The performance of stochastic designs in wellbore drilling operations
 502 Downloads
Abstract
Wellbore drilling operations frequently entail the combination of a wide range of variables. This is underpinned by the numerous factors that must be considered in order to ensure safety and productivity. The heterogeneity and sometimes unpredictable behaviour of underground systems increases the sensitivity of drilling activities. Quite often the operating parameters are set to certify effective and efficient working processes. However, failings in the management of drilling and operating conditions sometimes result in catastrophes such as well collapse or fluid loss. This study investigates the hypothesis that optimising drilling parameters, for instance mud pressure, is crucial if the margin of safe operating conditions is to be properly defined. This was conducted via two main stages: first a deterministic analysis—where the operating conditions are predicted by conventional modelling procedures—and then a probabilistic analysis via stochastic simulations—where a window of optimised operation conditions can be obtained. The outcome of additional stochastic analyses can be used to improve results derived from deterministic models. The incorporation of stochastic techniques in the evaluation of wellbore instability indicates that margins of the safe mud weight window are adjustable and can be extended considerably beyond the limits of deterministic predictions. The safe mud window is influenced and hence can also be amended based on the degree of uncertainty and the permissible level of confidence. The refinement of results from deterministic analyses by additional stochastic simulations is vital if a more accurate and reliable representation of safe in situ and operating conditions is to be obtained during wellbore operations.
Keywords
Well stability Stochastic analysis Deterministic analysis Mud pressure Safe mud window Wellbore drilling Rock properties1 Introduction
An overview of experiences during the drilling and production of hydrocarbon from wells indicates rampant incidences arising from wellbore instability. The wellbore system becomes unstable when the integrity of the wellbore and surrounding formation can no longer hold or is threatened due to induced stresses or the weakening of the wellbore or formation materials. Wellbore instability poses a major problem during drilling, and its causes can be categorised into mechanical and chemical effects. Pašić et al. (2007) classify the factors contributing to wellbore instability as uncontrollable (natural) and controllable factors. Natural factors include the presence of naturally fractured or faulted formations, tectonically stressed formations, high in situ stresses, mobile formations, unconsolidated formations, naturally overpressured rock collapse and induced overpressure rock collapse; controllable factors include bottomhole pressure (mud density), well inclination and azimuth, transient pore pressures, physicochemical rock–fluid interaction, drill string vibrations, erosion and temperature. Other factors which affect wellbore stability are the orientation of in situ stress fields, the mechanical properties of rock and bedding planes, and pore pressure (Chen et al. 1997).
The wellbore trajectory and mud density (also known as mud pressure) are amongst the factors which have a significant impact on the stability. Deviated wells have a greater tendency to become unstable (Standifird 2006) and can be measured in terms of the inclination and azimuth of wells with respect to the principal stresses. Wellbore failure happens when the tensile or shear strength of the formation and bedding plane is exceeded. To prevent this, the rock and bedding plane must be kept intact.
The mud density (mud pressure) is a dominant parameter that greatly influences the stability of wells, especially while drilling is being performed (Pašić et al. 2007). Pressure exerted by the drilling fluid (mud) instigates an additional concentration of stresses in the surroundings of the wellbore. Since the presence of effective stresses impacts on rock material behaviour, including failure, stability is highly dependent on the management of the mud pressure. The magnitude of mud pressure applied has to be adequate to avoid damage. Optimal values are usually in the high range; however, if the pressure is too high it may result in tensile fracturing and fluid loss, which are typical causes of instability. On the other hand, a mud pressure that is below the threshold (critical) value may not be sufficient in providing the necessary stress counterbalance to forestall collapse due to a preponderance of shear failure. The appropriate range of safe mud pressure is dependent on the local factors controlling individual cases and may differ for each scenario.
A classic example as illustrated in Mohiuddin et al. (2007) is the dependence of mud pressure on well inclination and azimuth. The susceptibility of deviated wells implies that they are more likely to fail if the same conditions used for vertical wells are applied. This is demonstrated in Mohiuddin et al. (2007), where comparisons of the mud density requirement between vertical, directional and horizontal wells are presented, indicating that generally greater magnitudes of mud densities are needed for nonvertical wells. It is inferred that horizontal wells require the highest range of values of mud densities. The derived critical mud pressure data and contour plots can be applied directly when designing wellbore alignment. Applications of this sort (the production and utilisation of critical mud pressure contour plots) are shown by Tan and Willoughby (1993) and Tan et al. (2004). Time dependency of the critical mud pressure is realised if there are temporal changes in controlling parameters such as rock material properties (e.g. cohesive strength). Chen et al. (2003a) showed a significant variation in critical mud weight when the shale cohesive strength changes with time.
Wellbore stability is also impacted by chemical interactions between drilling fluids (mud) and the host rock. Activities including ion exchanges and modifications in swelling pressure and rock water content are examples of chemical alterations; they occur when there is a disparity between the water activity in the rock and the water activity in mud (Chen et al. 2003a). Where the mud water activity is lower, the reduction in pore pressure and the corresponding increase in effective stresses increase stability (Chen et al. 2003a). In Ma and Chen (2015), a collapse pressure wellbore stability model for shale reservoirs was developed based on the analytical solution of stress induced by mechanical, hydraulic and chemical effects. The model is proposed for the assessment of the collapse pressure of shale reservoirs, and unlike conventional models, it shows the occurrence of failure regions not only at the borehole surface, but also at the interior of the formation. They demonstrate that rock strength parameters decrease with exposure to drilling mud, and in the formation, pore pressure increases while solute concentration decreases when the solute concentration of the drilling mud is less than that of the fluid in the pore space. A decrease in rock strength and an increase in pore pressure impact on wellbore stability in shale reservoirs. As illustrated by van Oort (2003), fluid–rock interaction can be managed so as to improve well stability or prevent instability.
The effect of temperature on wellbore stability can be observed when there is thermal diffusion within the formation. An increase in the formation temperature through the application of hotter drilling fluids adds to the pore pressure, thereby increasing the risk of instability (Chen et al. 2003a). In hydratebearing sediments (HBS), an increase in temperature has been shown in FreijAyoub et al. (2007) to speed up the dissociation of hydrates, causing corresponding reductions in cohesion.
The risk of instability is influenced by fractured reservoir formations. Fractured rock masses are embedded with natural discontinuities comprising bedding planes and fractures, which affect their homogeneity and overall physical and mechanical properties. Hence, apart from the failure of the intact rock, wellbore instability may be instigated at the planes of natural discontinuities. Chen et al. (2003b), Chen and Tan (2001) and Zhang et al. (1999) studied the effect of fractured rock masses on wellbore behaviour. It was ascertained that the probability of instability due to high differential stresses was considerably increased by the presence of fractures. Fracture patterns have variable effects due to differences in spacing, size, alignment, connectivity and strength property. Mud infiltration into fractures reduces their friction angle, causing a significant increase in the tendency for instability (Chen and Tan 2001; Chen et al. 2003b). Instability in fractured rock masses are mainly initiated along planes of discontinuity.
Uncertainty is inherent in wellbore design and drilling. Within a wider context it is generally split into two or three categories: aleatory uncertainties, epistemic uncertainties, and errors (Bulleit 2008; Chalupnik et al. 2009). Whereas aleatory uncertainties occur from randomness or contingency, epistemic uncertainties arise due to deficiencies in human knowledge. According to Bulleit (2008), sources of uncertainty include time, statistical limits, model limits, randomness and human error. Our focus is on uncertainties principally caused by randomness in material properties and underground conditions. This can be caused by inherent inconsistencies and unclear information due to limitations in test data (Savoia 2012). Parameters affecting wellbore stability consist of rock strength, magnitude and orientation of principal stresses, well orientation, pore pressure and mud pressure (Moos et al. 2003). The variability of these parameters implies a great deal of uncertainty during wellbore design, drilling and operation. While the other parameters are often uncontrollable, mud pressure (also referred to as mud density or mud weight) is an operational measure necessary to maintain stability.
Because we have limited our wellbore design and analysis in this study to a single vertical well, the magnitude and orientation of principal stresses, pore pressure and well orientation are assumed to be consistent at a given depth. Hence, the variability in the rock formation will be viewed as changes in rock material strength and deformation properties; amongst these, the Poisson ratio is considered the most unpredictable and as such also chosen as one of the variables to be stochastically modelled.
The pattern describing the uncertainties of design variables are probability distributions that can be assigned based on the trends of statistical dispersions including Gaussian normal, lognormal, Bernoulli sequence and Poisson distributions. The uncertainty in material properties can thus be designated according to prescribed probability density functions. Although a deterministic approach can be employed to define the safe mud pressure by observing the stress responses, there are some inherent limitations, so it does not account for all the uncertainties mentioned above. This research aims at carrying out followup stochastic analyses to investigate the robustness of wellbore conditions and design and to test the reliability of results from preceding deterministic analyses.
