Practical Hints for Using Numerical Methods in

Rock Mechanics
Autor: Prof. Dr. habil. Heinz Konietzky, (TU Bergakademie Freiberg, Geotechnical Insti-
tute)

1 Introduction.............................................................................................................2
2 Initial- and Boundary Conditions .............................................................................2
3 Meshing Rules ........................................................................................................4
4 Meshing Techniques ..............................................................................................5
5 Model Size ..............................................................................................................7
6 Continuum versus Discontinuum Scale Effects ...................................................8
7 2D versus 3D ........................................................................................................10
8 Specifics for Simulation of Dynamic Processes....................................................11
9 Mesh dependency in the post-failure region .........................................................13
10 Parallel computing .............................................................................................15
11 Choice of numerical method / code ...................................................................16
12 Conceptual and numerical model ......................................................................18
13 Important terms..................................................................................................24
14 Current and future trends ...................................................................................25
15 Literature............................................................................................................25

Editor: Prof. Dr.-Ing. habil. Heinz Konietzky Layout: Angela Griebsch

TU Bergakademie Freiberg, Institut fr Geotechnik, Gustav-Zeuner-Strae 1, 09599 Freiberg [email protected]
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

1 Introduction
Although numerical software products become more and more easy to handle and
find broad acceptance and use in engineering practice as well as in geo-engineering
sciences, careful choice and use of these powerful tools are necessary to avoid
wrong calculation results. Quite a lot of aspects have to be considered, like:

Choice of appropriate numerical technique and tool

Initial and boundary conditions
Appropriate model size
Choice of appropriate constitutive models and parameters
Meshing (structure, density, element type) and mesh-dependency
2D versus 3D
Coupling (hydro-thermal-mechanical)
Modelling sequence
Continuum versus Discontinuum approach
Calculation efficiency versus accuracy
Static versus dynamic simulations

Also, different solution schemes (implizite versus explizite), different element-types,

different couplings etc. often need application of special simulation procedures.

Therefore, careful inspection of model set-up and in-depth analysis of simulation re-
sults is necessary. Comparison with other methods and experience or measurements
is strongly recommended.

2 Initial- and Boundary Conditions

The solution of differential equations (both, analytical and numerical) requires the
specification of initial and/or boundary conditions. Boundary conditions describe time-
independent or time-dependent mechanical, thermal, hydraulic, chemical condi-
tions, which are applied to inner or outer boundaries. Initial conditions describe me-
chanical, thermal, hydraulic or chemical conditions at the begin of the simulation
(at zero point in time) either at the boundaries and/or the inner model area.

Boundary conditions (examplary):

stresses
forces
velocities
accelerations
displacements
water pressures
temperatures

Page 2 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Initial conditions (examplary):

primary stress state
primary deformation state
primary pore water pressure / joint water pressure
initial temperature distribution

Displacement boundary conditions are also called Dirichlet conditions, force and
stress conditions, however, are called Neumann conditions.

For static calculations in geomechanics, depending on the application and modelling

task, displacement and stress boundaries are applied. In most cases (not always!)
displacement boundary conditions deliver too small deformations inside the model
and stress boundary conditions over-predict deformations. However, if boundaries
are far enough away from the interesting inner model area both types of boundary
conditions deliver similar results (results converge with increasing model size).
If analytical rough estimates or practical experience are not available, a preceding
parameter study shall be performed to investigate the influence of model size and
boundary conditions on the results of the specific model.

Fig. 1: Normalized stresses and displacements at two observation points in dependence of the type
of boundary condition and the ratio of model size to excavation size [ITASCA 2011]

Exemplary, Fig. 1 demonstrates, how displacement and stress values inside the
model with two excavations alter in relation to model size. Furthermore, this figure
Page 3 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

illustrates the influence of the displacement and stress boundary conditions on the
results in comparison to the exact analytical solution. One recognizes, that with in-
creasing distance between the boundaries and the inner model area both types of
boundary conditions converge and come finally close to the analytical solution.
Boundary conditions are normally applied in form of normal and/or tangential compo-
nents or in Cartesian coordinates. Also, over different regions of the boundary differ-
ent types of boundary conditions can be applied.

