BOTDA long-distance dynamic strain extraction method based on ESRCNN

Abstract

The invention discloses an ESRCNN-based BOTDA long-distance dynamic strain extraction method, and relates to the technical field of strain measurement, and the method comprises the steps: constructing a plurality of low-resolution BGS images and high-resolution BGS images to train an ESRGAN network, so as to enable the ESRGAN network to learn the mapping relation between the low-resolution BGS and the high-resolution BGS; training a stress extraction CNN network through the high-resolution BGS image and a corresponding stress label so as to enable the stress extraction CNN network to establish a mapping relation between the high-resolution BGS and stress; and processing Brillouin gain spectrum data measured by the BOTDA by using the trained ESRCNN to obtain stress distribution along the optical fiber. According to the invention, the strain extraction speed of the BOTDA system can be improved, and the potential of the BOTDA in the field of dynamic strain detection can be improved.

Description

Technical Field

[0001] The present invention relates to the field of strain measurement technology, and in particular to a BOTDA long-distance dynamic strain extraction method based on ESRCNN. Background Art

[0002] The significance of dynamic strain measurement using the Brillouin Optical Time Domain Analysis (BOTDA) system lies in its ability to achieve long-distance, distributed, high-spatial-resolution, and high-precision strain sensing. This technology has broad application prospects in fields such as large-scale infrastructure health monitoring and aircraft status monitoring. Dynamic strain measurement is crucial for assessing structural safety, early warning, and status assessment because it can monitor in real time the strain changes of structures under the influence of various environmental factors. Dynamic strain measurement can distinguish strain caused by moving loads, which is particularly important for health monitoring of structures such as bridges. This is crucial for real-time processing of monitoring data, timely early warning, identification of dynamic loads, and safety status assessment in online monitoring systems. Therefore, dynamic strain measurement using the BOTDA system has important practical significance for ensuring the long-term stability and safety of structures. How to achieve fast, long-distance, and high-precision dynamic strain measurement is an urgent problem to be solved.

[0003] BOTDA technology plays an important role in these fields due to its high sensitivity and long-distance measurement capabilities. It can provide high-precision strain data to help improve the safety and reliability of the system. At low gain, the stimulated Brillouin gain spectrum can be represented by a Lorentzian line shape, and as the gain increases, the gain spectrum evolves from a Lorentzian line shape to a Gaussian line shape. Specifically, the stimulated Brillouin gain spectrum at low gain can be represented by a Lorentzian line shape, and the gain spectrum peak frequency can be obtained by curve fitting. The measured BFS is also known as v′ B , which is consistent with v B The corresponding strain value can be obtained by calculating the difference between the two. The specific calculation formula is as follows:

[0004] v′ B =C ε,BFS Δε+v B (1)

[0005] Among them, C ε,BFS≈0.046MHz / με, representing the strain sensitivity of the BFS in a standard single-mode fiber. Dynamic strain measurement places higher demands on the data acquisition and processing speed of the BOTDA system. For example, when sweeping the probe light, to avoid Brillouin signal overlap, the repetition rate of the pump pulse light should be less than the speed of light divided by twice the fiber length, which limits the measurement time. To enhance the signal-to-noise ratio of the Brillouin signal and improve measurement accuracy, multiple averaging operations are required. If the fiber is polarization-maintaining, the number of averaging operations is relatively small. For standard single-mode fiber, a polarization scrambler is required to eliminate the influence of polarization noise, requiring thousands of averaging operations, which significantly limits the measurement time. When sweeping the distributed Brillouin gain spectrum, the switching time of the probe light frequency is typically determined by the frequency switching time of the electrically modulated signal. The switching time of the microwave signal output by the microwave source is typically on the order of milliseconds or even longer, significantly delaying the acquisition time. To measure the complete Brillouin gain spectrum or a large strain / temperature range, the sweep range must be expanded. At the same time, if a more accurate Brillouin gain spectrum curve is required, a smaller frequency sweep interval is required, which increases the number of frequency sweeps of the probe light and limits the acquisition time.

