An adaptive variable bias torque compensation method for dual-motor servo system

Abstract

The present invention discloses an adaptive variable bias torque compensation method for a dual-motor servo system. First, a dynamic model of the dual-motor servo follower system is established; then, a control system is established based on the dynamic model and a control input is obtained; secondly, whether the system is in a commutation state is determined based on the control input; finally, if the system is not in a commutation state, no bias torque is applied; if the system is in a commutation state, the momentum conservation theorem is used to predict the collision speed between the drive motor and the load gear based on the current angular velocity of the motor and the load, and then the magnitude of the bias torque to be applied is given by a constructed neural network model. The present invention makes the uncontrollable load in the system commutation process controllable, and at the same time designs an adaptive variable bias torque compensation method based on a neural network to reduce the problem of jitter caused by constant bias torque under different working conditions, which leads to a decrease in tracking accuracy, thereby achieving precise control of the dual-motor servo system.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor servo control, and in particular relates to an adaptive variable bias torque compensation method for a dual-motor servo system. Background Art

[0002] With the development of modern science and technology, the demand for large-scale, high-power servo systems in military, industrial and agricultural fields is increasing. Due to technical and price limitations, the driving power of a single motor is difficult to apply to these occasions. To address this problem, the driving capability of the servo system can be improved by using multiple small-power motors to jointly drive large-inertia loads. Compared with a single-motor system, this drive method improves the driving capability of the servo system and reduces the design cost and difficulty. However, the multi-click servo follow-up system introduces multiple gear transmission links, and tooth clearance is inevitable during the system deceleration process. The existence of tooth clearance turns the system into an incompletely controllable system. When the gear is between the tooth clearances, the drive motor and the load gear are not in contact. Therefore, during the startup and reversing process of the system, due to the influence of tooth clearance nonlinearity, there are return errors and jitter phenomena.

[0003] For servo systems with multiple motors, the backlash elimination process is often achieved by adding a bias torque. Currently, there are two common ways to add bias torque. One is to add a constant bias torque, applying two bias torques of equal magnitude and opposite directions to the two motors involved in driving the servo system. However, when the motors approach the commutation state, the motor with the negative bias torque applied will turn in advance to engage with the load gear on the other side, ensuring continuous control. The other is to apply a variable bias torque, applying two bias torques of equal magnitude and opposite directions only when the motors approach the commutation state, reducing the energy loss caused by applying a constant bias torque. Both of the above-mentioned applied bias torques are fixed values, but the operating conditions of the servo system are not constant. Under certain conditions, the bias torque may be too small to achieve the backlash elimination effect, or it may be too large to adversely affect the system control accuracy. The main work of this paper is to improve the applied bias torque, changing the constant bias torque into an adaptive variable bias torque. The bias torque changes adaptively according to the changes in the system state, so that the bias torque still has a good anti-backlash effect under various operating conditions of the system. Summary of the Invention

[0004] The object of the present invention is to provide an adaptive variable bias torque compensation method for a dual-motor servo system to solve the problem that the anti-backlash effect of the constant bias torque becomes poor or excessive when the system operating conditions change.

[0005] To achieve the purpose of the present invention, the present invention provides an adaptive variable bias torque compensation method for a dual-motor servo system, comprising the following steps:

[0006] Step 1: Analyze the dual-motor servo follower system, use the dead zone model to describe the backlash link, and establish the dynamic model of the dual-motor servo follower system;

[0007] Step 2: Analyze the motion state of the motor's backlash link and establish a motion model for backlash generation;

[0008] Step 3: Design a BP neural network model with the motor state as input and the bias current as output;

[0009] Step 4: Design a control system based on the established dual-motor servo follower system dynamics model to enable the actuator to accurately track the position signal;

[0010] Step 5: Determine whether the dual-motor servo follow-up system has entered a backlash state based on the control input of the established dual-motor servo control system. If it has entered a backlash state, use the BP neural network to calculate the bias current and apply a pair of bias currents of equal magnitude and opposite directions to the dual motors.

