Optical fiber shape coordinate prediction method and system based on neural network

Abstract

The invention discloses an optical fiber shape coordinate prediction method and system based on a neural network, and relates to the field of sensing and measurement, and the method comprises the steps: carrying out the modeling of multi-core optical fibers with different bending degrees through finite element software, extracting the strain values of the fiber cores and the spatial coordinates of the optical fibers, and constructing a training data set; constructing a network model by taking the fiber core strain value of the multi-core optical fiber as input and the space coordinate of the optical fiber as output; inputting the training data set into the network model for model training until a termination condition is met, and obtaining a trained optical fiber shape coordinate prediction model; and measuring the distributed strain of each fiber core of the multi-core optical fiber by using a distributed optical fiber sensing system as a test set, inputting the test set into the trained optical fiber shape coordinate prediction model, and outputting a corresponding optical fiber shape coordinate prediction value. According to the method, complex numerical solution and possibly introduced errors can be avoided, more efficient and high-precision optical fiber shape sensing is realized, and the method has a good application prospect in the field of optical fiber shape sensing / space shape sensing.

Description

Technical Field

[0001] The present invention relates to the field of sensing and measuring technology, and more specifically to a method and system for predicting optical fiber shape coordinates based on a neural network. Background Art

[0002] With the rapid development of information technology, the application demand of spatial shape sensing in aerospace, industrial machinery, biomedicine and other fields is becoming more and more extensive. Compared with traditional shape sensing technology, shape sensing technology based on fiber optic sensing has significant advantages such as small size, light weight, high sensitivity, anti-electromagnetic interference, and easy integration. Fiber optic shape sensing technology uses multiple fiber bundles, fiber grating (FBG) arrays or multi-core optical fibers with specific spatial arrangements to measure the differential strain responses of different optical fibers under the same deformation, thereby obtaining the bending information of the optical fiber, and then using the shape reconstruction algorithm to finally obtain the spatial position and shape of the optical fiber itself or the object to be measured connected to it. Therefore, the process of realizing fiber optic shape sensing generally includes three steps, namely strain sensing measurement, bending information solution and shape reconstruction algorithm iterative calculation. However, when the strain sensing measurement results have been obtained, the two numerical calculation steps of bending information solution and shape reconstruction algorithm iterative calculation are very cumbersome and time-consuming, and the curvature calibration error and the error accumulation effect of the reconstruction algorithm in the numerical calculation process will affect the accuracy of shape sensing.

[0003] At present, the methods of realizing shape sensing based on various neural networks have become a research hotspot. However, the proposed methods of realizing fiber shape sensing based on neural networks can reduce numerical calculations to a certain extent, but most of them are aimed at quasi-distributed strain measurement of FBG, and do not involve distributed strain measurement of multi-core optical fibers. At the same time, the existing neural networks have not established a direct mapping relationship between optical fiber strain and shape coordinates. For example, only the corresponding relationship between optical fiber spectrum and bending curvature and azimuth angle is established, and a certain degree of numerical calculation (i.e., substitution into shape reconstruction algorithm) is still required to finally realize the reconstruction of optical fiber shape.

[0004] Therefore, how to avoid the complex numerical solution process and the errors that may be introduced and achieve the efficiency and accuracy of optical fiber shape sensing is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides a method and system for predicting optical fiber shape coordinates based on a neural network, which solves the problems existing in the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for predicting optical fiber shape coordinates based on a neural network comprises the following steps:

[0008] Finite element software is used to model multi-core optical fibers with different bending degrees, extract the core strain values ​​and the spatial coordinates of the optical fibers, and construct a training data set;

[0009] The network model is constructed by taking the core strain value of the multi-core optical fiber as input and the spatial coordinates of the optical fiber as output;

[0010] The training data set is input into the network model for model training until the termination condition is met, thereby obtaining a trained optical fiber shape coordinate prediction model;

[0011] The distributed strain of each core of a multi-core optical fiber measured by a distributed optical fiber sensing system is used as a test set. The test set is input into a trained optical fiber shape coordinate prediction model, and the corresponding optical fiber shape coordinate prediction value is output.

