A neural network-based motor resolver angle adaptive prediction compensation method

Abstract

The application discloses a motor rotary transformer angle adaptive prediction compensation method based on a neural network and belongs to the technical field of motor control, and solves the technical problems of long time and poor precision of the existing rotary transformer sensor zero position angle calibration mode, the method determines the calibration of the rotary transformer zero position angle by using the controller and the dynamic test bench of the motor itself, and comprises the following steps: S1. A motor to be calibrated is dragged to a set rotating speed by a motor measured and controlled machine; S2. A d-axis current Id is given by motor control; S3. A rotary transformer zero position angle delta is calibrated and modified; S4. A q-axis current Iq is given by setting a program; S5. The output torque of the motor to be calibrated is read and recorded as T+; S6. The q-axis current -Iq is given by setting a program; S7. The output torque of the motor to be calibrated is read again and recorded as T-; S8. The operation of S4 to S7 is repeated N times, and the rotary transformer zero position angle is added with a preset precision each time; and S9. The input end of a neural network model is the sum of the positive torque and the negative torque |T + +T ‑ |, the output end is the zero position angle delta, and the input zero value is inputted, so that the real rotary transformer zero position angle delta can be predicted.

Description

Technical Field

[0001] This invention relates to the field of motor control technology, and more specifically, to a method and apparatus for adaptive prediction and compensation of motor resolver angle based on neural networks. Background Technology

[0002] According to the vector control of permanent magnet synchronous motors, in order to maximize the output torque of the motor, the electromagnetic field generated by the stator windings must always be orthogonal to the rotor's permanent magnetic field, which requires accurate measurement of the rotor position angle. Ideally, the motor's design phase can ensure that the resolver's zero-position deviation coincides with the A-axis. However, in reality, due to machining and installation deviations during motor production, the resolver's installation and positioning are inconsistent, resulting in inconsistent resolver deflection angles for each motor. Therefore, each motor needs to undergo resolver zero-position deflection angle calibration during offline testing. Currently, this is mainly done manually, which is time-consuming and has low accuracy.

[0003] Motor vector control requires accurate control of the amplitude and phase of the current vector. When the motor operates at a high frequency, the time delays of various parts of the motor control system related to current sampling and rotor position sampling will cause the actual feedback current of the motor to fail to track the given current after the vector control reaches a steady state. It will also cause the calculated PWM duty cycle to fail to be accurately applied to the next PWM cycle. As the motor operating frequency increases, the error between the feedback current and the given current increases, which will have a significant impact on the control accuracy and operating performance of the motor under high-frequency operating conditions. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the above-mentioned shortcomings of the prior art. The purpose of the present invention is to provide a method for adaptive prediction and compensation of motor resolver angle based on neural network.

[0005] The technical solution of this invention is: an adaptive prediction and compensation method for motor resolver angle based on neural networks, characterized in that the zero-position deflection angle of the resolver is determined by using the motor's own controller and a dynamic test bench, including the following steps:

[0006] Step S1. Supply DC rated voltage to the motor control, and the motor control is in torque mode; at the same time, the motor to be calibrated is dragged to the set speed by the test and control computer. The set speed cannot be in the field weakening speed zone. The test and control computer records the output torque of the motor to be calibrated.

[0007] Step S2. Control the given d-axis current Id via motor control;

[0008] Step S3. Calibrate and modify the zero-position deflection angle of the resolver, denoted as δ;

[0009] Step S4. Set the given q-axis current Iq in the program;

[0010] Step S5. Read the output torque of the motor to be calibrated and record it as T+;

[0011] Step S6. Set the given q-axis current -Iq in the program;

[0012] Step S7. Read the output torque of the motor to be calibrated again and record it as T-;

[0013] Step S8. Repeat steps S4 to S7 N times, each time calibrating and modifying the zero-position deflection angle of the resolver, and automatically adding a preset precision to the zero-position deflection angle of the resolver.

[0014] Step S9. Establish a neural network model for prediction, with the input being the sum of the positive and negative torques recorded N times |T + +T - The output is the zero-position deflection angle δ recorded N times; input zero value into the established neural network model; and the true zero-position deflection angle δ of the resolver can be predicted, with the accuracy being the preset accuracy.

[0015] As a further improvement, step S10 is also included, which modifies the preset accuracy of step S8 to a specified high accuracy, and then repeats the operation of steps S3 to S9 to advance the prediction accuracy to the specified high accuracy.

[0016] Furthermore, the high precision is specified as 0.01.

[0017] Furthermore, the preset precision is 0.1.

[0018] Furthermore, N in step S8 is not less than 100.