[0011] Furthermore, step 1 establishes a dynamic model of the dual-motor servo follower system. The model is used to design the dual-motor servo control system, including a single-motor mathematical model and a mechanical transmission model of the dual-motor system. The specific steps are as follows:

[0012] Step 1-1: First, establish a mathematical model of a single motor in a synchronous rotating coordinate system;

[0013] Among them, the stator voltage model is

[0014]

[0015] Where u d 、u q are the dq axis components of the stator voltage respectively; i d 、i q are the dq axis components of the electronic current; R is the electronic resistance; w e is the electrical angular velocity; L d 、L q are the dq axis inductance components respectively; f is the permanent magnet flux linkage;

[0016] Among them, the electromagnetic torque equation is

[0017]

[0018] Where p n is the pole pair number, T e is the electromagnetic torque;

[0019] Step 1-2, secondly, establish the transmission model of the dual-motor servo follower system;

[0020] Among them, the mechanical motion equation on the motor side is

[0021]

[0022] In the formula is the equivalent moment of inertia of motor j, is the viscous friction coefficient equivalent to motor j, w j is the speed of motor j, T cj is the torque of pinion j, r cj is the reduction ratio of motor j; T ej is the electromagnetic torque of motor j.

[0023] Among them, the dead zone model of tooth clearance is

[0024]

[0025] Where k is the rigidity coefficient, α j is the tooth gap size, z j is the relative displacement between the motor side and the load side;

[0026] Among them, the mechanical motion equation on the load side is

[0027]

[0028] Where J m is the equivalent moment of inertia of the large gear, b m Equivalent viscous friction coefficient of the gear, T L is the load moment, r m is the reduction ratio between the large gear and the small gear, w m is the speed of the large gear of the system.

[0029] Furthermore, in step 2, the motion state of the motor backlash link is analyzed, a motion model of backlash is established, and the expected speed after the driving motor gear collides with the load gear is calculated. The specific steps are as follows:

[0030] Step 2-1: When the motor side gear collides with the load side gear, it is an elastic collision. The collision process satisfies the conservation of angular momentum and angular kinetic energy. The corresponding formula is:

[0031]

[0032] Where J1 and J2 are the moments of inertia of the back-drive motor side and the load side, respectively; w1 and w2 are the angular velocities of the back-drive motor side and the load side before the collision, respectively; w′1 and w′2 are the angular velocities of the back-drive motor side and the load side after the collision, respectively;

[0033] Step 2-2: Combine the formulas in step 2-1 to obtain the angular velocity of the back-drive motor side and the load side after the collision:

[0034]

[0035] Step 2-3: According to the formula in step 2-2, the speed of the reverse drive motor after the collision can be calculated from the speeds of the motor and load before the collision. The load before the collision is still in contact with the gear of the forward drive motor, so the angular velocity of the load before the collision is equal to the angular velocity of the forward drive motor divided by the reduction ratio. To reduce the vibration of the load after the collision, the direction of the velocity remains unchanged after the collision, but the magnitude is reduced to 1 / 2 of the velocity before the collision.

[0036] Furthermore, in step 3, the trained neural network model is used to calculate the bias torque using the expected speed of the backdrive motor side gear before collision and the angular velocity when the bias torque is applied as input. The specific steps are as follows:

[0037] Step 3-1. First, design a three-layer fully connected neural network using the BP algorithm. The BP neural network structure includes 4 input layer nodes, 5 hidden layer nodes, and 1 output layer node. Each layer node is only connected to the adjacent layer nodes. There is no connection between nodes within each layer, and the nodes in each layer are fully connected.

[0038] The input and output formula of the hidden layer is

[0039]

[0040] Where, is the output node of the first hidden layer, is the hidden layer weight coefficient, It is the input of the activation function of the first layer of the hidden layer, and the activation function uses the hyperbolic tangent function:

[0041]

[0042] The input and output formula of the output layer is

[0043]

[0044] Where, O out is the output node, is the output layer weight coefficient, Net out It is the activation function input of the output layer.

