The effect of stresses on the sound velocity in rocks: Theory of Acoustoelasticity and Experimental Measurements Xiaojun Huang, Daniel R. Burns, and M. Nafi Toksöz, ERL, MIT Abstract The theory of acoustoelasticity provides direct link between the change of elastic wave velocities and residual stresses in solids. The general theory of acoustoelasticity is reviewed. A number of experimental measurements of the effect of stresses on the sound velocities in various types of rocks are compiled and compared to the acoustoelastic theory. The theory of acoustoelasticity agrees within 1% of error with experiments for stress levels that are representative for in-situ reservior conditions. With the measurements of Nur and Simmons, acoustoelastic theory is found to agree with Sayers’s microcrack model within 2% of error, much smaller than experimental error which was 10%. We may safely conclude that the theory of acoustoelasticity is a macroscopic version of the microcrack model and applicable to in-situ rocks. 1 Introduction Knowledge of in-situ stresses is of great interest to the petroleum industry for many purposes. In-situ stress is an important factor in engineering applications such as enhanced recovery of hydrocarbons, prevention of sand production and borehole instability. In exploration, in-situ stresses affect seismic anisotropy, caprock integrity, and whether or not fluids can flow along or across faults. The most accurate way by far for determining in-situ stresses is by conducting lowvolume hydraulic fracturing experiments (Zoback and Haimson, 1982). Apart from the fact that such measurements are expensive and can be done at only a few locations along the wellbore, hydraulic fracturing experiments are based on a number of assumptions that barely hold in many real situations (Haimson, 1988). In recent years, some efforts have focused on initiating an alternative mechanism to assess in-situ stresses using sonic well logs. Following the establishment of a theoretical framework on wave propagating along a borehole surrounded with prestressed formation (Sinha and Kostek, 1996), a multi-frequency inversion scheme was developed and successfully applied to assess in-situ stress profiles using a set of field sonic logs (Huang et al., 1999). Both Sinha and Huang’s work are in the context of the theory of acoustoelasticity, the basic theory that relates the change of sound wave velocities to initial stresses. This theory invokes third-order elastic (TOE) constants to account for the nonlinear strain response to 1 stresses of finite magnitude. Developments in acoustoelasticity have been stimulated by interest in the measurement of TOE constants of crystals and applied or residual stresses in polycrystalline materials. Due to the presence of compliant mechanical defects (cracks, microfractures, grain joints, etc), rocks in general display stronger nonlinear elastic behavior than crystals or polycrystalline materials [Meegan et al. (1993), Johnson et al. (1993), and Johnson and McCall (1994)]. In this paper, we will compile a number of experiments on stress-induced changes on sound velocities in different types of rocks, and investigate how well the theory of acoustoelasticity applies to rocks. We shall first review the general theory of acoustoelasticity, and construct the theoretical framework on the sound velocity change as a function of applied stresses. It is followed by a discussion on many experiments on metal materials that confirm experimentally the theory of acoustoelasticity. We then apply the theory of acoustoelasticity to a variety of published experimental data on rocks. Finally, we shall conduct a comparison between the theory of acoustoelasticity and Sayers’s microcrack model. 2 Theory of Acoustoelasticity In the continuum theory, to establish a theoretical basis for the change of wave velocities by initial stresses in a body, a nonlinear theory is required. The nonlinearity is caused by a large deformation of the body and, in addition, a nonlinear stress-strain relation for the material. The theory of acoustoelasticity was developed in the 1960s by introducing the second-order effect of strain (third-order elastic stiffness) and separating the small motion superimposed on a largely deformed body. In this section, we review the derivation of basic equations of acoustoelasticity (Pao et al., 1984). We first introduce three material configurations: the natural, the initial and the final states. The natural and initial configurations refer to states when the material is free of stresses and statically deformed, respectively. The final configuration denotes the material state with wave-induced dynamic deformation superimposed on the static load (Fig 1). A physical variable in the natural, initial, or final state is designated by a superscript label 0, i, or