Satellite electromagnetic docking process generalization simulation method based on FCNN

Abstract

The invention discloses an FCNN-based satellite electromagnetic docking process generalization simulation method, and belongs to the field of spacecraft electromagnetic docking dynamics simulation, and the method comprises the following steps: firstly, generating magnetic field data of different models of electromagnetic actuators under random excitation current through finite element simulation, constructing and training an FCNN-D model, and obtaining a FCNN-D model; the prediction from actuator parameters, spatial position and current input to the magnetic field intensity is realized; secondly, collecting stress data of the actuator in an external magnetic field, and constructing an FCNN-F model by taking the model, the pose and the current of the actuator as input to predict electromagnetic force / torque; then dynamic simulation is carried out, the force / torque calculated by the FCNN-F is input in a stepping mode, the electromagnetic force is calculated according to the current pose in each step of simulation, and the satellite pose is updated until boundary conditions are met; and finally, the docking process is visualized. The FCNN is utilized to avoid an electromagnetic force / torque model, the defect that the application scale of electromagnetic simulation is limited is overcome, and a full-scale and high-precision electromagnetic docking simulation method is obtained.

Description

Technical Field

[0001] The present invention relates to the field of spacecraft electromagnetic docking dynamics simulation, and particularly to a method for simulating the interaction of electromagnetic docking actuators and the motion process of spacecraft electromagnetic docking based on a fully connected neural network. Background Art

[0002] With the development of microsatellites and space stations, on-orbit interactions of spacecraft are becoming increasingly frequent. Traditional jet propulsion docking technologies have problems such as fuel consumption and plume impact, which have a non-negligible impact on the lifespan of spacecraft. Electromagnetic docking, with its advantages of no fuel consumption and low docking impact, well meets the current requirements of high docking frequencies and small-scale on-orbit maneuvers. The main problems faced by electromagnetic docking lie in the strong nonlinear characteristics of electromagnetics and the coupling characteristics of electromagnetics and kinematics. However, there is no analytical solution for the Biot-Savart law in engineering in terms of electromagnetics. At the same time, most of the current magnetic dipole models used for theoretical analysis are only applicable at specific scales and cannot meet the accuracy requirements of the small-scale docking process. In addition, most of the current simulation analyses in related fields independently conduct two main parts: the electromagnetic finite element simulation analysis for electromagnetic actuators and the dynamics and control system simulation analysis for the on-orbit motion of microsatellites. The former is mostly limited to the static field of simple motion, and the latter mostly uses simple models obtained by linearization approximation, making it difficult to adapt to the highly coupled characteristics of the electromagnetics and kinematics of electromagnetic actuators. Therefore, it is necessary to seek a complex dynamic model that couples electromagnetics and kinematics and is universal for the entire motion scale through data-driven and numerical analysis methods.

[0003] Currently, most simulation analyses in related fields independently conduct the design of microsatellite electromagnetic actuator models and distributions and the interaction between electromagnetic actuators on the tracking satellite and the target satellite, making it difficult to combine and unify the two. The current single-satellite single-research process efficiency in the aerospace industry is difficult to meet the current space resource competition requirements mainly based on the construction of low-orbit microsatellite constellations. Therefore, it is particularly necessary to develop a general and highly integrated theoretical research platform for microsatellites with electromagnetic actuators to simulate the docking process of different types of satellites under different working conditions, so as to improve the research and development and management efficiency of series small satellites. Summary of the Invention

[0004] Aiming at the limitations of current electromagnetic simulation methods, the present invention proposes a generalization simulation method for satellite electromagnetic docking process based on FCNN, aiming to establish a multi-dimensional simulation method for the electromagnetic docking process of CubeSats with four electromagnetic docking actuators (energized solenoid electromagnets with iron cores) installed on the docking panel at different relative distance scales, different actuator models and configurations. (Hereinafter, the electromagnetic docking actuator will be simply referred to as the actuator).

