Vibration sensing method based on distributed optical fiber sound wave sensing system

Abstract

The invention discloses a vibration sensing method based on a distributed optical fiber sound wave sensing system, which comprises the following steps of: injecting continuous light waves into a sensing optical fiber, and applying dynamic polarization disturbance to obtain a polarization state evolution sequence of Rayleigh backscattered light; performing differential processing on the polarization state sequence to extract polarization state variation caused by vibration; and inputting the signal subjected to differential processing into a pre-trained deep neural network model, and directly outputting vibration distribution information along the sensing optical fiber by the model, including vibration position, frequency and amplitude. The system comprises a narrow linewidth laser source module, a polarization modulation and disturbance module, a circulator, a sensing optical fiber, a polarization detection module and an intelligent signal processing module. According to the method, the problem of fading noise in traditional phase-sensitive distributed optical fiber sensing is solved by utilizing polarization state difference and deep learning.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing technology, and in particular to a vibration sensing method based on a distributed fiber optic acoustic wave sensing system. Background Technology

[0002] Distributed fiber optic acoustic sensors are a novel sensing technology that uses optical fiber itself as the sensing medium to detect and locate vibration, acoustic waves, or dynamic strain signals at any location along the fiber in real time. Due to their outstanding advantages such as resistance to electromagnetic interference, corrosion resistance, ease of networking, and long monitoring distance, they have been widely used in perimeter security, oil and gas pipeline monitoring, power cable monitoring, rail transit, and geological structure health monitoring.

[0003] Currently, mainstream distributed fiber optic acoustic sensing technology is primarily based on phase-sensitive optical time-domain reflectometers (OTDRs). These systems sense external vibrations by injecting coherent optical pulses into the sensing fiber and detecting changes in the phase or intensity of the backscattered Rayleigh light. However, this technology inherently suffers from several insurmountable drawbacks: due to the coherence of the laser, the Rayleigh backscattered light undergoes random interference during its return journey, resulting in random maxima (bright fringes) and minima (dark fringes) of the detected light intensity at different locations within the fiber—a phenomenon known as coherent fading. At these fading points, the signal-to-noise ratio (SNR) drops sharply, severely degrading or even completely disabling vibration detection sensitivity and creating a detection blind zone. Furthermore, improving spatial resolution requires shortening the incident pulse width, but this reduces pulse energy, resulting in extremely weak backscattered signals at distant points and a poorer SNR. Conversely, increasing the sensing distance necessitates increasing the pulse width, but this sacrifices spatial resolution. This inherent contradiction severely limits its application in scenarios requiring high-resolution, long-distance monitoring. Existing technologies (such as CN110375841A) propose using a combination of swept-frequency pulses and unmatched filtering with Rayleigh graph correlation to suppress fading noise. While this approach has some effectiveness, its technical path still relies on the traditional framework of "pulse emission" and "backscattered light intensity / phase detection," and its back-end processing still depends on traditional graph correlation algorithms. This limits its ability to identify complex vibration modes, and there is still significant room for improvement in system performance and intelligence. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention utilizes polarization state difference and deep learning to overcome the fading noise problem in traditional phase-sensitive distributed fiber optic sensing.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution: a vibration sensing method based on a distributed optical fiber acoustic wave sensing system, which obtains the polarization state evolution sequence of Rayleigh backscattered light by injecting continuous light waves into the sensing optical fiber and applying dynamic polarization perturbation; differential processing is performed on the polarization state sequence to extract the polarization state change caused by vibration; the differentially processed signal is input into a pre-trained deep neural network model, which directly outputs the vibration distribution information along the sensing optical fiber, including vibration position, frequency, and amplitude. The system includes: a narrow linewidth laser source module, a polarization modulation and perturbation module, a circulator, a sensing fiber, a polarization detection module, and an intelligent signal processing module.

