Available online at www.sciencedirect.

com

Tunnelling and Underground Space Technology

Tunnelling and Underground Space Technology 23 (2008) 618628

www.elsevier.com/locate/tust

Inuence of stress path on tunnel excavation response Numerical tool selection and modeling strategy
M. Cai
Geomechanics Research Centre, MIRARCO, Laurentian University, Sudbury, Ontario, Canada P3E 2C6 Received 18 October 2007; received in revised form 26 November 2007; accepted 27 November 2007 Available online 21 February 2008

Abstract The actual stress path in a rock mass during tunnel excavation is complex. To capture the correct tunnel excavation response, it is important to correctly resemble the stress path in situ in the numerical tools. FLAC and Phase2 are two powerful two-dimensional continuum codes for modeling soil, rock, and structural behavior, in the elds of geotechnical, geomechanics and in civil and mining engineering. FLAC is based on explicit nite dierence formulation while Phase2 is based on implicit nite element formulation. When the two codes are applied to the analysis of tunnel excavation problems, dierence in results might occur simply due to the dierent formulation methodologies used in these codes. It is shown that for linear elastic tunnel excavation problems, both codes provide the same result because stress path is unimportant. For tunnel excavation in elasto-plastic materials using long-round drill and blast method, there is signicant dierence in terms of yielding zone distribution by the two codes if conventional modeling approach is used, especially when the rock strength is low relative to the in situ stress magnitude. The mechanism of the dierence is investigated and recommendation provided for choosing appropriate tools and modeling strategies for tunnel excavation problems. The importance of honoring the true stress path in tunnel excavation response simulation is illustrated using a few examples. 2007 Elsevier Ltd. All rights reserved.
Keywords: Stress path; Tunnel excavation; FLAC; Phase2; FEM; FDM; Numerical model

1. Introduction FLAC (Itasca, 2005) and Phase2 (Rocscience Inc., 2004) are two popular and powerful numerical analysis tools for modeling soil, rock, and structural behavior, in the elds of geotechnical, geomechanics and in civil and mining engineering. These codes provide material models such as MohrCoulomb, HoekBrown failure criteria which are suitable for geotechnical materials, supply with features to simulate underground excavation, and have structural elements that can represent the soil and rock support systems installed. Some of the applications of the tools in rock mechanics and rock engineering can be found in Hoek (2001), Martin (1993), Cai et al. (2001, 2007a,b).

E-mail address: [email protected] 0886-7798/$ - see front matter 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.tust.2007.11.005

Young and new professional engineers can not understand the hardship in the 1970s and 1980s to perform a simple 2D numerical analysis of underground excavations. Two to four weeks were often required for just preparing the model mesh (Brady and Johnson, 1989). Nowadays, easy-to-use user interface in the numerical tools has made it possible for someone with or without strong knowledge background of numerical modeling theory to conduct a tunnel excavation analysis in just a few minutes. Colorful outputs make the interpretation job much easier but at the same time they also create an illusion that the obtained results are remarkable and correct, regardless what input parameters and modeling approaches are used in the underlying analysis. Another thing that is often taken for granted by some users of the numerical tools is the lack of a complete understanding of the solution schemes used in the tools. By just

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

619

looking at the user interface and modeling procedures of FLAC and Phase2, an impression that they are the same or at least similar can be formed. The fact that Phase2 employs the implicit Finite Element Method (FEM) while FLAC uses the explicit Finite Dierence Method (FDM) may be forgotten or unrecognized by some users. Furthermore, the general attitude towards the relative merits of FEM and FDM methods is that although the governing equations are derived dierently, the resulting equations are identical for the two methods. The underlying implication is that for a specic boundary value problem in tunneling, FLAC and Phase2 should give the same result. But is this always true? In underground engineering such as civil tunnel construction and mine excavation, rocks can be removed by employing dierent excavation techniques such as full-face drilling and blasting, TBM, mechanical excavation using roadheaders, and staged excavation etc. During excavation by conventional drill and blast method, the work face is perforated by long drill holes. These holes are then lled with explosives and blasted. A dynamic unloading condition is often created by this type of excavation method. TBM excavation uses several rolling cutters and feeds high pressure of the head against the face of the tunnel, and when combined with a rotation of the head, the excavation force can lead to the crushing of the rock at the face, thus the excavation. Stress redistribution occurs due to the excavation and is often modeled by FLAC or Phase2 for rock support design. The question to be asked is: will these two numerical tools, which are widely used in the rock mechanics community, produce the same result for a tunnel excavation problem? This paper strives to illustrate that the results obtained by FLAC and Phase2 can be very dierent, depending on the material properties and the method of excavation used. It will be explained that the dierence can stem simply from the solution scheme dierence adopted in these two tools, which has not been discussed in depth in the rock engineering community. Knowing such a dierence is important in interpreting obtained numerical results and choosing the right tool and modeling approach for a particular problem under investigation. 2. Dierence in solution schemes between FLAC and Phase2 2.1. FLAC FLAC uses dynamic equations of motion in its explicit, time-marching scheme, even for static problems. The solution of solid body problems in FLAC invokes the equations of motion (Newtons law of motion), constitutive relations, and boundary conditions. The solid body is divided into a nite dierence mesh composed of quadrilateral elements. In a calculation cycle, the new velocities and displacements are obtained from stresses and forces using the equations of motion, and then, strain rates are obtained from velocities and new stresses from strain rates. Since the

