A model predictive control-based electromechanical composite constant force system control method

Abstract

The application discloses a kind of electromechanical compound constant force system control method based on model predictive control, including including following steps: S1, training set data acquisition;S2, based on BP neural network establishes the dynamics model of system;S3, the data collected in step S1 is input into the dynamics model of system to complete model training;S4, model predictive control controller design, the state of electromechanical compound constant force system is identified using the model trained in step S3, and the optimal control force is calculated.The application adopts the above-mentioned electromechanical compound constant force system control method based on model predictive control, improves the constant force output precision and response speed of electromechanical compound constant force spring support, and further realizes high-precision vertical gravity compensation, improves the authenticity of spacecraft ground simulation.

Description

Technical Field

[0001] This invention relates to the field of zero-gravity simulation technology for spacecraft on the ground, and in particular to a control method for an electromechanical composite constant force system based on model predictive control. Background Technology

[0002] In recent years, my country has made rapid progress in the aerospace field, achieving great success with missions such as Chang'e 5, the Tianwen Mars probe, and the Chinese space station. At the same time, the difficulty of space missions and the complexity of spacecraft are also increasing. Therefore, technologies such as effectively revealing the impact of various disturbances through ground testing and verifying special spacecraft control methods have become crucial for improving the reliability of spacecraft operations.

[0003] High-fidelity ground testing is of great significance for complex, large spacecraft. Currently, the most mature solutions for ground testing only achieve five degrees of freedom, namely, horizontal translation along two axes and rotation along three axes through air buoyancy technology. The sixth degree of freedom, namely vertical zero-gravity translation, is currently mostly achieved using an electromechanical composite constant-force spring support system. However, this system suffers from high nonlinearity, low control accuracy, and slow response, and these problems urgently need to be addressed. Summary of the Invention

[0004] The purpose of this invention is to provide a control method for an electromechanical composite constant force system based on model predictive control, which improves the constant force output accuracy and response speed of the electromechanical composite constant force spring support, thereby achieving high-precision vertical gravity compensation and improving the realism of spacecraft ground simulation.

[0005] To achieve the above objectives, this invention provides a control method for an electromechanical composite constant force system based on model predictive control, comprising the following steps:

[0006] S1. Training set data collection;

[0007] S2. Establish a dynamic model of the system based on a BP neural network;

[0008] S3. Input the data collected in step S1 into the dynamic model of the system to complete the model training;

[0009] S4. Model predictive control controller design: The model trained in step S3 is used to identify the state of the electromechanical composite constant force system and calculate the optimal control force.

[0010] Preferably, in step S1, the collected data includes a random current I that is set to vary from time to time for the linear motor. k With the roller support of the constant force spring system moving, the displacement h of the roller support at N different times is collected. k Speed ​​vk acceleration a k and the system output force measured by the load force sensor The displacement is the distance between the initial position and the current position of the roller support.

[0011] Preferably, in step S2, a dynamic model of the electromechanical composite constant force system is established using a BP neural network, and I... k h k v k a k As the input to the BP neural network, denoted as Will As the output of the BP neural network.

[0012] A backpropagation (BP) neural network consists of an input layer, hidden layers, and an output layer. Therefore:

[0013] The input of the input layer is

[0014] Input of hidden layer and output They are respectively:

[0015]

[0016]

[0017] Where, ω ij The connection weights θ between input layer neurons and hidden layer neurons represent the connection weights between them. j is the threshold of the hidden layer neuron node, f() is the hidden layer transfer function, M is the dimension of the input layer, and i takes the value of 4;

[0018] Input of the output layer and output They are respectively:

[0019]

[0020]

[0021] Where, ω jk α represents the connection weights between hidden layer neurons and output layer neurons. k The threshold of the output layer neuron node is represented by g(), the output layer transfer function is represented by P, and the dimension of the hidden layer is represented by j = 1, 2, ..., P.

[0022] Preferably, step S3 includes the following steps:

[0023] (1) The I collected in step S1 k hk v k a k and As the training set for the BP neural network recognition model, I k h k v k a k For input, The output is the truth value;

[0024] (2) The I collected in step 1 k h k v k a k As the input to the input layer of the BP neural network recognition model, denoted as k = 1, 2, ..., N, where the input information is the current I. k Displacement h k Speed ​​v k acceleration a k That is, four-dimensional data, therefore i = 1, 2, 3, 4;

[0025] (3) Input the training set The input is fed into a backpropagation (BP) neural network to obtain the network output.

[0026] (4) The total error E of the BP neural network identification model is obtained using the following cost function:

[0027]

[0028] The parameters of the BP neural network identification model are adjusted using the gradient descent method to reduce its total error E. When the total error E reaches its minimum, the training of the BP neural network identification model is complete.

