Keywords: neural networks, data compression, principal component analysis, soil-structure interaction, mining tremors.

Mine-related underground shocks excite seismic waves that reach the surface of the earth and induce building vibrations. Although these tremors are strictly connected with the human activity, they differ considerably from other paraseismic vibrations. They are not subject to human control and they are random events with respect to the time, place and magnitude likewise earthquakes. The strongest rockbursts may cause some damages to buildings.

Comparison of a huge number of records of vibrations (accelerations and velocities) measured at the same time on the ground and on the building foundation level leads to conclusion that they differ significantly. Additionally the evaluation of rockburst transmissions to the building is very difficult because of their ambiguous nature. The evaluation of the precise relation between ground and foundation records of accelerations as well as velocities is not possible.

The more precise estimation of the harmfulness of the mine-induced vibrations to actual buildings can be performed on the basis of the foundation vibrations. With respect to the fact that in many cases for example in the design procedure of new structures as well as in the dynamic analysis of existing buildings, the measured ground vibrations are only the accessible, the prediction of foundation vibrations is necessary.

Taking into account the difficulties in the soil-structure interaction analysis in the case of vibrations induced by mining tremors, the application of neural networks for the prediction of building foundation vibrations on the basis of ground vibrations taken from measurements is proposed. Results from measurements in situ on the ground level and on the building foundation were used as the neural networks patterns. The influence of mining tremors parameters such as the mining tremor energy and the epicentral distance on the soil-structure interaction effect also is taken into consideration.

The application of vibration records in the time domain leads to some computational difficulties related to the "size" of the data. Hence the pre-processing (compression) of the experimental data using principal component analysis is proposed. The analysis carried out leads to conclusion that the introduction of the principal component analysis method enables us to compress vibration data corresponding to the ground records of accelerations as well as velocities. Because of the strong correlation of the successive values of accelerations and velocities, the full ground acceleration and velocity vibrations in the time domain could be compressed to the first principal components only. Hence the design of considerably smaller neural networks for the prediction of building foundation vibrations on the basis of ground vibrations taken from measurements is possible.

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 £125 +P&P)