Disturbance compensated vector look-up table model predictive fault-tolerant control method for five-phase permanent magnet motor

Abstract

The application discloses a kind of disturbance compensation five-phase permanent magnet motor vector look-up table model predictive fault-tolerant control methods, belong to motor drive system fault-tolerant control field.This method is based on stator separation winding to realize full-speed rotor position estimation, constructs double plane decoupling model and disturbance observer to complete flux linkage observation and system disturbance estimation, solves optimal given vector by flux linkage sector identification look-up table, constructs the hierarchical cost function of weighted fusion of sliding mode robustness and traditional model prediction, compensates by embedding disturbance estimation value into current prediction model, solves optimal vector combination and duty cycle after PWM modulation to realize control.This method can adapt to various open-circuit faults, suppresses harmonic and torque ripple, reduces stator copper loss, solves the problems of traditional model prediction, poor fault-tolerant adaptation, weak robustness, greatly improves control accuracy and real-time performance.

Description

Technical Field

[0001] This invention relates to the field of fault-tolerant control of motor drive systems, and more particularly to a vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor with disturbance compensation. Background Technology

[0002] In recent years, multiphase permanent magnet motor drive systems have been widely used in high-reliability fields such as aerospace, new energy transportation, and deep-sea exploration due to their high power density and significant fault tolerance. As the core of the drive system, the motor's control performance directly affects the operational accuracy and safety stability of the entire equipment. Maintaining continuous and stable motor operation, especially during open-circuit faults, has become a key focus in the industry. Therefore, researching high-precision, robust, and fast-response fault-tolerant control technologies is of great significance for improving the service capability of motor systems in complex environments.

[0003] The control strategy for motor drive systems is currently dominated by model predictive control (MMCC), which predicts and optimizes future system states by establishing mathematical models. However, traditional MMCC still has significant shortcomings in practical applications. First, traditional methods require exhaustive calculations by traversing a large number of voltage vectors in each control cycle. As the number of motor phases increases, the computational burden on the control algorithm increases significantly, resulting in limited system real-time performance and difficulty in meeting the requirements of high-frequency operation. Second, traditional predictive models rely heavily on accurate motor physical parameters. However, in actual operation, load disturbances, parameter perturbations, and harmonic interference often lead to large deviations in predicted current, making the system's disturbance rejection capability and control robustness weak. In addition, the generation of current setpoints under fault conditions usually relies on complex real-time online calculations, which further consumes controller resources and makes it difficult to achieve fast and smooth switching between normal and fault states. While existing improvement schemes have alleviated the above problems to some extent, they still have limitations. For example, some compensation methods based on analytical models require extremely high model accuracy; once the stator resistance or inductance fluctuates, the compensation effect drops significantly. Other studies, while incorporating offline optimization, have failed to adequately integrate real-time disturbance observation, resulting in inconsistent control accuracy under non-ideal conditions. Furthermore, existing fault-tolerant adaptation schemes often lack specific optimizations for computational speed, leading to significant dynamic fluctuations in control algorithm switching after fault diagnosis.

[0004] Therefore, existing fault-tolerant control methods for motors still need further improvement in terms of balancing control accuracy, robustness, and real-time performance, in order to meet the dual requirements of modern industry for high precision and high reliability of motor drive systems. Summary of the Invention

[0005] The purpose of this invention is to overcome the aforementioned deficiencies in the prior art and provide a disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor. This method significantly reduces computational complexity and improves response speed by utilizing an optimal vector selection table and a current setpoint mapping database. Simultaneously, it enhances disturbance rejection capability and control robustness by using an observer to compensate for system disturbances in real time. Thus, without the need for complex online calculations and additional hardware, this method enables high-precision, low-energy-consumption, and fully adaptive robust operation of the five-phase permanent magnet motor under normal and various open-circuit fault conditions.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: Step S1: Collect the operating parameters of the five-phase permanent magnet motor, including the five-phase inverter switch status signal, stator phase current, real-time speed, load torque, stator phase voltage, and stator disconnect winding induced voltage. Step S2: Calculate the current characteristics and unbalance of each phase of the stator, fuse and normalize to obtain standard fault characteristics, build a database and train a random forest model, sample features in real time and input them into the model to determine the motor operating status. Step S3: Based on the induced voltage of the stator split winding, the initial rotor position of the motor is detected by square wave voltage injection and polarity correction, and the full-speed rotor position is estimated by combining the split winding flux calculation. Using the generalized Clark and Park transformation matrices under normal conditions and the corrected reduced-order Clark and Park transformation matrices under fault conditions, a decoupling model is constructed by multiplying the left side of the voltage and current model in the natural coordinate system by the transformation matrix. The input to the decoupling model is the stator phase current collected in step S1, and the direct and quadrature axis currents of the fundamental plane and the third harmonic plane are separated by combining the estimated rotor position. , , , and output , ; Step S4: Construct an observer by combining the optimal composite vector from the previous cycle with the output from step S3. , After being input into the observer, the corrected flux observation value and disturbance estimate value are obtained by back-deriving the corrected current observation value and then output. Step S5: Based on the switch state signal and stator phase voltage collected in step S1, obtain the basic voltage vector. Based on the magnitude of the voltage vector in the fundamental subspace and harmonic subspace, and combined with the principle of volt-second balance, adjust the action time of the basic voltage vector in the fundamental subspace corresponding to the switch state to construct the synthetic voltage vector in different states. Step S6: Divide the flux linkage sector using the synthesized voltage vector, and establish a given vector selection table by outputting signals from two two-stage comparators. Then, select the optimal vector by reading the given vector selection table. Step S7: Optimal current setpoint generation, with given torque... Compared with the real-time rotational speed acquired in step S1 The system inputs an offline-established optimal current command mapping database and outputs a fundamental plane direct-axis current command in real time based on the motor's operating status. and quadrature axis current given The The speed is calculated by the outer loop PI controller based on the deviation between the given speed and the real-time speed; Step S8: Construct a robust prediction cost function based on the sliding mode principle, and weight and fuse it with the traditional model prediction cost function to form a total cost function. Solve for the minimum value of the weighted and fused total cost function and calculate the duty cycle. Construct two sets of candidate vector combinations based on the optimal vector selected in step S6 and set constraints. Embed the total disturbance estimate output in step S4 into the current prediction model for disturbance compensation and calculate the predicted current. Substitute the optimal current setpoint generated in step S7 into the total cost function and take the partial derivative of the prediction current tracking error cost function to obtain the optimal vector combination and its final duty cycle. Step S9: Based on the final duty cycle calculation results, obtain the optimal synthetic vector and voltage vector action time. Use the triangular carrier signal to compare with the switching points of each optimal vector combination to generate the required PWM pulse signal and complete the motor control.

[0007] Furthermore, step S2 is specifically as follows: Step S21: Based on the stator phase current signals, extract the harmonic amplitude of each phase current and calculate the phase current imbalance; the phase current imbalance is calculated in the following way: First, calculate the average value of the five-phase current: The The effective value of the five-phase stator current is obtained by taking the square root of the average value of the squares of the five-phase stator currents collected over one electrical cycle. Next, calculate the phase current imbalance: ; Step S22: The harmonic amplitude of each phase current and the phase current imbalance are fused into a fault feature. Specifically, the five-phase stator current signal is subjected to Fourier transform to extract the characteristic harmonic amplitude of each phase, and the characteristic harmonic amplitude of all phases and the phase current imbalance are defined as a fused fault feature vector. After normalization of speed and load parameters, standard fault characteristic values ​​are obtained, specifically based on the current real-time speed of the motor. Load torque and the rated speed of the motor Rated load torque Using normalized coefficients The fused fault feature vector is normalized to obtain standard fault feature values; Step S23: Slide to collect standard fault feature values ​​corresponding to normal motor state, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault, construct a fault diagnosis database, and train a random forest fault diagnosis model based on the fault diagnosis database. Step S24: Execute steps S1 and S21 to S22 in real time to obtain real-time standard fault characteristic values, and obtain diagnostic variables through sliding window sampling; Step S25: Input the diagnostic variables into the trained random forest fault diagnosis model and output the motor fault diagnosis results. The fault diagnosis results include normal state, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault.