[0008] As a preferred embodiment, the specific steps of the method are as follows: Step 1: Data Acquisition and Polarization State Sequence Acquisition The polarization modulation and perturbation module generates dynamic perturbations, while the polarization detection module continuously acquires the polarization state information of the Rayleigh backscattered light returned from the sensing fiber at a high sampling rate, forming a polarization state evolution sequence that varies with time and space. Step 2: Calculation of Differential Polarization State Sequence The polarization state evolution sequence obtained in step 1 is subjected to differential processing to calculate the change in polarization state at adjacent time points, generating a differential polarization state sequence that highlights the dynamic signal caused by external vibration. Step 3: Vibration information demodulation based on deep neural network The differential polarization state sequence obtained in step 2 is input into a pre-trained deep neural network model, which directly outputs the vibration position, vibration frequency, and vibration amplitude information distributed along the sensing fiber.

[0009] As a preferred embodiment, the output end of the narrow linewidth laser source module is electrically connected to the input end of the polarization modulation and perturbation module to provide continuous probe light; The output of the polarization modulation and perturbation module is connected to the first port optical path of the circulator, and is used to inject the probe light with applied dynamic polarization perturbation into the sensing fiber. The second port of the circulator is connected to one end of the sensing optical fiber. The third port of the circulator is connected to the optical path of the input end of the polarization detection module, and is used to guide the Rayleigh backscattered light returned by the sensing fiber to the detection end. The output of the polarization detection module is electrically connected to the input of the intelligent signal processing module, and is used to convert the polarization state information of the optical signal into an electrical signal. The intelligent signal processing module is used to control the polarization modulation and perturbation module and to perform the differential processing and deep neural network analysis.

[0010] As a preferred embodiment, the system further includes a self-calibration module. When the system starts up or runs intermittently, this module controls the polarization modulation and disturbance module to traverse a set of preset polarization states and record the corresponding detection signals to construct a system error compensation lookup table and to perform real-time compensation on the measurement data during normal sensing.

[0011] As a preferred embodiment, in step 1, the driving signal for the dynamic polarization disturbance is a pseudo-random binary sequence with a symbol rate not less than twice the highest vibration frequency to be detected in the system, so as to ensure sufficient excitation and sampling of the polarization state changes caused by vibration.

[0012] As a preferred embodiment, in step 2, the differential processing specifically involves: preprocessing the polarization state evolution sequence using a variational mode decomposition algorithm to obtain a series of intrinsic mode functions, and selecting the mode components containing the target vibration frequency band for differential calculation to suppress noise interference from non-target frequency bands.

[0013] As a preferred embodiment, in step 3, the deep neural network model adopts an encoder-decoder structure, wherein the encoder consists of a one-dimensional convolutional layer, used to extract high-dimensional features from the difference sequence; the decoder consists of a deconvolutional layer and a fully connected layer, used to map the high-dimensional features back to the spatial dimension, and directly generate a vibration amplitude distribution map along the optical fiber.

[0014] As a preferred option, the training of the deep neural network model adopts a transfer learning strategy: first, it is pre-trained on a large-scale dataset containing a variety of typical vibration events, and then fine-tuned using small sample data of specific application scenarios to improve the model's adaptability and detection accuracy in the target scenario.

[0015] As a preferred embodiment, before performing step 3, a feature enhancement step is also included: performing a short-time Fourier transform on the differential polarization state sequence to generate a time-frequency spectrum, and concatenating the time-frequency spectrum with the original differential sequence in the channel dimension, so as to provide both time-domain and frequency-domain features simultaneously.

[0016] As a preferred embodiment, the method further includes a long-term monitoring mode: in step 3, the deep neural network model is configured to simultaneously output a vibration cumulative effect index, which is obtained by integrating the vibration amplitude over a predetermined time window and is used to assess the long-term trend of structural fatigue.