calculation cycle requires that the neighboring elements should not aect each other, the adopted timestep must be smaller than a critical value for numerical stability. For plasticity analysis, FLAC checks the element state (elastic or plastic) in each cycling step. First, an elastic trial for the stress increment is computed from the total strain increment and the stresses are checked against the yield criterion. If the corresponding stresses violate the yield criteria, plastic deformation takes place. The stresses are corrected by using the plastic ow rule to ensure that they lie on the yield surface. For an elastic problem, a total of 30005000 cycling steps are needed to reduce the unbalanced force to a negligible value and thus a static solution is obtained. For an elasto-plastic problem, plastic yielding is checked once the cycling starts. Hence, the stress path diers signicantly from the elastic one if yielding is detected and stress state corrected. A typical elasto-plastic problem requires 800010,000 cycling steps to accomplish the solution. 2.2. Phase2 Phase2 is an implicit FEM program developed initially for underground excavation simulation, and subsequently new functions and capabilities have been added allow it to be used for slope stability analysis and groundwater ow simulation. In the FEM approach, the domain is discretized into a nite number of elements with a xed number of nodes. Displacements inside an element are approximated using a shape function that links the nodal displacement values, which are the system unknowns. The original partial dierential equations are replaced by an assembly of algebraic system of equations. A global stiness matrix is formed and stored. Solving the system of equations determines the node displacement values which in turn can be used to obtain stresses and strains in each element. For plasticity analysis, Phase2 tackles the problem by a series of static equilibrium solutions involving iterative procedures. An elastic solution is rst obtained and the stress state in each element is checked against the yield criteria. If
Table 1 Comparison of FLAC and Phase2 FLAC Solution scheme Computer memory requirement Non-linear problem handling Physical process Explicit Low No iteration necessary Computationally stable Always follow the physics if the timestep criterion is guaranteed Delete or assign null element Yes Yes (interface element) Good 1986 Phase2 Implicit High Iteration required Diverge may occur Need to be demonstrated that it follows the physical process Excavation (assign very low modulus) Yes Yes (joint element) Excellent 1990

Excavation method Structural elements Discontinuity model User interface First release

620

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

t
T0
E, , F A
Vp

the stresses violate the yield criteria, plastic deformation takes place. The nal solution is obtained by iteration and the quality of the solution depends on the convergence criterion and the algorithm (solver type) used to return stresses to the yield surface. 2.3. Comparison
T

Fig. 1. Stress wave propagation in a beam.

y Host rock

Tunnel

x
Excavated by deleting in the numerical modeling

Fig. 2. Tunnel excavation simulation.

A comparison of the explicit and implicit solution methods can be found in the FLAC manual Theory and Background. A comparison, specic to features of FLAC and Phase2, is presented in Table 1. One thing that is of interest for discussion is the ability of the tools to model the true physical process in underground excavation. Since the structure response is stress path depended for nonlinear materials (Cai et al., 2002; Kaiser et al., 2001; Ruistuen and Teufel, 1996), if a solution scheme cannot generate stress paths that represent the actual ones, then, a dierent structural response should be expected. This is the case for numerical tools that employ explicit and implicit solution schemes. Explicit schemes can always follow the true stress path during deformation but implicit schemes cannot. This can be demonstrated by stress propagation in a long beam subjected to the loading and boundary conditions shown in Fig. 1. Assume that the load (F) is applied to the left end A instantly at time T0. Before the load is applied, the axial stress in the beam is zero. At the time when the load is applied, load at end A is equal to F but load at end B is still zero. A stress wave is generated in the beam right after the

