Review and analysis of supervised machine learning algorithms for hazardous events in drilling operations

doi:10.1016/j.psep.2020.09.038

Process Safety and Environmental Protection

Volume 147, March 2021, Pages 367-384

https://doi.org/10.1016/j.psep.2020.09.038 Get rights and content

Abstract

Results of bibliometric analysis and a detailed review are reported on the use of supervised machine learning to study hazardous drilling events. The bibliometric analysis attempts to answer pertinent questions related to progress in the use of supervised machine learning for hazardous events due to drilling fluid density/mud weight. The analysis indicates artificial neural network as the most popular algorithm among researchers. Also, deep learning, random forest and support vector machine have gained momentum in recent use.

A critical review of literature on hazardous events and supervised machine learning algorithms are reported. This review was done to observe how the algorithms were used, their relative successes, limitations, as well as input parameters which aided in detection or estimation by the machine learning algorithms. An introduction to deep learning and a review of literature on the use of deep learning with respect to operations involving drilling parameters is presented. The review on deep learning and drilling parameters covered the following operations: lithology identification, drilling rig state determination, generating logging/other drilling parameters and detecting abnormality in data.

The study highlights need of publicly accessible large database with data from different oilfields for development of machine learning algorithms. These algorithms could be used globally for the enhancement of machine learning for new fields or fields with limited data. The availability of such large database would aid researchers in improving or customizing deep learning algorithms in line with the unique needs of drilling activities.

Introduction

The activities of the oil and gas industry can pertain to upstream, midstream or downstream. Upstream activities involve reservoir characterization, drilling and production of crude products. Midstream mainly involves processing, storage, marketing and transportation of the output from upstream. Downstream activities include receiving outputs from the midstream, refining the oil and performing distribution of petroleum products (PSAC, 2018). Several challenges encountered in the oil and gas industry can benefit from the use of machine learning. For example, in the area of drilling, machine learning can be applied towards pore pressure prediction (Ahmed et al., 2019a), in reservoir characterization, machine learning can be used in predicting reservoir properties at locations without core or appropriate log data (Osarogiagbon et al., 2015) and machine learning can also be used to forecast oil production rate (Mamudu et al., 2020).

Hazardous events are undesirable as they can lead to loss of time, loss of money, loss of human abilities and loss of lives. Sadly, the oil and gas industry has had its share of disastrous accidents which can be traced to certain events. A popular case is the Macondo blowout which resulted in an estimated loss of over 14 billion dollars (Mason, 2019). During drilling, hazardous events which are directly caused or highly influenced by the use of wrong drilling fluid density includes kick, formation fracture, lost circulation and stuck pipe (Abimbola et al., 2015). If these events are not properly monitored and controlled, they could lead to accidents. Predicting the occurrence of these events can be challenging due to the number of influencing parameters. Kick can occur due to several classes of factors related to hydrostatic head (e.g. abnormal pore pressure, insufficient mud density, lost circulation), cement (e.g. inadequate bonding, casing centralization), or pressure control equipment for managed pressure drilling (Tamim et al., 2019). This shows that robust mathematical models might be required to predict or detect kick as a function of causing factors.

The impact of hazardous events on oil and gas operations cannot be over-emphasized. Occurrence of blowouts or shutting/relieving a well to prevent blowout represents huge losses to the oil and gas industry. Thus, there are several publications which present methodologies for early detection, prediction, and mitigation of hazardous events in the oil and gas industry. Some studies showing methodologies for analysis and detection of drilling related hazards can be found in (Abimbola et al., 2015; Sun et al., 2018a). This work aims to summarize the efforts of authors in using supervised machine learning in the area of drilling with regards to hazardous events. This can thus provide pointers to where there are obvious rooms for improvement. Machine learning has attracted considerable interest in the oil and gas industry and Table 1 lists selective review papers on the use of machine learning for petroleum exploration/production applications.

Although several literature reviews on the use of machine learning or artificial intelligence for petroleum applications have been reported, to the best of our knowledge, (i) no bibliometric analysis on trend in usage of supervised machine learning and hazard events in drilling related activities of the petroleum industry is available, (ii) No general review on the use of deep learning on drilling parameters have been done. In addition, this work aims to present the limitations in the current usage of supervised machine learning for hazardous events with respect to drilling.

