Modal analysis using the Bessel harmonics of an extrinsic Fabry-Perot interferometric sensor (EFPI) and neural networks

Abstract

A demodulation system employing neural networks is used to process the non-linear signal from an extrinsic Fabry-Perot interferometric (EFPI) sensor. A sinusoidal strain is theoretically shown to produce well-defined Bessel harmonics in the EFPI signal. The neural network demodulator (NND) uses a Fourier Series Neural Network to separate the Bessel harmonic components of the EFPI signal and a Back-Propagation Neural Network is used to predict the strain levels through the analysis of the Bessel harmonics. The NND is first simulated in a computer program and then actually employed in an experimental setting to determine the frequency response of a 25 cm composite cantilever beam. A function generator was used to drive a PZT actuator attached to the composite beam and resulting periodic strain was measured by the EFPI; the frequency of the composite beam was varied between 10 Hz and 900 Hz. The NND demodulated the EFPI signal and determined the frequency response of the composite beam. The results show that the NND accurately reproduced the natural frequencies and mode shapes of the cantilever beam.