Phase2

3 (MPa) -30.0 -27.0 -24.0 -21.0 -18.0 -15.0 -12.0 -9.0 -6.0 -3.0 0.0

FLAC

Minimum principal stress (Pa) 0.00E+00 6.00E+06 1.20E+07 1.80E+07 2.40E+07 3.00E+07 Contour interval=3.00E+06

Fig. 3. Comparison of minimum principal stress distribution in Phase2 and FLAC (rx = 60 MPa, ry = 11 MPa; elastic response).

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

621

load is applied and the compressive stress front will travel q 1mE in the beam, where E, m, q are at a speed of V p q1 m 2 m 2 the Youngs modulus, Poissons ratio, and density of the beam, respectively. It will take T = L/Vp for the stress wave to travel from A to B, where L is the beam length. Once the stress wave reaches B, a portion of the compressive stress wave will be reected and the stress wave travels in the beam back and forth a few times before nal equilibrium is reached. This example shows that applying a load to a beam creates a dynamic loading process; the time involved is so short that most people would think it as a static problem. The physical process described above can be captured by FLAC. If Phase2 is used, one obtains only the nal result of the equilibrium state and the process of stress wave propagation cannot be captured. Since Phase2 is an FEM program, it does not mean that FEM programs cannot simulate the physical process described above. In fact, FEM codes with explicit solution algorithm, such as ABAQUS (www.abaqus.com), ELFEN (www.rockeld.co.uk), and ANSYS (www.ansys.com) can simulate this type of dynamic loading processes as well.

In underground excavation, dynamic loading and unloading conditions often exist. For example, a full-face, long-round drill and blast excavation of a tunnel in rocks will create a dynamic unloading condition to the host rocks. If FLAC and Phase2 are used to model a tunnel excavation problem (Fig. 2), the tunnel region is simply deleted or assigned to an excavated material in FLAC and Phase2, respectively, by following standard or conventional modeling approach. For reasons discussed above, it is anticipated that results from FLAC and Phase2 could be very dierent when the conventional modeling approach is used. The dierence in modeling results by these two tools are illustrated and discussed in the next section. 3. Yield zone distribution simulation in tunnel excavation by FLAC and Phase2 illustration examples 3.1. Simulation model A circular 10 m diameter tunnel excavation problem is considered. The outer boundary width and height are twenty times of the tunnel diameter with xed boundary

Phase2

FLAC

Shear Tension

state Elastic Elastic, Yield in Past At Yield in Tension

t = 5 MPa

t = 5 MPa

t = 10 MPa

t = 10 MPa

t = 15 MPa

t = 15 MPa

Fig. 4. Comparison of tensile yielding zone distribution between Phase2 and FLAC models for rt = 5, 10, 15 MPa (rx = 60 MPa, ry = 11 MPa).

622

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

condition. The Youngs modulus and Poissons ratio of the rock mass are 60 GPa and 0.25, respectively. It is assumed that the tunnel is excavated by long-round drill and blast method so that the tunnel portion in the model is deleted or excavation in one stage. In situ stress eld and tensile and shear strength are chosen as variable parameters to study the excavation response simulated by FLAC and Phase2. 4-node and 3-node elements are used in FLAC and Phase2 models, respectively. Ideally, the same mesh should be used in both the FLAC and Phase2 models to isolate the eect of mesh size (and shape) but it is deemed dicult to do so due to the dierent mesh generation algorithms used in these two codes. As an alternative, very ne meshes are used in both models to minimize the mesh inuence. As can be seen from Fig. 3, the distributions of the minimum principal stress in both models are very similar (elastic response), except that contours near the tunnel boundary in the FLAC model are missing due to a deciency in contour generation in FLAC.
Phase2

