The drilling optimization concept corresponds to enhancing the drilling efficiency by identifying the optimum drilling parameters with respect to bit wear. On many occasions, human interference can reduce the scale of a problem and enhance efficiency, but it has limitations, especially when working with big data and complex machines. This work utilizes in-cutter force sensing data and a scaled-drilling rig to apply an AI-based solution and facilitate drilling optimization.

There are several experimental challenges to building an advisory system: tuning the PID controller, continuous communication and feedback between hardware and software, data synchronization, and system efficiency. The core capability that enables drilling optimization is in-cutter sensing, which allows evaluating the forces acting on a PDC cutter. Initially, an extensive experimental study is conducted to record and process the drilling data with sharp cutters. Weight on bit, torque, rotational speed and rate of penetration are measured and sampled at the same frequency as the in-cutter force sensing. Rock samples with different mechanical properties are utilized and tested in atmospheric conditions.

The AI-based solution utilizes the data from the PDC cutter and the scaled-drilling rig structure to identify the optimum range of the drilling parameters depending on the mechanical properties of the rock samples. Artificial Neural Network (ANN) is utilized to predict the rate of penetration for samples with different uniaxial compressive strength. The supervised machine-learning models are trained on input variables such as weight on bit, torque, rotational speed, uniaxial compressive strength, vibrations, and more importantly the measured force at the PDC cutter.

A physics constraint is applied for torque, weight on bit and vibrations to guarantee that the output of the ML model is within the operating ranges. For each sample, approximately 100 data points are extracted per variable, with 70% for training, 15% for validation and 15% for testing. Cross-validation is used to enhance the robustness of the ML model. The results show that prediction performance is enhanced when the in-cutter sensing measurement is implemented as an input variable. The reliability of the ML models is tested on samples with different mechanical properties, with the results indicating an accurate prediction of the rate of penetration.

The paper demonstrates an innovative workflow that combines AI and in-cutter sensing data to enhance the drilling efficiency with respect to system's mechanical limitations. The model provides a reliable and rapid decision to identify the optimum drilling parameters with respect to physics constraints.

You can access this article if you purchase or spend a download.