Flexible antenna deformation monitoring and performance prediction system

Abstract

A flexible antenna deformation monitoring and performance prediction system comprises a deformation monitoring module and a deformation performance prediction module. The variable monitoring module comprises a fiber grating sensor array, a grating demodulation sub-module and a computer; wherein the fiber grating sensor array collects a deformation data signal of the flexible antenna, and is connected with the grating demodulation sub-module through the FC joint to demodulate wavelength information of the flexible antenna; wavelength information obtained by the grating demodulation sub-module is subjected to data signal interaction with a computer through an Ethernet interface, the computer processes data in real time, wavelength change information is processed into curvature data information in real time through the deformation performance prediction module, and a whole curved surface is reconstructed through a curved surface reconstruction algorithm. The computer respectively displays data of the deformation monitoring module and the deformation performance prediction module in real time through the deformation monitoring curved surface display interface and the antenna performance prediction interface; when the flexible antenna deforms, the current bending deformation state can be displayed in real time, and the antenna performance in the current bending state can be predicted in real time through deployment of the neural network; the deformation state of the antenna can be displayed in real time in a complex environment, and the performance under the current deformation can be quickly responded, so that the condition of performance loss can be well avoided when the flexible antenna is worn and used.

Description

Technical Field

[0001] The present invention belongs to the field of fiber grating sensing and deep learning, and specifically relates to a flexible antenna deformation monitoring and performance prediction system. Background Art

[0002] Flexible antennas are flexible microwave circuits with tightly coupled mechanical structures and electromagnetic properties. Due to their bendable and stretchable characteristics, they are easily subjected to mechanical stress and deformation during wearing, which causes the antenna structure to deform, thereby affecting its radiation characteristics and resonant frequency, leading to a serious deterioration of electromagnetic performance. In order to explore the electromechanical coupling relationship, it is necessary to know the corresponding flexible antenna performance under different deformation states. In addition to testing the antenna, it is also necessary to accurately measure the deformation state of the flexible antenna.

[0003] In the 1960s, two major scientific discoveries occurred in the field of optical fiber: the invention of the laser in 1960 and the establishment of the low-loss optical fiber theory in 1966. These two discoveries formed the basis for the emergence of optical fiber sensing technology. In the 1970s, with the development of optical fiber and optical fiber communication technology, optical fiber sensing technology began to take off. Over the past few decades, shape sensing has been an area of ​​interest for many researchers and industries. Shape sensing has been widely used in medical devices, industrial engineering, aerospace and other fields. Fiber optic shape sensors have gradually become a new hot spot in the field of shape sensing due to their strong anti-electromagnetic interference ability, long-distance multiplexing, high reliability and accuracy, and simple wiring. With its high sensitivity, small size, strong anti-interference ability, and no electromagnetic interference, it is widely used in structural monitoring in aerospace, marine and civil structures.

[0004] By measuring the strain of the fiber Bragg grating and combining it with the deformation reconstruction algorithm, the deformation of the flexible antenna can be accurately sensed indirectly, and the corresponding antenna performance changes can be analyzed to correct the deteriorated electrical performance caused by environmental factors. This research contributes to the high-precision and intelligent development of flexible antennas, and has far-reaching significance for the health monitoring, performance and life prediction of antenna structures.

[0005] Neural network is a computational model inspired by biological nervous system and widely used in artificial intelligence and machine learning. Its basic structure consists of many interconnected "neurons", simulating the working mode of neurons in the brain to process and transmit information. After training with a large amount of data, it can improve the real-time performance prediction of flexible antennas under different deformations while ensuring a certain accuracy, greatly reduce the time of calculation through mainstream electromagnetic simulation software, and make the application of flexible antennas more extensive.

[0006] The patent "A method for monitoring aircraft wing deformation based on fiber grating sensor network" (publication number: CN116625261A) proposes a method for reconstructing inverse finite element surfaces based on a fiber optic sensor network with triangular units, which can achieve real-time deformation monitoring, but lacks information on changes in electromagnetic performance parameters of radiating devices such as skin antennas with deformation.

