Rotary direct-drive servo valve nonlinear compensation method and system based on neural network

Abstract

The invention belongs to the technical field of prestage driving error compensation of a rotary direct-drive servo valve, and provides a rotary direct-drive servo valve nonlinear compensation method based on a neural network in order to solve the problem that a rotary servo valve has various nonlinear errors in practical application. Constructing a nonlinear compensation neural network, determining a preset target displacement of a valve core of the rotary direct-drive servo valve, inputting the preset target displacement of the valve core into the nonlinear compensation neural network to obtain a corresponding compensation current, converting the compensation current into an error compensation torque, transmitting the error compensation torque to a controller of the rotary direct-drive servo valve, and outputting the error compensation torque to the controller of the rotary direct-drive servo valve. And the displacement of the valve element is adjusted according to the error compensation torque, so that multiple nonlinearity, caused by reversing hysteresis of the sliding valve, a negative opening dead zone of the sliding valve, motor rotating friction and the like, of the system is compensated, and the reliability and stability of the system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of driving error compensation for the pre-stage of a rotary direct drive servo valve, and particularly relates to a non-linear compensation method and system for a rotary direct drive servo valve based on a neural network. Background Art

[0002] A rotary servo valve driven by a limited-angle motor is a valve driven by a motor, usually used to precisely control the flow rate and pressure of a fluid. It converts the position into the opening degree of the valve through the rotation of the motor, allowing precise flow regulation. Such servo valves are common in industrial automation and hydraulic systems, and have the characteristics of fast response and high control accuracy.

[0003] In a rotary servo valve driven by a limited-angle motor, due to factors such as spool commutation hysteresis, spool negative overlap dead zone, and motor rotational friction, there are various non-linear errors in the system, resulting in a deviation between the actual output torque and the target torque. Utilizing the self-learning ability of the neural network and the arbitrary approximation characteristics of non-linear functions, non-linear compensation can be effectively performed, thereby improving the accuracy and reliability of control. Through training, the neural network can identify and correct these non-linear characteristics to achieve more precise dynamic control. Summary of the Invention

[0004] The present invention provides a non-linear compensation method and system for a rotary direct drive servo valve based on a neural network to solve at least one of the above technical problems existing in the prior art.

[0005] According to a first aspect, a non-linear compensation method for a rotary direct drive servo valve based on a neural network includes the following steps:

[0006] Obtain the operating parameters of the rotary direct drive servo valve and establish a dynamic model of the rotary direct drive servo valve. The operating parameters at least include: actual spool displacement, armature current, motor rotation angle, and motor speed;

[0007] Based on the operating parameters, establish a non-linear compensation neural network. The non-linear compensation neural network includes an input layer, a hidden layer, and an output layer; the input layer includes five nodes, corresponding to the actual spool displacement, armature current, motor rotation angle, motor speed, and preset target spool displacement respectively; the hidden layer includes six nodes, corresponding to non-linear sign functions respectively; the output layer obtains the compensation current of the rotary direct drive servo valve according to the processing results of the input layer and the hidden layer;

[0008] Determine the preset target spool displacement of the rotary direct drive servo valve, input the preset target spool displacement into the non-linear compensation neural network to obtain the corresponding compensation current, and simultaneously determine the error compensation torque according to the compensation current and the dynamic model of the rotary direct drive servo valve;

[0009] Transmit the error compensation torque to the controller of the rotary direct drive servo valve, and adjust the spool displacement according to the error compensation torque.

[0010] Preferably, the dynamic model is:

[0011]

[0012] In the formula, x represents the spool displacement parameter; v represents the spool rotation speed parameter; T and T f respectively represent the torque proportional to the control current and the torque disturbance; f1 is a linear function with motor parameters as variables, f2 is a non - linear function affected by non - linear errors; t represents time.

[0013] Preferably, after establishing the non - linear compensation neural network based on the operating parameters, it further includes:

[0014] Obtain the historical operation data set of the rotary direct drive servo valve. The historical operation data set includes at least one data sample, and the data sample is the historical operation parameters and the historical compensation current obtained after inputting the historical operation parameters;

[0015] Train the non - linear compensation neural network based on the historical operation data set to obtain a non - linear compensation neural network with converged parameters.

