# Drilling performance monitoring and optimization: a data-driven approach

- 1.5k Downloads

## Abstract

Drilling performance monitoring and optimization are crucial in increasing the overall NPV of an oil and gas project. Even after rigorous planning, drilling phase of any project can be hindered by unanticipated problems, such as bit balling. The objective of this paper is to implement artificial intelligence technique to develop a smart model for more accurate and robust real-time drilling performance monitoring and optimization. For this purpose, the back propagation, feed forward neural network model was developed to predict rate of penetration (ROP) using different input parameters such as weight on bit, rotations per minute, mud flow (GPM) and differential pressures. The heavy hitter features identification and dimensionality reduction are performed to understand the impacts of each of the drilling parameters on ROP. This will be used to optimize the input parameters for model development and validation and performing the operation optimization when bit is underperforming. The model is first developed based on the drilling experiments performed in the laboratory and then extended to field applications. From both laboratory and field test data provided, we have proved that the data-driven model built using multilayer perceptron technique can be successfully used for drilling performance monitoring and optimization, especially identifying the bit malfunction or failure, i.e., bit balling. We have shown that the ROP has complex relationship with other drilling variables which cannot be captured using conventional statistical approaches or from different empirical models. The data-driven approach combined with statistical regression analysis provides better understanding of relationship between variables and prediction of ROP.

## Keywords

Data-driven technology Drilling Drill-bit dysfunction Drilling performance monitoring## Abbreviations

- AI
Artificial intelligence

- ANN
Artificial neural network

- MLP
Multilayer perceptron

- ROP
Rate of penetration (ft/h)

- WOB
Weight on bit (klb)

- RPM
Rotation per minute

- GPM
Gallons per minute

- DOC
Depth of cutting (in)

- DiffPress
Differential pressure (Psi)

- SPP
Standpipe pressure (Psi)

## Introduction

- 1.
Change in drilling fluid rheology.

- 2.
Use of oil-based mud in water reactive clays/shale.

- 3.
Developing electric potential between formation and drill bit.

- 4.
Modifications in drill bit hydraulics.

Unlike conventional techniques mentioned above, this work is aimed to look at the process of bit balling based on data collected during the process and evaluate different parameters that may result in bit balling. Since the techniques used to quantify the bit balling in terms of drilling parameters are just relied on X–Y cross-plots and there is no analytical or semi-analytical technique that can quantify the bit balling as a function of WOB, torque, ROP, RPM, mud flow and surface-controlled pressures such as swivel, choke and borehole pressure, we have chosen artificial intelligence (AI) to build this relationship. The main objective for the proposal is to provide a holistic outline for using real-time drilling data for early detection of bit balling and optimizing the drilling parameters instantaneously to prevent bit dysfunction.

## Background

Previously, different methodologies have been devised that used direct or indirect approaches to evaluate ROP. ROP is a function of drilling parameters, drilling fluid type and most importantly the properties of rock being drilled. That’s why it is the direct indicator of rock’s mechanical property. This leads to improved drilling parameters, bit design and fluid type to achieve desirable ROP in all kinds of formations. One of the first attempts for the drilling optimization was presented in the study of Graham and Muench (1959), where they analytically evaluated the weight on bit and rotary speed combinations to derive empirical mathematical expressions for bit life expectancy and for drilling rate as a function of depth, rotary speed and bit weight. Galle and Woods (1963) produced graphs and procedures for field applications to determine the best combination of drilling parameters. Bourgoyne and Young (1974) came with an idea of evaluating ROP as a function of eight variables, where these parameters were the result of multiple regression analysis. The equation developed by them was valid for roller cone bits. They used minimum cost formula, showing that maximum rate of penetration may coincide with minimum cost approach, if the technical limitations were ignored. In the mid-1980s, operator companies developed techniques of drilling optimization in which their field personnel could perform optimization at the site referring to the graph templates and equations. In 1990s, different drilling planning approaches were brought to surface [Carden et al. (2006)]. New techniques identified the best possible well construction performances. Later on, “Drilling the Limit” optimization techniques were also introduced Schreuder and Sharpe (1999). Toward the end of the millennium, real-time monitoring techniques started to take place, e.g., drilling parameters started to be monitored from off locations. A few years later, real-time operations/support centers started to be constructed. Some operators proposed advanced techniques in monitoring of drilling parameters at the rig site. Following the early developments in rotary drilling systems, some operators proposed advanced techniques in monitoring of drilling parameters at the rig site.

- 1.
Developing dynamic predictive models for bit dysfunction diagnosis in different laboratory tests.

- 2.
Developing a workflow for early detection of bit failure in the field.

To achieve these objectives, the actual laboratory and field data consist of WOB, ROP, Torque, RPM, Swivel, Choke and Borehole Pressure as a function of time which is provided by National Oilwell Varco used for building the intelligent models to predict ROP in both laboratory and field conditions and the predicted behavior of ROP as function of time is used for drill bit dysfunction diagnosis.