[0017] (III) Beneficial Effects

[0018] Compared with the prior art, the present invention provides a vibration sensing method based on a distributed fiber optic acoustic wave sensing system, which has the following advantages: This invention obtains the polarization state evolution sequence of Rayleigh backscattered light by injecting continuous light waves into a sensing fiber and applying dynamic polarization perturbations. By detecting polarization state changes instead of traditional intensity or phase information, it effectively avoids coherent fading problems. The polarization state sequence is differentially processed to extract the polarization state changes caused by vibration. The differential signal is input into a pre-trained deep neural network model, which directly outputs the vibration position, frequency, and amplitude. This differential processing strategy effectively extracts dynamic vibration signals, suppresses static environmental noise, and achieves a higher detection signal-to-noise ratio. This invention also uses a deep neural network to directly extract vibration features from complex signals, avoiding the reliance on empirical parameters in traditional algorithms. Furthermore, it can simultaneously acquire multi-dimensional information such as vibration position, frequency, amplitude, and event type, providing more comprehensive monitoring data. Attached Figure Description

[0019] Figure 1 This is a simplified flowchart of the method of the present invention; Figure 2 This is a functional flowchart of the system method of the present invention; Figure 3 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0020] To better understand the purpose, structure, and function of this invention, the vibration sensing method based on a distributed fiber optic acoustic wave sensing system will be further described below with reference to the accompanying drawings and specific embodiments. Example 1

[0021] refer to Figure 1-3 This invention relates to a vibration sensing method based on a distributed fiber optic acoustic wave sensing system. It involves injecting continuous light waves into the sensing fiber and applying dynamic polarization perturbations to obtain the polarization state evolution sequence of Rayleigh backscattered light. The polarization state sequence is then differentially processed to extract the polarization state change caused by vibration. The differentially processed signal is input into a pre-trained deep neural network model, which directly outputs the vibration distribution information along the sensing fiber, including vibration location, frequency, and amplitude. The system includes: a narrow linewidth laser source module, a polarization modulation and perturbation module, a circulator, a sensing fiber, a polarization detection module, and an intelligent signal processing module.

[0022] Specifically, the present invention is based on a distributed fiber optic acoustic wave sensing system, in which the output end of the narrow linewidth laser source module is electrically connected to the input end of the polarization modulation and perturbation module to provide continuous probe light; The output of the polarization modulation and perturbation module is connected to the optical path of the first port of the circulator, and is used to inject the probe light with applied dynamic polarization perturbation into the sensing fiber. The second port of the circulator is connected to one end of the sensing fiber optic path. The third port of the circulator is connected to the input optical path of the polarization detection module, which is used to guide the Rayleigh backscattered light returned by the sensing fiber to the detection end. The output of the polarization detection module is electrically connected to the input of the intelligent signal processing module, and is used to convert the polarization state information of the optical signal into an electrical signal. The intelligent signal processing module is used to control the polarization modulation and perturbation module, and to perform differential processing and deep neural network analysis.

[0023] The system also includes a self-calibration module. During system startup or intermittent operation, this module controls the polarization modulation and disturbance module to traverse a set of preset polarization states and records the corresponding detection signals to construct a system error compensation lookup table. During normal sensing, it performs real-time compensation on the measurement data.

[0024] It should be noted that in this invention's system, the narrow-linewidth laser source module uses a distributed feedback laser with a linewidth of less than 1 kHz. The polarization modulation and perturbation module consists of a polarization controller and a lithium niobate electro-optic polarization modulator, receiving drive signals generated by a pseudo-random binary sequence generator. The circulator is a three-port fiber optic circulator, and the sensing fiber uses G.652.D standard single-mode communication fiber, the length of which can be adjusted according to application requirements. The polarization detection module uses a Stokes parametric instrument based on free-space optics, including a polarization beam splitter, a quarter-wave plate, and four photodetectors, capable of measuring the complete Stokes vector in real time. The intelligent signal processing module consists of a high-speed data acquisition card and a workstation equipped with a GPU, running a custom deep learning inference framework. Example 2

[0025] This invention relates to a vibration sensing method based on a distributed fiber optic acoustic wave sensing system, the specific steps of which are as follows: Step 1: Data Acquisition and Polarization State Sequence Acquisition The polarization modulation and perturbation module generates dynamic perturbations, while the polarization detection module continuously acquires the polarization state information of the Rayleigh backscattered light returned from the sensing fiber at a high sampling rate, forming a polarization state evolution sequence that varies with time and space. Step 2: Calculation of Differential Polarization State Sequence The polarization state evolution sequence obtained in step 1 is subjected to differential processing to calculate the change in polarization state at adjacent time points, generating a differential polarization state sequence that highlights the dynamic signal caused by external vibration. Step 3: Demodulation of vibration information based on deep neural networks.