1.1 Review of wellbore stability studies
Some probabilisticbased approaches have been adopted in studies of wellbore stability. One of such methods is quantitative risk assessment (QRA) (e.g. Ottesen et al. 1999; McLellan and Hawkes 1998), which was employed by Moos et al. (2003) to determine the effect of uncertainties in input parameters (rock and reservoir properties) on well stability and optimal mud weight windows. Ottesen et al. (1999) had earlier introduced a QRAbased statistical technique—specifically for wellbore stability analyses—to measure uncertainties in input data and the probabilities of their effect in relation to mud pressures. An approach akin to this was applied by McLellan and Hawkes (1998) in modelling sand production. The input parameters used in Moos et al. (2003) are uniaxial compressive strength (UCS), pore pressure and in situ principal stresses (the vertical and two horizontal components). The response surfaces, typifying the wellbore behaviour, were calculated as quadratic polynomial functions of each input parameter. Monte Carlo simulations were used to compute uncertainties in wellbore collapse and lost circulation pressures. Quantitative risk assessment, based on the Monte Carlo method, was also applied by Moos et al. (2003) to assess uncertainties in seismic velocities and velocity transforms (velocitydensity functions and velocityeffective stress functions), as they impact estimations of density, effective stresses and pore pressure. This information can be applied in determining the sealing pressure of rocks (reservoir), the mud pressure window, and the required number of drilling casings. This method of probabilistic technique often requires an extensive and densely populated sample size.
Latter studies (AlAjmi and AlHarthy 2010; AlKhayari et al. 2016; Niño 2016; Sheng et al. 2006) have included some aspects of sensitivity analyses using, for instance, the oneatatime (OAT) technique, to identify critical parameters. To quantify uncertainties in input data (rock properties), Niño (2016) applied four approaches: expert judgement; spatial variability; indirect measurement [a procedure borrowed from Holzberg (2001)]; and inconsistency of data sources, using Monte Carlo simulation. Monte Carlo simulation was applied in the model output uncertainty analyses and used to derive safe mud windows based on probability estimates. Sensitivity analyses were also completed using both the oneatatime (OAT) method and the response surface methodology (RSM). The maximum horizontal stress and cohesion were key to determining collapse pressure, since they had the most influence, while the maximum and minimum stresses played a similar role in estimating the fracture pressure. The Poisson ratio and vertical stress were perceived to have trivial effects on responses. Similarly, critical mud pressures have been estimated through probabilistic wellbore stability analysis where Monte Carlo sampling techniques were used to capture uncertainties in in situ stresses, wellbore trajectory, cohesion, friction angle, Poisson ratio (AlAjmi and AlHarthy 2010; AlKhayari et al. 2016; Sheng et al. 2006), pore water pressure and rock strength (Sheng et al. 2006). Wellbore trajectory was determined as the most influential parameter causing wellbore collapse (AlKhayari et al. 2016), while other critical parameters impacting on wellbore stability were identified as friction angle, cohesion and maximum horizontal stress (AlAjmi and AlHarthy 2010). Comparisons between Mogi–Coulomb and Mohr–Coulomb failure criteria indicate that the former produces greater (conservative) magnitudes of minimum overbalance pressures (AlAjmi and AlHarthy 2010).
In addition, deterministic wellbore stability models have been developed for the following purposes: to define safe mud pressure windows (Aslannezhad et al. 2016); for wellbore stability assessments which allow correlations through the use of limited available input data to derive others such as in situ stresses and some rock mechanical properties (Simangunsong et al. 2006); for well path optimisation (Ma et al. 2015); and to compare the outputs of failure models such as Mogi–Coulomb and Mohr–Coulomb failure criteria (Aslannezhad et al. 2016; Ma et al. 2015), and Mohr–Coulomb, Drucker–Prager and modified Lade failure criteria (Simangunsong et al. 2006). For instance, Ma et al. (2015) derived a semianalytical model for wellbore stability analysis from the analytical solution of stress distribution around a borehole, rock failure criteria and a breakout width model. This model was used to compare the performance between Mohr–Coulomb and Mogi–Coulomb failure criteria, to calculate the mud weight extrema and to establish the most stable well path. The Mohr–Coulomb criterion is shown to be more conservative than the Mogi–Coulomb criterion, and contrary to conventional methods, the optimal stable well path using this method is shown to be vertical for normal faulting (NF), normal to strikeslip faulting (NFSS) and strikeslip faulting (SS) stress regimes.
1.2 Focus of study
To address the high variability in underground conditions, stochastic methods are being used to reflect the temporal and spatial changes during drilling and operations. The uncertainty is applicable to a wide range of parameters comprising pore pressure, uniaxial compressive strength (UCS), in situ stresses, Poisson ratio, void ratio, tensile strength, angle of internal friction, cohesion, elastic modulus, etc. Hitherto, elastic modulus, Poisson ratio and void ratio are parameters that are largely omitted in wellbore stability investigations. A plausible reason for the noninclusion of void ratio is that direct measurements and data are not readily available, especially within subsurface environments. Furthermore, the Monte Carlobased probabilistic techniques commonly employed often require an extensive and densely populated sample size.
Consequently, this study considers the ramifications of uncertainties in void ratio, Poisson ratio and elastic modulus. These are sensitive properties with significant corollaries that reflect in the trend of other properties. The elastic modulus is a function of the compressive strength and strain of the rock. It is a deformation parameter as it determines the extent of material distortion for given imposed stress conditions. The void ratio is a measure of consistency and packing of the rock grains and has several ramifications through, for example, estimates of porosity, specific gravity, density and saturation. The Poisson ratio defines the attributes of alterations in the morphology of the rock under imposed stress conditions. The relationships between the elastic, bulk and shear moduli are readily quantifiable where appropriate estimates of the Poisson ratio and its uncertainties are available. In place of quadratic polynomial functions, the response surface in this study is characterised explicitly by a finite element geomechanical wellbore model. A linkage allows the exchange of information between the finite element wellbore model and the stochastic model. Traditional Monte Carlo simulations are also replaced by quasiMonte Carlo simulations which circumvent the need for large samples.

create a platform that engenders quantitative comparisons of outputs between deterministic and stochastic predictions of safe mud windows,

demonstrate the performance of quasiMonte Carlo integration/sampling as a suitable method of achieving low discrepancies and decreased clustering during selection,

apply concurrent alterations of input variables during sampling (each selected from a repository of Gaussian distributed values),

illustrate the potential of a procedure that integrates deterministic and stochastic numerical models to obtain synchronised and optimised solutions, and

present the distinct combination of Poisson ratio, elastic modulus and void ratio as characteristic input rock properties.
2 Numerical procedure
2.1 Domain description
Rock property  Sandstone (top 1000 m)  Shale  Sandstone  Chert 

Mass density ρ_{m}, kg/m^{3}  2500  2500  2500  2500 
Wet bulk density ρ_{b(wet)}, kg/m^{3}  2128  2271  2215  2304 
Dry bulk density ρ_{b(dry)}, kg/m^{3}  1880  2119  2024  2174 
Elastic modulus E, Pa  2.32e+10  3.8e+9  2.32e+10  5.5e+10 
Poisson ratio \(\upsilon\)  0.225  0.18  0.225  0.20 
Porosity n  0.25  0.15  0.257  0.13 
Void ratio e  0.330  0.176  0.346  0.149 
Permeability k, m^{2}  1.97e−13  2.2e−20  1.97e−13  2.2e−20 
Specific gravity G_{s}  2.2–2.8  2.4–2.8  2.2–2.8  2.6–2.7 
Deep ocean drilling operations are typically carried out to extensive depths below the ocean floor. The average depth of oceans ranges from 1205 m for the Arctic Ocean to 3970 m for the Pacific Ocean with a maximum depth of up to 11,034 m recorded for the Challenger Deep located in the Pacific Ocean. Although the terms ocean and sea are often used interchangeably, a sea actually refers to that portion of an ocean that is partially enclosed by land and is shallower. For this model, a depth (2000 m) in between the lower and upper limits of average values is used. A rock depth of 1000 m below the ocean floor is selected as the top of the segment (Fig. 3). In essence, the top of the rock segment is considered to be 3000 m below sea level.
2.2 In situ and induced stresses
For Eq. (6), h is the depth of the cover, that is, the depth measured from the surface to the point of interest; K varies between 0.4 and 1.5 for depths below 1000 m (Eshiet and Sheng 2013). We have calculated our K values based on Eq. (6) since it produces more realistic estimates.
Pore pressure within a formation can be determined using data from acoustic and resistivity well logs whereby the sonic transient time and formation resistivity are measured against depth. A formation pore pressure can also be simply calculated from the hydrostatic pressure gradient which shows a linear increase in hydrostatic pressure with depth. However, this does not account for deviations from the normal trend line or the normal compaction trend due to disparities in rock properties, for instance, in areas of abnormal compaction, porosity and fluid movement. An overpressure condition can be easily generated in locations of high density and decreased porosity. Whereas the sonic transient time decreases linearly with depth due to reduced porosity, resistivity is shown to increase nonlinearly with depth. This trend was established by Hottmann and Johnson (1965). The divergence of the measured sonic transient time and resistivity from those observed from normal compaction trends in hydrostatic conditions is used as an indicator of the abnormal fluid pressure in the area. This is based on the premise that deviations from the normal pore pressure in an area are caused by changes in the petrophysical properties such as porosity, density and fluid flow (Azadpour et al. 2015).
There are several methods of estimating porepressurebased empirical derivations and petrophysical properties, for example Eaton’s, Bower’s, and Miller’s compressibility and resistivity methods, and the Tau model (Azadpour et al. 2015; Zhang 2011). With an exception of the compressibility and resistivity methods which use the rock compressibility and resistivity, respectively, to calculate pore pressure, other techniques are based on compressional velocity and sonic transit time obtained from well logs. Eaton’s method is presently the most widely adopted technique and is based on empirical derivations using sonic transit times.