[0045] The bias current applied to the output layer node The output layer activation function uses a non-negative sigmoid function, that is,

[0046]

[0047] Step 3-2, then design the weight update algorithm of the hidden layer; by defining the performance index function, use the gradient descent method to correct the weight coefficient in the negative gradient direction;

[0048] Define the system performance index function as

[0049]

[0050] According to the gradient descent method, the weight correction is adjusted by searching the negative gradient direction of the weight coefficient according to E(k). The basic formula is:

[0051]

[0052] Where η is the weight correction coefficient, 0<η<1, is the weight correction value on the mth layer, O o,m is the output node on the mth layer of the hidden layer, Net o,m is the activation function input of the hidden layer m, E o,m is the error on the mth layer.

[0053] The network output layer and hidden layer weight learning algorithm formula is obtained as follows:

[0054]

[0055] Step 3-3, then create a training set for neural network training;

[0056] The training set is

[0057]

[0058] Where w i is the current angular velocity of motor j, is the current angular acceleration of motor j, Predetermine the angular velocity for motor j; is the applied bias current;

[0059] The back-drive motor does not receive the load torque from the load gear during the period of crossing the tooth gap and can be regarded as no-load operation. Therefore, the training set can be collected under the no-load operation of the motor. i and angular acceleration Add a constant current of small amount i i , after the displacement of Δθ, the velocity is measured Get the dataset Where Δθ is the system backlash;

[0060] Step 3-4: Finally, train the neural network; use the weight update algorithm in step 3-2 to train the neural network using the prepared training set.

[0061] Furthermore, in step 4, a dual-motor servo control system is designed to enable the actuator to accurately track the position signal, which specifically includes the following steps:

[0062] Step 4-1: First, design the dual-motor servo follow-up system position control system based on three-loop PID control. The control system is divided into current loop, speed loop, and position loop from the inside out to achieve accurate control of motor current, speed, and position;

[0063] Step 4-2, first design the current loop control. In the current loop control, take i d =0 control strategy, its simplified stator voltage model is:

[0064]

[0065] Design the current loop control input to be

[0066]

[0067] Where, P I is the current loop proportional coefficient, D I Current loop differential coefficient, I I is the current loop integral coefficient;

[0068] Step 4-3, then design the speed loop control. The simplified mechanical motion equation and torque equation are:

[0069]

[0070] The speed loop input is designed to be

[0071]

[0072] Step 4-4, then design the position loop control

[0073]

[0074] Step 4-5. Finally, the designed three-loop PID control system is used for the position control of the dual-motor servo follower system. After the dual-motor servo follower system receives the position command, the speed loop, position loop, and current loop generate a controlled voltage command to enable the system to perform an action. The feedback link simultaneously feeds back the current, speed, and position state quantities to the three control links, and corrects the output of the control links so that the system can accurately track the position command.

[0075] Furthermore, in step 5, the backlash state of the dual-motor servo follower system is judged, calculated, and a bias current is applied, which specifically includes the following steps:

[0076] Step 5-1: First, determine whether the dual-motor servo servo system has entered the backlash state;

[0077] Since the backlash state mainly occurs when the motor accelerates or decelerates, the relative size of the drive motor speed and the load speed can be used to determine whether the system has entered the backlash state. According to the formula in step 4-3, the motor acceleration is related to the electromagnetic torque, and the electromagnetic torque is proportional to the q-axis current. The backlash state can be determined by detecting the current loop control input. If the current loop control input is near zero and about to cross zero, it indicates that the drive motor is about to accelerate or decelerate, and the dual-motor servo servo system is about to enter the backlash state. If the current loop control input is not near zero, it indicates that the dual-motor servo servo system is not in the backlash state, and there is no need to apply a bias current to the control.

[0078] Step 5-2 then calculates the bias current based on the BP neural network;

[0079] According to the neural network data set in step 3-3, the current angular velocity of the motor w is used as input in the data set. i and angular acceleration Can be measured directly; motor preset angular velocity is the motor speed before the gear collision. From the formula in step 2-2, we know that the angular velocity before the gear collision can be calculated from the angular velocity of the system before and after the collision. Assuming that the system speed is unchanged before the collision, the system speed after the collision is set to 90% of the pre-collision speed based on the premise of minimum system jitter;

[0080] The BP neural network accepts the current state input of the system to calculate the bias current, which is directly applied to the current loop input of the two motors, where the bias current of the back-drive motor is negative.