3.2. Yielding in tension Figs. 4 and 5 present the distributions of tensile yield elements in the FLAC (right column) and Phase2 (left column) models. The horizontal, vertical, and tunnel axis parallel in situ stress components are assumed to be rx = 60 MPa, ry = 11 MPa, rz = 45 MPa, respectively. In this simulation, the shear strength is set to a high value to ensure that only tensile yielding occurs in both models. Zero residual tension model is used, which means that if the rock fails in tension, its residual tensile strength is zero. The peak tensile strength (rt) varies from 5 to 30 MPa in the simulation. It is seen that when the rock tensile strength is low at 5 MPa, the dierence of the yielding zone distribution is large between the FLAC and Phase2 models (Fig. 4a and b). As the tensile strength increases, the yield zone patterns in the two models gradually converge. In addition, both Phase2 and FLAC predict the tensile yielding in the sidewall center where the tensile stress is the highest. HowFLAC

b
state Elastic Elastic, Yield in Past At Yield in Tension

Shear Tension

t = 20 MPa

t = 20 MPa

t = 25 MPa

t = 25 MPa

t = 30 MPa

t = 30 MPa

Fig. 5. Comparison of tensile yielding zone distribution between Phase2 and FLAC models for rt = 20, 25, 30 MPa (rx = 60 MPa, ry = 11 MPa).

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

623

ever, the shallow tensile failure on the sidewall, distributed over a large region away from the sidewall center, is not captured by the Phase2 model. When the peak tensile strength is 30 MPa, both models predict no tensile yielding (Fig. 5e and f). To explain the result dierence shown in Figs. 4 and 5, we need to further examine the solution scheme dierence in FLAC and Phase2, which had been briey discussed in Section 2. In reality, when a tunnel is excavated by longround drill and blast method, some of the strain energy in the system is converted into kinetic energy that needs to be dissipated. A sudden excavation creates large unbalanced forces right at the excavation boundary and the unbalanced forces need to be dissipated. FLAC, by its explicit nature, can model the stress redistribution process directly because inertial terms are included. To illustrate the stress propagation and dissipation process after excavation, the minimum principal stress distributions in a linear elastic model over several cycle steps (10, 30, 60, 100, 200) are shown in Fig. 6. Large tensile stresses are observed over a wide area at step 10. If a plastic tensile failure model (Rankine model) is used, immediate yielding would occur if the tensile stress is larger than the tensile strength. After 200 steps, the r3 P 5 MPa contour extended to a depth of 0.5 m. To reach a nal solution shown in Fig. 3b, about 5000 cycling steps are required. For the results shown in Figs. 4 and 5, the plastic tensile model is active right at the beginning in the FLAC model cycling, which means that yield elements are generated at an early stage in the cycling. In contrast, Phase2 detects the yield element based on an initial estimate using the linear stiness matrix, employing the initial minimum principal stress distribution

shown in Fig. 3a. To proof this point, we rst set the tensile strength in the FLAC model to a high value and cycle the model 8000 steps to reach equilibrium. An elastic solution is thus obtained and the minimum principal stress distribu-

Phase2

Shear Tension

t = 5 MPa

FLAC
state Elastic Elastic, Yield in Past At Yield in Tension

t = 5 MPa
Fig. 7. Comparison of tensile yielding in the Phase2 and FLAC models (rx = 60 MPa, ry = 11 MPa). In the FLAC model, an initial high tensile strength is set and after an equilibrium state is reached, the tensile strength is set to 5 MPa and the model is cycled to equilibrium.

Step 10

Step 30

Step 60

Step 100

Step 200

Minimum principal stress (Pa) 0.00E+00 2.00E+06 4.00E+06 6.00E+06 8.00E+06 1.00E+07 Contour interval= 1.00E+06

Fig. 6. Distribution of the minimum principal stress in the FLAC model at cycling step 10, 30, 60, 100, and 200 (elastic response, rx = 60 MPa, ry = 11 MPa).

624

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

tion is shown in Fig. 3b. Next, we change the rock tensile strength to 5 MPa and run the model for another 5000 steps. Immediately after cycling following the change of tensile strength properties, tensile failure is detected and plastic correction conducted. The resulting tensile yield element distribution is presented in Fig. 7b, which is dierent from the pattern shown in Fig. 4b but very similar to the distribution pattern obtained by Phase2 shown in Fig. 7a. In other words, if an elastic solution is rst obtained in FLAC, followed by conducting the plastic solution, then, the FLAC result will be similar to that obtained from the Phase2 model. Hence, the observed dierence in yielding zone distribution by the FLAC and Phase2 models is attributed to the solution scheme dierence adopted in the numerical tools. The solution scheme dierence leads to a stress path dierence. To further demonstrate the point, the inuence of in situ horizontal to vertical stress ratio (Ko) is conducted and the results are presented in Fig. 8. When Ko = 5.45, both the FLAC and Phase2 models predict tensile yielding on the sidewalls. When Ko is less than 3, the FLAC model still predicts tensile yielding due to dynamic unloading generated by tunnel excavation but the Phase2 model shows no tensile failure in the model (not shown in Fig. 8). The evolution of the minimum principal stress in the rst 200 cycling steps, at the center of the sidewall for Ko = 1, is shown in Fig. 9. It is seen that during the rst 20 cycling

steps, the minimum principal stress is tensile. If the rock tensile strength is low, tensile failure will result at an early stage, before the nal equilibrium is reached.

