Real-Time Data Analytics Allows for Bit-Wear Monitoring and Prediction

Topics: Drilling

You have access to this full article to experience the outstanding content available to SPE members and JPT subscribers.

To ensure continued access to JPT's content, please Sign In, JOIN SPE, or Subscribe to JPT

Severe bit damage is an issue in West Texas land drilling because of abrasive sand formation and interbedded hard stringers. Operational performance and rig costs often are affected by bits damaged beyond repair (DBR), low rates of penetration (ROPs) with worn bits, and inefficient decision-making regarding tripping. A real-time data-analytics application is developed that aims to provide information to operators to expedite decision-making.


As bottomhole-assembly (BHA) design and bit selection have become standardized, a historical data set of surface mechanics data and bit records has been accumulated from 40 bit runs. By combining conventional physical modeling of drilling mechanics and supervised machine learning, a hybrid analysis is conducted to separate bit-failure patterns from normal formation transitions and drilling-parameter adjustments. A metric-based algorithm is constructed for real-time monitoring of bit performance and for predicting bit wear.

A lightweight web-based framework is used for deployment in real time. A shadow mode trial on three wells in the same pad was conducted and generated satisfactory results.

Agile-Development Framework

The agile-development framework is an integrated platform for fast technology prototyping, which consists of the following components:

  • The Amazon cloud server is the platform for the application repository, computing-engine execution, and data stream and storage. Most real-time data-analytics applications are delivered as software-as-service, where the algorithm-processed data is stored on the cloud server and pertinent results are delivered to end-users over the web. User interaction with the web application is limited to default viewing options, and other data exchanges such as value inputs are kept to a minimum to reduce complexity.
  • MATLAB and Python are the engines for algorithm development, prototyping, and early-stage deployment. MATLAB is a powerful engineering computing/coding program with strong capability in data management, visualization, and debugging. It provides various toolboxes for signal processing, machine learning, and statistical analysis. Python is an alternative to MATLAB with similar functionality and the advantage of being open source. Codes of data preprocessing, core algorithms, and data stream input/output (I/O) are realized in the MATLAB environment running on the Amazon cloud server.
  • Wellsite Information Transfer Standard Markup Language (WITSML) I/O is the industrial standard for drilling-data stream and management. In this application, surface mechanics drilling data are streamed into MATLAB using a WITSML wrapper.
  • Plotly is a solution for visualization on the web and user interface. Computation results from MATLAB are delivered to a web page by calling Plotly functions within the MATLAB environment.

Historical Data Learning

The intermediate section of West Texas consists of approximately 6,000-ft vertical drilling through shale, abrasive sandstone, hard carbonates, and sandy formations to the kickoff point. Because of severe bit wear, a trip is often planned after encountering an abrasive formation to fit a second bit to maximize drilling economics. The challenge is how to determine the wear condition of the bit downhole and to optimize the decision of when to trip. If the trip is too early and a green bit is pulled, tripping time often cannot be compensated by the economics of using a second new bit. If the trip is too late, the bit may be DBR, resulting in an extra charge for the operator. In practice, the decision often relies on an individual engineer or the company representative’s subjective judgement and may be subject to distractions such as differing drilling parameters or varying formation responses.

In order to quantify bit wear, a data set of 40 bit runs drilled in 2016 and 2017 was collected. The input data consisted of bit dull grade, bit records, BHA information, and surface mechanics drilling parameters.

A hybrid approach combining both engineering metrics and machine learning was adopted to quantify bit wear. The engineering metrics have clear physical meanings and robustness for adjustment in real time. Machine learning enables the historical data set to be used fully to capture consistent patterns of bit behavior in a statistical sense.

Several mechanics-based metrics have been widely reported in the literature of drilling engineering.