[0016] Preferably, training the non - linear compensation neural network based on the historical operation data set includes:

[0017] Input the historical spool target displacement and historical operation parameters in the historical operation data set into the non - linear compensation neural network to obtain the corresponding historical compensation current;

[0018] Determine the historical error compensation torque based on the historical compensation current and the dynamic model;

[0019] Control the movement of the spool of the rotary direct drive servo valve based on the historical error compensation torque, and determine the historical spool error displacement according to the historical spool target displacement;

[0020] Feed back the historical spool displacement error to the non - linear compensation neural network to adjust the weight value of the non - linear compensation neural network;

[0021] Use the historical spool error displacement as the new input of the non - linear compensation neural network model after weight adjustment, and continue to adjust the weight value until the error between the historical spool target displacement and the historical spool displacement error is lower than the preset threshold, and save the model parameters of the non - linear compensation neural network at this time to obtain a non - linear compensation neural network with converged parameters.

[0022] Preferably, the weight values of the non-linear compensation neural network are adjusted, and the specific formula is:

[0023]

[0024] In the formula, x(k) is the spool displacement, is the output of the model training, Y(j) = [y(1)...y(n)] T is the neuron input vector, and

[0025] W(i,j) = [w(i,1)w(i,2)...w(i,n)] T is the weight coefficient during the (k - 1)th training, n is the number of input variables of the input layer, and i is the number of neurons in a layer.

[0026] Preferably, it further includes:

[0027] Calculating the gradient of each neuron in the output layer of the non-linear compensation neural network model by using the chain rule, and propagating it backward layer by layer until a preset number of iterations is reached.

[0028] According to a second aspect, a non-linear compensation system for a rotary direct drive servo valve based on a neural network, which can execute the method described in the first aspect and any preferred embodiment, includes: a controller, a magnetostrictive displacement sensor, a current sensor, and an angular displacement sensor;

[0029] The controller is configured to receive the error compensation torque and control the movement of the spool according to the error compensation torque;

[0030] The magnetostrictive displacement sensor is configured to provide a spool displacement signal, and the spool displacement signal includes the actual spool displacement and the error spool displacement;

[0031] The current sensor is configured to measure the armature current;

[0032] The angular displacement sensor is configured to measure the motor rotation angle and the motor speed.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] The present invention can achieve the compensation of many non-linearities existing in the system caused by spool commutation hysteresis, spool negative overlap dead zone, motor rotation friction, etc. without changing the front-stage drive hardware of the existing rotary direct drive servo valve, improving the reliability and stability of the system. Description of the Drawings

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0036] Figure 1 It is a schematic flowchart of a nonlinear compensation method for a rotary direct-drive servo valve based on a neural network provided by an embodiment of the present invention;

[0037] Figure 2 It is a schematic flowchart of the working process of a nonlinear compensation neural network provided by an embodiment of the present invention. Specific embodiments

[0038] Combined with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other implementation manners obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the present invention.

[0039] It should be noted that the structures, ratios, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the limited conditions under which the present invention can be implemented. Therefore, they do not have any technical substance. Any modification of the structure, change of the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should fall within the scope covered by the technical content disclosed in the present invention. It should be noted that in this specification, relational terms such as first and second are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0040] The rotary direct-drive servo valve is a high-performance hydraulic control element. It is a hydraulic control valve that receives an analog electrical signal and correspondingly outputs a modulated flow rate and pressure, and has the advantages of fast response, high control accuracy, etc., and is widely used in fields such as aerospace and robotics. However, traditional rotary direct-drive servo valves adopt complex structures such as spool valves, nozzle flapper valves, and jet pipe valves. In the actual application process, due to factors such as spool valve commutation hysteresis, spool valve negative overlap dead zone, and motor rotation friction, there are various nonlinear errors in the system, resulting in a deviation between the actual output torque and the target torque. To solve the above problems, the present invention provides a nonlinear compensation method and system for a rotary direct-drive servo valve based on a neural network, which can timely adjust the nonlinear errors generated during the application of the rotary direct-drive servo valve and improve the overall stability.

