A chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive.

Abstract

This invention provides a chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive, employing a zero-delay, rapid calculation three-vector model for predictive control of the motor; the input of the zero-delay, rapid calculation three-vector model is the target current; the current parameter Pr(k) in the zero-delay, rapid calculation three-vector model is [RL]. s ψ f ] T The parameters are estimated in real time using a forgetting factor recursive least squares method parameter identification algorithm; where R is the stator resistance of the motor; L s For the stator inductance of the motor; ψ f The rotor flux linkage of the motor is used. The forgetting factor recursive least squares parameter identification algorithm is implemented based on a q-axis discretized mathematical model and a least squares recursive formula with a forgetting factor. This method can effectively reduce current harmonic distortion rate, improve torque stability, and suppress the influence of external environmental changes such as temperature on motor parameters, achieving accurate parameter identification. The model predictive control algorithm can then better control the robot chassis, making its movement smoother and ensuring welding stability.

Description

Technical Field

[0001] This invention relates to the field of humanoid welding technology, and more specifically, to a chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive. Background Technology

[0002] Welding technology, like the tailor of industrial manufacturing, ensures robust industrial production capacity and has wide applications in the manufacturing sector. In recent years, the demand for marine resource development has continued to grow, leading to a corresponding increase in the demand for subsea pipeline welding. Currently, the most common method for subsea pipeline welding is wet welding. However, wet welding requires welders to operate underwater, and manual welding suffers from low efficiency and inconsistent quality, which can no longer meet the requirements of modern industrial production. Furthermore, the ability to perform welding operations simultaneously on land and underwater is also a problem that needs to be solved.

[0003] To address the aforementioned problems, Chinese patent "An Underwater Welding Robot and Its Operating Process" (publication number CN114669933A) describes a device that can adjust the welding torch posture according to the welding shape and spatial position. However, this device moves via a propulsion device, and underwater currents prevent it from consistently reaching the welding position, thus compromising welding accuracy and stability. Furthermore, because it operates via a propulsion system, it cannot perform welding operations on land. Another example is Chinese patent "A Chamber-Type Local Dry Underwater Welding Robot" (publication number CN112743192A). This device employs a chamber-type local dry underwater welding technique, ensuring welding quality. However, this device is large and moves via a triangular tracked wheel structure. While it can operate both underwater and on land, it cannot operate on negatively inclined magnetic surfaces or in confined spaces. Additionally, the tracked wheel structure is susceptible to underwater current fluctuations, leading to unstable welding.

[0004] Therefore, it is necessary to design a chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive, which can perform welding operations not only on land magnetic surfaces but also on underwater magnetic surfaces. The high control performance of the motor algorithm can ensure welding stability and improve welding quality. Summary of the Invention

[0005] To overcome the shortcomings and deficiencies of the existing technology, the purpose of this invention is to provide a chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive. This method can effectively reduce the current harmonic distortion rate, improve torque stability, and suppress the influence of external environmental changes such as temperature on motor parameters, achieve accurate parameter identification, and enable the model predictive control algorithm to better control the robot chassis, making the robot chassis move more smoothly and ensuring welding stability.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a control method for the chassis of an amphibious humanoid welding robot based on magnetic wheel drive, used to control the magnetic wheel driven by the motor in the chassis of the amphibious humanoid welding robot;

[0007] Predictive control of the motor is performed using a deadbeat-free, rapid calculation three-vector model; the input to the deadbeat-free, rapid calculation three-vector model is the target current I*; the current parameter Pr(k) in the deadbeat-free, rapid calculation three-vector model is [RL]. s ψ f ] T The parameters are estimated in real time using a forgetting factor recursive least squares method parameter identification algorithm; where R is the stator resistance of the motor; L s For the stator inductance of the motor; ψ f For motor rotor flux linkage;

[0008] The forgetting factor recursive least squares parameter identification algorithm is implemented based on the q-axis discretized mathematical model and the least squares recursive formula with forgetting factor.

