| Fiber optic sensors have advantages such as resistance to electromagnetic interference,small size,light weight,and resistance to high temperature and pressure.Coupled with fiber optic demodulators,they are widely used in fields such as nuclear pipelines,aerospace engines,and mechanical processing,such as real-time monitoring of pressure,strain,and vibration parameters.The monitoring of pressure and other parameters in the fields of nuclear field pipelines and aerospace engines may require a collection rate of tens of kHz,which puts high demands on the speed of fiber optic demodulators.In order to be compatible with demodulation of various sensors such as fiber Fabry Perot and fiber Bragg grating,this thesis develops a full spectrum demodulation system.In order to achieve high-frequency dynamic real-time monitoring,the system needs to solve the problem of massive spectral processing.In order to achieve a fast and high-precision full spectrum demodulation system,the main work and contributions of this thesis are as follows:1.In order to reduce the noise of the system and improve the quality of the signal output,this thesis uses the Singular value decomposition denoising algorithm to denoise the demodulated signal.Simulation and experiments show that Singular value decomposition denoising can achieve full frequency band denoising,and the actual signal-to-noise ratio is improved by 9.983 dB while retaining the main characteristics of the demodulation signal.In order to solve the frequency aliasing problem of traditional wavelets,an anti aliasing single subband reconstruction filter was designed and implemented.Simulation and experiments have shown that this filter can effectively suppress harmonic signals.And compared to traditional wavelets,it has more accurate subband extraction characteristics and is more in line with the wavelet theoretical frequency band.Its pass stop band attenuation can reach up to 29.42 dB,with a transition band of 0.00377π,and has stricter phase characteristics than FIR/IIR filters.Subsequently,combining the advantages of Singular value decomposition denoising algorithm and anti aliasing single subband reconstruction,the subbands of steady signal and high-frequency vibration signal are accurately extracted,while the base noise is suppressed,and the signal-to-noise ratio is increased by 3.47 dB and 4.2307 dB respectively.2.In order to achieve fast demodulation and handle massive spectral problems,this thesis conducts a parallel acceleration research on the demodulation algorithm of all optical fiber sensors based on the GPU-CUDA platform.In order to solve the problems of traditional FIR filter,such as time-consuming and low filtering efficiency,the idea of overlapping reservation subsection convolution and Convolution theorem is used to divide the spectrum into blocks for parallel processing,which greatly improves the spectral filtering speed,and the maximum acceleration ratio is 107.96.In addition,in response to the time-consuming problem of spectrum refinement transformation algorithms and least squares fitting peak finding algorithms,the algorithm was optimized by parallel processing of a single spectrum with multiple threads,with the highest acceleration ratios reaching 200 and 38.59,respectively.Finally,three demodulation techniques,wavelength demodulation,single peak demodulation,and spectrum transformation demodulation,were parallelized,and the speed advantage of GPU parallel demodulation compared to serial and concurrent demodulation on CPU was verified through experiments.The computational speed of the three parallel demodulation processes has increased by 15.34,57.82,and 77.04 times,respectively.Finally,a realtime demodulation system based on CPU+GPU architecture was built,achieving a 12 channels real-time demodulation rate of over 10 kHz. |
PDF Full Text Request