An integrated method and system for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-arm inverter

By using a fixed-time extended state observer and a health factor-driven integrated method for fault diagnosis and fault-tolerant control of three-phase four-arm inverters, the problems of switching impact and threshold dependence on empirical tuning caused by independent links for diagnosis and fault tolerance in existing technologies are solved. This method achieves impact-free continuous transition and analytically generated diagnostic time limits, ensuring that the fault phase current is controlled.

CN123122391APending Publication Date: 2026-11-17GUILIN UNIV OF ELECTRONIC TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611501614.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The existing open-circuit fault diagnosis and fault-tolerant control of three-phase four-arm inverters have two independent links for diagnosis and fault tolerance, resulting in switching impact and transient process. The diagnosis threshold relies on empirical tuning and lacks analytical basis, and the diagnosis time lacks a clear upper limit.

Method used

A diagnostic threshold is constructed using a fixed-time extended state observer. A health factor is generated through the phase current model and the extended state observer. Fault diagnosis and fault-tolerant control are integrated by combining the feedforward injection prediction model. The health factor is used to continuously drive the generation of reference current and the weighting of cost function to avoid hard switching of control structure.

Benefits of technology

It achieves a shockless and continuous transition for fault diagnosis and fault-tolerant control, generates diagnostic thresholds without false alarms, has a clear upper bound for diagnostic time, automatically compensates for faults by feedforward injection of lumped disturbance estimates, and constrains the fault phase current by reference values ​​to avoid runaway.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122391A_ABST
    Figure CN123122391A_ABST
Patent Text Reader

Abstract

This invention discloses an integrated method and system for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-arm inverter, belonging to the field of motor drive and fault-tolerant control technology. The method establishes a phase current model of a three-phase four-arm inverter driving a permanent magnet synchronous motor. A fixed-time extended state observer with double-power correction terms is constructed using the phase current as the state variable. A diagnostic threshold is generated from the upper bound of the observer's convergence time. The residual is obtained by maintaining the peak value of the lumped disturbance estimate. A delay duration after the residual exceeds the limit is used to determine the faulty phase. A health factor is generated from the residual using continuous saturation mapping. The lumped disturbance estimate is fed forward into the prediction model, and the current tracking error term in the cost function is weighted by the health factor. The optimal output drive signal is then selected from sixteen switching states. This method does not contain switching logic, the diagnostic delay has a provable upper bound independent of the load, and there is no transient process after a fault.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0002] This invention belongs to the field of motor drive and fault-tolerant control technology, specifically relating to an integrated method and system for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-bridge inverter. Background Technology

[0004] Permanent magnet synchronous motor drive systems are widely used in electric vehicles, CNC machine tools, aerospace actuators, and home appliances. Their power conversion stages typically employ three-phase two-level voltage source inverters. Under prolonged high-frequency switching and temperature cycling conditions, the power switching transistors in the inverter may experience open-circuit failure due to bonding wire detachment and chip solder layer fatigue. This failure mode accounts for a significant proportion of power converter failure statistics. When the switching transistors are open-circuited, half a cycle of the faulty phase current is cut off, the three-phase currents are no longer symmetrical, the motor output torque exhibits a pulsating component at the fundamental frequency, and the speed fluctuates periodically. The current amplitude of the other two phases that were not cut off increases and they are subjected to overcurrent stress, which may cause secondary faults during continuous operation.

[0005] To improve the continuous operation capability of the drive system after the switching transistors are open-circuited, existing technologies combine fault-tolerant topology with fault-tolerant control. A common form of fault-tolerant topology is a three-phase four-arm inverter, which adds a fourth arm in addition to the conventional three-phase arms. The midpoint of the fourth arm is connected to the motor neutral point, allowing independent control of the three-phase voltages. Therefore, even if any phase arm fails, the three-phase current can still be controlled. Fault-tolerant control typically consists of two parts: fault diagnosis and fault-tolerant strategies. The diagnosis part uses the current averaging method, current vector trajectory method, or normalized current method, while the fault-tolerant strategy part uses switching transistor redundancy replacement, current reference reconstruction, or direct torque fault-tolerant control.

