ARCHIVES
VOL. 10, ISSUE 1 (2026)
Transfer learning and meta-learning approaches for generalizing rheology models across multiple mud formulations
Authors
Ichenwo John Lander, Ogwu Philip
Abstract
The successful wellbore hydraulics and
drilling processes are heavily reliant on drilling fluid rheological properties
prediction accuracy. Traditional machine learning models often cannot
generalize across different mud formulations, leading to poor prediction
results and requiring large amounts of experimental data. This paper examines
the use of TL, MLL and a hybrid of the two in improving the generalization of
rheology models to different systems in OBM, WBM and SBM. In this study, the
random forest, support vector machine, and artificial neural network are the
fundamental machine learning models that will be used. ANN delivered the best
predictive accuracy (R2 = 0.11). However, it was not enough. TL was
implemented by pre-training ANN models on well characterized mud data, with the
best R2 at 0.41, which is more than 250 percent of the improvement
compared to baseline models. MLL carried out in the MAML model achieved R2
scores of 0.28, 0.33 and 0.38 in 5, 10 and 20 shot learning situations
respectively, and showed good adaptability with minimal data requirements. The
TL+MLL method, on the other hand, was the best among all the other methods to
achieve maximum R2 of 0.52 at 20 shots with enhanced generalization.
The paper provides an identification of a smart and adaptable rheology model
and promotes the growth of the approach to incorporate real-time field
measurements and other fluid characteristics, such as gel strength and
filtration control, to be applicable in the field of the drilling operation.
Download
Pages:32-36
How to cite this article:
Ichenwo John Lander, Ogwu Philip "Transfer learning and meta-learning approaches for generalizing rheology models across multiple mud formulations". International Journal of Advanced Engineering and Technology, Vol 10, Issue 1, 2026, Pages 32-36
Download Author Certificate
Please enter the email address corresponding to this article submission to download your certificate.