Measurement for monitoring, probing and recording evidence - Tunnel Engineering: Basics and Additional Services for Design and Construction - page 146

ANNs are particularly suitable for the analysis of data in tunnelling due to their capability
of completely independently learning, from example data sets, non-linear relationships
of any complexity, such as could no longer be comprehended by an expert. The prob-
lem with conventional solutions, that methods for extensive problems can only be found
with difficulty, does not occur because no mathematical formulation has to be undertaken.
Suspected rules and relationships in data fields can thus be easily tested or determined.
Tunnelling situations can be simulated with networks that have been trained once and the
effects on the surroundings can be determined. Furthermore, comparison of forecast with
newly recorded data can be used to control the tunnel drive as with statistical analysis and
can give warnings about changes in the ground or the condition of the machine.
Neuro-fuzzy. Neuro-fuzzy denotes the combination of artificial neural networks with fuzzy
technology. The combination of the two methods offers several advantages, which are
now briefly described.
- The influence of individual input parameters and the relationship between them can be
determined using an ANN so that the input parameters required for a fuzzy system can
be reduced to fewer but characteristic parameters.
- The result of the learning process is a normally controllably fuzzy system with a meth-
od of working, which in contrast to the ANN is understandable. This avoids the “black
box” behaviour of an ANN and the result can be manually optimised.
- In contrast to conventional fuzzy systems, it is no longer necessary to develop each rule
laboriously by hand and optimise it until the desired result is achieved. Particularly for
large systems with many input variables and correspondingly many rules, this can rapidly
become confusing. The system can learn the rules independently from example data.
Neuro-fuzzy models are predestined for the evaluation of tunnelling data since the advan-
tages of both systems are combined and also enable automatic process control through
connections to conventional control systems. For the present example of settlement analy-
sis, the high number of input quantities can be selected with the ANN and then passed to
a fuzzy controller for process control.
4.2.5 IT systems for the recording and evaluation of geotechnical data
IT is particularly useful both for data transmission and for the evaluation and monitoring
of geotechnical data. Quite apart from the rapid recording, the systematics, continuity
and comparability with rapid access back to previous measurements can be mentioned
as advantages. Systems have therefore been developed in recent years for the automatic
recording, processing and visualisation of the measured data.
Even with such ample information being made available on the computer screens of contrac-
tors, site supervision, employers and specialist consultants, the staff responsible for site man-
agement and supervision still has to make decisions on site. Particularly modern tunnelling
procedures, with the measurements being part of the verification of structural safety, demand
evaluation and making of decisions on site. Fig. 4-10 shows an example of a system for
automatic recording of measured data in real time (in conventional tunnelling) with graphi-
cal processing and saving of the data. This is accompanied by simple calculations like, for
example, temperature compensations for the measured readings. The data can be transmitted
by modem to external locations (site supervision, specialist consultant) [254].