[0006] The aforementioned existing technologies have the following shortcomings. First, diagnosis and fault tolerance are two independent links. The diagnosis section outputs a fault flag, and the fault tolerance section switches the control structure and current reference based on this flag. The control quantity jumps during the switching instant, resulting in current surges and torque ripples, and an additional transition process is required. Second, the threshold for the diagnosis section is empirically tuned and lacks analytical correlation with the observer's convergence performance. When motor parameters or loads change, the threshold becomes inapplicable. Under light load conditions, the fault characteristic amplitude decreases, significantly prolonging the diagnosis time, and under heavy load conditions, the false alarm rate increases. Summary of the Invention

[0008] The present invention aims to solve the technical problems in the existing three-phase four-arm inverter open-circuit fault tolerance technology, namely that diagnosis and fault tolerance are two independent links with switching impact and transition process, the diagnosis threshold depends on empirical tuning and lacks analytical basis, and the diagnosis time lacks a clear upper limit.

[0009] The technical solution of this invention is: an integrated method for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-arm inverter, comprising the following steps:

[0010] Step S101: Establish the phase current model of the three-phase four-bridge inverter driving the permanent magnet synchronous motor. The midpoint of the fourth bridge arm of the three-phase four-bridge inverter is connected to the neutral point of the permanent magnet synchronous motor. The voltage of the x-th phase is vxn = Vdc·(Sx-Sd), where Vdc is the DC bus voltage, Sx is the switching state of the x-th phase bridge arm, and Sd is the switching state of the fourth bridge arm. Both Sx and Sd are 0 or 1, and x takes values ​​a, b, or c. The phase current model is dix / dt = (1 / L)·(vxn-R·ix-ex)+ξx, where ix is ​​the stator current of the x-th phase, L is the stator inductance, R is the stator resistance, ex is the back electromotive force of the x-th phase, and ξx is the lumped disturbance of the x-th phase. The lumped disturbance includes stator resistance parameter mismatch, stator inductance parameter mismatch, unmodeled dynamics, and the difference between the actual phase voltage and the commanded phase voltage caused by faults.

[0011] Step S102: Construct a fixed-time extended state observer. Using the phase current in the phase current model as the state variable and the lumped disturbance as the extended state, the fixed-time extended state observer is defined as dz1x / dt=(1 / L)·(vxn-R·z1x-ex)+z2x+β1·εx+λ1·|εx|^α·sg(εx)+λ2·|εx|^β·sg(εx) and dz2x / dt=β2·εx+λ3·sg(εx)+λ4·|εx|^2 ·sg(εx), where z1x is the estimated value of the stator current of the x-th phase, z2x is the estimated value of the lumped disturbance of the x-th phase, εx=ix-z1x is the observation error of the x-th phase, β1 and β2 are the linear correction term gains, λ1 and λ2 are the power term gains with powers less than 1 and greater than 1, λ3 is the sign term gain, λ4 is the second-order power term gain, and sg(εx)=εx / (|εx|+δ) is the continuous sign function. The upper bound of the convergence time of the fixed-time extended state observer is Tmax=2 / (κ1·(1-α))+2 / (κ2·(β-1)), where κ1 and κ2 are the equivalent fixed-time gains. The upper bound of the convergence time is independent of the initial value of the observation error, that is, no matter how large the current deviation is when the fault occurs, the observation error converges within the upper bound of the convergence time.

[0012] Step S103: Generate a diagnostic threshold from the upper bound of the convergence time. The diagnostic threshold is θ = γth·ξmax, where ξmax is the upper bound of the amplitude of the lumped disturbance during normal operation, and γth is a safety factor greater than 1. The threshold is determined by the product of the upper bound of the amplitude of the lumped disturbance during normal operation and the safety factor. The upper bound of the amplitude is determined by the convergence accuracy bound of the fixed-time extended state observer. Therefore, the threshold has an analytical relationship with the convergence performance of the fixed-time extended state observer and does not require empirical tuning.

[0013] Step S104: Determine the faulty phase using the peak hold residual. The residual is obtained by peak hold of the absolute value of the lumped disturbance estimate. The peak hold condition is that the residual immediately follows the absolute value of the lumped disturbance estimate when it rises, and decays exponentially with respect to the time constant τres when it falls. The time constant τres is greater than the fundamental period Te of the stator current. The phase whose residual is greater than the diagnostic threshold and continues for a certain decision delay is determined as the faulty phase. Since the characteristics of an open-circuit fault are only observable during the conduction half-cycle of the faulty phase current, the time required to determine the faulty phase satisfies tdiag <= Tmax + Te, and this time is independent of the load torque.

[0014] Step S105: Generate a health factor from the residual. The health factor is μx = sat((θsat - rx) / (θsat - θ)), where μx is the health factor of phase x, rx is the residual of phase x, θsat is the saturation threshold, θ is the diagnostic threshold, and sat is a saturation function that restricts the independent variable to between 0 and 1, θsat = 1.5·θ. The health factor takes the value 1 during normal operation and 0 during a fault, and changes continuously with the residual between the two, thus avoiding discrete flag bit jumps.

