Temperature field and curing degree field monitoring method and system in co-curing process of conformal skin antenna

Abstract

The invention provides a method and a system for monitoring a temperature field and a curing degree field in a co-curing process of a conformal skin antenna, and mainly solves the problem that data of the whole temperature field and the curing degree field cannot be obtained in the prior art. The scheme comprises the following steps: arranging a sensor and establishing a connection relationship between the sensor and a data line; respectively constructing a temperature predictor and a curing degree predictor comprising a parameterized physical information neural network, and carrying out parallel data-free training on the predictors; inputting the thermal convection coefficient and the coordinates of the to-be-monitored area into the two trained predictors, and preliminarily predicting the temperature and the curing degree of a measured point in the to-be-monitored area in co-curing; constructing an inverse estimator which is the same as the trained temperature predictor, and updating a heat convection coefficient of the inverse estimator by utilizing a preliminary prediction result; and inputting the updated heat convection coefficient and the coordinates of the to-be-monitored area into the two trained predictors to predict a final temperature field and a final curing degree field. The method can accurately and quickly obtain the data of the whole temperature field and the curing degree field, and can be used for improving and manufacturing a co-curing molding process of the conformal skin antenna.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of composite material manufacturing, and particularly relates to a temperature field and curing degree field monitoring method and system, which can be used for improving and assisting in manufacturing of a co-curing forming process of a conformal skin antenna. BACKGROUND

[0002] The conformal skin antenna is a composite component having the functions of detection, navigation, communication, positioning, etc. of a traditional antenna and meeting the design and aerodynamic requirements of a fuselage, including an outer skin composite material, an antenna, a support material and an inner skin composite material. In the preparation process, the antenna and the support material are embedded into the inner skin composite material and the outer skin composite material to be co-cured and formed by using a hot press tank process.

[0003] The field monitoring method of the co-curing process of the conformal skin antenna refers to real-time monitoring of the state of the conformal skin antenna to be cured in the curing process and recording and processing physical field data. The existing field monitoring methods can be divided into direct monitoring and indirect monitoring. Direct detection is to obtain the physical field data of the material surface and the interior by using contact sensors and non-contact sensors, but this method can only monitor the area covered by the sensors and cannot monitor the full field data, and a large number of contact sensors are embedded, which will affect the curing forming quality. Indirect monitoring is to predict the full field through a mechanism model or a data-driven model based on a small amount of monitoring data, which can significantly increase the field monitoring range and reduce the influence on the curing forming quality compared with direct monitoring.

[0004] The patent document with the publication number CN 118810076 A discloses a "composite component curing temperature field monitoring method", which corrects a physical information neural network model by using a small amount of sensor data to monitor the temperature field of the composite component. Although this method can realize full field monitoring, the model accuracy will be affected because the process parameters such as the heat convection coefficient cannot be updated in real time, and the curing degree cannot be monitored.

[0005] The patent document with the publication number CN 109827995 A discloses a "resin-based composite material curing degree online self-monitoring method", which uses a curing technology that can accurately measure the electrical energy transmitted to the interior of the composite component to monitor the input electrical power value and the average temperature change of the overall component, establishes a mechanism model based on the law of conservation of energy, and obtains the average curing degree of the overall component in real time. Although this method can represent the curing degree of the material, the mechanism model itself has poor generalization ability, and only the average curing degree of the material can be represented, so the curing degree field cannot be represented.

[0006] In summary, the existing field monitoring methods cannot accurately obtain the entire temperature field and curing degree field data in the case of deploying only a small amount of sensors. SUMMARY

[0007] The present application aims to overcome the defects of the prior art, and provides a temperature field and curing degree field monitoring method and system for a conformal skin antenna co-curing process, which can accurately obtain monitoring data of the entire temperature field and curing degree field with only a small number of sensors deployed.

[0008] The technical key of the present application is to establish a physical information neural network containing process parameters, update the network structure and process parameters based on the temperature and curing degree physical information loss of the conformal skin antenna co-curing process, and characterize the entire temperature field and curing degree field. The implementation scheme includes:

[0009] 1. A temperature field and curing degree field monitoring method for a conformal skin antenna co-curing process, wherein the conformal skin antenna is uniformly arranged with fiber grating sensors inside, and the method comprises the following steps:

[0010] (1) arranging the sensors and establishing the connection relationship between the sensors and the data transmission lines;

[0011] (2) constructing a temperature predictor and a curing degree predictor each comprising at least two parameterized physical information neural networks;

[0012] (3) simultaneously performing data-free training on the temperature predictor and the curing degree predictor based on the temperature physical information loss and the curing degree physical information loss using a back propagation algorithm, to obtain the trained temperature predictor and the trained curing degree predictor;

[0013] (4) inputting the heat convection coefficient and the space-time coordinates of the to-be-monitored region into the trained temperature predictor and the trained curing degree predictor, respectively, to preliminarily predict the temperature and the curing degree of the measured points in the to-be-monitored region during the co-curing process;

[0014] (5) collecting the actual temperature signals of the measured points during the conformal skin antenna co-curing process, and transmitting the filtered and demodulated signals to an upper computer to obtain the measured temperature data of the measured points;

[0015] (6) constructing an inverse estimator identical to the trained temperature predictor in (3) and fixing the network weights and biases of the inverse estimator, and setting the heat convection coefficient as an updateable learning variable;

[0016] (7) calculating the root mean square loss of the predicted temperature data in (4) and the measured temperature data of the measured points in (5), and inputting the loss into the inverse estimator in (6) to update the heat convection coefficient using a back propagation algorithm;

[0017] (8) input the updated heat convection coefficient and the spatiotemporal coordinates of the monitoring region into the trained temperature predictor and the solidification degree predictor again, respectively, to predict the temperature field and the solidification degree field of the whole monitoring region during the co-solidification process.