[0005] The present invention establishes the following models based on the FCNN neural network:

[0006] FCNN-D model: The FCNN-D model is used to analyze the spatial distribution of the magnetic field generated by actuators of any model.

[0007] FCNN-F model: The FCNN-F model is used to analyze the electromagnetic force / torque situation of actuators of any model in a known external magnetic field.

[0008] A generalization simulation method for the satellite electromagnetic docking process based on the FCNN network includes the following steps:

[0009] Step S1, perform electromagnetic finite element simulations on a large number of actuators with different models by applying a large number of random excitation currents, obtain the spatial distribution of the electromagnetic fields they generate to form an initial magnetic field dataset, and process it to form an available magnetic field dataset.

[0010] Step S2, build a new fully connected neural network as the FCNN-D model, take the actuator model, spatial point position, and excitation current as the inputs of the network, take the magnetic field strength as the output of the network, and set the structure, parameters, activation function, and loss function of the neural network.

[0011] Step S3, use the dataset obtained in Step S1 to train the FCNN-D model, and judge whether the requirements are met through evaluation indicators. If the requirements are met, save the training results; if not, adjust the model and retrain.

[0012] Step S4, perform electromagnetic finite element simulations on a large number of actuators with different models and excitation currents to obtain the electromagnetic force / torque they receive at different positions and postures in the external magnetic field to form an initial force field dataset, and process it to obtain an available force field dataset.

[0013] Step S5, build a new fully connected neural network as the FCNN-F model, take the actuator model, its position and posture, and excitation current as the inputs of the network, take the electromagnetic force / torque received by the actuator as the output of the network, and set the structure, parameters, activation function, and loss function of the neural network.

[0014] Step S6, use the dataset obtained in Step S4 to train the FCNN-F model, and judge whether the requirements are met through evaluation indicators. If the requirements are met, save the training results; if not, adjust the model and retrain.

[0015] Step S7, build a dynamic model for the electromagnetic docking of cubic satellites, and set boundary conditions and the dynamic simulation step size.

[0016] Step S8: Input the actuator model to be analyzed and its initial spatial position and attitude into the FCNN-F model, set the excitation current signal to input the FCNN-F model, import the known external magnetic field, solve for the electromagnetic force / torque initially received by the actuator, perform dynamic simulation iteration, update the spatial position and attitude of the cube satellite using the output of the FCNN-F model, calculate the new position and attitude of the actuator and re-enter it into the FCNN-F model with other inputs unchanged, solve for the electromagnetic force / torque received by the actuator at the new position, perform step-by-step dynamic simulation, repeat the above process until the boundary conditions set in step S7 are reached, and determine whether the docking is successful or not.

[0017] Step S9: Visualize the electromagnetic docking motion process of the cube satellite calculated in step S8.

[0018] In step S1, the actuator model parameters include the installation positions of the centers of four solenoid electromagnets on the docking panel of the cube satellite, which are represented by four sets of two-dimensional coordinates; the inner and outer diameters of the solenoid, the length of the solenoid, the number of turns of the coil, the core material, the core diameter, and the core length can be represented by a 15-dimensional vector. The excitation current includes the excitation currents of 4 actuators and can be represented by a 4-dimensional vector. The spatial electromagnetic field distribution can be represented by a set of mappings of a series of corresponding three-dimensional coordinates of spatial points and the three-dimensional vectors of the magnetic field strength at those points, and this actuator should belong to the target cube satellite.

[0019] Use the well-known electromagnetic finite element simulation method to perform a large number of simulations. Each time, perform three-dimensional modeling according to the randomly set actuator model parameters and excitation current and save the corresponding simulation results of the spatial electromagnetic field distribution.

[0020] After classifying and binding the magnetic field strength, the corresponding spatial point positions, the set actuator model, and the excitation current in the initial magnetic field dataset, randomly arrange and perform normalization and standardization processing to form an available magnetic field dataset.