[0015] Step S106: Feedforward injection prediction model and health-weighted optimization. The lumped disturbance estimate is used as the prediction model for the feedforward injection model to predict the current controller. The prediction model is ix(k+1)=ix(k)+Tc·[(1 / L)·(vxn(k)-R·ix(k)-ex(k))+z2x(k)], where Tc is the control period and k is the discrete time. The cost function of the model predicting the current controller is g=Σ((μx+wfl)·(ix*(k+1)-ix(k+1))^2)+λsw·gsw, where ix* is the reference value of the x-th phase current, wfl is the lower limit of the health weight greater than 0, λsw is the switching penalty weight, and gsw is the number of switching times of the candidate switching state relative to the current switching state. The reference value for the x-th phase current is ix*=-1.5·iqs·μx·sin(θx) / (Σ(μy·sinθy^2)+εden), where iqs is the q-axis current reference value output by the speed loop, θx is the rotor electrical angle of the x-th phase, μy is the health factor of the y-th phase, sinθy is the sine value of the rotor electrical angle of the y-th phase, and εden is a small protection quantity in the denominator that is greater than 0. The minimum value of the cost function is taken from all candidate vectors composed of sixteen switching states of the eight switches of the three-phase four-arm inverter, and the switching state corresponding to the minimum value is used as the drive signal of the three-phase four-arm inverter.

[0016] The beneficial effects of this invention are as follows: Diagnosis and fault tolerance are driven by the same set of outputs from the same fixed-time extended state observer. The health factor is continuously entered into the reference current generation and cost function weighting, thus eliminating the hard switching of the control structure driven by the fault flag bit. There is no current surge or torque ripple transition process after a fault. The diagnostic threshold is generated analytically by the observer's convergence performance without relying on empirical tuning, resulting in a zero false alarm rate under parameter mismatch conditions. The diagnostic time satisfies tdiag<=Tmax+Te, where Tmax is explicitly calculated by the observer gain and power, and Te is the fundamental period. Therefore, regardless of the load torque magnitude, the diagnostic time does not exceed the upper bound. The lumped disturbance estimate is fed forward into the prediction model, causing the prediction model during normal operation to degenerate into a standard form. During a fault, the difference between the actual phase voltage and the commanded phase voltage caused by the fault is automatically compensated. The current tracking error term in the cost function is weighted by the health factor and set with a weight lower limit greater than 0. This ensures that the faulty phase no longer participates in the minimum copper loss reference current allocation, while also guaranteeing that the faulty phase current is still constrained by the reference value, thus preventing the faulty phase current from going out of control. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of the integrated method for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-arm inverter according to the present invention.

[0019] Figure 2 This is a schematic diagram of the overall simulation structure of the integrated system for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-arm inverter according to the present invention.

[0020] Figure 3 This is a fault diagnosis process curve diagram of the A-phase upper tube of the three-phase four-arm inverter of the present invention when it is open-circuited.

[0021] Figure 4 The graph shows a comparison of the rotational speed and electromagnetic torque response of the method of this invention, the traditional hard-switching fault-tolerant method, and the non-fault-tolerant baseline method. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings and specific parameters.

[0024] Reference Figure 2This embodiment uses a drive system consisting of a surface-mounted permanent magnet synchronous motor and a three-phase four-arm voltage source inverter as an example. The nominal parameters of the motor and inverter are: number of pole pairs pn=4, permanent magnet flux linkage ψf=0.175Wb, nominal stator resistance Rs0=2.875ohm, nominal stator inductance L0=8.5mH, nominal moment of inertia J0=8.0e-4kg·m^2, nominal coefficient of viscous friction B0=1.0e-4N·m·s / rad, and DC bus voltage Vdc=311V. Arms a, b, and c of the three-phase four-arm inverter are connected to the three phases of the motor, and the midpoint of arm d is connected to the neutral point of the motor. The eight switching transistors constitute sixteen switching state combinations. The fixed-time extended state observer parameters are set to α=0.5 and β=1.5, with linear correction term gains β1=2000 and β2=1.0e6, double power correction term gains λ1=λ2=800, sign term gain λ3=2.0e4, second power term gain λ4=2.0e4, continuous boundary layer thickness δ=1.0e-2, and saturation limit amplitude of lumped disturbance estimate z2max=1.0e5A / s. From the convergence time upper bound formula, Tmax=2.00ms. During normal operation, the upper bound of lumped disturbance amplitude is set to ξmax=1200A / s, with a safety factor of γth=4.0, resulting in a diagnostic threshold θ=4800A / s; a saturation threshold of θsat=1.5·θ=7200A / s; a decision delay of 0.2ms; a peak hold time constant of τres=0.3s; and a latch-up initiation time of 20ms. The speed loop proportional gain Kps = 0.25, integral gain Kis = 8.0, q-axis current limit is 14A, integral limit is 60; the model predictive current control control period Tc = 50µs (control frequency 20kHz), health weight lower limit wfl = 0.30, switch switching penalty weight λsw = 1.0e-3, denominator protection small quantity εden = 1.0e-3. The simulation uses a fixed-step discrete solver with a simulation step size of 1e-5s, a total simulation time of 1.2s, a reference speed of 100rad / s, a steady-state load torque of 2.0N·m, and a fault occurrence time of 0.6s, set as an open circuit on the upper pipe of phase A bridge arm. The controller's observer, diagnostic timing, and decision window are all executed according to the control period Tc.