## Methodology

### Artificial neural networks (ANNs)

### Multilayer perceptron

## Results and discussions

### Drill bit test data

#### Preprocessing

- 1.
Charge pressures (Psi)

- 2.
Rotations per minute (RPM)

- 3.
Mud flow (GPM)

- 4.
Weight on bit “WOB” (klb)

- 5.
Bore hole pressure (Psi)

- 6.
Swivel pressure (Psi)

- 7.
Choke pressure (Psi)

- 8.
Torque (klb ft)

- 9.
Penetration (in)

- 10.
Depth of cutting (in)

Since the laboratory tests are obtained in more defined conditions, we have used laboratory data to first develop, train and validate the model. This step served as our proof of concept study, and then, we have expanded our studies using field data that have been obtained in more complex environment in comparison with laboratory conditions for actual application of drilling performance monitoring and optimization in a real time. As discussed earlier using the pair plots, we did not identify any outliers and we did not have any missing data that require any imputation technique. An initial analysis was completed to determine correlation between different parameters within the database. It is found that most of all the parameters have correlations < 90% so we decide to keep them during the model development.

### Development of model

Neural network architecture

Feature | Value/model |
---|---|

Neurons | 15 |

Solver | ADAM |

Activation | Relu |

### Post processing

### Case of drilling dysfunction

Bit balling is characterized as slowness of penetration rate. Many parameters contribute to slow ROP, for example, formation characteristics, bit type, drilling fluid properties, drill bit hydraulics, operating conditions, etc. As discussed earlier, the first half of the data is used for training and verification of the model in each experiment as we do not expect the bit failure or malfunction occurs at early time of the bit usage. The failure and malfunction usually happen after new drill bit has been used for a while in the drilling job. The trained model is then used for prediction of the ROP in the second half of the data that we expect the failure or malfunction might occur. As long as the actual experimental measure of the ROP not used for training the model follows the predictions of ROP obtained using the trained model, we do not expect any bit malfunction or failure. However, as soon as the actual measure of ROP starts deviating from the model predictions, this can be used as indication of bit not performing as expected and seen in early stage of drilling job, i.e., the first half of the data.

### Heavy hitter features (HHF) identification

## Field case study

## Conclusions

From both laboratory and field test data provided, we have proved that the data-driven model built using MLP technique can be successfully used for drilling performance monitoring and optimization. The model can be also used for uncertainty quantification and sensitivity analysis in the laboratory conditions where the limitations of the operation conditions in the laboratory or time required to complete the test do not allow the full sensitivity analysis or uncertainty quantification studies. The model can also be used in the field in a real time to monitor the bit performance raise the flag in case of bit underperforming to avoid any possible bit malfunction or failure. We have shown that the ROP has complex relationship with other drilling variables which cannot be captured using conventional statistical approaches or from different empirical models. The data-driven approach combined with statistical regression analysis provides better understanding of relationship between variables and prediction of ROP.

## Notes

### Acknowledgements

Authors of this paper would like to acknowledge the National Oilwell Varco for providing the data and permission for this publication.

## References

- Ansari A, Fathi E, Belyadi F, Takbiri-Boroujeni A, Belyadi H, Walker K (2017) Data-based smart model for real time liquid loading diagnostics in Marcellus Shale via machine learning. SPE unconventional resources conference to be held 13–14 March 2018 in Calgary, Alberta, CanadaGoogle Scholar
- Belyadi H, Fathi E, Belyadi F (2016) Hydraulic fracturing in unconventional reservoirs: theories, operations, and economic analysis. Gulf Professional Publishing, HoustonGoogle Scholar
- Bourgoyne AT Jr, Young FS Jr (1974) A multiple regression approach to optimal drilling and abnormal pressure detection. Soc Petrol Eng J 14(4):371–384CrossRefGoogle Scholar
- Carden RS, Grace RD, Shursen JL (2006) Drilling practices. Petroskills-OGCI, Course Notes, Tulsa, OK, pp 1–8Google Scholar
- Galle EM, Woods AB (1963) Best constant weight and rotary speed for rotary rock bits. Drilling and production practice, API 1963, pp 48–73Google Scholar
- Graham JW, Muench NL (1959) Analytical determination of optimum bit weight and rotary speed combinations. In: SPE 1349-G, fall meeting of the society of petroleum engineers, Dallas, TXGoogle Scholar
- Roy S, Cooper GA (1993) Prevention of bit balling in shales: some preliminary results. Society of Petroleum Engineers. https://doi.org/10.2118/23870-pa CrossRefGoogle Scholar
- Schreuder JC, Sharpe PJ (1999) Drilling the limit—a key to reduce well costs. SPE 57258, Asia Pacific improved oil recovery conference, Malaysia, October 1999Google Scholar

## Copyright information

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.