Self-Learning Probabilistic Detection and Alerting of Drillstring Washout and Pump Failure Incidents During Drilling Operations

Abstract

The mechanical failure of drilling equipment is an operational risk that can be limited through a robust detection and alerting system, particularly for Drill String Washouts (DSW) and Mud Pump Failures (MPF). The detection of these issues focuses primarily on the time signatures of the real-time and modeled pump pressure in relation to the flow rate trends. Together, these parameters describe the state of the equipment which can be assessed through real-time alerts. A new methodology for real-time detection of washout and pump failure incidents during drilling operations was developed. The methodology behind the detection system uses a Bayesian network that models the drilling hydraulics and their associated failure modes. The network aggregates data from real-time rig floor sensors (standpipe pressure, pump rate, flow out, etc.), contextual information (rig state, mud properties, etc.), and predictions from hydraulic modeling. Cumulatively, they are the determinants of a probabilistic belief system indicative of DSW and MPF. The probabilistic model outputs belief values for DSW and MPF between zero and one. Given past and present trends, the model increases accuracy though self-learning and self-calibration that adjusts for poor sensor data, drilling conditions, and model uncertainty. The Bayesian network was integrated into decision support software with real-time alerting capabilities. The software was then validated by an operator's 100+ onshore wells in North America, some of which contained MPF and DSW incidents with varying degrees of severity. Several case studies drawn from these wells are presented in the paper. Each failure event that exceeded a programmed threshold for a specified period of time generated an alert in the form of a PDF report containing real-time sensor traces and DSW and MPF prediction outputs. The alerts were also displayed on a dashboard on the rig site user interface. Software thresholds were optimized to reduce false alert reports presented to the driller. Through continuous improvement and validation, DSW and MPF detection reached a level of accuracy which, in some cases, detected the warning signs of a failure hours before the problem was noticed at the rig site. Conclusively, the value added by the early detection of mechanical failures is the decreased amount of non-productive time due to pump downtime and maintenance, as well as trips and fishing jobs due to washed out pipe.

Characteristic Analysis of Downhole Pressure Signal while Drilling under Drillstring Washout Conditions

“…Wang Xinhu (2016) [10] studied the causes of drill pipes fatigue in deep well and provided optimization suggestions for subsequent drillstring design. A. Ambrus (2018) et al [11] established a self-learning probability detection and alarm system for drillingstring washout and pump failure based on Bayesian networks, it can improve the accuracy by self-learning and calibration based on sensor data, drilling conditions and model uncertainty.…”