[0026] The differential polarization state sequence obtained in step 2 is input into a pre-trained deep neural network model, which directly outputs the vibration position, vibration frequency, and vibration amplitude information distributed along the sensing fiber.

[0027] The method of the present invention also includes a long-term monitoring mode: in step 3, the deep neural network model is configured to simultaneously output a vibration cumulative effect index, which is obtained by integrating the vibration amplitude over a predetermined time window and is used to assess the long-term trend of structural fatigue.

[0028] Specifically, after the system starts up, it first executes a self-calibration procedure: controlling the polarization modulator to traverse 12 preset polarization states, recording the corresponding Stokes parameter reference values, and constructing an error compensation lookup table. Then it enters normal monitoring mode, where the polarization detection module continuously acquires the S1, S2, and S3 parameters of the backscattered light at a sampling rate of 50MHz, forming a polarization state evolution sequence; Then, the collected raw data is preprocessed: The signal is decomposed into 5 intrinsic mode functions using the variational mode decomposition algorithm; Modal components with a frequency range of 10Hz-10kHz were selected for subsequent processing; Calculate the Euclidean distance between Stokes parameters at adjacent time points (time interval Δt = 20 μs).

[0029] The difference sequence is then input into the pre-trained encoder-decoder network, where: Encoder: 5 layers of one-dimensional convolution, with 32, 64, 128, 256, and 512 filters per layer, respectively; Decoder: 5 layers of deconvolution to gradually restore spatial resolution Output layer: The three branches output the vibration location heatmap, frequency spectrum, and amplitude distribution, respectively.

[0030] Furthermore, in step 1, the driving signal for dynamic polarization perturbation is a pseudo-random binary sequence with a symbol rate no less than twice the highest detected vibration frequency of the system, ensuring sufficient excitation and sampling of the polarization state changes caused by vibration. The differential processing specifically involves preprocessing the polarization state evolution sequence using a variational mode decomposition algorithm to obtain a series of intrinsic mode functions. Modal components containing the target vibration frequency band are selected for differential calculation to suppress noise interference from non-target frequency bands. The deep neural network model employs an encoder-decoder structure, where the encoder consists of one-dimensional convolutional layers used to extract high-dimensional features from the differential sequence; the decoder consists of deconvolutional layers and fully connected layers used to map the high-dimensional features back to the spatial dimension, directly generating a vibration amplitude distribution map along the optical fiber. The deep neural network model is trained using a transfer learning strategy: first, it is pre-trained on a large-scale dataset containing various typical vibration events, and then fine-tuned using small sample data from specific application scenarios to improve the model's adaptability and detection accuracy in the target scenario.

[0031] It should be further explained that before performing step 3, there is also a feature enhancement step: performing a short-time Fourier transform on the differential polarization state sequence to generate a time-frequency spectrum, and concatenating the time-frequency spectrum with the original differential sequence in the channel dimension, so as to provide both time-domain and frequency-domain features.

[0032] To better understand the phenomenon, simulated vibration with a frequency of 500 Hz and an amplitude of 150 ne was applied at a distance of 9.8 km along a 20 km sensing fiber. The system successfully located the vibration point with a spatial resolution of 8 meters, a frequency measurement error of less than 0.5%, and a signal-to-noise ratio improvement of more than 15 dB compared to traditional methods. The system ran continuously for 72 hours, and the vibration cumulative effect index accurately reflected the patterns of human activity during the experiment.

[0033] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A vibration sensing method based on a distributed fiber optic acoustic wave sensing system, characterized in that, By injecting continuous light waves into the sensing fiber and applying dynamic polarization perturbation, the polarization state evolution sequence of Rayleigh backscattered light is obtained; the polarization state sequence is differentially processed to extract the polarization state change caused by vibration; the differentially processed signal is input into a pre-trained deep neural network model, which directly outputs the vibration distribution information along the sensing fiber, including vibration position, frequency and amplitude. The system includes: a narrow linewidth laser source module, a polarization modulation and perturbation module, a circulator, a sensing fiber, a polarization detection module, and an intelligent signal processing module.

2. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 1, characterized in that, The specific steps of the method are as follows: Step 1: Data Acquisition and Polarization State Sequence Acquisition The polarization modulation and perturbation module generates dynamic perturbations, while the polarization detection module continuously acquires the polarization state information of the Rayleigh backscattered light returned from the sensing fiber at a high sampling rate, forming a polarization state evolution sequence that varies with time and space. Step 2: Calculation of Differential Polarization State Sequence The polarization state evolution sequence obtained in step 1 is subjected to differential processing to calculate the change in polarization state at adjacent time points, generating a differential polarization state sequence that highlights the dynamic signal caused by external vibration. Step 3: Vibration information demodulation based on deep neural network The differential polarization state sequence obtained in step 2 is input into a pre-trained deep neural network model, which directly outputs the vibration position, vibration frequency, and vibration amplitude information distributed along the sensing fiber.

3. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 2, characterized in that, The output of the narrow linewidth laser source module is electrically connected to the input of the polarization modulation and perturbation module to provide continuous probe light. The output of the polarization modulation and perturbation module is connected to the first port optical path of the circulator, and is used to inject the probe light with applied dynamic polarization perturbation into the sensing fiber. The second port of the circulator is connected to one end of the sensing optical fiber. The third port of the circulator is connected to the optical path of the input end of the polarization detection module, and is used to guide the Rayleigh backscattered light returned by the sensing fiber to the detection end. The output of the polarization detection module is electrically connected to the input of the intelligent signal processing module, and is used to convert the polarization state information of the optical signal into an electrical signal. The intelligent signal processing module is used to control the polarization modulation and perturbation module and to perform the differential processing and deep neural network analysis.

4. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 2 or 3, characterized in that, The system also includes a self-calibration module. When the system starts up or runs intermittently, this module controls the polarization modulation and disturbance module to traverse a set of preset polarization states and record the corresponding detection signals to construct a system error compensation lookup table and to compensate the measurement data in real time during normal sensing.

5. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 2, characterized in that, In step 1, the driving signal for dynamic polarization perturbation is a pseudo-random binary sequence with a symbol rate no less than twice the highest vibration frequency to be detected in the system, so as to ensure sufficient excitation and sampling of the polarization state changes caused by vibration.

6. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 2, characterized in that, In step 2, the differential processing specifically involves: using a variational mode decomposition algorithm to preprocess the polarization state evolution sequence, decomposing it into a series of intrinsic mode functions, and selecting the mode components containing the target vibration frequency band for differential calculation to suppress noise interference from non-target frequency bands.

7. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 1, characterized in that, In step 3, the deep neural network model adopts an encoder-decoder structure, wherein the encoder is composed of a one-dimensional convolutional layer, which is used to extract high-dimensional features from the difference sequence. The decoder consists of deconvolutional layers and fully connected layers, used to map high-dimensional features back to spatial dimensions and directly generate a vibration amplitude distribution map along the optical fiber.

8. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 7, characterized in that, The deep neural network model is trained using a transfer learning strategy: first, it is pre-trained on a large-scale dataset containing a variety of typical vibration events, and then fine-tuned using small sample data from specific application scenarios to improve the model's adaptability and detection accuracy in the target scenario.

9. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 2, characterized in that, Before performing step 3, a feature enhancement step is also included: performing a short-time Fourier transform on the differential polarization state sequence to generate a time-frequency spectrum, and concatenating the time-frequency spectrum with the original differential sequence in the channel dimension, so as to provide both time-domain and frequency-domain features.

10. The vibration sensing method based on a distributed fiber optic acoustic wave sensing system according to claim 2, characterized in that, The method also includes a long-term monitoring mode: in step 3, the deep neural network model is configured to simultaneously output a vibration cumulative effect index, which is obtained by integrating the vibration amplitude over a predetermined time window and is used to assess the long-term trend of structural fatigue.

Images