Khan, Khaqan (Saudi Aramco) | Almarri, Misfer (Saudi Aramco) | Al-Qahtani, Adel (Saudi Aramco) | Syed, Shujath Ali (Baker Hughes, a GE Company) | Negara, Ardiansyah (Baker Hughes, a GE Company) | Jin, Guodong (Baker Hughes, a GE Company)
Rock mechanical properties are required as an input in many petroleum engineering applications, such as borehole stability analysis, hydraulic fracturing design, and sand production prediction. Their determination is commonly from various laboratory testing performed on subsurface rock samples. Due to the scarcity of reservoir samples and test cost, rock mechanical data are always very limited. Therefore, empirical correlations are very often used to estimate the mechanical properties from downhole logging measurements. Alternatively, the data-driven analytics techniques have been developed for predicting rock properties from other formation properties that can be determined directly from logs.
This paper presents a study of developing correlation equations and data-driven models that are used to predict the unconfined compressive strength (UCS) from logging data. Various rock mechanical tests including UCS, single- and multi-stage triaxial tests are performed on sandstone samples from three wells in one region. UCS values are obtained either from the UCS testing directly or from the Mohr-Coulomb failure analysis indirectly. Rock properties, such as mineralogy, porosity, grain and bulk density, ultrasonic wave velocities, are measured for each tested sample, which are used to build the correlations and data-driven analytical models for predicting UCS. Results shows that the empirical correlations are not universal and often cannot be used without some modifications, while the data-driven model is more generalized in application. In addition, data quality is very crucial for building correlations or predictive models.
Negara, Ardiansyah (Baker Hughes, a GE Company) | Ali, Syed (Baker Hughes, a GE Company) | AlDhamen, Ali (Baker Hughes, a GE Company) | Kesserwan, Hasan (Baker Hughes, a GE Company) | Nair, Asok (Baker Hughes, a GE Company) | Aleid, Zahra (Baker Hughes, a GE Company)
Maintaining a stable borehole is one of the major tasks during drilling operations. During the drilling, borehole breakout and drilling induced fractures are the two main instability problems which may lead to stuck pipe, sidetracking, and loss of circulation. To evaluate the stability of a wellbore, a constitutive model is required to estimate the stresses around the wellbore coupled with a failure criterion to predict the ultimate strength of reservoir rocks. The Mohr-Coulomb failure criterion is one of the commonly accepted criteria for rock strength estimation at a given state of stress. This failure criterion is mainly contributed from the cohesion and coefficient of internal friction parameters, which are determined by laboratory measurements. The laboratory measurements, although more reliable, are expensive and timeconsuming. This paper discusses artificial intelligence models particularly multilayer perceptron (MLP) and support-vector regression (SVR) for predicting cohesion and coefficient of internal friction from elemental spectroscopy and petrophysical properties.
Elemental spectroscopy, density, porosity, cohesion, and coefficient of internal friction data presented in this paper are based on various geological formations. Cohesion and coefficient of internal friction are determined through a rock mechanical test in the laboratory, while elemental spectroscopy data were obtained from X-ray fluorescence (XRF) analysis. We divide the data set into training and testing data. Training data is used to train MLP and SVR then establishes the cohesion prediction models. Similarly, training data is used to train and construct the MLP and SVR-based coefficient of internal friction models. Both models are then examined using the testing data.
Cohesion and coefficient of internal friction predicted from MLP and SVR match well with the laboratory measurements. Two quantitative measures for estimation accuracy are used including coefficient of determination and mean absolute percentage error. Cross-correlation plots of predicted cohesion and coefficient of internal friction and the experimental results show very good coefficient of determination and relatively small error. The results demonstrate that amongst the MLP and SVR models, the models whose inputs are grain density, porosity, and elemental spectroscopy are the best models. From a practical point of view, the application of artificial intelligence techniques as a new method for indirect estimation of rock failure parameters are beneficial especially when the amount of core samples are relatively few.
