Research On Key Technologies Of Few-mode Fiber Based Distributed Curvature Sensing

Due to its small size,lightweight and easy integration,optical fiber based shape sensors have become the critical technologies for high-precision medical equipment and robots,as well as in aviation and aerospace fields.Existing optical fiber shape sensors are generally based on quasi-distributed bending sensing using optical fiber grating.Due to the limitation of the optical fiber grating preparation process,the number of monitoring points is limited,and distributed bending measurement cannot be realized.Therefore,distributed bending sensors have significant research value as a critical technology to realize long-distance high-precision shape sensing.The existing distributed bending sensors utilize a sandwich structure or a multi-core optical fiber to increase the bending induced strain of the optical fiber.Then the curvature is obtained by measuring the distributed Brillouin frequency shift of the optical fiber.Although these solutions can improve the bending induced strain,the strain of the optical fiber is subject to both bending magnitude and direction.It means that strain of multiple optical fibers at different locations must be measured simultaneously to calculate the actual bending magnitude.And due to the limitation of spatial resolution,it is necessary to ensure that the optical fiber does not undergo rapid twisting.Otherwise,the bending information cannot be precisely retrieved.These shortcomings limit their use in practical applications.This thesis proposes and demonstrates a novel distributed bending sensing scheme.This scheme uses the Brillouin optical time domain analyzer to measure the Brillouin frequency shift caused by the bending of a few-mode fiber under the quasi-single mode state.Few-mode fiber has a larger mode field than a single mode fiber and is,therefore,more sensitive to bending.Besides,compared to multimode fiber,the few-mode fiber operating in the quasi-single mode can avoid high-order mode interference.Using this scheme,a distributed measurement of the radius of a multi-layer cylinder is achieved,and the surface shape of the object is restored.Compared with the existing distributed fiber bending sensors,this solution has the following advantages:it is insensitive to the bending direction,easy to access the standard single-mode fiber system,and has high measurement accuracy for the case of a small bending radius.To solve the temperature and strain cross-sensitive problem of Brillouin frequency shift,the spontaneous Raman scattered light in the few-mode fiber is measured simultaneously.And the wavelet denoising algorithm is used to improve the accuracy of Raman temperature measurement.The worst resolution of the square of fiber curvature is0.1047 cm^-2 while the temperature resolution is 1.308°C at the end of a 2 km few-mode fiber.To further improve the bending measurement accuracy,the image processing algorithm based on convolutional neural networks?CNN?is studied.The CNN is trained using simulated Brillouin signal and the actual noise of the system.During the training process,the CNN realizes the recognition and separating of Brillouin signals and noise by continuously adjusting the network parameters.After training,the CNN is applied to the optical time domain analyser signal,leading to a signal-to-noise ratio improvement of about13 dB.Compared with the traditional denoising algorithm,the proposed algorithm has little side effect on the effective information of the signal.Besides,its processing time is much shorter than traditional algorithms and the acquisition time of Brillouin optical time domain analyzer,which provides a good foundation for real-time bending measurement.