[0008] Furthermore, in step S3, the generalized Clark and Park transformation matrices in the normal state are respectively: coefficients of a generalized matrix Based on the constraint of constant amplitude, the spatial electrical angle spacing of the five-phase stator windings is... , The estimated rotor position angle of the motor; In step S3, the corrected reduced-order Clark and Park transformation matrices under the single-phase open-circuit fault state are as follows: In step S3, the corrected reduced-order Clark and Park transformation matrices for open-circuit fault states of adjacent two phases are as follows: In step S3, the corrected reduced-order Clark and Park transformation matrices for the non-adjacent two-phase open-circuit fault state are as follows: .

[0009] Furthermore, the specific implementation method for full-speed rotor position estimation in step S3 is as follows: The five-phase permanent magnet motor has a detection separation winding that is coaxially and independently wound with electrical isolation from the five-phase main power winding in the stator slot. In step S1, a high-frequency square wave voltage is injected into one phase separation winding of the five-phase motor, and the induced voltage of the other four phase separation windings is collected. Based on the magnitude of the induced voltage of each phase and the determination of the extreme value, the 360° electrical angle is divided into five angle intervals corresponding to the winding axis, that is, the maximum value corresponds to the nearest winding axis and the minimum value corresponds to the farthest winding axis, and the interval where the rotor magnetic shaft is located is initially identified. The inverter applies a polarity correction pulse to the stator winding of the motor, and the rotor magnetic pole polarity is corrected according to the amplitude change of the induced voltage of the separation winding, eliminating the 180° electrical angle error and obtaining the accurate initial rotor position in the static state. When the motor is running, the stator flux linkage measurement value is directly obtained by integrating the induced voltage of the isolated winding. The flux linkage measurement value is then projected onto the stationary coordinate system, and the real-time rotor position is obtained from the stationary position to the high speed range by solving the arctangent function.

[0010] Furthermore, the observer construction step in step S4 is as follows: Step S41: Construct a dual-model comparison between the real motor system and the observer system. The real motor system is the decoupled model constructed in step S3, and the observer system is a model system constructed with the same structure as the real system but using nominal parameters. Step S42: Extract the stator current of the real system in real time and decouple it through step S3 to obtain the real current. , The observer system synchronously calculates the observed current using nominal parameters. , ; Step S43: Calculate the current deviation vector: When the current deviation is not equal to 0, there is a disturbance in the real system; Step S44: The observer uses the current deviation as feedback to correct the observer current through the gain matrix L. The correction equation is: Let be the discrete system matrix of the observer. Here is the gain matrix. For constant terms, and for and The system state variables at time t. This is the optimal composite vector for the previous period. , These represent the direct-axis and quadrature-axis components of the corrected current observations, respectively. These represent the direct-axis and quadrature-axis components of the corrected flux linkage observations, respectively. , These are the estimated disturbance values ​​for the direct axis and the quadrature axis, respectively. , They represent , The time period refers to the current period and the previous period; the gain matrix The observer error system is obtained by setting the eigenvalues ​​of the observer error system at the desired position using the pole placement method. The observer error system is the difference system between the real motor system and the observer system. Step S45: Output the disturbance estimate, corrected current observation, and corrected flux linkage observation.

[0011] Furthermore, the gain matrix The specific solution is as follows: First, by subtracting the discrete state equation of the observer from the actual physical equation of the system, an error dynamics equation is constructed. That is, the observer error system, where This is a six-dimensional state error vector. for The output matrix is ​​used to extract current observations from state variables. This is achieved by designing the gain matrix. Make the characteristic matrix All eigenvalues ​​are located within the unit circle in the z-plane, ensuring that the observation error recursively converges to zero rapidly with the sampling period.

[0012] Secondly, the physical symmetry of the five-phase permanent magnet motor in the rotating coordinate system is utilized to... 1-th order matrix The internal structure is simplified by assuming complete decoupling of the observation loops of the d-axis and q-axis. This means the d-axis current residual only corrects its physically corresponding current, flux linkage, and disturbance states, and the gain coefficient remains equal to the corresponding term in the q-axis. This simplifies the original 12 unknown elements into a current-corrected gain. Magnetic flux correction gain and perturbation update gain Three sets of independent parameters significantly reduce the computational complexity of the subsequent characteristic equations.

[0013] Finally, the desired poles of the observer are preset according to the system's requirements for dynamic response and disturbance rejection capability. And based on the characteristic polynomial Expanding this yields a sixth-order algebraic equation for z. By comparing the coefficients of this polynomial with the target expectation polynomial... The corresponding coefficients are matched to establish a system of linear equations, which are then solved simultaneously to uniquely determine the core gain parameter. The specific value.

[0014] Furthermore, the construction of the synthesized voltage vector in step S5 is as follows: In normal conditions, a basic voltage vector is selected. The large and medium vectors, which have the same direction in the fundamental subspace, have opposite mapping directions in the third harmonic subspace. Utilizing the volt-second balance principle, within one control cycle... Internal command large vector action time Medium vector action time satisfy And make the equivalent voltage of the synthesized vector in the three-level harmonic subspace zero, thus obtaining ,according to Construct 10 composite vectors, the It is the median vector. For large vectors, combine two zero vectors to construct an extended control set to eliminate the third harmonic component; Single-phase open-circuit faults are selected with the constraint of minimizing copper loss. The fundamental voltage vector pairs with opposite projection directions in the subspace, or the self-selection method. For a vector whose projection is zero, a control period is set using the volt-second balance principle. The duration of action of the two fundamental vectors Let the composite vector be in The net voltage of the subspace is zero, and the synthesized current satisfies the third harmonic current. The constraints consist of six non-zero composite voltage vectors and two zero vectors, due to the fundamental voltage vector. and exist The mapping in the subspace is zero, therefore both can be directly used as virtual vectors in phase A fault mode. The duty cycles of the remaining four virtual vectors within one control cycle satisfy the formula. , It is the median vector. It is a large vector; After an open-circuit fault occurs between two adjacent phases or between two non-adjacent phases, the number of remaining healthy phases drops to 3, and the control degree of freedom is only 2, making it impossible to maintain the third harmonic space at the same time. At this time, vector synthesis is not performed, and the 6 non-zero basic voltage vectors that the inverter can currently output and their corresponding 2 zero vectors are directly used as the control set for subsequent model prediction fault-tolerant control.

[0015] Furthermore, the fundamental voltage vector is obtained by transforming the voltage vector corresponding to the switching state into the fundamental subspace through the Clark transformation of the corresponding mode. - Harmony wave space - In this process, a spatial voltage vector satisfying the equivalent constraint of the stator magnetomotive force is formed. The synthesized voltage vector is obtained by making... - The voltage vector in the subspace is calculated with a mapping of 0. The different states are as follows: Under normal conditions, the basic voltage vectors include 10 large vectors, 10 medium vectors, 10 small vectors, and 2 zero vectors, with corresponding amplitudes of 0.6472. 0.4 0.2472 , 0; the synthesized voltage vector includes phase angles of respectively The amplitude is 0.5527. 10 voltage vectors; During a single-phase open-circuit fault, the basic voltage vector includes two moduli of 0.6155. The vector consists of a large vector, six medium vectors, six small vectors, and two zero vectors; the medium vectors include two vectors with a magnitude of 0.4472. And two modules with a length of 0.4412 The vector, wherein the smaller vector comprises four vectors with a magnitude of 0.3425. And two modules with a length of 0.1453 The vector; the composite voltage vector includes a magnitude of 0.5527. The phase angles are respectively The six voltage vectors; During a two-phase open-circuit fault, the basic voltage vector includes four modules with a magnitude of 0.3914. The large vector, with two moduli of 0.1843. The middle vector and two zero vectors.

