Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Scientific Reports volume 14, Article number: 26624 (2024 ) Cite this article rig drilling
As technology advances and options increase, the selection process becomes more difficult and time-consuming, drilling is not an exception. Directional drilling is the norm of the industry. Most wells drilled around the globe are directional, thus directional tools never stopped emerging in the market. The most widely used technologies for drilling deviated wells are positive displacement mud motors (PDM) and rotary steerable systems (RSS). Operators differ in which technology do they prefer because each has its advantages. Generally speaking, RSS technology is continuing to occupy the market due to various reasons that will be discussed later on.
The challenge for operators lies in determining which tool will provide the optimal performance for specific well conditions. Historically, this decision has been based on expert judgment and past experience, which can introduce biases and inconsistencies. As the drilling industry evolves, massive data is generated which should be employed for a better decision-making process1. With the rise of digital drilling, this data now offers an opportunity to improve the decision-making process through machine learning (ML) techniques.
This paper addresses the need for a data-driven approach to directional tool selection by developing a machine learning model capable of predicting the most suitable tool—RSS or PDM—for any given wellbore section. The novelty of this study lies in its ability to predict and compare total section time and cost for both PDM and RSS by analyzing historical well data. This model not only eliminates subjective bias but also adapts to specific well conditions, offering tailored recommendations for cost-effective drilling.
To qualitatively define the favorable conditions for using PDM versus RSS, based on a comprehensive literature review and field experience.
To develop and validate a machine learning model that predicts section time and cost for directional drilling using PDM and RSS.
To implement a data-driven framework that guides the selection of the optimal directional drilling tool, improving both operational efficiency and cost savings.
Since the introduction of RSS, drilling professionals and researchers have been comparing the effectiveness of PDM and RSS in different environments. The operation principle of PDM and RSS differs in that, PDM principle is to create side-force through bit offset or tilt by incorporating a bent housing and allow the well to change trajectory where the drill string does not rotate. On the other hand, RSS has two variations: Push-the-bit and point-the-bit. Push-the-bit type is centered around producing bending force on bottom hole assembly (BHA) by activating of three pads (ribs), this force is almost proportional to the extent of produced curvature, where, point-the-bit type RSS concept is initiating a bit deflection through using a pivot or a cam to move a steering shaft eccentrically inside a housing called the steering clutch2.
The evolution of PDM & RSS has shown continuous improvement that allowed us to drill record-breaking extended-reach wells using both technologies. For example, high-performance PDM (HP-PDM) have been in the market since ’90s and are characterized by a uniform thickness of elastomer rubber which enable it to give higher torque to the bit, thus it can handle more aggressive bits and yield better rate of penetration (ROP)3. Stator sections in conventional PDMs deform proportionally to the applied torque from the bit, thus its mechanical efficiency gets lower, while HP-PDM can deliver consistent rotary speeds under high torque conditions4.
The mechanical nature of PDM bent housing gives it an advantage over RSS in terms of maximum possible dog leg severity (DLS). So, for the same material and dimensions and thus the same bending strength, maximum DLS achieved by PDM is usually higher. Although newer versions of RSS can achieve DLS comparable to that of PDM but at higher prices even when compared to relatively older RSS. RSS costs more than PDM whoever the service provider is. This can be attributed to better operational performance, relative recency, more expensive electronic components – compared to pure mechanical components of PDM – and naturally the know-how price.
While there are well-defined scenarios where the high cost of RSS makes it impractical5,6, and others where RSS is clearly the superior choice7—such as in complex trajectories with high rig operating costs—the majority of cases fall into a gray area where determining the optimal tool is less straightforward. Rather than viewing directional drilling tools in isolation, they must be considered as integral components of the entire drilling system, influencing overall performance8. In the following sections, we delve into the key qualitative criteria that guide the selection between these two technologies.
PDM add more burden to the mud pumps as it has its intrinsic differential pressure, this would be a factor while choosing liner size for mud pumps and if not optimized may jeopardize hole cleaning efforts9.
RSS (push or point) and PDM require different bit specifications10,11. PDM has a wide range of variation in terms of torque/speed combination, this makes the process of bit selection more flexible than that of RSS. Some formations need a specific type of bit to be drilled, which may not be compatible with RSS. These facts may necessitate the use of PDM in some situations3.
During the drilling, weight on bit (WOB) has different effects on the building capability of PDM and RSS. While PDM is deployed, higher WOB assists in increasing build-up rate (BUR). This is not the case while using push-the-bit RSS. Applying high WOB while using RSS will not allow enough time for side cutting of the bit and thus reduce achievable DLS. The mechanical setting of PDM, although it gives an advantage in terms of achievable DLS while slide drilling, is a major limiter for rotary speed3. The higher the bent housing angle, the lower is the permissible surface rotary speed.
Using RSS, the effect of rotary speed is not the same as the current knowledge of conventional rotary assemblies, increasing rotary speed while using RSS increases the achievable BUR as it improves the side cutting in a similar manner to time-drilling applications.
Regarding drilling performance, bit ROP (instantaneous ROP) achieved using RSS is usually higher than that of PDM while performing directional work (build, drop, or turn). This is mainly due to eliminating sliding intervals, in which PDM usually suffer from hanging problem, although it may be solved to a certain extent using down hole agitators9,12.
However, cases were reported where ROP of PDM was 69% higher than that of RSS on average. This may be attributed to increased bit rotary speed in case of PDM3.
While being used throughout this paper, average ROP is defined as the total footage drilled per operation hour regardless of if the bit was actually drilling or other operations (e.g., connection, tool face adjustment, downlinking…etc.) interfered. Average ROP by definition is always less than on-bottom ROP. Although RSS is in most cases faster than PDM within intervals of directional work, this may not be the case for hold intervals. PDM has the advantage of extra downhole rotary speed pursued by its design. Rotary speed is one factor that affects ROP positively. So, the contrast of average ROP depends on the ratio of directional work intervals to the total interval length. Besides on-bottom time drilling, there are other operations that consume time:
Orienting the tool face (PDM)13.
Survey measurements, connections, wash, and ream (both).
It should be noted that while sending downlinks to RSS, the directional driller may continue drilling with reduced ROP or stop drilling entirely depending on how soon he needs the tool to respond to the sent orders14.
In some times, the overall improvement in ROP as a result of RSS deployment does not match the increased cost15 and motorized RSS is considered16. Motorized RSS are used to improve drilling performance17 and in some reported cased the improvement reached 39%18.
Hole quality is one of the most important end results required for a successful drilling operation. Hole quality affects how deep a well can be drilled, at what rate and the usable lifetime of drill string components. It also affects casing running, logging and completion operations19,20.
Tortuosity which simply describes unwanted deviation from a planned curve or hold Sect21 can be measured on macroscopic (between standard survey stations) or microscopic scale (using continues survey measurements or high frequency borehole imaging tools). A uniform curve has a zero tortuosity, where changing build or turn rates will increase it. So, drilling with PDM gives much more tortuosity than drilling with RSS5,20,22. Also, macroscopic tortuosity measured for PDM run will be less than actual one, as the survey station could have a slide/rotary transition which induce more hidden tortuosity23.
