Application of self-organizing feature maps for diagnostics of vibroacoustic systems

Failure diagnostics for the structures suffered vibrations in acoustic frequency range is presented. Normalized spectral characteristics of structure response measured in checkpoints are used as indicators to be analyzed. Self-organizing feature maps (Kohonen networks), for which output variables are not required, detect faults. Simultaneous application of different networks duplicating each other makes it possible to improve the quality of recognition.. Principal component analysis is employed to reduce the number of variables under study. It extracted few latent variables that explained approximately all the set of observed ones. An aircraft panel with different combinations of attached defective dynamic suppressors is considered to demonstrate features of the approach. Tests demonstrated high effectiveness of the presented way of recognition and showed the advantages of neural networks over cluster analysis in recognition problems.