[0006] Recent research on dynamic BOTDA data processing has explored the use of advanced deep learning models, such as the Enhanced Super Resolution Convolutional Neural Network (ESRCNN), to improve the accuracy and efficiency of data analysis. These models aim to: 1. reconstruct a super-resolution Brillouin gain spectrum (BGS) from a low-resolution BGS. This approach reduces the number of probes required and the time-domain sampling rate, thereby enhancing BOTDA performance while improving both resolution and accuracy. 2. Accurately and rapidly extracting effective BFS information from the noisy BGS and converting it into a strain distribution along the fiber, significantly increasing the strain measurement speed of BOTDA systems while maintaining or improving measurement accuracy. The ESRCNN network establishes a complex, nonlinear mapping between the BGS image s and the strain distribution v. The data generated by BOTDA systems is massive and complex, involving extensive time and frequency domain information, posing a challenge to traditional data processing methods. The ESRCNN network, combining the characteristics of super-resolution neural networks and convolutional neural networks, possesses powerful feature extraction and modeling capabilities, enabling it to process this complex data. This enables faster and more accurate dynamic strain distribution measurements. The network uses the Generative Adversarial Network (GAN) network to establish a mapping between low-resolution BGS and high-resolution BGS data, helping to recover high-resolution BGS data from BGS data with large frequency sweep steps. This reduces the number of sweep frequencies during strain measurement without losing spatial and frequency domain information. The cascaded CNN network enables efficient and rapid strain extraction, reducing the time required by traditional Lorentz curve fitting by several orders of magnitude. Through these mechanisms, the ESRCNN network is able to better extract dynamic strain from BOTDA data, improving strain detection accuracy. This can effectively accelerate the responsiveness of real-time monitoring and early warning systems, providing more reliable data support for key areas such as structural health monitoring, earthquake prediction, and energy pipeline safety.

[0007] Therefore, how to extract long-distance dynamic strain from BOTDA based on ESRCNN is a technical problem that needs to be solved urgently. Summary of the Invention

[0008] The embodiment of the present invention provides a BOTDA long-distance dynamic strain extraction method based on ESRCNN, which can improve the strain extraction speed of the BOTDA system and enhance the potential of BOTDA in the field of dynamic strain detection.

[0009] In a first aspect, the present invention provides a BOTDA long-distance dynamic strain extraction method based on ESRCNN, comprising:

[0010] Construct multiple low-resolution BGS images and high-resolution BGS images to train the ESRGAN network, so that the ESRGAN network can learn the mapping relationship between low-resolution BGS and high-resolution BGS;

[0011] The stress extraction CNN network is trained using high-resolution BGS images and corresponding stress labels, so that the stress extraction CNN network can establish a mapping relationship between high-resolution BGS and stress.

[0012] The trained ESRCNN is used to process the Brillouin gain spectrum data obtained by BOTDA measurement to obtain the stress distribution along the optical fiber.

[0013] In some examples, a training dataset comprising pairs of low-resolution BGS images and high-resolution BGS images is constructed from simulated data.

[0014] In some examples, constructing a training dataset comprising pairs of low-resolution BGS images and high-resolution BGS images from simulated data comprises:

[0015] The low-resolution BGS images with sampling frequency steps of 20 MHz, 15 MHz, 10 MHz, and 5 MHz and the high-resolution BGS images with a sampling frequency step of 1 MHz form image pairs.

[0016] In some examples, constructing a plurality of low-resolution BGS images and a high-resolution BGS image to train the ESRGAN network includes:

[0017] Convert the image pairs into a format suitable for neural network processing and perform normalization;

[0018] Set the low-resolution BGS image as input and the high-resolution BGS image as output, and use the data loading function provided by the deep learning framework to load and batch process the image pairs.

[0019] In some instances, when acquiring stress signatures corresponding to high-resolution BGS images, dynamic strain loading utilizes the rotation of an eccentric to cause fiber stretching. The rotation speed of the eccentric is adjusted to determine the frequency of the dynamic strain. The strain offset range is set to 0 με to 5000 με with a step size of 50 με, and the Brillouin linewidth variation range is set to 25 MHz to 80 MHz with a step size of 5 MHz.

[0020] In some instances, the input layer size of the CNN network matches the output layer size of the GAN network during training, and a noisy BGS with a signal-to-noise ratio of approximately 5dB is added to the training set.

[0021] In some instances, during CNN training, BGSs are stacked by distance, and their amplitudes are attenuated according to the loss coefficient along the optical fiber. The number of pixels in the final distance-frequency shift image is N = frequency sampling interval × distance window length / distance sampling interval.

