Automatic calibration of a fiber-optic strain sensor using a self-learning system

This paper describes a fiber-optic strain sensor and a procedure for automatic calibration as applied to the measurement of longitudinal strain. The sensor exploits variation in the intermodal interference pattern in a few-mode birefringent fiber, producing a far-field light distribution varying with the measurand. An array of photovoltaic diodes carries out sampling of the sensor output. A small-size connectionist network integrated within the sensor computes strain values from samples, dealing with the implicit, nonlinear dependencies between the parameter and the sampling data. The automatic calibration method is based on the principle of self-learning. It involves supervised sampling, optimal selection of training inputs, and automated modulation of weights in the neural processor. The method aims at a processor which recombines the photodiode signal into a function fitting the measurand uniformly in the measurement range