Machine learning is becoming increasingly common across oil and gas, but the more important question is not whether the industry can apply machine learning. It is where it can actually add engineering value.
Cementing is one area where that question is particularly relevant.
Displacement efficiency depends on a combination of fluid properties, pumping conditions, pipe geometry, inclination, and movement. These variables do not operate independently, and small changes in one condition can affect how fluids behave throughout the displacement process.
That complexity creates an opportunity for data-driven approaches to complement established engineering methods.
Recent research by Hu Dai, Ph.D. and Zhibin Sun, published in the Journal of Energy Engineering, examines that opportunity by applying machine learning to the prediction of non-Newtonian fluid displacement efficiency in vertical and inclined pipes. The paper was published online September 10, 2026, and is included in the journal’s December 2026 issue.
The Challenge Is Bigger Than a Prediction
At first glance, using machine learning to predict displacement efficiency may sound straightforward: provide enough data, train a model, and generate a prediction.
Engineering applications are rarely that simple.
Drilling fluids and cement slurries exhibit non-Newtonian behavior, meaning their response can change depending on operating conditions. During displacement, engineers are dealing with several variables simultaneously, including velocity, density, viscosity, pipe inclination, and rotation.
The challenge is therefore not just predicting an outcome. It is understanding which factors matter, how they interact, and whether those relationships make physical and engineering sense.
That is one of the more interesting aspects of this research.
The study uses an XGBoost model trained on experimental displacement data and then applies SHAP analysis to examine the influence of individual variables. The model achieved an R² of 0.772 on the held-out test dataset.
In other words, the work does not stop at asking, “What does the model predict?”
It also asks, “What is driving that prediction?”
That distinction matters.
Data Can Reinforce What Engineers Already Know
The research identified pipe rotation, imposed velocity, and the density and viscosity of the displaced and displacing fluids among the factors influencing displacement efficiency. The findings also support the significance of higher pumping rates and casing rotation under the conditions examined in the study.
For experienced cementing engineers, those relationships are not necessarily surprising.
And that is precisely the point.
A successful machine-learning application does not have to produce a result that overturns established engineering knowledge to be valuable. It can provide a data-driven way to quantify relationships, explore interactions, and test assumptions across a broader set of conditions.
This creates a stronger relationship between engineering judgment and data.
Rather than asking engineers to choose between traditional modeling and machine learning, the more useful question may be how the two can work together.
Where Machine Learning Starts to Become Interesting
The bigger opportunity comes as engineering organizations accumulate more usable data.
Experimental results, laboratory measurements, simulation outputs, field data, and historical cases can contain patterns that are difficult to evaluate one variable at a time. Machine-learning methods can help identify those patterns and provide another layer of analysis.
For cementing, that could eventually support questions such as:
How do changes in fluid properties affect displacement under different operating conditions?
How do pumping velocity and pipe movement interact?
Where do certain operating conditions create more favorable displacement behavior?
How can historical data be used to inform future engineering decisions?
These are not questions that machine learning answers automatically. They require reliable datasets, appropriate model development, validation, and engineering interpretation.
But that is where the potential lies.
The Model Is Only as Good as the Engineering Behind It
There is also an important lesson in what the study does not claim.
The current research focuses on displacement inside vertical and inclined pipes using experimental data. The authors identify expanding the dataset and extending the work to annular displacement flows as areas for future development.
That limitation is important because machine learning does not eliminate the need for representative data.
A model trained on a narrow range of conditions cannot automatically be assumed to represent every well geometry, fluid system, or operating environment.
For digital engineering to be useful in the field, models need to be grounded in sound engineering principles and tested against relevant conditions.
That is why the combination of domain expertise, quality data, and machine learning is so important.
From Digital Models to Better Engineering Decisions
The future of machine learning in oil and gas may not be about replacing the engineer with an algorithm.
It may be about giving engineers better ways to work with the amount and complexity of data already available.
In cementing, that means moving beyond simply generating more data or more models. The opportunity is to turn data into information that engineers can understand, challenge, and ultimately use.
Research such as this provides an early look at what that could look like.
Machine learning can help uncover relationships within complex systems. Engineering knowledge provides the context to understand those relationships. Together, they create an approach that is both data-driven and grounded in the realities of field operations.
That is where the real opportunity for machine learning in cementing lies: not replacing engineering expertise, but extending what engineers can see, evaluate, and understand.
Explore the Research
The full research paper, “Predicting Inside-Pipe Displacement Efficiency with a Machine-Learning Model Using Experimental Data,” by Hu Dai, Ph.D. and Zhibin Sun, is available through the ASCE Library.
View the published paper:
https://ascelibrary.org/doi/10.1061/JLEED9.EYENG-6939
Looking Ahead
The value of machine learning in cementing will depend on more than predictive accuracy. It will depend on how effectively these models can be validated, interpreted, and integrated with the engineering knowledge already used to plan and execute cementing operations.
As datasets expand and research moves toward more representative wellbore and annular conditions, machine learning could provide new ways to examine complex displacement behavior and uncover relationships across variables that are difficult to evaluate independently.
For the industry, the opportunity is not to replace established engineering methods, but to build on them—bringing together experimental evidence, engineering models, and data-driven analysis to better understand what is happening downhole.
That intersection of engineering expertise and emerging data-driven methods is where the next generation of digital cementing research can continue to develop.
Interested in applying data-driven engineering to your cementing workflows? Connect with our team today!