An Innovative Automatic Alarm System for Early Kick or Influx Events by a Machine Learning Based Abnormal Flowback Detection

Abstract

Abstract During drilling operations, kick or influx events are threatening the safety of operating personnel and the environment; thus, early and accurate detection of the potential kick or influx becomes essential. Kicks frequently happen while making connections because when the pumps are turned off, the circulating flow rate decreases to zero, and the pressure exerted on the formations moderates to the hydrostatic pressure. Flowback fingerprinting, an analysis method for interpreting and comparing flowback patterns, has been widely used in the industry recently for early kick and loss detection during connection operations. However, the flowback fingerprinting, usually conducted manually by skilled engineers, lacks timeliness and accuracy in some scenarios. This paper presents an innovative solution to enable detecting abnormal flowbacks automatically by leveraging data-driven and machine learning based algorithm, including smart safe envelope, mud transfer filter and trip tank change detection. The new system automatically monitors the flowback volume vs. time data and compares them with a safe envelope in real time. The safe envelope is formed by a machine learning process, which performs clustering and curve-fitting based on historical normal flowbacks in the current well. Uncertainty is also included in the process to determine the upper and lower bound of the envelope. During each connection period, if the flowback curve deviates from the safe envelope, it will be identified as abnormal flowback and alarms will notify the users. During the operation, the safe envelope is automatic updated adaptively based on the characteristics of current flowback, e.g., the hole depth and the flow rate before turning off the pumps. This fully automated system greatly improves the accuracy and timeliness of abnormal flowback detection while minimizing the frequency of false alarms. Besides, the smart alarms could be sent to users by mails or mobile app which makes the reaction and decision-making more in time and convenient. The new system has been evaluated on previously drilled wells. In this paper, the capability of the new system is presented in one case studies by means of streaming actual well data, in which alarms were successfully triggered before the influx events occurred. The case studies demonstrate the capability and benefits for the entire workflow and how they can significantly reduce operation risks and non-productive time.