[0007] Peng Kangning, "Antenna Performance Analysis and Optimization Based on Machine Learning" (Nanjing University of Posts and Telecommunications, 2022. DOI: 10.27251 / d.cnki.gnjdc.2022.000827.) predicted S by neural network method. 11 Parameters and radiation patterns, using different design size information of the antenna to predict changes in its electromagnetic performance, but the antenna involved is a traditional antenna and has not been deformed, and the impact of deformation on electromagnetic performance is not considered.

[0008] During the use of the flexible wearable antenna, the antenna will inevitably deform. In order to ensure the normal operation of the antenna, it is necessary to monitor the radiation performance of the antenna in real time. The existing simulation software obviously cannot meet the requirements in terms of real-time performance. Summary of the invention

[0009] In order to overcome the defects of the above-mentioned prior art, the purpose of the present invention is to provide a deformation monitoring and performance prediction system based on a flexible antenna. Aiming at the problem of deformation performance deterioration in the design and preparation process of flexible wearable antennas, the deformation monitoring module, the deformation performance prediction module and the neural network are deployed to perform deformation monitoring and performance prediction on the flexible antenna, and the current bending deformation and the antenna performance in the current bending state are displayed in real time to optimize the composite performance of the flexible antenna in a wearable environment; it has the advantages of strong anti-interference and high precision.

[0010] In order to achieve the above purpose, the technical means adopted by the present invention are:

[0011] A flexible antenna deformation monitoring and performance prediction system comprises a deformation monitoring module and a deformation performance prediction module; the deformation monitoring module comprises a fiber grating sensor array 2, a grating demodulation submodule 5, and a computer 6; wherein the fiber grating sensor array 2 collects deformation data signals of a flexible antenna 1, and is connected to the grating demodulation submodule 5 via an FC connector to demodulate the wavelength information of the flexible antenna 1; the wavelength information obtained by the grating demodulation submodule 5 is exchanged with the computer 6 via an Ethernet interface for data signal, the computer 6 processes the data in real time, and the wavelength change information is processed into curvature data information in real time by the deformation performance prediction module, and the entire surface is reconstructed by a surface reconstruction algorithm, and the computer 6 respectively displays the data of the deformation monitoring module and the deformation performance prediction module in real time through a deformation monitoring surface display interface 3 and an antenna performance prediction interface 4.

[0012] The fiber Bragg grating sensor array 2 has the following characteristics: when detecting wavelength changes, under the condition of constant temperature, when the structure to be measured is subjected to pressure or tension and produces bending deformation, it is considered to be pure bending in the non-interference state; the deformation curvature obtained at each position of the fiber Bragg grating sensor array 2, the relationship between the wavelength change and the curvature is as follows:

[0013]

[0014] Where λ is the central wavelength of the fiber Bragg grating, P e is the elastic-optical coefficient and h is the geometric parameter of the measured structure, both of which are constants, k is the curvature, Δλ is the change in the central wavelength, and its value is |λ0-λ|, λ0 is the initial wavelength.

[0015] The sensors in the fiber grating sensor array 2 are arranged vertically and horizontally to obtain sensing information in two dimensions, vertically and horizontally, so as to distinguish the bending direction of the flexible antenna.

[0016] The two-dimensional sensing information is obtained to distinguish the bending direction of the flexible antenna as follows: the flexible antenna is placed in a three-dimensional coordinate system, the antenna radiation surface is in the positive direction of the Z-axis of the antenna floor, and if the difference between the antenna radiation surface and the antenna floor is positive, the bending is inward, and if the difference is negative, the bending is outward; the antenna feed line is parallel to the X-axis, the XOZ plane is the E plane, and the YOZ plane is the H plane.

[0017] The deformation performance prediction module processes the wavelength change information into curvature data information in real time, and reconstructs the entire surface by using the surface reconstruction algorithm. The specific method is:

[0018] Step 1: Export the performance data of the flexible antenna under different deformation parameters from the electromagnetic simulation software, including center frequency, bandwidth, standing wave ratio and maximum gain;

[0019] Step 2: Preprocess the flexible antenna performance data derived in step 1 to obtain a deformation data set of the flexible antenna, and divide the data set into a training set and a test set;

[0020] Step 3: Input the deformation data set of the antenna obtained in step 2 into a fully connected neural network model including a fully connected layer, a batch normalization layer and a tanh activation function, and output the predicted data including the center frequency, bandwidth, standing wave ratio and maximum gain;

[0021] Step 4: Iteratively train the fully connected neural network model in step 3 and perform optimization;

[0022] Step 5: Input the measured deformation data in the deformation monitoring module into the fully connected neural network model trained in step 4 to predict the current performance parameters of the flexible antenna and output the results.