[0018] Further, the temperature predictor of (2) comprising at least two physical information neural networks comprises:

[0019] 2a) a first standard deep neural network with input of 3-dimensional tensor, output of 1-dimensional vector and containing input layer, hidden layer and output layer is selected ;

[0020] 2b) a second standard deep neural network with input of 2-dimensional vector, output of 1-dimensional vector and containing input layer, hidden layer and output layer is selected ;

[0021] 2c) the physical information loss of temperature is calculated according to the heat-chemical control equation and its initial conditions and boundary conditions ;

[0022] 2d) the physical information loss of temperature is embedded in the above two standard deep neural networks , respectively, to obtain two temperature physical information neural networks , , ;

[0023] 2e) the above two temperature physical information neural networks , are connected in parallel to form a temperature predictor , which is used to predict the temperature.

[0024] Further, the solidification degree predictor of (2) comprising at least two physical information neural networks comprises:

[0025] 2f) a third standard deep neural network with input of 3-dimensional tensor, output of 1-dimensional vector and containing input layer, hidden layer and output layer is selected ;

[0026] 2g) a fourth standard deep neural network with input of 2-dimensional vector, output of 1-dimensional vector and containing input layer, hidden layer and output layer is selected ;

[0027] 2h) the physical information loss of solidification degree is calculated according to the heat-chemical control equation and its initial conditions ;

[0028] 2i)In the above two standard deep neural networks , , loss of physical information of the degree of cure , ;

[0029] 2j)The above two deep neural networks of the degree of cure , are connected in parallel to form a degree of cure predictor , which is used to predict the degree of cure.

[0030] Further, the root mean square loss of the predicted temperature data in the (4) and the measured temperature data of the measured points in the (5) in the (7) is calculated and input to the inverse estimator in the (6), and the heat convection coefficient is updated using the backpropagation algorithm, which includes:

[0031] 7a)The training points of the heat-chemical loss with the determined independent variable of temperature, the training points in the loss of the initial condition of temperature, the training points in the loss of the boundary condition of temperature, and the existing heat convection coefficient are input into the trained temperature predictor , and the predicted temperature of all measured points is output;

[0032] 7b)According to the measured temperature of the measured points and the predicted temperature of all measured points , the root mean square loss is calculated:

[0033] 7c)The root mean square loss is input into the inverse estimator, and the gradient of the root mean square loss with respect to the heat convection coefficient , and the gradient of the root mean square loss with respect to the heat convection coefficient are calculated respectively by the existing Adam backpropagation algorithm;

[0034] 7d)According to the adaptive adjustment of the learning rate set in the Adam backpropagation algorithm, and based on and , the heat convection coefficient and of the inverse estimator are updated;

[0035] ​​7e) repeating steps 7a) - 7d) until the root mean square loss is minimized, resulting in an updated optimal heat convection coefficient .

[0036] 2. A temperature field and degree of cure field monitoring system for a co-curing process of a conformal skin antenna, comprising the following modules:

[0037] a predictor building module for building a temperature predictor comprising at least two parameterized physical information neural networks and a degree of cure predictor ;

[0038] a low-precision prediction module for predicting the temperature and the degree of cure of the measured points of the region to be monitored in the co-curing process

[0039] a data-free training module for simultaneously performing data-free training on the built temperature predictor and the degree of cure predictor ;

[0040] a sensor filtering module for filtering the collected signals of the fiber Bragg grating sensors

[0041] a sensor demodulation module for demodulating the filtered collected signals of the fiber Bragg grating sensors

[0042] a data preprocessing module for storing, displaying and converting the filtered and demodulated collected signals of the fiber Bragg grating sensors into measured temperature data that can participate in calculation ;

[0043] an inverse estimator building module for building an inverse estimator with the same network structure, weights and parameters as the temperature predictor, and only taking the heat convection coefficient as the learning variable

[0044] an inverse estimator loss calculation module for calculating the inverse estimator loss according to the predicted temperature data of the measured points and the measured temperature data

[0045] a heat convection coefficient updating module for updating the heat convection coefficient according to the calculated inverse estimator loss, to obtain an optimal heat convection coefficient

[0046] a high-precision prediction module for predicting the temperature field and the degree of cure field of the entire region to be monitored in the co-curing process

[0047] Compared with the prior art, the present application has the following advantages:

[0048] The present application can realize full-field prediction of temperature and curing degree under the condition of deploying only a small number of sensors, because the predictor containing the parameterized physical information neural network is constructed and data-free training is performed on it.