[0041] An embodiment of the present invention provides a non - linear compensation system for a rotary direct - drive servo valve based on a neural network, including: a controller, a magnetostrictive displacement sensor, a current sensor, and an angular displacement sensor;

[0042] The controller is configured to receive an error compensation torque and control the movement of the spool according to the error compensation torque; the magnetostrictive displacement sensor is configured to provide a spool displacement signal, and the spool displacement signal includes the actual displacement of the spool and the error spool displacement; the current sensor is configured to measure the armature current; the angular displacement sensor is configured to measure the rotation angle and the rotation speed of the motor.

[0043] As Figure 1 shown, applied to the non - linear compensation system for a rotary direct - drive servo valve based on a neural network provided in the above - mentioned embodiment, a flow schematic diagram of a non - linear compensation method for a rotary direct - drive servo valve based on a neural network is provided, including the following steps:

[0044] S1: Obtain the operating parameters of the rotary direct - drive servo valve and establish a dynamic model of the rotary direct - drive servo valve. The operating parameters at least include: the actual displacement of the spool, the armature current, the rotation angle of the motor, and the rotation speed of the motor.

[0045] In this embodiment, the dynamic model is:

[0046]

[0047] In the formula, x represents the spool displacement parameter; v represents the spool rotation speed parameter; T and T f respectively represent the torque proportional to the control current and the torque disturbance; f1 is a linear function with motor parameters as variables, and f2 is a non - linear function affected by non - linear errors; t represents time.

[0048] S2: Establish a non - linear compensation neural network based on the operating parameters. The non - linear compensation neural network includes an input layer, a hidden layer, and an output layer; the input layer includes five nodes, corresponding to the actual displacement of the spool, the armature current, the rotation angle of the motor, the rotation speed of the motor, and the preset target displacement of the spool respectively; the hidden layer includes six nodes, corresponding to non - linear sign functions respectively; the output layer obtains the compensation current of the rotary direct - drive servo valve according to the processing results of the input layer and the hidden layer.

[0049] In this embodiment, the non - linear compensation neural network is set with an input layer, including five nodes, corresponding to the preset target displacement of the spool, the actual displacement of the spool, the armature current (fed back by the current sensor), the motor rotation angle (fed back by the angular displacement sensor), and the motor rotation speed respectively.

[0050] Set the hidden layer, including six nodes, corresponding to non - linear sign functions.

[0051] Set an output layer to output a compensation current required for an output error compensation torque.

[0052] Optionally, after establishing a non - linear compensation neural network based on the operating parameters, it further includes: obtaining a historical operating data set of a rotary direct - drive servo valve, where the historical operating data set includes at least one data sample, and the data sample is historical operating parameters and a historical compensation current obtained after inputting the historical operating parameters; training the non - linear compensation neural network based on the historical operating data set to obtain a non - linear compensation neural network with converged parameters.

[0053] In this embodiment, the training process of the non - linear compensation neural network can be completed through the MATLAB neural network toolbox. This method can also be trained with other tools, and the present application is not limited thereto.

[0054] Optionally, training the non - linear compensation neural network based on the historical operating data set includes: inputting the historical spool target displacement and historical operating parameters in the historical operating data set into the non - linear compensation neural network to obtain a corresponding historical compensation current; determining a historical error compensation torque based on the historical compensation current and the dynamic model; controlling the movement of the spool of the rotary direct - drive servo valve based on the historical error compensation torque, and determining a historical spool displacement error according to the historical spool target displacement; feeding back the historical spool displacement error to the non - linear compensation neural network to adjust the weight value of the non - linear compensation neural network; using the historical spool displacement error as the input of the non - linear compensation neural network model after weight adjustment, and continuing to adjust the weight value until the error between the historical spool target displacement and the historical spool displacement error is lower than a preset threshold, and saving the model parameters of the non - linear compensation neural network at this time to obtain a non - linear compensation neural network with converged parameters.