[0021] In step S2, the fully connected neural network model uses the supervised learning method and consists of an input layer; the first, second, third, and fourth hidden layers; and an output layer.

[0022] The input layer contains 22 input nodes corresponding to the actuator model parameters, the excitation current, and the spatial point positions respectively.

[0023] Each of the four hidden layers contains 512 neurons, and the activation function uses the ReLU function.

[0024] The output layer contains three output nodes corresponding to the three dimensions of the magnetic field strength vector respectively.

[0025] The neural network parameters include the learning rate, batch size, number of training epochs, and dropout rate, and the loss function uses the cross-entropy loss.

[0026] In step S3, the training process is to use 60% of the data randomly selected from the available magnetic field dataset obtained in step S1 to train the neural network, update the weights and biases of each neuron in the four hidden layers, and use the remaining 40% of the data to test the model. The evaluation index is the mean square error (MSE). When the relative error is lower than 5%, the model is considered to meet the accuracy requirements; otherwise, the number of model layers, the number of neurons in each layer, and the model parameters are adjusted and retrained until the requirements are met.

[0027] In step S4, the available spatial position of the actuator can be represented by the three-dimensional coordinates of the geometric center of the actuator in the global coordinates, and the spatial attitude of the actuator can be represented by the Euler angles of the body coordinate system of the cube satellite carrying the actuator relative to the global coordinate system. The electromagnetic force / torque received by the actuator can be represented by a six-dimensional vector including three dimensions of the electromagnetic force vector and three dimensions of the electromagnetic torque vector. This actuator should belong to the tracking cube satellite.

[0028] A large number of simulations are carried out using the well-known electromagnetic finite element simulation method. Each time, three-dimensional modeling is performed according to the randomly set actuator model parameters, excitation current, and the spatial position and attitude of the actuator, and the simulation results of the electromagnetic force / torque received by the corresponding actuator are saved.

[0029] The known external magnetic field is the electromagnetic field distribution obtained by matching and searching in the initial magnetic field dataset obtained in step S1 according to the actuator model parameters and excitation current set in each electromagnetic finite element simulation analysis. Setting the actuator parameters and excitation current of the target cube satellite and the tracking cube satellite (hereinafter referred to as the target satellite and the tracking satellite) to the same value is considered for the unity and symmetry of the cube satellite cluster.

[0030] After classifying, binding, randomly arranging, and normalizing and standardizing the electromagnetic force / torque received by the actuator, the corresponding spatial position and attitude of the actuator, and the set actuator model and excitation current in the initial force field dataset, an available force field dataset is formed.

[0031] In step S5, the fully connected neural network model uses the supervised learning method and consists of an input layer, the first, second, third, and fourth hidden layers, and an output layer.

[0032] The input layer contains 25 input nodes corresponding to the actuator model parameters, excitation current, and spatial position and attitude respectively.

[0033] Each of the four hidden layers contains 512 neurons, and the ReLU function is used as the activation function.

[0034] The output layer contains six output nodes corresponding to the electromagnetic force / torque received by the actuator.

[0035] The neural network parameters include the learning rate, batch size, number of training epochs, and dropout rate, and the cross-entropy loss is used as the loss function.

[0036] In step S6, the training process is to train the neural network using 60% of the data randomly selected from the available force field dataset obtained in step S4, update the weights and biases of each neuron in the four hidden layers, and use the remaining 40% of the data to test the model. The evaluation metric is the mean squared error (MSE). When the relative error is lower than 5%, the model is considered to meet the accuracy requirements; otherwise, the number of model layers, the number of neurons in each layer, and the model parameters are adjusted and retrained until the requirements are met.

[0037] In step S7, the electromagnetic docking dynamics model of the CubeSat adopts a simple Newton's law of motion model for ground tests targeting a fixed target satellite. The fixed target satellite is for facilitating the study of the relative motion of the chasing satellite. The key parameters of force and torque in Newton's law of motion are given by the FCNN-F model.

