09.01.2013 Views

Numerical methods in rock mechanics

Numerical methods in rock mechanics

Numerical methods in rock mechanics

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

416<br />

The hybrid FEM/BEM was first proposed <strong>in</strong> [238],<br />

then followed <strong>in</strong> [239,240] as a general stress analysis<br />

technique. In <strong>rock</strong> <strong>mechanics</strong>, it has been used ma<strong>in</strong>ly<br />

for simulat<strong>in</strong>g the mechanical behaviour of underground<br />

excavations, as reported <strong>in</strong> [241–245]. The<br />

coupl<strong>in</strong>g algorithms are also presented <strong>in</strong> detail <strong>in</strong> [48].<br />

The hybrid DEM/BEM model was implemented<br />

only for the explicit dist<strong>in</strong>ct element method, <strong>in</strong> the<br />

code group of UDEC and 3DEC. The technique<br />

was created <strong>in</strong> [246–248] for stress/deformation analysis.<br />

In [249–250] a development of hybrid discrete-cont<strong>in</strong>uum<br />

models was reported for coupled hydro-mechanical<br />

analysis of fractured <strong>rock</strong>s, us<strong>in</strong>g comb<strong>in</strong>ations<br />

of DEM, DFN and BEM approaches. In [251], a<br />

hybrid DEM/FEM model was described, <strong>in</strong> which<br />

the DEM region consists of rigid blocks and the<br />

FEM region can have non-l<strong>in</strong>ear material behaviour.<br />

A hybrid beam-BEM model was reported <strong>in</strong> [252]<br />

to simulate the support behaviour of underground<br />

open<strong>in</strong>gs, us<strong>in</strong>g the same pr<strong>in</strong>ciple as the hybrid BEM/<br />

FEM model. In [253], a hybrid BEM-characteristics<br />

method is described for non-l<strong>in</strong>ear analysis of <strong>rock</strong><br />

caverns.<br />

The hybrid models have many advantages, but special<br />

attention needs to be paid to the cont<strong>in</strong>uity or<br />

compatibility conditions at the <strong>in</strong>terfaces between<br />

regions of different models, particularly when different<br />

material assumptions are <strong>in</strong>volved, such as rigid and<br />

deformable block–region <strong>in</strong>terfaces.<br />

2.7. Neural networks<br />

L. J<strong>in</strong>g, J.A. Hudson / International Journal of Rock Mechanics & M<strong>in</strong><strong>in</strong>g Sciences 39 (2002) 409–427<br />

All the numerical modell<strong>in</strong>g <strong>methods</strong> described so far<br />

are <strong>in</strong> the category of ‘1:1 mapp<strong>in</strong>g’, i.e., the Level 1<br />

<strong>methods</strong> <strong>in</strong> Fig. 1. The neural network approach is a<br />

‘non-1:1 mapp<strong>in</strong>g’ <strong>in</strong> the Methods C and D categories of<br />

the Level 2 <strong>methods</strong> <strong>in</strong>dicated <strong>in</strong> Fig. 1. The <strong>rock</strong> mass is<br />

represented <strong>in</strong>directly by a system of connected nodes,<br />

but there is not necessarily any physical <strong>in</strong>terpretation of<br />

the geometrical or mechanistic location of the network’s<br />

<strong>in</strong>ternal nodes, nor of their <strong>in</strong>put and output values.<br />

Such a ‘non-1:1 mapp<strong>in</strong>g’ system has its advantages and<br />

disadvantages.<br />

The advantages of neural networks are that<br />

(1) The geometrical and physical constra<strong>in</strong>ts of the<br />

problem, which dom<strong>in</strong>ate the govern<strong>in</strong>g equations<br />

and constitutive laws when the 1:1 mapp<strong>in</strong>g<br />

techniques are used, are not such a problem.<br />

(2) Different k<strong>in</strong>ds of neural networks can be applied<br />

to a problem.<br />

(3) There is the possibility that the ‘perception’ we<br />

enjoy with the human bra<strong>in</strong> may be mimicked <strong>in</strong> the<br />

neural network, so that the programs can <strong>in</strong>corporate<br />

judgements based on empirical <strong>methods</strong> and<br />

experiences.<br />

The disadvantages are that<br />

(1) The procedure may be regarded as simply supercomplicated<br />

curve fitt<strong>in</strong>g—because the program has<br />

to be ‘taught’.<br />

(2) The model cannot reliably estimate outside its range<br />

of tra<strong>in</strong><strong>in</strong>g parameters.<br />

(3) Critical mechanisms might be omitted <strong>in</strong> the model<br />

tra<strong>in</strong><strong>in</strong>g.<br />

(4) There is a lack of any theoretical basis for<br />

verification and validation of the techniques and<br />

their outcomes.<br />

Neural network models provide descriptive and<br />

predictive capabilities and, for this reason, have been<br />

applied through the range of <strong>rock</strong> parameter identification<br />

and eng<strong>in</strong>eer<strong>in</strong>g activities. Recent published works<br />

on the application of neural networks to <strong>rock</strong> <strong>mechanics</strong><br />

and <strong>rock</strong> eng<strong>in</strong>eer<strong>in</strong>g <strong>in</strong>cludes the follow<strong>in</strong>g publications.<br />

* Stress–stra<strong>in</strong> curves for <strong>in</strong>tact <strong>rock</strong> [254].<br />

* Intact <strong>rock</strong> strength [255,256].<br />

* Fracture aperture [257].<br />

* Shear behaviour of fractures [258].<br />

* Rock fracture analysis [259].<br />

* Rock mass properties [260,261].<br />

* Microfractur<strong>in</strong>g process <strong>in</strong> <strong>rock</strong> [262].<br />

* Rock mass classification [263,264].<br />

* Displacements of <strong>rock</strong> slopes [265].<br />

* Tunnel bor<strong>in</strong>g mach<strong>in</strong>e performance [266].<br />

* Displacements and failure <strong>in</strong> tunnels [267,268].<br />

* Tunnel support [269,270].<br />

* Surface settlement due to tunnell<strong>in</strong>g [271].<br />

* Earthquake <strong>in</strong>formation analysis [272].<br />

* Rock eng<strong>in</strong>eer<strong>in</strong>g systems (RES) modell<strong>in</strong>g [273,274].<br />

* Rock eng<strong>in</strong>eer<strong>in</strong>g [275].<br />

* Overview of the subject [276].<br />

As evidenced by the list of highlighted references<br />

above, the neural network modell<strong>in</strong>g approach has been<br />

widely applied and is considered to have significant<br />

potential—because of its ‘non-1:1 mapp<strong>in</strong>g’ character<br />

and because it may be possible <strong>in</strong> the future for such<br />

networks to <strong>in</strong>clude creative ability, perception and<br />

judgement, and be l<strong>in</strong>ked to the Internet. However, the<br />

method has not yet provided an alternative to conventional<br />

modell<strong>in</strong>g, and it may be some time before it can<br />

be used <strong>in</strong> the comprehensive Box 2D mode envisaged <strong>in</strong><br />

Fig. 1 and described <strong>in</strong> [277].<br />

3. Coupled thermo-hydro-mechanical (THM) models<br />

The coupl<strong>in</strong>gs between the processes of heat transfer<br />

(T), fluid flow (H) and stress/deformation (M) <strong>in</strong><br />

fractured <strong>rock</strong>s have become an <strong>in</strong>creas<strong>in</strong>gly important<br />

subject s<strong>in</strong>ce the early 1980s [278,279], ma<strong>in</strong>ly due to the

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!