[0023] The specific method of step 2 is:

[0024] The performance data obtained in step 1, including the center frequency, bandwidth, standing wave ratio and maximum gain, are subjected to histogram analysis, and the values ​​of the performance data are counted. For the data whose center frequency, bandwidth, standing wave ratio and maximum gain are all positive numbers, they are scaled to between [0, 1]; for the data whose center frequency, bandwidth, standing wave ratio and maximum gain contain both positive and negative numbers, they are scaled to between [-1, 1], where positive and negative represent the bending direction; data larger than the preset value is reduced to prevent the gradient explosion phenomenon from hindering the training during the neural network training process; at the same time, data smaller than the preset value is amplified to ensure that the training is not slow due to too small gradients; the above preprocessed data form a data set and the data set is divided into a training set and a test set.

[0025] The fully connected neural network model described in step 3 uses the Adam optimizer to adjust the training step size according to the distribution of the training data.

[0026] The specific method of step 4 is:

[0027] The fully connected neural network model is trained according to the output of step 3. When the loss function gradually stabilizes, if the loss function value of the training set is less than half of the loss function value of the test set, the overfitting phenomenon of the network can be alleviated by increasing the abstention ratio between fully connected layers, adding L2 regularization constraints, adding truncated initialization during initialization, and reducing the number of layers or the number of neurons in each layer.

[0028] When the loss function is unstable and exceeds twice the average during multiple trainings, multiple short-cycle trainings are performed to reduce the overfitting of the fully connected neural network and preserve the fully connected neural network model with an obvious downward trend in the loss function; the training is performed until the loss function is stable and there is no phenomenon where the loss function value of the training set is less than half of the loss function value of the test set.

[0029] The specific method of step 5 is:

[0030] The measured deformation data in the deformation monitoring module is input into the trained fully connected neural network model to predict the current performance parameters of the flexible antenna and output the results.

[0031] In step 5, when data is input into the neural network, the input data is scaled according to the same scaling factor as in step 2; after the fully connected neural network prediction is completed and the output is obtained, the input data is inversely scaled according to the same scaling factor as when it was input.

[0032] When the predicted result obtained in step 5 is less than 0, a value of 0 is assigned.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. In the present invention, the signal processing submodule is used to integrate the fiber Bragg grating to demodulate the wavelength change, and the wavelength change information is processed into curvature data information in real time. The entire surface is reconstructed by the surface reconstruction algorithm, thereby realizing system miniaturization.

[0035] 2. The present invention solves the shortcoming of slow result generation of mainstream electromagnetic simulation software by reconstructing the wavelength information detected by the fiber grating sensor and displaying the current surface and sensor data in real time.

[0036] 3. The present invention trains a neural network model and combines it with current deformation data to predict and output antenna performance prediction data, thereby improving the real-time performance of the antenna's current performance display and forming a flexible antenna monitoring system that integrates deformation monitoring and performance prediction.

[0037] 4. The deformation performance prediction module of the present invention utilizes the deformation data output by the deformation detection module based on a fully connected neural network to predict the changes in four parameters including center frequency, bandwidth, standing wave ratio and maximum gain in real time, and compares them with the data of various deformation states of the corresponding type of antenna to obtain a rapid estimate of the performance parameters under the deformation state.

[0038] 5. The present invention uses a neural network model obtained by training a neural network with simulation and measured data, which can display the antenna performance under different deformation states in real time, including the prediction of center frequency, bandwidth, standing wave ratio VSWR and maximum gain.

[0039] In summary, the flexible antenna deformation monitoring and performance prediction system involved in the present invention has the advantages of strong anti-interference and high precision in hardware implementation; when the flexible antenna is deformed, it can not only display the current bending deformation in real time, but also through the deployment of neural networks, it can also display the antenna performance in the current bending state in real time, which plays an important reference role in adjusting the flexible antenna to the optimal performance state. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is the overall structure diagram of the deformation monitoring and antenna performance prediction system of the present invention.