[0038] The boundary conditions are as follows: First, set a simulation time sufficient to complete the entire process of the CubeSat electromagnetic docking motion. Second, set an index that indicates the CubeSat docking is successful and terminates the simulation program in advance. Third, set an index that forces the termination of the simulation program in case of other special situations such as the CubeSat rubbing or colliding.

[0039] In step S8, the training of FCNN-D and FCNN-F should have been completed when performing this step.

[0040] The excitation current sequence is an actuator excitation current time sequence of interest. The length of this sequence depends on the set dynamics simulation time and time step. Each simulation time step should correspond to an excitation current (a four-dimensional vector) in chronological order. Denote the excitation current at the k-th simulation step as i(k).

[0041] When the electromagnetic docking simulation reaches the k-th step, the following process is carried out:

[0042] Input the actuator model to be analyzed into the FCNN-D model, and input i(k). Generate a set of sampling point spatial positions in a large enough area according to the spatial position data density of the initial magnetic field dataset obtained in step S1. Input each generated spatial position into the FCNN-D model, and run the FCNN-D model repeatedly to calculate the spatial distribution of the electromagnetic field generated by the actuator of the target satellite, which should be a set of mappings of the three-dimensional coordinates of a series of generated sampling points and the three-dimensional vector of the magnetic field strength at that location, denoted as H(k), where the magnetic field strength at a specific location is denoted as H(k,x,y,z).

[0043] Input the actuator model to be analyzed, the spatial position and attitude of the tracking star, and input i(k). Import H(k) calculated in process A as the known external magnetic field. Denote the spatial position and attitude of the tracking star in the k-th step of simulation as p(k), and the calculation result of FCNN-F as F(k).

[0044] According to the established electromagnetic docking dynamics model of the cube satellite, use inertial integration to calculate the change in the spatial position and attitude of the tracking star under the action of F(k), denoted as δp(k), and the six-degree-of-freedom velocity increment of the tracking star as δv(k). Then, the position and attitude of the tracking star in the (k + 1)-th step of simulation can be obtained as p(k + 1) = p(k) + δp(k), and the six-degree-of-freedom velocity is v(k + 1) = v(k) + δv(k).

[0045] Repeat processes A, B, and C until the boundary conditions are reached to obtain the time series of the spatial position, attitude, and velocity of the tracking star.

[0046] In step S9, based on the time series of the spatial position, attitude, and velocity of the tracking star obtained in step S8 and the given cube satellite model, the motion situation during the electromagnetic docking of the two cube satellites can be visualized, and based on this, it can be judged whether the docking is successful and the docking quality.

[0047] The advantages of the present invention compared with the prior art are as follows:

[0048] A generalization simulation method for the satellite electromagnetic docking process based on a fully connected neural network proposed by the present invention has high generalization ability. Through the neural network trained with high-precision finite element simulation data, when the platform conducts simulation analysis on a new actuator, it is not necessary to perform complex finite element analysis again, and the electromagnetic field generated by the actuator and the interaction between actuators with high precision can be given, which is convenient for comparing the electromagnetic performance of different models of actuators and improving the actuator design efficiency.

[0049] The interaction between neural networks and the co-simulation of the neural network and the numerical analysis platform organically combine the actuator model and configuration design, and the interaction between the electromagnetic actuators on the tracking star and the target star, realizing the unity of cube satellite electromagnetic docking management. At the same time, a high-precision electromagnetic docking dynamics model applicable to the full space scale with high coupling of electromagnetics and kinematics adapted to actual electromagnets is established, further enhancing the generalization ability and versatility of this method. Description of the Drawings

[0050] Figure 1 It is a schematic flow chart of a generalization simulation analysis method for the cube satellite electromagnetic docking process;

