⚡ electrical engineering

Integrity Assessment of Buried Pipelines under Surcharge Loading Based on a GA-BP Neural Network

This study develops a GA-BP neural network model, validated by a 3D finite element simulation, to rapidly assess the integrity of buried natural gas pipelines under surcharge loading and successfully identified a potential failure risk in an in-service pipeline within the Penglai Dengying Block.

Original authors: Lin Chen, Yabo Lü, Liyuan Guo, Jiaqiang Ou, Wu Liu, Hongtao Xiao

Published 2026-10-21

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Lin Chen, Yabo Lü, Liyuan Guo, Jiaqiang Ou, Wu Liu, Hongtao Xiao

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Beneath the bustling surface of modern cities and the quiet expanses of rural landscapes, a vast network of steel arteries carries the natural gas that heats homes and powers industries. These pipelines are buried deep enough to avoid accidental damage, yet they are not immune to the weight of the world above them. When construction crews dig trenches or park heavy machinery near these lines, they create a temporary but intense pressure known as surcharge loading. This extra weight pushes down on the soil, which in turn squeezes the pipe. If the pressure becomes too great, the metal can stretch beyond its limits, leading to leaks or catastrophic failure. For engineers, the challenge has long been predicting exactly how much weight a specific pipe can handle before it breaks, a task that usually requires complex, time-consuming computer simulations that are difficult to run quickly in the field.

A team of researchers from PetroChina Southwest Oil & Gasfield Company and Southwest Petroleum University has developed a new way to solve this problem, combining advanced computer modeling with a type of artificial intelligence to create a rapid safety check. They began by building a detailed digital replica of a buried natural gas pipeline and the soil surrounding it. Using powerful software, they simulated how the ground and the pipe interact when heavy loads are placed on the surface above. The model accounted for the fact that soil is not a rigid block but a material that shifts and deforms, pressing against the steel in complex ways. By testing thousands of variations in their computer, they discovered that the stress on the pipe is not uniform; it concentrates heavily at the bottom of the pipe directly beneath the heavy load, creating a specific zone of danger that shifts depending on how deep the pipe is buried and how much internal pressure the gas exerts.

To turn these complex findings into a tool that engineers could use instantly, the researchers trained a specialized computer program to recognize patterns in the data. Instead of running a full, slow simulation every time a new construction site appeared, they taught this program to look at just a few key details—the size of the pipe, how deep it is buried, the pressure of the gas inside, and the weight of the load above—and then predict the maximum stress the pipe would experience. The program learned from one hundred carefully generated examples, refining its ability to guess the outcome with high accuracy. The result was a system that could predict the stress levels with a degree of certainty that closely matched the slow, detailed computer models, but in a fraction of the time.

The team then put this new method to the test on three real, active pipelines in the Penglai Dengying Block. They fed the specific conditions of each site into their prediction model to see if the pipes were safe under potential construction loads. Two of the pipelines showed stress levels well within the safe limits, meaning they could withstand the expected pressure from nearby construction. However, the third pipeline told a different story. The model predicted that under certain conditions, specifically when the internal gas pressure was high and a heavy load was placed directly above, the stress on the pipe would reach 219.11 megapascals. This number exceeded the safety limit of 216 megapascals for that specific pipe, indicating a genuine risk of failure. The analysis revealed that this danger was primarily due to the pipe being buried only one meter deep, which left it with very little soil to cushion the blow of the heavy load.

This discovery highlights a critical vulnerability in pipeline safety: the depth of burial is just as important as the strength of the steel. For the third pipeline, the researchers concluded that construction crews must strictly limit the weight of any equipment or soil piles placed nearby, and additional reinforcement measures should be installed to protect the line. The study demonstrates that by combining detailed physics simulations with smart prediction tools, engineers can move from guessing about risks to knowing them with precision. This approach allows for the rapid assessment of pipeline integrity, ensuring that the energy flowing beneath our feet remains secure even as the world above continues to build and change.

Technical Summary: Integrity Assessment of Buried Pipelines under Surcharge Loading Based on a GA-BP Neural Network

Problem Statement
The rapid expansion of natural gas pipeline networks, driven by global urbanization and energy demands, has increased the frequency of pipelines being subjected to construction surcharge loading. These loads, arising from temporary soil stockpiles and heavy machinery operation near buried pipelines, have become a significant cause of pipeline leakage and failure. While theoretical models and finite element analysis (FEA) have been used to study pipe-soil interactions, existing methods often rely on simplifying assumptions regarding loading conditions and contact mechanics, or they are computationally too intensive for rapid integrity assessments at construction sites. Consequently, there is a need for a method that accurately characterizes the mechanical response of pipelines under surcharge loading while enabling rapid risk identification.

