A 3 sensor multipoint optical fibre water sensor utilising artificial neural network pattern recognition

Abstract

A multipoint sensor on a 1 km continuous length of fibre has been investigated and proven to be capable of detecting the presence of air, ethanol and water at each of three independent sensing points using OTDR techniques. Artificial neural network signal processing techniques have allowed the resulting OTDR signals to be accurately determined using pattern recognition. Each of the U-bend evanescent wave absorption sensors were developed with 62.5 /spl mu/m polymer-clad silica fibre, which had its cladding removed in the sensing region. Although the length of the fibre used in this investigation was 1 km, longer or shorter lengths may be used as required. Earlier results from a single U-bend sensor have shown that a multilayer perceptron is required to adequately classify the data. Initial results have shown that it is possible to train a network to recognise trends such as ageing of the bare fibre when immersed in water, and therefore possible to separate out such effects from genuine changes in the measurand. It is envisaged that a more sophisticated multipoint U-bend evanescent wave sensor system will be developed, with the resulting complex signals being processed using Artificial neural network pattern recognition techniques. This will result in the development of a 'smart system', with the ability to interpret and separate relevant measurand data from the data received from cross coupling signals from external or interfering parameters as well as faults or defects detected in the fibre.