Brittleness index is one of the critical geomechanical properties to understand the rock’s drillability in drilling operations and screen effective hydraulic fracturing candidates in unconventional reservoirs. Brittleness index can generally be obtained from stress or strain based relationships. It can also be estimated from conventional well logs or rock mineralogical composition. Brittleness index measurements from stress/strain based relationships require laboratory tests, which are time-consuming and core samples are available only at discrete depths. While well logs can estimate a continuous profile of brittleness index along the borehole, it is derived from empirical correlation specific to a rock type. More recent advancements in logging tools have enabled the determination of elemental spectroscopy downhole. This information combined with petrophysical properties such as density and porosity can capture brittleness characteristics of rocks. This paper presents the use of support-vector regression (SVR) to construct a data-driven brittleness index prediction from the elemental spectroscopy and petrophysical properties.
The relationship of brittleness index with elemental spectroscopy, density, and porosity is often complex and nonlinear. The SVR described in this paper is used to correlate the elemental spectroscopy, density, and porosity to the brittleness index, thereafter building a data-driven brittleness index prediction model. The dataset of brittleness index, elemental spectroscopy, density, and porosity used in this study are based on various geological formations. Laboratory tests such as unconfined compressive strength, confined compressive strength, and Brazilian test were conducted. Brittleness indices were calculated based on data generated from these tests. Elemental spectroscopy data were obtained from X-ray fluorescence (XRF) analysis. The data are then separated into two categories: training and testing data. Training data are used to train the SVR and establish the brittleness index prediction model, while the testing data are used for validation.
In total, 28 cases were run with different combinations of petrophysical and mineralogical properties, number of training dataset, and SVR kernel functions. The results reveal that the SVR-based brittleness indices match very well with the laboratory-measured brittleness indices. Cross-correlation plots of regression models between the predicted and the measured brittleness indices show high values of coefficient of determination. The small error and high values of coefficient of determination denote the SVR models’ good performances. The prediction accuracy improves as more data are included to train the algorithm. From the comparison of SVR-kernel-function-based models, we observe that the RBF-based model performs better than the polynomial-based model. The RBF-based model yields better accuracy than the polynomial-based model using the same number of training dataset. Referring to the RBF-based model with 80% training dataset, it was observed that elemental spectroscopy has more influence than the other rock properties on the prediction. The promising results stemming from this study confirm that SVR can be further applied to build a brittleness index prediction model based on mineralogy logs and petrophysical logs.
Unconfined compressive strength (UCS) of rock is a key parameter in drilling and stimulation of oil and gas wells such as wellbore stability and fracturing operations. Better estimates of UCS could increase the efficiency of drilling and stimulation operations. Current techniques for UCS determination either rely on laboratory measurements or empirical relationships using well logs. The laboratory measurements, although more reliable, are expensive and time-consuming. On the other hand, the empirical relations derived from core data and well logs are very unique because they are developed for a specific rock type; hence has limited applications. This paper presents a data-driven model for UCS prediction from petrophysical properties and elemental spectroscopy using artificial intelligent technique, namely support-vector regression (SVR).
Elemental spectroscopy, density, porosity, and UCS data presented in this paper are based on various geological formations. UCS is determined by the uniaxial compression test in the laboratory, while elemental spectroscopy was obtained from X-ray fluorescence (XRF) analysis. We first use SVR to establish a correlation between elemental spectroscopy, density, and porosity with UCS. We separate these data into two categories: training and testing data. Training data is used to train SVR and establishes the UCS prediction model. The model will generate UCS prediction using testing data and compared with the laboratory-measured UCS.
