No abstract

No abstract

This paper is a follow up to the URTeC (2019-343) publication where the training of a Machine Learning (ML) model to predict rate of penetration (ROP) is described. The ML model gathers recent drilling parameters and approximates drilling conditions downhole to predict ROP. In real time, the model is run through an optimization sweep by adjusting parameters which can be controlled by the driller. The optimal drilling parameters and modeled ROP are then displayed for the driller to utilize. The ML model was successfully deployed and tested in real time in collaboration with leading shale operators in the Permian Basin. The testing phase was split in two parts, preliminary field tests and trials of the end-product. The key learnings from preliminary field tests were used to develop an integrated driller's dashboard with optimal drilling parameters recommendations and situational awareness tools for high dysfunction and procedural compliance which was used for designed trials. The results of field trials are discussed where subject well ROP was improved between 19-33% when comparing against observation/control footage. The overall ROP on subject wells was also compared against offset wells with similar target formations, BHAs, and wellbore trajectories. In those comparisons against qualified offsets, ROP was improved by as little as 5% and as much as 33%. In addition to comparing ROP performance, results from post-run data analysis are also presented. Detailed drilling data analytics were performed to check if using the recommendations during the trial caused any detrimental effects such as divergence in directional trends or high lateral or axial vibrations. The results from this analysis indicate that the measured downhole axial and lateral vibrations were in the safe zone. Also, no significant deviations in rotary trends were observed.

Kick events are some of the most life-threatening and environmentally disastrous events during drilling. Identifying potential kick events in time is valuable. Several hybrid algorithms combining physics and data analytics have been developed to help identify potential kick events from trends in real-time drilling data. These algorithms encompass all drilling operations like drilling, tripping, circulating, and making connections. The goal is to enable management by exception in real-time monitoring of wells. Real-time drilling data acquired from the sensors on the rig and other static metadata like drill string and casing specifications are used in these hybrid algorithms. The exact data channels from the real-time data used differ based on the drilling operation the algorithm is associated with, but the most important ones are flow rate in (FLOWIN), flow out percentage (FLOWOUT), pump strokes per minute (SPM), active pit volume (PITACT), and trip tank volumes (TTKVOL). Additional data channels like lag depth and rig states are computed from these existing channels to help the algorithms. Other provisions have been made in the algorithms to automatically account for human operations like lining up with trip tanks and addition/removal of pits from active pit volume. To identify the potential kicks: data from trip tank volumes are utilized while tripping, flowback signatures calculated from PITACT are used during connections, and real-time hole displacement (HDISP) calculated from PITACT and lag depth is used while drilling. All the algorithms have been parameterized to facilitate easy tuning of thresholds. An alerting system has been implemented that triggers an alarm when the algorithms identify a potential kick event. This system can also send notifications to a real-time support team or field personnel. Historical offset wells provided by an operator were used for testing the algorithms and fine-tuning the thresholds. We achieved a 12% false-positive rate while correctly identifying all the true well control kick events. False positives are defined as events identified by the algorithms as potential kicks but were due to some explainable operation that was accounted for in the algorithm scope. The user feedback from the alerting system was also used to improve the accuracy of the algorithms. The calculated data channels used to identify the potential kicks can also be displayed as real-time traces or in other suitable visualizations like a trip sheet or flowback fingerprints. These algorithms were designed to be run with minimal user input. This makes them suitable for use by real-time support centers and field personnel. The ability to calculate the lag depth and incorporate that into the analysis is novel and improves the accuracy. The provisions in the algorithm to account for human operations that affect pit volumes also add to the novelty.

Operators have established procedural requirements as part of their HSE, risk and cost control policies. Automating procedural compliance enables real-time operations oversight while reducing risks associated with manual processes. A novel technique was developed and implemented to automatically check rigs’ procedural compliance with an operator's guidelines and alert users in real time. Textual procedures are initially converted into structuralized JSON (JavaScript Object Notation) format making it easier for the user to form the logics and generate a format usable for the computer. Each procedure consists of starting and ending triggers and compliance rules. As WITS data is fed to the engine, they are automatically compared against the given guidelines. Triggers are detected in WITS data using automated rig activity detection (e.g. wash up, or rotary drilling). Subsequently, compliance rules are checked and, if all criteria are met, it is detected as complied; otherwise, an alert is triggered. The following procedures have been developed and deployed: zeroing of weight on bit and mud motor differential pressure for more accurate ROP (rate of penetration) optimization, lowering BHA (bottom hole assembly) back to bottom with reduced RPM, connection practice procedure, and capturing steady-state hook-load values for improved torque and drag monitoring. Each procedure was identified to influence future optimization efforts to a varying degree. Consider if the weight on bit is not zeroed properly it can cause incorrect driller's road-map recommendations for new wells. The framework has been extensively tested and refined with the assistance of a Permian operator until it reached an acceptable accuracy: 95% for back-to-bottom procedure, 99% for zeroing WOB (weight on bit), and 90% for zeroing DIFF (differential pressure). Reports have been generated to display the statistics of the compliance for each rig and crew. A Permian operator, who has field tested the procedural compliance application, estimates that it will save millions of dollars per year by avoiding costly equipment damage as well as from significant time savings by standardizing a back-to-bottom procedure. Real-time procedural compliance is a novel technique to check whether the rig personnel follow the operator's procedures. It has a flexible back-end interpreter, and each procedure is provided separately. This ensures the procedure is not attached to the original code and can be easily modified. Through the use of a real-time compliance engine, every operator can load their respective procedures and check for compliance in real-time.