New Open-Source Tool Automates Simulation of Pipeline Pitting Corrosion Monitoring

Pipelines carrying oil, gas, and chemicals form the hidden arteries of modern industry, and their failure can trigger environmental disasters, supply disruptions, and enormous financial losses. Among the many threats to pipeline integrity, localized pitting corrosion stands out as one of the most dangerous: tiny, deep cavities eat through the metal wall while the surrounding surface remains largely intact, making them notoriously difficult to detect. A research team led by Tao Jiang and Yan Li has now unveiled an open-source software tool, VDMM-BESP, published in the journal SoftwareX, that promises to transform how engineers simulate, analyze, and ultimately predict this insidious form of corrosion damage.

The tool targets a monitoring technique known as Voltage Difference Matrix Mapping, or VDMM, which is also widely recognized as the field signature method. The principle is elegant: an array of electrodes is attached to the outer surface of a pipeline, a controlled excitation current is injected, and the voltage differences between adjacent electrode pairs are measured continuously. Because corrosion pits distort the flow of current inside the metal wall, they leave characteristic fingerprints in these voltage patterns. By tracking how the voltage signature evolves over time, operators can monitor wall-thickness changes and detect corrosion defects without ever penetrating or interrupting the pipeline. The technique has found applications in buried pipelines, refineries, chemical plants, and even steel bridge structures, where fatigue crack monitoring benefits from the same physics.

Despite its proven value, the VDMM technique has long suffered from a practical bottleneck. How sensitive the measurements are to a hidden pit depends on a tangle of interacting factors: the geometry of the pipeline, the layout of the electrode matrix, the magnitude of the excitation current, and the size and shape of the defect itself. Exploring that parameter space with physical experiments is prohibitively expensive and slow. Researchers have therefore turned to general-purpose finite element software, but this approach demands tedious manual modeling, repeated parameter edits, careful extraction of electrode potentials, and hand-crafted feature calculations. Running hundreds or thousands of simulated scenarios this way is nearly impossible, and the resulting data rarely feed into a coherent analysis or inversion pipeline.

VDMM-BESP, built in MATLAB and released under the MIT license on GitHub, closes that gap by coupling COMSOL Multiphysics, a commercial finite element solver, with MATLAB-based data processing, visualization, and machine learning. The software requires MATLAB R2021a or later, COMSOL Multiphysics 6.2 or later with the LiveLink for MATLAB module, and optionally MySQL 8.0 for large-scale data storage. Its graphical user interface, developed with MATLAB App Designer, organizes the workflow into three modules: batch electric field simulation, data analysis, and pitting parameter inverse model construction. The modules can be chained into a complete pipeline from parameter setup to exported prediction model, or used independently depending on the research need.

The physics underlying the simulation module rests on the uniqueness theorem of electrostatics. The pipe is modeled as a homogeneous, isotropic conductor with constant electrical conductivity, and the potential distribution inside the metal satisfies the Laplace equation. Boundary conditions enforce current continuity at the injection and extraction electrodes, while the interfaces with internal and external media are treated as ideal insulators so that no current leaks away. Pitting defects are simplified to single, isolated semi-ellipsoidal cavities. The solver uses a conjugate gradient iterative scheme with a residual tolerance of 0.01 and a physics-controlled free tetrahedral mesh, with quadratic shape functions for the potential field. A mesh convergence study showed that the primary feature-value components computed on the default extremely fine mesh differ from those on the finest mesh by less than one percent, giving users confidence in the numerical stability of the results.

What makes the tool genuinely powerful is its batch capability. Users can customize four categories of parameters: pipeline geometry including length, radius, and wall thickness; pitting morphology defined by the radius and depth of the semi-ellipsoidal defect; electrical parameters such as material conductivity and excitation current; and the electrode matrix layout including rows, columns, and spacing. Every parameter can be fixed or swept across a range, and the software offers both a full-scan mode that computes the Cartesian product of all values and a custom combination mode for user-specified sets. For each configuration, the program builds the model, applies the parameters, solves the field, extracts potentials at the electrode positions, and computes the feature coefficient matrix, a normalized measure of the relative change in local electric field caused by the pit. In a demonstration run, 750 simulation models were computed and six electrode matrix configurations were extracted from each, yielding 4,500 samples. The authors estimate that the same work done manually would have taken 52 hours and 30 minutes; the software completed it in 7 hours and 46 minutes, a reduction of roughly 85.2 percent. Robust error handling allows unattended operation: if an extreme configuration fails during meshing or solving, the case is skipped, logged, and the batch continues.