This work is structured as follows: Section 2 gives an introduction to machine learning, Section 3 gives an introduction to deep learning, Section 4 describes factors that can influence the use of machine learning, Section 5 describes bibliometric analysis used to obtain trend in use of machine learning algorithms, Section 6 describes results of bibliometric analysis, Section 7 presents review of articles on machine learning and hazardous events in drilling engineering, Section 8 presents review of articles on deep learning and drilling parameters, Section 9 presents gaps in machine learning implementations, and conclusion is given in Section 10.

Section snippets

Introduction to machine learning

It is very convenient to have a simple and well-defined equation which solves a problem. However, it is often difficult to do this for many real-life problems. For example, while it is easy for a person to identify members of a family from their looks, there is no simple equation for this (Russell and Norvig, 2010). These forms of challenges are a key reason why computation is moving in the direction of simulating how human reason. Although machine learning and artificial intelligence are used

Introduction to deep learning

Deep learning results from the application of multi hidden layer neural network for learning (Sze et al., 2017) (Q. Zhang et al., 2018b). The beauty of deep learning is that it works by exploiting several levels of interrelationships among input parameters. With this, deep learning can learn important features of the input parameters at different hierarchical levels of input data interaction. Deep learning has been successfully applied to several domains such as image recognition and audio

Factors that can influence the progress in the use of machine learning

Considering that machine learning requires a computing device to learn from data, the factors to be considered are training data and computing packages.

Training data presents two major forms of problems which are: the availability of sufficient data for training and the time/resources required to train massive amount of data. Fortunately this challenges can be ameliorated by public availability of data sets and the use of transfer learning. One question probably could be how large should a

Bibliometric analysis methodology

The databases considered for this work are Scopus and OnePetro. The main advantage of using Scopus is that it is the largest abstract and citation database of peer reviewed literature (Elsevier, 2020). OnePetro was used because it primarily focuses on oil and gas activities.

It is very tough to anticipate the precise words that authors used in capturing their intentions. Based on this, the search words were selected to obtain a fair representation of most documents of our interest. The database

The results of Bibliometric analysis

The bibliometric search was done on 27th July 2020 and the total search result for the different categories of supervised machine learning algorithms is presented in Fig. 4.

Fig. 4 clearly shows artificial neural network leading in popularity for hazardous events in drilling. A breakdown of the relative use of the algorithms over the years is shown in Fig. 5. This gives an indication of usage trend over time.

It can be observed in Fig. 5 that the relative use of deep learning, support vector

Review of machine learning and hazardous events in drilling engineering

When pressure gradient due to drilling fluid density is below pore pressure gradient, kick can occur. On the other hand, when pressure gradient due to drilling fluid density is more than formation fracture gradient, formation fracture can also occur. When these events are not properly controlled, blow out or well collapse could occur (Abimbola et al., 2014; Khakzad et al., 2013; Khoshnaw et al., 2014; J. Zhang and Yin, 2017). This therefore necessitates a drilling fluid density window of

Literature Review on deep learning and drilling parameters

In this section, we aim to observe how deep learning have been utilized with drilling parameters.

Table 8 shows deep learning algorithms been used with drilling parameters for drilling rig state determination (e.g. tripping), drilling event identification (e.g. kick), lithology identification, generating logging/other drilling parameters and detecting occurrence of abnormality in data.

Gaps in the use of machine learning in drilling events as observed in literature survey

Several issues were observed which were chiefly based on training data as well as the nature of deep learning algorithms. These are:

1
For most cases, each article used data different from that used in other articles. This can likely translate to machine learning algorithms which may only be successful for the field data they trained with.
2
Nearly all data set used for machine learning implementation had data points/instances less than a million, with many of the dataset having instances less than a

Conclusions

Results of bibliometric analysis in the area of supervised machine learning in hazard related events during drilling clearly indicate a growing trend in the use of machine learning. The results of a review of the literature on supervised machine learning for hazardous drilling events -- kick, fracture, lost circulation, and stuck pipe -- are reported. In addition, some studies in the application of supervised machine learning on pore pressure, equivalent circulation density and bottom hole

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors gratefully acknowledge the Niger Delta Development Commission for their financial support, as well as the financial support provided through Canada Research Chair Program in Offshore Safety and Risk Engineering. We also want to thank Dr. Olalere Sunday Oloruntobi for his technical contribution.