Methodology
The study employs a hybrid approach combining high-fidelity numerical simulation with machine learning:

Finite Element Modeling (FEM): A three-dimensional ABAQUS model was developed to simulate a buried natural gas pipeline (L360 steel) interacting with surrounding soil.
- Constitutive Models: The pipeline material was modeled using the Ramberg–Osgood model to capture nonlinear plastic behavior, while the soil was modeled using the Drucker–Prager elastoplastic yield criterion.
- Pipe-Soil Interaction: Nonlinear contact was simulated using surface-to-surface contact with a "hard" contact formulation in the normal direction and a penalty formulation with a friction coefficient of 0.3 in the tangential direction.
- Validation: The model was validated through mesh-independence analysis (selecting 824,600 elements) and comparison with experimental data from literature, yielding a mean error of 6.32%.
- Mechanical Analysis: The model analyzed stress distributions under various conditions, revealing that maximum von Mises stress typically occurs at the pipe bottom within the surcharge-affected region.
Machine Learning Prediction (GA-BP): To overcome the computational cost of repeated FEA runs, a Genetic Algorithm-Backpropagation (GA-BP) neural network was developed.
- Data Generation: 100 simulation samples were generated using Latin hypercube sampling, varying five input parameters: pipeline diameter, surcharge position, internal pressure, burial depth, and surcharge pressure.
- Model Architecture: The GA was used to optimize the initial weights and thresholds of the BP neural network to avoid local optima. The final architecture utilized a single hidden layer with seven neurons.
- Training: 80% of the data was used for training, and 20% for testing. The model's performance was evaluated against Support Vector Machine (SVM) and other algorithms.
Integrity Assessment Framework: The prediction model was integrated with the strength design criteria from GB 50251-2015 (Code for Design of Gas Transmission Pipeline Engineering). A failure criterion based on the von Mises yield theory was established to determine if the predicted maximum stress exceeded the allowable stress ( $\sigma_{Allow}$ ).

Key Contributions

Nonlinear Coupled Model: Establishment of a 3D finite element model incorporating nonlinear pipe-soil contact and elastoplastic soil behavior to characterize von Mises stress distribution under surcharge loading.
Hybrid Prediction Model: Development of a GA-BP neural network trained on simulation data to rapidly and accurately predict maximum pipeline stress, achieving a coefficient of determination ( $R^2$ ) of 0.9691.
Engineering Assessment Framework: Creation of a practical integrity assessment procedure that integrates the prediction model with regulatory design factors, enabling the generation of 3D stress contour maps for rapid risk evaluation.

Results

Stress Distribution: The FEA results indicated pronounced stress variations within the surcharge-loaded region (17.5–32.5 m). The stress distribution exhibited a symmetric double-peak pattern at the pipe bottom and crown, with the maximum stress consistently occurring at the pipe bottom.
Model Performance: The GA-BP model demonstrated superior predictive accuracy compared to the SVM model, with a Root Mean Square Error (RMSE) of 10.5775 and an $R^2$ of 0.9691. The fitness value converged to approximately 0.035 after 42 generations.
Case Study Application: The framework was applied to three in-service pipelines in the Penglai Dengying Block:
- Pipelines No. 1 and No. 2: Under surcharge pressures of 0–0.4 MPa and internal pressures of 2–8 MPa, the predicted maximum stresses (171.06 MPa and 137.05 MPa, respectively) remained below the allowable limit of 216 MPa.
- Pipeline No. 3: Under the same loading conditions, the predicted peak stress reached 219.11 MPa, exceeding the allowable limit of 216 MPa. This risk was attributed to the pipeline's shallow burial depth (1 m) and large diameter (406.4 mm).

Significance and Claims
The paper claims to provide a solution for the "rapid integrity assessment" of buried pipelines subjected to construction surcharge loading, a gap identified in previous literature. By coupling ABAQUS simulations with a GA-BP neural network, the study offers a method that balances the accuracy of finite element analysis with the speed required for engineering decision-making. The successful application to real-world pipelines in the Penglai Dengying Block demonstrates the framework's utility in identifying specific failure risks (as seen with Pipeline No. 3) and guiding engineering decisions, such as restricting surcharge loading and implementing reinforcement measures. The study concludes that this integrated approach enables the rapid identification of construction-related surcharge risks without the need for time-consuming, repeated numerical simulations.

More like this