[0081] Compared with the existing technology, the significant improvements of the present invention are: 1) the implementation size of the bias current can be calculated in real time according to the motor operating conditions, realizing adaptive adjustment; 2) the size of the bias current changes according to the changes in the motor operating conditions, improving the tracking performance of the servo system; 3) the bias current realizes adaptive change, further reducing the energy loss of the system caused by the bias current.

[0082] In order to more clearly illustrate the functional characteristics and structural parameters of the present invention, further description is given below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0084] Figure 1Flowchart of the adaptive bias current compensation method of the dual-motor servo system of the present invention;

[0085] Figure 2 This is a block diagram of the single motor control structure of the dual motor servo system of the present invention;

[0086] Figure 3 It is a structural block diagram of the dual-motor servo system control structure of the present invention;

[0087] Figure 4 is a graph showing the relationship between the motor current and the bias in the present invention;

[0088] Figure 5 This is a neural network structure diagram of the present invention. DETAILED DESCRIPTION

[0089] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0090] like Figure 1 As shown, an adaptive variable bias torque compensation method for a dual-motor servo system includes the following steps:

[0091] Step 1: Establish a dynamic model of the dual-motor servo follower system. The model is used to design the dual-motor servo control system. The model mainly includes a single-motor mathematical model and a mechanical transmission model of the dual-motor system.

[0092] Step 1-1: First, establish a single motor dynamics model. Figure 2 In the control block diagram shown in FIG, the d-axis current is set to 0 in the rotating coordinate system, and an approximate linear system can be obtained. The simplified dynamic model of the single motor in the synchronous rotating coordinate system is:

[0093]

[0094] where u q is the q-axis component of the stator voltage; i q is the q-axis component of the stator current; R is the stator resistance; w e is the electrical angular velocity; ψ f is the permanent magnet flux; p n is the number of pole pairs; J is the moment of inertia of the motor; b is the viscous friction coefficient of the motor; T e is the electromagnetic torque; T c is the load torque.

[0095] Step 1-2: Next, establish the transmission model of the servo system. The torque transmission of gear engagement adopts the dead zone model, so the transmission model of the dual-motor servo system is

[0096]

[0097] In the formula is the equivalent moment of inertia of motor j, is the viscous friction coefficient equivalent to motor j, T cj is the torque of the large gear on the small gear, r cj is the reduction ratio between the motor and the pinion, k is the rigidity coefficient, α j is the tooth gap size, z j is the relative displacement between the motor side and the load side, J m is the equivalent moment of inertia of the large gear, b m Equivalent viscous friction coefficient of the gear, T L is the load moment, r m It is the reduction ratio between the large gear and the small gear.

[0098] Step 2: Analyze the motion state of the motor's backlash link, establish a motion model for backlash, and calculate the expected speed after the drive motor gear collides with the load gear. The specific steps are as follows:

[0099] Step 2-1: When the motor side gear collides with the load side gear, it is an elastic collision. The collision process satisfies the conservation of angular momentum and angular kinetic energy. The corresponding formula is:

[0100]

[0101] Where J1 and J2 are the moments of inertia of the back-drive motor side and the load side, respectively; w1 and w2 are the angular velocities of the back-drive motor side and the load side before the collision, respectively; w1′ and w′2 are the angular velocities of the back-drive motor side and the load side after the collision, respectively.

[0102] Step 2-2, then combine the two equations of formula 1.24 to obtain the angular velocity of the back-drive motor side and the load side after the collision:

[0103]

[0104] Finally, according to formula 1.25, the reverse-drive motor speed before the collision can be calculated from the load-side speed before and after the collision. The load-side gear is still in contact with the forward-drive motor before the collision, so the load-side angular velocity before the collision equals the forward-drive motor angular velocity divided by the reduction ratio. To minimize post-collision load vibration, the velocity after the collision remains unchanged in direction and is reduced to 90% of the pre-collision velocity.

[0105] Step 3: Use the trained neural network model to calculate the bias torque using the expected speed of the backdrive motor side gear before collision and the angular velocity when the bias torque is applied as input. The specific steps are as follows:

[0106] Step 3-1, first design a three-layer fully connected neural network using BP algorithm. The BP neural network structure is shown in the figure below: Figure 5 As shown, there are 4 input layer nodes, 5 hidden layer nodes, and 1 output layer node. Nodes in each layer are only connected to nodes in the adjacent layer. There is no connection between nodes in each layer, and nodes in each layer are fully connected.