[0049] The present application improves the prediction accuracy of the temperature field and the curing degree field, because the inverse estimator containing the parameterized physical information neural network is constructed with the heat convection coefficient as the learning variable, and the real-time update of the heat convection coefficient by the inverse estimator is used to calibrate the predictor. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 Flow chart of the temperature field and curing degree field monitoring method for the co-curing process of the conformal skin antenna of the present application;

[0051] Figure 2 Overall structure diagram of the monitoring device used in the method embodiment of the present application;

[0052] Figure 3 Positioning diagram of the fiber Bragg grating sensor arranged in the method embodiment of the present application;

[0053] Figure 4 Structure diagram of the monitored airborne conformal skin antenna in the method embodiment of the present application;

[0054] Figure 5 Structure diagram of the temperature predictor and the curing degree predictor constructed in the method embodiment of the present application;

[0055] Figure 6 Block diagram of the temperature field and curing degree field monitoring system for the co-curing process of the conformal skin antenna of the present application; DETAILED DESCRIPTION

[0056] The specific embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0057] The temperature field and curing degree field monitoring method and system of the present application are performed for the co-curing process of the conformal skin antenna, which includes the airborne conformal skin antenna, the missile-borne conformal skin antenna, the ship-borne conformal skin antenna, etc. The present application describes the temperature field and curing degree field monitoring method and system for the co-curing process of the conformal skin antenna by taking the airborne conformal skin antenna as an example.

[0058] Embodiment one, temperature field and curing degree field monitoring method for the co-curing process of the airborne conformal skin antenna

[0059] As Figure 4As shown in the figure, the airborne conformal skin antenna 2 comprises an outer skin 21 of glass fiber reinforced epoxy resin-based prepreg laminates, a flexible array antenna 22, an aramid paper honeycomb support 24 and an inner skin 25 of carbon fiber reinforced epoxy resin-based prepreg laminates, the outer shape of the inner skin 25 is a curved surface in the shape of an inverted U, the aramid paper honeycomb support 24 is provided with a fixing groove 23 for the flexible array antenna, and is glued to the inner skin, the flexible array antenna 22 is placed in the fixing groove 23, and the outer skin 21 is glued to the flexible array antenna 22.

[0060] Referring to Figure 1 The implementation steps of the present example include the following:

[0061] Step 1, constructing a monitoring device.

[0062] As Figure 2 shown, the monitoring device comprises a plurality of Bragg fiber grating sensors 1, an airborne conformal skin antenna 2, a plurality of fiber grating sensor interfaces 3, a stabilized power supply 4, an upper computer 5, a fiber grating demodulator 6, a plurality of sealing blocks 7 and a heat pressure tank 8; the plurality of Bragg fiber grating sensors 1, the airborne conformal skin antenna 2 and the plurality of fiber grating sensor interfaces 3 are arranged inside the heat pressure tank 8, the stabilized power supply 4, the upper computer 5 and the fiber grating demodulator 6 are distributed outside the heat pressure tank 8, and the plurality of sealing blocks 7 are welded on the heat pressure tank 8 to isolate the heat exchange between the inside and outside of the heat pressure tank wall and ensure the uniform temperature inside the heat pressure tank.

[0063] Step 2, establishing a connection relationship between the sensors and the airborne conformal skin antenna and the data transmission line in the monitoring device.

[0064] A part of the plurality of Bragg fiber grating sensors 1 are arranged on the surface of the airborne conformal skin antenna 2, and the other part are arranged inside the airborne conformal skin antenna 2, each Bragg fiber grating sensor is connected to the external fiber grating demodulator 6 through different corresponding interfaces 3 and a plurality of sealing blocks 7, and the fiber grating demodulator 6 is connected to the upper computer 5 through a data line.

[0065] The present example selects, but is not limited to, 52 identical Bragg fiber grating sensors, the grating length of each is 10 mm, and the fiber diameter is 125 mm, wherein 20 are uniformly arranged on the surface 211 of the outer skin 21 of the airborne conformal skin antenna, 16 are uniformly arranged inside 212 the outer skin 21 of the airborne conformal skin antenna, and 16 are uniformly arranged inside 251 the inner skin 25 of the airborne conformal skin antenna, the Bragg fiber grating sensors are co-cured with the airborne conformal skin antenna to be cured after being arranged, and are not removed, as Figure 3 shown.

[0066] Step 3: Construct temperature predictors including two parameterized physical information neural networks and cure degree predictor .

[0067] Reference Figure 5 , the implementation of this step includes the following:

[0068] 3.1) Select the first standard deep neural network with 1 standard input layer, multiple standard hidden layers and 1 standard output layer , the input of the network is a 3D tensor , the output is dimensional vector , which has multiple standard hidden layers containing trainable neuron weights and neuron biases;

[0069] 3.2) Select a second standard deep neural network with 1 standard input layer, multiple standard hidden layers and 1 standard output layer , the input of the network is a 2-dimensional vector , the output is dimensional vector , which has multiple standard hidden layers containing trainable neuron weights and neuron biases;

[0070] 3.3) Physical information loss for calculating temperature based on the thermochemical governing equations and their initial and boundary conditions :

[0071] 3.3.1) At any 10 locations along the outer skin surface of the airborne conformal skin antenna with equal thickness no greater than the thickness of the airborne conformal skin antenna, uniformly sample 2000 data points covering the entire airborne conformal skin antenna test area. , and project it vertically onto a fixed plane coordinate system in space The data points formed by the projection are the thermo-chemical losses with temperature as the independent variable. Training points ;

[0072] 3.3.2) According to the determined Training Point Calculate the thermo-chemical losses with temperature as the independent variable in the thermo-chemical governing equation :