[0051] Figure 2 It is a schematic diagram of the double-star docking dynamics coordinate system of a specific implementation case of the present invention;

[0052] Figure 3 Schematic diagram of electromagnetic actuator modeling based on AnsysMaxwell software for a specific implementation case of the present invention; Specific implementation manner

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

[0054] As Figure 1 shown, the specific implementation process of the method proposed by the present invention is as follows:

[0055] (1) As Figure 2 shown, define the global motion reference coordinate system S 1 , the target star-fixed coordinate system S t , the tracking star-fixed coordinate system S c . Define the spatial position and attitude of the actuator or the tracking star in S 1 , and define the yOz plane of the cube star-fixed coordinate system as the installation plane of the actuator. Define the two-dimensional coordinates of the electromagnet installation position in this plane.

[0056] (2) Corresponding to steps S1, S2, and S3, build the FCNN-D model with the structure of a fully connected neural network, perform electromagnetic finite element simulation to obtain a data set, and train and optimize the FCNN-D model.

[0057] Perform electromagnetic finite element simulation analysis in AnsysMaxwell to obtain the data set for FCNN-D training:

[0058] A1. Adopt the well-known electromagnetic finite element simulation analysis process of AnsysMaxwell to model and simulate an actuator. Among the 15 dimensions of the actuator model parameters and the 4 dimensions of the excitation current, they are defined in the form of global variables;

[0059] A2. Generate at least 50,000 non-duplicate random 15-dimensional vectors within a reasonable range. The vector dimensions correspond to the 15 dimensions of the actuator model parameters respectively, and the same number of non-duplicate random 4-dimensional vectors. The vector dimensions correspond to the 4 dimensions of the excitation current respectively. The above two random vector sets will be used as the list sets for parametric scanning of the corresponding global variable parameters. Then generate the same number of non-duplicate random 3-dimensional vectors. The vector dimensions correspond to the 3 dimensions of the spatial point position respectively. Merge the three types of vectors in a random order to obtain a set formed by at least 50,000 random 22-dimensional vectors. This vector set will be used as the input part of the data set of the FCNN-D model.

[0060] A3. Start the electromagnetic finite element simulation analysis. Traverse the randomly selected actuator models obtained in B, and save the spatial distributions of the electromagnetic fields generated under the action of random excitation currents respectively. The magnetic field intensity vector at any position in the space can be read from them. Denote the set of the simulation results of the electromagnetic field spatial distributions as the field database.

[0061] A4. Traverse all 19-dimensional vectors in the input part of the FCNN-D model dataset, search in the field database to obtain the magnetic field intensity at a random position in the electromagnetic field generated by a certain randomly selected actuator model under a certain randomly selected excitation current, and then a set formed by at least 50,000 three-dimensional magnetic field intensity vectors can be obtained. This set will be used as the output part of the FCNN-D model dataset.

[0062] Train the above FCNN-D model and optimize the model according to the training results:

[0063] Randomly select 60% of the data in the FCNN-D model dataset as the training set of the FCNN-D model to train the model. Use the remaining 40% of the data as the validation set of the FCNN-D model, and adopt the mean square error (MES) as the evaluation index to test the training effect. The training and testing process of the fully connected neural network is a well-known technology and will not be elaborated in the present invention. Examine the training effect of the FCNN-D model. When the relative error is lower than 5%, it is considered that the model meets the accuracy requirements; otherwise, adjust the number of hidden layers, the number of neurons in each layer, and the model parameters of the model and retrain until the requirements are met.

[0064] (3) Corresponding to steps S4, S5, and S6, build the FCNN-F model with the structure of the fully connected neural network, perform electromagnetic finite element simulation to obtain the dataset, and train and optimize the FCNN-F model.