The Biot effective stress coefficient defines the change in the bulk volume of a material as the pore pressure fluctuates and may be determined by means of several empirical relations. For rock, this coefficient generally increases with porosity and is shown to have values up to 0.9 for rock porosities of approximately 0.18 (Alam et al. 2010; Luo et al. 2015). With respect to this study, the formation being considered is predominantly sandstone with an average porosity of approximately 0.26. Hence, Biot’s effective stress coefficient with an estimated value of 1.0 is assumed.
By adopting Sheorey’s formulation (Eq. 6), we have assumed a vertically and transversely isotropic rock formation. This assumption is extended to the initial stress condition, which—for a simplified case—is also taken to be transversely isotropic implying insignificant disparities between the maximum and minimum horizontal stresses (\(\sigma_{\text{h}} \approx \sigma_{\text{H}}\)). While the vertical stress is computed from an integration of the weight of the overburden determined from the densities of water and the various rock types, the horizontal stresses are estimated using the horizontaltovertical stress ratio (Eq. 6) and are functions of the elastic modulus and depth. Profiles of the initial in situ stress distributions are shown in Figs. 5 and 6.
2.3 Formation rock properties
Based on the range of typical values for specific gravity (e.g. 2.2–2.8 for sandstone and 2.4–2.8 for shale), 2.5 was taken to be a representative average. Mean values of other properties including Poisson ratio \(\upsilon\), elastic modulus \(E\), and permeability \(k\), are given in Table 1. The elastic modulus should generally increase with depth (Moayed and Bolandi 2012) which amongst other factors may be attributed to the increase in consolidation (Moayed et al. 2012); nevertheless, because of the short interval under consideration we have used consistent initial values for each rock type.
Pore pressure along the rock segment ranges from 29.7 MPa at 3000 m to 31.3 MPa at 3183 m, giving an average value of 30.4 MPa. Ideally to ensure equilibrium and to deter fluid flow into the wellbore, the applied mud pressure should, at least, match the maximum pore pressure within the reservoir. Once a well bore is drilled, a pore pressure gradient is naturally established with the lowest magnitude occurring at the wellbore. This phenomenon is essential for enabling fluid flow towards the well. Hence, mud pressure is used to control the pressure gradient and fluid flow. It is also used to maintain well stability by preventing well collapse due to excessive shear and compressive stresses at the periphery of drilled cavities. The magnitude of mud pressure applied is therefore subject to many factors. Excessive high mud pressure may lead to tensile failure and loss of fluid during circulation. On the other hand, insufficient mud pressures may instigate compressive failure and wellbore breakouts. Mud pressures that are too low may not be sufficient to prevent uncontrollable inward flow and well collapse. A pressure gradient was established by setting the pore pressure at the wellbore surface to 23.95 MPa in order to initiate fluid flow.
3 Modelling methodology
The mud pressure is the principal parameter to be investigated due to its role in well stability. The determination of a window that provides a safe range of mud pressures that can be applied without compromising the integrity of the wellbore during drilling is the underlying purpose of this work. This is accomplished in two main stages: deterministic and stochastic analyses.
3.1 Deterministic analysis
This method is used to define an initial range of safe mud pressures. The safe mud pressure window is restricted to the specific string casing length being considered, which implies that a repeated analysis should be performed for each successive interval of depth. Also, as previously mentioned, the deterministic method largely relies upon accurate measurements of the geomechanical conditions around the well and cannot account for uncertainties under this setting.
The deterministic analysis was conducted by finite element numerical method (using Abaqus 6.10) and the radial strain taken as the response parameter. A depth approximately 3000 m below sea level for an interval spanning 183 m was adopted as the target area. It is assumed that the magnitude of safe mud pressures increases progressively with depth. The mud pressure and radial strain were taken as input and output parameters, respectively. With these, a response curve is derived at the end of each set of simulations. The applied mud pressure was varied between 0 and 60 MPa, with each value plotted against the maximum radial strain at the onset of failure. Failure of wellbores occurs in two main modes: compressive and tensile. Compressive failure is attributed to wellbore breakouts which happen when the wellbore stress exceeds the rock compressive strength. This is often mitigated by increasing the mud pressure/weight to counterbalance and decrease the compressive stresses at the wellbore vicinity. Tensile failure occurs when the excessive mud pressure increases tensile stresses to magnitudes exceeding the rock tensile strength. An indication of tensile failure is the initiation and propagation of fractures. The magnitude of the maximum radial strain is, thus, matched against the corresponding exerted mud pressure, and the region where neither breakout/collapse, nor fracture occurs is delineated as the stable region. The compressive and tensile failure criterion is governed by the elastic theory of deformation of materials whereby failure is deemed to have occurred when the material yield strength is surpassed. The result is therefore conservative as the postyield behaviour of the material is not considered.
3.2 Stochastic analysis
The stochastic analysis is carried out to verify the reliability of results from the deterministic study and where necessary redefine it for better accuracy. Additional variables are introduced that allow for risk assessment and optimisation of the results. Variability in the domain characteristics such as the rock material properties and behaviour presents uncertainties in resulting outputs. An iterative procedure ensures that various scenarios or combination of scenarios are accounted for. Repetitive calculations that entail the variation of different combinations of input parameters produce corresponding outcomes that depict the state of the wellbore for a given set of initial and boundary conditions. This heuristic approach is common in optimisation techniques, but is essential in determining required probability and cumulative density distributions.
Stochastic methods are statistical approaches for determining probabilities of specific outcomes. The main object of stochastic analysis as applied in this study was geared towards defining the safe mud pressure window under a given set of conditions. To achieve this it is mandatory to predict the probability of obtaining a predefined outcome for given mud pressures. The Monte Carlo sampling method remains the most widely used stochastic technique. The simple Monte Carlo method requires a large sample size resulting in greater computational cost. Hence, in its simple form, it may not be suitable where there are constraints in the extent of the sample domain and computational resources.
The stochastic analysis was performed using the optimisation software, HyperStudy, by linking it with the finite element solver. Thus, by altering the study setup within the HyperStudy domain, the finite element solver is repeatedly fed different sets of input parameter values. For this work we focused on the spatially and temporally changing rock properties at the proximity of the wellbore. Amongst these properties Poisson’s ratio was identified as a parameter that may have greater inconsistency because of uncertainties in its estimation. The Poisson ratio plays a major role in rock deformation and impacts on stress distributions and, in general, wellbore stability. Whereas in the deterministic analysis only mud pressure is varied, for the stochastic study Poisson’s ratio, void ratio, elastic modulus (which are inputs representing the material property design variables) and mud pressure are altered in a manner predefined by an assigned statistical distribution pattern, in this case the normal distribution. HyperStudy generates samples through the Hammersley algorithm. A sample size of up to 200 was produced and each passed on to the finite element solver at every run.
Many variances of the Monte Carlo method which require much reduced sample sizes are available such as Latin hypercube sampling, orthogonal array sampling, adaptive importance sampling and generalised antithetic sampling. QuasiMonte Carlo methods provide good or even better alternatives to random sampling methods (i.e. Monte Carlo sampling). These are also known as lowdiscrepancy sequences. Though quasiMonte Carlo integration functions in a similar manner to Monte Carlo integration, it uses sequences of quasirandom numbers for the numerical integration. They reduce clustering during sampling, resulting in a wider spread, and also enable an accelerated convergence rate (Caflisch 1998). In order to take advantage of these features, Hammersley sampling (Hammersley and Hanscomb 1964), which is one of such methods, was employed in this study.
The domain of the design variables are characterised by a normal distribution, typically comprising mean values, standard deviations and variances. For stochastic analyses it was necessary, in some cases, to modify the variances to ensure the desired dispersion is maintained which should ideally spread between lower and upper bound values. These may sometimes require an amendment of the mean value.
3.2.1 Study setup
3.2.1.1 Parameterised file model
The Solver input file created by Abaqus was parameterised to obtain a HyperStudy template file with an ASCII text format. Details of the model are included in template statements that enable the replacement of data fields with parameters. The use of parameters permits the automatic alteration of design variables within predefined bounds. The solver inputparameterised file precludes the need for importing the complete Abaqus model environment (.cae file).
3.2.1.2 Design variables
To define the design variables, the following were specified: the initial, lower bound and upper bound values, the data type, data continuity and distribution, and distribution properties. A continuous rather than a discrete dispersion of data was used, and the normal distribution was used to characterise the statistical scatter of each design variable.
3.2.1.3 Responses
Responses were defined with respect to the most important output variables required for observation. Usually, values of the output variables are subsequently fed into the main study approaches (e.g. DOEs and stochastic analyses).
3.2.2 Description of study approaches
Two interrelated but independent categories of analysis will suffice for this investigation: design of experiment (DOE) and stochastic analysis. The scope of this study is limited to stochastic analyses.