  • Wear factor (WF) shows that the wear of a polycrystalline-diamond-compact cutter is related to the penetration force applied and the sliding speed between it and the formation. WF can be viewed as an estimated wear rate in terms of volume removed from the cutter per unit time.
  • Aggressiveness relates the torque generated by a bit to the amount of weight applied on bit and can be viewed as a bit-friction factor, which quantifies how well the bit can bite into the formation.
  • Mechanical specific energy is the amount of energy input from the surface to remove a unit volume of rock and is known to be a good indicator of drilling efficiency.

These three metrics, along with other commonly used measures, are indicators of bit/rock interaction, which means that both bit-cutter condition and formation strength and hardness are captured. Which metric is more sensitive to cutter structural damage and which better facilitates separating bit wear from formation change are the main tasks of historical-data analysis using machine learning.

An initial screening by visualizing metrics vs. depth and labeled with dull grade identifies that the WF is most sensitive to cutter wear. Its value variation can be quantified to distinguish the different influence from formation change and cutter wear damage. As shown in Fig. 1, the influence of formation change on the wear factor value is within a median range and shows steady behavior within the same formation sequence, illustrated by the green line. On the other hand, cutter wear damage pushes the WF value to a high range and often exhibits an exponential peak pattern corresponding to catastrophic failure, illustrated by the red line.

Fig. 1—Comparison of green-bit and worn-bit performances.


After WF is chosen as the main metric, further analysis shows that its value variation with respect to time can capture two types of bit failure. One type results from accumulated wear, where WF value steadily increases and then peaks at a high value, indicating that severe cutter damage has occurred. Another type of failure results from a fast wearing out at a hard stringer, where the WF value stays in a low range until jumping to a peak value, indicating a sudden failure.

With the clear physical meaning of the chosen metric and its pattern identified through bit-wear behavior, the remaining task is to quantify the value of WF with a specific bit-wear condition and construct a robust algorithm to predict severe bit wear in a time frame. This is a classic supervised learning problem, where each bit run is an independent sample, WF is the data of each sample, and bit dull grade is the label of states. Two rounds of learning using a decision tree are conducted. In the first round, the maximum value of WF in each bit run is used as the data for each sample.

In the second round, accumulated on-bottom drilling hours are calculated separately and used as data vectors for each sample of bit run. Decision-tree training is conducted to determine the feature on a time frame to trigger each state of bit wear. The outcome is a hierarchical criterion of tolerated, accumulated on-bottom drilling hours at the WF ranges to predict severe bit wear.

By using this technique, the following success rate of prediction was achieved with respect to the given training data set:

  • Predicted severe wear: 93%
  • Predicted median wear: 83%
  • Predicted green condition or minor wear: 90%

Real-Time Solution

The results are converted into a real-time algorithm to predict bit wear. Time-based drilling-mechanics parameters are first processed with calculated WF and accumulated on-bottom drilling hours at each feature range. A hierarchical criterion is constructed to trigger warnings of severe bit wear. The information is delivered over a website. Key parameters are plotted vs. depth. A comparison with data from offset wells also is presented in the same plots. Formation top information is plotted as reference. WF value variations and accumulated hours at identified value ranges trigger a warning.

For a limited time, the complete paper SPE 189602 is free to SPE members.

This article, written by Special Publications Editor Adam Wilson, contains highlights of paper SPE 189602, “Real-Time Bit-Wear Monitoring and Prediction Using Surface Mechanics Data Analytics: A Step Toward Digitization Through Agile Development,” by Yu Liu, Justin Kibbey, Yanbin Bai, and Xianping Wu, SPE, Shell, prepared for the 2018 IADC/SPE Drilling Technology Conference and Exhibition, Fort Worth, Texas, USA, 6–8 March. The paper has not been peer reviewed.

Real-Time Data Analytics Allows for Bit-Wear Monitoring and Prediction

01 December 2018

Volume: 70 | Issue: 12


Don't miss out on the latest technology delivered to your email weekly.  Sign up for the JPT newsletter.  If you are not logged in, you will receive a confirmation email that you will need to click on to confirm you want to receive the newsletter.