[0009] The q-axis discretization mathematical model is as follows:

[0010] Where y(k)=I q (k);

[0011] Input matrix to the system;

[0012] η(k)=[β q1 β q2 β q3 β q4 ] T The parameter matrix to be identified;

[0013] Among them, I q (k-1) is the q-axis current at the previous moment; I d (k) is the d-axis current at the current moment; u q (k) is the q-axis voltage at the current moment; ω e (k) is the current electric angular velocity of the rotor; β q1 β q2 β q3 β q4 These are the coefficients that change over time;

[0014] The least squares recursive formula with forgetting factor is as follows:

[0015]

[0016] in, This is the estimated value at the current moment. P(k) is the estimated value at the previous time step; P(k-1) is the covariance matrix at the current time step; L(k) is the gain matrix; E(k) is the identity matrix.

[0017] The forgetting factor recursive least squares parameter identification algorithm refers to:

[0018] y(k) is measured; and the estimated value at the current time is calculated. Covariance matrix P(k) and gain matrix L(k);

[0019] When the convergence value of the parameter matrix η(k) to be identified is calculated Then, calculate the motor parameters to be identified, Pr(k) = [RL]. s ψ f ] T :

[0020]

[0021] The relative error err of the motor identification parameters is calculated based on the identification parameters Pr(k-1) of the previous time step and the identification parameters Pr(k) of the current time step. The relative error err is used to determine whether the parameters have fully converged.

[0022] If the parameters have converged completely, the current identified parameter Pr(k) is substituted into the three-vector model of the zero-difference beat calculation for predictive control; if the parameters have not converged, the current cycle is skipped, and the identified parameter of the previous cycle is substituted into the three-vector model of the zero-difference beat calculation for predictive control.

[0023] Preferably, in the q-axis discretized mathematical model, the coefficient β q1 β q2 β q3 β q4 With d-axis current L d q-axis current L q Stator resistance R, motor rotor flux linkage ψ f The mathematical relationship between the control algorithm period T and the control algorithm period T is as follows:

[0024]

[0025] Preferably, in the recursive least squares parameter identification algorithm for the forgetting factor, when k=1, the estimated value is first initialized. Initialize the covariance value P(0) = αE(k); then start the forgetting factor recursive least squares parameter identification algorithm.

[0026] The relative error err is:

[0027]

[0028] If err < threshold, it means the parameters have fully converged; otherwise, it means the parameters have not converged.

[0029] Preferably, the input to the deadbeat-free rapid calculation three-vector model is the target current I*;

[0030] The zero-difference-rate rapid calculation three-vector model selects an optimal vector u from six effective voltage vectors Ui, i = [1,2,3,4,5,6] based on the target current I*. p_1 and a suboptimal vector u p_2 And calculate the optimal vector u p_1 and suboptimal vector u p_2 The duration of action; based on the optimal vector u p_1 and suboptimal vector u p_2 The duration of the voltage vector is determined by the inverter, which generates a corresponding voltage vector that acts on the motor.

[0031] Preferably, in the deadbeat-free rapid calculation three-vector model, an optimal vector u is selected from the six effective voltage vectors Ui, i = [1,2,3,4,5,6]. p_1 and a suboptimal vector u p_2 , refers to:

[0032] First, based on the deadbeat control principle of dq-axis current, the optimal voltage vector u for the next moment is calculated from the full voltage vector range. p_dq (k+1):

[0033] u p_dq (k+1)=AI dq (k)+BI dq * +C;

[0034]

[0035] Among them, u p_d (k+1), u p_q (k+1) represent the globally optimal voltage components along the d-axis and q-axis at the next time step, respectively; I d (k), I q (k) represents the d-axis and q-axis currents collected at the current moment; I dq * Indicates the target current; A, B, and C represent coefficients, respectively;

[0036] Then, the globally optimal voltage vector u p_dq (k+1) is transformed into the value u in the two-phase stationary coordinate system. p_αβ (k+1); Judgment value u p_αβ The sector where (k+1) is located; based on the value u p_αβThe sector containing (k+1) selects the optimal vector u. p_1 and suboptimal vector u p_2 .