References (98)

A.K. Abbas et al.
Implementing artificial neural networks and support vector machines to predict lost circulation
Egypt. J. Pet.
(2019)
M. Abimbola et al.
Dynamic safety risk analysis of offshore drilling
J. Loss Prev. Process Ind.
(2014)
M. Abimbola et al.
Safety and risk analysis of managed pressure drilling operation using Bayesian network
Saf. Sci.
(2015)
S.A. Adedigba et al.
Data-driven dynamic risk analysis of offshore drilling operations
J. Pet. Sci. Eng.
(2018)
O.E. Agwu et al.
Artificial intelligence techniques and their applications in drilling fluid engineering: a review
J. Pet. Sci. Eng.
(2018)
Z. Alyafeai et al.
A fully-automated deep learning pipeline for cervical cancer classification
Expert Syst. Appl.
(2020)
R. Ashena et al.
Bottom hole pressure estimation using evolved neural networks by real coded ant colony optimization and genetic algorithm
J. Pet. Sci. Eng.
(2011)
C.P. Ezeakacha et al.
Experimental and statistical investigation of drilling fluid loss in porous media: part 2 (Fractures)
J. Nat. Gas Sci. Eng.
(2019)
S. Feng et al.
Using deep neural network with small dataset to predict material defects
Mater. Des.
(2019)
M. He et al.
Deep convolutional neural network for fast determination of the rock strength parameters using drilling data
Int. J. Rock Mech. Min. Sci.
(2019)

H.H. Alkinani et al.

Dynamic neural network model to predict the rate of penetration Prior to drilling

53rd US Rock Mechanics/Geomechanics Symposium

(2019)

Husam H. Alkinani et al.

Applications of artificial neural networks in the petroleum industry: a review

Husam H. Alkinani et al.

Application of artificial neural networks in the drilling processes: can equivalent circulation density be estimated prior to drilling?

Egypt. J. Pet.

(2019)

M.Z. Alom et al.

A state-of-the-art survey on deep learning theory and architectures

Electronics

(2019)

K. Arulkumaran et al.

Deep reinforcement learning: a brief survey

IEEE Signal Process. Mag.

(2017)

O. Bello et al.

Application of artificial intelligence methods in drilling system design and operations: a review of the state of the art

J. Artif. Intell. Soft Comput. Res.

(2015)

Y. Ben et al.

Building a General and sustainable machine learning solution in a real-time drilling system

IADC/SPE International Drilling Conference and Exhibition

(2020)

J. Brahma et al.

Design of safe well on the top of Atharamura anticline, Tripura, India, on the basis of predicted pore pressure from seismic velocity data

J. Pet. Explor. Prod. Technol.

(2018)

C.E. Bravo et al.

State of the art of artificial intelligence and predictive analytics in the E&P industry: a technology survey

Spe J.

(2014)

W.U. Chao et al.

A prediction of borehole stability while drilling preliminary prospecting wells based on seismic impedance

Pet. Explor. Dev.

(2015)

W. Chen et al.

Automatic drilling dynamics interpretation using deep learning

SPE Annual Technical Conference and Exhibition

(2019)

G. Chen et al.

A new method of lithology classification based on convolutional neural network algorithm by utilizing drilling string vibration data

Energies

(2020)

J. Chung et al.

Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling

(2014)

A. Creswell et al.

Generative adversarial networks: an overview

IEEE Signal Process. Mag.

(2018)

G. Ditzler et al.

Multi-layer and recursive neural networks for metagenomic classification

IEEE Trans. Nanobioscience

(2015)

Y. Du et al.

A new item-based deep network structure using a restricted Boltzmann machine for collaborative filtering

Front. Inf. Technol. Electron. Eng.