3 Meshing Rules
For 2- and 3-dimensional discretization (meshing) of objects, three fundamental as-
pects should be considered:
Choice of appropriate element type
Appropriate mesh density
Choice of appropriate meshing technique

In principle, it can be distinguished between the following element types:

Volume elements (e. g. triangular or rectangular elements in 2D and tetrahe-

dral or squared elements in 3D) typical for rock mass or massive concrete
Shell elements (planar elements with negligible thickness, but explicit consid-
eration of moments and membrane stresses) typical for shotcrete and thin
masonry or concrete walls
Bar elements (1-dimensional elements) typical for anchors, piles or struts

Furthermore, for one element type, different shape functions can be applied, which
implies different interpolation functions and finally leads to different accuracy (non-
linear interpolation).

Two competitive demands have to be considered in respect to mesh density

(gridpoint distance):

Mesh density : this leads to improved resolution and higher accuracy, but:
this leads, on the other side, to increasing calculation time and storage de-
mand

For meshing the following general practical rules are guilty:

Several elements of low order (lineare shape function) give equivalent results
compared to less number of elements of higher order (non-linear shape func-
tion)
Refined meshing is necessary, whenever high stress and deformation gradi-
ents are expected, that means for instance at free surfaces, areas with high
stiffness contrasts, areas with support elements or areas with load entry points

Page 4 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Elements should be designed preferably equidistant, meaning the ratio of

maximum edge length to minimum edge length should be smaller than about
10, also very small angles between edges should be avoided
Elements should preferably aligned according to the expected stress trajecto-
ries
Transition between coarse and fine meshed areas should be smooth (prefera-
bly without discrete jumps in element size)
In general: Quad-elements gives more accurate results than triangular-shaped
elements

4 Meshing Techniques
In a broader sense, meshing comprises two phases: set up of geometrical model and
subsequent filling with elements (meshing).

A distinction is made between:

free meshing: delivers an unstructured mesh, which just satisfies a few gen-
eral user defined criteria (e. g. max. edge length pre-defined value)
mapped meshing: delivers a structured mesh, e. g. adjusted according to the
model geometry and/or expected stress trajectories

The following conventional meshing techniques are used:

Generation of geometry (e. g. CAD-based with Boolean Algebra) and subse-
quent automatic meshing via free or mapped meshing
Generation via primitives: compilation via several basic bodies filled with ele-
ments (already meshed volumes)
Deformation of a basic mesh and extension by copying / extrusion / mirroring
Set-up of model (geometry + mesh) via imported point, edge and element data

Today, 2D meshing with different techniques is unproblematic, performed fast and

nearly fully automatic. However, 3D meshing, especially with top-quality cuboidal el-
ements and complicated geometry, like for example tunnel crossings, are still compli-
cated to perform, quite time consuming and often manual interaction is necessary
during the meshing process.

For 3D models in most cases the following procedure is applied:

Generation of geometrical model with a CAD-Program
Export of CAD model data into a standard format (e.g. IGES, STEP, STL etc.)
Import of CAD geometry into a meshing tool, execution of meshing and export
of mesh into a format, which can be used as input for the numerical simulation
tool (solver)
Input of mesh into the numerical simulation tool and solving the problem

Page 5 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Several of the numerical simulation tools have already integrated CAD tools and
mesher.

The following Figures 2 5 illustrate exemplary some of the above mentioned mesh-
ing techniques.

Fig. 1: With zones filled basic volumes, which can be composed to a final model (LEGO-system),
(ITASCA 2012)

Fig. 2: Assemblage of subnets via docking and mirroring

Page 6 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Fig. 3: Generation of final mesh via trimming, deformation and partial deletion of virigin (initial) mesh

Abb. 4: Generation of final mesh by graduel geometrical adaption and mesh refinement

5 Model Size
The ratio between model size (overall dimension of numerical mesh) and object size
(e.g. dimension of excavation, pillar, slope etc.) plays an important role in respect to
correct simulations (see also DGGT 2014). An optimum of two competitive require-
ments have to be found (see also chapter Meshing rules):

Preferably large ratio model size / object size to minimize boundary influence
Preferably small ratio model size / object size to minimize calculation time and
memory space.

Page 7 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

A second aspect, especially important for settlement and uplift prediction, is the prob-
lem, that overall model size has a non-vanishing influence of these deformation val-
ues. In case of large overall model dimension elastic or simple elasto-plastic models
would lead to unrealistic high deformation values. To avoid these problems either
depth-dependent stiffness values or so-called small-strain stiffness has to be includ-
ed.

