Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 92
Edited by: B.H.V. Topping and Y. Tsompanakis
Paper 29

The use of Artificial Neural Networks for the Estimation of the Effect of the Geology on Air Losses during Compressed Air Tunnelling

R. Boubou, F. Emeriault and R. Kastner

Lyon university, INSA - Lyon, LGCIE, France

Full Bibliographic Reference for this paper
R. Boubou, F. Emeriault, R. Kastner, "The use of Artificial Neural Networks for the Estimation of the Effect of the Geology on Air Losses during Compressed Air Tunnelling", in B.H.V. Topping, Y. Tsompanakis, (Editors), "Proceedings of the First International Conference on Soft Computing Technology in Civil, Structural and Environmental Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 29, 2009. doi:10.4203/ccp.92.29
Keywords: tunnel, excavation, compressed air tunnelling, air loss, air pressure, artificial neural network, geology.

The use of the compressed air technique in tunnelling operations can ensure the tunnel face stability and therefore minimize the ground settlement. Tunnelling and soil parameters are the essential factors required to determine the ideal air pressure to be applied inside the tunnel.

This paper presents an artificial neural network (ANN) methodology to predict the air losses in compressed air tunnelling taking into account some geological and geometrical parameters. Data from the excavation of Contract 3 of the Toulouse subway line B tunnel are used to evaluate the performance of the proposed method. This contract consists of three different tunnels, excavated by the same tunnel boring machine (TBM), almost parallel to each other and approximately at the same depth.

Considering as input parameters the applied air pressure and the respective area represented by the five different types of soils observed by the TBM pilot, an ANN model is employed to predict the measured air loss along the tunnel drive. Two different ways of selecting the data used to train and validate the ANN have been analysed and compared with measurements recorded on the three tubes. These two cases prove the ability of the ANN model to predict the air loss; better results are nevertheless obtained with the random data distribution.

Ordered data distribution corresponds to a rather operational way of training the ANN: data obtained during the excavation of a first part of tunnel drive are used for training; the resulting ANN is then employed for predicting the air loss for the remaining part of the tunnel drive. It appears that this procedure can be used with homogenous conditions and therefore it can help to improve the TBM performance.

The random selection of the training data improves the prediction results: the geological parameters of the training data are close to those of the validation while with ordered data selection, data from the first part of the tunnel drive can be very different from those of the second part.

Using the trained ANN, it is possible to determine the relation between air loss and the applied air pressure for the most frequent soil profiles encountered. It appears that for clayey profiles (100% or 90% of the face area), larger air loss is induced by the increase of the applied air pressure. For a 100% sandy profile, no clear relationship is obtained probably because additional data is missing in the TBM pilot reports: foam is used in this case to stabilize the tunnel face during excavation and reduce the air loss that could result from the permeability of sand.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £78 +P&P)