[0022] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0023] First, the ESRCNN network, through the image super-resolution effect of its GAN network module, can reconstruct a high-resolution Brillouin gain spectrum from a low-resolution Brillouin gain spectrum, achieving super-resolution image reconstruction. Specifically, the GAN network generates a high-resolution image from a low-resolution image through its generator, and the discriminator determines the image's authenticity, thus making the generated image similar to the real image in terms of structure and style. This approach not only reduces the number of required probes and the temporal sampling rate, but also improves the performance of the BOTDA system, reducing the number of measurements and increasing spatial and frequency domain resolution. Furthermore, the GAN network can learn complex feature maps to produce clearer and more realistic high-resolution images, especially in terms of visual perception, making the restored image more realistic. Therefore, the GAN network plays a significant role in the super-resolution of the BOTDA system, improving measurement efficiency and enhancing the accuracy and reliability of the measurement results. It can extract more detailed strain information from the gain spectrum acquired by the BOTDA system. Its efficient feature extraction and learning mechanism significantly improves the resolution and accuracy of strain data, thereby reducing the influence of errors and noise in the strain measurement process. Secondly, the CNN module of the ESRCNN network offers significant benefits in terms of speed and accuracy compared to traditional curve fitting methods. Specifically, CNN leverages its powerful feature extraction capabilities to directly extract distributed strain information from the BGS, eliminating the need for a complex BFS-to-strain conversion process. Compared to traditional equation-solving methods, CNN-based methods exhibit greater noise tolerance and robustness, maintaining high accuracy even at low signal-to-noise ratios (SNRs). Furthermore, the application of CNN significantly accelerates data processing. This speed increase enables the BOTDA system to respond more quickly, which is particularly important for applications requiring real-time monitoring. Furthermore, the CNN achieves very low mean standard deviation (SD) and root mean square error (RMSE) when extracting temperature and strain information, demonstrating its superior accuracy. Therefore, the application of the ESRCNN network can improve the strain extraction speed of the BOTDA system and enhance its potential in the field of dynamic strain detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 is a schematic diagram of a method provided by an embodiment of the present invention;

[0026] Figure 2 : The Brillouin gain spectrum obtained by the BOTDA system provided in an embodiment of the present invention, wherein: (a) the distributed Brillouin gain spectrum of the optical fiber under test with a sweep frequency step of 1 MHz; (b) the gain spectrum distribution along the optical fiber under test after strain is applied with a sweep frequency step of 5 MHz;

[0027] Figure 3 This is a flow chart of ESRCNN network strain extraction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0029] In the following description, specific embodiments of the present invention will be described with reference to steps and symbols performed by one or more computers, unless otherwise specified. Therefore, these steps and operations will be mentioned several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit that represents electronic signals of data in a structured form. This operation converts the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise change the operation of the computer in a manner familiar to testers in the field. The data structure in which the data is maintained is a physical location in the memory, which has specific characteristics defined by the data format. However, the principles of the present invention are described in the above text, which does not represent a limitation, and testers in the field will understand that the various steps and operations below can also be implemented in hardware.

[0030] As used herein, the terms "module" or "unit" may be considered software objects executed on the computing system. The various components, modules, engines, and services herein may be considered implementation objects on the computing system. While the devices and methods herein are preferably implemented in software, they may also be implemented in hardware and remain within the scope of protection of the present invention.

[0031] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0032] There are the following technical problems when using the ESRCNN network:

[0033] 1. The dataset design for ESRCNN network model training is very important. Since the ESRCNN network is a structure formed by the cascade of the ESRGAN network and the stress extraction CNN, the model training needs to be divided into two parts. First, the high-resolution BGS generation of the GAN network must be considered, and at the same time, the training set for the high-precision CNN stress extraction design must be extracted.

[0034] 2. Preprocessing of the input data of the ESRCNN network model is an important step to ensure the effectiveness of network training and the stability of the results. For the task of accurately extracting dynamic strain, standardization is a very important step in data preprocessing, which can accelerate the convergence of the network. For example, the pixel value range of an image is between 0 and 255, and normalization to the interval [-1,1] can make the network converge faster. In addition, data enhancement cannot be ignored, as it can help the model learn more robust features. During the training process of GAN, the use of transformations such as random rotation and translation can effectively improve the diversity of generated samples. Secondly, since the measurement of strain is related to the length of the original stretched optical fiber, it is necessary to design labels for the CNN network training set. Therefore, the preprocessing method of the network model input data is a technical issue worthy of in-depth study.

[0035] In an embodiment of the present invention, a BOTDA long-distance dynamic strain extraction method based on ESRCNN is provided. Figure 1 As shown, the following steps are included:

[0036] S101: Construct multiple low-resolution BGS images and high-resolution BGS images to train the ESRGAN network, so that the ESRGAN network learns the mapping relationship between low-resolution BGS and high-resolution BGS;

[0037] S102: training a stress extraction CNN network using the high-resolution BGS image and the corresponding stress label, so that the stress extraction CNN network establishes a mapping relationship between the high-resolution BGS and stress;

[0038] S103: Using the trained ESRCNN to process the Brillouin gain spectrum data obtained by BOTDA measurement, the stress distribution along the optical fiber is obtained.

