Abstract
Pipelines are observed one of the economic modes of transport for transporting oil, gas, and water between various locations. Most of the countries in the world transport petroleum and other flammable products through underground pipelines. The underground and aboveground pipelines are facing various damages due to corrosion, dents, and ruptures due to the environment and operational fluid conditions. The danger of leaks and accidents increases as a result of these damages. Pipelines must be evaluated on a regular basis to make sure they are fit for transmission. By evaluating the effects of damages and the possibility of catastrophic failures using a variety of techniques, pipeline integrity is controlled. Applying the relative risk scoring (RRS) technique, pipeline failures are predicted. One of the probabilistic techniques used to forecast risk based on an impartial assessment is machine learning. With different parameters like corrosion, leakage, materials, atmosphere, surface, earth-movements, above-ground and underground facilities, etc., the RRS method provides an accuracy of 97.5% in identifying the risk and gives a precise classification of risk, whether the pipeline has a high, medium, or low risk without any delay on the prediction compared with Naive Bayes, decision tree, support vector machine, and graph convolutional network.
Similar content being viewed by others
Introduction
India's economy is growing rapidly, necessitating increased hydrocarbon transport capacity. An item of machinery called a pipeline is made to move material constantly or irregularly from one place to another. Modern technologies prefer pipelines over other modes of transportation due to financial and safety reasons. Pressure is applied as highly flammable hydrocarbon material is transported across pipelines across the nation, frequently close to dense populations and places with a high environmental consciousness. Allocating a lot of money to preventive measures activities for mitigation and detection will help you analyse the risk presented by international pipelines effectively. To protect individuals, the general public, the environment, and property, more safety must be provided. An effective algorithm is required to simplify the processes and lower the failure rate of gas pipelines1.
Pipelines pose risks such as jet fire, unconfined vapour cloud explosion, flash fire, delay ignition, pool fire, and toxicity. There are many integrity management tools, like cathodic protection, inline inspection, hydro testing, surveillance, direct assessment and evaluations, pipeline equipment health monitoring, and thickness assessment2,3. The pipeline threats are categorised as fixed, non-stationary, and stationary. The non-stationary threats are outer erosion, inner erosion, and strain erosion splitting. Pipeline defects can occur during the manufacturing process, as well as during welding or fabrication4. Some of the failures that occur as a result of gasket toxicity (joint failure, guide apparatus malfunction, and clinch force out failure) must also be considered when considering the threat to stable equipment.
Many other threats should be considered as time-independent threats while analysing the threats in equipment and external factors. As with computer hackers, some third parties may damage pipeline equipment or perform incorrect operations, causing the pipeline to malfunction. Natural disasters5 like lightning, heavy rains, floods, weather-related events, and earth movements also affect the quality of pipelines. Due to the severity of the composite risk from all threats, pipeline sections may be prioritised for integrity assessment.
The overall risk value for a specific pipeline section is determined by the likelihood of failure and its consequences, taking into account all relevant dangers. Risk ranking6 will be determined for pipeline segments found to be at high risk in order to arrange the integrity evaluation. Prior to the execution of any pipeline framework advancements, an exact cycle will be put in place to make sure that impending changes are evaluated for their potential risk effects on the pipeline. The RRS method concentrates on using the relative risk scoring method of machine learning to perform the integrity and qualitative risk assessment of pipelines while taking into account all of these situations and factors.
The focal objectives of this research are:
-
1.
There are other studies that are dependent on specific aspects, such as leakage, corrosion, etc., but none of them offer a specific solution for all of the problems that are widespread.
-
2.
The Research focus to assess the risk of pipelines by calculating the Relative Risk Score (RRS) method.
-
3.
The RRS approach takes many parameters into account that could impact a pipeline. The methodology that is being suggested here aims to lessen the harm happens due to harmful gases, chemicals, and petroleum products inflict on people and other living things.
-
4.
To demonstrate the superior effectiveness of the novel approach the RRS method contrast it with Naive Bayes, Decision tree, SVM, and GCN.
The arrangement of this research paper is as per the following: The existing methodologies and literature background for pipeline assessment are presented in Segment 2. The terminologies related to the pipeline assessment are explained in Segment 3. The proposed RRS algorithm using machine learning is elaborated in Segment 4. The experimental setup and the discussion of results are presented in segment 5, and the research is concluded in segment 6.
Literature review
Risk assessment of chemical pipelines
The use of machine learning techniques in pipeline risk assessment has been a topic of increasing interest in recent years. Sohaib et al.5 proposed a method for detecting leaks in circular water storage tanks in the chemical sector using acoustic emissions. Support vector machines are used to locate the exact location of the crack or leakage. Mazumder et al.7 used machine learning algorithms to analyze the risk of failure of a steel pipeline. The research developed an alternative to statistically intensive analytical methods to estimate the steel pipeline failure threat. However, the research could not be fully realized due to the lack of adequate data for use in actual situations.
