Hybrid Model for Detecting Abnormal Drilling Behaviors

Unplanned and unexpected events during drilling a well do not only lead to a massive loss of resources by increasing the amount of non-productive time, but also cause the necessity of plugging a well and starting a contingency side-track, which will add environmentally and economically risks to the originally planned project. Therefore, detecting the undesirable downhole drilling trouble at the earlier stages may help avoid the matters above. Several surface drilling parameters can be used to predict the downhole drilling problems in real-time. Nevertheless, torque and standpipe pressure are considered to be the most critical and useful parameters. Therefore, several methods utilizing the two indicated surface parameters for detecting the downhole drilling problems were published in the last decade. However, these methods have flaws, mainly related to delays in receiving the necessary information, uncertainties associated with involved data, human error by potential incomplete data sets (due to sensor misreading), as well as human error interpretation of the data. Thus, linking sequential pattern recognition for possible drilling event determination is impacted. Consequently, recognizing drilling parameter anomalies in real-time using one single approach, such as data-driven or model-driven, can lead to an excessive increase in the nonproductive time due to the generation of undue false alarms. Thus, integrating a stochastic model with a datadriven model will reduce the associated uncertainties and make the predictive model more effective. From this perspective, the ultimate goal of this thesis is to develop a hybrid model that provides better accuracy in detecting abnormal behavior of measured drilling parameters such as standpipe pressure and torque. A standalone application based on a hybrid model was developed during the thesis work by the implementation of statistical calculations based on actual and predicted data channels. As a result, uncertainty windows are created and compared to the actual data points in order to detect abnormal drilling behavior and triggering alerts to provide warnings to the user. In order to evaluate and determine the shortcomings of the developed workflow, the developed hybrid model, a case study was conducted. The final results of the case study reveal that the workflow is reliable and easy to use.