[0041] Figure 2 This is a diagram of the arrangement and connection processing of the fiber grating sensor array of the present invention; wherein, Figure 2 (a) is a schematic diagram of the installed fiber Bragg grating sensor. Figure 2 (b) Fiber Bragg grating sensor installation positioning diagram.

[0042] Figure 3 It is the design index of the fiber grating sensor array of the present invention.

[0043] Figure 4 This is a physical picture of the fiber grating sensor array of the present invention.

[0044] Figure 5 The following is a photo of the physical system of the present invention.

[0045] Figure 6 This is the deformation monitoring interface of the present invention.

[0046] Figure 7 It is the performance prediction interface of the present invention.

[0047] Figure 8 The neural network training deployment process of the present invention.

[0048] Fig. 9 It is the fully connected block structure of the present invention.

[0049] Fig.10 It is the neural network structure of the present invention.

[0050] Fig.11 It is the S parameter test data of the antenna in the bending state; among them, Fig.11 (a) is bending parallel to the feeding direction, Fig.11 (b) Bending perpendicular to the feeding direction.

[0051] Fig.12 The deformation diagram of the antenna and cylindrical model is generated; among them, Fig.12 (a) The bending direction is perpendicular to the feeding direction. Fig.12 (b) The bending direction is parallel to the feeding direction.

[0052] Figure 1 In: 1. Flexible antenna; 2. Fiber Bragg grating sensor array; 3. Deformation monitoring surface display interface; 4. Antenna performance prediction interface; 5. Grating demodulation module; 6. Computer. DETAILED DESCRIPTION

[0053] The present invention is described in detail below with reference to the accompanying drawings.

[0054] See also Figure 1The present invention proposes a flexible antenna deformation monitoring and performance prediction system, including a deformation monitoring module and a deformation performance prediction module; the deformation monitoring module includes a fiber grating sensor array 2, a grating demodulation submodule 5, and a computer 6. The fiber grating sensor array 2 collects deformation data signals of the flexible antenna 1, and is connected to the grating demodulation submodule 5 through an FC connector to demodulate the wavelength information of the flexible antenna 1; the wavelength information obtained by the grating demodulation submodule 5 interacts with the computer 6 through an Ethernet interface, and the computer 6 processes the data in real time, and processes the wavelength change information into curvature data information in real time through the deformation performance prediction module, and reconstructs the entire surface through the surface reconstruction algorithm, and the computer 6 displays the data of the deformation monitoring module and the deformation performance prediction module in real time through the deformation monitoring surface display interface 3 and the antenna performance prediction interface 4. See Figure 5 , which is the physical construction of the flexible antenna deformation monitoring and performance prediction system of the present invention, corresponding to Figure 1 .

[0055] The fiber Bragg grating sensor array 2 has the following characteristics: when detecting wavelength changes, under the condition of constant temperature, when the structure to be measured is subjected to pressure or tension and produces bending deformation, it is considered to be pure bending in the non-interference state; the deformation curvature obtained at each position of the fiber Bragg grating sensor array 2, the relationship between the wavelength change and the curvature is as follows:

[0056]

[0057] Where λ is the central wavelength of the fiber Bragg grating, P e is the elastic-optical coefficient and h is the geometric parameter of the measured structure, both of which are constants, k is the curvature, Δλ is the change in the central wavelength, and its value is |λ0-λ|, λ0 is the initial wavelength.

[0058] The sensors in the fiber grating sensor array 2 are arranged vertically and horizontally to obtain sensing information in two dimensions, vertically and horizontally, so as to distinguish the bending direction of the flexible antenna.

[0059] See also Figure 2 , is a specific diagram of the fiber grating sensor array, arranged vertically and horizontally. Figure 2 (b) shows the installation and positioning of the fiber Bragg grating sensor. The center spacing of each sensor is 30 mm. Figure 2 (a) is a schematic diagram of the installed fiber Bragg grating sensor.

[0060] See also Figure 3 , are the specific parameters of the designed single fiber grating sensor. There are 8 sensor array elements in total, and the central wavelength of each element is 1531nm, 1534nm, 1537nm, 1540nm, 1543nm, 1546nm, 1549nm, and 1551nm.