[0073] ,

[0074] in, for Training points quantity, is the curing rate coefficient, is the apparent activation energy, is the universal gas constant, 、 、 、 、 is the curing degree constant with different values, 、 Respectively training points The curing degree and temperature at 、 、 are three different intermediate variables, 、 Along Direction and The heat transfer coefficient in the direction, is the material density, is the specific heat capacity of the material, is the resin volume fraction of the material, is the resin density in the material, is the total heat released by the resin;

[0075] 3.3.3) 2000 sampling data points have been obtained middle The point is taken out and projected vertically onto a fixed plane coordinate system in space In the figure, the data points formed by the projection are the temperature initial condition loss Training points in ;

[0076] 3.3.4) According to the determined Training Point and initial conditions of the thermo-chemical governing equations, calculating the loss of initial temperature conditions ,

[0077] ,

[0078] in, is the temperature at time zero, for Training Point the number of For the training points temperature;

[0079] 3.3.5) 2000 sampling data points have been obtained The boundary points at the position of equal thickness of the airborne conformal skin antenna are taken out and vertically projected onto a plane coordinate system fixed in space The data points formed by the projection are the temperature boundary condition losses training points ;

[0080] 3.3.6) According to the determined training points and the boundary conditions of the thermo-chemical control equation, the loss of temperature boundary conditions is calculated :

[0081] ,

[0082] wherein, and are different numerical values of the heat convection coefficient, takes the value 45, takes the value 67.5, is the external temperature of the material, is the thermal conductivity coefficient of the material, is the training points number, is the temperature of the th training point ;

[0083] 3.3.7) According to the above thermo-chemical loss of temperature , the loss of temperature initial conditions and the loss of temperature boundary conditions , the loss of physical information of temperature is calculated :

[0084] ;

[0085] 3.4) The loss of physical information of temperature is respectively embedded in the above two standard deep neural networks , to obtain two temperature physical information neural networks , Compared with existing neural networks, these two temperature physical information neural networks do not need to rely on measured data during training due to the inclusion of the loss of physical information of temperature;

[0086] 3.5) The above two temperature physical information neural networks , are connected in parallel to form a temperature predictor for predicting the temperature , wherein, , are the outputs of the two temperature physical information neural networks, respectively. Due to the fusion of the two temperature physical information neural networks , Output 、 , thus improving the temperature prediction accuracy;

[0087] 3.6) Use a third-standard deep neural network with 1 standard input layer, multiple standard hidden layers, and 1 standard output layer , the input of the network is a 3D tensor , the output is dimensional vector , whose multiple hidden layers contain trainable neuron weights and neuron biases;

[0088] 3.7) Use a fourth-order deep neural network with a standard input layer, multiple standard hidden layers, and a standard output layer. , the input 2D vector of the network , the output is dimensional vector , whose multiple hidden layers contain trainable neuron weights and neuron biases;

[0089] 3.8) Calculation of physical information loss of curing degree based on thermo-chemical governing equations and their initial conditions :

[0090] 3.8.1) Based on the determined training points and the thermo-chemical governing equation, calculating the thermo-chemical loss of curing degree as the independent variable ,

[0091] ,

[0092] in, For training points quantity, is the curing rate coefficient, is the apparent activation energy, is the universal gas constant, 、 、 、 、 is the curing degree constant, 、 Respectively training points Temperature and curing degree, 、 、 are three different intermediate variables, 、 Along Direction and The heat transfer coefficient in the direction, is the material density, Cp is the specific heat capacity of the material, φ is the resin volume fraction of the material, ρ is the resin density in the material, Q is the total heat release of the resin;

[0093] 3.8.2) Calculate the loss of the initial condition of the degree of cure according to the determined training points positions and the initial condition of the thermal-chemical control equation:

[0094] ,

[0095] wherein, C(0) is the degree of cure at time zero, N is the number of training points , and Ci is the degree of cure at the position of the i-th training point of the material; 3.8.3) Calculate the loss of the physical information of the degree of cure according to the loss of the thermal-chemical control equation and the loss of the initial condition of the degree of cure:

[0096]

[0097] ;

[0098] 3.9) Embed the loss of the physical information of the degree of cure in the above two standard deep neural networks , respectively, to obtain two degree of cure physical information neural networks , , Compared with existing neural networks, the two degree of cure physical information neural networks do not need to rely on measured data during training because they contain the loss of the physical information of the degree of cure;

[0099] 3.10) Parallelly connect the above two degree of cure physical information neural networks , to form a degree of cure predictor for predicting the degree of cure , wherein, , are the outputs of the two degree of cure physical information neural networks respectively, and the degree of cure predictor improves the prediction accuracy of the degree of cure because it combines the outputs , of the two degree of cure physical information neural networks , . ​​​​​

[0100] Step 4: Physical information loss based on temperature and curing degree physical information loss , using the back propagation algorithm to predict the temperature and cure degree predictor Also perform data-free training.

