Kick Detection During Offshore Drilling using Artificial Intelligence

Abstract

An uncontrolled or unobserved influx or kick during drilling has the potential to induce awell blowout, one of the most harmful incidences during drilling both in regards to economicand environmental cost. Since kicks during drilling are serious risks, it is important both toimprove kick detection performance and capabilities, and to develop automatic flux detectionmethodology. There are clear patterns during an influx incident. However, due to complexprocesses and sparse instrumentation, it is difficult to predict the behavior of kicks or lossesbased on sensor data combined with physical models alone. Emerging technologies within DeepLearning are however quite adept at picking up on and quantifying, subtle patterns in timeseries given enough data.In this paper, new models for kick detection is developed by using Long Short-Term Memory(LSTM) and Bidirectional LSTM (BiLSTM), two types of Deep Recurrent Neural Network, forkick detection and influx size estimation during drilling operations. The proposed detectionmethodology is based on simulated drilling data and involves detecting and quantifying theinflux of fluids between fractured formations and the wellbore in a large range of dynamicdrilling simulations.The results show that the proposed methods are effective both to detect and estimate the influxsize during drilling operations so that corrective actions can be taken before any major problemoccurs. The results further indicate that these methods can be used on readily available sensordata on the drill rig. Making it a suitable technology for both modern and older drilling rigs.