Yang et al.8 demonstrated urban gas data-driven pipeline accidents and consequences assessment using machine learning. The accidents in UPN may cause environmental disasters. Urban pipeline maintenance is related to the major facets of data. The work uses conventional assessment for risk models like the Kent index method and analytic evaluation indicators. The graph convolutional network (GCN) technique is used to assess the risk associated with pipelines. Liu and Bao9 reviewed automated conditions for the assessment of pipelines with machine learning. Pipelines, which transport intense substances, play a significant role in societal safety and commercial prosperity. Pipeline condition assessments are required to identify the risk.
Wu et al.10 presented FTAP: A feature-transferring autonomous machine learning pipeline. Successful machine learning11 frequently involves significant involvement with algorithms and expert knowledge in the field. The ML methods heavily rely on domain-specific information. FTAP improves efficiency and performance. It has also achieved success in distance domain transfer learning. He et al.12 used Geographical Information System at the threat location to develop a genetic and back propagation prototype to forecast the occurrence of a geographical calamity and avert pipeline damage. The GA-BP model is used to estimate weights of the indication factor by utilising the benefits of analysing data and predictive analytics, which avoid the subjective authority of earlier calculation methods.
Malinowska et al.13 showed off a model that uses Mamdani fuzzy inference for the study. In this method, the hazards in the pipeline due to the estimated horizontal strains, the solidity of the material, the time span of an unfavourable effect, and the importance of the targets are evaluated on the basis of one level. The prototype is used to assess the risk of an arranged gas pipe with systematic variables. This can also be used to stimulate further scientific variations of the examined item. Bu et al.14 investigated how soil-atmosphere coupling affects gas pipeline15 leakage. The joining process of methane leakage was calculated using arithmetic stimulation, and the effects of many factors on methane convergence in FDZ and SDZ were investigated. This analysis provides guidelines for the planning of gas pipes, improving the installation of detectors, and gas leakage16 maintenance.
Liu et al.17 demonstrated a dynamic danger estimation prototype depending on system dynamics (SD). Most of the pipeline risk assessments provide a static outline of the process. This model is to manage both the issue of given activity and changes there within a period. The solution provided the dynamic developments of principles of buried pipelines threat are compatible with real circumstances. This prototype adequately distinguishes the temporal and spatial principles of threat evolution.
Adumene et al.18 presented a method that combines the semi-empirical corrosion prototype with material used and parametric uncertainties. The pipe failure rises with a rise in factor of difference at the minimum limit of futile, while reduces in the maximum limit of futile. This method provides an organized structure for choosing material and threat-based integrity management plans for sea pipelines.
Froeling et al.19 demonstrated the danger of dangerous hydrogen jet fire transported through gas pipelines20,21. This analysis focuses on threats associated with a dangerous hydrogen fire, such as an invisible spray with a gas pipeline22. Using analytical software, it performs a detailed fire imitation and condition of the artwork. When compared to a gas pipeline, the ignition effects influence the threat for reducing pipe diameters and causing hydrogen transference to capitulate an increased hazard. Taleb-Berroune et al.23 suggested a prototype for the risk assessment for the deterioration of pipelines utilising adaptive bow-tie (ABT) analysis. The ABT model used for this analysis is engrossed in microbiologically impacted pipeline deterioration, as well as a corrosion economic risk profile. This prototype serves as a procedure to recognize, evaluate, and maintain the corrosion of the pipeline.
Wang et al.24 initiated a susceptibility assessment technique for the danger assessment of the gas pipe system. It integrates the features of threat assessment and susceptible analysis methods. Risk susceptibility classifies its critical components into three categories: the operating status of the pipeline, transmission performance, and network features. A utility proposition is employed to identify the depth of the outturn. This method balances the lower part of the threat, trustability, and susceptibility curves to adequately recognise the unfavourable joints and pipelines damaging gas supply in a pipeline network.
Zulkifli and Salleh25 analysed the effect of movement and pipe construction actions at various pressures and pipe widths on the pipeline in the UTHM biodiesel plant. The Computational Fluid Dynamics (CFD) technique helped them to examine the velocity and pressure dispersion, while the Interaction of Fluid Structure (FSI) method enabled them to examine the stress distribution on the pipes according to their thicknesses. The study found that the breadth of the pipe affects the flow rate and pressure in the region of the pipe, and pressure diffusion is reduced as quality improves. Additionally, when pressure increases, the storm created within the pipe also increases.
Risk assessment of buried pipelines
Li et al.26 proposed a method to analyse and maintain gas pipelines externally based on Bayesian networks. Buried gas pipelines27 are often exposed to natural calamities, leading to corrosion. To identify the cause of pipeline deterioration, the study used a defective tree prototype and implanted the maintenance plan in the Bayesian network. This approach provides an adequate maintenance plan for pipelines and reduces losses caused by external corrosion.
Jabbari et al.28 used a down-covered logical hierarchy procedure to assess the risk of poisonous gas discharge and outburst in gas pipelines29. An analysis of danger was carried out using MATLAB software, and the mass of each item in basic risk (BRI) was outlined in a reference frame. The study found that five-state norms determine the level of threat. This method is applicable to the security chief when making decisions associated with the hazardous evaluation of a gas pipeline.