[0065] Perform electromagnetic finite element simulation analysis in Ansys Maxwell to obtain the dataset for training FCNN-F:

[0066] B1. Adopt the well-known Ansys Maxwell electromagnetic finite element simulation analysis process to model and simulate a certain actuator. Among them, 15 dimensions of the actuator model parameters, 4 dimensions of the excitation current, 3 dimensions of the spatial coordinates of the center point of the actuator installation panel, and 3 dimensions of the Euler angles of the normal vector of the actuator installation plane relative to the unit vector of the x-axis of the global coordinate system are defined in the form of global variables for parametric scanning. The specific positions of each actuator during modeling can be solved through simple constraints and well-known attitude calculation formulas.

[0067] Such as Figure 3As shown, the spatial position and attitude of the tracking star actuator are established. The electromagnetic actuator below in the figure is the actuator on the target star and only serves as a position reference, and no excitation current is applied to it.

[0068] B2. Generate at least 50,000 random 15-dimensional vectors after deduplication within a reasonable range. The vector dimensions respectively correspond to the 15 dimensions in the actuator model parameters; the same number of random 4-dimensional vectors after deduplication, the vector dimensions respectively correspond to the 4 dimensions in the excitation current, and the same number of random 6-dimensional vectors after deduplication, the vector dimensions respectively correspond to the 6 dimensions of the spatial position and attitude of the actuator. The above three sets of random vectors will be used as the list sets for parametric scanning of the corresponding global variables. Merge the three types of vectors in a random order to obtain a set formed by at least 50,000 random 25-dimensional vectors, and this vector set will be used as the input part of the dataset of the FCNN-F model.

[0069] B3. Write an Ansys Maxwell finite element simulation recording script using a well-known process, and add a part for reading the external magnetic field file in the script. In the field database obtained when establishing the FCNN-F model, search for the electromagnetic field distribution in the space corresponding to the actuator model and excitation current search space traversed in the B2 vector set, and import this file into the simulation scenario as the known external magnetic field.

[0070] B4. Start the electromagnetic finite element simulation analysis, and save the electromagnetic force / moment received by a random model actuator at a random spatial position and attitude in the known electromagnetic field generated by the same model actuator under the same excitation current under the action of a random excitation current obtained from each parametric scan. Subsequently, a set formed by at least 50,000 6-dimensional electromagnetic force / vector vectors can be obtained, and this set will be used as the output part of the dataset of the FCNN-F model.

[0071] Train the FCNN-F model and optimize the model according to the training results.

[0072] (4) Corresponding to step S7, set the simulation parameters and perform dynamic modeling.

[0073] Set the simulation parameters, that is, the boundary conditions. Set the simulation time to 10 s, set the simulation step size to 0.01 s, and set the condition for ending the simulation in advance when the four pairs of electromagnets in the two cube satellite electromagnetic docking actuators are in contact with each other, that is, the distance is close to 0 and the radial misalignment of each pair of electromagnets does not exceed 2 mm. Detect the envelopes of the two cube satellites. When their envelopes intersect but do not meet the docking success condition, it is considered that an accident has occurred, and the simulation will be forced to terminate at this time. When it is detected that the simulation reaches the limit time or the cube satellites move to a sufficient distance so that the interaction between the actuators is close to 0, the simulation terminates at this time.

[0074] Corresponding to step S8, based on the optimized FCNN-D model and FCNN-F model after training, with the help of numerical analysis platforms such as Python, a joint simulation of the electromagnetic docking process of the cube satellite to be analyzed is carried out. Through step-by-step iteration, the entire motion process is obtained, and whether the docking is successful is judged.

