Some failures in tunnels constructed using the NATM have prompted discussion about the applicability of the method in certain situations – and sometimes heated debates on the NATM itself. In an address at a Geomechanics Colloquium in Salzburg, Professor Kovari of Switzerland even cast doubt the very existence of the NATM itself. There have been repeated calls by the Austrian tunnelling industry to universities and research institutions for more research in this area.

After a lengthy preparation phase, the Joint Research Initiative (JRI) entitled ‘Numerical Simulation in Tunnelling’, or SITU, was established in 1997 by the Austrian National Science fund. It involves research groups in three Austrian universities – Vienna, Graz and Innsbruck. The total budget over the five year life span of the project is over $1.8m.

The aim of the project is to develop new and improve existing computer simulation methods to help tunnel engineers better to understand ground behaviour during NATM advance and to provide software tools for prediction. The ultimate aim is that such tools are used routinely at the tunnel site.

‘Closed loop’ numerical simulation

Numerical simulation is an attempt to recreate real ground behaviour during tunnel advance by excavating a virtual tunnel which exists in a computer memory only. The advantages of this are obvious: many methods of advance can be tested and the ground behaviour observed; in the case of ‘failure’ of the virtual tunnel, the only loss would be computer time.

Computer simulations are, of course, widely used in the construction industry. One would not dream, for example, of constructing a bridge without performing extensive calculations and design work beforehand. With tunnelling the situation is different. Although in most cases computer analyses are carried out before construction starts, these are mainly two-dimensional simplifications of the actual three-dimensional problem, with a lot of guesswork involved in determining the input parameters. The reason is that tunnels are constructed in a given material, which is highly stressed (sometimes near failure) and whose mechanical properties are not well known.

Obviously, these properties depend greatly on the geological conditions and in many cases our information on geological conditions is patchy. Here we rely on results of exploration either through drilling or other methods such as seismics and geoelectrics. But even if we did have all the relevant geological data, we would still need to think about the way in which we input the material behaviour into a numerical model. These are the reasons why computer predictions are often wrong and unexpected ground behaviour is encountered during tunnel construction. It is one of the aims of the JRI to develop software which will assist the site engineer in this process.

Table 1 shows that input data for a numerical simulation can be divided into two sets: data that are determined by given conditions in the ground where the tunnel is to be constructed and the data that the tunnel designer supplies.

As mentioned previously data set 1 is usually associated with a degree of uncertainty.

The aim of the project is to develop a software tool that can be used at the site in two different ways:

First, in the Verification loop, a prediction is made with the available data and the predicted behaviour compared with observed behaviour (for example, displacement measurements). If the comparison is not good, input data set 1 can be modified and the analysis repeated. This method can also be used to make a prediction about geological conditions ahead of the tunnel face . Second is to modify the initial design. An example is a change in the method of advance or an increase in support to reduce settlements above shallow tunnels or to ensure the safe advance through highly squeezing rock.

Automated data acquisition

As mentioned previously, the knowledge of geological conditions is an essential pre-requisite for numerical simulation. It is possible to gather such information during the advance of the tunnel itself or from a pilot tunnel by observing geological features at the tunnel face at each advance. This has been standard practice in the NATM. It is desirable, however, for numerical simulation to have this information in digital form. It can then also be used to establish a complete numerical database of the tunnel for future reference. One project within the JRI is concerned with making this process automatic as much as possible.

After each tunnel advance a vehicle with two scanning video cameras mounted on each side is driven in front of the tunnel face and the surface is scanned. This takes only a few minutes. The data are either stored on disk or transferred directly via the internet to the office computer. The system is able to produce high resolution, true colour stereo images of the tunnel face in digital form. This information may then be viewed in 3-D and interpreted on a workstation by an experienced geologist using a geological editor.

Some logic which is able to detect simple joint structure automatically is built into the system to minimise the amount of work the geologist has to do. It is envisaged that this digital information concerning geological structures will be fed directly into the numerical model. First field trials of the system were carried out and results are encouraging.

Material models for numerical simulation

For the success of a numerical simulation it is essential to be able to describe accurately in a mathematical way the behaviour of the ground (soil/rock) and the shotcrete which is an integral part of the NATM. With respect to the description of ground behaviour, long established theories such a the Hoek/Brown and Mohr-Coulomb failure criteria have to be questioned.

Innovative ideas for describing the behaviour of jointed rock, for example, are damage tensor theories, which consider the effect of crack generation and movements of joints on the deformability of the rock mass. For shallow tunnels in soft ground, it has been found that small strain stiffness has to be considered in order to obtain realistic settlement throughs. In some cases, the generation of shear bands preceding failure has to be considered also. In the case of shotcrete, the complex chemical, thermal and mechanical processes that occur during and after placing have not been investigated in enough detail in the past. To ascertain the safety of the tunnel or to predict surface settlements this knowledge is essential.

Groundwater very often plays an important role, and where compressed air is used, the ability to predict air loss through the shotcrete and the tunnel face is required.

Our first site experiences

Numerical simulation was applied at two tunnel sites in Austria – the Siberg railway tunnel and the Semmering road tunnel. On the latter, the effect of the advance method (full and staged top heading advance) on the surface settlements was investigated. The simulation showed that an appreciable reduction in settlements could be achieved with a staged top heading advance.

The site experiences can be summarised as follows: for shallow tunnels it is essential to perform 3-D analysis to obtain realistic settlements. 3-D can be carried out on site within a reasonable time so that they can affect the decision making process. The time required to prepare the input data was half a day and the numerical simulation of up to 45 excavation stages was performed on a standard site PC overnight. The accuracy of predictions has still to be improved, for example by using better material models for the ground and shotcrete.

User friendly interfaces and visualisation

If one wants to ‘sell’ new technology to tunnelling engineers, especially on site, it is imperative that the tools are easy to use and that results are understandable. A major JRI effort has been in user friendly interfaces and visualisation tools. The information obtained from numerical simulation software is 3-D and include settlement troughs; tunnel wall displacements; stresses; etc.

It is therefore obvious that 3-D visualisation techniques are used, including virtual reality. If photo realistic textures are used for the tunnel walls, observers will actually feel they are in a real tunnel and the engineer may perform a virtual walk through the tunnel. In contrast to a real tunnel, the observer may, however, move freely inside the rock mass, observing geological features, damaged zones in the rock, stresses etc. A particular challenge is to display simulation results in 3-D. In the JRI we are experimenting with innovative methods to display such complex information. One example is the display of stress concentrations as a fog which increases in density as the stress increases.

Related Files
Shows how data from the designer are used to change the initial design
Verification loop comparing predicted behaviour with that observed