f, respectively. The positions of a particle in the body at natural, initial, and final states are measured by position vectors ξ, X, and x, respectively, all directed from the origin of a common Cartesian coordinate system. The components of ξ and other physical quantities which refer to the natural configuration are denoted by Greek subscripts; those of X and others refer to the initial configuration by upper case Roman subscripts; and those of x and others refer to the final configuration by lower case Roman subscripts. Thus ξα , XJ , and xj (α, J, j = 1, 2, 3) are the components of position vectors in three respective configurations (Fig 1). The predeformation driven by T i , the initial Kirchhoff (the second Piola-Kirchhoff) stress tensor that refers to the natural configuration, is static and T i must satisfy the equations of equilibrium ∂ ∂ui i [Tβγ (δαγ + α )] = 0. ∂ξβ ∂ξγ 2 (1) Figure 1: Coordinates for a material point at the natural (ξ), initial (X) and final (x) configuration of a predeformed body. 3 The equation of motion for the body at the final state is f ∂ f ∂ 2 uf f ∂uα Tβα + Tβγ = ρ0 2α , ∂ξβ ∂ξγ ∂t (2) where T f is the final Kirchhoff stress tensor that refers to the natural configuration and ρ0 is the mass density which refers to the natural state. Subtracting Eq (1) from Eq (2), we obtain the equation of motion for the incremental displacement in natural coordinates u(ξ, t), ∂ ∂ui ∂ 2 uα i ∂uα [Tαβ + Tβγ + Tβγ α ] = ρ0 2 . ∂ξβ ∂ξγ ∂ξγ ∂t (3) So far the only assumptions made in the derivations are that the predeformation is static and that the dynamic disturbance is small. We have not asked how the particles are carried from the position ξ to X, nor have we imposed restrictions on the constitutive property of the material. Thus the equations of motion Eq (2) are applicable to a body in general form of predeformation, finite or infinitesimal, elastic or inelastic. Now we introduce the constitutive relationship in the context of acoustoelasticity. One basic assumption in the theory is that the material is hyperelastic, i.e. the material remains elastic throughout the deformation. A body in deformation is accompanied by a change of internal energy W (per unit mass) or free energy F (per unit mass). The law of balance of energy states the following: dW = θdS + Tαβ dEαβ /ρ0 , dF = −Sdθ + Tαβ dEαβ /ρ0 , (4) where θ is the temperature, S the entropy, and F = W − θS. For a hyperelastic body, W is a function of strain E and S, and F is a function of E and θ. Therefore, we have Tαβ = ρ0 ( ∂W ∂F )S = ρ0 ( )θ . ∂Eαβ ∂Eαβ (5) The subscript S indicates an adiabatic thermodynamic process, and θ an isothermal process. The function W (E) may be expanded about the state of zero strain, 1 1 ρ0 W (E) = cαβγδ Eαβ Eγδ + cαβγδη Eαβ Eγδ Eη + · · · . 2 6 (6) f i A constitutive equation for Tαβ or Tαβ is thus obtained by neglecting the higher order terms: 1 1 f f f i i i i f Eη , Tαβ = cαβγδ Eγδ + cαβγδη Eγδ Eη . Tαβ = cαβγδ Eγδ + cαβγδη Eγδ 2 2 (7) From the difference of these two equations, a constitutive equation for the incremental stress tensor Tαβ is derived, Tαβ = cαβγδ Eγδ + cαβγδη eiγδ eη , 4 (8) where the infinitesimal strain tensor ei and e are used where 2eαβ = ∂uα ∂ξβ ∂u + ∂ξαβ . The difference f i of the Lagrangian strain tensor in the final and initial states Eαβ and Eαβ respectively, is given approximately by 1 ∂uα ∂uβ ∂uiλ ∂uλ ∂uiλ ∂uλ f i + + + ). Eαβ = Eαβ − Eαβ = ( 2 ∂ξβ ∂ξα ∂ξα ∂ξβ ∂ξβ ∂ξα (9) In terms of displacement gradients, the constitutive equation is Tαβ = cαβγδ (δργ + ∂uiρ ∂uρ ∂uiγ ∂u ) + cαβγδη . ∂ξγ ∂ξδ ∂ξδ ∂ξη (10) i In Eq (10), only terms linear in ∂u or ∂u are retained. ∂ξ ∂ξ Substituting the constitutive equations for Tαβ into Eq (3), we obtain the equation of motion for u(ξ, t), 2 ∂ ∂uγ i ∂uα 0 ∂ uα [T + Γαβγδ ]=ρ . ∂ξβ γβ ∂ξγ ∂ξδ ∂t2 (11) This is the equation for the dynamic disturbance with reference to the natural coordinates. The initial stress can be arbitrarily distributed and the material can have intrinsic anisotropy. The coefficient Γαβγδ = Γγδαβ is of lower order symmetry than cαβγδ . It is the effective elastic moduli of the material after predeformation. Γαβγδ ∂uiγ ∂ui = cαβγδ + cαβρδ + cρβγδ α + cαβγδη eiη ∂ξρ ∂ξρ (12) Within the framework of the aforementioned theory, the assumptions made that lead to Eq (3) are as follows: • The predeformation is static and the body is at equilibrium in the initial state. • The superposed dynamic motion is small. For a homogeneously predeformed medium, i.e. T i and ∂ui /∂ξ are constant throughout the body, the equation of motion [Eq (11)] is reduced to ∂ 2 uγ ∂ 2 uα = ρ0 2 , ∂ξβ ∂ξδ ∂t (13) i Aαβγδ = Tβδ δαγ + Γαβγδ (14) Aαβγδ where are constant coefficients. A plane sinusoidal wave is represented by uα = Uα exp[i(κνβ ξβ − ωt)], 5 (15) where U is a constant complex vector, ω the angular frequency, κ(= 2π/wavelength) the wave number, and ν (a unit vector) the wave normal. The wave speed is given by v = ω/κ. On substituting Eq (15) into Eq. (13), we obtain a system of equations for the amplitude vector U: [Aαβγδ νβ νδ − ρ0 v 2 δαγ ]Uγ = 0. (16) The associated characteristic equation is |Aαβγδ νβ νδ − ρ0 v 2 δαγ | = 0. (17) Once the initial stress and initial displacement gradients are specified, and the values of ρ0 , cαβγδ , and cαβγδη in a medium are given, the eigenvalues (wave velocities) and eigenvectors (polarization directions) of Eq (16) can be solved for each direction ν of propagation. 