3.2.2.1 Design of experiment
3.2.2.2 Design variables
One operation parameter and three controlled design parameters were selected as input variables: mud pressure, elastic modulus, Poisson ratio and void ratio. Mud pressure is a well operation parameter that represents the mud weight commonly applied to maintain well integrity during drilling and extraction process. It is also referred to as the bottomhole pressure. The magnitude and gradient of this pressure must exceed the formation pressure gradient to avert the inflow of formation fluid to the wellbore and well breakout; however, excessive mud pressure increases the potential for tensile fracturing and fluid loss. The elastic modulus and Poisson ratio are deformation parameters that dominate control of the rock strain characteristics, particularly at the elastic range. The void ratio is a measure of consistency and packing of the rock grains and can be used to estimate porosity, specific gravity, density and saturation.
Ideal bounds and statistical distribution of design variables
Design parameter  Statistical properties  

Lower bound  Upper bound  Initial value  Mean  Standard deviation \(\sigma_{{\bar{x}}}\)  Variance \((\sigma_{{\bar{x}}} )^{2}\)  
Input variable  
Elastic modulus E  1.18e+9 Pa  4.52e+10 Pa  2.32e+10 Pa  2.32e+10 Pa  2.32e+9 Pa  5.38e+18 Pa^{2} 
Poisson ratio \(\upsilon\)  0.1  0.35  0.225  0.225  2.25e−2  5.06e−4 
Void ratio e  0.0526  0.639  0.346  0.346  0.0346  0.001197 
Operation parameter  
Mud pressure P_{M}, MPa  0.0  60.0  0.0 
3.2.2.3 Responses/output
List of response/output factors
Stress  Strain  Displacement 

von Mises stress  Vertical strain  Vertical displacement 
Vertical stress  Radial strain  Radial displacement 
Radial horizontal stress  Tangential strain  Tangential displacement 
Tangential horizontal stress  Shear strain (1–2 plane)  
Shear stress (1–2 plane)  Shear strain (1–3 plane)  
Shear stress (1–3 plane)  Shear strain (2–3 plane)  
Shear stress (2–3 plane)  Max. principal strain  
Tresca stress  Mid. principal strain  
Third invariant deviatoric stress  Min. principal strain  
Hydrostatic pressure 
The state of a wellbore can be checked by evaluating the stress, strain and deformation conditions. These can be applied in determining criteria for wellbore failure. Wellbore failure is often described in two modes: compressive and tensile failure (Sheng et al. 2006). Compressive failure happens where the compressive strength of the rock is exceeded by the wellbore stresses resulting in well breakout. Likewise, tensile failure occurs where the rock tensile strength is exceeded causing hydraulic fracturing and loss of circulation fluid. Because wellbore stability is directly dependent on the extent at which compressive or tensile deformation has occurred, it is more straightforward to adopt a strain or deformation criterion as a measure of the wellbore failure. For this study critical radial compressive and tensile strain values were used to ascertain the advent of rock failure and wellbore instability.
4 Results and discussion
It is imperative that wellbore instability be considered as an integral factor during drilling and other well operations. These instabilities are attributed to mechanical and/or chemical effects (Pašić et al. 2007). Mechanical effects may be caused by, for instance, lack of caution during drilling, excessive stresses around the wellbore or weak formation rock. On the other hand, chemical effects arise due to the often complex interactions between the formation rock, formation fluid and drilling fluids. The combined impact of both mechanical and chemical effects is frequently manifested in field conditions. The conventional approach to ensure that the rock surrounding the wellbore during drilling or production remains intact is by the radial application of mud pressure using fluids with specialised properties. Knowing the correct magnitude of mud pressure (also referred to as mud weight since the pressure is a function of its density) to exert is crucial in order not to instigate instabilities that may lead to wellbore failure. A deterministic approach can be used to mark the limits beyond which well failure would occur. Theses limits are defined in terms of the range of safe mud pressures, implying an upper and lower bound. The upper bound represents the highest mud pressure value. Pressures above this magnitude result in wellbore failure or jeopardise its stability. Similarly, the lower bound represents the lowest mud pressure value below which wellbore failure will occur. The actual window of safe mud pressure is case specific and highly dependent on the type of rock encountered in the reservoir, the drilling practice and fluid flow in the reservoir.
In several instances the lower bound is set as the minimum allowable mud pressure to counterbalance compressive stresses that lead to compressive/shear failure of the wellbore; this is referred to as well breakout. For permeable formations, the minimum allowable mud pressure should also prevent an inflow of the reservoir fluid. For this to be achieved, the minimum allowable mud pressure must be greater than the formation pore pressure. At the opposite end, the maximum allowable mud pressure is defined as the highest magnitude of mud weight that can be applied without causing tensile failure, loss in circulation or hydraulic fracturing of the formation (Hilgedick 2012; Moos et al. 2003). The above conditions are likely to apply where the pore pressure gradient is low or normal such that a considerably low mud pressure is sufficient to restrict the influx of reservoir fluids. A low or normal pore pressure gradient also invariably suggests that the pore fluid velocity at the vicinity of the wellbore face is low. If the pore pressure gradient is steep, the associated pore fluid velocity near the wellbore will be high. Sufficiently high pore pressures can generate effective tensile stresses causing tensile failure where the rock strength is exceeded. This has been observed by French et al. (2012) and Secor (1965), where natural hydraulic fractures and dilation are reported to occur when the pore pressure surpasses the least in situ compressive stress by a magnitude equivalent to the rock tensile strength. Dilation and fracture were also shown to take place in response to a high strain rate. Likewise, high pore fluid velocities create tensile stresses that may cause tensile failure if the rock tensile strength is exceeded. Where there is a decline in permeability or at very high flow rates, the increased drag forces cause the effective radial stresses to become negative, leading to tensile failure (Eshiet and Sheng 2013; Morita et al. 1998; Nouri et al. 2002).
Hence, the in situ pore pressure or pore pressure gradient influences the reservoir characteristics, especially near the wellbore region, and plays a dominant role in determining the regime of stresses generated. This weighty effect of the prevailing pore pressure condition implies that at high in situ pore pressures there is likely going to be a reverse in the impact of mud pressure applied on the wellbore wall. Where the mud pressure is too low to counteract the increased flow rate caused by high pore pressure gradients, the corresponding large drag forces will instigate rock failure in tension. At the other extreme of the spectrum, under parallel pore pressure conditions, if the mud pressure is too high, viscous and, in some instances, applied rapidly, the excess over the compressive forces which counterbalances the in situ pore pressure at the wellbore wall will cause rock failure in compression or shear when the rock compressive strength is exceeded.
Imposed drawdown conditions are shown to generate tensile forces due to high fluid flows in the vicinity of the wellbore. It is particularly observed where the rock material permeability is significantly low or the flow rate is very high resulting in large drag forces and negative effective radial stress. This phenomenon is consistent with Nouri et al. (2002) and Morita et al. (1998). Consequently, tensile stresses and strains will be generated at low mud weights which are insufficient to counteract the impact of influxes. On the other hand, compressive stresses and strains are generated if mud weights are excessively applied. The mud weight should be regulated such that it is not too low thereby allowing tensile failure, nor high enough to cause compressive failure.
This investigation is performed on a reservoir formation subjected to high in situ pore pressure and pore pressure gradient. The pore pressure at the wellbore face is 23.95 MPa with an initial overburden pore pressure gradient of 0.01 MPa/m and initial lateral pore pressure gradient of 2.13 MPa/m. The lateral pore pressure gradient is indicative of the reservoir drawdown. Both vertical and horizontal pore pressure gradients are considerably large (see Zhang 2011); the stress and strains generated are thus expected to be in accordance with field conditions under high naturally occurring pore pressure.
Establishing criteria for compressive failure and tensile failure of the wellbore is complicated because of the complexity of reservoir formations. The state of the rock is lithology dependent, and there are several factors that determine formation rock behaviour. Usually, tensile and compressive rock failure takes place when the respective rock tensile and compressive strength is exceeded by prevalent stresses. In wellbore stability analyses, initial rock failure does not necessarily jeopardise the stability of the wellbore, at least in the short term. Eventually, the onset of rock failure initiates a progression of mechanisms culminating in a critical state where an unstable condition is reached. An excellent measure of this is the critical radial inward and outward deformation or the critical radial tensile and compressive strain. Their actual values can be set as thresholds of the extent of strain that is tolerable, which is a function of the consequences in relation to the well stability. This may be case and site specific, varying with each drilling system and field condition. As a consequence of the primary focus of this research, an array of predefined critical tensile and compressive strains was tested. They were grouped in the following matching pairs: ±4.0e−4, ±6.0e−4, ±7.0e−4, ±8.0e−4, ±9.0e−4, and ±1.0e−3, where the positive and negative signs denote tension and compression, respectively.
There are several criteria that could be used to determine wellbore instability. Some of these are linked to established rock strength or failure criteria, such as Mohr–Coulomb, Mogi–Coulomb, Drucker–Prager and Modified Lade, where the main parameters are stressrelated variables. The displacement or deformation of the wellbore is an alternative measure of its instability (e.g. Sheng et al. 2006). This focuses on the extent of wellbore deformation without recourse to the rock strength failure criteria. Adopting this form of criterion, radial strain is arbitrarily chosen, in this case, as an indicator of wellbore performance. This is also made to be consistent with the direction of application of mud pressure, since it is basically applied in the radial direction. In a similar manner, other displacement or strain parameters orientated tangentially or vertically to the wellbore may be suitable for this purpose.