[0101] 4.1) Randomly initialize the thermal convection coefficient with different values and ,in The initialization range is 30 – 60, The initialization range is 45 – 90;

[0102] 4.2) Thermo-chemical losses with temperature as the determined independent variable Training points , temperature initial condition loss Training points in , temperature boundary condition loss Training points in and heat convection coefficient and Input to the temperature predictor Meanwhile, the independent variable determined to be the thermo-chemical loss of curing degree Training points , loss of initial conditions of curing degree Training points in and heat convection coefficient and Input cure predictor middle;

[0103] 4.3) Temperature Predictor and cure degree predictor At the same time, the temperature of all training points is output separately and curing degree , and bring them into the temperature physical information loss and curing degree physical information loss middle;

[0104] 4.4) Calculate the temperature physical information loss separately through the Adam back propagation algorithm About the Temperature Predictor Network Weights and bias Gradient , and the loss of physical information of curing degree About the network weights of the solidification predictor and bias Gradient ;

[0105] 4.5) Adaptively adjust the learning rate according to the settings in the Adam back-propagation algorithm, and based on and These two gradients update the temperature predictor parameters 、 and the curing degree predictor network parameters 、 ;

[0106] 4.6) Repeat steps 4.1) - 4.5) until the temperature physics information is lost and curing degree physical information loss Minimum, get the optimal network weight and bias The trained temperature predictor with the optimal network weights and bias The trained solidification predictor of .

[0107] Step 5: The heat convection coefficient and the spatiotemporal coordinates of the area to be monitored , respectively input into the trained temperature predictor and cure degree predictor The temperature of the measured point in the area to be monitored during the co-curing process is preliminarily predicted. and curing degree .

[0108] Step 6: Collect the actual temperature signal of the measured point during the co-curing process of the airborne conformal skin antenna, and transmit it to the host computer after filtering and demodulation to obtain the measured temperature data of the measured point. .

[0109] 6.1) Use fiber grating sensors to collect the actual temperature signal of the measured point during the co-curing process of the airborne conformal skin antenna, filter it, and obtain the center wavelength of the filtered reflected light at the current temperature. :

[0110] ,

[0111] in, is the effective refractive index of the fiber Bragg grating sensor, is the spatial periodic modulation length of the refractive index in the fiber Bragg grating sensor;

[0112] 6.2) According to the center wavelength of the reflected light after filtering , the fiber Bragg grating demodulator is used to demodulate the temperature signal of the measured point during the co-curing process of the filtered airborne conformal skin antenna, and the demodulated temperature of the measured point is obtained. :

[0113] ,

[0114] wherein, is the temperature sensitivity coefficient of the fiber grating sensor, is the center wavelength of the reflected light of the fiber grating sensor at 20 degrees Celsius, the center wavelength of the grating is between 1520 mm - 1580 mm, corresponding to the temperature range 20 °C - 180 °C in the autoclave co-curing process.

[0115] Step 7, build the inverse estimator and train the inverse estimator.

[0116] Build the inverse estimator exactly the same as the trained temperature predictor in step 4, and fix its network weights and network biases, set the heat convection coefficient as an updatable learning variable;

[0117] Calculate the root mean square loss of the predicted temperature data in step 5 and the measured point measured temperature data in step 6 , input it to the inverse estimator in step 7, train the inverse estimator using the back propagation algorithm, and update the heat convection coefficient to :

[0118] 7.1) The training points of the heat-chemical loss whose independent variable has been determined as temperature , the training points in the temperature initial condition loss , the training points in the temperature boundary condition loss and the existing heat convection coefficient are input into the trained temperature predictor , and the predicted temperature of all measured points is output , wherein is 45, is 67.5; 7.2) According to the measured temperature of the measured point and the predicted temperature of all measured points , calculate the root mean square loss

[0119] :

[0120] ,

[0121] wherein, is the time sequence number of the current time, is the number of measurement points of the fiber grating sensor, a total of 52, is the time, the ​​The temperature measured by each fiber Bragg grating sensor measurement point after filtering and demodulation, The temperature predictor Time, The predicted temperature at each fiber Bragg grating sensor measurement point;

[0122] 7.3) The root mean square loss Input into the inverse estimator, and calculate the root mean square loss respectively through the existing Adam back propagation algorithm About the heat convection coefficient Gradient , and the root mean square loss About the heat convection coefficient Gradient ;

[0123] 7.4) Adaptively adjust the learning rate according to the settings in the Adam back-propagation algorithm, and based on and Update the heat convection coefficient of the inverse estimator and ;

[0124] 7.5) Repeat steps 7.1) – 7.4) until the RMS loss Minimum, get the updated optimal thermal convection coefficient .

[0125] Step 8: Update the thermal convection coefficient and the spatiotemporal coordinates of the area to be monitored , respectively input into the temperature predictor in (3) and cure degree predictor The temperature field of the entire monitored area during the co-curing process is predicted and curing degree field .

[0126] The flowchart representation or method representation of the above embodiments can be understood as representing a module, segment or portion of code including one or a group of executable instructions configured to implement the steps of a specific logical function or process. The present invention is not limited to the disclosed preferred embodiments, and its implementation can be performed in a different order than that shown or discussed, including performing functions in a substantially simultaneous manner according to the functions involved.