(107)

0.800

Minimum principal stress (Pa)

Tensile 0.400

0.000

-0.400

-0.800 Compressive -1.200

10

12

14

16

18

20

Cycling step (x 10)

Fig. 9. Evaluation of the minimum principal stress at the sidewall center during the rst 200 steps in the FLAC model (rx = 60 MPa, ry = 60 MPa; elastic solution).

FLAC
state Elastic Elastic, Yield in Past At Yield in Tension

t = 5 MPa

(a) Ko= 5.45 (x = 60 MPa y = 11 MPa)

(b) Ko= 3 (x = 60 MPa y = 20 MPa)

FLAC

(c) Ko= 2 (x = 60 MPa y = 30 MPa)

(d) Ko= 1 (x = 60 MPa y = 60 MPa)

Fig. 8. Inuence of in situ stress ratio on the tensile yielding in the FLAC model (for Phase2 model, no tensile yielded elements exist for the stress condition shown in (b)(d)).

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

625

3.3. Yielding in shear To understand the impact of dynamic unloading on shear failure of rocks in tunnel excavation by drill and blast method, the same circular tunnel excavation simulation was carried out using reduced shear strength parameters. The MohrCoulomb failure criterion is used with a perfect plastic model that has three strength parameters: c cohesion; / friction angle; w dilation angle. The in situ stresses used in the calculations are: rx = 60 MPa, ry = 11 MPa, rz = 45 MPa. The tensile strength is intentionally set to a high value so that only shear failure is observed in the modeling results. Fig. 10 presents the simulation results by FLAC (right column) and Phase2 (left column), showing the shear yielding distribution around the tunnel. It is seen that when the shear strength is relatively high (c = 30 MPa),

the yielding zone distributions in both models are similar (Fig. 10a and b). When the shear strength is reduced to 25 and 20 MPa, the shear failure patterns in both models start to dier. Some shallow shear yielding elements are observed in the FLAC model all over the tunnel boundary but not consistently show up in the Phase2 model. The deepest shear failure depths on the roof and oor as well as in the sidewalls in both models seem to agree with each other. The reason for this agreement is that to cause rocks on the roof and oor to fail in shear, the maximum tangential stress has to reach a higher value (e.g., 85.8 MPa for c = 20 MPa and / = 40). This requires approximately 400 cycling steps in FLAC. At this point, the overall stress distribution in the FLAC model is close to the elastic model result by Phase2. Subsequent cycling in FLAC leads to stress redistribution in the model in a fashion that resembles the stress iteration in

Phase2

FLAC

Shear Tension

c = 30 MPa = 40o = 10o

state Elastic At Yield in Shear or Vol. Elastic, Yield in Past

c = 25 MPa o = 40 o = 10

c = 20 MPa = 40o = 10o

Fig. 10. Comparison of shear yielding zone distribution between Phase2 and FLAC models for dierent strength levels (rx = 60 MPa, ry = 11 MPa).

626

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

Phase2 so that the deep located shear failure zones in both models are similar.

3.4. Yielding in both tension and shear Fig. 11 presents the tensile and shear yielding zone distributions in both FLAC and Phase2 models for c = 20 MPa, / = 40, w = 10, rt = 5 MPa, under the in situ stress eld of rx = 60 MPa, ry = 11 MPa, rz = 45 MPa. Again, dynamic unloading is assumed and the tensile yielding zone is concentrated on the sidewalls and shear yielding zone on the roof and oor. The resulting yield zone is very close to a simple composition of the yield zones in tension and shear (Figs. 4b and 10f). The major dierence between the two modeling results is the tensile yielding on the sidewalls. 3.5. Discussion Modeling of tunnel excavation using Phase2 and FLAC by deleting the excavation area is like excavating the tunnel by using long-round drill and blast construction method. If a dynamic unloading condition exists and when rocks behave in an elasto-plastic manner, the results obtained from these two numerical tools can be drastically dierent, especially when tensile strength of the rock is low. FLAC, due to its explicit modeling scheme, seems to capture the rock dynamic unloading process well. The insight from this investigation also explains why more rock damage is expected in tunnels excavated by drill and blast method. Previous models for describing the mechanisms by which blasting damages rock have either rated the role of the dynamic stress wave as more important in blasting fragmentation (Dally et al., 1975) or the gas action (Nilson et al.,

