A Data Driven Approach to Generating Synthetic Time-Lapse Casing Corrosion Logs

Downhole casing corrosion monitoring is crucial in production engineering to ensure well integrity and extend serviceability. Engineers perform corrosion time-lapse logs at intervals based on corrosion severity. With increasing well numbers and complexity, the demand for logs increases, challenging cost-effective surveillance. This paper introduces a machine learning application using One-Versus-Rest (OvR) Multi-class Classification to predict the next casing corrosion time-lapse log for previously logged wells. This study investigates Galvanic corrosion in oil and water wells, which propagates originally from the outermost casing to inner casings. Corrosion severity is determined by the degree of metal loss and is typically categorized into high, medium, and low classes. A machine learning model is applied to classify inputs at each depth zone into one of these classes, generating a comprehensive corrosion log. The model architecture consists of binary classifiers based on Light Gradient-Boosting (LightGBM) algorithm, predicting class membership probabilities. The final corrosion class prediction is assigned based on the maximum probability among the classifiers. Corrosion progression was found to be predominantly influenced by environmental exposure, particularly at interfaces between casing and surrounding formations. Factors such as corrosion history, structural attributes, depth-specific characteristics, and broader geological and geographical features were identified as primary contributors to model's performance. A machine learning framework was developed to estimate corrosion severity classification across depth zones and casings. The model demonstrated high accuracy in predicting metal loss growth over time in critical high corrosive depth zones, showcasing its potential to identify early signs of corrosion-related risks and avoid unnecessary logging costs, thereby providing a valuable tool for the oil and gas industry to optimize well maintenance and integrity. The method presented a novel approach to optimizing and potentially reducing costly corrosion time-lapse logging requirements while ensuring the safety of operations. By leveraging AI to generate a portion of the time-lapse logs, production engineers can balance budgetary constraints while focusing physical logging efforts where they are most needed.