Study On Data Fusion Of Fiber Optical Bragg Grating Sensing

With engineering applications of fiber bragg gratings and widely use of multiplexingtechnology of FBGs, signals of fiber gratings sensing array and fiber gratings networkbecome more and more complicated. Sensing signals processing now becomes one ofbottlenecks which restrict large-scale applications of fiber grating sensors. During theengineering applications of fiber grating sensors, interference, nosie and cross-sensitivity andso on result in measurement error. To reduce those error, package and reference grating arewidely used, those methods take good effects at the same time increase the cost of applicationand make the fiber grating sensor system complicated, and also go against large-scaleapplications of fiber grating sensorsIn this paper we study the characteristics of fiber grating sensors, apply data fusiontechnology to fiber sensing data to enhance the measurement accurancy. This method makesfull use of structural feature of demodulation at terminal, and also makes full use of sensorsinformation. At the same time changing data fusion algorithm is much easier than changingthe structure of sensing system, the cost is decreased effectively, and it fits for large-scaleapplications.This paper was supported by National Natural Science Foundation No.61107052, themain contents of this paper are as following:(1) Introduce the fundamental of fiber grating sensing and the development of itsapplications, illustrate the fiber gratings multiplexing technology and its demodulationtechnology. And then, fundamental and research contents of data fusion technology arestated, after this, classification method and important algorithms are discussed.(2) Analyse the peculiarity of outliers of grating sensing and common outlier algorithm,propose an new method to deal outliers with statistic algorithm. this method makes fulluse of sensors’data and take low algorithmic complexity, it is easy to implement, andconvenience to process data fusion later.(3) Improve the proposed outlier algorithm based on statistics, take weighted trimmed meanalgorithm to deal with outliers, and then integrate the outlier algorithm with weightedminimum mean square algorithm as a new data fusion algorithm, this algorithm processtwo-times fusion, first time fusion to deal with outliers, second time fusion to get optimalfusion result. Experimental result shows this algorithm can restrict outliers’effects to datafusion, output variance0.0768is much less than direct fusion.(4) According as the peaks shift of the TFBG core mode and cladding modes are changeddifferently when TFBG used by measuring strain, a method using RBF neural network toprocess data fusion to estimate the strain are proposed. Experimental result shows thismethod is easy to apply，its relative error is about1.24×103%and have a very good prospects.