Yin et al.30 presented an upgraded quantifiable danger assessment for gas pipelines, considering high-importance areas. Given that most gas discharge incidents in China occur in crowded areas, the study established two models: a failure probability31 model and a risk consequence model. This approach can adequately recognize significant areas and produce reliable outcomes.
Chen et al.32 proposed a danger analysis method for buried pipes based on an upgraded cloud variable weight thesis. The study offered a new way of using the cloud variable weight hypothesis to examine the hazard amount and consequential danger factor of a pipeline by setting up a threat evaluation index system for the pipeline. This method assists the supervisor in determining the threat level and sore point in the pipeline.
Wang et al.33 developed a technique for evaluating the existence of deteriorated pipeline balances and assessing the threat of pipeline defaulter conditions in the face of an unexpected current attack. The study used a Monte Carlo simulation approach to determine the impedance caused by electrochemical reactions and the unreliability of prototype variables. The balance course of the corrosive ecological variables remains remarkably constant, except for the dynamic stray current immersed by the pipeline.
Mederios et al.34 presented a technique to deal with multi-dimensional risks occurring in gas pipelines based on unexpected utility. The study accomplished multi-dimensional risk evaluation of gas pipelines in decision-making and used a non-expected utility method in the MRDU prototype. The deflection of usefulness was surveyed, and this included the benefit from an RDU (rank dependent utility)-based danger proposal. The study conducted similar outcome analysis and sensitivity testing, and this method provides great support to the decision-makers with regard to natural gas pipeline sections.
Ullah et al.35 assessed the rockburst patterns s of the Jinping-II hydropower project in China to enhance employee security in mining and geotechnical works. In the present work, the following methods were used to predict short-term rock burst threat: t-distributed stochastic neighbour embedding (t-SNE), K-means clustering, and extreme gradient boosting (XGBoost). The implied model's results provide an excellent standard to guide future short-term rock burst levels forecasting with excellent precision.
Kamran et al.36 proposed a model to reducing rockburst-related mortality. In this study, firstly,isometric mapping (ISOMAP) algorithm is used.then, ISOMAP was categorized using the fuzzy c-means algorithm (FCM) and at last, in order to predict different levels of the short-term rockburst dataset, KNearest Neighbour (KNN) was used. In the experiment's dataset, the suggested model properly categorised 96% of the rockburst occurrences.
Kamran et al.37 utilized the algorithms like Catboost and light gradient boosting machine (LightGBM) techniques with the objective to reduce the number of casualties and property damage associated with deep underground engineering tasks. Here, Catboost and light gradient boosting machine (LightGBM) techniques to examine several intriguing elements of mine fire statistics. The results show that LightGBM algorithms, having an accuracy of 92% and 89%, respectively, outperformed Catboost in terms of performance.
Kumaran et al.38 introduces a novel approach to foresee the stability of underground coal pillars utilising integrated unsupervised and supervised learning to roughly simulate the complicated behaviour of coal pillars. Kumaran et al.39 proposed a novel approach to forecast mine fire levels using a variety of machine learning approaches.
After examining various literature, pipelines are exposed to various physical and chemical environmental issues. These result in the explosion of chemicals, which lead to the spread of various diseases to humans, fire accidents, and natural calamities like land pollution, landslides, earthquakes, etc. The literature that is examined only with minimal parameters that affect the pipeline and does not contain important parameters to which the pipeline is exposed. In a specific assessment, the need for integrity and qualitative risk assessment of pipelines with various parameters such as corrosion, leakage, coating, materials, atmosphere, surface, earth movements, population, above-ground and underground facilities, and so on is identified. The existing methodologies focused only on the risk of pipelines and assessing that particular risk. This may lead to catastrophic disasters where the pipeline is being implanted. So, an efficient relative risk score (RRS) method is proposed by assessing the pipeline using the RRS method with various parameters that affect the pipelines, which is greatly helpful and safe to transmit materials through the pipelines.
Related terminologies
Pipeline integrity management (PIM)
Pipelines are the best-grounded and cheapest mode for transporting oil, fluids, and natural gas. Pipeline networks2 are large and complicated; they consume time and are often intensive for inquiring about a lot of pipelines. PIM is executed to reduce the possibility of remissness caused by the debasement and to maintain the programmability and security of pipelines. Despite the warnings to avoid transporting natural gas, fluids, and oil, these pipes are subjected to deteriorating conditions. The PIM are used to define the systematic approach, identify the possible risk by way of danger, and apply the preventive methods.
The research focuses on the basic objectives of pipeline integrity management (PIM) to improve confidence among the public in pipeline safety and operator management. It improves the operational processes to maintain the integrity of the pipeline. The primary goal of PIM is to ensure pipeline reliability, prevent incidents, and maintain the operation license. These integration activities generate the data with 3 V’s (a huge amount of volume, velocity, and variety) based on the pipeline’s length and the sensors and tools that are used to access the condition of pipelines. The pipeline should ensure not only the mechanical condition but also the operator's reliable operation, delivery duty, image, and estimation.