In reality, rock tensile strength is significantly lower than its compressive strength. Nonetheless, in this analysis, hypothetical criterion values are used and the relative magnitudes of the pair were found to be irrelevant provided they remain consistent throughout each set of calculations. The emphasis, primarily, is to provide an accurate qualitative description of the trend.
4.1 Deterministic study
The isotropic state of the horizontal stresses represents the in situ stress condition of the target formation. This equilibrium condition is affected once the wellbore is drilled, leading to a redistribution of the horizontal stresses. In cylindrical coordinates, these are represented by radial and tangential stresses. With respect to the wellbore axis, radial stresses act in both inward and outward directions resulting in corresponding compressive and tensile radial strains. The mud pressure is merely meant to act as a counterbalance. It is worth noting that stress measurements are not taken at a single point, but throughout the target segment of the wellbore.
A further increase in strain criterion to ±1.2e−3 (Fig. 10e) establishes a safe mud window delineated by a lower bound of 26.0 MPa and an upper bound of 35.5 MPa. While the compressive integrity is still maintained, tensile failure and instability occur when the mud pressure declines below the lower limit. At the other end of the spectrum, compressive failure and instability occur at mud pressures above 35.5 MPa, whereas, under equivalent conditions, the maximum tensile strains are considered insignificant. It becomes immediately obvious that for a radial strain criterion greater than ±1.0e−3, a margin of safe mud pressure can be clearly delineated (Fig. 10d, f). Hence, if a radial tensile strain criterion of +1.0e−3 is combined with a radial compressive strain criterion of −1.1e−3, it is then possible to define a safe mud window, denoted by the base of the curve (Fig. 10f).
4.2 Stochastic study
Stochastic techniques promise to be a more rigorous approach in dealing with uncertainties in design and implementation. This can be manifested in the form of risks which are quantifiable. The extent to which risks are accurately assessed depends on the complexity of the process and the interchanging factors. Risks of drilling, completion and production of wells are mainly associated with wellbore instability. These forms of instability are defined in various ways. In this context, an unplanned fracturing of the reservoir by excessive pressure from injected fluid leading to loss in circulation is described as an unstable condition; well collapse or convergence caused by high compressive and shear stresses around the wellbore is defined as an unstable condition; also, reservoir erosion due to the weakening and detachment of rock is another unstable condition.
The causes of wellbore instability are undeniably attributed to drilling, completion and production practices. Nonetheless, the mechanical properties, physical properties and the structural constituents of the formation have an equal weighting. In this regard, the stochastic analysis employed in this study does not only emphasise the aspects of the operation during drilling, completion and operation, it also focuses, to a large extent, on the rock properties and their variations. Rock formations are naturally mostly heterogeneous. This heterogeneity is often difficult to estimate the wide range of incidences and the randomness at which they occur; however, their probability of occurrences may be encompassed in a stochastic procedure linking the likelihood of the existence of a combination of a set of controlling parameters on the performance of the wellbore. Factors affecting wellbore stability can be categorised as the in situ stress field, rock properties, pore pressures and mud pressure (Chen et al. 1996). An examination of the mechanisms of these factors indicates that changes in their magnitude, orientation and distribution are likely to have a profound impact on the stability of wellbores.
For this stochastic analysis the in situ stress field is incorporated in terms of the magnitude and orientation, but precluding residual stresses, thermal stresses and the history of tectonic events. The stress field is assumed to be hydrostatic and lithostatic with an extensional regime as described in Anderson (1951) and Eckert and Liu (2014), where \(\sigma_{\text{v}} > \sigma_{\text{H}} > \sigma_{\text{h}}\). A vertical borehole is used for the analysis implying drilling in the downward direction with zero deviation angle. The elastic modulus and Poisson ratio as elastic deformation parameters are used to represent the rock mechanical properties since the well integrity is significantly influenced by the allowable strain. For the physical property, the void ratio is used, as it characterises the compactness of the rock. A pore pressure gradient of 10 kPa/m is applied, measured as the hydrostatic pressure from sea level to the depth of the subsurface, which falls within the range for normal pressure conditions. Drilling and completion operations are restricted to mud pressure conditions applied in various ways as to represent balanced, underbalanced and overbalanced drillings.
The input parameters used to evaluate wellbore stability are derived from data that tend to be inconsistent. The uncertainty in data is thus manifested in the results of predicted safe mud pressures. The variability of each input parameter should be accounted for in a manner that properly represents their uncertainties. This can be accomplished by employing quantitative risk analysis (QRA), where, as applied by Moos et al. (2003) and McLellan and Hawkes (1998), cumulative distribution density (CDF) and probability density function (PDF) curves are used to measure uncertainties in input parameters. Where actual/real data are accessible, CDF and PDF curves that more accurately portray disparities in values of input parameters are used. In the absence of reliable data, the values of input parameters are varied between upper and lower bounds with the PDF defined by an appropriate distribution function. Examples of continuous distribution supported on bounded intervals are uniform distributions, truncated normal distributions, logitnormal distributions, logarithmic distributions and triangular distributions. An example of a continuous distribution supported on semiinfinite intervals is the lognormal distribution, and an example of a continuous distribution that takes values over the whole “real line” is the generalised normal (generalised Gaussian) distribution. At one end, if there is absolutely no indication of the likely values, the uniform distribution is adopted because of the high degree of caution required. At the other end, it is pertinent to use the triangular distribution where a specific value has been identified as most likely to occur.
Statistical data for samples of design variables
Statistical data  Elastic modulus (Pa)  Poisson ratio  Void ratio 

Points  50  50  50 
Minimum  1.18e+9  0.101  0.051 
Mean  2.32e+10  0.221  0.339 
Maximum  4.52e+10  0.333  0.581 
Range  4.40e+10  0.232  0.530 
Average deviation  8.11e+9  0.0438  0.0986 
Standard deviation  1.01e+10  0.0545  0.122 
Standard deviation/mean  0.434 (dimensionless)  0.246  0.360 
Variance  1.01e+20  0.00297  0.0149 
Skewness  −5.97e−4  −0.0797  −0.165 
Kurtosis  −0.421 (dimensionless)  −0.441  −0.467 
Min. bound 99%  1.18e+9  0.101  0.0510 
Min. bound 95%  4.44e+9  0.118  0.112 
Min. bound 90%  6.51e+9  0.128  0.130 
Max. bound 99%  3.99e+10  0.314  0.536 
Max. bound 95%  4.20e+10  0.322  0.549 
Max. bound 90%  4.52e+10  0.332  0.580 
The shape of the reliability plots for the design variables is somewhat similar to the sshaped curve typically described by the sigmoid function (a form of the logistic function). This type of curve is usually created through logistic regression models. The major difference between the reliability plots and the sshaped curve is the negative relationship between reliability and rock properties. As the rock property increases, the probability of a random choice of value being greater than a given magnitude decreases (Fig. 12). This is consistent for all three cases.
 1.
the distribution of responses due to the application of mud pressures (P_{M} = 15 MPa) significantly lower than the pore pressure at the vicinity of the wellbore (P_{P} = 23.95 MPa), \(P_{\text{M}} \ll P_{\text{P}}\) (Fig. 16);
 2.
the distribution of responses due to the application of mud pressures (P_{M} = 23.95 MPa) equivalent to the pore pressure at the vicinity of the wellbore (P_{P} = 23.95 MPa), \(P_{\text{M}} = P_{\text{P}}\) (Fig. 17); and
 3.
the distribution of responses due to the application of mud pressures (P_{M} = 50 MPa) significantly higher than the pore pressure in the vicinity of the wellbore (P_{P} = 23.95 MPa), \(P_{\text{M}} \gg P_{\text{P}}\) (Fig. 18).