[0029] Preferably, in step S4, the specific steps for calculating the optimal control force are as follows:

[0030] (1) Collect the displacement h, velocity v, and acceleration a of the roller bracket at the current moment;

[0031] (2) Generate M random current inputs I k ;

[0032] (3) Input the displacement h, velocity v, acceleration a and the M generated currents collected in step S1 into I k The data are divided into M groups, which are used as network inputs and fed into the BP neural network dynamics model trained in step S3 to obtain M corresponding network outputs.

[0033] (4) Obtain the M items Each is related to the desired system output force F aim Compare and find the one that minimizes the reward function. Further find the corresponding current input I k This current is used as the control current I of the linear motor. The reward function R is defined as follows:

[0034]

[0035] Therefore, the present invention employs the above-mentioned model predictive control-based electromechanical composite constant force system control method, and its technical effects are as follows:

[0036] (1) The electromechanical composite constant force spring bracket can achieve near-zero stiffness force output through the spring and cam, and the active force compensation method of linear motor and sensor reduces the influence of factors such as processing error and spring accuracy, thereby achieving more accurate force output.

[0037] (2) In order to achieve higher precision constant force output, the dynamic model of the electromechanical composite constant force spring support system is identified by using a BP neural network. That is, by measuring the position, velocity, acceleration of the constant force spring support, the compensation force of the linear motor and the output force of the corresponding electromechanical composite constant force spring support system, the dynamic model of the electromechanical composite constant force spring support system is identified.

[0038] (3) When controlling the electromechanical composite constant force spring support system, the position, velocity and acceleration of the constant force spring support are measured and the linear motor compensation force is calculated according to the expected output force of the system for compensation control. Among them, the BP neural network can identify the dynamic model of the electromechanical composite constant force spring support system with strong nonlinearity and irregularity. The use of model predictive control can improve the response speed and control accuracy of the system.

[0039] (4) This invention improves the constant force output accuracy and response speed of the electromechanical composite constant force spring bracket, thereby achieving high-precision vertical gravity compensation and improving the realism of spacecraft ground simulation.

[0040] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0041] Figure 1 This is a front view of the main structure of the electromechanical composite constant force spring support system.

[0042] Figure 2 A three-dimensional schematic diagram of the main structure of the electromechanical composite constant force spring support system;

[0043] Figure 3A schematic diagram illustrating the principle of a BP neural network for pre-learning predictive control of a model.

[0044] Figure 4 This is a block diagram illustrating the principle of a control method for an electromechanical composite constant force system based on model predictive control.

[0045] Figure Labels

[0046] 1. First spring; 2. Second spring; 3. Knife cam; 4. Roller; 5. Linear motor; 6. Spindle; 7. Force sensor at the output end of the linear motor; 8. Second nut; 9. First nut; 10. Constant force output column; 11. Air bearing guide sleeve; 12. Load force sensor; 13. First mounting plate; 14. Second mounting plate; 15. Roller bracket; 16. Crossbar; 17. Air bearing guide rod. Detailed Implementation

[0047] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0048] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0050] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. These other embodiments are also covered within the scope of protection of this invention.

[0051] It should also be understood that the specific embodiments described above are only used to explain the present invention, and the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0052] Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0053] All prior art documents cited in this specification are incorporated herein by reference in their entirety and are therefore part of the disclosure of this invention.

[0054] Example 1

[0055] As shown in the figure, this invention provides a control method for an electromechanical composite constant force system based on model predictive control. This control method relies on an electromechanical composite constant force spring support system, which includes a housing. An electromagnetic active constant force compensation system and a mechanical constant force spring system are disposed on the housing. The mechanical constant force spring system is symmetrically distributed on both sides of the electromagnetic active constant force compensation system. The electromagnetic active constant force compensation system includes a spindle 6, on which a first spring 1 is sleeved. A linear motor 5 is disposed between the first spring 1 and the spindle 6 and fixed to the spindle 6. A linear motor output force sensor 7 is disposed at one end of the linear motor 5 and fixed to the spindle 6. A roller bracket 15 is disposed at one end of the first spring 1 and a first nut 9 is disposed at the other end.

[0056] The mechanical constant force spring system includes a first mechanical constant force spring system and a second mechanical constant force spring system. The first mechanical constant force spring system and the second mechanical constant force spring system have the same symmetrical distribution structure. A roller 4 is provided on one side of the roller bracket 15 on the first mechanical constant force spring system. A knife-type cam 3 is provided on one side of the roller 4. A crossbar 16 is provided on the knife-type cam 3 and passes through the housing. A second spring 2 is sleeved on the crossbar 16. One end of the second spring 2 is connected to a first mounting plate 13 provided on the housing, and the other end is provided with a second nut 8, which is sleeved on the crossbar 16.