[0016] Furthermore, step S6 is specifically as follows: Step S61: Identify the flux linkage sector. The boundary of each sector is the angle bisector of the angle between adjacent composite voltage vectors. In the case of a two-phase open-circuit fault, it is the angle bisector of the angle between adjacent basic voltage vectors. Under normal conditions, 10 sectors are divided; under single-phase fault, 6 sectors are divided; under two-phase fault, 6 sectors are divided; under non-adjacent two-phase fault, 6 sectors are divided. Step S62: Offline, for each flux sector under each operating state, determine the effect of each voltage vector on the increase or decrease of flux amplitude and electromagnetic torque, and establish a given vector selection table that corresponds one-to-one with the flux sector, flux deviation binary signal, torque deviation binary signal and the optimal voltage vector. Step S63: Calculate the actual electromagnetic torque using the corrected flux linkage observation value and corrected current observation value output in step S4. Compare the actual flux linkage deviation and torque deviation with the given flux linkage value and torque value. Input the flux linkage deviation and torque deviation into the hysteresis comparator to complete the binarization process. Output the flux linkage deviation binary signal and the torque deviation binary signal, which are only 1 and -1 respectively. Step S64: When the motor is running online, the optimal voltage vector is selected by looking up the corresponding indirect given voltage vector in the table based on the current flux sector, flux deviation binary signal, and torque deviation binary signal. ; Step S65: Select After that, Two adjacent vectors are selected as the candidate optimal voltage vectors, denoted as... and Determine candidate vector combinations.

[0017] Furthermore, the binarization processing of flux deviation and torque deviation in step S63 specifically involves: The flux linkage deviation is obtained by subtracting the stator flux linkage setpoint from the corrected flux linkage observed value. When the corrected flux linkage observed value is less than the setpoint, the air gap magnetic field is insufficient. A value of 1 is used to increase magnetization; when the corrected flux linkage observation is greater than or equal to the given value, A value of -1 is used to reduce magnetization; the torque deviation is obtained by subtracting the given electromagnetic torque value from the actual value. When the actual value of the electromagnetic torque is less than the given value, A value of 1 is set to increase torque; when the actual electromagnetic torque is greater than or equal to the given value, A value of -1 is used to reduce torque.

[0018] Furthermore, the optimal current given mapping database in step S7 is constructed offline, and the specific construction steps are as follows: Step A1: Solving for the current in the constant torque operating region. When the motor speed is lower than the base speed, the electromagnetic torque equation is used. The stator current modulus is obtained by solving the Lagrange multiplier method. The minimum current vector trajectory is used to obtain the maximum torque-current ratio solution set under different torque requirements. This represents the number of pole pairs of the motor. , These represent the inductance components of the motor in the direct and quadrature axes of the synchronous rotating coordinate system, respectively. is the flux linkage constant. , These are the fundamental plane direct-axis and quadrature-axis currents, respectively; the fundamental speed is the maximum speed at which the motor outputs its rated torque under rated voltage and rated magnetic flux. Step A2: Solve for the current in the constant power operating region. When the motor speed is higher than the base speed, verify the voltage limit constraint equation in real time. By injecting a negative direct-axis current component to cancel the magnetic flux of the permanent magnet, the field weakening control solution set under different speed and torque requirements is solved within the dual constraints of current and voltage. , These are the fundamental plane direct-axis and quadrature-axis voltages, respectively. This represents the voltage limit of the inverter under the current DC bus voltage constraint. This refers to the maximum current amplitude that the inverter is allowed to pass through. Step A3: Grid-based discrete storage. Set the sampling step size for the torque axis and speed axis. Fill the two-dimensional data array with the maximum torque-current ratio solution set and the field weakening control solution set obtained above according to the state coordinate points to complete the establishment of the optimal current given mapping database.

[0019] Furthermore, the specific steps in step S8 are as follows: Step S81: Construct two sets of candidate vector combinations, combination 1 is , , Combination 2 is , , , It is a zero vector; Step S82: Set constraints for each candidate vector combination: one control cycle. Only the combination of the three inner vectors satisfies And since the duration of action is non-negative, the phase duty cycle is defined. The auxiliary vector duty cycle is Zero vector duty cycle ,in for Duration of action For the duration of action of the auxiliary voltage vector, The zero vector action time; Step S83: Based on the dual-plane decoupling model of the five-phase permanent magnet motor, according to the vector in the fundamental wave space of each combination... The voltage components are combined by weighting the voltage components according to the duty cycle to form the total voltage components; Step S84: Substitute the combined total voltage component into the current prediction model with disturbance compensation obtained through first-order Euler discretization to obtain the predicted current value for the next cycle. , ; Step S85: With the objectives of minimizing the tracking error of the actual current to the optimal current setpoint and ensuring strong system robustness and stability, construct a total cost function. Then, analyze the total cost function... , Find the partial derivative and set it to 0 to obtain the duty cycle for each combination. , and the corresponding total cost function value; Step S86: Compare the total cost function values ​​of the two combinations and select the combination with the smaller total cost function value as the optimal vector combination. , , Its corresponding , To achieve the optimal duty cycle, and simultaneously synthesize the optimal output voltage. .

[0020] Furthermore, the total cost function in step S8 is specifically as follows: Among them, sliding mode robustness item It is used to eliminate the impact of motor parameter mismatch and fault disturbance on control performance, and drive the system to converge quickly to a stable sliding surface; These are dimensionless weighted coefficients. For current error tracking, where , , , They are respectively The direct-axis and quadrature-axis components of the fundamental plane total voltage at time intervals. , They are respectively The fundamental plane direct-axis and quadrature-axis stator currents at any given time.

[0021] Furthermore, the current error tracking term is specifically as follows: Under normal operating conditions, the cost function only includes the fundamental subspace current tracking error term, and its expression is: Under single-phase open-circuit fault conditions, the cost function adds a stator copper loss minimum constraint term to the current tracking error term, and the expression is: in, This is the copper loss constraint weighting coefficient. For stator copper loss, , Stator phase resistance, , , , This refers to the remaining healthy phase stator current after the fault. Under the two-phase open-circuit fault condition, the cost function, based on the current tracking error term and the stator copper loss minimum constraint term, is supplemented with a torque ripple suppression term, and the expression is: in, This is the copper loss constraint weighting coefficient. This is the torque ripple suppression weighting coefficient. The stator copper loss of the remaining healthy phase after a two-phase failure. This represents the electromagnetic torque ripple value. , This is the predicted value of the electromagnetic torque at the next moment.

[0022] Furthermore, the current prediction model is specifically as follows: Under normal conditions, In the case of a single-phase fault, When two adjacent phases are in a fault state, When there is a fault in two non-adjacent phases, .

[0023] Compared with the prior art, the present invention, employing the above technical solution, has the following beneficial effects: (1) The disturbance-compensated five-phase permanent magnet motor vector lookup table model prediction fault-tolerant control method provided by the present invention adopts a random forest model and integrates normalized fault features for state determination, effectively eliminating the interference of speed and load fluctuations on the diagnostic results, realizing rapid fault identification and seamless switching of control strategies, and avoiding the dynamic performance degradation problem that occurs in the fault switching process of the prior art.

[0024] (2) The disturbance-compensated five-phase permanent magnet motor vector lookup table model prediction fault-tolerant control method provided by the present invention transforms the complex iterative optimization process in traditional model prediction control into a simple lookup table operation by constructing the optimal vector setpoint selection table offline, which greatly reduces the computational burden of the controller and solves the defects of large computational load and poor real-time performance of traditional methods.