[0012] Optionally, when constructing a training data set, by changing the bending state of the multi-core optical fiber, several groups of core strain values ​​and optical fiber spatial coordinate data are extracted and split into a training data set and a validation data set in a ratio of 8:2.

[0013] Optionally, a multi-core optical fiber finite element model obtained by modeling multi-core optical fibers with different bending degrees using finite element software consists of a cladding and a core. The multi-core optical fiber finite element model is meshed in a sweeping manner, and when forces of different magnitudes and directions are applied to different positions of the multi-core optical fiber finite element model, bending deformation occurs.

[0014] Optionally, the network model includes: a CNN module, an LSTM module, a Dropout layer, an attention mechanism, and two independent fully connected layers;

[0015] CNN module, used to extract network input data features;

[0016] LSTM module, used to mine the temporal features of the output features of the CNN module;

[0017] Dropout layer, used to prevent overfitting;

[0018] Attention mechanism, used to process the acquired temporal features and weaken the influence of secondary features;

[0019] Two independent fully connected layers are used to process the output of two dimensions respectively, that is, the output of each fully connected layer takes the value of the last time step of the sequence as the predicted value.

[0020] Optionally, the CNN module includes: batch normalization layer, convolution layer and pooling layer;

[0021] Batch normalization layer, used to normalize the input data;

[0022] Convolutional layers, used to increase feature dimensions;

[0023] The pooling layer is used to perform maximum pooling operations on the output features of the convolutional layer.

[0024] Optionally, model training specifically includes the following steps:

[0025] Initialize the network and set the initial values ​​of network hyperparameters;

[0026] The network propagates forward and calculates the loss function, and determines the next step by judging whether the termination condition is met;

[0027] If the termination condition is not met, the gradient calculation is performed, the network is back-propagated and forward-propagated again; the network hyperparameters are adjusted, the model is optimized using the validation data set, and the network weights are continuously updated through iterations;

[0028] If the termination condition is met, the model training is considered to be completed, and a trained optical fiber shape coordinate prediction model is obtained.

[0029] Optionally, the distributed optical fiber sensing system is a Brillouin optical time domain reflection system, a Brillouin optical time domain analysis system or an optical frequency domain reflection system.

[0030] A system for implementing any of the above-mentioned methods for predicting optical fiber shape coordinates based on a neural network, comprising:

[0031] A data acquisition module is used to model multi-core optical fibers with different bending degrees using finite element software, extract the core strain value and the spatial coordinates of the optical fiber, and construct a training data set;

[0032] A model building module, used to build a network model by taking the core strain value of the multi-core optical fiber as input and the spatial coordinates of the optical fiber as output;

[0033] A model training module is used to input the training data set into the network model for model training until the termination condition is met to obtain a trained optical fiber shape coordinate prediction model;

[0034] The prediction module is used to measure the distributed strain of each core of a multi-core optical fiber using a distributed optical fiber sensing system as a test set, input the test set into a trained optical fiber shape coordinate prediction model, and output corresponding optical fiber shape coordinate prediction values.

[0035] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and system for predicting optical fiber shape coordinates based on a neural network, which has the following beneficial effects:

[0036] (1) The present invention adopts CNN and LSTM networks and integrates the attention mechanism to establish a direct mapping relationship between multi-core optical fiber distributed strain and shape coordinates, which can effectively avoid complex numerical calculations and possible errors, thereby improving the efficiency and accuracy of optical fiber shape sensing, and has good prospects in applying optical fiber distributed strain measurement to realize shape sensing;

[0037] (2) From the perspective of time series characteristics, the present invention adopts CNN to obtain the time series characteristics in the distributed strain information of multi-core optical fibers, and the LSTM network predicts the time series characteristics. Then, the attention mechanism is used to weaken the secondary time series characteristics, and finally the prediction result of the optical fiber shape coordinates is obtained. This scheme can achieve better optical fiber shape sensing. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0039] Figure 1 A flowchart of a method for predicting optical fiber shape coordinates based on a neural network provided by the present invention;

[0040] Figure 2 A schematic diagram of the structure of the network model provided by the present invention;

[0041] Figure 3 A flowchart of the model training provided by the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] The general steps to achieve optical fiber shape sensing are to measure the strain along the optical fiber, use the strain to solve the bending information (bending curvature and direction angle), and then use the shape reconstruction algorithm to iteratively solve the shape coordinates to finally obtain the optical fiber shape. However, this numerical solution method based on the geometric relationship of the sensing optical fiber core is cumbersome, and the curvature calibration error and the cumulative error effect of the reconstruction algorithm during the numerical calculation process will affect the accuracy of shape sensing.