[0127] Example 2: Temperature field and curing degree field monitoring system for airborne conformal skin antenna co-curing process

[0128] Reference Figure 6This example includes a predictor construction module 6A, a data-free training module 6B, a low-precision prediction module 6C, a sensor filtering module 6D, a sensor demodulation module 6E, a data preprocessing module 6F, an inverse estimator construction module 6G, an inverse estimator loss calculation module 6H, a thermal convection coefficient update module 6I, and a high-precision prediction module 6J. The thermal convection coefficient update module 6I includes a gradient calculation submodule 6I1 and a gradient update submodule 6I2. The relationships and functions of these modules are as follows:

[0129] The predictor construction module 6A is used to construct a temperature predictor including two parameterized physical information neural networks. and cure degree predictor , so as to improve the prediction speed and accuracy of temperature and curing degree by including physical information loss based on temperature in the temperature predictor and physical information loss based on curing degree in the curing degree predictor;

[0130] The data-free training module 6B is used to train the constructed temperature predictor and cure degree predictor Perform data-free training;

[0131] The low-precision prediction module 6C is used to use the trained temperature predictor and cure degree predictor , predict the temperature of the measured point in the monitored area during the co-curing process and curing degree , the predicted temperature Transmitted to the inverse estimator loss calculation module 6H;

[0132] The sensor filtering module 6D is used to automatically filter the collected signal of the fiber Bragg grating sensor and transmit the filtered collected signal to the sensor demodulation module 6E;

[0133] The sensor demodulation module 6E is used to automatically demodulate the collected signal after filtering by the fiber Bragg grating sensor, and transmit the filtered and demodulated collected signal to the data preprocessing module 6F;

[0134] The data preprocessing module 6F is used to store, display and convert the collected signal after filtering and demodulation of the fiber grating sensor into the measured temperature data that can be used for calculation. , the measured temperature data Transmitted to the inverse estimator loss calculation module 6H;

[0135] The inverse estimator construction module 6G is used to construct the inverse estimator using only the heat convection coefficient To learn variables, network structure, weights, parameters and temperature predictor Exactly the same inverse estimator;

[0136] The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data .

[0137] The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data

[0138] and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data . The inverse estimator loss calculation module 6H is configured to calculate an inverse estimator loss based on the measured point temperature data and the predicted temperature data .

[0139] It should be noted that the above-mentioned functional modules can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, they can be realized in whole or in part in the form of program instruction products, such as the program instruction products of PyCharm software for the functions of the predictor construction module 6A, the data-free training module 6B, the low-precision prediction module 6C, the inverse estimator construction module 6G, the inverse estimator loss calculation module 6H, the heat convection coefficient updating module 6I, and the high-precision prediction module 6J, and the program instruction products of MATLAB software for the function of the data preprocessing module 6F. The program instruction product includes one or a set of program instructions. When the program instructions are loaded and executed on a computer, the flow or function is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The program instructions can be stored in a computer-readable and writable storage medium or transferred from one computer-readable and writable storage medium to another.

[0140] The direct coupling or communication connection between the modules shown or discussed in the embodiments can be realized by indirect coupling or communication connection of some interfaces, devices or modules. The functional modules and sub-modules in the embodiments can be dynamically in one processing component, or each module can be physically present alone, or two or more modules can be dynamically in one processing component. When the above-mentioned dynamic components are realized in the form of software functional modules and sold or used as independent products, they can also be stored in a computer-readable and writable storage medium. The storage medium can be a memory, a magnetic disk or an optical disk, etc.

[0141] The above description is only two specific examples of the present application and does not constitute any limitation on the present application. Obviously, for those skilled in the art, after understanding the content and principles of the present application, various modifications and changes in form and details can be made without departing from the principles and structures of the present application, but these modifications and changes based on the idea of the present application are still within the protection scope of the claims of the present application.

Claims

1. A method for monitoring the temperature field and curing degree field during the co-curing process of a conformal skin antenna, wherein fiber Bragg grating sensors are uniformly arranged inside the conformal skin antenna, characterized in that: include: (1) Arrange sensors and establish connections between them and data transmission lines; (2) constructing a temperature predictor and a curing degree predictor respectively including at least two parameterized physical information neural networks; (3) Based on the temperature physical information loss and the curing degree physical information loss, the temperature predictor and the curing degree predictor are trained without data using the back propagation algorithm to obtain the trained temperature predictor and curing degree predictor; (4) The thermal convection coefficient and the spatiotemporal coordinates of the monitored area are input into the trained temperature predictor and curing degree predictor, respectively, to preliminarily predict the temperature and curing degree of the measured points in the monitored area during the co-curing process; (5) Collect the actual temperature signal of the measured point during the co-curing process of the conformal skin antenna, and transmit it to the host computer after filtering and demodulation to obtain the measured temperature data of the measured point; (6) Construct an inverse estimator identical to the trained temperature predictor in (3) and fix its network weights and biases, setting the above-mentioned heat convection coefficient as an updateable learning variable; (7) Calculating the root mean square loss between the predicted temperature data in (4) and the measured temperature data of the measured point in (5), inputting it into the inverse estimator in (6), and updating the thermal convection coefficient using the back propagation algorithm; (8) The updated thermal convection coefficient and the spatiotemporal coordinates of the monitored area are input into the trained temperature predictor and curing degree predictor respectively to predict the temperature field and curing degree field of the entire monitored area during the co-curing process.

2. The method according to claim 1, characterized in that The (1) arrangement of sensors and establishment of a connection relationship between the sensors and the data transmission lines may include: Selecting a plurality of fiber grating sensors and evenly arranging them on the surface of the conformal skin antenna to be cured; Then, these fiber grating sensors arranged on the surface of the conformal skin antenna and the existing fiber grating sensors inside the conformal skin antenna to be cured are connected to the external fiber grating demodulator through different corresponding interfaces, and are connected to the host computer through data cables.