[0025] Step S101: Establish the phase current model. The output terminals of inverter arms a, b, and c are connected to the A-phase, B-phase, and C-phase windings of the motor, respectively. The midpoint of arm d is connected to the neutral point of the motor. Therefore, the terminal voltage of the x-th phase winding is vxn = Vdc·(Sx-Sd). Taking phase A as an example, the phase current satisfies diA / dt = (1 / L)·(vA-R·iA-eA)+xiA, where the back electromotive force eA = -we·ψf·sin(θA), we is the electric angular velocity, and θA is the electric angle of the A-phase rotor. The lumped disturbance xiA includes parameter mismatch between stator resistance and stator inductance, unmodeled dynamics, and the difference between the actual phase voltage and the commanded phase voltage caused by the fault. The purpose of this step is to unify parameter mismatch and fault effects into an extended state, so that the subsequent observer design does not need to distinguish the source of the disturbance.

[0026] Step S102: Construct a fixed-time extended state observer. One fixed-time extended state observer is configured for each phase, with the stator current of that phase as the state variable z1x, the lumped disturbance of that phase as the extended state z2x, and the commanded phase voltage and the actual phase current as known inputs. The observer's correction terms consist of linear correction terms β1·εx and β2·εx, a power term λ1·|εx|^0.5·sg(εx) with a power less than 1, a power term λ2·|εx|^1.5·sg(εx) with a power greater than 1, a sign term λ3·sg(εx), and a second-order power term λ4·|εx|^2·sg(εx). The linear correction term provides the observer with a nominal bandwidth of approximately 2000 rad / s near the steady-state operating point, ensuring continuous tracking of slowly changing disturbances. Power terms with a power less than 1 exert a control effect far greater than the linear correction under large deviations, while power terms with a power greater than 1 provide additional convergence speed under small deviations. The sign function is made continuous by sg(εx) = εx / (|εx| + δ) to avoid chattering in fixed-step discrete implementations. The convergence time upper bound formula for this embodiment yields Tmax = 2.00 ms. This upper bound is independent of the initial observation error, ensuring that the observer establishes tracking of the lumped disturbance within 2.00 ms regardless of the current deviation at the time of the fault. This step provides a source with analytical convergence guarantees for the diagnostic quantities.

[0027] Reference Figure 1 Perform the following steps:

[0028] Step S103: Generate diagnostic threshold. During normal operation, the lumped disturbance only includes parameter mismatch and unmodeled dynamics, with an upper limit of amplitude of ξmax = 1200 A / s. Taking a safety factor γth = 4.0, the diagnostic threshold θ = 4800 A / s is obtained. After the switch is opened, the phase current is unidirectionally clamped, and a deviation with an amplitude close to the DC bus voltage appears between the actual phase voltage and the commanded phase voltage. This deviation is reflected by the observer as a lumped disturbance estimate with an amplitude much greater than the diagnostic threshold. The purpose of this step is to transform the threshold tuning into the calculation of the observer's convergence boundary, so that the threshold is automatically determined by the observer parameters without relying on empirical trial and error.

[0029] Step S104: Determine the faulty phase. (Refer to...) Figure 3 The residual is obtained by maintaining the peak value of the absolute value of the lumped disturbance estimate. It immediately follows the rise and decays exponentially with a time constant of 0.3s when it falls. This time constant is greater than the fundamental period of 15.71ms, so the residual can remain at the level corresponding to the fault characteristic regardless of where it appears in the fundamental period. The phase is determined to be faulty if the residual exceeds the diagnostic threshold and remains so for a decision delay of 0.2ms. Since the characteristics of an open-circuit fault only appear within the half-cycle of the fault phase current, the diagnostic time satisfies tdiag <= Tmax + Te. In this embodiment, Te = 15.71ms, resulting in tdiag <= 17.71ms. This step aims to determine the faulty phase and provide a calculable upper bound for the determination time that is independent of the load torque.

