Research On Key Techniques Of Fiberbragg Grating Spectrum Demodulation Forfiber Sensing Networks

With the continuous development of optical fiber sensing technology,the Fiber Bragg Grating(FBG)sensor network composed of multiple multiplexing modes is widely used in areas requiring multi-point monitoring,such as health detection of large engineering structures and perimeter security.However,with the increase of sensor multiplexing number and monitoring range of the FBG sensor network,spectral signals are prone to overlap and distortion,which reduces the demodulation accuracy of the FBG sensor network.Therefore,higher requirements are put forward for the precision of FBG spectral demodulation technology.The introduction of spectral shape multiplexing technology increases the overlap probability of the FBG reflectional spectrum.However,the demodulation accuracy of the current commonly used demodulation methods on the overlapping spectrum is limited,so the difficulty of accurate demodulation of the FBG overlapping spectrum restricts the multiplexing capability of the FBG sensor network.In addition,due to the influence of device performance and noise,the reflection spectrum of FBG will be distorted,resulting in a decrease in the detection accuracy of the FBG sensor network.In this context,the corresponding spectral demodulation method is designed to improve the multiplexing capability of the FBG sensor network.The main contents of this thesis include:Firstly,aiming at the cross-talk problem caused by spectral overlap,this thesis firstly constructed a spectral intelligent demodulation system with flexible bandwidth allocation by considering fault repair,spectral demodulation,and bandwidth allocation.Based on these,the Continuous Wavelet Transformation-Particle Swarm Optimization(CWT-PSO)algorithm is used to demodulate FBG overlapping spectra.The spectral signals are classified and demodulated according to the degree of spectral overlap.CWT-PSO algorithm can improve the demodulation rate and precision by reducing the particle search space.The results show that the demodulation time of the Particle Swarm Optimization(PSO)algorithm,Dynamic Particle Swarm Optimization(DPSO)algorithm,and CWT-PSO algorithm is 5.4 s,3.9 s,and 1.5 s respectively,and the demodulation rate of the CWT-PSO algorithm is faster when the demodulation has 8spectral signals with partial overlap and complete overlap.The root-mean-square errors of the PSO algorithm,DPSO algorithm,and CWT-PSO method are respectively 21 pm,3.4 pm,and 1 pm at the completely overlapping spectrum.CWT-PSO algorithm has higher demodulation accuracy and is not easy to fall into local optimal.Secondly,because of the defects of the traditional Gaussian fitting algorithm which is seriously affected by spectral types,this thesis first designs an FBG distorted spectrum classification algorithm based on a one-dimensional convolutional neural network to identify the spectral distortion types.On this basis,the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN-WTD)baseline correction algorithm is designed.The Gaussian white noise in the signal is eliminated and the baseline drift correction is achieved.The simulation results show that the accuracy of the distortion spectrum classification algorithm based on a one-dimensional convolutional neural network reaches 100%,and the loss value is close to 0.When the baseline slope of the CEEMDAN-WTD correction algorithm is 1.3 and the sub-peak threshold is 0.5,the signal survival rate can be improved by 66.7%.When the baseline slope is 0.25,the average demodulation accuracy of the empirical mode decomposition correction algorithm and CEEMDAN-WTD correction algorithm is 1.47 pm and 0.37 pm,respectively.The correction effect of the CEEMDAN-WTD correction algorithm is better and more stable.