Data-driven reliability evolution prediction of underground pipeline under corrosion

Abstract

Corrosion presents a substantial threat to both the structural integrity and the service life of pipelines. Despite the availability of existing models for assessing corrosion rate and pipeline reliability in the oil and gas industry, their applicability is constrained by the inherent complexity of the surrounding soil environment. In this study, a novel artificial intelligence-based hybrid model was developed to predict pipeline corrosion rate. The Extreme Learning Machine (ELM) was employed as the primary predictor. The Bald Eagle Search (BES) algorithm was integrated and enhanced by incorporating Lévy flight search and Simulated annealing (SA) algorithms, forming the LSBES algorithm to optimize the parameter learning of the ELM model. Three machine learning models were developed as benchmarks to evaluate the performance of the proposed hybrid model. The results demonstrate that the LSBES-ELM model demonstrates superior predictive accuracy and stability, with a mAP of approaching 95 % and a RE ranging from 0.0274 to 0.0761, surpassing the performance of both baseline ELM-based models (BES-ELM, ELM) and the non-optimized BP Neural Network (BPNN). Furthermore, the LSBES-ELM-MCS model was developed with the LSBES-ELM model and Monte Carlo simulation (MCS) to perform a dynamic assessment of the optimal distribution of factors influencing corrosion rates and the reliability evolution prediction of pipelines with various buried soil conditions. With target reliability, the optimal inspection interval for the case pipeline was projected to fall between 21 and 24 years. This study is expected to present significance for modeling corroded pipeline reliability and contribute to the broader goal of enhancing pipeline safety and longevity in the oil and gas industry.