[0025] (3) The disturbance-compensated five-phase permanent magnet motor vector lookup table model prediction fault-tolerant control method provided by the present invention introduces a disturbance observer to monitor the system state in real time, and actively compensates for load disturbances and model parameter perturbations by using current deviation feedback. This effectively solves the problem of excessive reliance on precise mathematical models in the prior art and significantly improves the control robustness of the system under complex conditions.

[0026] (4) The disturbance compensation five-phase permanent magnet motor vector lookup table model prediction fault-tolerant control method provided by the present invention establishes an offline optimal current given mapping database, and directly matches the output fundamental plane current given according to the real-time speed and given torque, avoiding complex online current reconstruction calculation under fault conditions, and ensuring the speed and accuracy of given value generation.

[0027] (5) The disturbance-compensated five-phase permanent magnet motor vector lookup table model prediction fault-tolerant control method provided by the present invention designs a hierarchical cost function for different operating states such as normal, single-phase open circuit and two-phase open circuit. While ensuring the current tracking accuracy, it also takes into account the minimization of stator copper loss and the suppression of torque ripple, thus overcoming the limitation of existing fault-tolerant strategies that are difficult to balance multiple performance indicators.

[0028] (6) The disturbance-compensated five-phase permanent magnet motor vector lookup table model prediction fault-tolerant control method provided by the present invention realizes the rotor position estimation from stationary to high speed based on the induced voltage of the separated winding. It can complete the initial position detection and real-time position calculation without the need for a position sensor, effectively solving the problem of positioning failure in the fault state of traditional sensorless control, and significantly improving the full-condition adaptability of rotor position detection. Attached Figure Description

[0029] Figure 1 This is a flowchart of a disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to the present invention. Figure 2 This is a control block diagram of a disturbance-compensated vector lookup table model predictive fault-tolerant control method for a five-phase permanent magnet motor according to the present invention. Figure 3 This is a block diagram illustrating the principle of the observer in this invention. Figure 4 This is a simulation waveform of single-phase fault and fault-tolerant current in this invention; Figure 5 The simulation waveforms for adjacent two-phase faults and fault-tolerant currents in this invention are shown. Figure 6 The simulation waveforms for non-adjacent two-phase faults and fault-tolerant currents in this invention are shown. Detailed Implementation

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0031] Example: like Figure 1 The diagram shown is a flowchart of the disturbance compensation method for a five-phase permanent magnet motor vector lookup table model prediction fault-tolerant control according to the present invention. The control block diagram is as follows. Figure 2 As shown, the specific steps include: Step S1: Collect the operating parameters of the five-phase permanent magnet motor, including the five-phase inverter switch status signal, stator phase current, real-time speed, load torque, stator phase voltage, and stator disconnect winding induced voltage. Step S2: Calculate the current characteristics and unbalance of each phase of the stator, fuse and normalize to obtain standard fault characteristics, build a database and train a random forest model, sample features in real time and input them into the model to determine the motor operating status. Step S3: Based on the induced voltage of the stator split winding, the initial rotor position of the motor is detected by square wave voltage injection and polarity correction, and the full-speed rotor position is estimated by combining the split winding flux calculation. Using the generalized Clark and Park transformation matrices under normal conditions and the corrected reduced-order Clark and Park transformation matrices under fault conditions, a decoupling model is constructed by multiplying the left side of the voltage and current model in the natural coordinate system by the transformation matrix. The input to the decoupling model is the stator phase current collected in step S1, and the direct and quadrature axis currents of the fundamental plane and the third harmonic plane are separated by combining the estimated rotor position. , , , and output , ; Step S4: Construct an observer by combining the optimal composite vector from the previous cycle with the output from step S3. , After being input into the observer, the corrected flux observation value and disturbance estimate value are obtained by back-deriving the corrected current observation value and then output. Step S5: Based on the switch state signal and stator phase voltage collected in step S1, obtain the basic voltage vector. Based on the magnitude of the voltage vector in the fundamental subspace and harmonic subspace, and combined with the principle of volt-second balance, adjust the action time of the basic voltage vector in the fundamental subspace corresponding to the switch state to construct the synthetic voltage vector in different states. Step S6: Divide the flux linkage sector using the synthesized voltage vector, and establish a given vector selection table by outputting signals from two two-stage comparators. Then, select the optimal vector by reading the given vector selection table. Step S7: Optimal current setpoint generation, with given torque... Compared with the real-time rotational speed acquired in step S1 The system inputs an offline-established optimal current command mapping database and outputs a fundamental plane direct-axis current command in real time based on the motor's operating status. and quadrature axis current given The The speed is calculated by the outer loop PI controller based on the deviation between the given speed and the real-time speed; Step S8: Construct a robust prediction cost function based on the sliding mode principle, and weight and fuse it with the traditional model prediction cost function to form a total cost function. Solve for the minimum value of the weighted and fused total cost function and calculate the duty cycle. Construct two sets of candidate vector combinations based on the optimal vector selected in step S6 and set constraints. Embed the total disturbance estimate output in step S4 into the current prediction model for disturbance compensation and calculate the predicted current. Substitute the optimal current setpoint generated in step S7 into the total cost function and take the partial derivative of the prediction current tracking error cost function to obtain the optimal vector combination and its final duty cycle. Step S9: Based on the final duty cycle calculation results, obtain the optimal synthetic vector and voltage vector action time. Use the triangular carrier signal to compare with the switching points of each optimal vector combination to generate the required PWM pulse signal and complete the motor control.

[0032] Furthermore, step S2 is specifically as follows: Step S21: Based on the stator phase current signals, extract the harmonic amplitude of each phase current and calculate the phase current imbalance; the phase current imbalance is calculated in the following way: First, calculate the average value of the five-phase current: The The effective value of the five-phase stator current is obtained by taking the square root of the average value of the squares of the five-phase stator currents collected over one electrical cycle. Next, calculate the phase current imbalance: ; Step S22: The harmonic amplitude of each phase current and the phase current imbalance are fused into a fault feature. Specifically, the five-phase stator current signal is subjected to Fourier transform to extract the characteristic harmonic amplitude of each phase, and the characteristic harmonic amplitude of all phases and the phase current imbalance are defined as a fused fault feature vector. After normalization of speed and load parameters, standard fault characteristic values ​​are obtained, specifically based on the current real-time speed of the motor. Load torque and the rated speed of the motor Rated load torque Using normalized coefficients The fused fault feature vector is normalized to obtain standard fault feature values; Step S23: Slide to collect standard fault feature values ​​corresponding to normal motor state, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault, construct a fault diagnosis database, and train a random forest fault diagnosis model based on the fault diagnosis database. Step S24: Execute steps S1 and S21 to S22 in real time to obtain real-time standard fault characteristic values, and obtain diagnostic variables through sliding window sampling; Step S25: Input the diagnostic variables into the trained random forest fault diagnosis model and output the motor fault diagnosis results. The fault diagnosis results include normal state, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault.

[0033] Furthermore, the specific implementation method for full-speed rotor position estimation in step S3 is as follows: The five-phase permanent magnet motor has a detection separation winding that is coaxially and independently wound with electrical isolation from the five-phase main power winding in the stator slot. In step S1, a high-frequency square wave voltage is injected into one phase separation winding of the five-phase motor, and the induced voltage of the other four phase separation windings is collected. Based on the magnitude of the induced voltage of each phase and the determination of the extreme value, the 360° electrical angle is divided into five angle intervals corresponding to the winding axis, that is, the maximum value corresponds to the nearest winding axis and the minimum value corresponds to the farthest winding axis, and the interval where the rotor magnetic shaft is located is initially identified. The inverter applies a polarity correction pulse to the stator winding of the motor, and the rotor magnetic pole polarity is corrected according to the amplitude change of the induced voltage of the separation winding, eliminating the 180° electrical angle error and obtaining the accurate initial rotor position in the static state. When the motor is running, the stator flux linkage measurement value is directly obtained by integrating the induced voltage of the isolated winding. The flux linkage measurement value is then projected onto the stationary coordinate system, and the real-time rotor position is obtained from the stationary position to the high speed range by solving the arctangent function.

