Sensitivity Analysis and Feature Selection for Drilling-Oriented Models

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper OTC 31680, “Successful Development and Deployment of a Global ROP-Optimization Machine-Learning Model,” by Timothy S. Robinson, SPE, Exebenus, Peter Batruny, SPE, Petronas, and Dalila Gomes, SPE, Exebenus, et al. The paper has not been peer reviewed. Copyright 2022 Offshore Technology Conference. Reproduced by permission. _ Drilling rate of penetration (ROP) is influenced by many factors, both controllable and uncontrollable, difficult to distinguish with the naked eye. Thus, machine-learning (ML) models such as neural networks have gained momentum in the drilling industry. Previous models were field-based or tool-based, which affected accuracy outside of the trained field. The authors of the complete paper aim to develop one generally applicable global ROP model, reducing the effort needed to redevelop models for every application. Introduction The authors have identified a need for an ROP model that can recommend parameters in real time, which ideally requires a general ROP model that can be applied with low prediction errors. Formation properties are available in real time from logging tools; however, incorporating logging data into a global ROP model is challenging. A distance usually exists between the bit and logging tools where the properties of the formation drilled in real time at the bit are not available. While strides have been made in ML interpretation of logs, human interpretation is still needed. Equipment and Processes A deep neural network was used for modeling ROP. This model was queried as a black box within the scope of a constrained optimization algorithm tasked with identifying combinations of controllable drilling parameters expected to increase ROP. An overview of the ROP optimization system powered by the ROP estimation model is shown in Fig. 1. The ROP model was fitted using training and validation data sets composed of data from independent wells. Four additional wells were held in reserve for final testing purposes, most importantly for assessing the model’s ability to generalize to completely unseen data. For optimization, a three-stage conditional parameter sweep was used where the value expected to optimize ROP was found sequentially for each of the controllable parameters while fixing the values of the other variables. The optimizing value for each parameter was used as a fixed value for the subsequent optimization steps. This approach maintains the simplicity of a grid search and converges to a local optimum quickly because of the relatively small number of search iterations required under this formulation. A variety of well types with global geographical diversity was present in the data, using different downhole drives and bottomhole assemblies. Twenty-nine wells were used for training the ROP model, with an additional five wells used for validation and a further four wells reserved for holdout testing. No overlap existed between the wells constituting the training, validation, and holdout data sets. The drilling data sets were preprocessed by methods detailed in the complete paper. To ensure consistency across all operations, an average ROP was calculated independently based on depth measurements and their associated timestamps. This ROP calculation was used as the target variable for the ROP estimation model and for assessing its predictive performance. Training and validating the neural network for ROP estimation using data from a variety of independent wells is very useful for assessing model generalization. This technique is superior to the common approach of randomly splitting data sets because it minimizes information leakage between training and validation data.

Pre-salt layers has long-term exploration and production possibilities, however the properties of such layers are highly challenging for exploration (i.e. these carbonate layers are highly abrasive, located deeper than 5000 m, have generally low permeability) and production (i.e. pre-salt layers are located in harsh oceanic conditions, hundreds of km offshore). These conditions mean increased exploration technological and economic difficulties. One of the partial solutions to decrease costs is to reduce the drilling operations time. Rate of penetration (ROP) has a significant effect on the overall drilling time, driving ROP modeling and optimization to be a viable solution for reducing drilling operations time in such environment. Several mathematical ROP models were developed in the last five decades in the petroleum industry, departing from rather simple but less reliable R-W-N (drilling-rate, weight-on-bit and rotary-speed) equations until the arrival to a comprehensive and complex approach: Bourgoyne and Young ROP Model (BYM) which was first published in 1974. The paper explains the equation, how it is applied in terms of ROP modeling, identifies the main drilling parameters driving each sub-functions, and introduces how they were developed. The paper base itself on the sub-functions of the equation, explains the normalization factors which have significant influence on the model, and also introduces simulations which aim to understand the approach by applying the equation in a pre-salt layer case study. Knowing that, the original publication was introduced in 1974, this paper also aims to identify rooms for improvement and/or alternate the sub-functions to match actual field data better than with the originally given drillability coefficient recommended boundaries stated in the first publication. This is accomplished through a real-world practical application of this ROP model in a pre-salt layer. The paper also assesses the limitation in terms of the applicability of this complex model for ROP analysis and optimization in such carbonate layers.