[0061] See also Figure 4 , for the designed fiber grating sensor 5×3 array, 4 groups are used Figure 3 Designed sensor. In order to allow the sensor to bend without breaking, the length between the centers of the two sensor array elements of 1540nm and 1543nm was extended to 60mm, and one of the sensors was disconnected between the positions of 1540nm and 1543nm, thus realizing a 5×3 array object.

[0062] The two-dimensional sensing information is obtained to distinguish whether the bending of the flexible antenna is inward or outward, and whether the bending is parallel to the feeder or perpendicular to the feeder; specifically: the flexible antenna is placed in a three-dimensional coordinate system, the antenna radiation surface is in the positive direction of the Z axis of the antenna floor, and if the difference between the antenna radiation surface and the antenna floor is positive, the bending is inward, and if the difference is negative, the bending is outward; the antenna feeder is parallel to the X axis, the XOZ plane is the E plane, and the YOZ plane is the H plane.

[0063] The deformation performance prediction module processes the wavelength change information into curvature data information in real time, and reconstructs the entire surface through the surface reconstruction algorithm. The specific method is: based on the fully connected neural network, the deformation data output by the deformation monitoring module is used to predict the changes of four parameters including the center frequency, bandwidth, standing wave ratio and maximum gain in real time, and compares them with the data of various deformation states of the corresponding type of antenna to obtain the performance parameters under the deformation state that are quickly estimated;

[0064] Step 1: Export the performance data of the flexible antenna under different deformation parameters from the High Frequency Structure Simulator electromagnetic simulation software, including center frequency, bandwidth, standing wave ratio and maximum gain;

[0065] Step 2: Preprocess the flexible antenna performance data derived in step 1 to obtain a deformation data set of the flexible antenna, and divide the data set into a training set and a test set;

[0066] The performance data obtained in step 1, including center frequency, bandwidth, standing wave ratio and maximum gain, are subjected to histogram analysis, and the values ​​of each performance data are counted. For data where the center frequency, bandwidth, standing wave ratio and maximum gain are all positive numbers, they are scaled to between [0,1]; for data where the center frequency, bandwidth, standing wave ratio and maximum gain contain both positive and negative numbers (input deformation, positive and negative represent the bending direction), they are scaled to between [-1,1], where positive and negative represent the bending direction; the specific method is to directly divide by the absolute maximum value in the data. This step will reduce the data that is greater than the preset value, which can effectively prevent the gradient explosion phenomenon from hindering the training during the neural network training process; at the same time, the data that is less than the preset value will be amplified to ensure that the training will not be slow due to too small gradients; the above preprocessed data form a data set and the data set is divided into a training set and a test set.

[0067] The information of positive and negative curvature is contained in the curvature value through the positive and negative differences, so the size of the network input data is 1.

[0068] Step 3: Input the deformation data set of the antenna obtained in step 2 into a fully connected neural network model including a fully connected layer, a batch normalization layer and a tanh activation function, and output prediction data including center frequency, bandwidth, standing wave ratio and maximum gain; the fully connected neural network model described in step 3 uses the Adam optimizer to take into account multiple gradient changes during training to prevent easily falling into the local optimal solution, and adjusts the training step size automatically according to the distribution of data during training.

[0069] Step 4: Iteratively train the fully connected neural network model in step 3 and perform optimization;

[0070] The fully connected neural network model is trained according to the output of step 3. When the loss function gradually stabilizes, if the loss function value of the training set is less than half of the loss function value of the test set, the overfitting phenomenon of the network can be alleviated by increasing the abstention ratio between fully connected layers, adding L2 regularization constraints, adding truncated initialization during initialization, and reducing the number of layers or the number of neurons in each layer.

[0071] Since the input data length of this problem is 1, which is relatively small and carries limited information, whether subsequent training can run normally is sensitive to the randomness of initialization. When the loss function is unstable and exceeds twice the average in multiple trainings, multiple short-cycle trainings are performed to reduce the overfitting of the fully connected neural network and save the fully connected neural network model with an obvious downward trend in the loss function; train until the loss function is stable and there is no phenomenon that the loss function value of the training set is less than half of the loss function value of the test set.