While RSS is capable of achieving uniform DLS without the stair-step profile of PDM, it must be stressed that directional driller – if not competent – can cause the same stair-step phenomenon. Assume that inclination is required to change from 63° to 65° along the next 50 ft from 5000 ft to 5050’, the RSS side-force has to be adjusted for a gradual inclination increase of 4° per 100 ft to reach 65° at 5050 ft. Directional driller must not for example use a higher force, reach 65° at 5035 ft and then hold the angle till 5050 ft, if so, there is no gain from using RSS in terms of hole quality.
During slide drilling with PDM, the hole should be –theoretically – in gauge, but during rotary intervals, the motor drills a slightly over-gauge hole. Although RSS is generally better than PDM in terms of local DLS, operational mode24 and human factor still affect the hole quality and resultant tortuosity23.
Less tortuosity of hole drilled with RSS compared to that of PDM results in less drag and thus extend the drilling envelope to farther lateral displacements3,5. In some cases25, friction factor (FF) using RSS was 28.6% lower compared with PDM, indicating better hole condition and less tortuosity. In addition, continues rotation – which is one factor with great effect on hole cleaning – gives an advantage to RSS over PDM as the drill string is continuously rotated while drilling with RSS. As a result of continuous rotation and better hole cleaning, pipe sticking is way less potential while using RSS. Nevertheless, while the probability is less, the consequences are more severe. If RSS is stuck and cannot be fished, the Lost-In-Hole (LIH) cost is much greater due to the price difference between RSS and PDM.
Vibration is a major concern while drilling directional wells. Types of vibrations can be summarized as:
Axial vibration is more prominent at shallow depth and can be damaging to surface equipment, this type of vibration can be overcome by using shock absorber above the bit. Lateral vibration is the most damaging type and can easily cause a twisted-off drill string component.
While torsional vibration can damage downhole sensors and drill string components. Torsional vibration can arise as a result of drill string/hole friction or from drill bit/formation interaction. If PDM is used, it acts as a dampener or decoupling device that prevents vibration transition from drill bit up the drill string while rotary drilling4,26,27, and no torsional vibration occurs in the first place during slide drilling phases. In area where stick/slip is known to be risky, high sampling rate recording may be essential to avoid missing severe micro stick/slip events28,29.
Vibration in general has a mechanical effect on PDM, where in push-the-bit RSS -which is more prone to stick/slip11 - it can damage RSS either mechanically (wash-out, fatigue or twist-off) or hydraulically by damaging hydraulic chamber behind the pads.
In some cases, decreasing surface RPM, will aid in alleviating downhole lateral vibration, that why PDM will be better than RSS in these cases30.
In addition to recommendations from service companies, some researchers found aligned results30. In their work, they suggested not exceeding 5 g root mean square (RMS) lateral vibration level to keep chances of twist-off low. Mitigating downhole vibration allows optimal application of drilling parameters and faster drilling operations31.
When discussing stresses on drill string, elaboration is required. One factor that affects stresses is rotary speed. From one perspective, using PDM, reduces the rotary speed due to two factors:
The bent housing angle dictates a certain rotary speed limit to avoid excessive stress on this weak point.
Due to downhole rotary speed inherent in PDM design, less surface rotary speed is required to achieve the same drilling performance.
In this regard, PDM causes less stresses than RSS4. But RSS gives a smoother hole and thus less friction forces (mainly T&D), in addition, sliding is the main initiator of helical buckling32, so RSS reduces the incidence of buckling through elimination of sliding3. BHA design also greatly affects the vibration tendency of a certain BHA, whatever directional tool is used33.
Generally, RSS is more preferred during drilling of long laterals due to its better weight transfer qualities34. PDM is not preferred as well, if critical geosteering is planned due to inherent micro-tortuosity and longer time required to correct well path35. Here is a merit of RSS over PDM due to close and continuous control of near bit inclination (NBI) as a result of its proximity to the bit and thus a more reliable well placement36,37. On the other hand, RSS (push the bit type) is also not the best option in cases where formation washouts are expected18, where formation is soft38,39 or has too high or too low UCS40,41, although some operators have applied customized solution for this problem9,42. Point-the-bit RSS system is better than push-the-bit one at these cases as a result of less sensitivity to hole wash-out and its ability to achieve required DLS39.
It must be noted that, although a certain DD tool (either PDM or RSS) would have a certain theoretical DLS capability43, practically, the maximum attainable DLS is usually less than the published values9,44,45.
In special applications like directional drilling with casing, RSS may be better in small holes due to less directional capabilities of small size PDM BHA46. Also for drilling near the lease boundary, a vertical-seeking RSS would be more preferrable due to its instantaneous correction of the inclination47. PDM on the other hand, operates more efficiently in case of open hole side-tracking, although RSS proved to be successful under certain conditions39,48.
RSS vs. PDM flow-chart guide for best-fit steering selection in GoM19,52.
There is a critical gap in the literature when it comes to the application of quantitative methods, particularly data-driven and machine learning techniques, to objectively evaluate and predict the optimal tool for specific well conditions.
Despite the growing integration of machine learning in various domains of the E&P53,54 and drilling industry55,56,57, no existing research, to the best of the authors’ knowledge, has applied these techniques to systematically predict and compare directional drilling tool performance in terms of cost and time. This research, therefore, represents a pioneering effort in the field, offering a novel, data-driven framework that bridges the gap between qualitative insight and quantitative prediction. By leveraging historical well data and advanced machine learning algorithms, this study transforms the traditionally subjective decision-making process into a scientifically grounded, automated tool-selection model. By addressing these gaps, the study aims to advance the existing body of knowledge and provide a more robust framework for directional tool selection. This groundbreaking approach sets a new precedent in the industry, positioning the research as a first-of-its-kind in the quest for optimized directional drilling operations.
Drilling operations produce vast amounts of data. Processing these data takes many forms and gives different outcomes. Old techniques like manual offset analysis were very useful in the last decades, but as drilling became more complex, more advanced techniques are required. Here comes the role of machine learning (ML) models. These models can handle a huge amount of data, maintaining computational resources and providing the best results58. Two main categories of ML models exist namely supervised models and unsupervised ones. In the first type, data are already labeled, and the model is trained on the training dataset to expect the outcome of validation and test data using several algorithms of regression or classification59. Whereas in unsupervised models, datasets are unstructured, and the model objective is to define a certain pattern in the dataset. Applications of ML in drilling are accelerating and most major oil and gas operators are becoming more inclined towards “Digital Drilling” as the drilling performance is correlated to the level of data implementation60.
Digitalization is invading oilfield. As an example, fully digital directional drilling with no physical presence of directional drillers on rig site has been adopted by many service providers61. Even more, morning reports are on way to full automation62, where human input would be minimum.
Many equations were empirically produced due to lack of satisfactory computational power in the past, but with advancements in ML applications, better data processing ways are attainable63.
Artificial Intelligence and ML were used in many drilling applications, including real-time well placement for better productivity64, pore pressure prediction65,66, prediction and classification of lost circulation events67,68,69, drilling dynamics and vibrations70, stuck pipe events avoidance and classifications71, drilling fluids specifications72,73, bit selection and optimization74,75,76, ROP prediction and optimization63,77,78,79,80, well cost estimation81, monitoring of lost time82, BHA design83 and walk tendency84, and well integrity85.
The industry is seeking better performance by making best use of digitalization to save rig time and well cost86,87, in one case, the efforts caused about 20% reduction in well time86. The current paper is aligned with the ultimate industry goal of optimizing drilling time and cost.