[0019] Beneficial effects

[0020] Compared with the prior art, the advantages of this invention are as follows:

[0021] 1. This invention can run automatically after establishing an algorithm model, without the need for human intervention. It has a high degree of automation and saves manpower and time costs compared to traditional methods.

[0022] 2. The accuracy of predictions in this invention can be increased by increasing the number of program loops to achieve higher prediction accuracy than manual methods. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of a stationary coordinate system ABC;

[0024] Figure 2 This is a schematic diagram of the stationary coordinate system αβ;

[0025] Figure 3 This is a schematic diagram of the rotor synchronous rotation coordinate system dq;

[0026] Figure 4 This is a schematic diagram showing the D-axis coinciding with the zero position of the resolver;

[0027] Figure 5 This is a schematic diagram of the rotor position angle;

[0028] Figure 6 This is a flowchart of the data collection process in this invention;

[0029] Figure 7 This is a diagram of a neural network model. Detailed Implementation

[0030] The present invention will be further described below with reference to specific embodiments shown in the accompanying drawings.

[0031] See Figures 1 to 7 Definition of motor resolver angle parameters:

[0032] 1) Zero-position deflection angle of resolver sensor

[0033] Taking a three-phase permanent magnet synchronous motor as an example, based on motor vector control technology, each coordinate system can be determined.

[0034] Stationary coordinate system ABC: The three phases of the stator windings are symmetrical, with their axes differing by 120 degrees. Using the stator's UVW phases as a reference, the stationary coordinate system ABC is determined, as follows: Figure 1 As shown.

[0035] 2) Stationary coordinate system αβ: The α-axis coincides with the A-axis, and the β-axis leads the α-axis by 90 degrees, such as... Figure 2 As shown.

[0036] 3) Rotor synchronous rotation coordinate system dq: The central axis of the N pole where the magnetic field is generated by the motor rotor is taken as the direct axis d-axis; the position leading the direct axis by 90 degrees is defined as the quadrature axis q-axis. The dq axis rotates at the rotor's synchronous angular velocity ω. Assuming the direction of the rotor's counterclockwise rotation is positive, such as... Figure 3 As shown.

[0037] 4) Resolver Zero Position: This refers to the zero position of the resolver position sensor. When the induced voltage in the sinusoidal output winding of the resolver is at its minimum, the rotor position is the electrical zero position, and the output voltage is the zero-position voltage. Assuming that when the dq-axis coordinate system rotates to the d'q' position, the angle actually measured and output by the resolver sensor is zero, then the d'-axis position is defined as the resolver zero position. Figure 4 As shown, the zero point of the resolver is fixed.

[0038] 5) The actual angle θ measured and output by the resolver sensor: Figure 4 When the d-axis coincides with the zero position of the resolver, and the rotor continues to rotate counterclockwise, the zero position of the resolver and the d-axis will form an angle θ, such as... Figure 5As shown, the included angle θ is the angle actually measured and output by the resolver sensor. When the rotor rotates to the point where the d-axis coincides with the zero position, the angle θ actually measured and output by the resolver sensor is 0. Figure 4 As shown.

[0039] 6) Resolver sensor zero-position offset angle δ: This is the angle between the resolver zero position and the A-axis, which is the angle that the motor needs to be calibrated. For example... Figure 4 As shown.

[0040] 7) Motor rotor position angle θr: the angle between the d-axis and the A-axis, such as... Figure 5 As shown. It can be seen that: θr=θ+δ. The torque equation of the PSM motor:

[0041]

[0042] When the resolver zero-position angle and the resolver sampling are correct, the motor output torque is T when Id and Iq are given; when Id and -Iq are given, the motor output torque is -T. Under a given current, the motor output torque T is maximum only when the resolver zero-position angle is correct.

[0043] Traditional standardization method:

[0044] Manual calibration: Using a low-voltage DC power supply and a resolver sensor calculation device, connect the U phase of the motor winding to the positive terminal, and the V and W phases to the negative terminals. When the power is turned on, the motor rotor rotates to a certain position. At this time, the resolver sensor calculation device reads the angle θ. If 90 < θ < 360, then the resolver sensor zero-position deflection angle is: δ = 360 - θ; if θ <= 90, then the resolver sensor zero-position deflection angle is: δ = θ.

[0045] Automatic calibration: The manual calibration process mentioned above is built into the motor controller. When the motor is unloaded, the DC voltage of the motor controller is given, and the current of the U phase is controlled to a certain value. The currents of the V phase and W phase are correspondingly negative values ​​(i.e., the upper half bridge of the U phase is closed, and the lower half bridge of the V / W phase is closed). Then the motor rotor will move to a certain fixed position, that is, the A axis and the d axis coincide. The angle read by the resolver is the zero position deflection angle of the resolver.