[0030] Step S105: Generate health factors. (Refer to...) Figure 3 The health factor is calculated using μx = sat((θsat-rx) / (θsat-θ)). It is set to 1 when the residual is less than the diagnostic threshold and to 0 when the residual is greater than the saturation threshold of 7200 A / s. The values ​​between these two values ​​change continuously with the residual. This step transforms discrete fault determination into continuous weights, ensuring that subsequent reference current generation and cost function weighting change continuously without abrupt changes in the control quantity.

[0031] Step S106: Feedforward injection of the prediction model and health-weighted optimization. The prediction model is calculated as ix(k+1)=ix(k)+Tc·[(1 / L)·(vxn(k)-R·ix(k)-ex(k))+z2x(k)], where the lumped disturbance estimate z2x is injected as a feedforward term. During normal operation, z2x approaches zero, and the prediction model degenerates into the standard form. During a fault, z2x automatically compensates for the difference between the actual phase voltage and the commanded phase voltage caused by the fault. The reference value for the current of phase x is calculated as ix*=-1.5·iqs·μx·sin(θx) / (Σ(μy·sinθy^2)+εden), which is the optimal solution for the minimum copper loss allocation problem. Under normal operation, the health factors of all three phases are 1, and the reference current degenerates into a three-phase symmetrical sinusoidal current. When the health factor of phase A is 0, the reference current degenerates into a minimum copper loss allocation in which phases B and C bear all the torque, and the neutral point current of the sum of the currents of the two phases is borne by the bridge arm d. The cost function is calculated as g=Σ((μx+wfl)·(ix*(k+1)-ix(k+1))^2)+λsw·gsw, where the lower limit of the health weight wfl=0.30 ensures that when the health factor of phase A is 0, the current tracking error term of this phase still participates in the optimization with a weight of 0.30, so that the current of phase A is still constrained by the reference value and does not go out of control. The minimum cost function is obtained by iterating through sixteen switch state combinations, and the corresponding switch state is used as the driving signal. This step enables the diagnostic quantity to directly drive the fault-tolerant control of the machine, with no switching logic throughout the process.

[0032] Performance Verification: Simulations were performed using the parameters described above, and the results were compared item by item with the fault-free baseline method and the traditional hard-switching fault-tolerant method. The fault-free baseline method continues to use the original controller after a fault, while the traditional hard-switching fault-tolerant method uses the normalized average current method for diagnosis and hard-switches to two-phase fault-tolerant control using the fault flag bit. The start-up and blocking times of both methods are taken as the same value as those of this invention to ensure consistent comparison conditions.

[0033] In fault diagnosis scenarios, refer to Figure 3Before the fault occurred, the maximum values ​​of the three-phase residuals during the normal operation period (30ms to 0.60s) were 128.1A / s, 130.1A / s, and 128.1A / s, respectively, all of which were much smaller than the diagnostic threshold of 4800A / s. During the startup period (0 to 20ms), the maximum values ​​of the three-phase residuals were 117.0A / s, 240.1A / s, and 239.9A / s, respectively. Because a 20ms startup interlocking time was set, no false alarms were generated. At t=0.60s, the upper pipe of phase A is open-circuited. The residual error of phase A exceeds the diagnostic threshold at 0.6090s and is determined at 0.6092s, with a diagnostic delay of 9.20ms, which does not exceed the upper limit of 17.71ms, accounting for 51.9% of the upper limit. After the fault, the average residual error of phase A is 5553.8A / s, while the average residual errors of phases B and C are 52.3A / s and 43.3A / s, respectively. The residual error of the faulty phase is 106.2 times that of the healthy phase. After the fault, phase A... The average health factor is 0.0082, and the average health factor of phases B and C is 1.0000. The ratio of faulty phases to healthy phases is 1:122. After the fault, the current range of phase A is -0.91A to 0.00A, that is, the positive half-cycle of phase A current is cut off due to the open circuit of the upper pipe. The current amplitudes of phases B and C are 4.74A and 4.75A respectively, which bear the entire torque output. The diagnostic flag only flips once in the entire simulation process and there is no repeated switching.