3 Acoustoelastic Experiments on Metal Materials An early review of the acoustoelastic experimental work and analytical foundations can be found in an article by Smith (Smith, 1963). Another review of the subject by Pao (Pao et al., 1984) summarized the theory in a more general form and various techniques which had been used up to that time for making acoustoelastic measurements. In a series of detailed experiments, Hsu (Hsu, 1974) utilized the ultrasonic pulse-echooverlap technique to make absolute wave-speed measurements of the two principal shear waves propagating normal to the loading direction in aluminum and steel. Hsu denotes the uniaxial stress direction as ξ3 . According to the theory of acoustoelasticity in Section 2, the velocity of the shear wave polarized in ξ1 direction v13 is 2 i v13 = (v13 )20 + ΠT33 , (18) where Π is a constant depending on the elastic moduli and the TOE constants of the material. i (v13 )20 is the shear velocity in the stress-free state and T33 is the applied uniaxial stress. Hsu observed that for aluminum and steel, the relative change in wave speed with applied stresses 2 is very small; therefore, v13 − (v13 )20 = (v13 + (v13 )0 )(v13 − (v13 )0 ) ' 2(v13 )0 (v13 − (v13 )0 ), thus Eq (18) becomes v13 = (v13 )0 + Π Ti . 2(v13 )0 33 (19) i Similarly, v23 is also a linear function of the applied uniaxial stress T33 . For three different types of steel and four different types of aluminum, Hsu found that the velocity changes were linear functions of the applied stresses (Figure 2). Additional wave-speed measurements using various wave modes were made by Egle and Bray (Egle and Bray, 1976) on rail-steel specimens parallel as well as perpendicular to the loading axis (Figure 3). Hirao et al conducted a surface wave experiment with a mild steel specimen (Hirao et al., 1981). Both theoretical and experimental results for the change of Rayleigh wave with respect to the uniaxial strain are plotted in Figure 4. 6 Figure 2: Shear velocity changes as function of applied compressive stress in three steel specimens. Vij denotes the shear wave polarized in the j th direction propagating in the ith direction (Hsu, 1974). Figure 3: Relative change in wave speed ( ∆V ) as a function of strain for a rail specimen V0 (Egle and Bray, 1976). 7 Figure 4: Relative variations of Rayleigh wave transit time and velocity versus the uniaxial strain (Hirao et al., 1981). Figure 5: Specimen geometry and scan lines for the aluminum panel with central hole measured by Kino et al (Kino et al., 1978). Figure 6: Specimen geometry and scan line for the double edge-notched aluminum panel measured by Kino et al (Kino et al., 1978). 8 Figure 7: Stress contour plots of results for the aluminum panel with central hole. Solid curve, theory; dotted curve, acoustoelastic experiment (Kino et al., 1978). Figure 8: Stress contour of the double edge-notched aluminum panel. Solid curve, photoelastic measurements; dotted curve, acoustoelastic measurements (Kino et al., 1978). 9 Kino et al (Kino et al., 1978) made use of the phase comparison technique to make precise time-of-flight measurements with longitudinal waves while the transducer is scanned on a two-dimensional grid across specimens of aluminum. Geometries of specimens and scan lines are shown in Figure 5 and 6. Stress fields of both specimens are inverted from measured longitudinal wave velocities using the aforementioned theory of acoustoelasticity, and compared with theoretical results and results by another measurement scheme (photoelasticity). Stress-field contour plots are displayed in Figures 7 and 8. These and many other experiments confirmed experimentally the theory of acoustoelasticity in the context of metal materials. 