[0055] In this embodiment, input the historical spool target displacement x d (1) and the corresponding historical actual displacement x(0) into the non - linear compensation neural network, and based on the historical operating data set, give the control torque T(1) corresponding to an incorrect compensation current and the historical spool error displacement x(1) controlled by the incorrect compensation current. Use the error between x d (1) - x(1) through the back - propagation process to adjust the weight value of the neural network.

[0056] In this embodiment, to adjust the weight value of the non - linear compensation neural network, the back - propagation method is adopted, and the weight value W of the neural network is adjusted by minimizing the following error value, and the specific formula is:

[0057]

[0058] where x(k) is the spool displacement, is the output of model training, Y(j) = [y(1)...y(n)] T is the neuron input vector, and W(i,j) = [w(i,1)w(i,2)...w(i,n)] T is the weight coefficient at the (k - 1)th training, n is the number of input variables of the input layer, and i is the number of neurons in a layer.

[0059] Adjust according to the calculation result of this formula. When the sum of squared errors is less than 0.001, output the optimal weight adjustment result.

[0060] Take the historical spool error displacement x(1) as the new input of the non - linear compensation neural network model after weight adjustment, generate new parameters x(2), I(2), θ(2), ω(2), and continue to adjust the weight values according to the new parameters until the error between the historical spool target displacement and the historical spool displacement error is lower than the preset threshold, and save the model parameters of the non - linear compensation neural network at this time to obtain a non - linear compensation neural network with convergent parameters.

[0061] Optionally, it further includes: calculating the gradient of each neuron in the output layer of the non - linear compensation neural network model using the chain rule and propagating it backward layer by layer until the preset number of iterations is reached.

[0062] In this embodiment, the gradient of the loss with respect to each parameter (weight and bias) is calculated by the chain rule. The non - linear compensation method includes: calculating the partial derivative of the loss function with respect to the output to obtain the gradient of each neuron in the output layer, and propagating the gradient backward layer by layer. In each layer, calculate the gradient of the current layer and calculate the gradient of the previous layer according to the output of the current layer and the gradient of the previous layer. Using the calculated gradient, adjust the weight and bias through an optimization algorithm. The backpropagation process is repeated for each training sample until the network converges or reaches the set number of iterations.

[0063] S3: Determine the preset target displacement of the spool of the rotary direct - drive servo valve, input the preset target displacement of the spool into the non - linear compensation neural network to obtain the corresponding compensation current, and simultaneously determine the error compensation torque according to the compensation current and the dynamic model of the rotary direct - drive servo valve.

[0064] S4: Transmit the error compensation torque to the controller of the rotary direct - drive servo valve and adjust the spool displacement according to the error compensation torque.

[0065] In this embodiment, as Figure 2As shown in the figure, the preset target displacement of the spool is input into the trained non-linear compensation neural network, and the corresponding compensation current is obtained at the output layer. Then, the controller calculates the error compensation torque based on the compensation current and the dynamic model of the rotary direct drive servo valve, and the controller controls the movement of the spool of the rotary direct drive servo valve according to the error compensation torque.

[0066] In summary, compared with the prior art, the present invention realizes the compensation of many non-linearities existing in the system caused by spool commutation hysteresis, spool negative overlap dead zone, motor rotational friction, etc. without changing the front-stage drive hardware of the existing rotary direct drive servo valve, and improves the reliability and stability of the system.

[0067] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A nonlinear compensation method for a rotary direct-drive servo valve based on a neural network, characterized in that: The steps include: Acquiring the operating parameters of the rotary direct-drive servo valve and establishing a dynamic model of the rotary direct-drive servo valve, wherein the operating parameters at least include: actual displacement of the valve core, armature current, motor rotation angle, and motor speed; A nonlinear compensation neural network is established based on the operating parameters, and the nonlinear compensation neural network includes an input layer, a hidden layer and an output layer; the input layer includes five nodes, which correspond to the actual displacement of the valve core, the armature current, the motor rotation angle, the motor speed and the preset target displacement of the valve core respectively; the hidden layer includes six nodes, which correspond to the nonlinear sign function respectively; the output layer obtains the compensation current of the rotary direct-drive servo valve according to the processing results of the input layer and the hidden layer; Determine a preset target displacement of a valve core of the rotary direct-drive servo valve, input the preset target displacement of the valve core into the nonlinear compensation neural network, obtain a corresponding compensation current, and determine an error compensation torque according to the compensation current and a dynamic model of the rotary direct-drive servo valve; The error compensation torque is transmitted to a controller of the rotary direct drive servo valve, and the valve core displacement is adjusted according to the error compensation torque.