[0075] Specific process of the joint simulation:

[0076] Generate the time series of the excitation current to be analyzed. Denote the excitation current at the k-th step of the simulation as i(k). In the example of the present invention, a step current signal with an amplitude of 1 A is applied to all electromagnets, and then the excitation current in each simulation step is i(k)=[1 1 1 1] T (unit: A), and then the time series of the excitation current can be expressed as

[0077] Start the joint simulation. When the simulation takes one step forward (when the simulation steps to the k-th step), perform the following operations:

[0078] C1. Input the type of the actuator to be analyzed into the FCNN-D model, and input i(k). Generate a set of spatial position points of a large enough area according to the sampling point density in the electromagnetic field spatial distribution file obtained when acquiring the FCNN-D model dataset. Input each generated spatial position into the FCNN-D model, run the FCNN-D model repeatedly, and save the spatial distribution of the electromagnetic field generated by the actuator of the target satellite as a csv file, which should be a set of mappings of the three-dimensional coordinates of a series of generated sampling points and the three-dimensional vector of the magnetic field strength at that place, denoted as H(k), representing the spatial magnetic field distribution generated by the target satellite under the action of the excitation current of the actuator at the k-th step.

[0079] C2. Input the type of the actuator to be analyzed, the spatial position and attitude of the tracking satellite into the FCNN-F model, and input i(k). Import H(k) as the known external magnetic field. Denote the spatial position and attitude of the tracking satellite at the k-th step of the simulation as p(k), and the calculation result of FCNN-F as F(k).

[0080] C3. According to the established dynamic model of cube satellite electromagnetic docking, the inertial integral is used to calculate the spatial position and attitude increment of the tracking satellite under the action of F(k), denoted as δp(k), and the six-degree-of-freedom velocity increment of the tracking satellite is denoted as δv(k). Then, the position and attitude of the tracking satellite at the (k + 1)-th simulation step can be obtained as p(k + 1) = p(k) + δp(k), and the six-degree-of-freedom velocity of the tracking satellite is v(k + 1) = v(k) + δv(k), thus realizing the alternating step-by-step closed-loop joint simulation process between the interactive neural network model and the data analysis platform. In the example of the present invention, Python is used for dynamic numerical analysis. The spatial position and attitude of the cube satellite and the six-degree-of-freedom velocity at the first simulation step are used as the simulation initial conditions, which are artificially set as [200 cm 20 cm -30 cm 10° -20° 15°], [0 0 0] m / s, [0 0 0] deg / s.

[0081] C4. Repeat the above operations until the simulation boundary conditions are reached. Then, the spatial position and attitude of the cube satellite and the six-degree-of-freedom velocity within each simulation step have been calculated. Subsequently, the six-degree-of-freedom motion time series of the tracking satellite can be expressed as

[0082] Detect the simulation termination conditions triggered during the above simulation process to determine whether the docking is successful.

[0083] (5) Corresponding to step S9, perform visualization processing on the entire motion process obtained in (4).

[0084] The visualization processing process is to read the six-degree-of-freedom time series of the tracking satellite obtained from the joint simulation and use the built-in plotting software in Python to plot the waveform of the spatial position and attitude change of the tracking satellite and the waveform of the six-degree-of-freedom velocity.