In total, 21 cases were run with different combination of input parameters. Good agreement was observed between the SVR-predicted UCS and the laboratory measurement. Two quantitative measures for estimation accuracy are calculated and examined including the coefficient of determination and the mean absolute percentage error. Considering limited number of available data used in this study, the SVR-predicted UCS produces very good coefficient of determination and small error. The results also demonstrate the significant influence of elemental spectroscopy on the UCS prediction because elements determine grain density, which contributes to the rock strength. This emphasizes the advantage of incorporating elemental spectroscopy, together with other petrophysical properties, for UCS prediction. The favorable results in this study demonstrate the promising capability of SVR to build a UCS-prediction model based on elemental spectroscopy and petrophysical properties. Further application of SVR can be adapted to predict UCS directly from mineralogy logs and conventional well logs.
Successful exploitation of shale reservoirs largely depends on the effectiveness of hydraulic fracturing stimulation program. Favorable results have been attributed to intersection and reactivation of pre-existing fractures by hydraulically-induced fractures that connect the wellbore to a larger fracture surface area within the reservoir rock volume. Thus, accurate estimation of the stimulated reservoir volume (SRV) becomes critical for the reservoir performance simulation and production analysis. Micro-seismic events (MS) have been commonly used as a proxy to map out the SRV geometry, which could be erroneous because not all MS events are related to hydraulic fracture propagation. The case studies discussed here utilized a fully 3-D simulation approach to estimate the SRV.
The simulation approach presented in this paper takes into account the real-time changes in the reservoir's geomechanics as a function of fluid pressures. It is consisted of four separate coupled modules: geomechanics, hydrodynamics, a geomechanical joint model for interfacial resolution, and an adaptive re-meshing. Reservoir stress condition, rock mechanical properties, and injected fluid pressure dictate how fracture elements could open or slide. Critical stress intensity factor was used as a fracture criterion governing the generation of new fractures or propagation of existing fractures and their directions. Our simulations were run on a Cray XC-40 HPC system.
The studies outcomes proved the approach of using MS data as a proxy for SRV to be significantly flawed. Many of the observed stimulated natural fractures are stress related and very few that are closer to the injection field are connected. The situation is worsened in a highly laminated shale reservoir as the hydraulic fracture propagation is significantly hampered. High contrast in the in-situ stresses related strike-slip developed thereby shortens the extent of SRV. However, far field nature fractures that were not connected to hydraulic fracture were observed being stimulated.
These results show the beginning of new understanding into the physical mechanisms responsible for greater disparity in stimulation results within the same shale reservoir and hence the SRV. Using the appropriate methodology, stimulation design can be controlled to optimize the responses of in-situ stresses and reservoir rock itself.
It is timely for our industry to introspect on ways for step improvement in the utility of the wireline and logging-while-drilling logs which remains central to any asset development. Interpretation is limited by our current understanding of rock-fluid physics in source rocks, which is still developing. The gap is clearly evident in unconventional source rock interpretation where approximations such as pseudo-Archie approach are used for saturation estimation. This paper presents the use of emerging knowledge in machine learning to demonstrate its applicability for improving the total organic carbon (TOC) estimation in an unconventional well and permeability prediction in a conventional well.
We have used the support-vector regression (SVR) technique, which is a new machine learning technique. Vast amount of logging data can be quickly processed using this technique. Limited core data is used to train the SVR algorithm. In this work, we first use the SVR technique to establish a correlation between conventional well logs (e.g., gamma ray, formation resistivity, neutron porosity, bulk density) and core measurements, thereafter building a rock property-prediction model as a function of well logs selected. Two field datasets from a South American well and a Mississippi Canyon well were selected to validate the method. Both wells contain a suite of logs and few core TOC and permeability data. Various combinations of conventional well logs were studied to check if the prediction accuracy can be improved.
The results of the two case studies reveal that the SVR technique provides accurate and reliable TOC and permeability predictions. We observed in the case study of TOC prediction that incorporating the carbon weight fraction log as an input, in addition to the conventional well logs, improves the TOC prediction because the carbon weight fraction log provides information about the amount of carbon, which eventually helps the SVR algorithm to learn better from the data. Meanwhile, for the case study of permeability prediction, we observed that the conventional well logs are sufficient to generate a good permeability prediction model. Additional logging data including the nuclear-magnetic-resonance (NMR) logging data do not improve significantly the prediction accuracy. In conclusion, SVR technique could be used to improve our log interpretation. This technique can be easily adapted to predict rock mechanical properties and especially useful for unconventional reservoirs where traditional models may not be applicable and new methods are still evolving. Such new data analysis technologies could optimize our logging service and core analysis planning.