[0034] Furthermore, in step S3, the generalized Clark and Park transformation matrices in the normal state are respectively: coefficients of a generalized matrix Based on the constraint of constant amplitude, the spatial electrical angle spacing of the five-phase stator windings is... , This refers to the rotor position angle of the motor; In step S3, the corrected reduced-order Clark and Park transformation matrices under the single-phase open-circuit fault state are as follows: In step S3, the corrected reduced-order Clark and Park transformation matrices for open-circuit fault states of adjacent two phases are as follows: In step S3, the corrected reduced-order Clark and Park transformation matrices for the non-adjacent two-phase open-circuit fault state are as follows: .

[0035] Furthermore, the observer construction step in step S4 is as follows: Step S41, as follows Figure 3 As shown, a dual-model comparison is constructed between the real motor system and the observer system. The real motor system is the decoupled model constructed in step S3, and the observer system is a model system constructed with the same structure as the real system but using nominal parameters. Step S42: Extract the stator current of the real system in real time and decouple it through step S3 to obtain the real current. , The observer system synchronously calculates the observed current using nominal parameters. , The specific calculation process is as follows: Under normal conditions, When a single-phase open circuit fault occurs, When there is an open circuit fault in two adjacent phases, When there is an open circuit fault in two adjacent phases, in, , They represent and time, , These are the model's nominal resistance and the motor's electrical angular velocity, respectively. , , , These are the fundamental inductance, and the effective fundamental inductance when phases A, AB, and AC are open-circuited, respectively. These are the fundamental flux linkage amplitudes, To control the cycle, , These are the fundamental component and the third harmonic component of the optimal composite vector, respectively. , These are the direct-axis and quadrature-axis components of the initial observed current, respectively. Step S43: Calculate the current deviation vector: When the current deviation is not equal to 0, there is a disturbance in the real system; Step S44: The observer uses the current deviation as feedback to correct the observer current through the gain matrix L. The correction equation is: Let be the discrete system matrix of the observer. Here is the gain matrix. For constant terms, and for and The system state variables at time t. This is the optimal composite vector for the previous period. , These represent the direct-axis and quadrature-axis components of the corrected current observations, respectively. These are the direct-axis and quadrature-axis components of the corrected flux linkage observations, respectively. , These are the estimated disturbance values ​​for the direct axis and the quadrature axis, respectively. , They represent , The time period refers to the current period and the previous period; the gain matrix The observer error system is obtained by setting the eigenvalues ​​of the observer error system at the desired position using the pole placement method. The observer error system is the difference system between the real motor system and the observer system. Step S45: Output the disturbance estimate, corrected current observation, and corrected flux linkage observation.

[0036] Furthermore, the discrete system matrix of the observer Gain matrix constant term The normal states are as follows: When the fault is a single phase, the fault conditions are as follows: The fault conditions for two adjacent phases are as follows: The fault conditions for non-adjacent two phases are as follows: Furthermore, the gain matrix The specific solution is as follows: First, by subtracting the discrete state equation of the observer from the actual physical equation of the system, an error dynamics equation is constructed. ,in This is a six-dimensional state error vector. for The output matrix is ​​used to extract current observations from state variables. This is achieved by designing the gain matrix. Make the characteristic matrix All eigenvalues ​​lie within the unit circle in the z-plane, ensuring that the observation error recursively converges to zero rapidly with the sampling period.

[0037] Secondly, utilizing the physical symmetry of the five-phase permanent magnet motor in the rotating coordinate system... 1-th order matrix The internal structure is simplified by assuming complete decoupling of the observation loops of the d-axis and q-axis. This means the d-axis current residual only corrects its physically corresponding current, flux linkage, and disturbance states, and the gain coefficient remains equal to the corresponding term in the q-axis. This simplifies the original 12 unknown elements into a current-corrected gain. Magnetic flux correction gain and perturbation update gain Three sets of independent parameters significantly reduce the computational complexity of the subsequent characteristic equations.

[0038] Finally, the desired poles of the observer are preset according to the system's requirements for dynamic response and disturbance rejection capability. And based on the characteristic polynomial Expanding this yields a sixth-order algebraic equation for z. By comparing the coefficients of this polynomial with the target expectation polynomial... The corresponding coefficients are matched to establish a system of linear equations, which are then solved simultaneously to uniquely determine the core gain parameter. The specific value.

[0039] Furthermore, the calculated discrete gain vector is injected into the real-time algorithm of the observer. The flux linkage gain term is used to force the correction of the flux linkage observation value through the closed-loop feedback force generated by the current residual when parameters such as stator resistance fluctuate. This eliminates the reliance on open-loop integral calculations. Simultaneously, the disturbance gain term captures residual voltage deviations caused by inaccuracies in the physical model and updates them to the disturbance position, ultimately achieving high-precision observation of the motor's state under various complex conditions and parameter perturbations.

[0040] Furthermore, the construction of the synthesized voltage vector in step S5 is as follows: In normal conditions, a basic voltage vector is selected. The large and medium vectors, which have the same direction in the fundamental subspace, have opposite mapping directions in the third harmonic subspace. Utilizing the volt-second balance principle, within one control cycle... Internal command large vector action time Medium vector action time satisfy And make the equivalent voltage of the synthesized vector in the three-level harmonic subspace zero, thus obtaining ,according to Construct 10 composite vectors, the It is the median vector. For large vectors, combine two zero vectors to construct an extended control set to eliminate the third harmonic component; Single-phase open-circuit faults are selected with the constraint of minimizing copper loss. The fundamental voltage vector pairs with opposite projection directions in the subspace, or the self-selection method. For a vector whose projection is zero, a control period is set using the volt-second balance principle. The duration of action of the two fundamental vectors Let the composite vector be in The net voltage of the subspace is zero, and the synthesized current satisfies the third harmonic current. The constraints consist of six non-zero composite voltage vectors and two zero vectors, due to the fundamental voltage vector. and exist The mapping in the subspace is zero, therefore both can be directly used as virtual vectors in phase A fault mode. The duty cycles of the remaining four virtual vectors within one control cycle satisfy the formula. , It is the median vector. It is a large vector; After an open-circuit fault occurs between two adjacent phases or between two non-adjacent phases, the number of remaining healthy phases drops to 3, and the control degree of freedom is only 2, making it impossible to maintain the third harmonic space at the same time. At this time, vector synthesis is not performed, and the 6 non-zero basic voltage vectors that the inverter can currently output and their corresponding 2 zero vectors are directly used as the control set for subsequent model prediction fault-tolerant control.

[0041] Furthermore, the fundamental voltage vector is obtained by transforming the voltage vector corresponding to the switching state into the fundamental subspace through the Clark transformation of the corresponding mode. - Harmony wave space - In this process, a spatial voltage vector satisfying the equivalent constraint of the stator magnetomotive force is formed. The synthesized voltage vector is obtained by making... - The voltage vector in the subspace is calculated with a mapping of 0. The different states are as follows: Under normal conditions, the basic voltage vectors include 10 large vectors, 10 medium vectors, 10 small vectors, and 2 zero vectors, with corresponding amplitudes of 0.6472. 0.4 0.2472 , 0; the synthesized voltage vector includes phase angles of respectively The amplitude is 0.5527. 10 voltage vectors; During a single-phase open-circuit fault, the basic voltage vector includes two moduli of 0.6155. The vector consists of a large vector, six medium vectors, six small vectors with magnitudes of 0.4472, and two zero vectors; the medium vectors include two with a magnitude of 0.4472. And two modules with a length of 0.4412 The vector, wherein the smaller vector comprises four vectors with a magnitude of 0.3425. And two modules with a length of 0.1453 The vector; the composite voltage vector includes a magnitude of 0.5527. The phase angles are respectively The six voltage vectors; During a two-phase open-circuit fault, the basic voltage vector includes four modules with a magnitude of 0.3914. The large vector, with two moduli of 0.1843. The middle vector and two zero vectors.

