Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SEVENTH INTERNATIONAL CONFERENCE ON THE APPLICATION OF ARTIFICIAL INTELLIGENCE TO CIVIL AND STRUCTURAL ENGINEERING
Edited by: B.H.V. Topping
Structural Response Extraction from Sound Vibration Measurements using Neural Networks
N. Bourahla, R. Taleb and T. Boukhemacha
Department of Civil Engineering, University of Blida, Algeria
N. Bourahla, R. Taleb, T. Boukhemacha, "Structural Response Extraction from Sound Vibration Measurements using Neural Networks", in B.H.V. Topping, (Editor), "Proceedings of the Seventh International Conference on the Application of Artificial Intelligence to Civil and Structural Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 63, 2003. doi:10.4203/ccp.78.63
Keywords: neural network, sound vibration, dynamic analysis, noise cancellation, dynamic testing.
In earthquake engineering accurate structural response measurements are of paramount importance for successful investigation of the behaviour of physical models. However in most cases, dynamic testing yields huge amount of data containing valuable information in clustery form which needs tedious processing. In strictly controlled conditions, measurements can be of high quality which makes the data reduction easier. Yet, the processing and interpretation phase is deemed to be a difficult task particularly for nonlinear behaviour of test specimens . However, in situation where the measured response amplitudes are very low such as in most cases of ambient vibration testing where the signals may be highly altered by extraneous noise, then the processing task becomes more demanding . This paper discusses the potential use of neural network and its application to the problems of data processing in seismic testing. For this purpose a procedure of extracting structural response parameters of simple specimens from acoustic vibration signals recorded by standard PC microphones with sound cards using neural networks is presented.
The test specimens are six cables having different diameters attached at their ends to form fixed ended spans. Damped free vibrations were initiated by shifting the mid-span from its equilibrium position and releasing it. In the test condition described in this paper, the acoustic wave propagating from the specimen to the microphone is highly altered by the nonlinear nature of the transmitting media and the background noise. Therefore the relationship between the recorded sound signals and the corresponding acceleration time-history responses is very complex. Elman Feedforward backpropagation neural networks were adopted to approximate this relationship. The neural network was trained using a data base composed of recorded sound signals (input) and acceleration time-histories numerically calculated at a specific point of the specimen (target signals). The effectiveness of the neural network model is investigated using a wide variety of specimens.
The approach is interesting in situation in which fast demonstration for educational purposes are needed. It can be easily adapted to other applications of wave propagation through undefined media for which additional accuracy or specific structural response extraction is sought.
purchase the full-text of this paper (price £20)