[0057] The mandrel 6 passes through the second mounting plate 14 on the housing. One end of the mandrel 6 is fixed to the constant force output column 10. One end of the constant force output column 10 is connected to one end of the roller bracket 15. The constant force output column 10 is connected to the load through the load force sensor 12.

[0058] One end of the constant force output column 10 is provided with a load force sensor 12. An air bearing guide sleeve 11 is sleeved on the outside of the housing. The other end of the roller bracket 15 is provided with an air bearing guide rod 17 between the housing and the first nut 9. The air bearing guide rod 9 and the air bearing guide sleeve 11 are coaxially arranged to achieve low friction air bearing guidance in the vertical direction.

[0059] A control method for an electromechanical composite constant force system based on model predictive control includes the following steps:

[0060] S1. Training set data collection

[0061] Linear motor 5 is set to use a constantly changing random current I. k With the roller bracket 15 of the constant force spring system in motion and placed, the displacement h of the roller bracket at N different times is collected. k Speed ​​v k acceleration a k and the system output force measured by load force sensor 12 The displacement is the distance between the initial position and the current position of the roller bracket 15.

[0062] S2. Establishing a dynamic model of the system based on a BP neural network.

[0063] Will I k h k v k a k As the input to the BP neural network, denoted as Will As the output of a BP neural network. A BP neural network consists of an input layer, hidden layers, and an output layer, then:

[0064] The input of the input layer is

[0065] Input of hidden layer and output They are respectively:

[0066]

[0067]

[0068] Where, ω ij The connection weights θ between input layer neurons and hidden layer neurons represent the connection weights between them. j is the threshold of the hidden layer neuron node, f() is the hidden layer transfer function, M is the dimension of the input layer, and i takes the value of 4;

[0069] Input of the output layer and output They are respectively:

[0070]

[0071]

[0072] Where, ω jkα represents the connection weights between hidden layer neurons and output layer neurons. k The threshold of the output layer neuron node is represented by g(), the output layer transfer function is represented by P, and the dimension of the hidden layer is represented by j = 1, 2, ..., P.

[0073] S3. Input the data collected in step S1 into the system's dynamic model to complete model training.

[0074] (1) The I collected in step S1 k h k v k a k and As the training set for the BP neural network recognition model, I k h k v k a k For input, The output is the truth value;

[0075] (2) The I collected in step 1 k h k v k a k As the input to the input layer of the BP neural network recognition model, denoted as k = 1, 2, ..., N, where the input information is the current I. k Displacement h k Speed ​​v k acceleration a k That is, four-dimensional data, therefore i = 1, 2, 3, 4;

[0076] (3) Input the training set The input is fed into a backpropagation (BP) neural network to obtain the network output.

[0077] (4) The total error E of the BP neural network identification model is obtained using the following cost function:

[0078]

[0079] The parameters of the BP neural network identification model are adjusted using the gradient descent method to reduce its total error E. When the total error E reaches its minimum, the training of the BP neural network identification model is complete.

[0080] S4. Model predictive control controller design: The model trained in step S3 is used to identify the state of the electromechanical composite constant force system and calculate the optimal control force.

[0081] Controller design such as Figure 4 As shown, each control cycle is:

[0082] (1) Collect the displacement h, velocity v, and acceleration a of the roller bracket at the current moment;

[0083] (2) Generate M random current inputs I k ;

[0084] (3) Input the displacement h, velocity v, acceleration a and the M generated currents collected in step S1 into I k The data are divided into M groups, which are used as network inputs and fed into the BP neural network dynamics model trained in step S3 to obtain M corresponding network outputs.

[0085] (4) Obtain the M items Each is related to the desired system output force F aim Compare and find the one that minimizes the reward function. Further find the corresponding current input I k This current is used as the control current I of the linear motor. The reward function R is defined as follows:

[0086]

[0087] The operating principle of the electromechanical composite constant force spring support system controlled by the above method is as follows:

[0088] The housing remains stationary, and the load force sensor 12 is mounted on top of the constant force output column 10. The load is mounted on top of the load force sensor 12. As the load moves dynamically along the vertical direction, the two rollers 4 roll on the working surfaces of the two blade cams 3 and compress the blade cams 3, thereby changing the length and angle of the second spring 2, and further changing the vertical component of the force exerted by the blade cams 3 on the rollers 4. The sum of this force and the force of the first spring 1 always remains a constant value.