6 Continuum versus Discontinuum Scale Effects

At a certain scale all geomaterials and construction materials should be considered
as discontinua, e.g. sand in form of grains or a fractured rock mass in form of rock
blocks as well as composite constructions like tunnel lining / rockmass, anchor /
rockmass, geotextile / ballast or pile / wall.

Whether discontinuum modelling shall be applied or not, depends on three factors:

Ratio of representativ elementary volume (REV) to total volume

Modelling task (interesting phenomena)
Availability of parameters
Availability of software

REV is defined as the smallest volume, from which measurements, parameters and
object reactions, respectively, can be obtained and which are representative for the
whole rock mass. That means: the REV is the smallest continuum mechanical vol-
ume, which is statistical equivalent to the in-situ discontinuum.

Fig. 6 shows the history of the two quantities npm and nfr as function of considered
volume. One recognizes, that the spread of the quantity decreases with increasing
volume (e.g. strength values, deformation module, permeability etc.) und finally the
quantity converges and reaches a nearly constant value.

Beyond a specific volume Vmin, which represents the REV, the quantity does not
change any more significantly even over a bigger volume towards Vmax. Like Fig. 6
also shows, the area of Vmin - Vmax can be quite different for different physical quanti-
ties. If several quantities have to be considered simultaneously the common intersec-
tion (overlap) have to be considered. In such a case Rfp hast to be considered as
REV.

Page 8 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Fig. 1: Parameter history as function of considered volume (scale effect).

intact rock

single joint set

two joint sets

many joint sets

heavy fractured
rock mass

Fig. 2: Relation between structure and considered volume using the example of a rock mass (scale
effect).

Depending on the considered volume the rock mass structure shall be modelled in
different ways (Fig. 7). In case of very small volumes, it is possible that only intact
rock exist and therefore a classical continuum mechanical approach with rock-
mechanical parameters is recommended. At bigger volumes several single joints or
joint sets are observed, which can be handled best with a discontinuum mechanical
approach (e.g. Discrete Element Method), where different material laws and parame-
ters are assigned to rock matrix and discontinuities in an explicit manner. If a huge
number of joints or joint sets exist (highly fractured rock mass), it make sense, to go
back to a continuum mechanical approach, where the effect of the discontinuities is
not directly (discrete) considered, but in a smeared manner by reduced strength and

Page 9 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

stiffness parameters. But, due to increased computer power today even complex
DFNs can be modelled via DEM-techniques (e.g. Sainsbury 2015).
The scale effect is highly important for rockmasses and has to be considered careful-
ly whenever constitutive laws and corresponding parameter are chosen. Rockmass
classification schemes can be used to deduce appropriate rockmass parameters
based on rockmechanical lab data and engineering-geological investigations accord-
ing to the following scheme:

Determination of rockmechanical parameters in the lab

Engineering-geological inspection (core analysis, joint-set statistics etc.)
Rockmass classification (e.g. RMR, Q, GSI, RQD) on the basis of the engi-
neering-geological inspection and the rockmechanical lab data
Deduction of rockmass parameters for suitable constitutive laws (e.g. Hoek-
Brown, Mohr-Coulomb etc.) on the basis of the rockmass classification

7 2D versus 3D
In the forefront of any numerical model set-up the following key questions should be
answered:
Is a 3-dimensional modelling necessary or is a 2-dimensional consideration
acceptable?
Are there symmetry lines or planes which would allow to reduce the model
size?

A complete 3-dimensional modelling is necessary, if:

the strike of the geological elements (layer, joint, fault, bedding plane ) does
not coincide with the long axis of the geotechnical construction
the axes material anisotropy do not coincide with the axes of the geotechnical
construction
the directions of principal stresses are neither parallel nor perpendicular to the
axes of the geotechnical construction
the dimensions of the geotechnical constructions and the geological bodies,
respectively, are nearly equal in the three spatial directions
several structural components intersect each other (e.g. tunnel crossings)
applied boundary conditions (forces, velocities, stresses .) have components
in all 3 spatial directions

If symmetry conditions are met, models may be reduced (simplified) to half or quarter
models or even reduced from 3D to 2D (axisymmetrical). However, one has to take
into account, that symmetry conditions have to be referred to several aspects, which
have to be met simultaneously:

in respect to the geometry

Page 10 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

in respect to the stress field

in respect to the constitutive law (planes of anisotropy, orientation of joints
etc.)
in respect to the modelling sequenz (construction process, excavation phases,
support measures etc.)
in respect to boundary conditions
in respect to support installation- and preservation measures (anchors, shot-
crete, piles, struts etc.)
in respect to couplings (hydraulic, thermal etc.)