Risk assessment
Danger estimation and management is the term used to describe the general method for identifying the danger and the risk factors that have the potential to cause damage. Analyzing and evaluating the problem that is associated with the hazard. The tools used for risk assessment are the risk matrix, decision tree, failure modes and effects analysis. Since the oil, fluids, and natural gas pipes are burnable, they will be dangerous and toxic. The outflow in the pipeline can cause catastrophic effects like fire explosions and environmental pollution. Risk assessments used to reduce risk include baseline, issue-based, and continuous risk assessments.
Pipeline assessment entails closely inspecting pipeline inner and outer sections to determine corrosion rates, flow modeling, and profile calculation. The two components of pipeline risks are leaks and ruptures. The risk in the pipeline40 can be monitored by using supervisory control and data acquisition systems, which collects data about the pipeline operations and transmit the data to engineers or technicians if some problem occurs. The security cameras and sensors, fibre-optic cables, and temperature sensors placed on or near the pipelines can also check for leakages and corrosion to prevent them from becoming dangerous.
Machine learning in PIM
Machine learning41 concentrates on integrity, maintenance, inspection, analysing the crack, and preventing corrosion in pipelines. In pipeline integrity management, the two categories of ML classification and regression are used. Classification is used for detecting leakage, identifying the defect type, and predicting the level of risk in pipelines. Regression is used for calculating the size of the defect and predicting the rate of debasement in pipelines. For pipeline integrity management, clustering identical pipeline42 segments based on identical operating conditions, materials used for establishment, and debasement mechanisms are taken into account for the assessment of risk.
Classifier techniques used in prediction
The Classification algorithm is a Supervised Learning technique utilized in order to categorize new observations, on the premise of training data. In classification, a system makes use of the dataset or observations that are provided to learn how to categorize fresh observations into various classes or groups. In this research, three classifier models such as support vector machine (SVM), Decision Tree and Naïve Bayes are used to identify the chance of pipeline failures, based on a variety of input characteristics.
Support vector machine (SVM)
Collecting and pre-processing the data is the initial stage in the use of SVM for pipeline risk assessment. Data collection on pipeline parameters, environmental variables, and information on previous pipeline failures are necessary.
where y_i is the class for the i-th data point, x_i is the input vector for the i-th data point, b is the bias term.
Decision tree
It functions by creating a tree-like model of choices and potential outcomes. Each node in a decision tree indicates a choice made in response to a particular characteristic or attribute of the data. Up until a stopping requirement is satisfied, the data are recursively divided into subsets based on the values of the characteristics. This terminating criterion can be a predetermined tree depth or the minimal quantity of samples needed in a leaf node. The leaf nodes represent the class labels. By estimating the possibility of a pipeline failure or leak based on different characteristics like the age of the pipeline, the material it is constructed of, the operating pressure, and the location, decision trees can be utilized for pipeline assessment.
The above Eq. 2 calculates how much information a feature provides us about a class
The above Eqs. 2, 3 and 4 calculate entropy and gini index. C also represents the number of events (groups). The entropy and Gini index is a measure of impurity or purity used while creating a decision tree.
Naïve Bayes
This is also using the different characteristics of pipeline to assess. It is possible to identify high-risk pipelines and set priorities for maintenance and inspection work by utilising a Naive Bayes algorithm for pipeline assessment. The Naive Bayes algorithm is excellent for huge datasets since it is easy to use and computationally effective. The Naive Bayes algorithm for pipeline evaluation involves utilising the Bayes theorem to determine the likelihood of a pipeline failure or leak given its characteristics.
Proposed RRS methodology
-
a.
System model of risk assessment of pipelines using machine learning
The research focuses on qualitative risk assessment and the integrity of pipelines using relative risk scoring methods in machine learning. Relative risk scoring is an index model in which important conditions and activities of the pipeline are assigned numerical values (scores) that contribute to risk. Depending on the needs of the assessment, multiple layers of the layered hierarchy in which the relative risk scoring algorithm is designed may be necessary. Failure or consequence factors are investigated. The relative contribution to the risk, consequence, or total risk will determine the best course of action for risk minimization.
The relative risk score (ℜ) is calculated by sum of the Index Sum (µ) divided by the Leak Impact Factor.
Index sum (µ) is the sum of Third-party index (Ɲ), Corrosion Index (ʩ), Design Index (ϑ) and the Incorrect Operation (Ѵ).
-
I.
Third party index
The Third-Party Index (Ɲ) is the summation of the seven factors.
where X1 is the minimum depth cover. In this case, X1 is obtained by dividing the no. of inches covered by three [X1 = A1/3], where A1 is the number of inches covered. X2 is the activity level, X3 is the facilities in aboveground, X4 is the line locating, X5 is the public education, X6 is the Right of the way Condition, X7 is the patrol and n is the number of factors in third party index.
-
II.
