Development of an enhanced torque and drag model using machine learning for optimizing drilling efficiency

The object of this research is drilling process. The key is to ensure safe and efficient drilling operations by proactively identifying and eliminating critical anomalies, such as stuck pipes, that cause downtime, increase costs and degrade performance. A machine learning model combining Multilayer Perceptron (MLP) and XGBoost was developed to predict critical parameters such as hook weight, minimum weight on bit, effective tension, and torque on bit. The model achieved 86 % accuracy in detecting drilling anomalies, including sinusoidal and spiral buckling. This enabled timely corrective actions and improving drilling efficiency. The model’s accuracy is due to its ability to process large datasets and capture complex, nonlinear relationships between drilling parameters. By training on both historical and real-time field data, it can learn patterns that are difficult to detect with traditional tools which allows to predict of drilling anomalies in real-time. The distinctive feature of this model is its adaptability to new data, as well as its ability to predict complex phenomena like helical buckling and torque fluctuations, which are challenging for traditional methods. Unlike conventional models that need manual tuning, this model continuously learns from data, improving over time and under varying conditions. The model can be applied practically in real-time drilling operations to optimize drilling parameters, reduce the risk of stuck pipes, and minimize non-productive time