[0072] Step 5: Input the measured deformation data in the deformation monitoring module into the fully connected neural network model trained in step 4 to predict the current performance parameters of the flexible antenna and output the results.

[0073] In step 5, when data is input into the neural network, the input data is scaled according to the same scaling factor as in step 2; after the fully connected neural network prediction is completed and the output is obtained, the input data is inversely scaled according to the same scaling factor as when it was input.

[0074] Scaling the input ensures that the neural network can receive data under the same metric and obtain the correct output; the inverse scaling of the output can convert the network output into actual physical units. In step 5, since the bandwidth parameter cannot be less than 0, when the predicted result is less than 0, it is assigned a value of 0.

[0075] After the curvature data information processed by the computer 6 is input into the deformation performance prediction module, the change of parameters including the center frequency, bandwidth, standing wave ratio and maximum gain can be predicted in real time, and compared with the data of the corresponding type of antenna in the undeformed state, the center frequency, bandwidth, standing wave ratio and maximum gain of the flexible antenna 1 in the deformed state and the relative change value can be quickly estimated.

[0076] See also Figure 6 This embodiment integrates data initialization function (calibration, init button), start / stop button (run / stop button), wavelength and deformation data export function (export), and can display the current deformation direction and deformation degree in real time. Frame represents the number of frames of the current collected data, radius represents the average bending radius of each sensor at present, and positive and negative represent its outward and inward bending. It has real-time channel detection function and displays the status and wavelength data of each channel of the sensor. Channel0-channel7 are 8 wavelength collection channels. Only 5 channels are used as shown in the figure, and channel 0 and channel 3 have 4 wavelength data corresponding to 4 sensor array elements, channels 2, 3, and 5 have 8 wavelength data corresponding to 8 sensor array elements, and channels without data default to 0; it has the function of automatically analyzing data to determine whether the bending is inward or outward, and whether the bending is along the E surface or the H surface, and it is intuitively displayed through the curved surface.

[0077] See also Figure 7 Through the detected deformation data, the changes in four parameters including center frequency, bandwidth, standing wave ratio and maximum gain can be predicted in real time. Figure 7 The center frequency, bandwidth, VSWR, maxgain in the image are compared with the static data of the corresponding type of antenna (delta). In deformationinfo, select the bending direction E / H plane, enter the bending radius, and click the solve button to get a quick estimate of the performance parameters under deformation (withdeformation).

[0078] Fully connected neural networks rely on GPU devices for training. When the model needs to be deployed on other platforms, in order to eliminate the dependence on GPU devices, other programming languages ​​are used to reconstruct the network. After the trained parameters are saved as a data file in a universal format, calculations can be performed without the GPU.

[0079] The proposed method is written in Python during training, and the network training mainly relies on the pytorch library. The deployed network is still written in Python, and the network parameters in the trained pth neural network model file are exported as npz files. The inter-layer calculation of the neural network is completed with the help of the numpy library, thus getting rid of the dependence on the GPU. In actual tests, on the i9-13900H processor, the neural network completes an inference in less than 1ms, which has sufficient real-time performance. Figure 8 , is the neural network training deployment process, the data set is the deformation curvature and direction, and the corresponding center frequency, bandwidth, standing wave ratio and maximum gain antenna parameters. After the trained neural network is deployed, the corresponding center frequency, bandwidth, standing wave ratio and maximum gain antenna parameter prediction data can be obtained by inputting the deformation data. The operation display interface corresponds to Figure 7 .

[0080] Fig. 9 It is a basic fully connected block in the designed neural network. Each of the fully connected blocks consists of a basic fully connected layer, a batch normalization layer, and a tanh activation function. In the basic fully connected layer, the matrix-vector multiplication and addition operation of wx+b is performed, and the batch normalization layer is used to eliminate the bias and the tanh activation function is used to limit the amplitude. In the figure, i and o represent a value, indicating the size of the data; x in each layer i and x o is a vector representing the input and output of the current layer; w and b represent the weight and bias of the network respectively. The weight is a matrix and the bias is a vector. They are fixed values ​​that are adjusted during training and do not change in the real-time system. In particular, in the batch normalization layer, mean and var represent the mean and variance of the data distribution. w bn and b bn The weights and biases in the representation layer. All four parameters are vectors. They are also adjusted during training and are fixed values ​​that do not change in the real-time system. Tanh is the mathematical tanh function, which calculates each value in the vector in turn.