SummaryReamers are an integral part of deepwater Gulf of Mexico (GOM) drilling and their performance significantly impacts the economics of well construction. This paper presents a novel programmatic approach to model rate of penetration (ROP) for reamers and improve drilling efficiency. Three field implementations demonstrate value added by the reamer drilling optimization (RDO) methodology.Facilitated by user interface panels, the RDO workflow consists of surface and downhole drilling data filtering and visualization, detection of rock formation boundaries, frictional torque (FTRQ) and aggressiveness estimation, ROP modeling with analytical equations and machine learning (ML) algorithms [regression, random forests, support vector machines (SVMs), and neural networks], and optimization of drilling parameters. ROP model coefficients and bit and reamer aggressiveness are dependent on lithology and computed from offset well data. Subsequently, when planning a nearby well, bottomhole assembly (BHA) designs are evaluated on the basis of drilling performance and weight and torque distributions between cutting structures to avoid early reamer wear and dysfunctions. Geometric programming establishes optimal drilling parameter roadmaps according to operational limits, downhole tool ratings, rig equipment power constraints, and adequate hole cleaning.Separate ROP models are trained for reamer-controlled and bit-controlled ROP zones, defined by the proportion of surface weight on bit (WOB) applied at the reamer, in every rock formation. This novel concept enables ROP prediction with the appropriate model for each well segment depending on which cutting structure limits drilling speed. In the first of the three RDO applications with field data from deepwater GOM wells, optimal bit-reamer distances are determined by analyzing reamer weight load in uniform salt sections. Next, ROP modeling for the addition or removal of a reamer from the BHA is used in contrasting well designs to conceivably alleviate a USD 16 million casing inventory surplus. Finally, active optimization constraints are investigated to reveal drilling performance limiters, justifying equipment upgrades for a future deepwater GOM well.The proposed innovative workflow and methodology apply to any drilling optimization scenario. They benefit the practicing engineer interested in drilling performance optimization by providing insights on how different cutting structure sizes affect ROP behavior and ultimately aiding in the selection of appropriate bit and reamer diameters and optimal operational parameters.

Drilling rate of penetration (ROP) is a major contributor to drilling costs. ROP is influenced by many different controllable and uncontrollable factors that are difficult to distinguish with the naked eye. Thus, machine learning (ML) models such as neural networks (NN) have gained momentum in the drilling industry. Existing models were either field-based or tool-based, which impacted the accuracy outside of the trained field. This work aims to develop one generally applicable global ROP model, reducing the effort needed to re-develop models for every application. A drilling dataset was gathered from exploration and development wells in both onshore and offshore operations from a variety of fields and regions. The wells were curated to have different water depths, down hole drive such as Rotary Steerable System (RSS), PDM, Standard Rotary, bit types (Mill Tooth, TCI, PDC) and inclinations (vertical or deviated). A deep neural network was used for modelling the relationship between ROP and inputs taken from real-time surface data, such as Torque, Weight-on-Bit (WOB), rotary speed (RPM), flow and pressure measurements. The performance of the ROP model was analyzed using historical data via summary statistics such as Mean Absolute Percentage Error, as well as graphical results such as residuals distributions, cumulative distribution functions of errors, and plots of ROP vs depth for independent holdout testing wells not included in the model fitting process. Analysis was done both in aggregate, and for each specific well. The ROP model was demonstrated to generalize effectively in all cases, with only minor increases in error metrics for the holdout test wells, where the Mean Absolute Percentage Error averaged across wells was ~20%, compared to 17.5% averaged across training wells. Furthermore, residuals distributions were centered close to zero, indicating low systematic error. This work proves the case for a "global" ROP prediction model applicable "out-of-the-box" to a broad set of drilling operations. A global ROP model has the potential to eliminate learning curves, reducing time and costs associated with having to develop a new model for every field. Furthermore, a model that effectively captures the relationships between parameters controllable by drillers and ROP can be used for automatically identifying drilling parameters that improve ROP. Preliminary field-testing of the ROP optimization system yielded positive results, with many examples of increased ROP realized after following drilling parameter recommendations provided by the software.