Phase2
Shear Tension

FLAC

state Elastic At Yield in Shear or Vol. Elastic, Yield in Past At Yield in Tension

Fig. 11. Comparison of tensile and shear yielding zone distribution between Phase2 and FLAC models (rx = 60 MPa, ry = 11 MPa).

a
Tensile yielding zone

FLAC

t = 5 MPa

c
70 60 50

E (GPa)

Modulus softening E=60 GPa

40 30 20

b
Phase2

10 0 0 1 2 3 4 5 6 Stage 7 8 9 10

E=60 0 GPa over 10 stages

t = 5 MPa
Tensile yielding zone
Fig. 12. Distribution of tensile yielding zone by material softening method in: (a) FLAC and (b) Phase2 (rx = 60 MPa, ry = 11 MPa).

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628

627

1985; Coursen, 1979). It is seen that in addition to dynamic stress loading and gas action, there is another important element by dynamic unloading that will contribute to the blasting-induced rock damage. In general, dynamic unloading-induced rock damage is shallow (<0.5 m), as revealed by the FLAC modeling result. The essence of the modeling result dierence is the stress path dependent nature of plastic materials. In fact, full-face tunnel excavation by FLAC and Phase2 creates two sets of different stress paths in the host rock due to the explicit and implicit solution schemes used in the codes, respectively. Tunnel excavation is a 3D problem. If FLAC3D (explicit, www.itascacg.com) and a Phase2 comparable 3D FEM (implicit) tool are used, similar result dierence should be expected. To honor the stress path due to face advance and alternative excavation method such as full-face TBM tunneling, various excavation modeling methods have been used in the numerical tools. One approach is called excavation relaxation method, which is available in ELFEN and can be realized in FLAC by writing a FISH function and in Phase2 by manual intervention. The elements are removed from the simulation and replaced by a set of forces equal to the internal forces of the group removed. The equivalent forces are then relaxed over a specic time period to simulate the gradual excavation relaxation process. Another approach uses the material softening method. Youngs modulus of the elements in the tunnel area are degraded to zero over a specic time period or over a few stages to simulate the gradual excavation eect. It is expected that the yield zone distribution obtained by either excavation relaxation or material softening method will be the same in both the FLAC and Phase2 models, because the same stress path is followed in both models. It should be noted that this stress path (by excavation relaxation or material softening) is dierent from the stress paths in FLAC or Phase2 by a sudden excavation of the tunnel. For reference, the tensile yield element distribution obtained from FLAC and Phase2 by softening the Youngs modulus from 60 to 0 GPa over ten stages is presented in Fig. 12a and b. The tensile strength is 5 MPa and the shear strength is set to a high value so that only tensile failure occurs in the model. The yield zones obtained from both codes are similar. Because the material softening generates a stress path which is dierent from the one-stage excavation, the tensile yield zone distribution shown in Fig. 12a or b is quite different compared to the result shown in Fig. 4a or b. 4. Conclusions For a given rock mass and in situ stress condition, excavation induced failure or damage to the rock mass depends on the stress path, which in turn depends on the excavation methods employed. The essence of this dependency is that the mechanical response of elasto-plastic materials is stress path dependent. FLAC and Phase2 are used to simulate a tunnel excavation problem assuming that the tunnel is excavated by