[0034] In scenarios requiring fault tolerance, refer to Figure 4 The peak-to-peak values ​​of electromagnetic torque pulsation for the three methods were 5.312 N·m, 9.307 N·m, and 1.823 N·m, respectively; the peak-to-peak values ​​of speed pulsation were 10.414 rad / s, 10.062 rad / s, and 0.930 rad / s, respectively; the lowest speed values ​​were 93.280 rad / s, 92.944 rad / s, and 99.545 rad / s, respectively; the speed drops were 6.72%, 7.06%, and 0.45%, respectively; and the speed recovery times were 598.0 ms, 599.6 ms, and 0.0 ms, respectively. The method of this invention reduces torque ripple by 65.7% compared to the fault-tolerant baseline method and by 80.4% compared to the traditional hard-switching fault-tolerant method; the method of this invention reduces speed ripple by 91.1% compared to the fault-tolerant baseline method and by 90.8% compared to the traditional hard-switching fault-tolerant method; the speed drop of the method of this invention is 0.45%, and there is no transition process requiring recovery. The above comparison shows that the torque ripple of the traditional hard-switching fault-tolerant method is higher than that of the fault-tolerant baseline method. This is because the switching of the control structure driven by the fault flag bit generates a current surge at the moment of switching, while the diagnostic quantities of the method of this invention continuously enter the controller, thus eliminating this surge.

[0035] In the scenario where diagnostic time is related to speed and load, the load rates are taken as 20%, 50%, 100%, and 150% of the steady-state load torque, and the reference speeds are taken as 50 rad / s, 100 rad / s, and 150 rad / s. At a reference speed of 50 rad / s, the diagnostic delays for 20%, 50%, and 100% load rates are 18.80 ms, 18.90 ms, and 20.30 ms, respectively; at a load rate of 150%, the delay is 0.95 ms, with an upper limit of 33.42 ms. At a reference speed of 100 rad / s, the diagnostic delays are 7.55 ms, 8.35 ms, 9.20 ms, and 10.05 ms, with an upper limit of 17.71 ms. At a reference speed of 150 rad / s, the diagnostic delays are 4.85 ms, 5.10 ms, 5.70 ms, and 6.45 ms, with an upper limit of 12.48 ms. The diagnostic delays under all operating conditions ranged from 0.95ms to 20.30ms, all not exceeding the upper limit of the corresponding reference speed. The most unfavorable operating condition accounted for 60.7% of the upper limit. The diagnostic delay range for the four load rates at the same reference speed was 1.60ms to 19.35ms. This difference was not caused by the load torque, but by the difference in the position of the fault occurrence within the fundamental frequency period. The evidence for this is the scan results below.

[0036] Maintaining a load rate of 100% and a reference speed of 100 rad / s, the fault occurrence time is scanned from 0.600 s at intervals of 0.002 s to 0.614 s, covering one fundamental frequency cycle. The diagnostic delays of the method of this invention are 9.20 ms, 7.20 ms, 5.20 ms, 3.20 ms, 1.30 ms, 1.10 ms, 1.25 ms, and 10.95 ms, with a range of 9.85 ms and a standard deviation of 3.86 ms, all not exceeding the upper limit of 17.71 ms. The diagnostic delays of the traditional hard handover fault-tolerant method are 16.55 ms, 14.55 ms, 12.55 ms, 10.55 ms, 8.55 ms, 6.55 ms, 35.95 ms, and 33.95 ms, with a range of 29.40 ms and a standard deviation of 11.30 ms; the ratio of the delay range is 3.0. The diagnostic time of the method of the present invention is 0.6092s for the first five fault occurrences, that is, the judgment action is locked at the same rotor position, indicating that the decision is determined by the rotor position rather than by the timing. This is the result of the diagnostic threshold being generated by the convergence boundary of the observer.

[0037] In a scenario where faulty phases and faulty tubes are covered, six fault types are set up: open circuit in the upper tube and open circuit in the lower tube of phases A, B, and C. The diagnostic delays are 9.20ms, 1.35ms, 1.15ms, 6.60ms, 3.95ms, and 1.20ms, respectively. The location results are phases A, A, B, B, C, and C, respectively, with a fault phase location accuracy of 6:6, or 100.0%. When the upper tube of phase A is open, the current of phase A can only be negative; when the lower tube of phase A is open, the current of phase A can only be positive. The signs of the residuals are opposite in the two cases, which can be used to distinguish between the upper and lower tubes. The location of the faulty phase is not affected by this distinguishing ability.

