Abstract
Real-time downhole data from coiled tubing (CT) milling operations improved performance metrics including rate of penetration (ROP) and stall rates. Those data enabled new diagnostics such as detection of milling target interfaces. At the same time, characterization of milling targets improves detection of interfaces between two contiguous milling targets and enables real-time diagnostics of the milling motor, bit, and target. These new capabilities enable more cost-effective operations and automation.
The torque and applied thrust relationship provides a means of characterizing the milling curve along which a milling bottomhole assembly (BHA)—motor and mill bit—and milling target operate. Torque-thrust data from milling three types of downhole targets—cement, through-tubing bridge plugs (TTBP), and composite bridge plugs (CBP)—are used to characterize that relationship for each BHA-target pair. Torque-thrust slopes for cement and mechanical plugs were calculated based on milling data from seven different wells. These data provide expected values for future milling operations and a reliable means to identifying when the BHA transitions from cement to a mechanical target.
The torque-thrust slope of cement (six samples), TTBP (three samples), and CBP (three samples) targets average −0.10, −0.01, and −0.03 ft-lbf/lbf, respectively. Cement milling follows a steeper torque-thrust curve than TTBP and CBP, which is explained by a higher friction coefficient between mill bit and cement. The TTBP has hard metal slips that must be milled to release the plug; the CBP has minimal metal content and is designed for easier millout of body and slips. Those differences in material and build explain the difference in torque-thrust curve slope between mechanical plugs. Changes to mill bit and milling target condition, pump rate fluctuations, and downhole condition variations also trigger deviations in the torque-thrust behavior. An algorithm based on a cumulative sum (CUSUM) statistical method detects small shifts in acquisition channels based on current and previous data. The algorithm considers individual surface and downhole channels, estimates group statistics, and triggers event detection when the CUSUM drifts beyond predefined standard deviations of the mean. The algorithm automates real-time detection and visualization of tagging top of targets, active milling, and stall events. The algorithm is augmented by known BHA specifications to anticipate stall conditions based on maximum recommended differential pressure, thrust, and torque.
The algorithm detects downhole events 9−27 seconds before they are visually perceptible, accelerating reaction time. Its causal design allows real-time detection and can be ported to CT acquisition software. It can calculate metrics including ROP and stall rates almost instantly, either in real-time or in post-job analysis. A control decision model is proposed for extending event detection—tagging a target, starting milling, anticipation of a stall, and stall events—to an automated CT milling operation.