long-round drill and blast method. The yield zone distributions obtained by these two tools demonstrate a signicant dierence when the rock fails in tension and the tensile strength is low. This departure of analysis result by FLAC and Phase2 is attributed to the fact that one tool can handle the dynamic unloading by default while the other cannot. FLAC employs explicit solution scheme but Phase2 takes an implicit solution approach and the explicit formulation can follow the physical process automatically. Modeling of tunnel excavation problems using FLAC or Phase2 is not simply an exercise of building a model, deleting a group of elements in the model, letting the model runs, and producing colorful outputs when the calculation is nished, with the click of a few icons. The users need to have a thorough understanding of the theory behind the tools in order to obtain results that mimic the reality. The danger of using FLAC and Phase2 as black boxes for numerical modeling can be seen from the examples presented in this study. Because the tensile strength of the rock is low relative to its compressive strength, rocks can fail easily in tension than in compression. This failure phenomenon is magnied near the tunnel boundary where dynamic unloading induced tensile stress is high. When shear failure dominates, the maximum depths of yielding zone by the two numerical tools are comparable. Shallow shear failure due to dynamic unloading is also possible. Knowing such a dierence is important in interpreting obtained numerical results and choosing the right tool and modeling approach for a particular problem under investigation. The most notable conclusion from this study is not to judge whether FLAC is superior to Phase2 or vice versa, but to point out the importance of selecting the right tool and modeling approach to represent the expected stress path as close to reality as possible. Stress path changes in a tunnel not only exist over a large time span such as tunnel face advancing and near-by mining excavation but also show up in a very short time span such as instant removing of a rock block. The correct rock mass response can only be captured if the stress path is correctly represented in a model. Acknowledgement The author would like to thank Derek Martin of University of Alberta for his comments and suggestions during the preparation of the manuscript. References
Brady, T.M., Johnson, J.C., 1989. Comparison of a nite-dierence code to a nite-element code in modeling an excavation in an underground shaft pillar. In: Pietruszczak, S., Pande, G.N. (Eds.), Proceedings of Third International Symposium on Numerical Modelling in Geomechanics (NUMOG III), Niagara Falls, pp. 608619. Cai, M., Kaiser, P.K., Martin, C.D., 2001. Quantication of rock mass damage in underground excavations from microseismic event monitoring. Int. J. Rock. Mech. Min. Sci. 38, 11351145.

628

M. Cai / Tunnelling and Underground Space Technology 23 (2008) 618628 Itasca 2005. FLAC-Fast Lagrangian analysis of continua, 5. Itasca Consulting Group Inc. Kaiser, P.K., Yazici, S., Maloney, S., 2001. Mining-induced stress change and consequences of stress path on excavation stability A case study. Int. J. Rock Mech. Min. Sci. 38 (2), 167180. Martin, C.D., 1993. The strength of massive Lac du Bonnet granite around underground opening. Ph.D. thesis, University of Manitoba, p. 278. Nilson, R.H., Proer, W.J., Du, R.E., 1985. Modelling of gas driven fractures induced by propellant combustion within an explosion cavity. Int. J. Rock. Mech. Min. Sci. Geomech. Abstr. 22, 319. Rocscience Inc., 2004. Phase2, 5. Rocscience Inc., Toronto, <www.rocscience.com>. Ruistuen, H., Teufel, L.W., 1996. Analysis of the inuence of stress path on compressibility of weakly cemented sandstones using laboratory experiments and discrete particle models. In: Aubertin, M. et al. (Eds.), Proceedings of Second North American Rock Mech. Symposium, pp. 15251531.

Cai, M., Kaiser, P.K., Uno, H., Tasaka, Y., 2002. Inuence of stress-path on the stressstrain relations of jointed rocks. In: Lin, Y. et al. (Eds.), Second International Conference on New Development in Rock Mechanics and Rock Engineering, pp. 6065. Cai, M., Kaiser, P.K., Morioka, H., Minami, M., Maejima, T., Tasaka, Y., Kurose, H., 2007a. FLAC/PFC coupled numerical simulation of AE in large-scale underground excavations. Int. J. Rock. Mech. Min. Sci. 44 (4), 550564. Cai, M., Morioka, H., Kaiser, P.K., Tasaka, Y., Minami, M., Maejima, T., 2007b. Back analysis of rock mass strength parameters using AE monitoring data. Int. J. Rock Mech. Min. Sci. 44 (4), 538549. Coursen, D.L., 1979. Cavities and gas penetration from blasts in stresses rock with ooded joints. Acta Astron. 6, 341363. Dally, J.W., Fourney, W.L., Holloway, D.C., 1975. Inuence of containment of the bore hole pressures on explosive-induced fractures. Int. J. Rock. Mech. Min. Sci. Geomech. Abstr. 12, 512. Hoek, E., 2001. Big tunnels in bad rock, 2000 Terzaghi lecture. J. Gotech. Geoenviron. Eng., ASCE 127 (9), 726740.