Corrosion Index
Corrosion Index (ʩ) is the summation of Atmospheric Corrosion (À), Internal Corrosion (ƛ), and Subsurface Corrosion (₰). Where À is the Atmospheric corrosion, ƛ is the internal corrosion and ₰ is the subsurface corrosion. Atmospheric corrosion (À) is calculated with the summation of three factors.
where Y1 is the atmospheric exposure, Y2 is the atmospheric type and Y3 is the atmospheric coating.
Internal corrosion (ƛ) is calculated with the summation of corrosion in the product (ɥ) and internal production (\(\gamma\)).
Subsurface corrosion (₰) is the summation of subsurface environment (ψ), Cathodic protection (ɰ) and coating ( ).
Subsurface environment (ψ) is the summation of corrosion in soil (ϒ) and mechanical corrosion (П).
Cathodic protection (ɰ) is the summation of effectiveness () and interference potential ().
Coating ( ) is the summation of fitness (Ʊ) and condition (ȡ).
where ʩ is the corrosion index.
-
III.
Design Index
Design index (ϑ) is the summation of safe factor (D1), fatigue (D2), surge potential (D3), verification of integrity (D4), land movements (D5).
where ϑ is the Design index, n is the number of factors, D1 is the safe factor, D2 is the fatigue, D3 is the surge potential, D4 is the verification of integrity, D5 is the land movements.
-
IV.
In-correct Operation Index
The In-correct Operation Index (Ѵ) is the sum of design (ɠ), construction (ɓ), operation (τ), and maintenance (ω).
where ɠ denotes the design, n denotes the number factors, \({M}_{1}\) is the Hazard identification, \({M}_{2}\) is the MAOP potential, \({M}_{3}\) is the safety system, \({M}_{4}\) is the material selection and \({M}_{5}\) is the checks.
where ɓ denotes the construction, n denotes the number factors, \({N}_{1}\) is the inspection value, \({N}_{2}\) is the Materials rating, \({N}_{3}\) is the joining value, \({N}_{4}\) is the backfills, \({N}_{5}\) is the handling and \({N}_{6}\) is the coating of pipelines.
where τ denotes the operation, n denotes the number factors, \({O}_{1}\) is the procedure, \({O}_{2}\) is communication test, \({O}_{3}\) is drug testing value, \({O}_{4}\) is the safety program, \({O}_{5}\) is the survey/maps/record, \({O}_{6}\) is the training and \({O}_{7}\) is the mechanical error preventer.
where ω denotes the Maintenance, n denotes the number factors, \({P}_{1}\) is the documentation, \({P}_{2}\) is the schedule and \({P}_{3}\) is the procedure.
From above Eqs. (15), (16), (17) and (18),
From Eqs. (6), (13), (14) and (19).
-
I.
Product hazard
Product hazard is calculated by the summation of acute hazard and chronic hazard.
where δ denotes the acute hazard, η is the reactivity, ƕ is the flammability and ؏ is the toxicity.
where denotes the Product hazard, δ is the acute hazard, ζ is the chronic hazard.
-
II.
Leak volume
Leak volume (Final spill score) = (effective score of spill size) × (adjustment factor larger openings).
where £ denotes the Leak volume, ξ is the score of the spill size and is the Adjustment factor larger openings.
-
III.
Dispersion
Dispersion is calculated by Operating pressure divided by 100.
where Ә denotes the dispersion and denotes Operating pressure.
-
IV.
Receptors
Receptor (ɮ) is the summation of population density (ρ), environment considerations () and high value areas (ℏ).
where ɮ denotes Receptors, ρ is the population density, is the Environmental considerations and ℏ is the High value areas.
From Eqs. (23), (24), (25) and (26),
where denotes the Product hazard, £ is the Leak volume, Ә is the Dispersion and ɮ is the Receptors.
-
b.
Architecture and work flow of risk assessment of pipelines using machine learning
Figure 1 presents the complete flow of the system. The system first analyses various parameters that are needed to calculate the risk score of pipelines. Using that the Third-party index, corrosion index, Design index are calculated. Finally, with the index sum and the leakage impact factor the risk level is decided. Table 1 displays the Symbols and semantics used for experimental purpose.
Experimentation and analysis
Experimental setup
This research concentrates on the quality and assesses the possible dangers of the pipeline. The assessment of the pipeline requires various factors, like environmental conditions, land movements, and other factors. The RRS methodology focuses on every factor that affects the pipeline and assesses the risk. Here, the relative risk score method is calculated by dividing the index sum by the leak impact factor. The dataset analysed for this research is published in Kaggle platform43 to collaborate with colab. Table 2 contains the risk level for the range of values of the relative risk score.
The above-mentioned Table 2 is the range from which the risk level of pipeline is calculated.
Results and discussion
The RRS is obtained by dividing the index sum by the leak impact factor. The index sum is calculated as the sum of the corrosion index, third-party index, design index, and in-correct operation index. The leak impact factor is calculated by the product of leak volume, product hazard, dispersion, and receptors. The risk level is marked based on the relative risk score range as depicted in Table 2. Table 3 depicts the statistical characteristics of the research data. Table 4 shows a sample of the calculated index sum, leak impact factor, relative risk score, and risk level.