Claims

1. A generalized simulation method for satellite electromagnetic docking process based on FCNN, characterized by: The method comprises the following steps: Step S1, applying a large number of random excitation currents to a large number of actuators of different models to perform electromagnetic finite element simulation, obtaining the spatial distribution of the electromagnetic field generated by them to form an initial magnetic field data set, and forming a usable magnetic field data set after processing. Step S2, build a new fully connected neural network as the FCNN-D model, take the actuator model, spatial point position, and excitation current as the network input, take the magnetic field strength as the network output, and set the structure, parameters, activation function, and loss function of the neural network. Step S3, using the data set obtained in step S1 to train the FCNN-D model to determine whether the requirements are met through evaluation indicators. If the requirements are met, the training results are saved. If not, the model is adjusted and retrained. Step S4, performing electromagnetic finite element simulation on a large number of actuators with different models and excitation currents to obtain the electromagnetic forces / forces received by them at different positions and postures in the external magnetic field to form an initial force field data set, and obtaining a usable force field data set after processing. Step S5, build a new fully connected neural network as the FCNN-F model, take the actuator model, its position and posture, and excitation current as the input of the network, take the electromagnetic force / torque exerted on the actuator as the output of the network, and set the structure, parameters, activation function, and loss function of the neural network. Step S6, using the data set obtained in step S4 to train the FCNN-F model, and judging whether the requirements are met through evaluation indicators, if the requirements are met, the training results are saved, if not, the model is adjusted and retrained. Step S7, building a dynamic model for electromagnetic docking between satellites, setting boundary conditions and dynamic simulation steps. Step S8, input the actuator model to be analyzed and its initial spatial position and posture into the FCNN-F model, set the excitation current signal to input the FCNN-F model, import the known external magnetic field, solve the electromagnetic force / torque initially received by the actuator, perform dynamic simulation iterations, use the output of the FCNN-F model to update the spatial position and posture of the cubic satellite, calculate the new position and posture of the actuator and re-input it into the FCNN-F model, keep other inputs unchanged, solve the electromagnetic force / torque received by the actuator in the new position, perform step dynamics simulation, repeat the above process until the boundary conditions set in step S7 are met, and determine whether the docking is successful or not. Step S9, visualizing the cubic satellite electromagnetic docking motion process calculated in step S8.

2. A generalized simulation method for satellite electromagnetic docking process based on FCNN as claimed in claim 1, characterized in that: There is data interaction between the FCNN-D model and the FCNN-F model described in step S2 and step S5 rather than independent existence. The specific connection is that the external magnetic field data relied on by the FCNN-F model calculation needs to be calculated by the FCNN-D model. Splitting a complex problem into two interconnected neural network models with clear physical meanings enhances the interpretability and credibility of the model.

3. A generalized simulation method for satellite electromagnetic docking process based on FCNN as claimed in claim 1, characterized in that: The introduced fully connected neural network, based on the universal approximation theorem of neural networks, provides a high-precision model applicable to all scales for the analysis of complex electromagnetic fields and electromagnetic forces with strong nonlinearity and high coupling with kinematics.

4. A generalized simulation method for satellite electromagnetic docking process based on FCNN as claimed in claim 1, characterized in that: The introduced fully connected neural network has a high generalization ability for models within a reasonable range due to its data-driven characteristics. It can bypass the complex modeling and analysis process and directly give the electromagnetic performance of electromagnetic docking actuator models that have never appeared, thereby improving the efficiency of model screening and optimization design of electromagnetic docking actuators.

5. The generalized simulation method of satellite electromagnetic docking process based on FCNN as claimed in claim 1, characterized in that: The idea of ​​joint simulation is to establish a rigid body kinematics simulation model in the numerical analysis platform and a series network model formed by connecting the FCNN-D neural network and the FCNN-F neural network in series for data interaction, including the numerical analysis platform transmitting the simulated pose information of the cubic satellite and the electromagnetic docking actuator to the series network model, and the series network model transmitting the simulated interaction between the electromagnets to the numerical analysis platform. In this process, when the numerical analysis platform is simulated step by step, the pose output of the previous simulation step will be used as the electromagnetic simulation electromagnetic docking actuator pose information of the series network model in the current simulation step, and the calculation results of the series neural network will be used as the external force and external torque of the dynamic model in the current simulation step to update the cubic satellite pose information of the current simulation step, and then the signal flow forms a dynamic closed-loop structure of step iteration, thereby making up for the deficiency that a single electromagnetic or dynamic simulation is difficult to adapt to the highly coupled characteristics of the electromagnetic kinematics of the electromagnet.

6. A generalized simulation method for satellite electromagnetic docking process based on FCNN as claimed in claim 1, characterized in that: During the joint simulation process, the excitation current is supported to traverse the input series network model in the form of a time series, and then the external control system is allowed to control the current, that is, to change the electromagnet current value of each simulation step according to the preset control law. The control law is a control method used for simulation of specific working conditions.

Images