Fig. 8 illustrates potential model reductions on the basis of pure geometrical consid-
erations, like used to simulate shaft- , tunnel- and borehole problems or building ex-
cavations.

Fig. 1: Examples for model reduction on the basis of the consideraion of pure geometrical symmetry
conditions (full model, half model and quarter model)

8 Specifics for Simulation of Dynamic Processes

The simulation of dynamic processes involves wave propagation and requires the
consideration of four important aspects:

The correct representation of movement of waves requires that the maximum

gridpoint distance hp,s has to be considerable smaller than the shortest wave
length p,s:

E 1
c (1 + )(1 )
hP 10 P= 10 S= (1)
f f
G
c
hS 10 S= 10 S= (2)
f f

hP: maximum gridpoint distance for P-wave (longitudinal wave / compression wave)
hS: maximum gridpoint distance for S-wave (transverse wave / shear wave)
cp: P-wave velocity

Page 11 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

cs: S-wave velocity

f: Frequency
E: Youngs modulus
: Poissons ratio
Model boundaries shall be designed in such a way, that reflections are avoid-
ed to a large extent (application of anechoic boundary conditions).
As a general rule, dynamic strength parameters are higher than static values
(e.g. factor of app. 1,5). Therefore, strength parameters have to be adjusted
for the dynamic calculation part.
The numerical algorithm has to include physical correct damping for wave
propagation. Often applied algorithms are:

o mass-proportional damping
o stiffness-proportional damping
o local damping
o Viscous damping
o Rayleigh damping
o Intrinsic damping (damping due to plastifications)

The damping behavior can be described by the seismic quality factor Q. The seismic
quality factor is frequency independent and effects increasing damping with increas-
ing frequency, which is characteristic for geomaterials. This requirement is satisfied
by the local damping scheme and, for a broader frequency range, also for the popu-
lar Rayleigh damping (Fig. 9).

= (3)

where:

f: frequency [Hz]
c: wave speed [m/s]
: damping coefficient [m-1]

Page 12 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Damping

local
Rayleigh

stiffness proportional

mass proportional
Frequency

Fig. 1: Damping vs. frequency for different damping procedure

The logical consequence is, that normally dynamic meshes shall have much higher
resolution (much smaller gridpoint distances). This results in larger computation
times.
Also, often it is necessary to filter the dynamic input signal to suppress higher fre-
quencies and to perform baseline-correction.

9 Mesh dependency in the post-failure region

The post-failure behavior is characterized by so-called localization. Localization
means the progressive accumulation (concentration) of microcracks within a small
macroscopic fracture plane (zone, band). This process is numerically represented by
the so-called strain softening, which is dependent on the grid structure (Fig. 10). That
means, that the stress-deformation behavior inside the post-failure region is heavily
dependent on the mesh resolution. In principal, there are three different solutions to
overcome this problem:
Calibration on a fixed grid structure (mesh resolution should be constant for
any model using one calibrated data set)
Adaptive re-meshing inside the localization (mesh refinement)
Extension of the constitutive model by internal scaling parameter (regulariza-
tion)

d
c,

ab c d
b,
a,

Fig.1: Stress-strain behavior with strain-softening and different mesh refinement a,b,c and d without
any procedure to avoid influence of mesh dependency

Page 13 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Similar behavior is sometimes observed, when static problems with pronounced plas-
tifications are calculated by using explicit methods and loading or unloading occurs
instantaneously. Therefore, in such cases properties, loads or geometrical changes
should be performed gradually. Exemplary, Fig. 12 shows displacement magnitudes
along the tunnel contour for a model according to Fig. 11 with isotropic virgin stress
field, but anisotropic material behavior with complex plastification pattern. As Fig. 12
illustrates, coarse mesh (here: 900 zones) does not deliver correct values, but mesh-
es with number of zones around 14000 or bigger give nearly identical results. Also,
without soft relaxation of boundary stresses at the tunnel contour (see dashed line in
Fig. 12) wrong displacements are obtained. That statement, that appropriate choice
of elements and mesh density is essential to obtain reasonable results is supported
by a lot of studies, e.g. Batoz et al. (1980), Jovanovic et al. (2010), Ljustina et al.
(2014) or Veyhl et al. (2010).