Statistical data for wellbore response: maximum radial tensile strain
Statistical data  Mud pressure P_{M}  

15 MPa  23.95 MPa  50 MPa  
Points  50  50  50 
Minimum  3.24e−4  1.35e−4  5.07e−5 
Mean  8.31e−4  3.57e−4  1.34e−4 
Maximum  8.03e−3  3.41e−3  1.31e−3 
Range  7.70e−3  3.27e−3  1.26e−3 
Average deviation  4.29e−4  1.81e−4  7.30e−5 
Standard deviation  1.09e−3  4.60e−4  1.79e−4 
Standard deviation/mean  1.308  1.289  1.334 
Variance  1.18e−6  2.12e−7  3.20e−8 
Skewness  6.184  6.208  6.052 
Kurtosis  41.158  41.435  39.781 
Min. bound 99%  3.24e−4  135e−4  5.07e−5 
Min. bound 95%  3.32e−4  1.42e−4  5.45e−5 
Min. bound 90%  3.70e−4  1.61e−4  5.57e−5 
Max. bound 99%  1.58e−3  6.72e−4  2.48e−4 
Max. bound 95%  2.05e−3  8.45e−4  3.65e−4 
Max. bound 90%  8.02e−3  3.40e−3  1.31e−3 
Statistical data for wellbore response: maximum radial compressive strain
Statistical data  Mud pressure P_{M}  

15 MPa  23.95 MPa  50 MPa  
Points  50  50  50 
Minimum  −2.47e−3  −2.52e−3  −1.05e−2 
Mean  −2.36e−4  −2.42e−4  −1.04e−3 
Maximum  −8.14e−5  −8.66e−5  −3.97e−4 
Range  2.38e−3  2.43e−3  1.01e−2 
Average deviation  1.37e−4  1.40e−4  5.63e−4 
Standard deviation  3.39e−4  3.47e−4  1.42e−3 
Standard deviation/mean  1.440  1.430  1.374 
Variance  1.15e−7  1.20e−7  2.02e−6 
Skewness  −6.072  −6.080  −6.178 
Kurtosis  39.927  40.007  41.066 
Min. bound 99%  −2.47e−3  −2.52e−3  −1.05e−2 
Min. bound 95%  −6.87e−4  −7.00e−4  −2.69e−3 
Min. bound 90%  −4.46e−4  −4.57e−4  −1.97e−3 
Max. bound 99%  −9.09e−5  −9.39e−5  −4.27e−4 
Max. bound 95%  −9.09e−5  −9.15e−5  −4.27e−4 
Max. bound 90%  −8.38e−5  −8.90e−5  −4.07e−4 
In contrast to the defined trend in tensile strain, the compressive strain increases with increases in mud pressure (Fig. 21). There is a remarkable difference in the progression in compressive strain between the lower and upper mud weight regimes with the rate of increase being considerably higher at higher mud pressures. Prior to the attainment of a mud pressure of 35 MPa, the lower bound compressive strain increases at a rate of 5.8e−7/MPa, then escalates sharply to a steep rate of 2.03e−5/MPa beyond this value (Fig. 21a). The same pattern occurs at the other end of the scale (the upper bound) (Fig. 21b), where the compressive strain increases at 6.4e−6 below 35 MPa and then accelerates to 5.25e−4 above 35 MPa. Consequently, the attenuation in tensile strain is followed by synchronised increments in compressive strain.
The stability of a wellbore can be described in terms of the tendency of failure of the rock surrounding the well. Where the rock around the wellbore fails, this immediately compromises the stability of the wellbore. Rock failure can be quantified in various ways. This include, for instance, the extent of brittle failure, the onset of a given strain criterion, and the proportion of rock and depth of area that reach the yield criterion. Because rock failure models are necessary for predicting rock behaviour, they are instrumental in determining wellbore stability conditions.
The choice of rock failure models defines the magnitude of minimum and maximum allowable mud pressures. For example, applying the Mogi–Coulomb rock failure model, the minimum allowable mud pressure is much less compared to when either the Mohr–Coulomb failure model or the Hoek–Brown failure model is adopted (Elyasi and Goshtasbi 2015). Also, predicted mud windows are narrower and conservative with the Hoek–Brown model (Elyasi and Goshtasbi 2015). The failure criterion selected therefore has a prominent influence on the defined safe mud window. Due to its simplicity, the linearelastic analysis is often used to predict the initiation of failure. In terms of stress, the onset of tensile or compressive failure happens when the respective tensile and compressive rock strength is exceeded. Whereas the criterion for tensile failure is defined by when the minimum effective stress is greater than the rock tensile strength, the criterion for compressive failure is determined by whichever compressive failure criterion is deemed appropriate (McLean and Addis 1990) (e.g. Mohr–Coulomb, Drucker–Prager and Hoek–Brown).
To preclude the dependency on stress evaluations, the use of strain parameters may be used to assess the condition of the wellbore by determining both the critical compressive strain where there is a high risk of wellbore collapse and the critical tensile strain where there is a high risk of hydraulic fracture. As previously mentioned, these are the criteria adopted in this study reflected in terms of the critical radial tensile strain and critical radial compressive strain which are also representative of the critical radial inward and outward deformations, respectively.
Safe mud windows at different confidence levels and stability criteria
Confidence level, %  Mud window P_{W}, MPa  

Stability criterion ±6.0e−4  Stability criterion ±8.0e−4  Stability criterion ±1.0e−3  
95  24–36  23–38  21–40.7 
90  22.5–38  20–40.5  18–43 
70  18–43  14–47  9.3–52 
Deterministic predictions of wellbore instability are associated with predefined constant values of rock properties and operating parameters. This concept is underpinned by the assumption of consistency in the behaviour of the wellbore system matched by different operating conditions. This presupposes that the stability of the well system can be ascertained based on advance knowledge of the wellbore/rock behaviour derived from established rock mechanical and failure models. Deterministic methods are founded on the principle of causality, wherein outcomes are entirely defined by a chain of relationships between cause and effect. Deterministic systems are therefore predictable (Kirchsteiger 1999) and consistent. This type of approach when applied to predictions and analyses of wellbore instability produces a set of welldefined responses under various conditions. Probabilistic methods involve the integration of uncertainties and randomness (Kirchsteiger 1999). The extent of these two components is largely dependent on the heterogeneity and inconsistency of material and prevailing conditions.
Deterministic predictions of safe mud weights
Stability criterion  Safe mud weight bound limits, MPa  Remarks  

Lower bound (tensile limit)  Upper bound (compressive limit)  
±4.0e−4  –  –  Total failure 
±6.0e−4  29.5  –  Total compressive failure 
±8.0e−4  28.0  –  Total compressive failure 
±1.0e−3  27.0  ≈ 35.0  Defined mud window 
±1.2e−3  26.0  35.5  Defined mud window 
Parallel comparisons between deterministic and stochastic predictions can hence be made following both outcomes. Using a stability criterion of ±1.0e−3, a mud window of 27.0–35.0 MPa is recommended through the deterministic method, whereas the stochastic predictions provide much broader margins even at high confidence levels (Table 7). At 95% level of certainty, the mud window is 21–41 MPa, which still accommodates more values of mud weight. This margin is further increased at lower levels of certainty and is shown to extend to as much as 9.3–52 MPa at 70% level of certainty (Table 7). In other words, the lower and upper limits are extended. Where lower stability criteria are adopted (≤ ±1.0e−3), results from deterministic estimates imply adverse conditions unsuitable for wellbore drilling and/or production.
Deterministic evaluations of wellbore stability are therefore conservative as they estimate much smaller ranges of mud weights that can be safely applied during wellbore operations. The conservative approximations are prompted by various factors: firstly, the dependency of deterministic models on chosen characteristic models that are formulated to mimic rock behaviour. These models are intrinsically built on the premise of continuum theories and only able to account for discontinuities to a limited extent upon explicit modifications; for instance, a linear deterministic prognosis is underpinned by an assumption of a linear behaviour of the rock formation. Secondly, deterministic approaches presuppose utter homogeneity of the system and even where heterogeneity is considered it is simplified, structured and predefined. Thirdly, this approach precludes uncertainty. It assumes precise knowledge of the in situ conditions, rock properties and rock behaviour (Moos et al. 2003). It also ignores the inevitable occurrence of errors and lack of information because of incomplete data. Most natural systems are variable and subject to temporal and spatial changes. This is typically reflected in underground rock formations. The divergence of rock properties therein makes such systems prone to substantial levels of uncertainties. Under the deterministic approach there is a “head or tail” kind of distinction in the status of the wellbore. The system is either safe or unsafe, eliminating any potential for risks. The margin of safety is invariably reduced where risks are to be avoided.
The probabilistic approach, on the other hand, recognises the existence of inherent uncertainties arising due to factors such as (Bulleit 2008) material heterogeneity, time, human error, statistical limits, restrictions in models and randomness. Uncertainty is considered the norm rather than the exception (Kirchsteiger 1999), and even when the level of statistical certainty is set to as high as 0.95, there is still a large degree of flexibility that stretches the limits of the band of safe mud pressures. The inconsistency in the design parameters implemented via variable values of input rock properties—elastic modulus, Poisson ratio and void ratio—permits the realisation of a broader range of safe operating conditions visàvis where consistent and/or uniform values of rock properties are employed.
5 Conclusions
A deterministic method has been applied in this study to assess the performance of wellbores when subjected to changing conditions in order to identify settings where the structural status of the wellbore can be declared as either stable or unstable. By adopting this procedure, a safe mud pressure window can be established which represents a range of applied mud weights that will not degrade the stability of the wellbore. To optimise this process, stochastic techniques which fully integrate fluctuations associated with randomness and inconsistencies in in situ conditions and rock properties have been invoked. Emphasis was given in particular to the impact of the variability of rock material properties to the reliability of the wellbore. The key design parameters considered are elastic modulus, Poisson ratio and void ratio.
 1.
The prevalence of any type of stress around the wellbore depends on the stresses generated by the applied mud pressure in conjunction with those generated by drawdown. High drawdown conditions produce tensile stresses induced by large flow drag forces. These are counterbalanced by compressive stresses generated when the excessive mud pressure is applied at the wellbore face.
 2.
The quasirandom Monte Carlo sampling—implemented via the Hammersley method—provides low discrepancies which are ascribed to an algorithm that enhances uniformity and spread over the design space. This is evidenced by the even distribution of the input design variables.
 3.
For each mud weight applied, the magnitude of most generated compressive and tensile strains fall within the lower range of the strain scale suggesting a very low tendency for the generation of strains with values near the upper strain bound limits. The cumulative density function (CDF) towards the lower limit of the strain scale is generally above 0.95.
 4.
Because the mud weight counterbalances tensile stresses caused by drawdown conditions, the intensity of tensile strains decreases with increments in mud pressure and the rate of this reduction is significantly greater at high ranges of mud pressures. On the other hand, the compressive strain increases with mud pressure and the rate of increase in strain is considerably higher at high ranges of mud pressures. Accordingly, the decline in tensile strain is followed by a synchronised progression in compressive strain.
