Competing multimode-informed reliability evolution of blended-hydrogen natural gas pipelines with crack-in-corrosion defects

This study presents a probabilistic modeling of a hybrid machine learning (ML) model using the Latin Hypercube Sampling-Monte Carlo Simulation (LHS-MCS) method to implement the competing multimode-informed reliability evolution of blended hydrogen-natural gas (BHNG) pipelines with crack-in-corrosion (CIC) defects. The samples generated by finite element (FE) simulation form a dataset for training and testing. The Gaussian Process Regression (GPR) model is optimized by integrating Circle chaotic mapping with Grey Wolf Optimization (GWO). Competitive failure modes are embedded in advanced reliability algorithms to perform probabilistic evaluations. The results indicate that the CIGWO algorithm significantly improves the prediction accuracy and global search capability of GPR. The CIGWO-GPR model implements an R² value of 0.9897 for corrosion-dominant failure pressure (Pcorrosion) and 0.9939 for crack-dominant failure pressure (Pcrack). At high hydrogen blending ratios, hydrogen-induced damage (HID) accelerates performance degradation, causing the pipelines to evolve unpredictably, driven by non-equilibrium probabilities. The proposed methodology supports the digital and probabilistic integration-based integrity management of BHNG pipelines.