A white-box machine learning model for predicting compressive strength of oilfield geopolymer cements

Agwu, Okorie Ekwe, Alatefi, Saad, Yusof, Muhammad Aslam Md and Agwu, Nwode 2026. A white-box machine learning model for predicting compressive strength of oilfield geopolymer cements. Results in Engineering 32 , 113016. 10.1016/j.rineng.2026.113016

PDF - Published Version Available under License Creative Commons Attribution Non-commercial. Download (3MB)

Abstract

As the world transitions toward carbon neutrality, low-carbon cement alternatives are increasingly important. Geopolymer cements, synthesized from aluminosilicate precursors activated by alkaline solutions, offer a promising substitute for oilfield well cementing. However, accurate and interpretable prediction of compressive strength is essential for safe field deployment. This study develops a white-box, explainable machine learning model based on a Bayesian regularized neural network (BRNN) to estimate the compressive strength of geopolymer cements. The model targets high-performance, eco-friendly alternatives to Ordinary Portland Cement, particularly for harsh downhole environments. It balances predictive accuracy with interpretability by providing an explicit mathematical relationship between key parameters—alkaline solution concentration, mole ratio, liquid-to-fly ash ratio, curing temperature, curing time and compressive strength. This formulation enables rapid estimation without retraining or computational iteration, supporting practical engineering use. The results obtained shows that the model demonstrated strong predictive capability achieving an R² of 0.9787, MSE of 8.33, RMSE of 2.86 MPa and MAE of 2.05 MPa. Sensitivity analysis shows curing temperature (r = +0.43), curing time (r = +0.417), and alkaline concentration (r = +0.411) as dominant positive factors, while mole ratio (−0.374) and liquid-to-fly ash ratio (−0.295) exhibit negative effects. Trend analysis confirms consistency with established geopolymerization mechanisms. Leverage analysis indicates that 98.5% of data points fall within the model’s applicability domain, confirming robustness. Overall, this work demonstrates that transparent, physics-consistent models can reliably support geopolymer cement optimization and sustainable drilling operations.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	2590-1230
Date of First Compliant Deposit:	28 September 2026
Date of Acceptance:	14 September 2026
Last Modified:	28 Sep 2026 10:45
URI:	https://orca.cardiff.ac.uk/id/eprint/189821

Actions (repository staff only)

Edit Item