Review and analysis of supervised machine learning algorithms for hazardous events in drilling operations
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)
- et al.
Implementing artificial neural networks and support vector machines to predict lost circulation
Egypt. J. Pet.
(2019) - et al.
Dynamic safety risk analysis of offshore drilling
J. Loss Prev. Process Ind.
(2014) - et al.
Safety and risk analysis of managed pressure drilling operation using Bayesian network
Saf. Sci.
(2015) - et al.
Data-driven dynamic risk analysis of offshore drilling operations
J. Pet. Sci. Eng.
(2018) - et al.
Artificial intelligence techniques and their applications in drilling fluid engineering: a review
J. Pet. Sci. Eng.
(2018) - et al.
A fully-automated deep learning pipeline for cervical cancer classification
Expert Syst. Appl.
(2020) - et al.
Bottom hole pressure estimation using evolved neural networks by real coded ant colony optimization and genetic algorithm
J. Pet. Sci. Eng.
(2011) - et al.
Experimental and statistical investigation of drilling fluid loss in porous media: part 2 (Fractures)
J. Nat. Gas Sci. Eng.
(2019) - et al.
Using deep neural network with small dataset to predict material defects
Mater. Des.
(2019) - et al.
Deep convolutional neural network for fast determination of the rock strength parameters using drilling data
Int. J. Rock Mech. Min. Sci.
(2019)
Quantitative risk analysis of offshore drilling operations: a Bayesian approach
Saf. Sci.
A survey of deep neural network architectures and their applications
Neurocomputing
Structural damage identification based on unsupervised feature-extraction via Variational Auto-encoder
Measurement
Estimating the fracture gradient coefficient using neural networks for a field in the Middle East
J. Pet. Sci. Eng.
Dynamic risk assessment of reservoir production using data-driven probabilistic approach
J. Pet. Sci. Eng.
Application of decision tree, artificial neural networks, and adaptive neuro-fuzzy inference system on predicting lost circulation: a case study from Marun oil field
J. Pet. Sci. Eng.
Robust acoustic event classification using deep neural networks
Inf. Sci. (Ny)
A data processing algorithm proposed for identification of breakout zones in tight formations: a case study in Barnett gas shale
J. Pet. Sci. Eng.
A new pattern recognition model for gas kick diagnosis in deepwater drilling
J. Pet. Sci. Eng.
Developing leading indicators-based decision support algorithms and probabilistic models using Bayesian network to predict kicks while drilling
Process. Saf. Environ. Prot.
Canadian Association of Radiologists white paper on artificial intelligence in radiology
Can. Assoc. Radiol. J.
Estimation of the fracture density in fractured zones using petrophysical logs
J. Pet. Sci. Eng.
A DBN-based risk assessment model for prediction and diagnosis of offshore drilling incidents
J. Nat. Gas Sci. Eng.
Real-time risk analysis method for diagnosis and warning of offshore downhole drilling incident
J. Loss Prev. Process Ind.
A dynamic and quantitative risk assessment method with uncertainties for offshore managed pressure drilling phases
Saf. Sci.
A survey on deep learning for big data
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
Front. Inf. Technol. Electron. Eng.
About Scopus
Discrete fracture network modelling in a naturally fractured carbonate reservoir in the Jingbei oilfield, China
Energies
Cited by (71)
Anomaly detection in multivariate time series of drilling data
2024, Geoenergy Science and EngineeringPerformance 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 ProtectionSelf-adaptive early warning of undesirable gas-liquid flow pattern in offshore oil and gas pipeline-riser system
2024, Process Safety and Environmental ProtectionMachine learning framework for estimating CO<inf>2</inf> adsorption on coalbed for carbon capture, utilization, and storage applications
2023, International Journal of Coal GeologyEarly warning of drillstring faulty conditions based on multi-model fusion in geological drilling processes
2023, Journal of Process Control