[0037] Preferably, substituting the estimated motor parameters into the zero-delay rapid calculation three-vector model means: substituting the estimated motor parameters into the calculation formulas for coefficients A, B, and C; the calculation formulas for parameters A, B, and C are as follows:

[0038]

[0039] Where T is the control period.

[0040] Preferably, the target current I* includes the d-axis component I of the target current. d * Target current q-axis component I q *; Set the target current d-axis component I d * = 0;

[0041] The target current q-axis component I q The method to obtain * is: set the target speed Speed * The actual speed, acquired and processed by the encoder, is input to the speed PI controller, which outputs the target current q-axis component I. q *

[0042] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0043] 1. This invention employs a novel forgetting factor recursive least squares parameter identification-rapid calculation three-vector model predictive current control algorithm to control the current (torque) of the chassis permanent magnet synchronous motor. Compared with the current PI closed loop in the existing traditional vector control algorithm FOC, this method has better dynamic response performance, can effectively cope with changes in current (torque), can better adapt to the pulse fluctuations generated during the welding process, ensure the stability of the chassis movement during welding, and does not require repeated adjustment of the parameters in the controller, thus simplifying the controller. Compared with the traditional model predictive current control algorithm, this algorithm does not need to calculate the value function to predict the voltage vector required at the next moment. Instead, it generates the voltage vector required at the next moment through deadbeat control of the dq axis current. This reduces the number of predictions and thus the calculation time, reducing the computational burden on the control system.

[0044] 2. The algorithm used in this invention combines the forgetting factor recursive least squares parameter identification algorithm; compared with the traditional offline parameter identification FOC algorithm, this algorithm identifies motor parameters online, can identify motor parameters in real time, has strong parameter robustness, and can effectively suppress the impact of motor parameter changes on the model predictive control algorithm. Attached Figure Description

[0045] Figure 1 This is a structural schematic diagram of the amphibious humanoid welding robot of the present invention;

[0046] Figure 2 This is a schematic diagram of the layout of the watertight control box of the chassis of the amphibious humanoid welding robot of the present invention;

[0047] Figure 3 This is a block diagram of the chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to the present invention;

[0048] Figure 4 This is a flowchart of parameter identification for the chassis control method of the amphibious humanoid welding robot based on magnetic wheel drive, as described in this invention.

[0049] Figure 5 This is a performance comparison chart between the chassis control method of the amphibious humanoid welding robot based on magnetic wheel drive of the present invention and traditional algorithms. Detailed Implementation

[0050] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0051] Example

[0052] This embodiment discloses a control method for the chassis of an amphibious humanoid welding robot based on magnetic wheel drive; the method is used to control the magnetic wheel driven by the motor in the chassis of the amphibious humanoid welding robot.

[0053] The structure of an amphibious humanoid welding robot is as follows: Figure 1 and Figure 2 As shown, it includes a chassis 1, a six-degree-of-freedom robotic arm, a robot head 13, a body 14 connecting the various parts of the robot, an underwater searchlight 5, and an underwater camera 6.

[0054] The chassis includes a chassis transmission structure, a support bracket for carrying the chassis transmission structure, a platform for connecting the upper robotic arm, a magnetic wheel 4, and a watertight control box 3. The chassis transmission structure includes a watertight high-torque permanent magnet synchronous motor and a harmonic reducer. The host computer outputs control commands to the motor driver, which controls the motor to rotate, transmitting power to the magnetic wheel via the harmonic reducer to achieve chassis movement. The watertight control box 3 mainly houses the brushless motor drive board, robot main controller 9, switch 11, 48V DC power module 7, 48V to 24V power converter 8, wire feeder motor driver 10, and a splitter. The watertight control box 3 uses watertight terminals to connect the host computer to electrical equipment and the electrical equipment to the chassis motor, preventing water ingress that could damage or short-circuit the controller and other electrical equipment.

[0055] The brushless motor driver board includes a core circuit based on the STM32F405RGT6 main control chip, a drive circuit based on the DRV8301 driver chip, a three-phase full-bridge inverter circuit based on parallel SiC-MOSFETs, and a sampling circuit composed of high-precision sampling resistors and operational amplifiers.