With the qualitative factors influencing tool selection thoroughly examined and the implementation of machine learning in the drilling industry well-documented, the foundation is set for developing a quantitative, reliable, and generalizable system. The following section outlines the materials and methods used to construct a data-driven, unbiased framework capable of determining the optimal directional drilling tool for any given subsurface and operational conditions.
For a successful directional drilling operation, any operator would decide to run PDM or RSS based on:
Better drilling performance (less total section time including next casing running), or.
More economic operations (less section cost), or.
The time of a certain section is the sum of drilling time (involving other operations like surveys, etc.), tripping time, time to make-up or lay-down BHA. To assess the effect of directional technology on total section time, the time taken to run casing after finishing the section is taken into account as it is - besides tripping time - representative of hole quality. To estimate the section cost, the timings stated above are transferred into cost. And any flat cost of directional tools should be added besides expected value of lost in hole cost (LIH cost multiplied by probability).
To calculate the total section cost and time associated with either RSS or PDM, the following equations are proposed by the authors (Eq. 1 through 3), note that in Eq. 3, the subscripts RO and RC refer to running in open hole and cased hole, where the subscripts PO, PC refer to pulling out of open hole and cased hole, respectively:
where C = Total cost ($); T = Total time (hr); Tt = Tripping time (hr); Cf = Flat charge of either PDM or RSS; TMU = Make Up time of Tool (hr); F = Footage drilled (ft); ROPAVG = Average run ROP including tool face adjusting, downlinks, surveys, connections, wash, and ream time (ft/hr); SRC = Tripping speed while running within cased hole (ft/hr); SRO = Tripping speed while running within open hole (ft/hr); SPC = Tripping speed while pulling within cased hole (ft/hr); SPO = Tripping speed while running within open hole (ft/hr); N = number of trips (with the same assigned subscript); TLD = Time of lay down (hr); Cr = Rig operating rate including service companies ($/hr); Ct = Tool operating rate (PDM or RSS), including all related BHA components ($/hr); TD = Section total depth (ft); Sc = Casing or liner running speed (ft/hr).
To accurately calculate time and cost, certain variables, such as BHA assembly and lay-down time, rig and tool rental rates, tool flat costs, tripping speed within cased hole, section footage, and total depth (TD), can be directly assigned using predetermined values. However, to mitigate uncertainty and avoid introducing bias into the decision-making process, three critical variables—average run ROP, tripping speed in open hole, and casing/liner running speed in open hole—cannot be assumed. To address these uncertainties, we developed a series of three advanced machine learning (ML) models. These models predict bit ROP, section average ROP, and tripping/casing running speeds, effectively integrating a broad range of drilling and operational parameters. This approach provides a rigorous, data-driven foundation for accurately estimating section time and cost for both PDM and RSS, thereby enabling informed, quantitative tool selection.
In the next three subsections, an explanation of the aspects for each model is given coupled with criteria of selecting features (feature engineering) for the specific model. Figure 2 represents a comprehensive flow chart for the workflow.
The first model is designed to predict bit ROP, a key determinant of drilling efficiency. It depends on many variables like: directional technology used (either PDM or RSS), formations to be penetrated, which in turn can vary in drillability between fields88 and even at different depths within the same field, hole size, mud weight and type, inclination, azimuth89 and DLS, drilling parameters (WOB, flow rate and rotary speed), and bit type (tri-cone or PDC) and condition (new or re-run)90. A regression-based ML approach was used to map the complex relationships between these input variables and the bit ROP, enabling accurate predictions across different formations and well designs.
Building on the predicted bit ROP, the second model calculates the section average ROP, a vital input for determining both section cost and time. Average ROP takes into consideration time for tool face adjusting, downlinks, surveys, connections, wash, and ream. Intuitively, average ROP is a fraction of bit ROP, how much is this fraction depends on few variables like:
Directional technology: PDM and RSS have offline time which does not contribute directly to drilling like survey, wash and ream time, time for adjusting tool face in case of PDM and sending downlink in case of RSS. This offline time may differ between PDM and RSS.
Rig: a rig equipped with TDS will have greater average ROP than one equipped with Kelly if both drill at the same instantaneous ROP. Also rig Horsepower (HP) and crew performance affect how efficient are the operations.
Hole size also may affect how fast wash and ream are done.
Bit type has measurable effect on wash and ream time due to some bit intrinsic features (e.g., back-reaming cutters within PDC bit).
Inclination and DLS also affect wash and ream time and frequency. Driller needs to be more careful at zones of higher inclination and DLS to avoid hole pack-off during wash and ream.
Mud type also has a potential effect of wash and ream time. Oil-based muds are known to improve friction profile within the well and thus better wash and ream. Various additives can be used with water-base muds as well – to reduce cost and environmental problems of OBM - which can improve borehole stability and thus help in obtaining high quality wellbore91,92.
The third model estimates tripping and casing running speeds, key components of overall section time. A high-quality hole will experience a smoother friction profile and thus better tripping and casing running speeds.
Directional technology is the main factor to be addressed in terms of hole condition as RSS is known for its superiority over PDM in terms of hole quality.
Formation also determines how good a trip or casing running would be. Troublesome and unstable formations will suffer longer tripping times. Formations differ among different fields, and a troublesome formation at certain field may be a no-risk formation at another.
The speed with which the hole was drilled is known between drilling professionals to have a major effect on how good a trip will be. A trade-off between ROP and tripping speed is sometimes necessary.
Bit type also affects tripping speed in different aspects. Tri-cone bits usually are easier to run compared to Polycrystalline diamond compact (PDC) ones, where a PDC bit with back-reaming cutters may be easier to pull.
Inclination and DLS also have a dominant effect on tripping speed. High inclination and DLS are associated with higher T&D and thus slower tripping and casing running on average.
The type and direction of the trip affect how fast it is. Liner running is usually faster than casing running, also running speeds are usually higher than pulling speeds.
In addition, the second trip-out of hole is usually better than the fast one if time-dependent wellbore instability is absent.
LIH probability is usually higher in case of PDM than RSS due to obvious reasons. The expected cost value of such an event can be estimated as PLIH multiplied by CLIH (where; PLIH is the probability of losing tool in hole and CLIH is the tool cost of lost in hole ($)) and added to the total cost if required. The authors recommend using a suitable value based on area experience.
Table 2 shows a summary for feature engineering of the three models grouped by data category.
While most data in Table 1 are straightforward, more elaboration may be required for the last one, i.e., tripping type and direction. A hole in a good shape would allow better tripping profile, either while tripping in or out and either while tripping drill string or running casing/liners. The tripping type (drill string, casing, or liner) and direction (in or out) are used amongst other variables to predict the tripping speed which is used as the hole quality quantifier. Tripping speed is never constant and for the sake of simplicity an averaged value for different intervals is used in this model.
In this section, we outline the systematic approach undertaken to develop our machine learning models for predicting drilling performance metrics. The workflow begins with data acquisition, where historical field data is collected to ensure a robust foundation for analysis. Following this, we detail the data pre-processing steps, which involve refining the dataset by addressing missing values and normalizing features to enhance model performance. Finally, we discuss the optimization of model hyperparameters, a crucial phase that fine-tunes the algorithm to accurately capture underlying patterns while minimizing overfitting. This structured methodology underpins the effectiveness of our predictive models and their applicability in real-world drilling operations.
