (2019)
  • H. Wang et al.

    A stochastic defect growth model for reliability assessment of corroded underground pipelines

    Process Saf. Environ. Prot.

    (2019)
  • Y.F. Wen et al.

    Corrosion rate prediction of 3C steel under different seawater environment by using support vector regression

    Corros. Sci.

    (2009)
  • M. Xiao et al.

    A system active learning Kriging method for system reliability-based design optimization with a multiple response model

    Reliab. Eng. Syst. Saf.

    (2020)
  • W.G. Zhang et al.

    Multivariate adaptive regression splines for analysis of geotechnical engineering systems

    Comput. Geotech.

    (2013)
  • J. Zhang et al.

    A novel projection outline based active learning method and its combination with Kriging metamodel for hybrid reliability analysis with random and interval variables

    Comput. Methods Appl. Mech. Eng.

    (2018)
  • Y. Zhang et al.

    Maximizing natural frequencies of inhomogeneous cellular structures by Kriging-assisted multiscale topology optimization

    Comput. Struct.

    (2020)
  • Cited by (94)

    View all citing articles on Scopus
    View full text