 5.
For a designated pair of compressive and tensile stability criteria, the probability of generated tensile strains exceeding the given tensile strain criterion reduces as the mud pressure is increased, while the probability of the compressive strains exceeding the specified compressive strain criterion rises as mud pressure is increased. However, the magnitude of the produced strains taken into account during stochastic analyses is lower in comparison with those from deterministic analyses, thus permitting a wider safe mud pressure window. The stochastic approach implicitly refines the definition of the compressive and tensile stability criteria by providing for uncertainties and variable geomechanical conditions.
 6.
As the mud weight increases, the frequency of incidences of unstable conditions triggered by excessive tensile strains reduces, whereas instances of unstable conditions initiated by excessive compressive strains increase. This phenomenon is reversed when the mud weight is reduced in that the number of occurrences of unstable conditions caused by excessive tensile strains increases while the population of instabilities instigated by excessive compressive strains decreases.
 7.
The size of the safe mud window is a function of the permissible confidence level or level of significance which indicates the degree of uncertainty. The size of this window is also dependent on the pair of stability criteria. The margin of a safe mud window is inversely proportional to the confidence level/level of significance suggesting that the higher the acceptable risk, the broader the margin. Likewise, wider margins are associated with increases in the threshold of stability criteria. Furthermore, predictions from deterministic models reveal that the size of safe mud windows increases with stability criteria.
 8.
Deterministic techniques do not account for risks or uncertainties. As such, there is a clear and twosided distinction between the statuses of the wellbore stability. The wellbore is either declared “safe” or “unsafe”. Stochastic techniques incorporate variations and uncertainties due to influencing factors including variabilities in design and operating parameters. The security of the wellbore is invariably linked with the degree of acceptable risks.
 9.
Estimates by deterministic models are conservative since the range of safe mud weights is considerably narrower when compared to predictions from stochastic analyses. By incorporating inconsistencies and risks stochastic models are able to broaden the margin of safe mud windows, thereby extending the range of mud pressures that can be employed during drilling and/or production.
References
 AlAjmi AM, AlHarthy MH. Probabilistic wellbore collapse analysis. J Pet Sci Eng. 2010;72(3–4):171–7. https://doi.org/10.1016/j.petrol.2010.09.001.CrossRefGoogle Scholar
 Alam MM, Borre MK, Fabricius IL, Hedegaard K, Røgen B, Hossain Z, et al. Biot’s coefficient as an indicator of strength and porosity reduction: calcareous sediments from Kerguelen Plateau. J Pet Sci Eng. 2010;70(3–4):282–97. https://doi.org/10.1016/j.petrol.2009.11.021.CrossRefGoogle Scholar
 AlKhayari MR, AlAjmi AM, AlWahaibi Y. Probabilistic approach in wellbore stability analysis during drilling. J Pet Eng. 2016; Article ID 3472158, 13 pages. http://dx.doi.org/10.1155/2016/3472158.
 Altair Engineering. Hyperworks 13.0 user guide: HyperStudy; 2014.Google Scholar
 Anderson EM. The dynamics of faulting and dyke formation with applications to Britain. 2nd ed. London: Oliver and Boyd; 1951.Google Scholar
 Aslannezhad M, Manshad AK, Jalalifar H. Determination of a safe mud window and analysis of wellbore stability to minimize drilling challenges and nonproductive time. J Pet Explor Prod Technol. 2016;6(3):493–503. https://doi.org/10.1007/s1320201501982.CrossRefGoogle Scholar
 Atkinson J. The mechanics of soils and foundations. 2nd ed. Abingdon: Taylor & Francis; 2007.Google Scholar
 Azadpour M, Manaman NS, KadkhodaieIlkhchi A, Sedghipour MR. Pore pressure prediction and modelling using welllogging data in one of the gas fields in south of Iran. J Pet Sci Eng. 2015;128:15–23. https://doi.org/10.1016/j.petrol.2015.02.022.CrossRefGoogle Scholar
 Biot MA. General theory of threedimensional consolidation. J Appl Phys. 1941;12(1):155–64. https://doi.org/10.1063/1.1712886.CrossRefGoogle Scholar
 Blocher G, Bruhn D, Zimmermann G, McDermott C, Huenges E. Investigation of the undrained poroelastic response of sandstones to confining pressure via laboratory experiment, numerical simulation and analytical calculation. Geol Soc Lond Spec Publ. 2007;284:71–87. https://doi.org/10.1144/SP284.6.CrossRefGoogle Scholar
 Bulleit WM. Uncertainty in structural engineering. Pract Period Struct Des Constr. 2008;13(1):24–30. https://doi.org/10.1061/(ASCE)10840680(2008)13:1(24).CrossRefGoogle Scholar
 Caflisch RE. Monte Carlo and quasiMonte Carlo methods. Acta Numer. 1998;7:1–49. https://doi.org/10.1017/S0962492900002804.CrossRefGoogle Scholar
 Chalupnik MJ, Wynn DC, Clarkson PJ. Approaches to mitigate the impact of uncertainty in development processes. In: International conference on engineering design, ICED’09 24–27 August. Stanford University, Stanford, CA, USA; 2009.Google Scholar
 Chen G, Chenevert ME, Sharma MM, Yu M. A study of wellbore stability in shales including poroelastic, chemical, and thermal effects. J Pet Sci Eng. 2003a;38(3–4):167–76. https://doi.org/10.1016/S09204105(03)000305.CrossRefGoogle Scholar
 Chen X, Tan CP. Wellbore behaviour in fractured rock masses. In: The 38th U.S. symposium on rock mechanics (USRMS), July 7–10, 2001, Washington, DC; 2001.Google Scholar
 Chen X, Tan CP, Detournay C. A study on wellbore stability in fractured rock masses with impact of mud infiltration. J Pet Sci Eng. 2003b;38(3–4):145–54. https://doi.org/10.1016/S09204105(03)000287.CrossRefGoogle Scholar
 Chen X, Tan CP, Haberfield CM. Wellbore stability analysis guidelines for practical well design. In: SPE Asia Pacific oil and gas conference, 28–31 October, Adelaide, Australia Adelaide, Australia; 1996. https://doi.org/10.2118/36972MS.
 Chen X, Tan CP, Haberfield CM. Guidelines for efficient wellbore stability analysis. Int J Rock Mech Min Sci. 1997;34(3–4):50.e1–14. https://doi.org/10.1016/S13651609(97)002669.Google Scholar
 Eckert A, Liu X. An improved method for numerically modelling the minimum horizontal stress magnitude in extensional stress regimes. Int J Rock Mech Min Sci. 2014;70:581–92. https://doi.org/10.1016/j.ijrmms.2014.04.020.Google Scholar
 Elyasi A, Goshtasbi K. Using different rock failure criteria in wellbore stability analysis. Geomech Energy Environ. 2015;2:15–21. https://doi.org/10.1016/j.gete.2015.04.001.CrossRefGoogle Scholar
 Eshiet KI, Sheng Y. Influence of rock failure behaviour on predictions in sand production problems. Environ Earth Sci. 2013;70(3):1339–65. https://doi.org/10.1007/s1266501322190.CrossRefGoogle Scholar
 Freeze RA, Cherry JA. Groundwater. Englewood Cliffs: PrenticeHall; 1979.Google Scholar
 FreijAyoub R, Tan C, Clennell B, Tohidi B, Yang J. A wellbore stability model for hydrate bearing sediments. J Pet Sci Eng. 2007;57(1–2):209–20. https://doi.org/10.1016/j.petrol.2005.10.011.CrossRefGoogle Scholar
 French ME, Boutt DF, Goodwin LB. Sample dilation and fracture in response to high pore fluid pressure and strain rate in quartzrich sandstone and siltstone. J Geophys Res. 2012;117:B03215. https://doi.org/10.1029/2011JB008707.CrossRefGoogle Scholar
 Gholami R, Rabiei M, Rasouli V, Aadnoy B, Fakhari N. Application of quantitative risk assessment in wellbore stability analysis. J Pet Sci Eng. 2015;135(185–200):185–200. https://doi.org/10.1016/j.petrol.2015.09.013.CrossRefGoogle Scholar
 Hammersley JM, Hanscomb DC. Monte Carlo methods. Monographs on applied probability and statistics. Netherlands: Springer; 1964.Google Scholar
 Hart DJ, Wang HF. Laboratory measurements of a complete set of poroelastic moduli for Berea sandstone and Indiana limestone. J Geophys Res. 1995;100(B9):17741–51. https://doi.org/10.1029/95JB01242.CrossRefGoogle Scholar
 Hilgedick SA. Investigation of wellbore stability in a North Sea field development. Doctoral Dissertation. Paper 2199, Missouri University of Science and Technology; 2012.Google Scholar
 Holzberg B. Quantification and treatment of uncertainties in wellbore stability analysis. MSc Dissertation, Pontifical Catholic University of Rio de Janeiro (PUCRio); 2001.Google Scholar
 Horsrud P, Sonstebo EF, Boe R. Mechanical and petrophysical properties of North Sea shales. Int J Rock Mech Min Sci. 1998;35(8):1009–20. https://doi.org/10.1016/S01489062(98)001624.CrossRefGoogle Scholar
 Hottmann CE, Johnson RK. Estimation of formation pressures from logderived shale properties. J Pet Technol. 1965;17(6):717–22. https://doi.org/10.2118/1110PA.CrossRefGoogle Scholar
 Howard RA. The foundations of decision analysis revisited. In: Edwards W, Miles Jr RF, Winterfeldt DV, editors. Advances in decision analysis: from foundations to applications. New York: Cambridge University Press; 2007.Google Scholar
 Jaky J. The coefficient of earth pressure at rest. J Soc Hung Eng Archit. 1944;78:355–8.Google Scholar
 Kirchsteiger C. On the use of probabilistic and deterministic methods in risk analysis. J Loss Prev Process Ind. 1999;12(5):399–419. https://doi.org/10.1016/S09504230(99)000121.CrossRefGoogle Scholar
 Liang QJ. Application of quantitative risk analysis to pore pressure and fracture gradient prediction. In: SPE annual technical conference and exhibition held in San Antonio, Texas, 29 September–2 October, San Antonio, Texas; 2002. https://doi.org/10.2118/77354MS.