10 4 Stress-induced Velocity Changes in Rocks The stress-induced velocity change in rocks is a well-established observation. It is classically modeled by a change in the alignments of cracks or any other alignment of micro-structural flaws or defects, induced by a changing stress (e.g. (Sayers et al., 1990)). Only a few studies take the approach of acoustoelasticity (Johnson and Rasolofosaon, 1996). It it advantageous to employ the theory of acoustoelastic theory in estimating in-situ stresses, for it provides direct links between stresses and the change of elastic wave velocities. In this section, we shall first test the theory of acoustoelasticity against experimental measurements of different types of rocks. We then make a connection between Sayers’s microcrack model and the theory of acoustoelasticity. Our goal here is not to present an exhaustive analysis of the experimental data available from the literature, but to illustrate the relevance of the above analysis. We analyze high precision data by Dillen (2000). We also take experimental measurements by Johnson (1993) and Lo et al (1986) as an example of velocity changes with confining pressures, and Nur and Simons (1969) with uniaxial stresses. 4.1 Colton Sandstone In Dillen’s thesis, the ultrasonic experiment is carried out on a cubic block of Colton sandstone. The sample consists of lithic quartz and feldspar. It is fairly homogeneous and has a porosity of about 3%. At zero stress state, the measured P-wave velocities in the X− and Z−directions are approximately equal and differed by 5% from the velocity in the Y −direction. We assume that the Colton sandstone is transversely isotropic with the symmetric axis in the Y −direction. Figure 9 shows the load cycle ABCD as a function of experiment time. The entire ABCD stress path has equal normal stresses in the X− and Z−directions. According to the characteristic equation (Eq (17)), when a plane wave is propagating in the X−direction, ν1 = 1, ν2 = 0 and ν3 = 0; therefore, Eq (17) becomes |Aα1γ1 − ρ0 v 2 δαγ | = 0. (20) In matrix form, it is where A11 − ρ0 v 2 A16 A15 0 2 A61 A66 − ρ v A65 A51 A56 A55 − ρ0 v 2 = 0, i A11 = c11 + T11 + (2c11 + c111 )ei11 + c112 ei22 + c113 ei33 , i A55 = c55 + T33 + (2c55 + c155 )ei11 + c144 ei22 + c344 ei33 , i A66 = c66 + T22 + (2c66 + c166 )ei11 + c266 ei22 + c366 ei33 . (21) (22) (23) (24) In Dillen’s experiment, one of the principal initial strains is in the X−direction; therefore, ei31 = ei21 = 0, thus A51 = A15 = A16 = A61 = 0. From Eq (21) we may obtain the 11 12 A 10 B C D stress X = stress Z stress (MPa) 8 6 4 2 0 0 stress Y 200 400 600 800 1000 1200 1400 experiment time (s) Figure 5.1: cycle Loading cycle of ABCD of the tri-axial pressure machine. To preserve Figure 9: Loading ABCD the tri-axial pressure machine. To preserve the intrinsic the isotropy intrinsic of transverse isotropy of the insample, the stress inis the X-direction is in the transverse the sample, the stress the X−direction equal to the stress equal to the stress in the Z-direction. Z−direction (Dillen, 2000). the X- velocity and Z-directions. During parts A, B,as C,follows and D, the X-force and the compressional propagating in X−direction Z-force increase. Part A shows a decrease of the stress in the Y -direction, A11 the Y -direction was kept conwhereas during parts B and C the . (25) vp2x stress = 0 in stant at 2 MPa and 4 MPa, respectively.ρFinally part D simulates increasing hydrostatic stress conditions up to 10 MPa. A11 is determined in Eq (22). According to the acoustoelasticity theory, initial strains ei11 , i i i ei22 and ei33 are linearly related with applied stresses T11 , T22 and T33 through Hook’s Law. i T i results i i T −1 i i Let E =5.2.3 [ei11 ei22 eExperimental ] and T = [T T T ] , we have E = C T. Noting that T11 = T22 33 11 11 33 in the experiment. C is the of elasticsignals moduli, Fig. (5.2) shows thematrix full waveform of compressional-waves as func tions of experiment time (top horizontal axis) c11 c12 c13 and transmission travel time (vertical axis). The bottom horizontal axis indicates the segments A, B, C, C = c12 c22 c12 (26) and D similar to those in Fig. (5.1). source and receiver are both P-wave c13 The c12 c11 transducers aligned in the X-direction. The transducer-sample coupling, described in the previous section, produces traces easily Substituting initial strains as a function of appliedclean stresses intowith Eq (22) anddiscernible then substituting first arrivals and amplitudes. To visualize shear-wave splitting, shear-wave Eq (22) into Eq (25), we find that the square of compressional velocity propagating in transducers were aligned in the X-direction and polarized in the Y Z-plane X−direction is a linear function of applied stresses T11 and T22 : in a direction of 45 degrees to the symmetry axis. Fig. (5.3) displays the i i vp2x = (vpx )20 + AT11 + BT22 , (27) 81 where (vpx )0 denotes the compressional wave propagating in the X−direction in the natural state. A and B are constants that are completely determined by elastic moduli and TOE constants of the rock. i i T22 is constant in loading period B and C, thus vp2x is a linear function of T11 . In period i A, normal stresses in X and Z−directions T11 raise from 5 MPa to 7 Mpa, and for the same 12 i period of time, T22 decreases