[0081] Fig.10The overall structure of the designed neural network. After the first set of fully connected layers, batch normalization layers, and tanh activation functions convert the single data into a vector representation of length 120, the data passes through 7 stacked fully connected blocks in turn to form a deep neural network. The final output will pass through a fully connected layer in turn to convert the vector representation of 120 into a vector representation of 4 prediction parameters, and then pass through the tanh activation function to give the final prediction result. After reverse normalization, the expected antenna performance parameter change can be obtained, and then it is superimposed on the parameters in the static state to obtain the current predicted antenna performance parameters. The meaning of each parameter in the figure and Fig. 9 Same as in.

[0082] See also Fig.11 , the S parameter changes of the circularly polarized flexible antenna when the antenna is bent parallel to the feeding direction and perpendicular to the feeding direction are tested. When the fabric antenna is bent parallel to the feeding direction and the bending radius is 8cm (a), the spacing between adjacent resonant frequencies increases from 40MHz to 110MHz; when the bending radius increases to 12cm, the resonant frequency is better matched at 1.45GHz than when the bending radius is 8cm, and the frequency matching at 1.34GHz decreases. As shown in Figure (b), when bent perpendicular to the feeding direction, when the bending radius is 8cm, the resonant frequency increases to 1.44GHz, and the spacing between adjacent resonant frequencies increases from 40MHz to 110MHz; when the bending radius is 12cm, the coupling of the two resonant frequencies is improved, and the bandwidths of the two frequencies overlap each other without separation trend, which significantly improves the S parameters at 1.37GHz and reduces the S11 parameter by 11dB, but the overall bandwidth is reduced by about 10MHz compared to the unbent state.

[0083] See also Fig.12 To obtain the data of the antenna under deformation, the deformation produced by conforming it to the cylindrical model is shown. Fig.12 (a) The bending direction is perpendicular to the feeding direction. Fig.12 (b) The bending direction is parallel to the feeding direction.

[0084] Multiple technologies are implemented together to complete the construction of fiber Bragg grating sensor monitoring and neural network in the present invention, and a system integrating monitoring and prediction is formed through deformation information. Taking the antenna used in the present invention as an example, the time for traditional simulation software to calculate the result is generally 5 minutes. Through the training of neural network, the time can be reduced to about 1s, which improves the real-time performance of prediction. The present invention greatly improves the real-time prediction of antenna performance under the deformation of flexible antenna, and can quickly estimate the performance under the current deformation.

Claims

1. A flexible antenna deformation monitoring and performance prediction system, comprising a deformation monitoring module and a deformation performance prediction module; the deformation monitoring module comprises a fiber grating sensor array (2), a grating demodulation submodule (5), and a computer (6); wherein: The fiber grating sensor array (2) collects deformation data signals of the flexible antenna (1), and is connected to a grating demodulation submodule (5) through an FC connector to demodulate wavelength information of the flexible antenna (1); the wavelength information obtained by the grating demodulation submodule (5) is exchanged with a computer (6) through an Ethernet interface for data signal exchange, and the computer (6) processes the data in real time, and processes the wavelength change information into curvature data information in real time through a deformation performance prediction module, and reconstructs the entire surface through a surface reconstruction algorithm, and the computer (6) displays the data of the deformation monitoring module and the deformation performance prediction module in real time through a deformation monitoring surface display interface (3) and an antenna performance prediction interface (4), respectively.

2. A flexible antenna deformation monitoring and performance prediction system according to claim 1, characterized in that: The fiber Bragg grating sensor array (2) has the following characteristics: when detecting wavelength changes, under the condition of constant temperature, when the structure to be measured is subjected to pressure or tension and produces bending deformation, it is considered to be pure bending in the non-interference state; the deformation curvature obtained at each position of the fiber Bragg grating sensor array (2), the relationship between the wavelength change and the curvature is as follows: Where λ is the central wavelength of the fiber Bragg grating, P e is the elastic-optical coefficient and j is the geometric parameter of the measured structure, both of which are constants, k is the curvature, Δλ is the change in the central wavelength, and its value is |λ0-λ|, λ0 is the initial wavelength.