Rate of penetration (ROP) has been a key indicator for drilling efficiency. Increasing ROP can lead to significant operational-time and cost savings in drilling. ROP can be estimated using the physics-based and data-driven models. The physics-based model, although it is derived from drilling physical principles, involves empirical coefficients and functional constrains for fitting that often lead to poor results. Recently, data-driven approach that utilize machine learning to analyze and learn patterns purely from the data could be use to overcome the drawback of the physics-based models. In this paper, we combine the insight from two physics-based models to determine the features that are used in the data-driven models. We rank the input features using adaptive boosting algorithm based on the input parameters in the Bingham's and Hareland and Rampersad's models. The results demonstrate that WOB, RPM, and rock strength have more influence on the ROP prediction compared to other drilling parameters in these two models. This outcome sets the base data-driven ROP model used in this study. Furthermore, ranking features using more drilling parameters are investigated, which inferring that flowrate, standpipe pressure, and torque also contribute on the ROP prediction. We include these new parameters, in addition to WOB, RPM, and rock strength, to develop an extended data-driven ROP model from the base model. We examine the base and extended models to estimate ROP using a dataset of a well drilled through a shale formation in North America. Statistical performance evaluation criteria including root-mean-square error and cross-correlation plots between the actual and estimated ROP are employed to evaluate the prediction accuracy and robustness, respectively. The results show good agreement between the actual and predicted ROP for both the base and extended models.

A central element to reduce drilling cost is to improve drilling operation by analyzing real-time data. Developing advanced real-time analysis tools is one way to improve the drilling operation. Two approaches which currently are used for optimizing the actual rotary drilling process are mechanical specific energy and inverted rate of penetration models. The mechanical specific energy method is defined as the work needed to destroy a given volume of the rock. It can act as a tool during the drilling operation to detect changes in drilling efficiency thus providing a method to optimize the drilling parameters to enhance instatanious rate of penetration. Rate of penetration models, on the other hand, can be used to calculate formation drillability considering the effects of drilling parameters, bits design and bit wear. Drilling optimization using rate of penetration models is done by changing the drilling parameters and/or bit design to find the optimum drilling scenario for an entire bit run. The mechanical specific energy log and the drillability ratio differ when mud weight is changed and when bits are worn. These two differences are due the fact that mechanical specific energy does not include bit wear as well as the effect of changing mud weight. By combining these methods and modifying the mechanical specific energy equations to incorporate these effects and the mechanical specific energy can be used as a real-time trending tool for bit wear estimations. In this analysis, wells from offshore Middle East and onshore North America are analyzed. The field results are very encouraging in that the bit wear for both roller cone and PDC bits can be predicted. The field validation of this new approach shows that the supplementary information on the bit wear status can in some cases benefit in the decision of when to pull the bit while it still is in the hole and thereby possibly improve overall economics of the drilling operation.