A comprehensive dataset was collected from wells drilled in three distinct fields between 2018 and 2020, employing both PDM and RSS technologies. The data encompassed a wide range of input features essential for training the machine learning models, including drilling parameters, lithology descriptions, mud and bit records, directional surveys, and drilling reports.
The initial dataset comprised 17 distinct feature categories, with a total of 10,691 data points per category, amounting to 181,747 data entries overall. Figures 3 and 4 illustrate the distribution histograms of the numeric input features utilized for each model.
Histogram of numerical data distribution used in model-1.
Histogram of numerical data distribution used in model-2 (to the left) and model-3 (to the right).
Before delving into the specifics of data processing and manipulation, it is important to note that the subsequent computational tasks were carried out using Pandas93 (a Python94 library) for data handling, the Python programming language for implementation, and the Scikit-learn library95 for machine learning functionalities. Additionally, eXtreme Gradient Boosting (XGBoost)96 was employed as the machine learning algorithm for this study. These tools provide a robust framework for efficiently managing data and applying various preprocessing techniques, with Scikit-learn facilitating feature scaling and model evaluation, while XGBoost serves as a powerful algorithm for model training, ensuring a solid foundation for the development of the machine learning models.
The data preprocessing phase involved carefully excluding intervals where drilling was conducted without the use of either RSS or PDM. Additionally, we eliminated data points exhibiting illogical values, such as negative rates of penetration (ROP < 0) or weight on bit (WOB < 0). As a result, the final dataset consisted of 8,366 rows and 17 columns of validated data, ensuring a high-quality foundation for our analysis.
Data was split between training (80%) and testing (20%) sets using (Stratified Shuffle Split) to maintain the distribution of features between both sets. Missing values were handled using median imputation for numerical features and most-frequent-value imputation for categorical features. Numerical features were then standardized using Standard Scaler technique to ensure uniform contribution across all features. For categorical variables, (One-Hot Encoding) was applied to convert categorical values into binary vectors, preventing any ordinal assumptions. No manual feature selection was performed, as all variables were deemed relevant based on domain knowledge. A custom-built transformation pipeline was employed to automate the entire process, using Scikit-learn ColumnTransformer function to handle both numerical and categorical data simultaneously. This allowed preserving the integrity of the data while making it suitable for machine learning tasks. The final output was an array of fully prepared data, ready for model training and evaluation.
For hyperparameter optimization, we employed GridSearchCV with 5-fold cross-validation to fine-tune the XGBoost model. In 5-fold cross-validation, the dataset is split into five equal parts, or “folds.” The model is trained on four of these folds while the remaining fold is used for testing. This process is repeated five times, with each fold serving as the test set once. The final performance is averaged across all five runs, which ensures that the model is not biased toward any particular subset and provides a more reliable estimate of its performance on unseen data. This method helps assess the model’s generalization ability and reduces the risk of overfitting.
GridSearchCV allows us to identify the best combination of hyperparameters that optimize the model’s accuracy. Specifically, we fine-tuned parameters such as the learning rate (which determines how fast the model learns), the maximum depth of trees (which controls model complexity), the number of estimators (how many trees are built), and the subsample and colsample_bytree (which determine the proportion of data and features used in each iteration). Through careful tuning, we ensured that the model captures essential patterns while minimizing overfitting, ultimately enhancing its accuracy and reliability.
Various machine learning models perform differently depending on the problem at hand, and the most reliable way to assess a model’s effectiveness is through rigorous training and testing on respective datasets. In this study, six models were evaluated: linear regression, decision tree, random forest, support vector machine (SVM), stochastic gradient descent (SGD), and XGBoost. Among these, XGBoost demonstrated the highest accuracy and the least susceptibility to overfitting, indicating its superior ability to generalize to new datasets without compromising performance.
In addition, XGBoost was chosen due to its ability to handle complex, high-dimensional data from diverse sources such as drilling parameters, lithology, and operational records. Its built-in regularization prevents any single feature from dominating the model, ensuring unbiased and reliable predictions. XGBoost natively handles missing data, minimizing the need for extensive preprocessing. Additionally, its support for hyperparameter tuning and gradient boosting optimizes model performance. Given the potential to integrate more fields and operational settings, XGBoost’s support for parallel and distributed computing will allow the model to scale efficiently as more data becomes available, making it an ideal choice for diverse drilling environments.
Model-1 demonstrates strong predictive capability for Bit ROP, achieving R-squared values of 0.97 and 0.63 for training and testing datasets, respectively, with Mean Absolute Errors (MAE) of 16.06 and 20.76 ft/hr, indicating minimal overfitting. Model-2, which utilizes predictions from Model-1 along with required features to estimate Average ROP, shows generalizability with MAEs of 18.24 and 20.12 ft/hr on training and testing datasets, respectively. Model-3 effectively predicts tripping speeds with an average MAE of 83.70 ft/hr (less than one stand per hr error) and a standard deviation of 22.96 ft/hr, though it has not been validated with independent testing data due to dataset limitations. The performance metrics for ML-1 and ML-2 are illustrated in Fig. 5, and the parity charts for both models are illustrated in Figs. 6 and 7.
Performance Metrics and R2 comparison (Training against Testing) for ML-1 and ML-2.
Performance of Model-2 (prediction of average ROP).
After validating the model, it was applied to three independent planned sections with diverse geological setting and well configurations to check model response to different inputs. Output results obtained from the model were plugged in Eq. 1 through 3 to estimate total section time and cost using both PDM and RSS.
Figures 8 and 9 show the result obtained using the model for three different cases with entirely different input features, which clearly assures the sensitivity of model output depending on circumstances of a certain drilling operation. For example, in case-2, the operator must drill the section using RSS, as it is almost the same price (2,000 $ less as per case cost scheme) and would finish the Sect. 33 h earlier than PDM. Where in cases 1 & 3, RSS section cost is projected to be 12,000$ and 15,000$ more expensive and expected to finish the Sect. 7 h and 9 h earlier than PDM respectively. And here the operator would decide the tool based on his custom requirements.
Results of section time for three different cases interpreted using the model.
Results of section cost for the same cases interpreted using the model.
Effect of rig equipment: One important result was that the model is sensitive to the rig used. Some rigs made astonishing results by converting to RSS both theoretical and actual. The big performance leap with about 100% increase in Bit ROP (in case of Rig_2, see Fig. 10), can be attributed to low HP mud pumps, which limits WOB that can be used PDM (higher WOB should be accompanied with higher flow rate to avoid stalling). After using RSS, the performance showed a great improvement.
The effect of transforming rigs with small hydraulic HP to RSS.
Criticality of domain knowledge for data analysis: Domain knowledge is of utmost importance when dealing with data-driven models. One observation was that a positive correlation exists between hole size and Bit ROP (Fig. 11), the noticed correlation is not a result of cause/effect relationship, but an intuitive result of the fact that bigger diameter sections are shallower, hence the actual correlation is a negative correlation with Measured Depth (MD), not a positive one with hole size.
Pearson correlation heat map between various features of Model-2.