[0056] The robot main control unit 9 includes connections to the chassis motor drivers, robotic arm motor drivers, wire feeder motor drivers, welding power signal receivers and transmitters, and an industrial control computer. The robot main control unit 9 is primarily responsible for sending control signals to the chassis motors, the six-joint robotic arm motors, the wire feeder motors, welding power analog signals, and control signals to the underwater searchlight 5 and the underwater camera 6. The wire feeder motor driver 10 is primarily responsible for supplying driving energy to the wire feeder motor and controlling it using the control signals issued by the robot main control unit. The switch 11 is primarily responsible for communication between the chassis motors, robotic arm motors, and the host computer.

[0057] Based on the chassis's power requirements, including chassis movement speed, turning speed, and turning radius, the target speeds of the four power motors can be obtained through chassis dynamics model analysis and calculation. These target speeds are then input into the chassis motor control algorithm to achieve open-loop control of the robot chassis motion. The speed closed-loop control of the amphibious humanoid welding robot chassis power motors, based on the forgetting factor recursive least squares method-rapid calculation three-vector model predictive current control algorithm, includes two closed-loop modules, an inverter module, a sampling module, a feedback module, a coordinate transformation module, and a parameter identification module. The two closed-loop modules are the outer speed closed loop and the inner current closed loop. The speed closed loop uses traditional PI algorithm control, while the inner current loop uses a novel parameter identification-three-vector model predictive current control algorithm.

[0058] Specifically, this invention employs a fast-calculation three-vector model without error for predictive control, such as... Figure 3 As shown, the input to the deadbeat-free rapid calculation three-vector model is the target current I*; the current parameter Pr(k) in the deadbeat-free rapid calculation three-vector model is [RL]. s ψ f ] T The parameters are estimated in real time using a forgetting factor recursive least squares method parameter identification algorithm; where R is the stator resistance of the motor; L s For the stator inductance of the motor; ψ f The motor rotor flux is used; the estimated motor parameters are substituted into the zero-delay speed calculation three-vector model for prediction; the motor control of the magnetic wheel is performed based on the predicted values ​​obtained from the zero-delay speed calculation three-vector model.

[0059] The target current I* includes the d-axis component I of the target current. d * Target current q-axis component I q * Target current q-axis component Iq The method to obtain * is: set the target speed Speed * The actual speed, acquired and processed by the encoder, is input to the speed PI controller, which outputs the target current q-axis component I. q *. This invention uses the d-axis component I of the target current. d The *=0 control strategy can achieve static decoupling of the motor's d-axis and q-axis currents, that is, it can control the motor's d-axis and q-axis currents independently, and can achieve better control of the motor speed and torque.

[0060] The parameter identification algorithm of the recursive least squares method with forgetting factor is implemented based on the q-axis discretized mathematical model and the least squares recursive formula with forgetting factor.

[0061] The process of constructing the q-axis discretized mathematical model is as follows: First, the three parameters of the motor are identified by the forgetting factor recursive least squares parameter identification algorithm.

[0062] Motor mathematical model

[0063]

[0064] Where R is the stator resistance; L d L q For the AC and DC axis stator inductances of the motor; I d I q u d u q These represent the AC and DC axis currents and voltages of the motor, respectively; ψ f For the rotor flux linkage of the motor; ω e This is the electric angular velocity of the motor rotor.

[0065] The above differential model is discretized using forward Euler discretization, separating the parameters and variables. The discretized result is in the following form:

[0066]

[0067] Among them, I d (k), I q (k) represent the d-axis and q-axis currents at the current time, respectively; I d (k-1), I q (k-1) represent the d-axis and q-axis currents at the previous moment, respectively; u d (k), u q (k) represents the d-axis and q-axis voltages at the current moment, respectively; ω e (k) is the current electric angular velocity of the rotor; β d1 β d2 β d3 β d4and β q1 β q2 β q3 β q4 It is a coefficient that changes with time, and it is related to L. d L q , R, ψ f The mathematical relationship between the control algorithm period T and the control algorithm period T is as follows:

[0068]

[0069] The motor parameter to be identified is L d L q , R, ψ f Therefore, the q-axis discretization mathematical model is chosen as the basic model for parameter estimation. Simplifying the model yields:

[0070]

[0071] Where y(k)=I q (k);

[0072] Input matrix to the system;

[0073] η(k)=[β q1 β q2 β q3 β q4 ] T The parameter matrix to be identified;

[0074] Among them, I q (k-1) is the q-axis current at the previous moment; I d (k) is the d-axis current at the current moment; u q (k) is the q-axis voltage at the current moment; ω e (k) is the current electric angular velocity of the rotor; β q1 β q2 β q3 β q4 These are the coefficients that change over time;

[0075] Based on the above recursive parameter identification model, the least squares recursive formula with forgetting factor is established as follows:

[0076]

[0077] in, This is the estimated value at the current moment. P(k) is the estimated value from the previous time step; P(k) is the covariance matrix at the current time step, and P(k-1) is the covariance matrix from the previous time step; L(k) is the gain matrix; E(k) is the identity matrix; initial values ​​are estimated. The initial value of covariance is P(0) = αE(k), where α is generally a large number between 1000 and 10000; λ is the forgetting factor, which is generally between 0.97 and 1.

[0078] The forgetting factor recursive least squares parameter identification algorithm, such as Figure 4 What is shown is:

[0079] y(k) is measured; and the estimated value at the current time is calculated. Covariance matrix P(k) and gain matrix L(k);

[0080] When the convergence value of the parameter matrix η(k) to be identified is calculated Then, calculate the motor parameters to be identified, Pr(k) = [RL]. s ψ f ] T :

[0081]

[0082] Calculate the relative error err of the motor identification parameters, and use the relative error err to determine whether the parameters have fully converged:

[0083]

[0084] Where Pr(k) represents the identification parameters at the current time; Pr(k-1) represents the identification parameters at the previous time.

[0085] If err < threshold (e.g. 0.01%), it means that the identification error in the algorithm is small and the parameters have fully converged. Substitute the identification parameter Pr(k) at the current time into the error-free beat calculation three-vector model for predictive control.

[0086] If err ≥ threshold, it means that the identification error in the algorithm is large and the parameters have not converged. In this case, the current cycle is skipped, and the identification parameters with large identification errors are not substituted into the controller. Instead, the identification parameters of the previous cycle are substituted into the zero-beat rapid calculation three-vector model for predictive control.

[0087] By incorporating the three motor parameters estimated in real time by the forgetting factor recursive least squares parameter identification algorithm into the zero-delay rapid calculation three-vector model predictive current control algorithm, the motor parameters affected by changes in environmental factors such as temperature can be effectively identified, thereby reducing the impact of parameter changes on the model predictive control algorithm.

[0088] Secondly, by analyzing the motor parameters identified by the above algorithm and substituting them into the zero-delay, fast-calculation three-vector model predictive current control algorithm to replace the original model's motor parameters, accurate identification of the motor model and control of the motor state can be achieved.

[0089] The zero-difference-rate rapid calculation three-vector model selects an optimal vector u from six effective voltage vectors Ui, i = [1,2,3,4,5,6] based on the target current I*. p_1 and a suboptimal vector u p_2 And calculate the optimal vector u p_1 and suboptimal vector u p_2 The duration of action; based on the optimal vector u p_1 and suboptimal vector u p_2 The duration of the voltage vector is determined by the inverter, which generates a corresponding voltage vector that acts on the motor.

[0090] Specifically, in the deadbeat-free rapid calculation three-vector model, the method for obtaining the six effective voltage vectors Ui is as follows:

[0091] Calculate the d-axis voltage values ​​u of the six effective voltage vectors Ui in the switching state. d [i] and q-axis voltage value u q [i]:

[0092]

[0093] Among them, U dc θ is the motor bus voltage; e Let θ be the electrical angle of the motor. e =pθ, where p is the number of pole pairs of the motor and θ is the detection angle; S_abc[i][j] is the j-th bit of the switch state value S_abc[i]. The switch state value S_abc[i] is shown in Table 1:

[0094] Table 1 shows the values ​​of the switch status S_abc[i].