from 5 MPa to 1 MPa. We may work out the relation between i i T11 and T22 in period A as i i T22 = 5 − 2T11 . (28) i Substituting Eq (28) into Eq (27), we may also find that vp2x is a linear function of T11 in loading period A, only with a different slope from that in periods B and C. Similar i dependence of vp2x on T11 is found in load period D with yet another slope. Similar analysis holds for dependence of P wave propagating in the Y −direction and for shear waves. Figures 10, 11, 12 and 13 show both experimental and theoretical results of compressional and shear velocities versus the normal stress in the X−direction during cycle ABCD. The fact that very small root mean square errors between theory and experiments suggests that the theory of acoustoelasticity holds for rocks in the stress range of the experiment. We may infer that in the regime of small stresses which is below 10 MPa in the above experiment, there is no permenant deformation in the rock, i.e., cracks will reopen when stresses are removed. This satisfied the assumption associated with the theory of acoustoelasticity which requires the rock to be elastic. 4.2 Chelmsford Granite, Chicopee Shale and Berea Sandstone A second experimental data set is analyzed for the following reason. The stress range up to 10 MPa to which the Colton sandstone was subjected, as described above, is too low to encompass in-situ stresses that occur in a hydrocarbon reservoir. An exception is overpressured reservoirs, showing anomalously high pore fluid pressures, resulting in correspondingly low effective stresses. Therefore, Johnson and Christensen (1993) and Lo’s (1986) experimental data with confining pressures up to 200 MPa on Millboro and Braillier shales and up to 100 MPa on Berea sandstone, Chicopee shale and Chelmsford granite are analyzed in the following. In both Johnson and Lo’s experiments, three compressional and six shear velocities are measured for each of the rock samples under vacuum dry condition (Johnson and Christensen, 1993) and Lo et al. (1986). Figure 14 illustrates velocities measured and symmetry planes of rock samples each of which is measured transversely isotropic in the natural state. Figures 15, 16 and 17 show that root mean square errors between experiments and theoretical predictions are below 1% for every shale sample at all confining pressures. The excellent agreement between theory and experiment is not surprising, for all shale samples have very low porosities thus crack growth and coalescence, primary factors attributing to the highly nonlinear and inelastic behavior in rocks, are very inactive. For the Berea sandstone sample, experiments and theory show sound agreement at confining pressure levels that are higher than 30 MPa (Figure 19). A significant portion of cracks in this rock are closed at the confining pressure of 30 MPa therefore the rock starts to have a similiar behavior to shales. The corresponding error between experiments and theory is also winthin 1%. For the granite sample, the required confining pressure to close the majority of cracks is 40 MPa (Figure 18). When stresses are applied to a rock, elastic and inelastic deformations are competing each other. The inelastic deformation includes permenant closure of cracks, development of 13 Period A Period B PX − wave velocity (m/s) 3150 PX − wave velocity (m/s) 3090 3080 3100 SRME=1.0709e−04 m/s 3070 3060 3050 3050 Experiment Theory 3040 3030 SRME=0.0054 m/s 5 5.5 6 6.5 stress T11 (MPa) Experiment Theory 3000 2950 7 2 4 Period C 10 Period D PX − wave velocity (m/s) 3200 PX − wave velocity (m/s) 3160 3140 3150 3120 SRME=0.0309 m/s 3100 SRME=2.7841e−04 m/s 3100 3050 3080 3000 Experiment Theory 3060 3040 6 8 stress T11 (MPa) 4 6 8 stress T11 (MPa) Experiment Theory 2950 2900 10 0 2 4 6 stress T11 (MPa) 8 10 Figure 10: Theoretical and experimental results of velocity of the compressional wave propagating in the X−direction versus the normal stress in the X−direction during the cycle ABCD (Dillen, 2000). 14 Period A Period B 2910 2870 PY − wave velocity (m/s) PY − wave velocity (m/s) SRME=0.0029 m/s 2900 2860 2890 2850 2880 2870 2840 2860 Experiment Theory 2850 2840 SRME=3.1357e−04 m/s 5 5.5 Experiment Theory 2830 6 6.5 stress T11 (MPa) 2820 7 2 4 Period C 10 Period D PY − wave velocity (m/s) 3050 PY − wave velocity (m/s) 2915 2910 3000 SRME=3.1716e−04 m/s 2905 SRME=0.0376 m/s 2950 2900 2900 2895 2850 Experiment Theory 2890 2885 6 8 stress T11 (MPa) 4 6 8 stress T11 (MPa) Experiment Theory 2800 2750 10 0 2 4 6 stress T11 (MPa) 8 10 Figure 11: Theoretical and experimental results of velocity of the compressional wave propagating in the Y −direction versus the normal stress in the X−direction during the cycle ABCD. 15 Period A Period B SXZ − wave velocity (m/s) 2080 SXZ − wave velocity (m/s) 2040 2035 2060 SRME=3.5727e−05 m/s 2030 2040 2025 2020 2020 2000 Experiment Theory 2015 2010 SRME=0.0027 m/s 5 5.5 6 6.5 stress T11 (MPa) Experiment Theory 1980 1960 7 2 4 Period C 10 Period D SXZ − wave velocity (m/s) 2100 SXZ − wave velocity (m/s) 2100 2080 2050 SRME=2.8773e−04 m/s 2060 SRME=0.0078 m/s 2000 2040 Experiment Theory 2020 2000 6 8 stress T11 (MPa) 4 6 8 stress T11 (MPa) Experiment Theory 1950 1900 10 0 2 4 6 stress T11 (MPa) 8 10 Figure 12: Theoretical and experimental results of velocity of the shear wave propagating in the X−direction and polarizing in the Z−direction versus the normal in the X−direction during the cycle ABCD (experiment by Dillen). 