A Neural Network Pattern Recognition Approach to Addressing a U-bend Optical Evanescent Wave Fibre Distributed Sensor

摘要

An optical fibre U-Bend evanescent wave absorption sensor system is reported which is capable of detecting the concentration of alcohol in a solution. The sensor is based on a continuous 250 meter 200μm core diameter Polymer Clad Silicon (P.C.S.) fibre which has had its cladding removed in the sensing area, so that the core is directly exposed to the measurand. An optical time domain reflectometer (OTDR) is used to address the sensing fibre, and is thus capable of resolving distance along its length allowing measurement at multiple points on a single fibre loop. Due to cross-coupling effects of interfering parameters, signals arising from optical fibre sensors can often be complex in nature and this is particularly so in the case of distributed sensors. It is difficult to interpret data from such systems using conventional detection techniques. The signal analysis for the sensor is performed using Artificial Neural Network pattern recognition techniques, which allow classification of the samples under test, thus allowing the true measurand to be recognised. Results are included that have been obtained from the sensors OTDR data. Also presented, are the resulting test outputs that have been obtained from a trained feedforward neural network designed to interpret the sensor data. The system was 100% successful in classification of all test samples analysed.