[0044] To this end, the embodiment of the present invention discloses a method for predicting optical fiber shape coordinates based on a neural network, such as Figure 1 As shown, the following steps are included:

[0045] Finite element software is used to model multi-core optical fibers with different bending degrees, extract the core strain values ​​and the spatial coordinates of the optical fibers, and construct a training data set;

[0046] The network model is constructed by taking the core strain value of the multi-core optical fiber as input and the spatial coordinates of the optical fiber as output;

[0047] The training data set is input into the network model for model training until the termination condition is met, thereby obtaining a trained optical fiber shape coordinate prediction model;

[0048] The distributed strain of each core of a multi-core optical fiber measured by a distributed optical fiber sensing system is used as a test set. The test set is input into a trained optical fiber shape coordinate prediction model, and the corresponding optical fiber shape coordinate prediction value is output.

[0049] Reference Figure 1 The scheme shown in this embodiment uses a neural network to establish a fiber shape coordinate prediction model, constructs a direct mapping relationship between the fiber distributed strain and the shape coordinates, avoids the complex numerical solution process and the errors that may be introduced, can achieve high-precision fiber shape sensing, and solve the above-mentioned problems.

[0050] Next, Figure 1 The process shown is described in detail to provide a deeper understanding of the technical solution.

[0051] 1. Data acquisition

[0052] Finite element software was used to model multi-core optical fibers with different bending degrees, extract the core strain values ​​and the spatial coordinates of the optical fibers, and construct a training data set. By changing the bending state of the multi-core optical fiber, 7200 sets of core strain values ​​and spatial coordinate data of the optical fibers were extracted and split into a training data set and a validation data set in a ratio of 8:2.

[0053] The multi-core optical fiber finite element model obtained by modeling consists of a cladding and a core. Since the core diameter is much smaller than the cladding diameter, the core in the model is simplified to a straight line with a 120° distribution. The finite element model has a diameter of 1.25 mm, a length of 50 mm, and an elastic modulus of 7×10 10 N / m 2, Poisson's ratio is 0.27. The multi-core optical fiber finite element model is meshed in a sweeping manner. When forces of different magnitudes and directions are applied to different positions of the multi-core optical fiber finite element model, bending deformation occurs. The strains of the three cores and the optical fiber shape coordinate values ​​are extracted in the finite element software, and the core strains are used as input data and the shape coordinate values ​​are used as output data to construct the training data set of the network model.

[0054] 2. Model construction

[0055] Most of the existing methods for realizing fiber shape sensing based on neural networks are aimed at quasi-distributed sensing of fiber Bragg gratings (FBGs), and no direct mapping relationship between strain measurement results and fiber shape coordinates is established. To this end, this embodiment constructs a network model, the input of which is the distributed strain of the three cores of a three-core optical fiber, and the output is the fiber space coordinates. Figure 2 As shown in the figure, the network model includes: a CNN module, an LSTM module, a Dropout layer, an attention mechanism, and two independent fully connected layers;

[0056] CNN module, used to extract network input data features;

[0057] LSTM module, used to mine time series features of the 128-dimensional output features obtained after 128 filters in the CNN module;

[0058] Dropout layer, used to prevent overfitting;

[0059] Attention mechanism, used to process the acquired temporal features and weaken the influence of secondary features;

[0060] Two independent fully connected layers are used to process the output of two dimensions respectively, that is, the output of each fully connected layer takes the value of the last time step of the sequence as the predicted value.

[0061] Furthermore, the CNN module includes: a batch normalization layer (BN), a convolutional layer (Con), and a pooling layer (Pool);

[0062] Batch Normalization layer (BN), used to normalize the input data;

[0063] Convolutional layer (Con), used to increase the feature dimension; the convolution kernel size in the convolutional layer is 7×1;

[0064] The pooling layer (Pool) is used to perform maximum pooling operations on the output features of the convolutional layer.