[0039] Furthermore, during the ESRCNN network training process, the GAN module must first be trained. This requires constructing multiple low-resolution BGS-high-resolution BGS images for the GAN module to learn the mapping relationship between the two, achieving rapid BGS resolution improvement, that is, completing the missing scanning frequency intensity. The recovered high-resolution BGS is then input into the subsequent CNN module for dynamic stress extraction. The training set label setting of this CNN module should be based on the optical fiber tensile strain calculation formula:

[0040]

[0041] Here, ΔL represents the elongation of the stretched fiber segment, and L represents the original length of the stretched fiber segment. When training the CNN module, it is necessary to first assign different stress labels to the BGS. This allows the CNN to establish a high-resolution BGS-stress mapping during training.

[0042] Figure 2 (a) shows an image of the high-resolution 3D Brillouin gain spectrum, which is the image generated by the super-resolution module. The distance represents the different sensing points on the fiber, while the frequency axis represents the frequency range of the modulator. To enhance the super-resolution convolutional neural network, the super-resolution module must be trained first. The following steps need to be followed for module training. This module is composed of a generative adversarial neural network, such as Figure 3 The ESRGAN module shown in Figure 1 extracts, restores, and refines image features through multiple convolutional layers.

[0043] Furthermore, applying the ESRCNN network to dynamic strain extraction in BOTDA sensing systems requires the following technical considerations. First, the training dataset should be constructed using simulated data to ensure the generalization capability of the trained network model. The GAN network module, which recovers low-resolution Brillouin gain spectra (BGS) to high-resolution BGS, requires constructing a dataset containing paired low-resolution and high-resolution BGS images. These paired images will be used to train the GAN, enabling it to learn how to generate high-resolution images from low-resolution images while minimizing the addition of redundant and spurious information. In practice, different sampling frequency steps are set. Low-resolution BGS with sampling frequency steps of 20MHz, 15MHz, 10MHz, and 5MHz are paired with high-resolution BGS data with a sampling frequency step of 1MHz to form data pairs. Low-resolution BGS is set as input and high-resolution BGS as output. The data loading functions provided by the deep learning framework, such as Dataset and DataLoader in PyTorch, can be used to load and batch process these paired image data so that they can be effectively input into the network during training. As the network training process progresses, the parameters in the module will be adjusted to the optimal state. This process requires continuous trial and error. In addition, the data preprocessing step is also critical, including converting the image into a format suitable for neural network processing, such as Tensor, and performing normalization so that the pixel values fall within an appropriate range, thereby accelerating network convergence and improving training results.

[0044] Figure 2 (a) shows the Brillouin gain spectrum measured using the BOTDA. This spectrum was measured with a 1 MHz sweep interval, scanning 200 frequencies. Each frequency trace required 1024 averagings, which was time-consuming. Figure 2 (b) shows a BGS measurement image with a scanning frequency interval of 5 MHz. It can be seen that there is a high degree of distortion at the location where the strain is applied, but the measurement time is greatly reduced. Submitting this image to CNN processing may also cause large measurement errors. Therefore, the super-resolution module of the ESRCNN network, that is, the GAN network module, can effectively input high-resolution data into the CNN, obtaining more refined and accurate processing results.

[0045] Furthermore, the cascaded extraction convolutional neural network module, i.e. Figure 3 When the CNN extraction module in the BOTDA sensing system is used for dynamic strain extraction, it is necessary to master the BGS characteristics of parameters such as strain and line width when they change over a large range in order to achieve accurate strain measurement. At the same time, extraction speed is also an important consideration. The input of the network is as follows Figure 2(b) shows a high-resolution Brillouin gain plot on the distance-frequency plane. The final output is a stress-distance image. Strain normalization ensures that the network outputs accurate results for fibers with varying stretch lengths. The final network output strain has a root mean square error of less than 50 microstrain.