On-bottom/average ROP relationship dependency on inclination: Although the relationship between Bit ROP and Average ROP is a direct linear relationship, the linearity coefficient changes with inclination. At low inclinations, Average ROP is very sensitive to Bit ROP, where it is less sensitive at higher inclination, which can be attributed to higher T&D at high inclination which require more carefulness during wash and ream and hence less average ROP (Fig. 12).
Relationship between Bit ROP and average ROP.
ROP/tripping speed relationship: Model-3 provides two interesting results, that there is a negative correlation between average ROP and tripping speed, which supports the current knowledge of adverse effect of fast drilling on tripping quality (Fig. 13).
One other interesting result is the positive correlation between hole inclination and tripping speed, which may seem counterintuitive, but this is mostly attributed due to this is linked with the fact that highest angles occur in the hold section within the study wells, which is usually easier for tripping, as more wells with different directional design are added to the model, it will correct itself.
Inclination and Bit ROP Pearson correlation with tripping speed.
The main take-away of the paper is that neither PDM, nor RSS is the optimum steering tool in every situation. The optimum technology changes on a well-by-well basis, and the aim of this paper is to present a data-driven model that uses past performance figures to automate the selection steps and makes it a timely and cost-effective process.
This study developed and implemented a series of machine learning models to optimize the selection of directional drilling technologies—specifically, Positive Displacement Motors (PDM) and Rotary Steerable Systems (RSS)—under varying operational conditions. Based on the results, the following conclusions can be drawn:
By leveraging historical data, the models accurately estimate total section time (including drilling and tripping phases) and associated costs based on specific inputs.
The developed models offer a rapid, reliable, and robust decision-making tool for operators, facilitating informed choices about the most efficient directional drilling technology in terms of cost and time.
The developed model proved to be responsive to change of input data and give clear and sharp recommendations.
Recommendations derived from model applications indicate that RSS technology demonstrates a favorable economic application on smaller rigs equipped with compact mud pumps.
The study emphasizes that well-specific conditions are critical determinants in selecting the optimal directional tool; no single tool universally excels across all scenarios.
The analysis identified that the relationship between on-bottom ROP and average ROP is inclination-dependent.
Domain knowledge and understanding were crucial in selecting relevant input features and interpreting model outputs, ensuring the machine learning model provided actionable and accurate recommendations for field operations.
The findings of this study indicate that incorporating an additional step in the planning phase for each directional well is essential for selecting the appropriate drilling tool, particularly in situations where no single technology emerges as the definitive choice. Operators often favor Positive Displacement Motors (PDM) for economically straightforward wells, while Rotary Steerable Systems (RSS) are preferred for more challenging conditions. Our model effectively identifies the optimal tool for a significant majority of contemporary wells, enhancing decision-making related to drilling time and cost efficiency.
However, it is important to acknowledge certain limitations that we consider as Prospects for future research directions.
Enhancing Data Quality for Future Insights: The dataset utilized for machine learning model-3 has a relatively small number of data points, particularly for tripping and casing running speeds. Instead of having data recorded every foot, we have average speeds over certain intervals, which may limit the model’s precision in predicting tripping speed. As more wells are drilled and data accumulates, the accuracy of this model is expected to improve.
Expanding Tool Availability for Improved Applications: The current models are based solely on data for PDM and RSS. The inclusion of data from motorized rotary steerable systems (MRSS) could enhance the model’s comprehensiveness. However, due to the limited use of MRSS in the field, we currently lack the necessary data to include this tool.
Looking ahead, the work presented here advocates for a paradigm shift in how directional drilling tools are selected, emphasizing the importance of data-driven methodologies to optimize performance across diverse drilling environments. As the industry continues to evolve, the integration of advanced machine learning techniques will undoubtedly play a pivotal role in shaping the future of drilling operations. Additionally, exploring the incorporation of other diagnostic tools could provide a more holistic approach to optimizing drilling operations. By addressing these avenues, the potential for significant advancements in drilling efficiency and cost-effectiveness can be realized, ultimately benefiting oil and gas operators.
The raw data supporting the conclusions of this article will be made available by the authors on request (contact: Muhammad Nour muhammadnour@icloud.com).
Flat charge of either PDM or RSS
Rig operating rate including service companies ($/hr)
Tool operating rate (PDM or RSS), including all related BHA components ($/hr)
Average run ROP including tool face adjusting, downlinks, surveys, connections, wash, and ream time (ft/hr)
Casing or liner running speed (ft/hr)
Tripping speed while pulling within cased hole (ft/hr)
Tripping speed while running within open hole (ft/hr)
Tripping speed while running within cased hole (ft/hr)
Tripping speed while running within open hole (ft/hr)
Make Up time of Tool (hr)
Time of lay down (hr)
Macpherson, J. D. et al. Drilling-systems automation: Current state, initiatives, and potential impact. SPE Drill. Complet. 28, 296–308. https://doi.org/10.2118/166263-PA (2013).
Voronin, A., Gilmanov, Y., Eremeev, D., Dubrovin, A., Abaltusov, N. & Perunov, A. An analysis of rotary steerable systems for sidetracking in open hole fishbone multilateral wells in Vostochno-Messoyakhskoye field. In Society of Petroleum Engineers - SPE Russian Petroleum Technology Conference 2017 (2017).
Malcore, E. & Murray, B. Do you really need that rotary steerable? An evaluation of applications and comparisons with high performance mud motors. In Presented at the Abu Dhabi International Petroleum Exhibition and Conference, SPE (2010).
Rønnau, H.-H. et al. Integration of a performance drilling motor and a rotary steerable system combines benefits of both drilling methods and extends drilling envelopes. In Presented at the SPE/IADC Drilling Conference, SPE (2005).
Nour, M., Elgibaly, A. A., Farahat, M. S. & Mahmoud, O. Novel torque and drag model for drilling two-dimensional high-angle wells. ACS Omega 7, 12374–12389. https://doi.org/10.1021/acsomega.2c00924 (2022).
Article CAS PubMed PubMed Central Google Scholar
Janwadkar, S. et al. Pioneering advanced directional drilling technologies overcome challenges of S and J type wells in North America.
Wiktorski, E., Kuznetcov, A. & Sui, D. ROP optimization and modeling in directional drilling process. In Day 1 Wed, April 05, 2017 (SPE, 2017).
Mensa-Wilmot, G., Gagneaux, J., Langdon, S. & Benet, P. Steering efficiency and directional responsiveness - definitions, considerations, drilling systems effects and contributions to performance drilling improvements. In All Days, Fort Worth (SPE, 2014).
Introduction of a Modified Rotary Steerable System in Papua New Guinea Improves Drilling Performance in Poor Quality Wellbores with Tectonic Breakout | SPE Asia Pacific Oil and Gas Conference and Exhibition | OnePetro. Accessed 21 Sept 2024. Available: https://onepetro.org/SPEAPOG/proceedings-abstract/11APOG/11APOG/SPE-147809-MS/148737.
Bailey, M. J., Thomas, A. X., Doudou, S., El Kotob, M., Al-Shukaili, A. H. & Al-Rawahi, I. S. Enhancing directional performance of pdc drill bits by removing extraneous geometry. In Day 1 Mon, November 09, 2020 D012S116R196 (SPE, 2020).
Reboul, S. et al. Improving the directional behavior of PDC bits affected by stick-slip: A statistical approach. In Day 7 Mon, March 29, 2021, Virtual: IPTC p. D071S027R001 (2021).