(2017)

Elsevier

About Scopus

(2020)

J. Fang et al.

Discrete fracture network modelling in a naturally fractured carbonate reservoir in the Jingbei oilfield, China

Energies

(2017)

Cited by (71)

Anomaly detection in multivariate time series of drilling data
2024, Geoenergy Science and Engineering
Different clusters of abnormal activities often arise within same temporal domain of drilling operations. This contrasts with employing simplified scenarios, such as anomaly detection models for specific issues like stuck pipe or loss circulation. The dynamic nature of drilling environments demands a holistic framework that can adapt to evolving conditions and detect anomalies beyond predefined categories. There exists a need to evaluate the performance of various data-driven models and achieve a broader applicability of anomaly detection instead of focusing on a single abnormal activity type. This study presents a generalized framework designed to detect anomalous events in drilling operations. Three different unsupervised learning algorithms, principal component analysis (PCA), isolation forest (IF) and lstm-autoencoder (LSTM-AE), are used to determine abnormal patterns and irregularities in the drilling data. PCA is chosen for its interpretability and efficiency in handling high-dimensional data by reducing the dimensionality. IF is selected for its effectiveness in isolating anomalies, making it robust for scenarios where anomalies exhibit distinct patterns. LSTM-AE is employed due to its capability to capture temporal dependencies and nonlinear patterns in time series data. A multivariate time series of drilling data from an offshore well is collected to use in this work. The dataset comprises approximately six months of drilling operations, including drilling and non-drilling periods. The input variables are chosen from the common controllable parameters continuously monitored by the drilling crew. The pre-processed drilling data with input features is organized into groups based on each drilling section. These unlabeled data is split into train, validation and test sets. Cross-validation (CV) is applied into train and validation sets to prevent overfitting and tune the hyperparameters. The test data is used to evaluate the generalizability and robustness of each model. Evaluation metrics such as precision, recall, F1 score, and receiver operating characteristic (ROC) curve are used to select the best-performing model. The thresholds for anomaly identification are determined by analyzing the reconstruction errors from PCA and LSTM-AE models, as well as anomaly scores from the IF algorithm. Subsequently, evaluation metrics are calculated using these thresholds, assigning true and false predictions to each data point. These predictions are then compared with the ground truth data labels obtained from the daily reports. Experimental results show that the proposed approach can effectively predict anomalous events in drilling operations. It is observed that there is a causal link between irregularities in time series drilling data and the occurrence of anomalous events in drilling operations. Implementing the proposed approach in real-time monitoring systems may improve overall efficiency and minimize risks in drilling operations. The presented methodology can be tested with diverse drilling datasets to enhance the generalizability and reduce bias, leading to a more comprehensive and reliable assessment across various scenarios.
Performance analysis of various machine learning algorithms for CO<inf>2</inf> leak prediction and characterization in geo-sequestration injection wells
2024, Process Safety and Environmental Protection
The effective detection and prevention of CO₂ leakage in active injection wells are paramount for safe carbon capture and storage (CCS) initiatives. This study assesses five fundamental machine learning algorithms, namely, Support Vector Regression (SVR), K-Nearest Neighbor Regression (KNNR), Decision Tree Regression (DTR), Random Forest Regression (RFR), and Artificial Neural Network (ANN), for use in developing a robust data-driven model to predict potential CO₂ leakage incidents in injection wells. Leveraging wellhead and bottom-hole pressure and temperature data, the models aim to simultaneously predict the location and size of leaks. A representative dataset simulating various leak scenarios in a saline aquifer reservoir was utilized. The findings reveal crucial insights into the relationships between the variables considered and leakage characteristics. With its positive linear correlation with depth of leak, wellhead pressure could be a pivotal indicator of leak location, while the negative linear relationship with well bottom-hole pressure demonstrated the strongest association with leak size. Among the predictive models examined, the highest prediction accuracy was achieved by the KNNR model for both leak localization and sizing. This model displayed exceptional sensitivity to leak size, and was able to identify leak magnitudes representing as little as 0.0158% of the total main flow with relatively high levels of accuracy. Nonetheless, the study underscored that accurate leak sizing posed a greater challenge for the models compared to leak localization. Overall, the findings obtained can provide valuable insights into the development of efficient data-driven well-bore leak detection systems.
Critical review on data-driven approaches for learning from accidents: Comparative analysis and future research
2024, Safety Science