Fig. 2: Mesh for -symmetric tunnel including boundary conditions (900 zones)

Fig. 12: Displacement magnitudes along tunnel contour vs. location of observation point

Page 14 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

10 Parallel computing
Huge numerical models (huge number of elements, zones, blocks, gridpoints etc.)
demand very long runtimes (days up to weeks). Therefore, parallelizing also called
high performance computing (= usage of several processors in parallel) is sugges-
tive. Parallelization can be realized in different ways: e.g. by hyper- and multi-
threading (shared memory computing) or physical partitioning of the model to several
processors (at best this could be several processors at one board, but also several
computers within a network or cluster distributed memory computing). Shared
memory computer reach maximum speed, if all processors on board are used. Fur-
ther speed increase can be reached if hybrid parallel computing is applied (combina-
tion of shared and distributed memory methods).
However, the calculation speed does not linearly increase with number of processors,
but the calculation speed follows Amdahls law:

1
Sm = (4)
fp
fs +
N
where:
Sm factor of speed increase
N number of processors
fp share of parallelized code
fs share of serialized code

It holds fs + fp = 1, where fp < 1, because at least the communication between proces-

sors has to be performed in a serial manner. Fig. 13 illustrates Amdahls law.

Sm

N
Fig. 13: Amdahls law: Increase in claculation speed Sm as fuction of number of processors N

To evaluate the efficiency of parallelization, a measure of efficiency can be defined

(Fig. 14):

Ts
=
N Tn

where:
Ts calculation time using 1 processor
Tn calculation time using N processors
N number of processors

Page 15 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

1 number of zones

N
Fig. 14: Calculation efficiency as function of number of processors N and number of zones

The efficiency of parallelization increases with increasing number of zones and de-
creases with rising number of processors.

11 Choice of numerical method / code

Numerical simulation techniques can be distinguished either according to:

temporal discretization into explicit and implicit methods

spatial discretization into mesh-based and meshless methods

Whereas implicit methods are faster than explicit methods in case of elastic simula-
tions, explicit methods are better suited whenever large deformations, strong non-
linearities or physical instabilities occur (e.g. crash or impact simulations, blasting,
rock cutting or drilling, rockfall etc.)

In respect to spatial discretization of mesh-based methods one can further distinguish

Finite Element (FE, XFE), Finite Difference (FD), Volume Element (VE) and Bounda-
ry Element (BE) methods. The most popular representatives of meshless methods
are the Discrete Element Methods (DEM) based on spheres or polyhedra and the
Smooth Particle Hydrodynamics (SPH) method.

Mesh-based methods can be distinguished into integral (BEM) and differential (FEM,
FDM, XFEM, VEM) methods. Integral methods have the following general character-
istics:

simple meshing (only at surfaces)

exact far-field solution
minimized discretization error
no problems with singularities
small demand of computational power
but: problems to handle anisotropies, inhomogeneities, couplings, non-
linearities

Differential methods have the following general characteristics:

Page 16 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

ability to depict any kind of anisotropies, inhomogeneities, couplings, non-

linearities
tremendous flexibility to solve continuum mechanical problems
higher demand for computational power
sometimes complicated meshing procedure
problems to simulate disintegration and mixture of material

Compared to classical integral or differential methods, meshless methods including

DEM are characterized by several additional features:

discrete elements should be able to follow all kinematic possible movements

(displacements and rotations)
automatic contact detection and elimination of contacts (e.g. via cell-space or
Verlet algorithm)
contact laws, which act, whenever a contact (direct touch or long-ranging in-
teraction) exist

Due to this additional features, especially the automatic contact detection algorithms,
these approaches are quite computational intensive, but offer unique simulation pos-
sibilities whenever disintegration, mixture or flow of material, mass movement, blast-
ing, rock cutting and drilling, rockfall or micro-mechanical problems at the grain size
level are of interest (e.g. Stahl & Konietzky 2011, Lunow & Konietzky 2009, Groh et
al. 2011, Wang & Konietzky 2009, Lisjak & Grasselli 2014 ). Fig. 15 gives an impres-
sion about the power of DEM methods to simulate desintegration processes.