[0107] The formula for the network input layer is

[0108]

[0109] The output layer formula is

[0110]

[0111] in, are the weight coefficients of the hidden layer and the output layer, respectively, g(x) and f(x) are the activation functions of the hidden layer and the output layer, respectively

[0112] Step 3-2: Next, design the weight update algorithm for the hidden layer. By defining the performance indicator function, use the gradient descent method to correct the weight coefficient in the negative gradient direction.

[0113] Define the system performance index function as

[0114]

[0115] According to the gradient descent method, the weight correction is adjusted according to the negative gradient direction of E(k) on the weight coefficient, and the weight learning algorithm formula of the network output layer and hidden layer is obtained as follows:

[0116]

[0117] Step 3-3, then make a neural network training set. According to the elastic collision formula derived in step 2, you need to prepare a data set

[0118]

[0119] When the motor is unloaded, the motor is accelerated with an acceleration of w′ i Uniform acceleration to set speed w i , applying a constant bias current i qi , after θΔ displacement, the velocity is measured

[0120] Step 3-4: Finally, train the neural network. Use the weight update algorithm from step 3-2 to train the neural network using the training set. Use the BP algorithm described above to iterate and adjust the weights until the performance function meets the requirements or the maximum number of iterations is reached.

[0121] In step 4, a dual-motor servo control system is designed to enable the actuator to accurately track the position signal. Specifically, the following steps are included:

[0122] Step 4-1: First, design the current loop, whose control input is

[0123]

[0124] Step 4-2, then design the speed loop input as

[0125]

[0126] Step 4-3, finally design the position loop control input as

[0127]

[0128] Step 5: Determine the backlash state of the dual-motor servo system, calculate and apply a bias current, specifically including the following steps:

[0129] Step 5-1, according to Figure 3 It can be seen that the backlash compensator receives the output signal of the speed loop, and the backlash compensator continuously detects the control output value of the speed loop during the operation of the system. Figure 3 As shown in the figure, the outermost layer is the position loop. The position controller receives the difference between position feedback and position command as input. The difference between the position controller output and velocity feedback is then fed into the current controllers of the two motors. The backlash compensator receives the output of the velocity controller and calculates the bias current required to eliminate backlash, which is then fed into the current controller. The current controller generates a voltage command to control the rotation of the motors. The motor output is then reduced by the reducer and used to drive the large gear. To synchronize the speeds of the two motors, the difference between the two motor speeds is fed back to the current controller to compensate for the speed difference.

[0130] Step 5-2: If the current loop control input is near zero, it means that the drive motor will decelerate and the dual-motor servo system will enter the backlash state. If the current loop control input is not near zero, it means that the dual-motor servo system is not in the backlash state and there is no need to apply bias torque to the control.

[0131] Step 5-3: Finally, the bias current is calculated based on the BP neural network and added to the current loop input.

[0132] According to the neural network data set in step 3-3, the input parameters of the BP neural network include the current angular velocity w of the motor. i and angular acceleration Can be directly measured, the motor's predetermined angular velocity Calculation is required. The motor's predetermined angular velocity is the motor speed before the gear collision. From the formula in step 2-2, we can know that the angular velocity before the gear collision can be obtained from the angular velocity of the system before and after the collision. It is assumed that the system speed does not change before the collision. Based on the premise of minimum system jitter, the system speed after the collision is set to 90% of the pre-collision speed.

[0133] according to Figure 4 As can be seen, the BP neural network receives the current state of the system as input and calculates the bias current. It then directly applies the bias current of equal magnitude and opposite direction to the current loop input terminals of the two motors. The bias current application stops when the system backlash state ends.