Data-driven intelligent technologies are promoting a disruptive digital transformation of human society. Industrial accident prevention is also amid this change. Although many emerging technologies, such as machine learning (ML), are extensively employed in workplace safety, these approaches need to fit the intended safety purpose of accident analysis, risk assessment, adverse outcome prediction, or anomaly detection. Hence, examining the “real-world” need for accident prevention and the advantages of emerging data-driven methodologies to better integrate them is necessary. This study provides a systematic review to clarify the current research status, existing problems, and future insights into these evolving technologies in accident prevention. We present notable gaps and barriers in data-driven accident prevention by analyzing 194 published studies from four perspectives: Paradigm, Model, Data Source, and Purpose. The results demonstrate (1) lack of a systematic framework to guide the application of Big Data (BD) in the field of safety; (2) few prior studies have considered model interpretability; (3) more proactive data needs to be incorporated into accident analysis; (4) safety-related data and domain knowledge need to be further integrated; (5) some recent data-driven techniques are unexplored in safety science. Further, the future research opportunities are discussed based on these findings. Such review may help clarify the mapping of data-driven tasks to safety goals to accelerate the uptake of data-driven technologies in safety or accident analysis research.
Self-adaptive early warning of undesirable gas-liquid flow pattern in offshore oil and gas pipeline-riser system
2024, Process Safety and Environmental Protection
It is an important function of submarine pipeline flow assurance system to monitor undesirable gas-liquid flow patterns in pipeline in real time, and release early warning ahead of the formation of these flow patterns. In order to tackle the problem of existing method that requires a large training set, and A method of self-adaptive early warning of transition to unstable regime is developed based on transient experiment of gas-liquid two-phase flow performed on a 150-m-long laboratorial pipeline-riser system. The responses of different differential pressure signals were investigated first, so as to pick out the optimum ones used for early warning. Then, the long-short time memory (LSTM) neural network combined with empirical mode decomposition (EMD) was adopted to judge transient condition. In the next step, a series of EMD-LSTM models were trained in order to predict whether undesirable flow pattern would occur, and the warning would be released if the result was ‘yes’. Within the range of experiments, all transient cases were correctly judged. The principles of proposed method were then tested and verified by data from a longer laboratorial pipeline-riser system and an offshore oil field, indicating good adaptability of the method.
Machine learning framework for estimating CO<inf>2</inf> adsorption on coalbed for carbon capture, utilization, and storage applications
2023, International Journal of Coal Geology
For the purpose of carbon capture, utilization, and storage, carbon dioxide (CO₂) injection in coal formations can enhance methane recovery and mitigate climate change. However, measuring CO₂ adsorption isotherms using experimental or mathematical models can be time-consuming, expensive, and inaccurate. Thus, this study presents a machine-learning framework that predicts CO₂ adsorption in coal formations based on various coal properties and testing conditions. Machine-learning (ML) framework was applied using a dataset of 1,064 points collected for different coal samples at different operating conditions to predict the CO₂ adsorption in coal surface. The ML techniques include decision tree regression (DT), random forests (RF), gradient boost regression (GBR), K-nearest neighbor (KNN), artificial neural network (ANN), function network (FN), and adaptive neuro-fuzzy inference system (ANFIS). The applied framework determines CO₂ adsorption as a function of coal's physical and chemical properties (moisture, ash, volatile matter, and fixed carbon content), the vitrinite reflectance of the coal samples, and testing conditions (pressure and temperature). Classical statical tools such as R², root mean square error (RMSE), and average absolute percentage error (AAPE) were used to evaluate the model's performance analysis.
The results demonstrated the ability to determine CO₂ adsorption for varying coal types and at different temperature and pressure conditions. The statistical measures suggested that RF, GBR, and KNN are very reliable ML models, with RF being the best. At low operating pressure (P < 4 MPa), CO₂ adsorption is impacted by any pressure changes, while it is stabilized at high-pressure values and becomes more dependent on the rock properties at high operating pressure (P > 4 MPa). The introduced ML framework offers a technique to evaluate the capability of different algorithms and accurately estimate CO₂ adsorption without the requirement of additional experimental measurements or complicated mathematical techniques.
Early warning of drillstring faulty conditions based on multi-model fusion in geological drilling processes
2023, Journal of Process Control
The early warning of drilling faults is of paramount importance to prevent or reduce environmental and property damages. In view of the complex drillstring kinematic characteristics and changeable formation environment, this paper proposes a new fault early warning method for the drillstring system based on a multi-model fusion and self-updating strategy. The major contributions are twofold: (1) A hybrid drillstring early warning method is proposed to identify faulty conditions based on the fusion of multi-dimensional prediction models of the drillstring system; (2) An event-triggered model self-updating strategy is proposed to mitigate the prediction performance degradation caused by formation uncertainties. The effectiveness and practicability of the proposed method are demonstrated by industrial case studies, and the results indicate that the proposed method outperforms other approaches.

