Research On Fiber Optic Sensing Demodulation And Structural Reverse Design Method Based On Neural Network Algorithm

Fiber optic sensing and demodulation algorithms are vital in developing highperformance structural health monitoring systems.However,most related FBG and fiberoptic FPI sensing and demodulation algorithms rely on complex and expensive devices.They have become difficult to apply in practical applications with increasing demands for scalable deployments,high-accuracy measurements,and fast response times in terms of computational complexity and reusability.In addition,the low efficiency and high computational consumption of fiber optic structure design based on numerical simulation methods have seriously hindered the development of high-sensitivity fibers for enhancing the performance of structural health monitoring systems.Starting from the above problems,this thesis’ s research content and main results are summarized as follows.(1)For the problems of high cost,complicated calculation,and limited range of FBG demodulation system,an FBG wavelength demodulation system based on array waveguide grating and artificial neural network is proposed and experimentally demonstrated.The array waveguide grating converts the sensing signal to a light-intensity signal.The nonlinear relationship between intensity and FBG peak wavelength is modeled using a neural network model.The proposed system automatically selects the array grating channels for demodulation based on the wavelength shift of the FBG sensor,with a demodulation precision of ±5.672 pm,and the demodulation range can be extended to 40 nm without any additional hardware cost.The demodulation method based on longperiod gratings and generation of adversarial networks is proposed to achieve highperformance absolute wavelength demodulation with a precision of ±6 pm for small data sets.(2)To address the problem that the demodulation of fiber optic Fabry-Perot interferometric sensors often relies on low multiplexing and complex demodulation algorithms,a fiber optic Fabry-Perot interferometric sensor demodulation system based on array waveguide gratings and neural networks is proposed and experimentally demonstrated.Considering that expensive and cumbersome spectral analyzers limit most demodulation systems,we propose a neural network-based fiber-optic Fabry-Perot interferometric sensor demodulation system with a high precision of ±14 pm for wavelength and ±0.07 μm for cavity length.A neural network-based reconstruction framework for interferometric spectra of fiber-optic Fabry-Perot sensors is proposed,which can achieve the high-accuracy reconstruction of interferometric spectra in the range of more than 30 nm based on extremely sparse sampling points to achieve efficient demodulation.(3)To address the inefficiency of numerical methods for fiber optic design,a neural network-based framework for inverse design of fiber optic structures is proposed,which can provide efficient optimization performance of fiber optic-related optical properties without considering the potential physical constraints involved.In addition,considering the effects of data scarcity,one-sided design knowledge,and low computational resources on the comprehensiveness of structural design,a collaborative optical property optimization framework is proposed to achieve cross-institutional structural inverse design without data interaction.This method consumes one-third of the resources of the numerical method and takes less than 1 second to perform,providing efficient and accurate optimization and prediction of optical properties.