Validation of the forward model lends credibility to the entire workflow. The team compared simulated feature values against published experimental data for square pits, using identical physical settings and extraction procedures. Across 25 data points, the mean relative error was 5.83 percent and the maximum was 13.57 percent, with the simulations reproducing the reported order of magnitude, spatial distribution, and trends with defect size. The validation script is publicly available in the repository, allowing independent scrutiny. Within the software itself, single-model runs produce numerical tables, three-dimensional bar charts of the feature distribution, pseudocolor plots of potential, and streamline plots of current density, and the feature values peak sharply in the pitted region, exactly as the physics predicts.

The data analysis module then turns raw simulation output into insight. Users can load results from Excel files, MATLAB data files, a MySQL database, or the most recent batch run, and filter them by parameter ranges. Pearson and Spearman correlation analyses quantify which input parameters most strongly influence the measured features, displayed as both a numerical coefficient matrix and an intuitive heatmap. Univariate and multivariate fitting tools follow, supporting linear and polynomial fits with optional logarithmic or reciprocal preprocessing, and bivariate models rendered as three-dimensional scatter plots with fitted surfaces. This statistical layer helps researchers identify the most sensitive feature variables before committing to expensive modeling, effectively guiding the design of voltage matrix imaging schemes.

The final module tackles the inverse problem: reconstructing pit dimensions from measured voltage features. Four algorithms are built in: backpropagation neural networks, random forests, gradient boosting decision trees, and simultaneous equations regression, in which a system of regression equations jointly maps feature values to unknown pitting parameters. Users control input features, prediction targets, hyperparameters, and the train-test split, and the tool reports the coefficient of determination, root mean square error, and mean absolute error alongside scatter plots of predicted versus true values. A particularly thoughtful feature is configurable Gaussian noise injection: users can specify a voltage noise standard deviation matching real acquisition conditions, and the software perturbs the simulated voltages and recalculates the features accordingly. This allows noise-matched models to be trained and evaluated, so that when field measurements from a compatible electrode configuration are available, the trained models can estimate pitting parameters directly and inform maintenance planning.

The implications reach well beyond the laboratory. Field technicians can compare electrode layouts and excitation schemes at the design stage without any finite element expertise, balancing detection accuracy against hardware cost. Educators can use the visual interface to teach the complete chain from electrostatic simulation to machine learning inversion. The authors are candid about limitations: the model assumes constant conductivity and ideal insulation, ignores material heterogeneity and contact impedance, and considers only isolated single pits, so results cannot be extrapolated to interacting multiple defects. Future work will extend the models to multi-defect coupling and complex geometries, and integrate deep neural network or Gaussian process surrogate models to replace repeated finite element solves, enabling rapid on-site evaluation. For now, VDMM-BESP stands as a rare example of a fully open, end-to-end research tool that turns a decades-old monitoring technique into a data-driven, machine-learning-ready platform for fighting one of industry’s most corrosive enemies.

Subject of Research: A software tool for batch finite element simulation and data-driven inversion of voltage difference matrix mapping in pipeline pitting corrosion monitoring

Article Title: VDMM-BESP: A MATLAB-based tool for batch simulation and data analysis of voltage difference matrix mapping in pipeline pitting monitoring

Article References: Jiang, T., Yao, W., Wang, S., Zhai, Y., Li, D., Wang, H., & Li, Y. (2026). VDMM-BESP: A MATLAB-based tool for batch simulation and data analysis of voltage difference matrix mapping in pipeline pitting monitoring. SoftwareX, 36, Article 103120. https://doi.org/10.1016/j.softx.2026.103120

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103120

Keywords: pipeline corrosion, pitting corrosion, field signature method, voltage difference matrix mapping, finite element simulation, MATLAB, COMSOL Multiphysics, machine learning, inverse modeling, non-destructive testing, open-source software, structural health monitoring