[0042] Furthermore, step S6 is specifically as follows: Step S61: Identify the flux linkage sector. The boundary of each sector is the angle bisector of the angle between adjacent combined voltage vectors. In the case of a two-phase open circuit fault, it is the angle bisector of the angle between adjacent voltage vectors. Under normal conditions, 10 sectors are divided; under single-phase fault, 6 sectors are divided; under two-phase fault, 6 sectors are divided; under non-adjacent two-phase fault, 6 sectors are divided. Step S62: Offline, for each flux sector under each operating state, determine the effect of each voltage vector on the increase or decrease of flux amplitude and electromagnetic torque, and establish a given vector selection table that corresponds one-to-one with the flux sector, flux deviation binary signal, torque deviation binary signal and the optimal voltage vector. Step S63: Calculate the electromagnetic torque using the corrected flux linkage observation value and corrected current observation value output in step S4. Compare the electromagnetic flux linkage deviation and torque deviation with the given flux linkage value and torque value. Input the flux linkage deviation and torque deviation into the hysteresis comparator to complete the binarization process. The output is only a +1 and a -1 binary signal of flux linkage deviation and a binary signal of torque deviation. Step S64: Based on the current flux linkage sector, flux linkage deviation binary signal, and torque deviation binary signal, select the corresponding indirect given voltage vector from the table; this is the optimal voltage vector. ; Step S65: Select After that, Two adjacent vectors are selected as the candidate optimal voltage vectors, denoted as... and Determine candidate vector combinations.

[0043] Furthermore, the formula for calculating electromagnetic torque is: in, , The fundamental wave plane is respectively , Axial electromagnetic components, , The fundamental wave plane is respectively , Axis current components; Furthermore, the binarization processing of flux deviation and torque deviation in step S63 specifically involves: The flux linkage deviation is obtained by subtracting the stator flux linkage setpoint from the corrected flux linkage observed value. When the corrected flux linkage observed value is less than the setpoint, the air gap magnetic field is insufficient. A value of 1 is used to increase magnetization; when the corrected flux linkage observation is greater than or equal to the given value, A value of -1 is used to reduce magnetization; the torque deviation is obtained by subtracting the given electromagnetic torque value from the actual value. When the actual value of the electromagnetic torque is less than the given value, A value of 1 is set to increase torque; when the actual electromagnetic torque is greater than or equal to the given value, A value of -1 is used to reduce torque.

[0044] Furthermore, the given vector selection table established in step S62 is specifically as follows: Under normal conditions, as shown in Table 1, Table 1. Normal State Given Vector Selection Table In the case of a single-phase fault, such as a phase A fault, as shown in Table 2, Table 2. Voltage vector selection results under single-phase fault-tolerant conditions. When two adjacent phases are in a fault state, such as when phases AB are in a fault state, as shown in Table 3. Table 3 Vector Selection Table for Two-Phase AB Fault For non-adjacent two-phase fault conditions, such as AC phase fault conditions, as shown in Table 4. Table 4. Selection of Given Vector under AC Two-Phase Fault Furthermore, the optimal current given mapping database in step S7 is constructed offline, and the specific construction steps are as follows: Step A1: Solving for the current in the constant torque operating region. When the motor speed is lower than the base speed, the electromagnetic torque equation is used. Solve for the stator current mode length The minimum current vector trajectory is used to obtain the maximum torque-current ratio solution set under different torque requirements. This represents the number of pole pairs of the motor. , These represent the inductance components of the motor in the direct and quadrature axes of the synchronous rotating coordinate system, respectively. Let be the flux linkage constant. , These are the fundamental plane direct-axis and quadrature-axis currents, respectively. Step A2: Solve for the current in the constant power operating region. When the motor speed is higher than the base speed, verify the voltage limit constraint equation in real time. By injecting a negative direct-axis current component to cancel the magnetic flux of the permanent magnet, the field weakening control solution set under different speed and torque requirements is solved within the dual constraints of current and voltage. , These are the fundamental plane direct-axis and quadrature-axis voltages, respectively. This represents the voltage limit of the inverter under the current DC bus voltage constraint. This is the maximum current amplitude that the inverter is allowed to pass through; Step A3: Grid-based discrete storage. Set the sampling step size for the torque axis and speed axis. Fill the two-dimensional data array with the maximum torque-current ratio solution set and the field weakening control solution set obtained above according to the state coordinate points to complete the establishment of the optimal current given mapping database.

[0045] Furthermore, the specific steps in step S8 are as follows: Step S81: Construct two sets of candidate vector combinations, combination 1 is , , Combination 2 is , , , It is a zero vector; Step S82: Set constraints for each candidate vector combination: one control cycle. Only the combination of the three inner vectors acts, satisfying And since the duration of action is non-negative, the phase duty cycle is defined. The auxiliary vector duty cycle is Zero vector duty cycle ,in for Duration of action For the duration of action of the auxiliary voltage vector, The zero vector action time; Step S83: Based on the dual-plane decoupling model of the five-phase permanent magnet motor, according to the vector in the fundamental wave space of each combination... The voltage components are combined by weighting the voltage components according to the duty cycle to form the total voltage components; Step S84: Substitute the combined total voltage component into the current prediction model with disturbance compensation obtained through first-order Euler discretization to obtain the predicted current value for the next cycle. , ; Step S85: With the objectives of minimizing the tracking error of the actual current to the optimal current setpoint and ensuring strong system robustness and stability, construct a total cost function. Then, analyze the total cost function... , Find the partial derivative and set it to 0 to obtain the duty cycle for each combination. , and the corresponding total cost function value; Step S86: Compare the total cost function values ​​of the two combinations and select the combination with the smaller total cost function value as the optimal vector combination. , , Its corresponding , To achieve the optimal duty cycle, and simultaneously synthesize the optimal output voltage. .

[0046] Furthermore, the total cost function in step S8 is specifically as follows: Among them, sliding mode robustness item It is used to eliminate the impact of motor parameter mismatch and fault disturbance on control performance, and drive the system to converge quickly to a stable sliding surface; These are dimensionless weighted coefficients. For current error tracking, where , , , They are respectively The direct-axis and quadrature-axis components of the fundamental plane total voltage at time intervals. , They are respectively The fundamental plane direct-axis and quadrature-axis stator currents at any given time.

[0047] Furthermore, the current error tracking term is specifically as follows: Under normal operating conditions, the cost function only includes the fundamental subspace current tracking error term, and its expression is: Under single-phase open-circuit fault conditions, the cost function adds a stator copper loss minimum constraint term to the current tracking error term, and the expression is: in, This is the copper loss constraint weighting coefficient. For stator copper loss, , Stator phase resistance, , , , This refers to the remaining healthy phase stator current after the fault. Under the two-phase open-circuit fault condition, the cost function, based on the current tracking error term and the stator copper loss minimum constraint term, is supplemented with a torque ripple suppression term, and the expression is: in, This is the copper loss constraint weighting coefficient. This is the torque ripple suppression weighting coefficient. The stator copper loss of the remaining healthy phase after a two-phase failure. This represents the electromagnetic torque ripple value. , This is the predicted value of the electromagnetic torque at the next moment.