[0134] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0135] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive variable bias torque compensation method for a dual-motor servo system, characterized in that: The following steps are involved: Step 1: Analyze the dual-motor servo follower system, use the dead zone model to describe the backlash link, and establish the dynamic model of the dual-motor servo follower system; Step 2: Analyze the motion state of the motor's backlash link and establish a motion model for backlash generation; Step 3: Design a BP neural network model with the motor state as input and the bias current as output; Step 4: Design a control system based on the established dual-motor servo follower system dynamics model to enable the actuator to accurately track the position signal; Step 5: Determine whether the dual-motor servo follower system has entered a backlash state based on the control input of the established dual-motor servo control system. If it has entered a backlash state, calculate the bias current using a BP neural network and apply a pair of bias currents of equal magnitude and opposite directions to the dual motors. In step 2, the motion state of the motor backlash link is analyzed, a motion model of backlash is established, and the expected speed after the drive motor gear collides with the load gear is calculated. The specific steps are as follows: Step 2-1: When the motor side gear collides with the load side gear, it is an elastic collision. The collision process satisfies the conservation of angular momentum and angular kinetic energy. The corresponding formula is: Where J1 and J2 are the moments of inertia of the back-drive motor side and the load side, respectively; w1 and w2 are the angular velocities of the back-drive motor side and the load side before the collision, respectively; w1′ and w′2 are the angular velocities of the back-drive motor side and the load side after the collision, respectively; Step 2-2: Combine the formulas in step 2-1 to obtain the angular velocity of the back-drive motor side and the load side after the collision: Step 2-3: According to the formula in step 2-2, the speed of the reverse drive motor after the collision is calculated from the speeds of the motor and load before the collision. The load before the collision is still in contact with the gear of the forward drive motor, so the angular velocity of the load before the collision is equal to the angular velocity of the forward drive motor divided by the reduction ratio. To reduce the vibration of the load after the collision, the direction of the velocity after the collision remains unchanged, but the magnitude is reduced to 1 / 2 of the velocity before the collision.

2. The adaptive variable bias torque compensation method for a dual-motor servo system according to claim 1, characterized in that: Step 1: Establish a dynamic model of the dual-motor servo follower system. The model is used to design the dual-motor servo control system, including a single-motor mathematical model and a mechanical transmission model of the dual-motor system. The specific steps are as follows: Step 1-1: First, establish a mathematical model of a single motor in a synchronous rotating coordinate system; Among them, the stator voltage model is Where u d 、u q are the dq axis components of the stator voltage respectively; i d 、i q are the dq axis components of the electronic current; R is the electronic resistance; w e is the electrical angular velocity; L d 、L q are the dq axis inductance components respectively; f is the permanent magnet flux linkage; Among them, the electromagnetic torque equation is Where p n is the pole pair number, T e is the electromagnetic torque; Step 1-2, secondly, establish the transmission model of the dual-motor servo follower system; Among them, the mechanical motion equation on the motor side is In the formula is the equivalent moment of inertia of motor j, is the viscous friction coefficient equivalent to motor j, w j is the speed of motor j, T cj is the torque of pinion j, r cj is the reduction ratio of motor j, T ej is the electromagnetic torque of motor j; Among them, the dead zone model of tooth clearance is Where k is the rigidity coefficient, α j is the tooth gap size, z j is the relative displacement between the motor side and the load side; Among them, the mechanical motion equation on the load side is Where J m is the equivalent moment of inertia of the large gear, b m Equivalent viscous friction coefficient of the gear, T L is the load moment, r m is the reduction ratio between the large gear and the small gear, w m is the system gear speed.