All of the above mentioned techniques have advantageous and disadvantageous

considering a specific modelling tasks. Also, a lot of problems can be solved with dif-
ferent techniques with nearly same quality and efficiency.

Currently under development are hybrid codes, which combine mesh-based and
meshless approaches.

Besides general purpose programs (e.g. Ansys, Abaqus, Nastran, Comsol etc.) spe-
cial designed codes for rock- and soil-mechanical simulations were developed, which
have the advantage, that appropriate constitutive models and structural elements (for
simulating anchors, piles, walls etc.) are already included. Also, these codes have
participated in a lot of benchmarks and validation procedures (e.g. DECOVALEX).

Page 17 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Fig. 15: DEM Simulation of cutting process for brittle rock (Lunow & Konietzky 2009)

12 Conceptual and numerical model

Each numerical modelling should be separated into 2 successive phases:

1. Phase: Conceptual model

2. Phase: Numerical model

The first phase comprises the development of the general modeling strategy and in-
cludes some more general decisions about simulation tools and approaches.

The second phase comprises the programming, specific allocation of constitutive

laws, initial and boundary conditions, parameters etc. and the corresponding model
runs including evaluation.

Phase 1: Conceptual model:

Phase 1 starts with an analysis of the modelling task and the available data base fol-
lowed by the allocation to the corresponding phase and finally the detailed analysis of
all aspects according to Fig. 16.

Each geotechnical task can be subdivided into several stages, like:

Pre-planning
Dimensioning
Construction
Monitoring
Backanalysis

Besides typical geotechnical project work, applied or fundamental geotechnical re-

search work exist.
Depending on the project phase, tasks, demands and applications, numerical simula-
tions will be quite different.

Page 18 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

In the pre-planning phase quite often only a very limited data base exist, the budget
is limited and the expectations about precision in prediction are lower. Therefore, in
this phase numerical modelling is restricted to simplified geometries, coarser mesh-
es, simple constitutive laws and consideration of less construction stages. The pre-
planning phase also includes the comparison of different concepts. The aim of this
phase is to detect the main geomechanical features / characteristics / pitfalls etc. of
the project, to get the correct order of magnitude in terms of stresses and defor-
mations and to obtain a deeper understanding of the geomechanical processes. Also
different construction methods can be compared and evaluated.

The phase of dimensioning / detailed planning requires a valuable and comprehen-

sive data base. Material behavior and interaction between rockmass and structure
have to be described by appropriate constitutive laws in detail.

The construction phase is characterized by application of numerical modelling in par-

allel to the in-situ construction steps. During that phase the model has to be up-
dated, corrected, improved etc. based on actual in-situ measurements and observa-
tions. The model is used to interpret / explain observations and can be used to pre-
dict the effect of short-termed changes during the construction or to perform backa-
nalysis in case of any disaster (failure).

The phase of monitoring and backanalysis normally starts, when the construction is
finished and characterized by the fact, that in-situ measurement data and observa-
tions in respect to the interaction rockmass construction are available. The aim of
modelling during these phases is the back-calculation of parameters or to analyze
failure situations.

The approach to perform fundamental or applied research can be quite different and
is highly dependent on the modeling task (often innovative approaches).

If one has defined the corresponding phase, several aspects according to Fig. 16
have to be considered. Especially, the following questions have to be discussed:

Should the problem be modelled 3-dimensional or is a 2-dimensional ap-

proach sufficient or is the assumption of axisymmetry (quasi-3D) acceptable?
If so, how can symmetry lines / planes be defined? Please note, that in re-
spect to symmetry not only geometry has to be considered, but also other fac-
tors like stress state, anisotropy of material etc.
What is most appropriate way to solve the modelling task: a continuum ap-
proach or a discontinuum approach? This is decisive for the choice of corre-
sponding software tools.
What type of constitutive model is appropriate (elastic, elasto-plastic, visco-
elasto-plastic, fracture mechanical, damage mechanical )?
Can the simulation be performed as pure mechanical one or should couplings
be considered, e.g. hydro-mechanical, thermo-mechanical, hydro-thermo-
mechanical, hydro-thermal-mechanical-chemical )?
Do we have a pure static incl. quasi-static problem or is it necessary to in-
clude dynamic effects (wave propagation etc.)?