16 Period A Period B SXY − wave velocity (m/s) 2020 SXY − wave velocity (m/s) 2030 2020 2010 SRME=2.7087e−04 m/s 2010 2000 1990 1990 1980 1980 Experiment Theory 1970 Experiment Theory 1970 1960 SRME=6.7788e−04 m/s 2000 1960 5 5.5 6 6.5 stress T11 (MPa) 1950 7 2 4 Period C 10 Period D 2030 SXY − wave velocity (m/s) 2100 SXY − wave velocity (m/s) 2040 SRME=7.6680e−05 m/s SRME=0.0208 m/s 2050 2020 2000 2010 Experiment Theory 2000 1990 6 8 stress T11 (MPa) 4 6 8 stress T11 (MPa) Experiment Theory 1950 1900 10 0 2 4 6 stress T11 (MPa) 8 10 Figure 13: Theoretical and experimental results of velocity of the shear wave propagating in the X−direction and polarizing in the Y −direction versus the normal in the X−direction during the cycle ABCD (experiment by Dillen). 17 VP33 VS3a VS3b VP11 VSV1 VSH1 VP45 VSV45 VSH45 Figure 14: Velocities measured and symmetry planes of rock samples each of which is measured transversely isotropic in the natural state (both Johnson and Lo’s experiments). 18 new cracks that results from local failures and permenant relative movement between rock grains. In order to account for inelastic deformation, the theory of acoustoelasticity has to be modified. From the above analysis, we find that the theory of acoustoelasticity applies to all shale samples of low porosity. For the sandstone and granite samples, in the intermediate stress regime which is about 10 MPa to 30 or 40 PMa, they undergo primarily inelastic deformations. When the stress level is 30 or 40 MPa higher, the theory of acoustoelasticity works well with all types of rocks that are studied in this paper, because most cracks are closed at that confining pressure level and thus elastic deformation becomes primary. Rocks are under high confining pressures when they are kilometers beneath the surface of the earth where we are interested to measure in-situ stresses. So for the purpose of estimating in-situ stresses, the theory of acoustoelasticity is applicable to all types of rocks. Braillier Shale 6 6 Vpaa Vpbb Vpcc 5.8 4 Vpd VSac VSbc 5.5 VSca VSdh VSdv 3.5 5.6 5 RMSE (km/s): 4.5 5 4.8 RMSE (km/s): 0.0144 0.0131 3 0.0064 4 3.5 0.0045 3 velocity (km/s) 5.2 velocity (km/s) velocity (km/s) 5.4 0.0043 2.5 2 4.6 RMSE (km/s): 4.4 0.0456 0.0203 0.0120 4.2 4 2.5 0.0114 2 1.5 1.5 0 100 200 confining pressure (MPa) 1 0 100 200 confining pressure (MPa) 1 0 100 200 confining pressure (MPa) Figure 15: Theoretical and experimental results for Braillier shale (experiment by Johnson and Christensen (1993). ∆: experiment; solid line: theory. RMSE denotes root mean square error between theory and experiment. 19 Millboro Shale 6 6 Vpaa Vpbb Vpcc 5.8 4 Vpd VSac VSbc 5.5 VSca VSdh VSdv 3.5 5.6 5 RMSE (km/s): 0.0124 0.0188 5 4.8 3 0.0271 velocity (km/s) velocity (km/s) 0.0375 5.2 RMSE (km/s): RMSE (km/s): 4.5 4 3.5 0.0045 3 0.0159 0.0065 velocity (km/s) 5.4 0.0154 2.5 2 4.6 2.5 4.4 2 0.0140 1.5 4.2 4 1.5 0 100 200 confining pressure (MPa) 1 0 100 200 confining pressure (MPa) 1 0 100 200 confining pressure (MPa) Figure 16: Theoretical and experimental results for Millboro shale (experiment by Johnson and Christensen (1993). ∆: experiment; solid line: theory. 20 Chicopee Shale 7 4 4 RMSE (km/s): RMSE (km/s): RMSE (km/s): 3.8 3.8 0.0049 0.0294 6.5 0.0294 0.0182 0.0101 3.6 0.0324 0.0082 3.6 0.0124 0.0063 3.4 3.4 3.2 3.2 5.5 velocity (km/s) velocity (km/s) velocity (km/s) 6 3 2.8 3 2.8 5 2.6 Vp11 Vp45 V 4.5 p33 4 0 50 100 confining pressure (MPa) 2.6 2.4 VSH1 VSV1 VSH45 2.2 2 0 50 100 confining pressure (MPa) 2.4 VSV45 VS3a VS3b 2.2 2 0 50 100 confining pressure (MPa) Figure 17: Theoretical and experimental results for Chicopee Shale (experiment by Lo et al (1986). ∆: experiment; solid line: theory. 21 Chelmsford Granite 7 4 RMSE (km/s): 3.9 0.0223 6.5 3.8 RMSE (km/s): 0.0190 RMSE (km/s): 3.7 0.0110 0.0137 0.0137 3.8 0.0232 0.0110 3.6 0.0110 3.7 3.5 3.6 3.4 0.0110 5.5 velocity (km/s) velocity (km/s) velocity (km/s) 6 3.5 3.4 3.3 3.2 5 3.3 4.5 4 40 Vp11 Vp45 Vp33 60 80 100 confining pressure (MPa) 3.1 3.2 VSH1 VSV1 VSH45 3.1 3 40 60 80 100 confining pressure (MPa) 3 2.9 2.8 40 VSV45 VS3a VS3b 60 80 100 confining pressure (MPa) Figure 18: Theoretical and experimental results and their relative errors for Chelmsford granite (experiment by Lo et al (1986). ∆: experiment; solid line: theory. 