View all citing articles on Scopus

View full text

Review and analysis of supervised machine learning algorithms for hazardous events in drilling operations

Abstract

Introduction

Section snippets

Introduction to machine learning

Introduction to deep learning

Factors that can influence the progress in the use of machine learning

Bibliometric analysis methodology

The results of Bibliometric analysis

Review of machine learning and hazardous events in drilling engineering

Literature Review on deep learning and drilling parameters

Gaps in the use of machine learning in drilling events as observed in literature survey

Conclusions

Declaration of Competing Interest

Acknowledgements

Egypt. J. Pet.

J. Loss Prev. Process Ind.

Saf. Sci.

J. Pet. Sci. Eng.

J. Pet. Sci. Eng.

Expert Syst. Appl.

J. Pet. Sci. Eng.

J. Nat. Gas Sci. Eng.

Mater. Des.

Int. J. Rock Mech. Min. Sci.

Saf. Sci.

Neurocomputing

Measurement

J. Pet. Sci. Eng.

J. Pet. Sci. Eng.

J. Pet. Sci. Eng.

Inf. Sci. (Ny)

J. Pet. Sci. Eng.

J. Pet. Sci. Eng.

Process. Saf. Environ. Prot.

Can. Assoc. Radiol. J.

J. Pet. Sci. Eng.

J. Nat. Gas Sci. Eng.

J. Loss Prev. Process Ind.

Saf. Sci.

Inf. Fusion

New approach to evaluate the equivalent circulating density (ECD) using artificial intelligence techniques

J. Pet. Explor. Prod. Technol.

Estimation of the fracture density in reservoir rock using regression analysis of the petrophysical data

Geod. Cartogr.

Comparative analysis of artificial intelligence techniques for formation pressure prediction while drilling

Arab. J. Geosci.

New model for pore pressure prediction while drilling using artificial neural networks

Arab. J. Sci. Eng.

Reducing consumed energy while drilling an oil well through a deep rig time analysis

Advances in Petroleum Exploration and Development

Dynamic neural network model to predict the rate of penetration Prior to drilling

53rd US Rock Mechanics/Geomechanics Symposium

Applications of artificial neural networks in the petroleum industry: a review

Application of artificial neural networks in the drilling processes: can equivalent circulation density be estimated prior to drilling?

Egypt. J. Pet.

A state-of-the-art survey on deep learning theory and architectures

Electronics

Deep reinforcement learning: a brief survey

IEEE Signal Process. Mag.

Application of artificial intelligence methods in drilling system design and operations: a review of the state of the art

J. Artif. Intell. Soft Comput. Res.

Building a General and sustainable machine learning solution in a real-time drilling system

IADC/SPE International Drilling Conference and Exhibition

Design of safe well on the top of Atharamura anticline, Tripura, India, on the basis of predicted pore pressure from seismic velocity data

J. Pet. Explor. Prod. Technol.

State of the art of artificial intelligence and predictive analytics in the E&P industry: a technology survey

Spe J.

A prediction of borehole stability while drilling preliminary prospecting wells based on seismic impedance

Pet. Explor. Dev.

Automatic drilling dynamics interpretation using deep learning

SPE Annual Technical Conference and Exhibition

A new method of lithology classification based on convolutional neural network algorithm by utilizing drilling string vibration data

Energies

Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling

Generative adversarial networks: an overview

IEEE Signal Process. Mag.

Multi-layer and recursive neural networks for metagenomic classification

IEEE Trans. Nanobioscience

A new item-based deep network structure using a restricted Boltzmann machine for collaborative filtering