3. The method according to claim 1, characterized in that Said (2) constructs a temperature predictor including at least two physical information neural networks, and its implementation includes: 2a) Select the input as 3D tensor and the output as The first standard deep neural network with a dimensional vector and an input layer, a hidden layer, and an output layer ; 2b) Choose input as 2D vector and output as The second standard deep neural network with a dimensional vector and an input layer, a hidden layer, and an output layer ; 2c) Physical information loss for calculating temperature based on the thermochemical governing equations and their initial and boundary conditions ; 2d) In the above two standard deep neural networks 、 The physical information loss of temperature is embedded in , get two temperature physical information neural networks 、 ; 2e) Combine the two temperature physical information neural networks 、 Connect in parallel to form a temperature predictor , used to predict the temperature ,in, 、 are the outputs of two temperature physical information neural networks respectively.

4. The method according to claim 1, wherein Said (2) constructing a solidification degree predictor including at least two physical information neural networks, the implementation of which includes: 2f) Select the input as 3D tensor and the output as A third-dimensional deep neural network with an input layer, a hidden layer, and an output layer. ; 2g) Select the input as a 2-dimensional vector and the output as The fourth standard deep neural network with a dimensional vector and an input layer, a hidden layer, and an output layer ; 2h) Calculate the physical information loss of the curing degree based on the thermochemical control equation and its initial conditions ; 2i) In the above two standard deep neural networks 、 The physical information loss of the embedding solidification degree , we get two solidification degree physical information neural networks 、 ; 2j) Combine the above two solidification degree physical information neural networks 、 Parallel connection to form a curing degree predictor , used to predict the degree of cure ,in, 、 are the outputs of two solidification degree physical information neural networks respectively.

5. The method according to claim 3, characterized in that Physical information loss in calculating the temperature according to the thermochemical governing equation and its initial and boundary conditions in 2c) , whose implementation includes: (2c1) At any position of equal thickness along the conformal skin antenna that does not exceed the thickness of the conformal skin antenna, uniformly sample data points. And project it vertically into a plane coordinate system fixed in space The data points formed by the projection are the temperature thermo-chemical losses Training points ; (2c2) According to the determined Training points Calculate the thermo-chemical losses with temperature as the independent variable in the thermo-chemical governing equation : ， in, for Training points quantity, is the curing rate coefficient, is the apparent activation energy, is the universal gas constant, 、 、 、 、 is the curing degree constant with different values, 、 Respectively training points The curing degree and temperature at 、 、 are three different intermediate variables, 、 Along Direction and The heat transfer coefficient in the direction, is the material density, is the specific heat capacity of the material, is the resin volume fraction of the material, is the resin density in the material, is the total heat released by the resin; (2c3) The sampled data points have been obtained middle The point is taken out and projected vertically onto a fixed plane coordinate system in space In the figure, the data points formed by the projection are the temperature initial condition loss Training points in ; (2c4) According to the determined Training Point and initial conditions of the thermo-chemical governing equations, calculating the loss of initial temperature conditions , ， in, is the temperature at time zero, for Training Point the number of For the training points temperature; (2c5) The sampled data points have been obtained The boundary points at the position of equal thickness of the conformal skin antenna are taken out and projected vertically onto a plane coordinate system fixed in space The data points formed by the projection are the temperature boundary condition losses Training points in ; (2c6) According to the determined Training Point and boundary conditions of the thermo-chemical governing equations, calculating the losses of the temperature boundary conditions , ， in, and are the thermal convection coefficients with different values, is the external temperature of the material, is the thermal conductivity of the material, for Training Point the number of For the training points temperature; (2c7) According to the above thermo-chemical losses with temperature as the independent variable , loss of initial temperature conditions and temperature boundary condition losses , calculate the physical information loss of temperature : 。 6. The method according to any one of claims 4 to 5, characterized in that The physical information loss of the curing degree calculated according to the thermochemical control equation and its initial conditions in 2h) , whose implementation includes: (2h1) Based on the determined training points and the thermo-chemical governing equation, calculating the thermo-chemical loss of curing degree as the independent variable , ， in, For training points quantity, is the curing rate coefficient, is the apparent activation energy, is the universal gas constant, 、 、 、 、 is the curing degree constant, 、 Respectively training points Temperature and curing degree, 、 、 are three different intermediate variables, 、 Along Direction and The heat transfer coefficient in the direction, is the material density, is the specific heat capacity of the material, is the resin volume fraction of the material, is the resin density in the material, is the total heat released by the resin; (2h2) Based on the determined training points The position and initial conditions of the thermo-chemical control equations are used to calculate the loss of the initial conditions of the curing degree. : ， in, is the degree of solidification at time zero, For training points quantity, For material training points Degree of cure at the location; (2h3) According to the above independent variables, the thermo-chemical loss of curing degree and loss of initial conditions of curing degree , calculate the physical information loss of curing degree : 。 7. The method according to any one of claims 1 to 6, characterized in that In (3), based on the temperature physical information loss and the curing degree physical information loss, the back propagation algorithm is used to simultaneously perform data-free training on the temperature predictor and the curing degree predictor, and its implementation includes: 3a) Randomly initialize the thermal convection coefficient with different values and , the thermo-chemical losses whose independent variable is temperature are determined Training points , temperature initial condition loss Training points in , temperature boundary condition loss Training points in and heat convection coefficient and Input to the temperature predictor middle; 3b) Simultaneously determine the independent variable as the thermo-chemical loss of cure Training points , loss of initial conditions of curing degree Training points in and heat convection coefficient and Input cure predictor middle; 3c) Temperature Predictor and cure degree predictor At the same time, the temperature of all training points is output separately and curing degree , and bring them into the temperature physical information loss and curing degree physical information loss middle; 3d) Calculate the temperature physical information loss separately through the existing Adam back propagation algorithm About the Temperature Predictor Network Weights and bias Gradient , and the loss of physical information of curing degree About the network weights of the solidification predictor and bias Gradient ; 3e) Adaptively adjust the learning rate according to the settings in the Adam backpropagation algorithm and based on and These two gradients update the temperature predictor parameters 、 and the curing degree predictor network parameters 、 ; 3f) Repeat steps 3a) - 3e) until the temperature physical information is lost and curing degree physical information loss Minimum, get the optimal network weight and bias The trained temperature predictor with the optimal network weights and bias The trained solidification predictor of .