[0038] In the parameter mismatch robustness scenario, nine combinations of stator resistance mismatch, stator inductance mismatch, and stator resistance mismatch (-30%, 0%, and +30%) were considered, operating under fault-free conditions. The number of false alarms for all nine combinations was zero, resulting in a false alarm rate of 0.0%. The maximum residual for each combination ranged from 122.8 A / s to 339.9 A / s, with the maximum residual under the nominal parameter condition being 130.1 A / s. This means that parameter mismatch increases the residual by at most 1.6 times relative to the nominal condition, still far below the diagnostic threshold of 4800 A / s. Under the condition of parameter mismatch, the diagnostic delays for the four combinations of resistance mismatch -30% and inductance mismatch -30%, resistance mismatch -30% and inductance mismatch +30%, resistance mismatch +30% and inductance mismatch -30%, and resistance mismatch +30% and inductance mismatch +30% were 8.95ms, 9.30ms, 9.05ms, and 9.30ms, respectively. The maximum residuals were 9348.0A / s, 8703.6A / s, 8879.4A / s, and 8454.1A / s, respectively. The diagnostic delays relative to the nominal parameter condition of 9.20ms did not exceed 0.10ms, indicating that the diagnostic time was not sensitive to parameter mismatch.

[0039] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An integrated method for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-arm inverter, characterized in that, include: A phase current model is established for a three-phase four-arm inverter driving a permanent magnet synchronous motor. The midpoint of the fourth arm of the three-phase four-arm inverter is connected to the neutral point of the permanent magnet synchronous motor. The voltage of the x-th phase is vxn = Vdc·(Sx-Sd), where Vdc is the DC bus voltage, Sx is the switching state of the x-th phase arm, and Sd is the switching state of the fourth arm. Both Sx and Sd are 0 or 1, and x takes the values ​​a, b, and c. A fixed-time extended state observer is constructed using the phase currents in the phase current model as state variables. The fixed-time extended state observer uses lumped disturbances as extended states. The lumped disturbances include stator resistance parameter mismatch, stator inductance parameter mismatch, unmodeled dynamics, and the difference between the actual phase voltage and the commanded phase voltage caused by faults. The correction terms of the fixed-time extended state observer include a first power term with a power less than 1 and a second power term with a power greater than 1. A diagnostic threshold is generated from the upper bound of the convergence time of the fixed-time extended state observer. The diagnostic threshold is the product of the upper bound of the lumped disturbance amplitude during normal operation and the safety factor. The residual is obtained by peak holding the absolute value of the lumped disturbance estimate output by the fixed-time extended state observer. The phase with a residual greater than the diagnostic threshold and a continuous decision delay duration is determined as the faulty phase. A health factor is generated by mapping the residual and the diagnostic threshold to continuous saturation. The health factor is 1 when the residual is less than the diagnostic threshold and 0 when the residual is greater than the saturation threshold. The saturation threshold is greater than the diagnostic threshold. The lumped disturbance estimate is injected as a feedforward term into the prediction model of the model predictive current controller. The current tracking error term of the corresponding phase in the cost function of the model predictive current controller is weighted by the health factor. The minimum value of the cost function is taken from all candidate vectors composed of the switching states of the three-phase four-arm inverter. The switching state corresponding to the minimum value is used as the drive signal of the three-phase four-arm inverter.

2. An integrated system for open-circuit fault diagnosis and fault-tolerant control of a three-phase four-arm inverter, characterized in that, include: A phase current model construction module is used to establish a phase current model of a three-phase four-arm inverter driving a permanent magnet synchronous motor. The midpoint of the fourth arm of the three-phase four-arm inverter is connected to the neutral point of the permanent magnet synchronous motor. The voltage of the x-th phase is vxn = Vdc·(Sx-Sd), where Vdc is the DC bus voltage, Sx is the switching state of the x-th phase arm, and Sd is the switching state of the fourth arm. x takes values ​​a, b, and c. A fixed-time extended state observer module is used to take the phase current in the phase current model as the state variable, and includes stator resistance parameter mismatch, stator inductance parameter mismatch, unmodeled dynamics, and faults. The lumped disturbance of the difference between the actual phase voltage and the commanded phase voltage caused by the fault is estimated in real time for the expanded state. The correction term of the fixed-time expanded state observer module includes a first power term with a power less than 1 and a second power term with a power greater than 1. The threshold generation and fault judgment module is used to generate a diagnostic threshold from the upper bound of the convergence time of the fixed-time expanded state observer module. The diagnostic threshold is the product of the upper bound of the lumped disturbance amplitude during normal operation and the safety factor. The residual is obtained by peak holding from the absolute value of the lumped disturbance estimate. When the residual is greater than the diagnostic threshold and continues to be determined for the decision delay time, the phase is determined to be a faulty phase. A health mapping module is used to generate a health factor by continuously saturating the residual and the diagnostic threshold. The health factor is 1 when the residual is less than the diagnostic threshold and 0 when the residual is greater than the saturation threshold. The saturation threshold is greater than the diagnostic threshold. A predictive control module is used to inject the lumped disturbance estimate as a feedforward term into the predictive model, weight the current tracking error term of the corresponding phase in the cost function with the health factor, take the minimum value of the cost function from all candidate vectors composed of the switching states of the three-phase four-arm inverter, and use the switching state corresponding to the minimum value as the drive signal of the three-phase four-arm inverter.

