Prediction of geopolymer pumpability and setting time for well zonal isolation - Using machine learning and...

Area of Science:

• Materials Science
• Chemical Engineering
• Petroleum Engineering

Background:

• Geopolymers and alkali-activated materials are emerging as sustainable alternatives to Portland cement.
• Their application in oil and gas well cementing for zonal isolation is under investigation.
• Limited lab-scale research exists, necessitating predictive models for broader application.

Purpose of the Study:

• To address the gap in understanding geopolymer behavior under diverse conditions.
• To apply various prediction models to limited experimental data for accurate pumpability prediction.
• To validate the generalizability of these models for geopolymer applications.

Main Methods:

• Utilized binary/multi-threshold classification models (logistic/probit, decision tree, random forest, SVM).
• Employed regression and continuous models (multi-linear regression, neural networks).
• Focused on predicting the critical geopolymer property: pumpability.

Main Results:

• The Decision Tree model demonstrated a simple, intuitive understanding of geopolymer behavior.
• Accurate prediction of geopolymer pumpability was achieved using the Decision Tree model.
• The probability of inaccurate predictions with high operational risk was found to be very low.

Conclusions:

• Geopolymer pumpability can be reliably predicted using machine learning models, particularly Decision Trees.
• The Decision Tree model offers a practical tool for assessing geopolymer suitability in well cementing.
• Further research should focus on model generalization and validation with extensive field data.