DSpace Repository

Distinction between processable and unprocessable seismic records using neural network and wavelet techniques

Show simple item record

dc.contributor.author Hoogenboezem C en
dc.date.accessioned 2016-09-22T07:17:25Z
dc.date.available 2016-09-22T07:17:25Z
dc.date.submitted 1998 en
dc.identifier.uri http://hdl.handle.net/20.500.11892/11327
dc.description.abstract Many lives are lost because of rock bursts in mines. If the rock-bursts can be predicted in time, these lives could be saved. To predict rock-bursts, it is necessary to have a good knowledge of the behaviour of the rock mass. Seismologists can build various models, describing the rock mass, using the information gathered by recording the seismic activity of the rock mass. To improve the accuracy of the models, more seismic information is required. This can be achieved by improving the sensitivity of the monitoring system so that the smaller seismic events can also be recorded. The more sensitive the system, however, the more false records are recorded. This results in more manual processing of the data, since the automatic processing routines will misinterpret the false data. If these false records can be removed automatically, the whole processing procedure can be automated. A study was conducted on techniques to automatically distinguish between true (usable) and false (unusable) records. The techniques evaluated for this purpose were wavelet transform techniques for feature extraction and neural network techniques for classification. Techniques that were evaluated and compared included the sum-of-squares of the raw data, some modified continuous wavelet transforms, the discrete wavelet transform, some visually defined feature extraction techniques and statistical moments. For classification the feed forward neural network was evaluated using different training algorithms. Mutual information theory was applied to reduce the number of features used for classification. Mutual information combined with statistical moments of some of the wavelet transforms produced a small number of features that still achieved comparatively good results. The overall success rate of the best combinations was in the order of 80% while the best results achieved were 84% and 83% for two different combinations. en
dc.language English en
dc.subject Electrical and Electronic engineering en
dc.title Distinction between processable and unprocessable seismic records using neural network and wavelet techniques en
dc.type Masters degree en
dc.description.degree MIng en

Files in this item

Files Size Format View

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record