New predictive methodology for severity assessment of corrosion and gouge defects in pipelines

Abstract

A major challenge in pipeline integrity is the absence of a standard method to directly compare corrosion severity with mechanical damage like gouges. This work fills that gap with a new machine learning-based approach. We developed A-Tanh, an adaptive activation function with a learnable slope that enables a simple, efficient model to reach near-perfect accuracy (R2 > 0.99) even with complex, varied data. These results are presented in a severity matrix, offering the first direct comparison of how dangerous different corrosion and gouge defects are relative to one another. Beyond confirming known risks, the model identifies critical patterns, such as how circumferential gouges become more dangerous as defect length increases. Ultimately, this tool offers a practical way for operators to move from analyzing single defects to implementing a smarter, risk-based maintenance plan. • Unified ANN model predicts burst pressure for corrosion and gouges • Adaptive A-Tanh function achieves prediction accuracy with R² > 0.99 • Novel severity matrix ranks defect criticality for decision support • Analysis confirms gouges pose higher risks than corrosion defects