3. A flexible antenna deformation monitoring and performance prediction system according to claim 1, characterized in that: The sensors in the fiber optic grating sensor array (2) are arranged vertically and horizontally to obtain sensing information in two dimensions, vertically and horizontally, so as to distinguish the bending direction of the flexible antenna.

4. A flexible antenna deformation monitoring and performance prediction system according to claim 3, characterized in that: The two-dimensional sensing information is obtained to distinguish the bending direction of the flexible antenna as follows: the flexible antenna is placed in a three-dimensional coordinate system, the antenna radiation surface is in the positive direction of the Z-axis of the antenna floor, and if the difference between the antenna radiation surface and the antenna floor is positive, the bending is inward, and if the difference is negative, the bending is outward; the antenna feed line is parallel to the X-axis, the XOZ plane is the E plane, and the YOZ plane is the H plane.

5. The flexible antenna deformation monitoring and performance prediction system according to claim 1, characterized in that: The deformation performance prediction module processes the wavelength change information into curvature data information in real time, and reconstructs the entire surface by using the surface reconstruction algorithm. The specific method is: Step 1: Export the performance data of the flexible antenna under different deformation parameters from the electromagnetic simulation software, including center frequency, bandwidth, standing wave ratio and maximum gain; Step 2: Preprocess the flexible antenna performance data derived in step 1 to obtain a deformation data set of the flexible antenna, and divide the data set into a training set and a test set; Step 3: Input the deformation data set of the antenna obtained in step 2 into a fully connected neural network model including a fully connected layer, a batch normalization layer and a tanh activation function, and output the predicted data including the center frequency, bandwidth, standing wave ratio and maximum gain; Step 4: Iteratively train the fully connected neural network model in step 3 and perform optimization; Step 5: Input the measured deformation data in the deformation monitoring module into the fully connected neural network model trained in step 4 to predict the current performance parameters of the flexible antenna and output the results.

6. A flexible antenna deformation monitoring and performance prediction system according to claim 5, characterized in that: The specific method of step 2 is: The performance data obtained in step 1, including the center frequency, bandwidth, standing wave ratio and maximum gain, are subjected to histogram analysis, and the values ​​of each performance data are counted. For data whose center frequency, bandwidth, standing wave ratio and maximum gain are all positive numbers, they are scaled to between [0, 1]; for data whose center frequency, bandwidth, standing wave ratio and maximum gain contain both positive and negative numbers, they are scaled to between [-1, 1], where positive and negative represent the bending direction; data larger than the preset value is reduced to prevent the gradient explosion phenomenon from hindering the training during the neural network training process; at the same time, data smaller than the preset value is amplified to ensure that the training is not slow due to too small gradients; The preprocessed data is divided into a training set and a test set.

7. The flexible antenna deformation monitoring and performance prediction system according to claim 5, characterized in that: The fully connected neural network model described in step 3 uses the Adam optimizer to adjust the training step size according to the distribution of the training data.

8. The flexible antenna deformation monitoring and performance prediction system according to claim 5, characterized in that: The specific method of step 4 is: The fully connected neural network model is trained according to the output of step 3. When the loss function gradually stabilizes, if the loss function value of the training set is less than half of the loss function value of the test set, the overfitting phenomenon of the network can be alleviated by increasing the abstention ratio between fully connected layers, adding L2 regularization constraints, adding truncated initialization during initialization, and reducing the number of layers or the number of neurons in each layer. When the loss function is unstable and exceeds twice the average during multiple trainings, multiple short-cycle trainings are performed to reduce the overfitting of the fully connected neural network and preserve the fully connected neural network model with an obvious downward trend in the loss function; the training is performed until the loss function is stable and there is no phenomenon where the loss function value of the training set is less than half of the loss function value of the test set.

9. The flexible antenna deformation monitoring and performance prediction system according to claim 5, characterized in that: In step 5, when data is input into the neural network, the input data is scaled according to the same scaling factor as in step 2; after the fully connected neural network prediction is completed and the output is obtained, the input data is inversely scaled according to the same scaling factor as when it was input.

10. The flexible antenna deformation monitoring and performance prediction system according to claim 5, characterized in that: When the predicted result obtained in step 5 is less than 0, a value of 0 is assigned.

Images