[0046] Furthermore, the cascaded CNN module needs to learn the BGS features when variables such as strain and linewidth vary over a large range to achieve accurate strain extraction. At the same time, the extraction speed also becomes a factor that needs to be considered. According to formula (1), the influence of the BGS spectrum line is mainly due to the change of BFS, which corresponds to the range of the BFS offset v'. The loading of dynamic strain can be used to cause a large stretch of the optical fiber by rotating the eccentric wheel. Adjusting the rotation speed of the eccentric wheel can determine the frequency of the dynamic strain. The strain offset Δε is set to a range of 0με~5000με with a step size of 50με. The Brillouin linewidth in standard single-mode optical fiber is basically around 60MHz under 20ns pulse conditions. Therefore, the Brillouin linewidth variation range can be set to 25MHz~80MHz. Considering the training resource consumption, the step size of this parameter can be set to 5MHz to reduce the consumption of training resources. During the CNN training process, it is important to note that the input layer size of the network needs to match the output layer size of the GAN network. At the same time, in order to increase the robustness of the network, a noisy BGS with a signal-to-noise ratio of about 5dB is added to the training set. These BGSs are then stacked by distance, with their amplitude attenuated according to the loss coefficient along the fiber. The resulting distance-frequency shift image has a pixel count N = frequency sampling interval × distance window length / distance sampling interval. Specifically, the training data labels require the use of normalized strain to ensure that the network outputs accurate results for fibers with varying stretch lengths. Once properly trained, the network can be applied to high-speed BOTDA strain measurement.

[0047] Figure 3 The paper demonstrates a network architecture for using ESRCNN to process Brillouin gain spectrum data from BOTDA measurements to determine stress distribution along an optical fiber. While the sampling step size of the input low-resolution BGS is 10 MHz, the sampling step size of the intermediate output high-resolution BGS is increased to 2 MHz, significantly improving accuracy, preventing information loss, and reducing the number of parameters. The resulting high-resolution BGS serves as input to the CNN network, allowing strain information to be accurately extracted through the CNN extraction module, reducing the process time by more than two orders of magnitude. This approach promises to enable long-distance dynamic strain measurement.

[0048] To enable long-distance network extraction of dynamic strain, a distributed Brillouin optical time-domain analysis sensing system was used to measure the strain distribution of a 50km optical fiber. The sampling frequency step size was set to 15MHz, and the scanning window width was 300MHz. The system ultimately recovered a three-dimensional Brillouin gain spectrum with a step size of 1MHz, enabling rapid strain extraction. Ultimately, dynamic strain measurement required recovering the vibration power spectrum to verify feasibility.

[0049] The above is a detailed introduction to a BOTDA long-distance dynamic strain extraction method based on ESRCNN provided by an embodiment of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A BOTDA long-distance dynamic strain extraction method based on ESRCNN, characterized in that: include: Construct multiple low-resolution BGS images and high-resolution BGS images to train the ESRGAN network, so that the ESRGAN network can learn the mapping relationship between low-resolution BGS and high-resolution BGS; The stress extraction CNN network is trained using high-resolution BGS images and corresponding stress labels, so that the stress extraction CNN network can establish a mapping relationship between high-resolution BGS and stress. The trained ESRCNN is used to process the Brillouin gain spectrum data obtained by BOTDA measurement to obtain the stress distribution along the optical fiber.

2. The method according to claim 1, characterized in that A training dataset consisting of pairs of low-resolution BGS images and high-resolution BGS images is constructed from simulated data.

3. The method according to claim 2, characterized in that The training data set comprising paired low-resolution BGS images and high-resolution BGS images is constructed from simulation data, including: The low-resolution BGS images with sampling frequency steps of 20 MHz, 15 MHz, 10 MHz, and 5 MHz and the high-resolution BGS images with a sampling frequency step of 1 MHz form image pairs.

4. The method according to claim 3, characterized in that The step of constructing a plurality of low-resolution BGS images and high-resolution BGS images to train the ESRGAN network includes: Convert the image pairs into a format suitable for neural network processing and perform normalization; Set the low-resolution BGS image as input and the high-resolution BGS image as output, and use the data loading function provided by the deep learning framework to load and batch process the image pairs.

5. The method according to claim 4, characterized in that When acquiring the stress signature corresponding to the high-resolution BGS image, dynamic strain loading utilizes the rotation of the eccentric to cause fiber stretching. The rotation speed of the eccentric is adjusted to determine the frequency of the dynamic strain. The strain offset range is set to 0με to 5000με with a step size of 50με, and the Brillouin linewidth variation range is set to 25MHz to 80MHz with a step size of 5MHz.

6. The method according to claim 5, characterized in that During CNN training, the input layer size of the network matches the output layer size of the GAN network, and a noisy BGS with a signal-to-noise ratio of about 5dB is added to the training set.

7. The method according to claim 6, characterized in that During the CNN training process, BGS are stacked by distance, and their amplitude is attenuated according to the loss coefficient along the optical fiber. The number of pixels of the final distance-frequency shift image is N = frequency sampling interval × distance window length / distance sampling interval.

Images