Challenges and Lessons of Implementing a Real-Time Drilling Advisory System
This paper discusses the technical challenges related to implementing a rigsite, real-time drilling advisory system and current solutions to these challenges.
This paper discusses the technical challenges related to implementing a rigsite, real-time drilling advisory system and current solutions to these challenges. The system uses a data-driven response-surface model based on physics-based calculations to optimize rate of penetration (ROP) while minimizing drilling-vibration dysfunction with regard to lateral (whirl) and torsional (stick/slip) vibrational modes. Minimizing these vibrations is important to mitigate bit damage that can lead to reduced ROP and increased bit trips.
The system is a rigsite software application that should be deployed in view of the driller. Fig. 1 above shows a driller-cabin deployment.
The software contains capabilities for real-time surface drilling-data acquisition, drilling-performance estimation, vibration analysis, surface trends for drilling performance, and drill-off-test guidance for drilling optimization. The system primarily serves as an open-loop advisory tool but retains capabilities for closed-loop autodriller and topdrive control. The user interface provides the rigsite personnel with drilling-performance surface trends (e.g., ROP, drilling efficiency, and stick/slip vibration), bit aggressiveness and depth-of-cut (DOC) calculations, and drilling-parameter set-point recommendations on the basis of the surface trends.
Data Input and Output. The system operates on 1-second data provided from the electronic data-recording equipment. Input data consist of data channels included among standard or spare Well Information Transfer Specification (WITS) Record 1 items—block height, weight on bit (WOB), rotary speed, mud-flow rate, hole depth, bit depth, torque, and differential pressure.
Drilling-Performance Estimation. The system filters and preprocesses the raw WITS Record 1 data to calculate the drilling performance variables: ROP, surface mechanical specific energy (MSE), motor MSE, DOC divided by WOB, torsional-severity estimate (TSE), bit aggressiveness, and DOC.
Performance Averaging and Modeling. The measures of drilling performance, primarily the ROP, drilling efficiency, and stick/slip indicator, are averaged over depth drilled to produce a mean or median value referred to as a “response point.” A clustering algorithm groups these response points in the 2D drilling-parameter space. The response-point groups, or “calibration points,” serve as estimates of the drilling set points to measure whether the drilling-parameter space is explored sufficiently to produce an accurate response-surface model.
Optimization. An objective function merges the response surfaces of multiple performance objectives into a single objective surface. This objective function makes a tradeoff between the ROP, drilling efficiency, and stick/slip terms.
Drill-Off-Test Guidance. When the system is recalibrated for a new hole section or new rock formation, it recommends a sequence of drilling set points to explore the parameter space. Once the space is explored sufficiently for an initial trend, the system makes set-point recommendations in a near-optimal direction on the basis of the surface to continue exploring the space.
Drilling-Performance Estimation From Data
Raw surface drilling data should be filtered, cleaned, and combined within drilling-performance equations to create an estimate of drilling performance. The drilling-performance measures of ROP, MSE, DOC, DOC/WOB, and bit aggressiveness have been commonly used in the industry for viewing as strip-chart data. Additional processing can be required to use these measures for real-time drilling optimization: removal of dynamic artifacts to estimate downhole performance, choice of an appropriate proxy for drilling performance, and verification that trending occurs within a stable autodriller control mode.
Data-Driven Adaptive Modeling and Disturbance Handling
The system uses data-driven statistical modeling by averaging current drilling inputs and outputs to create response points and fitting a 3D response surface to the response points. The surfaces represent a trend of two drilling parameters at steady-state conditions vs. a single drilling-performance variable. The two drilling parameters consist of the inferred topdrive rotational-speed set point and inferred autodriller set point for the current autodriller control mode. Because the system primarily focuses on real-time drilling mechanics, the drilling-performance variables include ROP, MSE, DOC, DOC/WOB, TSE, and bit aggressiveness. Drilling set points must be held steady for a sufficient depth of rock drilled and sufficient time to characterize steady-state drilling performance at the current set point.
Unmeasured disturbances may significantly alter the relationship between the drilling parameters and the drilling-performance variables without indications from surface measurements, and these are a primary challenge for creating an accurate response surface to correlate drilling parameters with performance.
One approach to manage the unmeasured disturbances is moving-window filtering. A moving-window algorithm only includes data over a fixed window of time extending from the current time to the past. The moving window slides in time to remain anchored to the present time while ignoring data before the window to maintain the same window length. Applying surface fitting to the moving window of response points enables the surface to adapt gradually to varying rock strength or bit wear.