 Luo X, Were P, Lui J, Hou Z. Estimation of Biot’s effective stress coefficient from well logs. Environ Earth Sci. 2015;73(11):7019–28. https://doi.org/10.1007/s1266501542198.CrossRefGoogle Scholar
 Ma T, Chen P, Yang C, Zhao J. Wellbore stability analysis and well path optimization based on the breakout width model and Mogi–Coulomb criterion. J Pet Sci Eng. 2015;135:678–701. https://doi.org/10.1016/j.petrol.2015.10.029.CrossRefGoogle Scholar
 Ma T, Chen P. A wellbore stability analysis model with chemical–mechanical coupling for shale gas reservoirs. J Nat Gas Sci Eng. 2015;26:72–98. https://doi.org/10.1016/j.jngse.2015.05.028.CrossRefGoogle Scholar
 Manger GE. Porosity and bulk density of sedimentary rocks. United States Geological Survey Bulletin 1144E. 1963.Google Scholar
 Mayne PW, Kulhawy FH. K _{o}OCR relationship in soil. J Geotech Eng Div Am Soc Civ Eng. 1982;108(6):851–72.Google Scholar
 McLean MR, Addis MA. Wellbore stability: the effect of strength criteria on mud weight recommendations. In: The 65th annual technical conference and exhibition, September 23–26, New Orleans, LA; 1990. https://doi.org/10.2118/20405MS.
 McLellan PJ, Hawkes CD. Application of probabilistic techniques for assessing sand production instability risks. In: SPE/ISRM Eurock ‘98 held in Trondheim, Norway, 8–10 July, Trondheim, Norway; 1998. https://doi.org/10.2118/47334MS.
 McWorter DB, Sunada DK. Groundwater hydrology and hydraulics. Fort Collins: Water Resources Publications; 1977.Google Scholar
 Moayed RZ, Bolandi MA. Effect of elastic modulus varieties in depth on subgrade reaction modulus of granular soils. In: Second international conference on geotechnique, construction materials and environment, Nov. 14–16, 2012, Kuala Lumpur, Malaysia, ISBN: 9784990595814 C3051; 2012.Google Scholar
 Moayed RZ, Tamassoki S, Izadi E. Numerical modelling of direct shear tests on sandy clay. World Acad Sci Eng Technol. 2012;6(1):1093–7.Google Scholar
 Mohiuddin MA, Khan K, Abdulraheem A, AlMajed A, Awal MR. Analysis of wellbore instability in vertical, directional, and horizontal wells using field data. J Pet Sci Eng. 2007;55(1–2):83–92. https://doi.org/10.1016/j.petrol.2006.04.021.CrossRefGoogle Scholar
 Moos D, Peska P, Finkbeiner T, Zoback M. Comprehensive wellbore stability analysis utilizing quantitative risk assessment. J Pet Sci Eng. 2003;38(3–4):97–109. https://doi.org/10.1016/S09204105(03)00024X.CrossRefGoogle Scholar
 Morita N, Burton RC, Davis E. Fracturing, frac packing, and formation failure control: can screenless completions prevent sand production? SPE Drill Complet. 1998;13(3):157–62. https://doi.org/10.2118/51187PA.CrossRefGoogle Scholar
 Niño FAP. Wellbore stability analysis based on sensitivity and uncertainty analysis. In: SPE annual conference technical conference and exhibition, Dubai, UAE, September 26–28; 2016. https://doi.org/10.2118/184480STU.
 Nouri A, AlDarbi MM, Vaziri H, Islam MR. Deflection criteria for numerical assessment of the sand production potential in an openhole completion. Energy Sources. 2002;24(7):685–702. https://doi.org/10.1080/00908310290086617.CrossRefGoogle Scholar
 Ottesen S, Zheng RH, McCann RC. Wellbore stability assessment using quantitative risk analysis. In: SPE/IADC drilling conference in Amsterdam, Holland, 9–11 March; 1999. https://doi.org/10.2118/52864MS.
 Pašić B, GaurinaMedimurec N, Davorin M. Wellbore instability: causes and consequences. Rudarsko Geolosko Naftni Zbornik. 2007;19(1):87–98.Google Scholar
 Radoslaw L, Michalowski F. Coefficient of earth pressure at rest. J Geotech Geoenviron Eng. 2005;131(11):1429–33. https://doi.org/10.1061/(ASCE)10900241(2005)131:11(1429).CrossRefGoogle Scholar
 Savoia M. The role of uncertainties in structural engineering problems: how to manage them in simulation and design. Italian Academy Fellows’ seminar working papers. Italian Academy for Advanced Studies in America, Columbia University; 2012.Google Scholar
 Secor DT. Role of fluid pressure in jointing. Am J Sci. 1965;263(8):633–46. https://doi.org/10.2475/ajs.263.8.633.CrossRefGoogle Scholar
 Shanmugam G, Higgins JB. Porosity enhancement from chert dissolution beneath Neocomian unconformity: Ivishak Formation, North Slope, Alaska. AAPG Bull. 1988;72(5):523–35.Google Scholar
 Sheng Y, Reddish D, Lu Z. Assessment of uncertainties in wellbore stability analysis. Mod Trends Geomech. 2006. https://doi.org/10.1007/9783540357247_31.Google Scholar
 Sheorey PR. A theory for in situ stresses in isotropic and transversely isotropic rock. Int J Rock Mech Min Sci Geomech Abstr. 1994;31(1):23–34. https://doi.org/10.1016/01489062(94)923124.CrossRefGoogle Scholar
 Sheorey PR, Mohan GM, Sinha A. Influence of elastic constants on the horizontal in situ stress. Int J Rock Mech Min Sci. 2001;38(8):1211–6.CrossRefGoogle Scholar
 Simangunsong RA, Villatoro JJ, Davis AK. Wellbore stability assessment for highly inclined wells using limited rockmechanics data. In: SPE annual technical conference and exhibition, 24–27 September, San Antonio, Texas, USA; 2006. https://doi.org/10.2118/99644MS.
 Standifird W. Stabilizing wellbore stability analysis [Online]. Exploration and Production. Houston: Hart Energy; 2006. http://www.epmag.com/EPMagazine/archive/Stabilizingwellborestabilityanalysis_5994. Accessed August 2006.
 Tan CP, Willoughby DR. Critical mud weight and risk contour plots for designing inclined wells. In: SPE annual technical conference and exhibition, 3–6 October, Houston, Texas; 1993. https://doi.org/10.2118/26325MS.
 Tan CP, Yaakub MA, Chen X, Willoughby DR, Choi SK, Wu B. Wellbore stability of extended reach wells in an oil field in Sarawak Basin, South China Sea. In: SPE Asia Pacific oil and gas conference and exhibition, 18–20 October, Perth, Australia; 2004. https://doi.org/10.2118/88609MS.
 Terzaghi K. Theoretical soil mechanics. New York: Wiley; 1943.CrossRefGoogle Scholar
 Terzaghi K, Peck RB, Mesri G. Soil mechanics in engineering practice. New York: Wiley; 1996.Google Scholar
 van Oort E. On the physical and chemical stability of shales. J Pet Sci Eng. 2003;38(3–4):213–35. https://doi.org/10.1016/S09204105(03)000342.CrossRefGoogle Scholar
 Vernik L, Zoback MD. Estimation of maximum horizontal principal stress magnitude from stressinduced well bore breakouts in the Cajon Pass Scientific Research borehole. J Geophys Res. 1992;97(B4):5109–19. https://doi.org/10.1029/91JB01673.CrossRefGoogle Scholar
 Woolley JA. Shock tube experiments measuring & modeling the slow pwave in low permeability rock cores. MSc Thesis. Colorado School of Mines; 2004.Google Scholar
 Zhang J. Pore pressure prediction from well logs: methods, modifications, and new approaches. Earth Sci Rev. 2011;108:50–63. https://doi.org/10.1016/j.earscirev.2011.06.001.CrossRefGoogle Scholar
 Zhang X, Last N, Powrie W, Harkness R. Numerical modelling of wellbore behaviour in fractured rock masses. J Pet Sci Eng. 1999;23(2):95–115. https://doi.org/10.1016/S09204105(99)000108.CrossRefGoogle Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.