Most coal seams in China are characterized by high‐gas pressure and high‐gas content, which seriously threaten the safety of coal mines. Gas extraction is an effective measure for preventing gas disasters. However, the drilling process in soft coal seams is inefficient; it is difficult for the boreholes to reach the design length. The rate of penetration (ROP) plays an important role in determining drilling efficiency and borehole length in soft coal seams. In this study, we establish a simplified ROP model for soft coal seams and evaluate the maximum ROP of Hebi No. 6 coal mine. First, the drill cutting transport capacity of the drilling system and drill cutting quantity generated during the drilling process are analyzed. By simplifying drill cutting transport and coal seam properties, the ROP model is established. Second, the Rock Failure Process Analysis software is used to analyze the deformation region of boreholes, which has a great influence on the drill cutting transport space. Finally, a field test is implemented in Hebi No. 6 coal mine in Henan Province, China, to verify the model, and we propose a measure to control the ROP below the maximum value to increase the length of the borehole. The test results indicate that the model has high accuracy, with errors less than 15%, and the lengths of boreholes in air‐return and transportation roadways are increased to 60 and 80 m, respectively, and the increase rate is greater than 50%.

Accurate prediction of Rate of Penetration (ROP) is critical for optimizing drilling efficiency and reducing costs in hydrocarbon exploration. Traditional ROP models often overlook formation temperature, despite its significant influence on rock mechanics and drilling fluid rheology, particularly in high-temperature sandstone reservoirs like those in the Niger Delta. This study employs three machine learning (ML) algorithms—Support Vector Regression (SVR), Random Forest (RF), and Artificial Neural Networks (ANN)—to evaluate the contribution of formation temperature to ROP prediction. A dataset of 1,200 drilling records from Niger Delta wells was used, incorporating parameters such as weight-on-bit (WOB), rotary speed (RPM), pump pressure, and formation temperature. Model performance was assessed using R², Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE), with ablation studies to isolate temperature’s impact. Inclusion of formation temperature improved ROP prediction accuracy across all models. The ANN achieved the highest performance (R² = 89.3%, RMSE = 0.387, MAE = 0.141), followed by RF (R² = 90.5%, RMSE = 2.737, MAE = 0.900) and SVR (R² = 87.8%, RMSE = 0.553, MAE = 0.169). Temperature omission led to significant performance degradation (R² reductions of 7–13%). Sensitivity analysis ranked temperature among the top three influential features. Formation temperature is a critical but underutilized parameter in ROP modeling. ML techniques, particularly ANN, demonstrate superior capability in capturing nonlinear temperature-dependent effects, offering actionable insights for real-time drilling optimization in thermally complex formations. This study provides a framework for integrating thermal data into predictive models to enhance drilling efficiency in the Niger Delta and analogous basins.

The drilling rate of penetration (ROP) is defined as the speed of drilling through rock under the bit. ROP is affected by different interconnected factors, which makes it very difficult to infer the mutual effect of each individual parameter. A robust ROP is required to understand the complexity of the drilling process. Therefore, an artificial neural network (ANN) is used to predict ROP and capture the effect of the changes in the drilling parameters. Field data (4525 points) from three vertical onshore wells drilled in the same formation using the same conventional bottom hole assembly were used to train, test, and validate the ANN model. Data from Well A (1528 points) were utilized to train and test the model with a 70/30 data ratio. Data from Well B and Well C were used to test the model. An empirical equation was derived based on the weights and biases of the optimized ANN model and compared with four ROP models using the data set of Well C. The developed ANN model accurately predicted the ROP with a correlation coefficient (R) of 0.94 and an average absolute percentage error (AAPE) of 8.6%. The developed ANN model outperformed four existing models with the lowest AAPE and highest R value.