Hussain, Maaruf (Baker Hughes) | Saad, Bilal (Baker Hughes) | Negara, Ardiansyah (Baker Hughes) | Elgassier, Mokhtar (Baker Hughes) | Agrawal, Gaurav (Baker Hughes) | Sun, Shuyu (University of Science & Technology) | Abdullah, King (University of Science & Technology)
It is often reported that around 60% of hydraulic fracturing stages are ineffective. If so, it is likely that the design accuracy is limited by the current state of modeling and hydraulic fracture (HF) simulations. Our study presents a new alternative – a full 3-D simulation with geomechanics coupled to fluid flow. With the conventional simulation, it is extremely hard to model opening of weak lamination (Lam) and nearly impossible to generate induced horizontal fractures against the vertical overburden stress. However, horizontal fractures are routinely evident in shale reservoirs as healed fractures observed along the bedding planes. Hence, the need and importance of a true 3-D simulator that could incorporate complex geology and dynamically simulate fracture propagation by accounting for realtime changes in geomechanics and fluid pressures. Case study uses shale reservoirs, which are heavily laminated with complex natural fractures (NFs). Numerical simulations consisted of four separate coupled modules - geomechanics, hydrodynamics, a geomechanical joint model for interfacial resolution, and an adaptive remeshing module. Reservoir stress condition, rock mechanical properties, and injected fluid pressure dictate how fracture elements could open or slide. Critical stress intensity factor was used as a fracture criterion governing the generation of new fractures or propagation of existing fractures and their directions. Simulation was run on a Cray XC-40 HPC system. Typical laminated shale reservoirs anisotropic geomechanical properties obtained from literature were used to estimate a 3-D geomechanical model and NF network. HF geometry was significantly different in the presence of weak bedding, compared to when bedding was strong enough to transmit crack tip stresses across the interface. Significant amounts of fracturing fluid can be diverted into creation of horizontal fractures, even when the pressure was below the vertical stress, once bedding discontinuities are activated. Choices of NF network and Lam thickness significantly affected observed fracture propagation. The value of 3-D modeling was clearly established. This method provides more accurate solutions for stimulation design optimization, e.g., landing points, number of stages, number of clusters, spacing between stages, and stimulated reservoir volume.
Production from unconventional reservoirs like shale gas has increased considerably in the past few years due to the advancement in twofold, i.e., horizontal drilling and hydraulic fracturing technologies. Although there has been some success in increasing gas production from shale reservoirs, unfortunately, the physicochemical processes that take place in the shale formations remain challenging and are not completely understood. Unlike conventional reservoirs, shale reservoirs are characterized by very small porosity and extremely low-permeability. Gas flow in this tight formation involves complex flow processes such as Knudsen diffusion, Klinkenberg effect, adsorption and desorption, strong rock-fluid interaction, rock deformation, etc. Furthermore, because of high pressure and high temperature reservoir conditions the gas behaves as real gas. In this work, our shale gas mathematical model is built based on the dual-porosity dual-permeability model that incorporates the complex flow processes mentioned above as well as the thermodynamic calculations. Peng-Robinson equation of state (PR-EOS) was used to calculate the gas density and compressibility factor by solving the cubic equation. In the numerical method implementation we combine the finite difference method with the experimenting pressure field approach to solve the pressure equations for the matrix and fracture systems in the dual-porosity dual-permeability model. This combination greatly reduces the computational cost when solving the large systems of pressure equations of the matrix and fracture. In this approach, a set of predefined pressure fields is generated in the solution domain such that the undetermined coefficients are calculated from these pressure fields. In the numerical example, we considered a shale reservoir with single production well. Comparison between real gas and ideal gas is studied and the result shows that considering the real gas behavior generates higher cumulative production, which implies that the gas transport capacity is higher than the ideal gas case. The result also indicates that considering real gas behavior in the model would increase the production and retard the decline curve. Therefore, it is very important to incorporate the real gas behavior into the model in order to be able to forecast the production accurately.