Page 19 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Which construction stages (excavation stages, installation of support

measures etc.) have to be considered, to take into account the stress path
dependency of the material behavior and structural engineering require-
ments?
What are the principal initial and boundary conditions (initial stresses, pore
water pressures, stress and /or displacement boundary conditions, static vs.
dynamic boundary conditions etc.)?

The answering of all questions / aspects according to Fig. 16 allows to formulate the
Conceptual Model.

Page 20 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Analysis of task and date base

Phases Aspects

2D or 3D symmetry
Pre-planning
lines/planes

Continuum vs.
Dimensioning
discontinuum

Construction Constitutive law

Monitoring ? Couplings (HTMC)

Backanalysis Static vs. dynamic

Calculation sequence
Applied research
boundary conditions

Fundamental research Support elements

Fig. 16: Conceptual model

Page 21 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Phase 2: Numerical model:

In a second step the Numerical Model (Fig. 17) hast to be set-up, which means to
perform the detailed programming including specification of initial and boundary con-
ditions, meshing, parameter specification etc.

The second phase starts with a detailed analysis of the data base and demands also
thinking ahead, how and which model results should be obtained and documented.
After that analysis, the model set-up starts with definition of the mesh, the initial and
boundary conditions, the calculation sequence, the specification of constitutive laws
und their parameters etc. in form of an input script or a menu-driven dialog. In case of
any umbiguity / problem or to test the behavior, it is helpful to set up small models (in
an extreme case a 1-zone-model) and to perform test simulations until the problem is
exhausted.
Then, the numerical simulation will be started and the model results will be stored for
further evaluation. However, before reporting is started, the modelling results have to
be checked carefully. Considering that, the following methods are available:

Plausibility check: that means to check, whether the calculated physical val-
ues are generally feasible (are they within a plausible range? is the general
deformation, stress and failure pattern logical?)
Comparison with experience: that means, are the model results located inside
the field of experience and if not, can they be explained logical on a physical
basis?
Direct comparison with measurements and / or observations in-situ. This ap-
proach is always the best choice and should always be used.
Comparison with other calculation methods, either alternative numerical simu-
lation approaches or semi-analytical solutions. This is already a demand at
least for projects of special importance.

If the check of modelling results is positive, either the project can be finished by writ-
ing a report or further simulations follow, e.g. in form of a parameter study, sensitivity
analysis, robustness analysis, optimization, comparison between variants etc.
If the check of the modelling results is negative, one has to check if it is a more prin-
cipal conceptual error (e.g. effect of water not considered, continuum approach not
able to duplicate significant discontinuum effects etc.) or a more detailed error inside
the numerical model (e.g. wrong parameter, wrong initial condition etc.) Depending
on this evaluation one has to jump back inside the scheme, has to make corrections
and then to execute all subsequent steps again.

It is strongly recommended to perform geotechnical simulations always in parallel to

the construction (project in-situ), that means starting already with the pre-planning
phase until the phase of monitoring and backanalysis. Finally, this leads to more
economic design, allows optimization and the chance to react short-termed to prob-
lematic situations, e.g. collapse. If such a strategy is applied, the numerical model will
be step-by-step modified and improved (adjusted) according to the actual observa-
tions and measurement results. This, again, allows more and more precise predic-
tions.

Page 22 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Conceptual Model

Detailed data analysis and

Small test models
conceptual idea of reporting

Definition
(i) model geometry and mesh structure
(ii) boundary conditions
(iii) initial coniditions
(iv) constitutive law and parameters
(v) calculation sequence

Numerical model

Numerical calculations according to simulation sequences

Check
(i) Plausibility
(ii) experience
(iii) measurements / observations
(iv) comparison with other methods

Reporting Additional calculations

Fig. 17: Numerical Model

Page 23 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

13 Important terms

Verification (verify = make good)

Process to proof the correct mathematical-physical implementation of the desired

algorithms / models etc. (in most cases done by comparison with analytical solutions
or proofed numerical solutions often called Benchmarks).