Modeling the rate of penetration (ROP) plays a fundamental role in drilling optimization since the achievement of an optimum ROP can drastically reduce the overall cost of drilling activities. Evolved Extreme learning machine (ELM) with the evolutionary algorithms and multi-layer perceptron with Levenberg-Marquardt training algorithm (MLP-LMA) were proposed in this study to predict ROP. This paper focused mainly on two aspects. The first one was the investigation of the whale optimization algorithm (WOA) to optimize the weights and biases between input and hidden layers of ELM to enhance its prediction accuracy. The other was to adopt a prediction methodology that seeks to update the predictive model at each formation in order to reduce the dimension of input data and mitigate the effect of non real-time data such as the formation properties on the bit speed prediction. The prediction models were trained and tested using 3561 data points gathered from an Algerian field. The statistical and graphical evaluation criteria show that the ELM-WOA exhibited higher accuracy and generalization performance compared with the ELM-PSO and MLP-LMA. Furthermore, ELM-WOA was compared with two well-known ROP correlations in the literature, and the comparison results reveal that the proposed ELM-WOA model is superior to the pre-existing correlations. The findings of this study can help for the achievement of an optimum ROP and the reduction of the non-productive time. In addition, the outputs of this study can be used as an objective function during the real-time optimization of the drilling operation.

Lost circulation costs are a significant expense in drilling oil and gas wells. Drilling anywhere in the Rumaila field, one the world’s largest oilfields, requires penetrating the Dammam formation, which is notorious for lost circulation issues and thus a great source of information on lost circulation events. This paper presents a new, more precise model to predict lost circulation volumes, equivalent circulation density (ECD), and rate of penetration (ROP) in the Dammam formation. A larger data set, more systematic statistical approach, and a machine-learning algorithm have produced statistical models that give a better prediction of the lost circulation volumes, ECD, and ROP than the previous models for events. This paper presents the new model, validates the key elements impacting lost circulation in the Dammam formation, and compares the predicted outcomes to those from the older model. The work previously presented by Al-Hameedi et al. (http://www.onepetro.org, 2017a; http://www.AADE.org, 2017b) provided a platform for predicting the severity of lost circulation incidents in the Dammam formation. Using the new models, the predictions closely track actual field incidents of lost circulation. When new lost circulation events were compared with predictions from the old and new models, the new model presented a much tighter prediction of events. Three equations for optimizing operations were developed from these models focusing on the elements that have the highest degree of impact. The total flow area of the nozzles was determined to be a significant factor in the ROP model indicating that nozzle size should be chosen carefully to achieve optimal ROP. Good modeling of projected lost circulation events can assist in evaluating the effectiveness of new treatments for lost circulation. The Dammam formation is a significant source of lost circulation in a major oilfield and warrants evaluation of the effectiveness of lost circulation treatments. These techniques can be applied to other fields and formations to better understand the economic impact of lost circulation and evaluate the effectiveness of various lost circulation mitigation efforts.

Summary The rate of penetration (ROP) obtained with a roller-cone bit is limited by either the cuttings-generation or cuttings-removal process. An ROP model that includes the effect of both the initial chip formation and cuttings-removal processes is presented. Laboratory observations show that the reduction in ROP at high borehole pressure is the result of both local cratering effects and global cleaning effects.