2. The nonlinear compensation method for rotary direct-drive servo valve based on neural network according to claim 1 is characterized in that: The kinetic model is: In the formula, x represents the valve core displacement parameter; v represents the valve core rotation speed parameter; T and T f They respectively represent the torque and torque disturbance which are proportional to the control current; f1 is a linear function with motor parameters as variables, f2 is a nonlinear function affected by nonlinear errors; t represents time.

3. The nonlinear compensation method for a rotary direct-drive servo valve based on a neural network according to claim 1, characterized in that: After the nonlinear compensation neural network is established based on the operating parameters, the method further includes: Acquire a historical operation data set of the rotary direct-drive servo valve, wherein the historical operation data set includes at least one data sample, and the data sample is a historical operation parameter and a historical compensation current obtained after inputting the historical operation parameter; The nonlinear compensatory neural network is trained based on the historical operation data set to obtain a nonlinear compensatory neural network with converged parameters.

4. The nonlinear compensation method for a rotary direct-drive servo valve based on a neural network according to claim 3 is characterized in that: The training of the nonlinear compensation neural network based on the historical operation data set includes: Inputting the historical valve core target displacement and the historical operation parameters in the historical operation data set into the nonlinear compensation neural network to obtain the corresponding historical compensation current; Determining a historical error compensation torque based on the historical compensation current and the dynamic model; Controlling the movement of the valve core of the rotary direct-drive servo valve based on the historical error compensation torque, and determining the historical valve core error displacement according to the historical valve core target displacement; Feeding back the historical valve core displacement error to the nonlinear compensation neural network, and adjusting the weight value of the nonlinear compensation neural network; The historical valve core error displacement is used as a new input of the nonlinear compensation neural network model after weight adjustment, and the weight value is continuously adjusted until the error between the historical valve core target displacement and the historical valve core displacement error is lower than a preset threshold value. The model parameters of the nonlinear compensation neural network model at this time are saved to obtain a nonlinear compensation neural network with converged parameters.

5. The nonlinear compensation method of a rotary direct-drive servo valve based on a neural network according to claim 4, characterized in that: The weight value of the nonlinear compensation neural network is adjusted. The specific formula is: Where x(k) is the valve core displacement, is the output of model training, Y(j)=[y(1)...y(n)] T is the neuron input vector, and W(i,j)=[w(i,1)w(i,2)...w(i,n)] T is the weight coefficient during k-1 training, n is the number of input variables in the input layer, and i is the number of neurons in a layer.

6. The nonlinear compensation method of a rotary direct-drive servo valve based on a neural network according to claim 4, characterized in that: Also includes: The chain rule is used to calculate the gradient of each neuron in the output layer of the nonlinear compensation neural network model and propagate backward layer by layer until a preset number of iterations is reached.

7. A nonlinear compensation system for a rotary direct-drive servo valve based on a neural network, capable of executing the nonlinear compensation method for a rotary direct-drive servo valve based on a neural network as described in any one of claims 1 to 6, characterized in that: include: Controller, magnetostrictive displacement sensor, current sensor and angular displacement sensor; The controller is used to receive the error compensation torque and control the movement of the valve core according to the error compensation torque; The magnetostrictive displacement sensor is used to provide a valve core displacement signal, wherein the valve core displacement signal includes an actual valve core displacement and an error valve core displacement; The current sensor is used to measure the armature current; The angular displacement sensor is used to measure the rotation angle and speed of the motor.

Images