Neural processing-type fiber-optic strain sensor

The authors present a neural-processing-type strain sensor insensitive to thermal variation and describe calibration of the device through modulation of the internal parameters of the processing system. The sensor exploits the variation of the far-field polarization pattern in a single-mode birefringent fiber under the influence of longitudinal strain. A temperature-compensating fiber element is built in, making the sensor assembly immune to thermal variation. Sampling of the sensor output and parallel distributed processing of the samples are integrated with the sensor. The processor contains both a training function and a generalization function. The training function modulates a small linear network built into the system. In the working phase, the generalization function is used to recover the measurement information. Provided the sensor is thermally compensated, the network gives the reading of the measurand with an error not exceeding 0.1%