Figure 2 depicts the range in which the values of the index sum lie. Here, the x-axis represents the values from the dataset, and the y-axis represents the ranges. The index sum values consist of corrosion index, third-party index, design index, and in-correct operation. All these four factors have several subfactors, in which the values are added and the final value is given as the index sum.
Figure 3 depicts the range in which the values of the leak impact factor lie. Here, the x-axis represents the values from the dataset, and the y-axis represents the ranges. The leak impact factor values consist of product hazards, leak volume, dispersion, and receptors. Each of these four factors has several subfactors, the values are multiplied and the final value given as the leak impact factor.
Figure 4 depicts the observations on a single attribute, which is univariate, and visualizes it through a histogram, i.e., only one observation. Here, the relative risk score is observed. The relative risk score is calculated by dividing the index sum by the leak impact factor.
Figure 5 also depicts the observations on a single attribute which is univariate and visualizes it through a histogram. Here, the risk of the pipeline is observed. From the relative risk score method, the risk of the pipeline is calculated as high, low, or medium based on the input values.
Figure 6 shows the relative risk score’s box plot. The main objective of the box plot is to find the minimum value, maximum value, and outliers. Outlier detection is the process of identifying an unknown observation in a given dataset.
Figure 7 depicts the comparison of various methodologies, i.e., support vector machines, decision trees, and naive bayes algorithms, with the RRS methodology. These algorithms are compared based on the execution time. In this way the RRS method shows the better execution time (24 ms) while comparing the other algorithms. This comparison proves that the submitted methodology is better than the existing methodologies in terms of performance.
In Table 5 the performance metrics like accuracy, precision, recall and F1 score of RRS methodology are discussed.
The above Fig. 8 depicts the graph for identifying the risk due to corrosion. The x-axis represents the corrosion in different algorithms, and the y-axis represents the accuracy in percentage. The percentage of pipeline corrosion is compared here using various algorithms such as Naive Bayes, support vector machine (SVM), and graph convolutional network (GCN). The SVM shows 92%, the Naive Bayes algorithm shows 92.7%, the GCN algorithm shows 93%, and the RRS method shows the greatest accuracy percentage of 97.5%. When comparing these three algorithms, the Relative Risk Scoring method shows a high percentage of accuracy in predicting corrosion. Based on this analysis, the graph proves that the RRS method is the best method for calculating corrosion accuracy.
The Fig. 9 depicts a graph for identifying the risk due to leakage. The x-axis represents the leakage in different algorithms, and the y-axis represents the accuracy in percentage. Support vector machine (SVM), the Naive Bayes algorithm, and the graph convolutional. The SVM shows 95%, Naive Bayes algorithm shows 94%, GCN algorithm shows 93%, and RRS method shows the greatest accuracy percentage of 97.5%. When these three algorithms are compared, the Relative Risk Scoring method has the highest percentage accuracy in detecting leakage. Based on this analysis, the graph shows that the RRS method is the most effective method for calculating leakage accuracy. From above Figs. 8 and 9, it is proven that SVM and GCN are individually better in the aspect of predicting leakage and corrosion but, RRS is better than them in both aspects.
The methodology was carefully designed and tested to ensure that it is reliable. The RRS methodology has achieved only 93% of recall which can be improved with further addition of experiments. The proposed methodology has several advantages, the risk of the pipeline is predicted using various parameters including corrosion, leakage and other damages so that the risk of the pipeline is not over calculated. Thus, the leakage, corrosion and other damages are not predicted separately. Hence, this method is more accurate to calculate the risk of the pipeline than other methodologies.
Conclusion
In conclusion, pipelines serve as the backbone of the global transportation network for products such as oil, water, and gas. However, the safe and efficient operation of pipelines requires comprehensive risk assessment. Our study presents the RRS method, a novel approach to risk assessment that considers a wide range of parameters, including those often overlooked in existing methodologies. The RRS algorithm provides more accurate results for calculating leakage, corrosion, and classification, with accuracies of 96.5%, 94.7%, and 94.3%, respectively. Compared to the Decision Tree algorithm, the RRS method executes much faster, reducing the time and cost associated with risk assessment. Overall, the RRS method represents a significant advancement in pipeline risk assessment. Its ability to provide more accurate results, faster execution time, and comprehensive consideration of all parameters makes it a more reliable and efficient approach. The RRS method has the potential to improve the safety and efficiency of pipeline operation in the future, providing a safer, more cost-effective, and more sustainable approach to transporting products across the globe. Thus, we conclude that the RRS method is a reliable and efficient approach for risk assessment of pipelines.
Data availability
The datasets generated and/or analysed during the current study are available in the Kaggle43 repository, [https://www.kaggle.com/datasets/vanitham20bsr059/oilchemical-pipeline-dataset].
References
Shaik, N. B. et al. Recurrent neural network-based model for estimating the life condition of a dry gas pipeline. Process Saf. Environ. Prot. 164, 639–650 (2022).
Aryai, V., Baji, H. & Mahmoodian, M. Failure assessment of corrosion affected pipeline networks with limited failure data availability. Process Saf. Environ. Prot. 157, 306–319 (2022).
