Machine Learning for Sulfide Stress Cracking Prediction

Abstract

Stress Corrosion Cracking (SCC) poses a significant threat to production systems, arising from the interaction of tensile stresses and corrosive environments. Sulfide Stress Cracking (SSC), particularly associated with hydrogen sulfide (H2S) gas, is highly relevant in oil and gas production. Corrosion‐resistant alloys, such as Duplex Stainless Steel (DSS), help mitigate this issue. However, understanding the impact of environmental conditions and loads on SSC in DSS remains challenging. Existing standards lack insights into specific environmental factors. Modeling SSC using physics‐based approaches is computationally intensive. To address this, a novel machine learning (ML) framework utilizing decision tree‐based models and probabilistic graphical models (Bayesian network, BN) is developed. The dataset for DSS is curated from published literature, and data imbalance is addressed using advanced data curation methods. The framework aims to unravel the intricate factors driving SSC in DSS, providing an accurate predictive tool for the oil and gas industry.