[0065] For an optical fiber of a certain shape, its shape reconstruction result is represented by spatial coordinates, and the distribution of shape coordinates has characteristics similar to the trend and irregularity of time series. LSTM network has significant advantages in time series prediction. This embodiment proposes a multi-core optical fiber shape coordinate prediction network model that integrates CNN-LSTM and attention mechanism from the perspective of time series characteristics, wherein a convolutional neural network (CNN) is combined with a long short-term memory (LSTM) network to extract the time series features in the strain information, and the attention mechanism is integrated to weaken the secondary time series features, thereby establishing a direct mapping relationship between the core strain and the shape coordinates, and realizing the direct prediction of the optical fiber shape coordinates. This method can effectively avoid complex numerical calculations and possible errors, improve the efficiency and accuracy of optical fiber sensing, and achieve better optical fiber shape sensing.

[0066] 3. Model training

[0067] like Figure 3 As shown in the figure, model training specifically includes the following steps:

[0068] Initialize the network and set the initial values ​​of the network hyperparameters; the hyperparameters are learning rate and optimizer;

[0069] The network propagates forward and calculates the loss function, and determines the next step by judging whether the termination condition is met;

[0070] If the termination condition is not met, the gradient calculation is performed, the network is back-propagated and forward-propagated again; the network hyperparameters are adjusted, the model is optimized using the validation data set, and the network weights are continuously updated through iterations;

[0071] If the termination condition is met, the model training is considered to be completed, and a trained optical fiber shape coordinate prediction model is obtained.

[0072] 4. Network testing

[0073] The distributed strain of each core of a multi-core optical fiber measured by a distributed optical fiber sensing system is used as a test set. The test set is input into a trained optical fiber shape coordinate prediction model, and the corresponding optical fiber shape coordinate prediction value is output to realize optical fiber shape sensing.

[0074] Among them, the distributed optical fiber sensing system is a Brillouin optical time domain reflection system, a Brillouin optical time domain analysis system or an optical frequency domain reflection system.

[0075] In this embodiment, the curves with curvature radius of 400 mm and 700 mm are measured, and the strain measurement results are used as the test set data of the neural network. The prediction results of the network model are compared with the numerical calculation results based on the Freynes equation, and the root mean square error (RMSE) and mean absolute error (MAE) of shape reconstruction are shown in Table 1.

[0076] Table 1

[0077]

[0078] It can be seen that compared with the numerical calculation method based on the Freynes equation, the neural network model of this embodiment can avoid complex numerical solutions. It only needs to input the distributed strain of the multi-core optical fiber into the network model to directly obtain the shape coordinates. Moreover, the prediction result of the optical fiber shape coordinates is more accurate, which improves the efficiency of optical fiber shape sensing.

[0079] and Figure 1 Corresponding to the method described above, the embodiment of the present invention also provides a fiber shape coordinate prediction system based on a neural network for Figure 1 The specific implementation of the method in the embodiment of the present invention is a fiber shape coordinate prediction system based on a neural network, which can be applied to a computer terminal or various mobile devices, and specifically includes:

[0080] A data acquisition module is used to model multi-core optical fibers with different bending degrees using finite element software, extract the core strain value and the spatial coordinates of the optical fiber, and construct a training data set;

[0081] A model building module, used to build a network model by taking the core strain value of the multi-core optical fiber as input and the spatial coordinates of the optical fiber as output;

[0082] A model training module is used to input the training data set into the network model for model training until the termination condition is met to obtain a trained optical fiber shape coordinate prediction model;

[0083] The prediction module is used to measure the distributed strain of each core of a multi-core optical fiber using a distributed optical fiber sensing system as a test set, input the test set into a trained optical fiber shape coordinate prediction model, and output corresponding optical fiber shape coordinate prediction values.

[0084] In summary, this embodiment establishes a neural network model that combines CNN and LSTM networks and integrates the attention mechanism, which can establish a direct mapping of multi-core optical fiber distributed strain and shape coordinates, avoid complex numerical solutions and possible errors, and achieve more efficient and high-precision optical fiber shape sensing. It has good application prospects in the field of optical fiber shape sensing / spatial shape perception.