Wells drilled in the Rumaila field are highly susceptible to lost circulation problems when drilling through the Hartha formation. This paper presents a comprehensive statistical work and sensitivity analysis models of the lost circulation events for more than 300 wells. Moreover, this study will demonstrate an integrated analysis regarding the most significant drilling parameters, which have a pivotal impact on the lost circulation to provide the greatest chance of mitigating or avoiding lost circulation in the Hartha formation. Lost circulation events are extracted from daily drilling reports, final reports, and technical reports. Key drilling parameters are analyzed using statistical software to understand the relationship between the mud losses and various drilling parameters such as MW, ECD, Yp, ROP, SPM, RPM, WOB, flow rate, and bit nozzles. The sensitivity analysis is conducted to examine the impact of each parameter in all models. In addition, variance inflation factor (VIF) method is used to test for the multicollinearity phenomena in each model to maximize the accuracy and to obtain a solid mathematical model. The volume loss model is conducted to predict lost circulation in the Hartha formation. As a proactive action, this model can be used to estimate the volume loss prior drilling the Hartha formation. Observations that have been made from the volume loss model are MW, ECD, and Yp have a significant impact on lost circulation respectively; however, SPM, RPM, and ROP have a minor effect on the volume loss model. Equivalent circulation density (ECD) model is obtained to estimate ECD in the Hartha formation, and from this model can be deduced that MW, ROP, and Q have a significant impact on ECD respectively; nevertheless, RPM and Yp have a minor impact on the ECD. The rate of penetration model is made to estimate ROP in the Hatha formation. It is concluded that WOB, SPM, and RPM have a significant impact on the ROP respectively, but MW, ECD, and Yp have a minor influence on the ROP. In addition, engineering solutions are developed to give a clear image regarding lost circulation, and it will provide a unique statistical study and coherent sensitivity analysis of all factors which have an essential or a small impact on this issue. Due to the lack of published studies for this formation, this study will provide a unique understanding for lost circulation events, drilling fluid properties, and operational drilling parameters, which have a prominent or a minor impact on the mud losses issue. Lost circulation will be illustrated in terms of causes, consequences, recommendations, and guidelines to reduce or avoid unwanted losses. In addition, this work can serve as a practical resource for drilling through the Hartha formation.

• Enhanced SSV airflow rate estimation of empirical models via data-driven modeling • Achieved high accuracy by customizing models individually for each room • Compared gray-box (physics-informed) with black-box (purely data-driven) models • Assessed data efficiency and model generalization using learning curve analysis • Revealed model robustness and key factors affecting airflow by sensitivity analysis Empirical models are widely used in building design and simulation to estimate airflow rates for single-sided natural ventilation (SSV). However, these models often suffer from limited accuracy. Data-driven modeling offers a viable solution to improve model accuracies by customizing models for various practical scenarios. This study investigates the feasibility and performance of two data-driven modeling approaches for SSV: gray-box modeling (physics-informed using polynomial regressions) and black-box modeling (purely data-driven using machine learning regressions). Taking SSV through a solitary bottom-hung window as a case study, training data were acquired from a full-scale test bench via tracer gas measurements conducted over 11 days. The results indicate that both modeling approaches significantly improve accuracy compared to existing empirical models from the literature. Notably, individual models tailored for each test room generally outperform generalized models developed for all rooms. This finding highlights the importance of customizing models on-site in real-world applications. When comparing the two modeling approaches, gray-box models require lower training ratios to achieve good generalization, demonstrating higher data efficiency than black-box models. Furthermore, sensitivity analysis reveals that gray-box models are more robust and reliable during interpolation and extrapolation, making them particularly suitable for scenarios with limited data availability. Incorporating opening area corrections and selecting root mean square error (RMSE) as the objective function are crucial for achieving optimal gray-box model performance. Overall, this study highlights the potential of physics-informed gray-box data-driven modeling as an advanced strategy for enhancing SSV airflow rate estimation, providing valuable insights to guide future research.

Rate of drilling bit penetration into the subsurface formations, referred to as rate of penetration (ROP), is considered as the main optimization parameter in the drilling operation. ROP optimization results in a faster and cheaper operation. The rate of bit penetration depends on several factors, including but not limited to rock properties, fluid characteristics and drilling operational parameters. Due to the diversity of affecting parameters, prediction of rate of penetration is a challenging task. There are various mathematical relations in the literature to estimate ROP. However, these relations are developed based on specific conditions, where data from a field or experimental tests are used to develop the relations. In this work, drilling data of three wells from one onshore oil fields are gathered and then the performances of some common ROP models are investigated. Results show that the simplified Bourgoyne and Young model can accurately predict the ROP in the mentioned field, which makes the model applicable for planning of the new offset wells in the future. It should be noted that the model requires fewer input data, which makes it more applicable. The results show the average R2 coefficient for the model is 0.91, which is higher than other models. The results confirmed the applicability of the model.