[0089] The load force sensor 12 measures the vertical force of the load in real time and feeds it back to the external control equipment. The linear motor 5 changes its output current according to the control equipment's instructions, thus changing its output force to compensate for the force error of the mechanical passive constant force component. The force sensor 7 at the output end of the linear motor 5 can provide real-time feedback on the output force of the linear motor 5, thereby enabling the linear motor to output a precise compensating force. The linear motor 5 and the air buoyancy mechanism work together to significantly reduce the constant force output error of the mechanical passive constant force component, improving the overall system's output accuracy by more than an order of magnitude.

[0090] By rotating the second nut 9 to adjust the position of the lower surface of the first spring 1, the compression of the first spring 1 during the initial operation of the constant force spring system can be changed, thereby adjusting the range of constant force output within a certain range.

[0091] The constant force output column 10, spindle 6, air-bearing guide rod 17, and air-bearing guide sleeve 11 installed at the center of the system constitute an air-bearing mechanism. This mechanism provides radial support, eliminating the overturning torque caused by the load's center of gravity deviation, and also eliminating the overturning torque of the central support plate caused by the deviation of the linear motor 5's output force application point from the center of gravity of the central support plate, thus ensuring the system's vertical movement. Furthermore, the air-bearing mechanism employs a non-contact air-bearing method, which significantly reduces friction caused by the guiding device, simplifies linear motor control, and improves the accuracy of zero-gravity simulation.

[0092] Therefore, the present invention adopts the above-mentioned model predictive control-based electromechanical composite constant force system control method, which improves the constant force output accuracy and response speed of the electromechanical composite constant force spring support, thereby realizing high-precision vertical gravity compensation and improving the realism of spacecraft ground simulation.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A control method for an electromechanical composite constant force system based on model predictive control, characterized in that, Includes the following steps: S1. Training set data collection; S2. Establish a dynamic model of the system based on a BP neural network; S3. Input the data collected in step S1 into the dynamic model of the system to complete the model training; S4. Model predictive controller design: The model trained in step S3 is used to identify the state of the electromechanical composite constant force system and calculate the optimal control force. In step S4, the specific steps for calculating the optimal control force are as follows: (1) Collect the displacement of the roller support at the current moment. ,speed acceleration ; (2) Generate random Current input ; (3) The displacement collected in step S1 ,speed acceleration and generated Current input Divided into The groups are used as network inputs and fed into the BP neural network dynamics model trained in step S3 to obtain... Each corresponding network output ; (4) The acquired indivual Each with the desired system output force Compare and find the one that minimizes the reward function. Further find the corresponding current input. This current is used as the control current of the linear motor. reward function The definition of is: 。 2. The control method for an electromechanical composite constant force system based on model predictive control according to claim 1, characterized in that, In step S1, the collected data includes setting the linear motor to a random current that varies over time. When the roller support of the constant force spring system moves, samples are collected respectively. Displacement of the roller support at different times ,speed acceleration and the system output force measured by the load force sensor The displacement is the distance between the initial position and the current position of the roller support.

3. The control method for an electromechanical composite constant force system based on model predictive control according to claim 1, characterized in that, In step S2, a dynamic model of the electromechanical composite constant force system is established using a BP neural network. , , , As the input to the BP neural network, denoted as ,Will As the output of the BP neural network A backpropagation (BP) neural network consists of an input layer, hidden layers, and an output layer. Therefore: The input of the input layer is ; Input of hidden layer and output They are respectively: ， ， in, This represents the connection weights between input layer neurons and hidden layer neurons. The threshold value for hidden layer neuron nodes. For hidden layer transfer functions, Let i be the dimension of the input layer, and take the value 4. Input of the output layer and output They are respectively: ， ， in, This represents the connection weights between hidden layer neurons and output layer neurons. The threshold value represents the output layer neuron node. This represents the output layer transfer function. Indicates the dimension of the hidden layer. .

4. The control method for an electromechanical composite constant force system based on model predictive control according to claim 1, characterized in that, Step S3 includes the following steps: (1) The data collected in step S1 , , , and As the training set for the BP neural network recognition model, among which, , , , For input, The output is the truth value; (2) The data collected in step 1 , , , As the input to the input layer of the BP neural network recognition model, denoted as , Since the input information is current Displacement ,speed acceleration That is, four-dimensional data, therefore ; (3) Input the training set The input is fed into a backpropagation (BP) neural network to obtain the network output. ; (4) Use the following cost function to obtain the total error of the BP neural network identification model. : ， The parameters of the BP neural network identification model are adjusted using the gradient descent method to reduce its total error. Decrease, when the total error When the minimum value is reached, the training of the BP neural network recognition model is complete.

Images