8. The method according to claim 1, characterized in that The actual temperature signal of the measured point collected during the co-curing process of the conformal skin antenna in (5) is filtered and demodulated, which includes: 5a) Use fiber Bragg grating sensors to filter the temperature signal of the measured point during the co-curing process of the conformal skin antenna to obtain the center wavelength of the filtered reflected light at the current temperature. : ， in, is the effective refractive index of the fiber Bragg grating sensor, is the spatial periodic modulation length of the refractive index in the fiber Bragg grating sensor; 5b) According to the center wavelength of the reflected light after filtering , the fiber Bragg grating demodulator is used to demodulate the temperature signal of the measured point during the co-curing process of the filtered conformal skin antenna, and the demodulated temperature of the measured point is obtained. : ， in, is the temperature sensitivity coefficient of the fiber Bragg grating sensor, is the central wavelength of the light reflected by the fiber Bragg grating sensor at 20 degrees Celsius.

9. The method according to any one of claims 1 to 7, characterized in that The root mean square loss between the predicted temperature data in (4) and the actual temperature data of the measured point in (5) is calculated in (7), and is input into the inverse estimator in (6), and the thermal convection coefficient is updated using the back propagation algorithm, which is implemented as follows: 7a) Thermo-chemical losses with temperature as the independent variable Training points , temperature initial condition loss Training points in , temperature boundary condition loss Training points in and the existing thermal convection coefficient Input the trained temperature predictor Output the predicted temperature of all measured points ; 7b) Based on the measured temperature of the measured point and the predicted temperature of all measured points , calculate the root mean square loss : ， in, is the time sequence number of the current moment, is the number of measurement points of the fiber Bragg grating sensor, For the Time, The temperature measured by each fiber Bragg grating sensor measurement point after filtering and demodulation, The temperature predictor Time, The predicted temperature at each fiber Bragg grating sensor measurement point; 7c) Convert the RMS loss Input into the inverse estimator, and calculate the root mean square loss respectively through the existing Adam back propagation algorithm About the heat convection coefficient Gradient , and the root mean square loss About the heat convection coefficient Gradient ; 7d) Adaptively adjust the learning rate according to the settings in the Adam back-propagation algorithm, and based on and Update the heat convection coefficient of the inverse estimator and ; 7e) Repeat steps 7a) – 7d) until the RMS loss Minimum, get the updated optimal thermal convection coefficient .

10. A temperature field and curing degree field monitoring system for the co-curing process of a conformal skin antenna, characterized in that: include: Predictor building block: used to build a temperature predictor that contains at least two parameterized physical information neural networks and cure degree predictor ; Low-precision prediction module: used to predict the temperature of the measured point in the monitored area during the co-curing process and curing degree ; No data training module: used to train the constructed temperature predictor and cure degree predictor Simultaneously perform data-free training; Sensor filter module: used to filter the collected signal of the fiber Bragg grating sensor; Sensor demodulation module: used to demodulate the collected signal after filtering of the fiber Bragg grating sensor; Data preprocessing module: used to store, display and convert the collected signal after filtering and demodulation of the fiber grating sensor into measured temperature data that can be used for calculation ; Inverse estimator building block: used to construct the inverse estimator based only on the heat convection coefficient To learn variables, network structure, weights, parameters and temperature predictor Exactly the same inverse estimator; Inverse estimator loss calculation module: used to predict temperature data based on the measured point The measured temperature data Compute the inverse estimator loss ; Thermal convection coefficient update module: used to update the inverse estimator loss according to the calculated Update thermal convection coefficient , get the optimal heat convection coefficient ; High-precision prediction module: used to predict the temperature field of the entire monitored area during the co-curing process and curing degree field .

11. The system according to claim 10, wherein: The heat convection coefficient updating module includes: Gradient calculation submodule: used to calculate the inverse estimator loss separately About the heat convection coefficient Gradient , and the inverse estimator loss About the heat convection coefficient Gradient ; Gradient update submodule: used to adaptively adjust the learning rate according to the settings in the Adam back-propagation algorithm, 、 , update the heat convection coefficient of the inverse estimator and , get the optimal heat convection coefficient .

Images