[0085] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0086] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting optical fiber shape coordinates based on a neural network, characterized in that: The following steps are involved: Finite element software is used to model multi-core optical fibers with different bending degrees, extract the core strain values ​​and the spatial coordinates of the optical fibers, and construct a training data set; The network model is constructed by taking the core strain value of the multi-core optical fiber as input and the spatial coordinates of the optical fiber as output; The training data set is input into the network model for model training until the termination condition is met, thereby obtaining a trained optical fiber shape coordinate prediction model; The distributed strain of each core of a multi-core optical fiber measured by a distributed optical fiber sensing system is used as a test set. The test set is input into a trained optical fiber shape coordinate prediction model, and the corresponding optical fiber shape coordinate prediction value is output.

2. The method for predicting optical fiber shape coordinates based on a neural network according to claim 1, characterized in that: When constructing the training data set, by changing the bending state of the multi-core optical fiber, several groups of core strain values ​​and optical fiber spatial coordinate data are extracted and split into a training data set and a validation data set in a ratio of 8:

2.

3. The method for predicting optical fiber shape coordinates based on a neural network according to claim 1, characterized in that: The multi-core optical fiber finite element model obtained by modeling multi-core optical fibers with different bending degrees using finite element software consists of a cladding and a core. The multi-core optical fiber finite element model is meshed in a sweeping manner. When forces of different magnitudes and directions are applied to different positions of the multi-core optical fiber finite element model, bending deformation occurs.

4. The method for predicting optical fiber shape coordinates based on a neural network according to claim 1, characterized in that: The network model includes: a CNN module, an LSTM module, a Dropout layer, an attention mechanism, and two independent fully connected layers; CNN module, used to extract network input data features; LSTM module, used to mine the temporal features of the output features of the CNN module; Dropout layer, used to prevent overfitting; Attention mechanism, used to process the acquired temporal features and weaken the influence of secondary features; Two independent fully connected layers are used to process the output of two dimensions respectively, that is, the output of each fully connected layer takes the value of the last time step of the sequence as the predicted value.

5. The method for predicting optical fiber shape coordinates based on a neural network according to claim 4, characterized in that: The CNN module includes: batch normalization layer, convolution layer and pooling layer; Batch normalization layer, used to normalize the input data; Convolutional layers, used to increase feature dimensions; The pooling layer is used to perform maximum pooling operations on the output features of the convolutional layer.

6. The method for predicting optical fiber shape coordinates based on a neural network according to claim 2, characterized in that: Model training specifically includes the following steps: Initialize the network and set the initial values ​​of network hyperparameters; The network propagates forward and calculates the loss function, and determines the next step by judging whether the termination condition is met; If the termination condition is not met, the gradient calculation is performed, the network is back-propagated and forward-propagated again; the network hyperparameters are adjusted, the model is optimized using the validation data set, and the network weights are continuously updated through iterations; If the termination condition is met, the model training is considered to be completed, and a trained optical fiber shape coordinate prediction model is obtained.

7. The method for predicting optical fiber shape coordinates based on a neural network according to claim 1, characterized in that: The distributed optical fiber sensing system is a Brillouin optical time domain reflection system, a Brillouin optical time domain analysis system or an optical frequency domain reflection system.

8. A system for implementing the method for predicting optical fiber shape coordinates based on a neural network as claimed in any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to model multi-core optical fibers with different bending degrees using finite element software, extract the core strain value and the spatial coordinates of the optical fiber, and construct a training data set; A model building module, used to build a network model by taking the core strain value of the multi-core optical fiber as input and the spatial coordinates of the optical fiber as output; A model training module is used to input the training data set into the network model for model training until the termination condition is met to obtain a trained optical fiber shape coordinate prediction model; The prediction module is used to measure the distributed strain of each core of a multi-core optical fiber using a distributed optical fiber sensing system as a test set, input the test set into a trained optical fiber shape coordinate prediction model, and output corresponding optical fiber shape coordinate prediction values.

Images