[0095] Voltage vector S_abc Voltage vector S_abc U1 001 U4 100 U2 010 U5 101 U3 011 U6 110

[0096] Selecting one optimal vector and one suboptimal vector from the six effective voltage vectors Ui means:

[0097] First, based on the deadbeat control principle of dq-axis current, the globally optimal voltage vector u for the next moment is calculated from the full voltage vector range. p_dq (k+1):

[0098] u p_dq (k+1)=AI dq (k)+BI dq * +C;

[0099]

[0100] Among them, u p_d (k+1), u p_q(k+1) represent the globally optimal voltage components along the d-axis and q-axis at the next time step, respectively; I d (k), I q (k) represents the d-axis and q-axis currents collected at the current moment; I dq * Indicates the target current; A, B, and C represent coefficients; A, B, and C are respectively:

[0101]

[0102] Where R is the stator resistance; L s It is the stator inductance; ω e It is the electric angular velocity of the motor; ψ f It is the magnetic flux of a permanent magnet; T s To control the cycle;

[0103] Substituting the identified motor parameters obtained from the previous calculations into the above formula, we can obtain the correction values ​​of these three matrices for subsequent model predictive control.

[0104] Then, the globally optimal voltage vector u p_dq (k+1) is transformed into the value u in the two-phase stationary coordinate system. p_αβ (k+1); Judgment value u p_αβ The sector containing (k+1); value u p_αβ The sector where (k+1) is located is determined by the sector determination method of the SVPWM algorithm;

[0105] According to the value u p_αβ The sector containing (k+1) selects the optimal vector u. p_1 and suboptimal vector u p_2 The optimal and suboptimal vectors can be selected by looking up the correspondence table between sectors and optimal and suboptimal vectors, as shown in Table 2.

[0106] Table 2. Correspondence between sectors and optimal and suboptimal vectors.

[0107] <![CDATA[u p_αβ (k+1) is located in the sector. <![CDATA[u p_1 ,in p_2 ]]> <![CDATA[u p_αβ (k+1) is located in the sector. <![CDATA[u p_1 ,in p_2 ]]> Ⅰ U4, U6 Ⅳ U3,U1 Ⅱ U6,U2 Ⅴ U1,U5 Ⅲ U2,U3 Ⅵ U5 U4

[0108] Optimal vector u p_1 and suboptimal vector u p_2 The method for calculating the duration of action is as follows:

[0109] Based on dq-axis current deadbeat control, the output voltage is determined by the optimal vector u. p_1 Suboptimal vector u p_2 It consists of the zero vector u0. Therefore, the control period is set to T. s :T s =t p1 +t p2+t0; where t p1 For the optimal vector u p_1 Duration of action; t p2 For suboptimal vector u p_2 The duration of action; t0 is the duration of action of the zero vector u0;

[0110] Set the optimal vector u for motor action p_1 Suboptimal vector u p_2 After the zero vector u0, the resulting d-axis current change rates are s p1_d s p2_d and s 0_d The resulting q-axis current change rate is s p1_q s p2_q and s 0_q The expression for each rate of change of current is as follows:

[0111]

[0112] Among them, u p1_d and u p1_q These are the optimal vectors u p_1 d-axis and q-axis voltage components; u p2_d and u p2_q These are the suboptimal vectors u p_2 d-axis and q-axis voltage components; I d I q These are the d-axis and q-axis components of the detected motor current, respectively.

[0113] The application time is allocated based on the deadbeat principle of dq-axis current, as expressed below:

[0114] I d (k+1)=s p1_d t p1 +s p2_d t p2 +s 0_d t0 = I d *

[0115] I q (k+1)=s p1_q t p1 +s p2_q t p2 +s 0_q t0 = I q *

[0116] By transforming the above formula, the optimal vector u can be obtained. p_1 Suboptimal vector u p_2 The formula for calculating the duration of action of the zero vector u0 is as follows:

[0117]

[0118] Among them, I d *、I q * Represents the d-axis and q-axis components of the required current, respectively; K is a coefficient, expressed as follows:

[0119] K = s p0_q s p2_d +s p1_q s p0_d +s p2_q s p1_d

[0120] -s p1_q s p2_d -s p2_q s p0_d -s p0_q s p1_d .

