Predicting stress corrosion cracking in downhole environments: a Bayesian network approach for duplex stainless steels

This study presents a Bayesian network (BN) for the holistic assessment of stress corrosion cracking (SCC) risk. The model is designed to interrogate the optimal operating conditions of duplex stainless steels (DSSs) in downhole environments, addressing the perceived overly conservative limits by current industry standards, particularly those from ISO 15156—Part 3. A knowledge-based dataset on DSS performance was compiled from diverse sources. Machine learning and deep learning techniques facilitated data pre-processing and identification of feature interactions, supporting the BN structure’s development. Extensive cross-validation demonstrated that the BN model accurately predicted the occurrence of both pitting corrosion and SCC with over 90% accuracy. Using the BN model, inference analyses were undertaken to examine SCC risks for DSSs under diverse sour conditions. The results indicate that DSSs could withstand more aggressive conditions than those currently permitted by ISO 15156—Part 3, suggesting potential for broader and more effective use in oilfield applications.