3. The method according to claim 1, characterized in that, The phase current model is dix / dt=(1 / L)·(vxn-R·ix-ex)+ξx, where ix is ​​the stator current of the x-th phase, L is the stator inductance, R is the stator resistance, ex is the back electromotive force of the x-th phase, and ξx is the lumped disturbance of the x-th phase; the back electromotive force is ex=-we·ψf·sin(θx), where we is the electric angular velocity, ψf is the permanent magnet flux linkage, and θx is the rotor electric angle of the x-th phase; the stator inductance of the permanent magnet synchronous motor takes the same value L.

4. The method according to claim 1, characterized in that, The fixed-time extended state observer is defined as dz1x / dt = (1 / L)·(vxn-R·z1x-ex)+z2x+β1·εx+λ1·|εx|^α·sg(εx)+λ2·|εx|^β·sg(εx) and dz2x / dt = β2·εx+λ3·sg(εx)+λ4·|εx|^2·sg(εx), where z1x is the estimated value of the stator current of the x-th phase, z2x is the estimated value of the lumped disturbance of the x-th phase, εx = ix - z1x is the observation error of the x-th phase, and β1 and β2 are linear corrections. The term gains are defined as follows: λ1 and λ2 are power term gains with powers less than 1 and power term gains with powers greater than 1, respectively; λ3 is the sign term gain; λ4 is the second-order power term gain; sg(εx)=εx / (|εx|+δ) is the continuous sign function; δ is the continuous boundary layer thickness greater than 0; α is a power greater than 0 and less than 1; and β is a power greater than 1. The upper bound of the convergence time is Tmax=2 / (κ1·(1-α))+2 / (κ2·(β-1)), where κ1 and κ2 are fixed-time gains, and the upper bound of the convergence time is independent of the initial value of the observation error.

5. The method according to claim 4, characterized in that, The diagnostic threshold is θ = γth·ξmax, where ξmax is the upper bound of the amplitude of the lumped disturbance during normal operation, and γth is a safety factor greater than 1; the estimated value of the lumped disturbance is saturated and limited, and the amplitude of the saturated limit z2max is greater than the saturation threshold; the characteristics of the open circuit fault are only observable within the conduction half-cycle of the fault phase current, and the time required to determine the fault phase satisfies tdiag <= Tmax + Te, where Te is the fundamental period of the stator current, and the time is independent of the load torque.

6. The method according to claim 5, characterized in that, The health factor is μx = sat((θsat-rx) / (θsat-θ)), where μx is the health factor of phase x, rx is the residual of phase x, θsat is the saturation threshold, θ is the diagnostic threshold, and sat is a saturation function that restricts the independent variable between 0 and 1, θsat = 1.5·θ; the peak value is maintained by the residual immediately following the absolute value of the lumped disturbance estimate when it rises, and exponentially decaying with time constant τres when it falls, where time constant τres is greater than the fundamental period Te.

7. The method according to claim 6, characterized in that, The prediction model is ix(k+1)=ix(k)+Tc·[(1 / L)·(vxn(k)-R·ix(k)-ex(k))+z2x(k)], where Tc is the control period and k is the discrete time; the cost function is g=Σ((μx+wfl)·(ix*(k+1)-ix(k+1))^2)+λsw·gsw, where ix* is the reference value of the x-th phase current, wfl is the lower limit of the health weight greater than 0, λsw is the switching penalty weight, and gsw is the number of switching times of the candidate switching state relative to the current switching state.

8. The system according to claim 2, characterized in that, In the predictive control module, the reference value of the current of the x-th phase is ix*=-1.5·iqs·μx·sin(θx) / (Σ(μy·sinθy^2)+εden), where iqs is the reference value of the q-axis current output by the speed loop, θx is the rotor electrical angle of the x-th phase, μy is the health factor of the y-th phase, sinθy is the sine value of the rotor electrical angle of the y-th phase, εden is the small denominator protection quantity greater than 0, and y takes values ​​of a, b, and c; the all candidate vectors are the sixteen switching state combinations of the eight switching transistors of the three-phase four-arm inverter, and the sixteen switching state combinations are obtained by combining the switching state of the fourth arm with the switching state of the three-phase arms.