[0121] Calculate the optimal vector u p_1 Suboptimal vector u p_2 After the interaction time of the zero vector u0, it is also determined whether t... p1 +t p2 +t0>T s If so, then control period T will be... s The value is equal to the inverter's switching cycle, and the operating time is reconstructed; the reconstruction formula is:

[0122]

[0123] To verify the technical effectiveness of this invention, the method was simulated on the Matlab / Simulink platform. The parameters of the motor are shown in Table 3, and the parameters of the speed loop PID are shown in Table 4.

[0124] The simulation conditions are as follows: the switching frequency is 10kHz, which means the sampling time of the actual system is 1e-5; the simulation step size is 1e-6, which means the accuracy calculated by the simulation platform, indicating that sampling is performed once every ten steps; the forgetting factor λ = 0.996.

[0125] Table 3 Motor Parameter Table

[0126] Table 4 PID Parameter Table

[0127] parameter P I D value 0.75 0.0001 0

[0128] Simulations compared the performance of DBQ3V-MPCC and the proposed method (FFRLS-DBQ3V-MPCC) when the motor's three parameters were 1.5 times their normal values, as well as the performance of the DBQ3V-MPCC algorithm when the parameters were at their normal values. The motor parameters were set to 1.5 times their normal values, the speed to 1000 rpm, and the load to 1.0 N·m.

[0129] Depend on Figure 5 It can be seen that when the parameters are 1.5 times the normal value, compared with DBQ3V-MPCC, the current harmonic distortion (THD) of FFRLS-DBQ3V-MPCC is reduced by 0.3% and the torque ripple rate (Tr) is reduced by 0.11%. However, compared with DBQ3V-MPCC under normal parameters, the THD of FFRLS-DBQ3V-MPCC is increased by 0.4% and the Tr is increased by 0.12%.

[0130] The results show that the FFRLS-DBQ3V-MPCC algorithm with abnormal parameters performs better than the DBQ3V-MPCC algorithm without parameter identification, but slightly worse than the DBQ3V-MPCC algorithm with normal parameters. Therefore, when parameters are abnormal, the FFRLS algorithm can accurately identify the parameters and apply them to DBQ3V-MPCC, reducing the total harmonic distortion of the three-phase current and thus reducing torque ripple.

[0131] This novel forgetting factor recursive least squares parameter identification-rapid calculation three-vector model predictive current control method, compared to the current PI closed-loop control in the traditional voltage vector control (FOC) algorithm, has the advantages of effectively reducing current harmonic distortion rate, improving torque stability, and effectively reducing the impact of parameter changes on the controller, thus improving the controller's parameter robustness. Furthermore, the method does not require repeated adjustment of controller parameters or SVPWM modulation of the generated voltage, and it allows for easier constraint of state variables in the system, making control simpler and faster.

[0132] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive, characterized in that: Used to control the motor-driven magnetic wheel in the chassis of an amphibious humanoid welding robot; Predictive control of the motor is performed using a deadbeat-free, rapid calculation three-vector model; the input to the deadbeat-free, rapid calculation three-vector model is the target current I*; the current parameter Pr(k) in the deadbeat-free, rapid calculation three-vector model is [RL]. s ψ f ] T The parameters are estimated in real time using a forgetting factor recursive least squares method parameter identification algorithm; where R is the stator resistance of the motor; L s For the stator inductance of the motor; ψ f For motor rotor flux linkage; The forgetting factor recursive least squares parameter identification algorithm is implemented based on the q-axis discretized mathematical model and the least squares recursive formula with forgetting factor. The q-axis discretization mathematical model is as follows: Where y(k)=I q (k); Input matrix to the system; η(k)=[β q1 β q2 β q3 β q4 ] T The parameter matrix to be identified; Among them, I q (k-1) is the q-axis current at the previous moment; I d (k) is the d-axis current at the current moment; u q (k) is the q-axis voltage at the current moment; ω e (k) is the current electric angular velocity of the rotor; β q1 β q2 β q3 β q4 These are the coefficients that change over time; The least squares recursive formula with forgetting factor is as follows: in, This is the estimated value at the current moment. P(k) is the estimated value at the previous time step; P(k-1) is the covariance matrix at the current time step; L(k) is the gain matrix; E(k) is the identity matrix. The forgetting factor recursive least squares parameter identification algorithm refers to: y(k) is measured; and the estimated value at the current time is calculated. Covariance matrix P(k) and gain matrix L(k); When the convergence value of the parameter matrix η(k) to be identified is calculated Then, calculate the motor parameters to be identified, Pr(k) = [RL]. s ψ f ] T : The relative error err of the motor identification parameters is calculated based on the identification parameters Pr(k-1) of the previous time step and the identification parameters Pr(k) of the current time step. The relative error err is used to determine whether the parameters have fully converged. If the parameters have converged completely, the current identified parameter Pr(k) is substituted into the three-vector model of the zero-difference beat calculation for predictive control; if the parameters have not converged, the current cycle is skipped, and the identified parameter of the previous cycle is substituted into the three-vector model of the zero-difference beat calculation for predictive control.