22 Berea Sandstone 5 3 Vp11 Vp45 Vp33 2.9 2.8 4.4 2.7 2.7 4.2 2.6 2.6 3.8 2.5 2.4 RMSE (km/s): 3.6 0.0177 3 20 RMSE (km/s): 2.3 2.2 0.0145 0.0082 2.2 0.0117 2.1 40 60 80 100 confining pressure (MPa) 2.4 0.0145 0.0282 3.2 2.5 RMSE (km/s): 2.3 0.0193 3.4 velocity (km/s) 2.8 4 2 20 VSV45 VS3a VS3b 2.9 4.6 velocity (km/s) velocity (km/s) 4.8 3 VSH1 VSV1 VSH45 0.0164 0.0164 2.1 40 60 80 100 confining pressure (MPa) 2 20 40 60 80 100 confining pressure (MPa) Figure 19: Theoretical and experimental results and their relative errors for Berea sandstone (experiment by Lo et al (1986). ∆: experiment; solid line: theory. 23 4.3 Barre Granite: Acoustoelastic Approach and Sayers’s Microcrack Model Nur and Simmons (Nur and Simmons, 1969) subject a cylinder of Barre granite to a uniaxial compressive stress normal to the axis of the cylinder. Compressional and shear wave velocities were measured for propagation normal to the cylinder axis at various angles β to the stress axis. It is more convenient to work with cylindrical coordinate given the geometry of the experimental setup. Note that in cylindrical coordinate, T = [Ti11 Ti22 Ti33 ]T = [Tirr Tiββ Tizz ]T , and E = [ei11 ei22 ei33 ]T = [eirr eiββ eizz ]T . The initial strain E is linearly related to the initial stress T through elastic moduli as mentioned previously. Components of E and T in Eq (22) can be those in cylindrical coordinates as defined above. Suppose the applied uniaxial stress, σ, coincides with β = 0o ; therefore, at angle β, the three normal stresses components of T in cylindrical coordinates are σcos2 β T = σsin2 β . (29) 0 Substituting Eq (29) into Eq (25), we obtain the compressional wave velocity as a function of applied stress σ and the angle β as vp2 = (vp )20 + Aσ + Bσcos2 β, (30) where A and B are constants determined by elastic moduli and TOE constants of the rock. (vp )0 is the compressional velocity in the natural state. We choose the average stress-free velocity (vp )0 at all angles to be 3.79 km/s, and then invert Nur’s data (10 MPa, 20 MPa and 30 MPa) for constants A and B. The results are A = 0.008 and B = 0.0225. An alternative way of analyzing stress-induced velocity changes in rocks is to think microscopically which has long been well received in geophysical community[e.g.Sayers (1988a), Sayers et al. (1990)]. Since the theory of acoustoelasticity, which is a macroscopic approach, also works reasonably well with rocks, as shown previously, we shall make comparison between the two approaches. Sayers applied the microcrack theory to the measurements of Nur and Simmons (Sayers, 1988b). His formula for the compressional velocity at angle β is vp = A + Bcos2β, (31) where A and B are unknown variables that depend not only on the elastic properties of the rock, but also the applied stress. He evaluated A and B by fitting the measurements of Nur and Simmons with the following results: Stress (MPa) A B 10 4.052 0.199 20 4.271 0.301 30 4.414 0.322 24 Figure 20 exhibits compressional wave velocity measurements of Nur and Simmons for Barre granite compared with the acoustoelastic theory and microcrack model prediction. An error analysis shows errors between experiment measurements and both models are mostly below 2% (Figure 21). Relative errors between the two models also suggest the two models agree with each other quite well (Figure 22). Note that the uniaxial assumption has 10% error in the experiment (Nur and Simmons, 1969). The microcrack model implicitly deals with the velocity dependence on the applied stress. A and B are inverted for each applied stress level and thus are dependents of applied stress. On the other hand, in the acoustoelastic approach, A and B are constants that depend only on the elastic properties of the rock. So it is not surprising that microcrack model fits the experiments slightly better than acoustoelastic model. However, for the ultimate purpose of inverting measured velocity changes for formation stresses, it is a disadvantage of the microcrack model not to work explicitly with the velocity dependence on the applied stress. 5 compressional wave velocity (km/s) 30 MPa 20 MPa 4.5 10 MPa 4 3.5 0 10 20 30 40 50 β (degrees) 60 70 80 90 Figure 20: Compressional wave velocity measurements of Nur and Simmons (Nur and Simmons, 1969) for Barre granite compared with the acoustoelastic theory and microcrack model prediction (Sayers, 1988b). Solid line: acoustoelasticity, Dash line: microcrack model. 25 Relative errors between theory and experiments 4 10 MPa 20 MPa 30 MPa 3.5 3 % error 2.5 2 1.5 1 0.5 0 0 10 20 30 40 50 β (degrees) 60 70 80 90 Figure 21: Relative error between experiment measurements and theory for Barre sandstone. Solid line: acoustoelasticity, Dash line: microcrack model. 