Shabab, Mohamad (University of Waterloo) | Jin, Guodong (Baker Hughes Dhahran Global Technology Center) | Negara, Ardiansyah (Baker Hughes Dhahran Global Technology Center) | Agrawal, Gaurav (Baker Hughes Dhahran Global Technology Center)
The need for improved data accuracy, cost effectiveness and delivery time to assist in decision making has gained importance in reservoir characterization and evaluation. This paper presents a robust and inexpensive data-driven method for predicting the formation permeability profile from conventional well logs (CWLs) using the support vector regression (SVR) technique with limited core measurements. The method's feasibility and applicability are demonstrated on one field data set from a North Sea well contained a complete suite of logs and extensive core measurements.
The relationship between formation permeability and well logs is often overwhelming complex and nonlinear. We use the SVR method to establish the correlation between CWLs and limited core permeability, thereafter building a permeability-prediction model as a function of selected well logs. The basic logging data used here include density, neutron, deep resistivity, compressional and Stoneley wave slowness. The permeability derived from these well logs generally compares well with the measured core permeability. Additional logging data including the clay weight fraction, thorium/potassium content, or nuclear-magnetic-resonance (NMR) bulk volume movable and irreducible are also separately integrated into those basic logs to determine if the prediction accuracy can be improved. There is no obvious difference among the predicted permeability profiles even these additional well logs are added, which could imply that the basic logs are sufficient to generate the permeability with good accuracy.
SVR method could be used to improve the log interpretation accuracy as shown in this study. It can be easily adapted to predict other rock electrical, mechanical and petrophysical properties when only conventional logs and few core measurements are available. It is especially useful for unconventional reservoirs where traditional models may not be applicable and new methods are still evolving. Such new data analysis technologies could optimize our logging service and core analysis planning.
Saad, Bilal (Baker Hughes) | Negara, Ardiansyah (Baker Hughes) | Hussain, Maaruf (Baker Hughes) | Elgassier, Mokhtar (Baker Hughes) | Sun, Shuyu (University of Science and Technology) | Abdullah, King (University of Science and Technology)
Hydraulic fracture stimulation designs are typically made of multiple stages placed along the lateral section of the well using various well completion technologies. Understanding how multiple hydraulic fractures propagate and interact with each other is essential for an effective stimulation design. The number and placement of stages are important factors for optimizing the performance of the laterals. This in turn depends on accuracy in determining fracture interference. We present advanced simulations for accurate placement of well stages. In this paper, we use a 3-D fully coupled geomechanical-fluid flow simulator which incorporates anisotropic geomechanical properties. Densely complex natural fractures and lamination are built into the model based on available core and log information. Multiple fractures are concurrently imployed to simulate real life scenarios. Fluid pressures are incrementally computed such that stress state changes dynamically with time as it happens in real field situation. Our simulations were run on Cray XC 40 HPC system. The results demonstrate that the stress shadow effects can significantly alter hydraulic fracture propagation behavior, which eventually affects the final fracture geometry. The results show that there are large differences in aperture throughout the stimulation which persists to the end of pumping. Furthermore comparison between cases with and without complex natural fractures (discrete fracture network (DFN)) and lamination was conducted with even and uneven spacing configurations. Fracture interference and spacing analysis conducted based on model with perforation frictions shows that while spacing between fractures is important, the largest impact was observed in the presence of lamination and DFN. The large differences in the way the fracture propagates highly depend on the DFN connectivity. Late-stage connection throughout the model implies later disconnection when the pressure drops. Though the computations are time intensive, we believe this is a valuable tool to use in the planning stages for asset development to increase production potential.