2. The chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to claim 1, characterized in that: In the q-axis discretized mathematical model, the coefficient β q1 β q2 β q3 β q4 With d-axis current L d q-axis current L q Stator resistance R, motor rotor flux linkage ψ f The mathematical relationship between the control algorithm period T and the control algorithm period T is as follows:

3. The chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to claim 1, characterized in that: The recursive least squares parameter identification algorithm for the forgetting factor first initializes the estimated value when k=1. Initialize the covariance value P(0) = αE(k); then start the forgetting factor recursive least squares parameter identification algorithm.

4. The chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to claim 1, characterized in that: The relative error err is: If err < threshold, it means the parameters have fully converged; otherwise, it means the parameters have not converged.

5. The chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to claim 1, characterized in that: The deadbeat-free rapid calculation three-vector model selects an optimal vector u from six effective voltage vectors Ui, i = [1,2,3,4,5,6] based on the target current I*. p_1 and a suboptimal vector u p_2 And calculate the optimal vector u p_1 and suboptimal vector u p_2 The duration of action; based on the optimal vector u p_1 and suboptimal vector u p_2 The duration of the voltage vector is determined by the inverter, which generates a corresponding voltage vector that acts on the motor.

6. The chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to claim 5, characterized in that: In the aforementioned deadbeat-free rapid calculation three-vector model, an optimal vector u is selected from six effective voltage vectors Ui, i = [1,2,3,4,5,6]. p_1 and a suboptimal vector u p_2 , refers to: First, based on the deadbeat control principle of dq-axis current, the optimal voltage vector u for the next moment is calculated from the full voltage vector range. p_dq (k+1): u p_dq (k+1)=AI dq (k)+BI dq * +C; Among them, u p_d (k+1), u p_q (k+1) represent the globally optimal voltage components along the d-axis and q-axis at the next time step, respectively; I d (k), I q (k) represents the d-axis and q-axis currents collected at the current moment; Idq* represents the target current; A, B, and C represent coefficients respectively; Then, the globally optimal voltage vector u p_dq (k+1) is transformed into the value u in the two-phase stationary coordinate system. p_αβ (k+1); Judgment value u p_αβ The sector where (k+1) is located; based on the value u p_αβ The sector containing (k+1) selects the optimal vector u. p_1 and suboptimal vector u p_2 .

7. The chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to claim 6, characterized in that: Substituting the estimated motor parameters into the error-free rapid calculation three-vector model means substituting the estimated motor parameters into the calculation formulas for coefficients A, B, and C; the calculation formulas for parameters A, B, and C are as follows: Where T is the control period.

8. The chassis control method for an amphibious humanoid welding robot based on magnetic wheel drive according to claim 1, characterized in that: The target current I* includes the d-axis component I of the target current. d * Target current q-axis component I q *; Set the target current d-axis component I d * = 0; The target current q-axis component I q The method to obtain * is: set the target speed Speed * The actual speed, collected and processed by the encoder, is input to the speed PI controller, and the target current q-axis component I is output. q *

Images