The drilling advisory system performs drilling optimization by balancing the objectives of increasing ROP and avoiding lateral and torsional drilling vibrations. A drilling-efficiency response surface—either surface MSE, motor MSE, or DOC/WOB—serves as a proxy for lateral vibrations vs. drilling parameters. The TSE response surface quantifies the effect of parameters on stick/slip torsional vibrations.
A key challenge for balancing these objectives is comparing terms with relative importance, such as ROP and drilling efficiency, with a term with absolute importance such as TSE. ROP and drilling efficiency have relative importance because optimal values for these are relative to the rock strength. The TSE has an absolute importance such that any TSE value equal to or greater than 1.0 indicates that the bit rotary speed drops to zero momentarily during the stick/slip cycle.
The system handles objective terms with relative importance—ROP and drilling-efficiency terms—by normalizing and scaling the surfaces with respect to the relative slope of the surfaces.
With the surfaces normalized to the same scale, the surfaces are then scaled with regard to the relative slope of each surface.
A scaling parameter acts as an adaptive weight that governs the importance of each objective term in the objective function. The parameter up-weights relative objective terms that show high variance with respect to the drilling parameters, and vice versa.
The system poses drilling optimization as an explore/exploit problem and uses a data-driven response-surface model to visualize the explore/exploit tradeoff. This tradeoff involves choosing when to vary parameters to learn the data-driven model and when to choose parameters to exploit the optimal point.
A key objective of the system is to identify improved or near-optimal drilling performance relevant for the current formation by analyzing trends. The drilling set points should be varied at a sufficient frequency to ensure that the trend remains relevant for current drilling operations. This drilling-set-point variation is also known as conducting a drill-off test. The response surfaces interpolate drilling performance between the varied set points, and the surface and recommendations guide drillers by use of set-point recommendations. Once the driller explores a sufficient space during the drill-off test, the recommendation remains in the vicinity of a near-optimal point given steady drilling conditions.
The current system uses two recommendation work flows: one for explore phase and one for exploit phase. The explore-phase recommendation work flow begins by estimating drilling performance at the current drilling parameters. The next two recommendations guide the driller to change the autodriller set point once and the topdrive set point once. Next, the system balances the exploit/explore tradeoff to continue expanding the surface area while making near-optimal recommendations on the basis of the surface trend.
Operating Near Dysfunction or Constraints
Experience from trials indicated that drillers may step drilling parameters systematically even without the system but did not always recognize the need to respond to dysfunction or parameter constraints while stepping parameters.
The trial cases demonstrated the need for the system to guide drilling operation out of dysfunction as possible if the driller starts exploring in dysfunction and to notify the driller when dysfunction persists for an extended time period.
Trials indicated the importance of making recommendations to drillers with a sufficient time period between recommendations to demonstrate stability and allow the driller to follow recommendations while managing drilling operations. Experience indicated that drillers were willing to continue making set-point changes with a minimum period of 5 minutes.
Consistent Driller Guidance
Feedback from trials indicated the need to provide the drillers a surface trend from the beginning of the explore phase. Early trials used response points instead of response surfaces to make explore-phase recommendations, and the recommendations could differ from the optimal direction per the surface. Driller feedback indicated the need to ensure that all system guidance is consistent and that explore recommendations are made on the basis of the surface trend. A new algorithm was developed to provide explore-phase recommendations consistent with the surface on the basis of projecting an optimal vector from the surface centroid.
Dynamic drilling dysfunction may be mitigated with proper selection of steady-state set points, and steady-state modeling on the basis of surface drilling data has proved to be effective for multivariable drilling optimization. The algorithm uses surface drilling data as a low-cost, vendor-neutral data source to estimate operation at the drilling assembly, which may be miles downhole from the surface wellsite. Model-based estimation of downhole drilling performance provides a proxy for downhole dynamics that are not measurable directly from the surface. Adaptive data-driven modeling enables gradual relearning as unmeasured disturbances—formation or bit-condition changes—alter the relationship between drilling parameters and performance. The objective function proves capable of making a tradeoff between relative measures of performance—ROP and drilling efficiency—and an absolute measure of performance, TSE.
This article, written by Special Publications Editor Adam Wilson, contains highlights of paper SPE 187447, “Challenges and Lessons From Implementing a Real-Time Drilling Advisory System,” by Benjamin J. Spivey, SPE, Gregory S. Payette, SPE, and Lei Wang, SPE, ExxonMobil Upstream Research Company; Jeffrey R. Bailey, SPE, ExxonMobil Development Company; Derek Sanderson, XTO Energy; and Stephen W. Lai, SPE, Behtash Charkhand, SPE, and Aaron Eddy, SPE, Pason Systems, prepared for the 2017 SPE Annual Technical Conference and Exhibition, San Antonio, Texas, USA, 9–11 October. The paper has not been peer reviewed.