[0046] Due to bearing friction and inertia, the traditional calibration process described above will cause a deviation in the alignment of the resolver zero position with the A-axis. Furthermore, due to mechanical tolerances, the angular offset measured for each electrical cycle of a single mechanical rotation may vary. Therefore, it is necessary to determine the resolver zero-position deflection angle by repeatedly calibrating for each electrical cycle and calculating the average angle.

[0047] This invention improves a neural network-based adaptive prediction and compensation method for motor resolver angle. It utilizes the motor's own controller and a dynamic test bench to determine the calibration of the resolver zero-position deflection angle, and includes the following steps:

[0048] Step S1. Supply DC rated voltage to the motor control, and the motor control is in torque mode; at the same time, the motor to be calibrated is dragged to the set speed by the test and control machine. The set speed cannot be in the field weakening speed zone. For example, the set speed is 3000 rpm. The test and control machine records the output torque of the motor to be calibrated.

[0049] Step S2. Control the given d-axis current Id via motor control;

[0050] Step S3. Calibrate and modify the zero-position deflection angle of the resolver, denoted as δ;

[0051] Step S4. Set the given q-axis current Iq in the program;

[0052] Step S5. Read the output torque of the motor to be calibrated and record it as T+;

[0053] Step S6. Set the given q-axis current -Iq in the program;

[0054] Step S7. Read the output torque of the motor to be calibrated again and record it as T-;

[0055] Step S8. Repeat steps S4 to S7 N times, where N is not less than 100. Each time, the zero-position deflection angle of the resolver is modified. The zero-position deflection angle of the resolver is automatically increased by a preset precision, which is expressed as δ = δ + preset precision in automatic programming. In this embodiment, the preset precision is 0.1, i.e., δ = δ + 0.1.

[0056] Step S9. Establish a neural network model for prediction, with the input being the sum of the positive and negative torques recorded N times |T + +T - The output is the zero-position deflection angle δ recorded N times; input zero value into the established neural network model; and the true zero-position deflection angle δ of the resolver can be predicted. At this time, the accuracy is the preset accuracy, that is, the accuracy is 0.1.

[0057] Furthermore, the method includes step S10, which modifies the preset accuracy in step S8 to a specified high accuracy, and then repeats the operations from steps S3 to S9 to advance the prediction accuracy to the specified high accuracy. In this embodiment, the specified high accuracy is 0.01, so the expression in step S8 is δ=δ+0.01, and the final prediction accuracy is 0.01.

[0058] This invention, once the algorithm model is established, can run automatically without human intervention, achieving a high degree of automation and saving manpower and time costs compared to traditional methods. The accuracy of predictions can be increased by increasing the number of program iterations to achieve even higher prediction accuracy than manual methods.

[0059] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention, and these will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. A method for adaptive prediction and compensation of motor resolver angle based on neural networks, characterized in that, The calibration of the resolver zero-position deflection angle is determined using the motor's own controller and a dynamic test bench, including the following steps: Step S1. Supply DC rated voltage to the motor control, and the motor control is in torque mode; at the same time, the motor to be calibrated is dragged to the set speed by the test and control computer. The set speed cannot be in the field weakening speed zone. The test and control computer records the output torque of the motor to be calibrated. Step S2. Control the given d-axis current Id via motor control; Step S3. Calibrate and modify the zero-position deflection angle of the resolver, denoted as δ; Step S4. Set the given q-axis current Iq in the program; Step S5. Read the output torque of the motor to be calibrated and record it as... ; Step S6. Set the given q-axis current -Iq in the program; Step S7. Read the output torque of the motor to be calibrated again and record it as... ; Step S8. Repeat steps S4 to S7 N times, each time calibrating and modifying the zero-position deflection angle of the resolver, and automatically adding a preset precision to the zero-position deflection angle of the resolver. Step S9. Establish a neural network model for prediction, with the input being the sum of the positive and negative torques recorded N times. The output is the zero-position deflection angle δ recorded N times; input zero value into the established neural network model; and the true zero-position deflection angle δ of the resolver can be predicted, with the accuracy being the preset accuracy.

2. The adaptive prediction and compensation method for motor resolver angle based on neural network according to claim 1, characterized in that, It also includes step S10, which modifies the preset accuracy of step S8 to a specified high accuracy, and then repeats the operation of steps S3 to S9 to advance the prediction accuracy to the specified high accuracy.

3. The adaptive prediction and compensation method for motor resolver angle based on neural networks according to claim 2, characterized in that, The specified high precision is 0.

01.

4. The adaptive prediction and compensation method for motor resolver angle based on neural network according to claim 1, characterized in that, The preset precision is 0.

1.

5. The adaptive prediction and compensation method for motor resolver angle based on neural network according to claim 1, characterized in that, In step S8, N is not less than 100.

Images