Li, X., Zhang, Y., Abbassi, R., Khan, F. & Chen, G. Probabilistic fatigue failure assessment of free spanning subsea pipeline using dynamic Bayesian network. Ocean Eng. 234, 109323 (2021).
Seghier, M. E. A. B., Höche, D. & Zheludkevich, M. Prediction of the internal corrosion rate for oil and gas pipeline: Implementation of ensemble learning techniques. J. Nat. Gas Sci. Eng. 99, 104425 (2022).
Teng, M. C. & Ke, S. S. Disaster impact assessment of the underground hazardous materials pipeline. J. Loss Prev. Process Ind. 71, 104486 (2021).
Spandonidis, C., Theodoropoulos, P., Giannopoulos, F., Galiatsatos, N. & Petsa, A. Evaluation of deep learning approaches for oil & gas pipeline leak detection using wireless sensor networks. Eng. Appl. Artif. Intell. 113, 104890 (2022).
Sohaib, M., Islam, M., Kim, J., Jeon, D. C. & Kim, J. M. Leakage detection of a spherical water storage tank in a chemical industry using acoustic emissions. Appl. Sci. 9(1), 196 (2019).
Mazumder, R. K., Salman, A. M. & Li, Y. Failure risk analysis of pipelines using data-driven machine learning algorithms. Struct. Saf. 89, 102047 (2021).
Yang, Y., Li, S. & Zhang, P. Data-driven accident consequence assessment on urban gas pipeline network based on machine learning. Reliab. Eng. Syst. Saf. 219, 108216 (2022).
Liu, Y. & Bao, Y. Review on automated condition assessment of pipelines with machine learning. Adv. Eng. Inform. 53, 101687 (2022).
Xu, H., Jiao, Z., Zhang, Z., Huffman, M. & Wang, Q. Prediction of methane hydrate formation conditions in salt water using machine learning algorithms. Comput. Chem. Eng. 151, 107358 (2021).
Wu, X. et al. FTAP: Feature transferring autonomous machine learning pipeline. Inf. Sci. 593, 385–397 (2022).
He, B. et al. Risk assessment of pipeline engineering geological disaster based on GIS and WOE-GA-BP models. Appl. Sci. 11(21), 9919 (2021).
Malinowska, A., Cui, X., Salmi, E. F. & Hejmanowski, R. A novel fuzzy approach to gas pipeline risk assessment under influence of ground movement. Int. J. Coal Sci. Technol. 9(1), 1–11 (2022).
Kraidi, L., Shah, R., Matipa, W. & Borthwick, F. An investigation of mitigating the safety and security risks allied with oil and gas pipeline projects. J. Pipeline Sci. Eng. 1(3), 349–359 (2021).
Su, Y., Li, J., Yu, B., Zhao, Y. & Yao, J. Fast and accurate prediction of failure pressure of oil and gas defective pipelines using the deep learning model. Reliab. Eng. Syst. Saf. 216, 108016 (2021).
Bu, F. et al. Real scenario analysis of buried natural gas pipeline leakage based on soil-atmosphere coupling. Int. J. Press. Vessels Pip. 199, 104713 (2022).
Liu, A., Chen, K., Huang, X., Li, D. & Zhang, X. Dynamic risk assessment model of buried gas pipelines based on system dynamics. Reliab. Eng. Syst. Saf. 208, 107326 (2021).
Adumene, S., Khan, F., Adedigba, S., Zendehboudi, S. & Shiri, H. Offshore pipeline integrity assessment considering material and parametric uncertainty. J. Pipeline Sci. Eng. 1(3), 265–276 (2021).
Medeiros, C. P., da Silva, L. B. L., Alencar, M. H. & de Almeida, A. T. A new method for managing multidimensional risks in natural gas pipelines based on non-expected utility. Reliab. Eng. Syst. Saf. 214, 107709 (2021).
Aalirezaei, A. & Kabir, G. A bibliometric analysis on oil and gas pipeline failure consequence analysis. Innov. Infrastruct. Solut. 6(4), 1–18 (2021).
Iqbal, H., Haider, H., Waheed, B., Tesfamariam, S. & Sadiq, R. Benchmarking of oil and gas pipeline companies in British Columbia: Integrating integrity management program and safety culture using a risk-based approach. Eng. Manag. J. 34, 526–542 (2021).
Froeling, H. A. J., Dröge, M. T., Nane, G. F. & Van Wijk, A. J. Quantitative risk analysis of a hazardous jet fire event for hydrogen transport in natural gas transmission pipelines. Int. J. Hydrog. Energy 46(17), 10411–10422 (2021).
Taleb-Berrouane, M., Khan, F. & Hawboldt, K. Corrosion risk assessment using adaptive bow-tie (ABT) analysis. Reliab. Eng. Syst. Saf. 214, 107731 (2021).
Wang, W. et al. Vulnerability analysis method based on risk assessment for gas transmission capabilities of natural gas pipeline networks. Reliab. Eng. Syst. Saf. 218, 108150 (2022).
