Damped Vibration Analysis of Extrinsic Fabry-Perot Interferometric Sensors using Artificial Neural Networks

Health monitoring of a structure entails regular strain sensing. Vibrational strain, characterized as functions of damped sinusoids, is a typical case of strain that can act on a structure. Past research has developed a demodulation technique, employing artificial neural networks (ANN) as the processing element, for extrinsic Fabry-Perot interferometric (EFPI) sensors, attached to a vibrating structure, exposed to un-damped sinusoidal strain. The work employed two ANN to perform the demodulation. The first ANN was trained to extract the harmonic content from the EFPI modulated output and the second ANN was trained to predict the maximum strain acting, from the predicted harmonic content, during a vibration event. This project extends the study to a damped sinusoidal strain acting on the sensor. The ANN demodulation system predicts the maximum strain level from the spectral content of the sensor output, during a vibration event. Instead of employing an ANN to extract the spectral content, as done in the past research, simple fast Fourier transforms (FFT) is used. This paper develops the demodulation technique using computer simulations. Results are presented for different ANN architectures employed. An algorithm fusion system is presented that shows an improved accuracy in maximum strain prediction.