What question did this study set out to answer?

The aim is to develop efficient sealing materials that enhance coalbed methane drainage while minimizing gas escape.

April 5, 2026Open Access

AI-based development and optimization of sealing materials for secondary grouting in methane recovery

Key Points

The aim is to develop efficient sealing materials that enhance coalbed methane drainage while minimizing gas escape.
Utilized a data-driven AI framework for predictive modeling and optimization.
Conducted experiments with variations in water-to-material ratio, phenol–formaldehyde dosage, crosslinker dosage, and foaming agent content.
Trained hybrid machine learning models to estimate viscosity, compressive strength, gel time, and expansion rate.
Employed genetic algorithms and particle swarm optimization for model tuning and performance comparison.
Performed VIKOR-based multi-criteria analysis for tailored material selection.
Achieved a maximum compressive strength of approximately 4.9 MPa with high apparent viscosity.
Lower-viscosity formulations (AV < 180 mPa·s) provided better injectability while maintaining acceptable compressive strength levels (3–4 MPa).
Identified Pareto-optimal compositions reflecting trade-offs between strength and fluency in FSMs.
The approach significantly reduces reliance on laboratory prototyping and enhances the exploration of material formulations.

Abstract

Efficient coalbed methane (CBM) drainage is frequently hindered by the presence of gas pathways in fractured strata, which reduces borehole sealing effectiveness and limits production rates. Developing fracture sealing materials (FSMs) that combine adequate fluidity for fracture penetration with rapid consolidation and long-term strength is therefore essential. To accelerate the design of such materials, this study introduces a data-driven artificial intelligence framework that links predictive modeling with multi-objective and multi-criteria optimization. A limited experimental dataset (25 samples), incorporating variations in water-to-material ratio (0.4 < WMR < 0.6), phenol–formaldehyde dosage (6.5 < PFD < 8.5%), crosslinker dosage (4.3 < CD < 4.7%), and foaming agent content (1 < FA < 1.4%), was used to construct hybrid machine learning models. Two evolutionary-tuned neural network surrogates were trained to estimate apparent viscosity (AV), compressive strength (CS), gel time (GT), and expansion rate (ER). Performance comparison demonstrates that genetic algorithm (GA)-based artificial neural network (ANN) yields superior prediction quality for AV and CS. In contrast, the particle swarm optimization (PSO)-based ANN achieves higher accuracy in forecasting GT and ER. The trained surrogates were subsequently integrated into the non-dominated sorting genetic algorithm III (NSGA-III) optimization algorithm to derive Pareto-optimal FSM compositions showing distinct trade-off behavior. Solutions favoring maximum CS (≈ 4.9 MPa) typically exhibit higher AV (≈ 305 mPa·s). Conversely, lower-viscosity formulations (AV < 180 mPa·s) offer better injectability while maintaining acceptable CS levels (≈ 3–4 MPa). To support practical material selection aligned with operational requirements, VIKOR-based multi-criteria analysis was performed. This analysis yielded tailored compromise solutions suitable for both highly pumpable and high-strength grouting conditions. This unified approach significantly reduces reliance on iterative laboratory prototyping. It enables efficient exploration of the formulation space and provides adaptable sealants capable of improving long-term CBM extraction under complex underground environments.

Mark Helpful

Bookmark

Relay

View Full Paper