Nour, M. Torque and drag control during drilling of horizontal and extended reach wells. Master thesis, Suez University, Suez (2015).
Ledgerwood, L. W. et al. The effect of bit type on reactive torque and consequent tool-face-control anomalies. SPE Drill. Complet. 31, 95–105. https://doi.org/10.2118/174949-PA (2016).
Cao, D., Hender, D., Ariabod, S., James, C., Ben, Y. & Lee, M. The development and application of real-time deep learning models to drive directional drilling efficiency. In Day 1 Tue, March 03, 2020 (SPE, 2020).
Blakney, D. et al. Combining state-of-the-art hybrid bit and positive displacement motors saves 863,670 CAD over 20 wells in Northern Alberta, Canada. Accessed 21 Sept 2024. Available: https://doi.org/10.2118/195237-MS.
Challenging Conventional Best Practices with an Updated Productive Drilling Solution | SPE Kingdom of Saudi Arabia Annual Technical Symposium and Exhibition | OnePetro. Accessed 21 Sept 2024. Available: https://onepetro.org/SPESATS/proceedings-abstract/18SATS/All-18SATS/SPE-192335-MS/215707.
Faster ROP in Hard Chalk: Proving a New Hypothesis for Drilling Dynamics | SPE/IADC Drilling Conference and Exhibition | OnePetro. Accessed 21 Sept 2024. Available: https://onepetro.org/SPEDC/proceedings-abstract/15DC/3-15DC/D031S016R001/183362.
Andoni, B. et al. Directional drilling and logging while drilling operation optimization methods in a brownfield-fast-drilling environment. In Day 2 Wed, February 28, 2024 D021S002R004 (OTC, 2024).
D’Ambrosio, P., Hart, S., Prochaska, E., Bouska, R. & Tinsley, D. Cost-effective ultralarge-diameter polycrystalline-diamond-compact-bit drilling in deepwater Gulf of Mexico. SPE Drill. Complet. 29, 47–63. https://doi.org/10.2118/163448-PA (2014).
Greenwood, J., Marck, J., Nair, V. N. & Munguia, J. Drilling performance evaluation using advanced BHA modeling and field validation. In Presented at the IADC/SPE International Drilling Conference and Exhibition D082S000R003–D082S000R003 (SPE, 2020).
Oliveira, V., Zahrani, B., Alrushud, A. & Mohammad, M. Evaluating wellbore tortuosity in horizontal wells caused by RSS steering mechanism. In Presented at the SPE Kingdom of Saudi Arabia Annual Technical Symposium and Exhibition (SPE, 2018).
Carpenter, C. Faster rate of penetration in hard chalk: Proving a new hypothesis for drilling dynamics. J. Petrol. Technol. 68, 59–60. https://doi.org/10.2118/0216-0059-JPT (2016).
Menand, S., Mills, K. A. & Suarez, R. Micro Dogleg detection with continuous inclination measurements and advanced BHA modeling. Accessed 21 Sept 2024. Available: https://doi.org/10.2118/183299-MS.
Nobbs, B., Menand, S. & Dao, N.-H. RSS and motor directional analysis: Modelling the effect of local Doglegs on RIH of completions and casing. In Day 1 Tue, November 17, 2020, Virtual: SPE D013S105R015 (2020).
Hummes, O. et al. Using advanced drilling technology to enable well factory concept in the Marcellus shale. Accessed 21 Sept 2024. Available: https://doi.org/10.2118/151466-MS.
Carpenter, C. Resolving torsional vibration in limestone reservoirs reduces equipment damage. J. Petrol. Technol. 71, 68–69. https://doi.org/10.2118/1219-0068-JPT (2019).
Lines, L. Technology update: A holistic approach to controlling torsional dynamics in the drillstring. J. Petrol. Technol. 68, 20–23. https://doi.org/10.2118/1016-0020-JPT (2016).
Bowler, A. et al. Continuous high-frequency measurements of the drilling process provide new insights into drilling-system response and transitions between vibration modes. SPE Drill. Complet. 31, 106–118. https://doi.org/10.2118/170713-PA (2016).
Sugiura, J. & Jones, S. A drill bit and a drilling motor with embedded high-frequency (1600 Hz) drilling dynamics sensors provide new insights into challenging downhole drilling conditions. SPE Drill. Complet. 34, 223–247. https://doi.org/10.2118/194138-PA (2019).
Raap, C., Craig, A. & Perez, D. Understanding and eliminating drill string twist-offs by the collection of high frequency dynamics data. In Presented at the SPE/IADC Middle East Drilling Technology Conference and Exhibition (SPE, 2011).
Sugiura, J. & Jones, S. Simulation and measurement of high-frequency torsional oscillation (HFTO)/high-frequency axial oscillation (HFAO) and downhole HFTO mitigation: Knowledge gains continue using embedded high-frequency drilling dynamics sensors. SPE Drill. Complet. 35, 553–575. https://doi.org/10.2118/199658-PA (2020).
Huang, W. & Gao, D. Combined effects of wellbore curvature, connector, and friction force on tubular buckling behaviors. SPE J. 24, 2083–2096. https://doi.org/10.2118/195680-PA (2019).
Bailey, J. R. R. et al. Design evolution of drilling tools to mitigate vibrations. SPE Drill. Complet. 28, 350–369. https://doi.org/10.2118/163503-PA (2013).
Sugiura, J. & Jones, S. Rotary steerable system enhances drilling performance on horizontal shale wells. Accessed 21 Sept 2024. Available: https://doi.org/10.2118/131357-MS.
Adams, A. A. et al. Continuous proportional steering for a re-entry multi-lateral well in cretaceous reservoir in UAE Saves rig days and achieves reservoir contact objectives in a one-bit run. In Day 4 Thu, March 21, 2019 D041S045R001 (SPE, 2019).
Hameed, M., Gouda, M., Abouzaid, A., Selim, M. & Alrushud, A. Reducing the uncertainties associated with unseen borehole inclination changes between directional survey stations on TVD calculations and post-geological interpretation. In Day 2 Tue, February 22, 2022 D021S045R006 (IPTC, 2022). https://doi.org/10.2523/IPTC-22218-EA.
Ruzhnikov, A. et al. ERD wells on remote artificial islands: Successful optimization strategy to boost a performance. Accessed 21 Sept 2024. Available: https://doi.org/10.2118/188627-MS.
Khattab, H. M. et al. High build up rate rotary steerable system leads to revolutionize onshore horizontal drilling in western desert of Egypt. Accessed 21 Sept 2024. Available: https://doi.org/10.2118/178186-MS.
Bryan, S., Cox, J., Blackwell, D., Slayden, F., & Naganathan, S. High Dogleg rotary steerable system: A step change in drilling process. Accessed 21 Sept 2024. Available: https://doi.org/10.2118/124498-MS.
Prediction of directional changes in well drilling based on formation rock strength | U.S. Rock Mechanics/Geomechanics Symposium | OnePetro. Accessed 15 Sept 2024. Available: https://onepetro.org/ARMAUSRMS/proceedings-abstract/ARMA08/All-ARMA08/ARMA-08-230/118981.
Oredolapo, O. M. & Onomo, C. P. A study of directional drilling using rotary steerable system in soft shallow formation Benin Sand – Rio Del Rey Area, Offshore Cameroon. In Day 2 Wed, October 30, 2019 (OTC, 2019). https://doi.org/10.4043/29750-ms.
