Operational safety, efficiency and reliability have been key priorities for any oil and gas project. During drilling and production operations, while several safety barriers, controls and mechanisms are incorporated into the overall project plan, the reliability of the cement layer is vital to prevent catastrophic incidents and failures like blowouts. While properties of cement have been studied extensively in the laboratory, evaluating its long-term downhole behavior remains challenging due to practical limitations of cost, time and access. This study leverages machine learning (ML) strategies to determine the mechanical properties of cement such as compressive strength (UCS) and mechanical stability over time using available experimental laboratory data. An extensive experimental dataset of over 1,100 oilwell cement samples was tested at temperatures ranging from 25ºC to 75ºC, and the compressive strength (UCS) was measured to determine stability. Experimental results revealed that while higher curing temperatures (75°C) accelerated early-stage UCS development (1-3 days) compared to lower temperatures (up to 15 days), they suffered significant thermal degradation, reducing peak UCS by at least 20%. Using this laboratory data, three predictive models were developed and evaluated: Random Forest (RF), Gradient Boost (XGBoost), and Support Vector Machines (SVM). All models performed well on testing, with high R² values of 0.9166, 0.8866 and 0.7156, indicating that machine learning strategies can reliably predict intermediate-to-long-term cement strength under varying downhole conditions. Ultimately, this research provides operations managers and petroleum engineers with data-driven predictive tools thereby enhancing well integrity management.

Keywords

Machine Learning, Oilwell Cement, Cement Compressive Strength, Neural Network and Deep Learning.