Validation (validate = become valid)

Process to determine, to what degree the underlying model represents reality in a
correct manner under chosen perspective (in most cases performed by comparison
with observations and measurements in-situ or in the lab).

reality
ana
lysis
simulation

modell conceptual
validation
model
ing
ramm
g
pro
numerical
model
model
verification

Fig.17: Role of verification and validation within the framework of simulation and software development

Calibration:
Process of adjustment model parameters in such a way, that measured values are
reproduced in a satisfying manner. The prerequisit is successful verification and vali-
dation. Calibration can be achieved by trial-and-error procedure, by mathematical
based backanalysis on the basis of special in-situ or labor tests or by mathematical
based optimization.

Sensitivity analysis:
This analysis investigates the sensitivity of the model output as function of varying
input parameters. This can be performed by a parameter study or in a more sophisti-
cated and effective manner by stochastic sampling with statistical evaluation.

Parameter studies:
The model is run systematically with different parameter sets. The model response is
evaluated as function of input parameters.

Uncertainty analysis:
Probabilistic modelling to determine the influence of fuzziness (range of variation) of
input parameters on the model response.

Page 24 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Robustness analysis:
Probabilistic modelling to determine the robustness (stability) of the model response
as function of varying (fluctuating) input values.

Reliability analysis:
This analysis investigates border violations (limit state violations) of the system be-
havior. The probability of failure is the quotient of number of model runs with failure to
total number of model runs.

14 Current and future trends

There is an ongoing development of numerical simulation techniques. Some of the
very active current research fields are listed below:

Multi-scale modelling
High performance computing
Automatic coupling of meshless and meshfree methods
3D-visualization within caves
Integration of numerical models into GIS
Coupling with optimization tools
Integration of time-dependent and time-independent damage and fracture me-
chanical approaches into classical elasto-plastic ones
Sophisticated HTMC-coupling

15 Literature

Batoz, J.-L. et al. (1980): A study of the three-node triangular plate bending ele-
ments, Int. J. Numerical Methods in Engineering, 15: 1771-1812

DGGT (2014): Empfehlungen des Arbeitskreises Numerik in der Geotechnik, Ernst &
Sohn

Groh, U. et al. (2011): Damage simulation of brittle heterogeneous materials at the

grain size level,Theoretical and Applied Fracture Mechanics, 55: 31-38

Itasca (2011): FLAC Manuals, Itasca Consulting Group, Inc., Minneapolis, Minnesota,
USA

Itasca (2012): FLAC3D Manuals, Itasca Consulting Group, Inc., Minneapolis, Minne-
sota, USA

Jovanovic, M. et al. (2010): Accuracy of the FEM analysis in the function of the finite
element type selection, Mechanical Engineering, 8(1): 1-8

Lisjak, A. & Grasselli, G. (2014). A review of discrete modelling techniques for fractur-
ing processes in discontinuous rock mass, Int. J. Rock Mech. Min. Sci., 6: 301-314

Page 25 of 26
 Practical Hints for Using Numerical Methods in Rock Mechanics
Only for private and internal use! Updated: 30 June 2015

Ljustina, G. et al. (2014): Rate sensitive continuum damage model sand mesh de-
pendence in finite element analysis, The Scientific World Journal, ID 260571

Lunow, Ch. & Konietzky, H. (2009): Two dimensional simulation of the pressing and
the cutting rock destruction., Proc. 2nd Int. Conf. on Computational Methods in Tun-
neling, Aedificatio Publishers, 1: 223-230

Sainsbury, D. et al. (2015): The use of Synthetic Rock Mass (SRM) modelling tech-
niques to investigate joint rock mass strength and deformation behavior
http://share.hydrofrac.wikispaces.net/file/view/Sainsbury_SRM.pdf (19.06.2015)

Stahl, M. & Konietzky, H., (2011): Discrete element simulation of ballast and gravel
under special consideration of grain-shape, grain size and relative density, Granular
Matter, 4: 417-428

Veyhl, C et al. (2010): On the mesh dependency of non-linear mechanical finite ele-
ment analysis, Finite Elements in Analysis and Design, 46: 371-378

Wang, Z.L. & Konietzky, H., (2009): Modelling of blast-induced fractures in jointed
rock mass, Engineering Fracture Mechanics, 76: 1945-1955.

Page 26 of 26