Osita, H. et al. Rotary steerable directional jetting service successfully delivering step change in directional well design for soft sediment formation in deepwater Nigeria. In Day 3 Wed, August 07, 2019 (SPE, 2019). https://doi.org/10.2118/198754-ms.
Saavedra Jerez, L., Cayeux, E. & Sui, D. Assisting directional drilling by calculating a safe operating envelope. SPE J. 1–17. https://doi.org/10.2118/217707-PA (2024).
Clayton, L. M., Lee, M. H. & Salmachi, A. Alleviating directional well trajectory problems via data analytics. In Day 1 Mon, October 17, 2022 D031S014R005 (SPE, 2022).
Li, J., Tudor, R., Ginzburg, L., Robello, G., Xu, H. & Grigor, C. Evaluation and prediction of the performance of positive displacement motor (PDM). J. Can. Pet. Technol. 40 (2001).
Strickler, R., Mushovic, T., Warren, T. & Lesso, B. Casing directional drilling using a rotary steerable system.
Rotary Steerable System Technology Case Studies in a High-Volume, Low-Cost Environment | SPE Middle East Oil and Gas Show and Conference | OnePetro. Accessed 21 Sept 2024. Available: https://onepetro.org/SPEMEOS/proceedings-abstract/07MEOS/All-07MEOS/SPE-105468-MS/143307.
Al-Majed, A., Soh, A., Al-Mutairi, Y., Abdelgayoum, A. & Al Sahli, B. Data driven optimization of sidetrack operation planning and execution. Accessed 21 Sept 2024. Available: https://doi.org/10.2523/IPTC-20286-ABSTRACT.
Pollock, J., Stoecker-Sylvia, Z., Veedu, V., Panchal, N. & Elshahawi, H. Machine learning for improved directional drilling. In Day 3 Wed, May 02, 2018 (OTC, 2018). https://doi.org/10.4043/28633-ms.
Salamah, F. W., Oliveira, V. C, Martinez, M. A., Zahrani, B. S. & Abouelnaaj, K. K. A comprehensive study of directional drilling systems in the Kingdom of Saudi Arabia. In Day 2 Tue, April 25, 2017 (SPE, 2017). https://doi.org/10.2118/188045-ms.
Zimmer, C., Person, J., Richter, D., Tilley, J. & Bittar, M. Drilling a better pair: New technologies in SAGD directional drilling. J. Can. Pet. Technol. (2012).
Weber, A., Gray, I. & Neuschaefer, R. Rotary steerable systems in the Gulf of Mexico gain acceptance. World Oil 53–62 (2007).
Tewari, S. & Dwivedi, U. D. Ensemble-based big data analytics of lithofacies for automatic development of petroleum reservoirs. Comput. Ind. Eng. 128, 937–947. https://doi.org/10.1016/j.cie.2018.08.018 (2019).
Kaleem, W., Tewari, S., Fogat, M. & Martyushev, D. A. A hybrid machine learning approach based study of production forecasting and factors influencing the multiphase flow through surface chokes. Petroleum 10, 354–371. https://doi.org/10.1016/j.petlm.2023.06.001 (2024).
Tewari, S. & Dwivedi, U. D. A real-world investigation of TwinSVM for the classification of petroleum drilling data.
Artificial intelligence applications promise improved drilling efficiency | Offshore. Accessed 22 Sept 22 2024. Available: https://www.offshore-mag.com/drilling-completion/article/14301447/artificial-intelligence-applications-promise-improved-drilling-efficiency.
Tewari, S., Dwivedi, U. D. & Biswas, S. Intelligent drilling of oil and gas wells using response surface methodology and artificial bee colony. Sustainability 13, 1664. https://doi.org/10.3390/su13041664 (2021).
Temizel, C. et al. A thorough review of machine learning applications in oil and gas industry. In Society of Petroleum Engineers - SPE/IATMI Asia Pacific Oil and Gas Conference and Exhibition 2021, APOG 2021 (2021).
Ekeregbe, M. P., Khalaf, M. S. & Samuel, R. Dull bit grading using video intelligence. In Proceedings - SPE Annual Technical Conference and Exhibition (2021).
Dupriest, F. E., Pastusek, P. E. & Prim, M. T. The critical role of digital data in a drilling performance workflow. In Society of Petroleum Engineers - SPE Intelligent Energy International 2012 Vol. 2 722–729 (2012).
Stepnov, A., Chatar, C., El Hawy, A., Cassel, J., Abouassi, B. & Khader, C. The digital directional drilling evolution. In Society of Petroleum Engineers - Abu Dhabi International Petroleum Exhibition and Conference 2019, ADIP 2019 (2019).
Thonhauser, G. Guest editorial: Digital drilling disruption - understand downhole, gain control. J. Pet. Technol. 70, 14–15. https://doi.org/10.2118/1118-0014-JPT (2018).
Sabah, M. et al. A machine learning approach to predict drilling rate using petrophysical and mud logging data. Earth Sci. Inform. 12, 319–339. https://doi.org/10.1007/S12145-019-00381-4/FIGURES/14 (2019).
Kristoffersen, B. S., Silva, T. L., Bellout, M. C. & Berg, C. F. Efficient well placement optimization under uncertainty using a virtual drilling procedure. Comput. Geosci. 26, 739–756. https://doi.org/10.1007/S10596-021-10097-4/METRICS (2022).
Bui, H., De Nicolias, N., Nye, R. & Estrada, J. Machine learning applications to improve pore pressure prediction in hazardous drilling environments. In Proceedings of the Annual Offshore Technology Conference (2022).
Farsi, M. et al. Predicting formation pore-pressure from well-log data with hybrid machine-learning optimization algorithms. Nat. Resour. Res. 30, 3455–3481. https://doi.org/10.1007/S11053-021-09852-2/FIGURES/19 (2021).
Al-Hameedi, A. T. T. et al. Mud loss estimation using machine learning approach. J. Pet. Explor. Prod. Technol. 9, 1339–1354. https://doi.org/10.1007/S13202-018-0581-X/FIGURES/37 (2019).
Alkinani, H. H., Al-Hameedi, A. T. T. & Dunn-Norman, S. Artificial neural network models to predict lost circulation in natural and induced fractures. SN Appl. Sci. 2, 1–13. https://doi.org/10.1007/S42452-020-03827-3/TABLES/6 (2020).
Mardanirad, S., Wood, D. A. & Zakeri, H. The application of deep learning algorithms to classify subsurface drilling lost circulation severity in large oil field datasets. SN Appl. Sci. 3, 1–22. https://doi.org/10.1007/S42452-021-04769-0/FIGURES/13 (2021).
Caballero, E. F., Lobo, D. M., Di Vaio, M. V., Silva, E. C. C. M. & Ritto, T. G. Support vector machines applied to torsional vibration severity in drill strings. J. Braz. Soc. Mech. Sci. Eng. 43, 1–12. https://doi.org/10.1007/S40430-021-03089-Z/TABLES/6 (2021).
Elmousalami, H. H. & Elaskary, M. Drilling stuck pipe classification and mitigation in the Gulf of Suez oil fields using artificial intelligence. J. Pet. Explor. Prod. Technol. 10, 2055–2068. https://doi.org/10.1007/S13202-020-00857-W/FIGURES/11 (2020).