[0048] Furthermore, the current prediction model is specifically as follows: Under normal conditions, In the case of a single-phase fault, When two adjacent phases are in a fault state, When there is a fault in two non-adjacent phases, A simulation model is built using MATLAB / Simulink simulation software, such as Figure 4The figure shows the simulation results of the phase current signal switching between normal, fault, and fault-tolerant modes under an open-circuit fault in phase A. The simulation operation procedure is as follows: the motor first runs in healthy mode for 0.5s, then the trigger signal of the fault diagnosis module is set to high level to disconnect the phase A winding. After the motor runs with the fault for 0.5s, the model predictive fault-tolerant control algorithm is activated to realize the fault-tolerant operation of the motor. Figure 5 and Figure 6 The simulation results are for normal-fault-tolerance switching of phases AB and AC, including the motor's phase current signal, electromagnetic torque signal, and speed signal. The motor speed setpoint is always 500 rpm in all three modes. The motor initially operates in healthy mode, then an open-circuit fault is introduced in phases AB and AC. The improved model prediction algorithm for compensating for the disturbance is activated 100 ms after the fault occurs.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A disturbance-compensated vector lookup table model predictive fault-tolerant control method for a five-phase permanent magnet motor, characterized in that, Includes the following steps: Step S1: Collect the operating parameters of the five-phase permanent magnet motor, including the five-phase inverter switch status signal, stator phase current, real-time speed, load torque, stator phase voltage, and stator disconnect winding induced voltage. Step S2: Based on the stator current signals of each phase, extract the harmonic amplitude of each phase current, calculate the current characteristics and unbalance of each phase current of the stator, and fuse and normalize to obtain the standard fault characteristics. The standard fault feature values ​​corresponding to the normal state of the motor, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault are collected by sliding acquisition to construct a fault diagnosis database. A random forest fault diagnosis model is trained based on the fault diagnosis database. After real-time sampling of features, the model is input to determine the motor operating state. The motor operating state results include normal state, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault. Step S3: Based on the induced voltage of the stator split winding, the initial rotor position of the motor is detected by square wave voltage injection and polarity correction, and the full-speed rotor position is estimated by combining the split winding flux calculation. Using the generalized Clark and Park transformation matrices under normal conditions and the corrected reduced-order Clark and Park transformation matrices under fault conditions, a decoupling model is constructed by multiplying the left side of the voltage and current model in the natural coordinate system by the transformation matrix. The input to the decoupling model is the stator phase current collected in step S1, and the direct and quadrature axis currents of the fundamental plane and the third harmonic plane are separated by combining the estimated rotor position. , , , and output , ; Step S4: Construct an observer by combining the optimal composite vector from the previous cycle with the output from step S3. , After being input into the observer, the corrected flux observation value and disturbance estimate value are obtained by back-deriving the corrected current observation value and then output. Step S5: Based on the switch state signal and stator phase voltage collected in step S1, obtain the basic voltage vector. Based on the magnitude of the voltage vector in the fundamental subspace and harmonic subspace, and combined with the principle of volt-second balance, adjust the action time of the basic voltage vector in the fundamental subspace corresponding to the switch state to construct the synthetic voltage vector in different states. Step S6: Divide the flux linkage sector using the synthesized voltage vector, and establish a given vector selection table by outputting signals from two two-stage comparators. Then, select the optimal vector by reading the given vector selection table. Step S7: Optimal current setpoint generation, with given torque... Compared with the real-time rotational speed acquired in step S1 The system inputs an offline-established optimal current command mapping database and outputs a fundamental plane direct-axis current command in real time based on the motor's operating status. and quadrature axis current given The The speed is calculated by the outer loop PI controller based on the deviation between the given speed and the real-time speed; Step S8: Construct a robust prediction cost function based on the sliding mode principle, and weight and fuse it with the traditional model prediction cost function to form a total cost function. Solve for the minimum value of the weighted and fused total cost function and calculate the duty cycle. Construct two sets of candidate vector combinations based on the optimal vector selected in step S6 and set constraints. Embed the disturbance estimate value output in step S4 into the current prediction model for disturbance compensation and calculate the predicted current. Substitute the optimal current setpoint generated in step S7 into the total cost function and take the partial derivative of the total cost function of the prediction current tracking error to obtain the optimal vector combination and its final duty cycle. Step S9: Based on the final duty cycle calculation results, obtain the optimal synthetic vector and voltage vector action time. Use the triangular carrier signal to compare with the switching points of each optimal vector combination to generate the required PWM pulse signal and complete the motor control.

2. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 1, characterized in that, Step S2 is as follows: Step S21: Based on the stator phase current signals, extract the harmonic amplitude of each phase current and calculate the phase current imbalance; the phase current imbalance is calculated in the following way: First, calculate the average value of the five-phase current: The The effective value of the five-phase stator current is obtained by taking the square root of the average value of the squares of the five-phase stator currents collected over one electrical cycle. Next, calculate the phase current imbalance: ； Step S22: The harmonic amplitude of each phase current and the phase current imbalance are fused into a fault feature. Specifically, the five-phase stator current signal is subjected to Fourier transform to extract the characteristic harmonic amplitude of each phase, and the characteristic harmonic amplitude of all phases and the phase current imbalance are defined as a fused fault feature vector. After normalization of speed and load parameters, standard fault characteristic values ​​are obtained, specifically based on the current real-time speed of the motor. Load torque and the rated speed of the motor Rated load torque Using normalized coefficients The fused fault feature vector is normalized to obtain standard fault feature values; Step S23: Slide to collect standard fault feature values ​​corresponding to normal motor state, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault, construct a fault diagnosis database, and train a random forest fault diagnosis model based on the fault diagnosis database. Step S24: Execute steps S1 and S21 to S22 in real time to obtain real-time standard fault characteristic values, and obtain diagnostic variables through sliding window sampling; Step S25: Input the diagnostic variables into the trained random forest fault diagnosis model and output the motor fault diagnosis results. The fault diagnosis results include normal state, single-phase open circuit fault, adjacent two-phase open circuit fault, and non-adjacent two-phase open circuit fault.

3. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 1, characterized in that, The specific implementation method for full-speed rotor position estimation in step S3 is as follows: The five-phase permanent magnet motor stator slot is provided with a detection separation winding that is coaxially and independently wound with electrical isolation from the five-phase main power winding; in step S1, a high-frequency square wave voltage is injected into one phase separation winding of the five-phase motor, and the induced voltage of the other four phase separation windings is collected. According to the magnitude of the induced voltage of each phase, the 360° electrical angle is divided into five angle intervals corresponding to the axis of the winding, and the interval where the rotor magnetic shaft is located is initially identified. By applying polarity correction pulses to the stator windings of the motor through the inverter, the rotor magnetic pole polarity is corrected according to the amplitude change of the induced voltage of the separated windings, eliminating the 180° electrical angle error and obtaining the accurate initial rotor position in a stationary state. When the motor is running, the stator flux linkage measurement value is directly obtained by integrating the induced voltage of the isolated winding. The flux linkage measurement value is then projected onto the stationary coordinate system, and the real-time rotor position is obtained from the stationary position to the high speed range by solving the arctangent function.

4. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 3, characterized in that, The observer construction steps in step S4 are as follows: Step S41: Construct a dual-model comparison between the real motor system and the observer system. The real motor system is the decoupled model constructed in step S3, and the observer system is a model system constructed with the same structure as the real system but using nominal parameters. Step S42: Extract the stator current of the real system in real time and decouple it through step S3 to obtain the real current. , The observer system synchronously calculates the observed current using nominal parameters. , ; Step S43: Calculate the current deviation vector: When the current deviation is not equal to 0, there is a disturbance in the real system; Step S44: The observer uses the current deviation as feedback, through the gain matrix... The observer current is corrected using the following equation: Let be the discrete system matrix of the observer. Here is the gain matrix. For constant terms, and for and The system state variables at time t. This is the optimal composite vector for the previous period. , These represent the direct-axis and quadrature-axis components of the corrected current observations, respectively. These represent the direct-axis and quadrature-axis components of the corrected flux linkage observations, respectively. , These are the estimated disturbance values ​​for the direct axis and the quadrature axis, respectively. , They represent , The time period refers to the current period and the previous period; the gain matrix The observer error system is obtained by setting the eigenvalues ​​of the observer error system at the desired position using the pole placement method. The observer error system is the difference system between the real motor system and the observer system. Step S45: Output the disturbance estimate, corrected current observation, and corrected flux linkage observation.

5. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 4, characterized in that, The construction of the synthesized voltage vector in step S5 is as follows: Under normal conditions, a basic voltage vector is selected, and the action time of the large and medium vectors is adjusted according to the volt-second balance principle. Construct 10 composite vectors, the It is the median vector. For large vectors, combine two zero vectors to construct an extended control set to eliminate the third harmonic component; For single-phase open-circuit faults, the constraint of minimizing copper loss is used. By adjusting the duration of the selected basic voltage vector based on the principle of volt-second balance, a third harmonic current is synthesized. The constraints consist of six non-zero composite voltage vectors and two zero vectors; Since open-circuit faults in two adjacent phases and open-circuit faults in two non-adjacent phases cannot maintain the third harmonic space, the six non-zero basic voltage vectors are directly used as the synthesized voltage vector.

6. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 5, characterized in that, Step S6 is as follows: Step S61: Identify the flux linkage sector. The boundary of each sector is the angle bisector of the angle between adjacent composite voltage vectors. In the case of a two-phase open-circuit fault, it is the angle bisector of the angle between adjacent basic voltage vectors. Under normal conditions, 10 sectors are divided; under single-phase fault, 6 sectors are divided; under two-phase fault, 6 sectors are divided; under non-adjacent two-phase fault, 6 sectors are divided. Step S62: Offline, for each flux sector under each operating state, determine the effect of each voltage vector on the increase or decrease of flux amplitude and electromagnetic torque, and establish a given vector selection table that corresponds one-to-one with the flux sector, flux deviation binary signal, torque deviation binary signal and the optimal voltage vector. Step S63: Calculate the actual electromagnetic torque using the corrected flux linkage observation value and corrected current observation value output in step S4. Compare the actual flux linkage deviation and torque deviation with the given flux linkage value and torque value. Input the flux linkage deviation and torque deviation into the hysteresis comparator to complete the binarization process. Output the flux linkage deviation binary signal and the torque deviation binary signal, which are only 1 and -1 respectively. Step S64: When the motor is running online, the optimal voltage vector is selected by looking up the corresponding indirect given voltage vector in the table based on the current flux sector, flux deviation binary signal, and torque deviation binary signal. ; Step S65: Select After that, Two adjacent vectors are selected as the candidate optimal voltage vectors, denoted as... and Determine candidate vector combinations.

7. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 6, characterized in that, The optimal current setting mapping database in step S7 is constructed offline. The database stores the optimal current setting using speed and torque as dual indexes. The specific steps for database construction are as follows: Step A1: Determining the Constant Torque Operating Region: The constant torque operating region is defined as the motor speed being lower than the base speed. This is determined using the electromagnetic torque equation. ,make The stator current modulus is obtained by solving the Lagrange multiplier method. The minimum current vector trajectory is used to obtain the maximum torque-current ratio solution set under different torque requirements. This represents the number of pole pairs of the motor. , These represent the inductance components of the motor in the direct and quadrature axes of the synchronous rotating coordinate system, respectively. is the flux linkage constant. , These are the fundamental plane direct-axis and quadrature-axis currents, respectively; the fundamental speed is the maximum speed at which the motor outputs its rated torque under rated voltage and rated magnetic flux. Step A2, Determining the Constant Power Operating Zone: The motor is in the constant power operating zone when its speed is higher than or equal to the base speed. This is determined by offline calculation of the right-angle shaft voltage and current based on real-time speed. , , , Real-time verification of voltage limit constraint equations Stator current amplitude constraint equation First, all offline steady-state current combinations satisfying both voltage and current constraints are solved to form a feasible solution set for field weakening control. Then, using the minimum stator current amplitude under a constant given torque as the optimization principle, the optimal orthogonal axis current combination is selected from the feasible solution set. All optimal current combinations for different torques are then summarized to form the optimal current solution set for field weakening control. , These are the fundamental plane direct-axis and quadrature-axis voltages, respectively. This represents the voltage limit of the inverter under the current DC bus voltage constraint. This is the maximum current amplitude that the inverter is allowed to pass through; Step A3: Grid-based discrete storage. Set the sampling step size for the torque axis and speed axis. Fill the two-dimensional data array with the maximum torque-current ratio solution set and the optimal current solution set for field weakening control according to the state coordinate points to complete the establishment of the optimal current given mapping database.

8. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 7, characterized in that, The specific steps in step S8 are as follows: Step S81: Construct two sets of candidate vector combinations, combination 1 is... , , Combination 2 is , , , It is a zero vector; Step S82: Set constraints for each candidate vector combination: one control cycle. Only the combination of the three inner vectors acts, satisfying And since the duration of action is non-negative, the phase duty cycle is defined. The auxiliary vector duty cycle is Zero vector duty cycle ,in for Duration of action For the duration of action of the auxiliary voltage vector, The zero vector action time; Step S83: Based on the dual-plane decoupling model of the five-phase permanent magnet motor, according to the vector in the fundamental wave space of each combination... The voltage components are combined by weighting the voltage components according to the duty cycle to form the total voltage components; Step S84: Substitute the combined total voltage component into the current prediction model with disturbance compensation obtained through first-order Euler discretization to obtain the predicted current value for the next cycle. , ; Step S85: With the objectives of minimizing the tracking error of the actual current to the optimal current setpoint and ensuring strong system robustness and stability, construct a total cost function. Then, analyze the total cost function... , Find the partial derivative and set it to 0 to obtain the duty cycle for each combination. , and the corresponding total cost function value; Step S86: Compare the total cost function values ​​of the two combinations and select the combination with the smaller total cost function value as the optimal vector combination. , , Its corresponding , To achieve the optimal duty cycle, and simultaneously synthesize the optimal output voltage. .

9. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 8, characterized in that, The total cost function in step S8 is specifically as follows: Among them, sliding mode robustness item It is used to eliminate the impact of motor parameter mismatch and fault disturbance on control performance, and drive the system to converge quickly to a stable sliding surface; These are dimensionless weighted coefficients. For current error tracking term, where , , , They are respectively The direct-axis and quadrature-axis components of the fundamental plane total voltage at time intervals. , They are respectively The fundamental plane direct-axis and quadrature-axis stator currents at any given time.

10. The disturbance-compensated vector lookup table model prediction fault-tolerant control method for a five-phase permanent magnet motor according to claim 9, characterized in that, The specific current error tracking item is as follows: Under normal operating conditions, the cost function only includes the fundamental subspace current tracking error term, and its expression is: Under single-phase open-circuit fault conditions, the cost function adds a stator copper loss minimum constraint term to the current tracking error term, and the expression is: in, This is the copper loss constraint weighting coefficient. For stator copper loss, , Stator phase resistance, , , , This refers to the remaining healthy phase stator current after the fault. Under the two-phase open-circuit fault condition, the cost function, based on the current tracking error term and the stator copper loss minimum constraint term, is supplemented with a torque ripple suppression term, and the expression is: in, This is the copper loss constraint weighting coefficient. This is the torque ripple suppression weighting coefficient. The stator copper loss of the remaining healthy phase after a two-phase failure. This represents the electromagnetic torque ripple value. , This is the predicted value of the electromagnetic torque at the next moment.

Images