Zulkifli, F. S. & Salleh, Z. M. Effect on flow and pipe structure behaviours from different pressure and pipe thickness for pipeline in UTHM biodiesel plant. Prog. Eng. Appl. Technol. 3(1), 902–914 (2022).
Chen, C., Li, C., Reniers, G. & Yang, F. Safety and security of oil and gas pipeline transportation: A systematic analysis of research trends and future needs using WoS. J. Clean. Prod. 279, 123583 (2021).
Li, Y. T., He, X. N. & Shuai, J. Risk analysis and maintenance decision making of natural gas pipelines with external corrosion based on Bayesian network. Pet. Sci. 19(3), 1250–1261 (2022).
Shi, Y. et al. A safety management system for natural gas pipeline in subsidence area of coal mine. Energy Sources Part A Recov. Util. Environ. Effects 44(3), 5766–5783 (2022).
Jabbari, M., Gholamnia, R., Esmaeili, R., Kouhpaee, H. & Pourtaghi, G. Risk assessment of fire, explosion and release of toxic gas of Siri-Assalouyeh sour gas pipeline using fuzzy analytical hierarchy process. Heliyon 7(8), e07835 (2021).
González-Arévalo, N. E. et al. Influence of aging steel on pipeline burst pressure prediction and its impact on failure probability estimation. Eng. Fail. Anal. 120, 104950 (2021).
Yin, Y. et al. Improved quantitative risk assessment of a natural gas pipeline considering high-consequence areas. Reliab. Eng. Syst. Saf. 225, 108583 (2022).
Chen, Y., Xie, S. & Tian, Z. Risk assessment of buried gas pipelines based on improved cloud-variable weight theory. Reliab. Eng. Syst. Saf. 221, 108374 (2022).
Wang, C., Li, W. & Wang, Y. Remaining lifetime assessment of gas pipelines subjected to stray current interference using an integrated electric-electrochemical method. Eng. Fail. Anal. 127, 105494 (2021).
Ullah, B., Kamran, M. & Rui, Y. Predictive modeling of short-term rockburst for the stability of subsurface structures using machine learning approaches: T-SNE, K-Means clustering and XGBoost. Mathematics 10(3), 449 (2022).
Kamran, M., Ullah, B., Ahmad, M. & Sabri, M. M. S. Application of KNN-based isometric mapping and fuzzy c-means algorithm to predict short-term rockburst risk in deep underground projects. Front. Public Health 10, 1023890 (2022).
Kamran, M. et al. Intelligent based decision-making strategy to predict fire intensity in subsurface engineering environments. Process Saf. Environ. Prot. 171, 374–384 (2023).
Kamran, M., Shahani, N. M. & Armaghani, D. J. Decision support system for underground coal pillar stability using unsupervised and supervised machine learning approaches. Geomech. Eng. 30(2), 107 (2022).
Kamran, M. & Shahani, N. M. Decision support system for the prediction of mine fire levels in underground coal mining using machine learning approaches. Min. Metall. Explor. 39(2), 591–601 (2022).
Guo, X., Stoesser, T., Nian, T., Jia, Y. & Liu, X. Effect of pipeline surface roughness on peak impact forces caused by hydrodynamic submarine mudflow. Ocean Eng. 243, 110184 (2022).
Phan, H. C. & Dhar, A. S. Predicting pipeline burst pressures with machine learning models. Int. J. Press. Vessels Pip. 191, 104384 (2021).
Shi, G. et al. Time-dependent economic risk analysis of the natural gas transmission pipeline system. Process Saf. Environ. Prot. 146, 432–440 (2021).
KAGGLE https://www.kaggle.com/datasets/vanitham20bsr059/oilchemical-pipeline-dataset.
Funding
This work was supported the Korea Environmental Industry & Technology Institute (KEITI), with a grant funded by the Korea government, Ministry of Environment (The development of IoT-based technology for collecting and managing big data on environmental hazards and health effects), under Grant RE202101551 and partially supported by the Institute of Information and Communications Technology Planning and Evaluation (IITP) funded by the Korea Government, Ministry of Science and ICT(MSIT) (Building a Digital Open Lab as open innovation platform) under Grant 2021-0-00546. This paper was supported by research funds for newly appointed professors of Jeonbuk National University in 2022.
Author information
Authors and Affiliations
Contributions
Conceptualization, S.V.E.; C.N.V. and S.A.K.; methodology, J.C; S.V.E.; C.N.V.; software, S.A.K.; validation, S.V.E.; C.N.V and J.C.; formal analysis, S.V.E. and C.N.V.; investigation, C.N.V; writing—original draft preparation, S.V.E.; C.N.V. and S.A.K; writing—review and editing, S.V.E.; C.N.V. and S.A.K; supervision, J.C.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Vanitha, C.N., Easwaramoorthy, S.V., Krishna, S.A. et al. Efficient qualitative risk assessment of pipelines using relative risk score based on machine learning. Sci Rep 13, 14918 (2023). https://doi.org/10.1038/s41598-023-38950-9
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-023-38950-9
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.