Gasser, M., Mahmoud, O., Ibrahim, F. & Abadir, M. Using artificial intelligence techniques in modeling and predicting the rheological behavior of nano-based drilling fluids. In Proceedings of the International Conference on Offshore Mechanics and Arctic Engineering – OMAE (2021).
Gasser, M., Naguib, A. & Mahmoud, O. New approach to predict the filtrate invasion of nanoparticle-based drilling mud using artificial neural network. In Proceedings of the International Conference on Offshore Mechanics and Arctic Engineering – OMAE (2022).
Al-Riyami, N., Revheim, O., Robinson, T. S., Batruny, P., Meor Hakeem, M. H. & Tze Ping, G. Drilling in the digital age: Case studies of field testing a real-time ROP optimization system using machine learning. In Proceedings of the SPE/IADC Middle East Drilling Technology Conference and Exhibition (2023).
Batruny, P., Zubir, H., Slagel, P., Yahya, H., Zakaria, Z. & Ahmad, A. Drilling in the digital age: Machine learning assisted bit selection and optimization (2021).
Tewari, S., Dwivedi, U. D. & Biswas, S. A novel application of ensemble methods with data resampling techniques for drill bit selection in the oil and gas industry. Energies 14, 432. https://doi.org/10.3390/en14020432 (2021).
Abdel Azim, R. Application of artificial neural network in optimizing the drilling rate of penetration of western desert Egyptian wells. SN Appl. Sci. 2, 1–13. https://doi.org/10.1007/S42452-020-2993-8/FIGURES/11 (2020).
Al-AbdulJabbar, A., Mahmoud, A. A. & Elkatatny, S. Artificial neural network model for real-time prediction of the rate of penetration while horizontally drilling natural gas-bearing sandstone formations. Arab. J. Geosci. 14, 1–14. https://doi.org/10.1007/S12517-021-06457-0/FIGURES/12 (2021).
Amadi, K., Iyalla, I., Prabhu, R., Alsaba, M. & Waly, M. Development of predictive optimization model for autonomous rotary drilling system using machine learning approach. J. Pet. Explor. Prod. Technol. https://doi.org/10.1007/S13202-023-01656-9/FIGURES/12 (2023).
Noshi, C. I. Application of data science and machine learning algorithms for ROP optimization in west Texas: Turning data into knowledge. Accessed 21 Sept 2024.
Yunus, M. N. M. et al. Machine learning vs. statistical models approach: A case for probabilistic AFE cost in drilling. In Society of Petroleum Engineers - ADIPEC 2022 (2022).
Mora, S. & Martinez, D. Drilling in slips: Strategies to measure the invisible lost time, technical limits definition, using standard analytics and machine learning algorithms. In Proceedings of the SPE/IADC Middle East Drilling Technology Conference and Exhibition (2021).
Skoff, G. et al. Machine learning-based drilling system recommender: Towards optimal BHA and fluid technology selection. In Day 3 Thu, March 09, 2023 D031S021R004 (SPE, 2023).
Noshi, C. I. & Schubert, J. J. Using supervised machine learning algorithms to predict BHA walk tendencies. Accessed 21 Sept 2024.
Salem, A. M., Yakoot, M. S. & Mahmoud, O. A novel machine learning model for autonomous analysis and diagnosis of well integrity failures in artificial-lift production systems. Adv. Geo Energy Res. 6, 123–142. https://doi.org/10.46690/ager.2022.02.05 (2022).
Caso, C. et al. Toward the redefinition of drilling plan and execution via a digital drilling ecosystem. In SPE/IADC Drilling Conference, Proceedings (2020).
D’Almeida, A. L. et al. Digital transformation: A review on artificial intelligence techniques in drilling and production applications. Int. J. Adv. Manuf. Technol. 119, 5553–5582. https://doi.org/10.1007/S00170-021-08631-W/TABLES/7 (2022).
Article PubMed PubMed Central Google Scholar
Adams, A. J., Grundy, K. C. & Kelly, C. M. Probabilistic well-time estimation revisited: Five years on. SPE Drill. Complet. 31, 200–218. https://doi.org/10.2118/173028-PA (2016).
Nour, M., Farahat, M. S. & Mahmoud, O. The influence of directional well azimuth on bit performance: A statistical approach. In Volume 10: Petroleum Technology V010T11A015 (American Society of Mechanical Engineers, 2022).
Mazen, A. Z., Rahmanian, N., Mujtaba, I. & Hassanpour, A. Prediction of penetration rate for PDC bits using indices of rock drillability, cuttings removal, and bit wear. SPE Drill. Complet. 36, 320–337. https://doi.org/10.2118/204231-PA (2021).
Mahmoud, O., Nasr-El-Din, H. A., Vryzas, Z. & Kelessidis, V. C. Nanoparticle-based drilling fluids for minimizing formation damage in HP/HT applications. In SPE International Formation Damage Control Symposium Proceedings (2016).
Mahmoud, O. & Nasr-El-Din, H. Formation damage assessment and filter cake characterization of NPs/Ca-bentonite fluids for drilling harsh environments using computed-tomography scan. In Society of Petroleum Engineers - SPE Trinidad and Tobago Section Energy Resources Conference 2018 (2018).
McKinney, W. Python for Data Analysis: Data Wrangling with Pandas, NumPy, and IPython 2nd edn, 524 (O’Reilly Media Inc, 2018).
Linge, S. & Langtangen, H. P. Programming for Computations - Python: A Gentle Introduction to Numerical Simulations with Python 3.6. In Texts in Computational Science and Engineering. (Springer International Publishing, 2020).
Géron, A. Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow (2019, O’reilly). In Hands-On Machine Learning with R 510–510 (2017).
XGBoost Documentation—xgboost 1.6.0-dev documentation. Available: https://xgboost.readthedocs.io/en/latest/index.html (2022).
The authors would like to acknowledge the late Prof. M.S. Farhat for his invaluable contributions and guidance throughout this work. His expertise and mentorship have left a lasting impact on this research.
Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).
Department of Petroleum Engineering, Faculty of Petroleum and Mining Engineering, Suez University, Suez, 11252, Egypt
Muhammad Nour & Said K. Elsayed
Department of Petroleum Engineering, Faculty of Engineering and Technology, Future University in Egypt (FUE), Cairo, 11835, Egypt
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
Conceptualization, M.N.; methodology, O.M., M.N. and S.K.E.; software, M.N.; validation, O.M. and S.K.E.; formal analysis, M.N.; investigation, O.M. and S.K.E.; resources, M.N.; data curation, M.N.; writing—original draft preparation, M.N.; writing—review and editing, O.M. and S.K.E.; visualization, M.N.; supervision, O.M. and S.K.E.; project administration, M.N.; funding acquisition, N/A. All authors have read and agreed to the published version of the manuscript.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Nour, M., Elsayed, S.K. & Mahmoud, O. A supervised machine learning model to select a cost-effective directional drilling tool. Sci Rep 14, 26624 (2024). https://doi.org/10.1038/s41598-024-76910-z
DOI: https://doi.org/10.1038/s41598-024-76910-z
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Scientific Reports (Sci Rep) ISSN 2045-2322 (online)
pile driving rig Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.