Estimation of rocks' failure parameters from drilling data by using artificial neural network.

Area of Science:

• Geotechnical Engineering
• Petroleum Engineering
• Artificial Intelligence in Geoscience

Background:

• Accurate rock mechanical properties (cohesion, friction angle) are crucial for drilling optimization and wellbore integrity.
• Conventional methods rely on compressional tests of rock samples, which face challenges in availability, continuity, representability, and cost.
• There is a need for alternative, in-situ techniques to estimate these parameters.

Purpose of the Study:

• To investigate an alternative technique for estimating rock failure parameters (cohesion and friction angle) using instantaneous drilling data.
• To develop and validate artificial neural network (ANN) models for predicting these parameters.

Main Methods:

• Utilized over 2200 data points, each containing failure parameters and five drilling records (rate of penetration, weight on bit, torque).
• Developed ANN models trained, tested, and validated on datasets (60/20/20 split).
• Optimized and evaluated models using correlation coefficient (R) and average absolute percentage error (AAPE).

Main Results:

• The friction angle model achieved R values around 0.86 and AAPE around 4% across datasets.
• The cohesion model achieved R values around 0.89 and AAPE around 6% across datasets.
• Both models demonstrated good fits with actual values, indicating reliable estimation.

Conclusions:

• ANN models can reliably estimate in-situ rock mechanical properties from drilling data.
• This approach provides instantaneous and cost-effective parameter estimation, reducing reliance on sample-based laboratory testing.
• The developed models offer a valuable tool for real-time drilling performance optimization and risk reduction.