Coupling Psychological Factors With Machine Learning to Improve Rig Technical Training

Abstract

Abstract Technical training is an essential activity for optimizing rig operations. Recently, the use of drilling simulators has revolutionized the way training is done and, accompanied with on-site assistance, it has ensured near optimal performance from the trained crews. This paper explains how machine learning and physiology can be used to improve rig technical training by monitoring the operator's stress, identifying the key operations where situational awareness is low and targeting these operations with dedicated exercises. The developed methodology is based on a study of human psychological indicators captured through light biometric devices. These indicators are fed to a machine learning algorithm that calculates a stress index for the observed operator and uses this index to identify key operations where the operator lacks focus, is under high stress or feels a lack of preparation. The measured indicators are skin temperature, specific face movements, heart rate, and sweat. The model uses machine vision to identify key physiological parameters and a convolutional neural network to interpret them. Finally, a third algorithm correlates the stress index to specific operations. The system can be used either in simulation environment or on the rig itself during operational studies. The primary results show high detection accuracy with minimal errors. Using this methodology for well control simulation, the main periods of high stress and low concentration were correctly identified. The repeated tests showed that different drillers or supervisors respond differently to the situation and may be stressed out by different operations. This highlighted a key drawback of the training that focuses on the same main operations for all participants. By customizing the second training session for each participant's needs, the high stress levels were significantly reduced. From the initial trials, a key point needed to be highlighted: for the study to be as non-intrusive as possible, the biometric devices used for monitoring stress need to be as light as possible. This led to a review of the devices used and a compromise between accuracy and lightness. As with advanced military training, targeted training for drilling rig crews can deeply impact the outcome of the training and preparedness of the crew. Today, biometric devices combined with machine learning models finally, allow for an accurate detection and evaluation of human stress. Using this analysis methodology to customize training will prove essential soon and may revolutionize the way rig crews are trained.