26 Relative errors between acoustoelastic theory and microcrack model 3.5 30 MPa 3 2.5 % error 2 10 MPa 1.5 1 20 MPa 0.5 0 0 10 20 30 40 50 β (degrees) 60 70 80 90 Figure 22: Relative error between acoustoelatic theory and microcrack model for Barre sandstone. 27 5 Conclusion Rocks exhibit strong nolinear stress-strain behavior. As a result, applied or residual stresses in rocks affect sound velocity considerably. In the context of the widely accepted microcrack approach, velocity changes are considered to result from crack closure driven by stresses. Because the direct dependence of velocities changes on stresses, the relationship necessary to invert sound velocity change for formation stresses, is not available. On the other hand, in the framework of acoustoelasticity, velocity changes are explicitly related with stresses. In the past 40 years, the theory of acoustoelasticity has been confirmed and widely employed to evaluate initial stresses for metal materials. Very few works have been conducted for rocks. We compiled various rock measurements and have found that the theory of acoustoelasticity agrees within 1% of error with experiments using confining pressures higher than certain levels for most rocks. The necessary confining pressure level is easily achieved under insitu conditions; therefore, the theory of acoustoelasticity is applicable to rocks in estimating formation stresses through in-situ sound velocity measurements. Using the measurements of Nur and Simmons, we compared acoustoelastic theory with the microcrak model and found they agree with each other within 2% of error, much smaller than experimental error (10%). We may safely conclude that the theory of acoustoelasticity is a macroscopic version of the microcrack model. References Menno Dillen. Time-lapse seismic monitoring of subsurface stress dynamics. PhD thesis, Technische Universiteit Delft, 2000. D. M. Egle and D. E. Bray. J. Acoust. Soc. Am., 60:741, 1976. B. Haimson. Preface. Proc. 2nd International Workshops on Hydraulic Fracturing Stress Measurements, pages 1–4, 1988. M. Hirao, H. Fukuoka, and K. Hori. J. Appl. Mech., 48:119, 1981. Nelson N. Hsu. Acoustical birefringence and the use of ultrasonic waves for experimental stress analysis. Experimental Mechanics, 14:169–176, 1974. Xiaojun Huang, Bikash K. Sinha, Daniel R. Burns, and Nafi M. Toksoz. Formation stress estimation using standard acoustic logging. Soc. Explor. Geophys. Expanded Abstracts, pages BG2.6(53–56), 1999. J. E. Johnson and N. I. Christensen. Compressional to shear velocity ratios in sedimentary rocks. Int. J. Rock Mech. Min. Sci. and Geomech. Abstr., 30:751–754, 1993. P. A. Johnson, P. N. J. Rasolofosaon, and B. E. Zinszner. Advances in Nonlinear Acoustics, chapter Measurement of nonlinear elastic response in rock by the resonant bar method, pages 531–536. World Science, Singpore, 1993. 28 P.A. Johnson and K. R. McCall. Observation and implications of nonlinear elastic wave response in rock. Geophys. Res. Lett., 21:165–168, 1994. P.A. Johnson and P.N.J. Rasolofosaon. Nonlinear elasticity and stress-induced anisotropy in rocks. J. Geophys. Res., 101:3113–3124, 1996. G. S. Kino, J.B. Hunter, G.C. Johnson, A. R. Selfridge, D. M. Barnett, G. Hermann, and Steele C. R. Acoustoelastic imaging of stress fields. J. Appl. Phys., 50(4):2607–2613, 1978. Tien-when Lo, Karl B. Coyner, and M. Nafi Toksoz. Experimental determination of elastic anisotropy of Berea sandstone, Chicopee shale, and Chelmsford granite. Geophysics, 51(1): 164–171, 1986. G. D. Jr. Meegan, P. A. Johnson, R. A. Guyer, and K. R. McCall. Observations of nonlinear elastic wave behavior in sandstones. J. Acoust. Soc. Am, 94:3387–3391, 1993. Amos Nur and Gene Simmons. Stress-induced velocity anisotropy in rock:an experimental study. J. Geophys. Res., 74:6667–6674, 1969. Yih-Hsing Pao, Wolfgang Sachse, and Hidekazu Fukuoka. Physical Acoustics, volume 17, chapter Acoustoelasticity and Ultrasonic Measurements of Residual Stresses, pages 61– 143. Academic Press, Inc, 1984. C. M. Sayers. Inversion of ultrasonic wave velocity measurements to obtain the microcrack orientation distribution function in rocks. Ultrasonics, 26:73–77, 1988a. C. M. Sayers. Stress-induced elastic wave anisotropy in fractured rock. Ultrasonics, 26: 311–317, 1988b. C. M. Sayers, J. G. Van Munster, and M. S. King. Stress-induced ultrasonic anisotropy in Berea sandstone. J. Rock Mech. Min. Sci. & Geomech. Abstr., 27(5):429–436, 1990. Bikash K. Sinha and Sergio Kostek. Stress-induced azimuthal anisotropy in borehole flexural waves. Geophysics, 61:1899–1907, 1996. R. T. Smith. Ultrasonics, 1:135, 1963. M. D. Zoback and B. Haimson. Status of the hydraulic fracturing method for in-situ stress measurements. Proc. 23rd U. S. Symp. on Rock Mechanics, pages 143–156, 1982. 29