3. The adaptive variable bias torque compensation method for a dual-motor servo system according to claim 1, characterized in that: In step 3, the trained neural network model is used to calculate the bias torque using the expected speed of the backdrive motor side gear before collision and the angular velocity when the bias torque is applied as input. The specific steps are as follows: Step 3-1. First, design a three-layer fully connected neural network using the BP algorithm. The BP neural network structure includes 4 input layer nodes, 5 hidden layer nodes, and 1 output layer node. Each layer node is only connected to the adjacent layer nodes. There is no connection between nodes within each layer, and the nodes in each layer are fully connected. The formula for the network input layer is In the formula is the input node; The input and output formula of the hidden layer is Where, is the output node of the first hidden layer, is the hidden layer weight coefficient, It is the input of the activation function of the first layer of the hidden layer, and the activation function uses the hyperbolic tangent function: The input and output formula of the output layer is Where, O out is the output node, is the output layer weight coefficient, Net out Input to the activation function of the output layer; The bias current applied to the output layer node The output layer activation function uses a non-negative sigmoid function, that is, Step 3-2, then design the weight update algorithm of the hidden layer; by defining the performance index function, use the gradient descent method to correct the weight coefficient in the negative gradient direction; Define the system performance index function as According to the gradient descent method, the weight correction is adjusted by searching the negative gradient direction of the weight coefficient according to E(k). The basic formula is: Where η is the weight correction coefficient, 0<η<1, is the weight correction value on the mth layer; The network output layer and hidden layer weight learning algorithm formula is obtained as follows: Step 3-3, then create a training set for neural network training; The training set is Where w i is the current angular velocity of motor i, is the current angular acceleration of motor i, Predetermine the angular velocity for motor i; i qi is the applied bias current; The back-drive motor does not receive any load torque from the load gear during the period of crossing the tooth gap and can be regarded as no-load operation. Therefore, the training set can be collected under the condition of no-load operation of the motor. When the motor is running at no load, i and angular acceleration Add a constant current of small amount i i , after Δθ displacement, the velocity is measured Get the dataset Where Δθ is the system backlash size; Step 3-4: Finally, train the neural network; use the weight update algorithm in step 3-2 to train the neural network using the prepared training set.

4. The adaptive variable bias torque compensation method for a dual-motor servo system according to claim 3, characterized in that: In step 4, a dual-motor servo control system is designed to enable the actuator to accurately track the position signal. Specifically, the following steps are included: Step 4-1: First, design the dual-motor servo follow-up system position control system based on three-loop PID control. The control system is divided into current loop, speed loop, and position loop from the inside out to achieve accurate control of motor current, speed, and position; Step 4-2, first design the current loop control. In the current loop control, take i d =0 control strategy, its simplified stator voltage model is: Design the current loop control input to be Where, P I is the current loop proportional coefficient, D I Current loop differential coefficient, I I is the current loop integral coefficient; Step 4-3, then design the speed loop control. The simplified mechanical motion equation and torque equation are: The speed loop input is designed to be Step 4-4, then design the position loop control Step 4-5. Finally, the designed three-loop PID control system is used for the position control of the dual-motor servo follower system. After the dual-motor servo follower system receives the position command, the speed loop, position loop, and current loop generate a controlled voltage command to enable the system to perform an action. The feedback link simultaneously feeds back the current, speed, and position state quantities to the three control links, and corrects the output of the control links so that the system can accurately track the position command.

5. The adaptive variable bias torque compensation method for a dual-motor servo system according to claim 4, characterized in that: In step 5, the backlash state of the dual-motor servo follower system is judged, calculated, and bias current is applied, which specifically includes the following steps: Step 5-1: First, determine whether the dual-motor servo servo system has entered the backlash state; Since the backlash state mainly occurs when the motor accelerates or decelerates, it is possible to determine whether the backlash state has occurred based on the relative size of the drive motor speed and the load speed. According to the formula in step 4-3, the motor acceleration is related to the electromagnetic torque, and the electromagnetic torque is proportional to the q-axis current. The backlash state can be determined by detecting the current loop control input. If the current loop control input is near zero and about to cross zero, it means that the drive motor is about to accelerate or decelerate, and the dual-motor servo servo system is about to enter the backlash state. If the current loop control input is not near zero, it means that the dual-motor servo servo system is not in the backlash state and there is no need to apply bias current to the control. Step 5-2 then calculates the bias current based on the BP neural network; According to the neural network data set in step 3-3, the current angular velocity w of the motor is used as input in the data set. i and angular acceleration Can be directly measured; motor predetermined angular velocity is the motor speed before the gear collision. From the formula in step 2-2, we can know that the angular velocity before the gear collision is obtained by the angular velocity of the system before and after the collision. Assuming that the system speed is unchanged before the collision, the system speed after the collision is set to 90% of the pre-collision speed based on the premise of minimum system jitter; The BP neural network accepts the current state input of the system to calculate the bias current, which is directly applied to the current loop input of the two motors, where the bias current of the back-drive motor is negative.

Images