A high-torque-density double-sided air-gap axial brushless DC motor and a high-precision servo driving system thereof

Abstract

The application discloses a high-torque-density double-sided air-gap axial brushless DC motor and a high-precision servo driving system thereof, and belongs to the field of axial flux motor servo driving, and solves the problems of unobservable conventional electrical quantity caused by double-sided air-gap asymmetry, torque ripple and net unbalanced axial force caused thereby, and insufficient position estimation precision under low-resolution Hall. The concentrated windings on each stator tooth of the middle stator are divided into two sections close to two rotors in the axial direction, and constitute double three-phase windings with a spatial phase shift of 30 electric angles and independently extracted; the driving system identifies the double air-gap asymmetry by the difference between the two sets of winding extension back electromotive force amplitude synchronous average values in the same identification window of the two sets of winding current instructions, applies differentiated quadrature axis current weight, differential mode direct axis current and angle domain periodic compensation according to the double air-gap asymmetry, and corrects the position observation model parameters. The application can be used for electric rudders and servo actuators.

Description

Technical Field

[0001] This invention relates to an axial flux brushless DC motor and its servo drive, and particularly to a dual-rotor intermediate stator type double-sided air gap axial brushless DC motor and its high-precision servo drive system. Background Technology

[0002] Axial flux brushless DC motors have their main magnetic flux along the motor's axial direction. A typical structure consists of two disc rotors sandwiching a central stator in an axial sandwich arrangement. An air gap is formed between each end face of the central stator and the inner disc faces of the two rotors, a so-called double-sided air gap. Compared to radial flux motors, this structure has a shorter magnetic circuit, smaller winding end protrusion, and higher winding utilization. With the same outer diameter and copper usage, it can achieve greater electromagnetic torque. It features a compact structure, small size, light weight, high torque, and low speed. The yokeless segmented armature structure, which eliminates the stator yoke, further improves torque density, making it suitable as a power source for applications sensitive to axial space, weight, and torque density, such as electric servo motors and servo actuators. During operation, this type of motor requires a servo drive system to complete commutation, current regulation, and position closed-loop control. Its torque stability and positioning accuracy depend on both the motor itself and the control strategy of the drive system.

[0003] In actual operation, it is difficult for the air gaps on both sides of this type of motor to remain equal for a long period of time. The axial dimensional chain tolerances of the housing and end cover, the axial clearance and preload deviation of the bearings, and the assembly deviations of the shaft and rotor hub will cause initial air gap deviations. During operation, the temperature rise of the intermediate stator, support, housing and the two rotor discs on both sides is not synchronized, and their thermal expansion differences will cause dynamic drift of the axial dimensions. Under the action of kilonewton-level axial magnetic pull, the two rotor discs on both sides will also produce unequal disc-shaped bending deformation and end face runout, making the air gap non-uniformly distributed along the radius and circumference. In ideal symmetry, the axial magnetic pulls of the permanent magnet arrays on both sides on their respective rotors are opposite in direction and equal in value, thus canceling each other out. The net unbalanced axial force acting on the bearing is the difference of two kilonewton-level large numbers. Therefore, it is extremely sensitive to any factor that disrupts the symmetry of the magnetic circuit on both sides. Even a small air gap deviation can form a non-negligible additional load on the bearing. Meanwhile, the unequal air gaps on both sides cause the edge magnetic flux linked by the two sets of windings to no longer be equal. The low-order magnetic flux harmonics that originally canceled each other out by the phase shift of the winding space are no longer completely canceled out. The end face runout and disc deformation of the rotor disk also modulate the sum of the air gap lengths with the mechanical rotation angle. Both of these factors introduce mechanical first and second order, as well as electrical first, second, and sixth order pulsation components into the electromagnetic torque. These components are mapped into harmonics of the same frequency in the phase current under the regulation of the current loop, increasing the effective value of the current and the additional copper loss, exciting mechanical vibration and noise, and shortening the bearing life together with the net unbalanced axial force. Existing technical means to address air gap eccentricity are mostly focused on structural optimization in the design stage, such as pole arc coefficient optimization, skewed poles, non-uniform air gap design, and finite element offline analysis. These are static prevention methods and cannot track the dynamic air gap asymmetry caused by thermal deformation and axial force fluctuations during operation. Although the master-slave vector control used in counter-rotating motors can adapt to asymmetrical loads on both sides, the torque and speed pulsation is large and the dynamic performance is not ideal. The more fundamental obstacle is that the main magnetic flux of the intermediate stator structure of the dual rotor passes through the air gaps on both sides in series. The sum of the lengths of the air gaps on both sides remains unchanged when the rotor assembly is translated along the axial direction or the intermediate stator is offset along the axial direction. As a result, the combined back electromotive force of the main magnetic flux and the conventional single set of three-phase windings hardly changes. The asymmetry of the double-sided air gap during operation is almost unobservable for conventional electrical quantities, which makes any online compensation lose the usable feedback quantity.

[0004] Meanwhile, electric servo motors have stringent requirements for cost, size, and reliability. High-resolution photoelectric encoders or rotary transformers are expensive, have large axial dimensions, and poor vibration resistance. In engineering, three switching-type Hall elements are often used as position sensors on the motor side, but they only provide 6 discrete states per electrical cycle, with a position resolution of 60 electrical angles. The instantaneous position error during startup and low-speed phases can reach ±30 electrical angles. Directly using them for commutation and position servoing will lead to low-frequency torque pulsation, noise, and insufficient positioning accuracy. Existing position estimation methods based on average speed or interpolation have inherent phase lag and are only effective when approaching uniform speed. Although observer-based or phase-locked loop-based methods can achieve higher accuracy, they are sensitive to motor model parameters. The unequal flux linkages of the two windings caused by the asymmetry of the double-sided air gap, and the flux linkage drift caused by the temperature rise of the permanent magnet, cause mismatch in the back EMF model, which increases the position estimation error. The position estimation error directly causes commutation phase deviation, further aggravating torque pulsation and causing misjudgment based on current harmonics. There is a mutually reinforcing relationship between the two.

[0005] Accordingly, the technical problem to be solved by the present invention is: how to make the dynamic double-sided air gap asymmetry generated during operation into an observable quantity that can be extracted from the end electrical quantity without increasing the high-resolution position sensor and air gap displacement sensor on the motor side, and to compensate online for the torque pulsation and net unbalanced axial force caused by it; and how to obtain the rotor position estimate that is not affected by the air gap asymmetry and permanent magnet flux temperature drift under low-resolution Hall sensing conditions, thereby eliminating the mutual aggravation of the above two aspects. Summary of the Invention

[0006] The purpose of this invention is to provide a high torque density double-sided air-gap axial brushless DC motor and its high-precision servo drive system. This addresses the problem in existing dual-rotor intermediate stator axial flux motors where the main flux passes sequentially through the two air gaps, and the sum of the lengths of the two air gaps remains constant during axial translation of the rotor assembly or axial offset of the intermediate stator. This results in the double-sided air gap asymmetry, dynamically generated by assembly tolerances, thermal deformation, and axial force fluctuations during operation, making conventional electrical quantities almost unobservable, and the resulting torque pulsation and net unbalanced axial force acting on the bearings cannot be compensated online. Simultaneously, it solves the problem that, under low-resolution position sensing conditions using three switch-type Hall elements, the position estimation accuracy is deteriorated by the unequal flux linkages of the two windings caused by double-sided air gap asymmetry and the temperature drift of the permanent magnet flux linkage. Furthermore, the position estimation error is further aggravated by commutation phase deviation, leading to torque pulsation and misjudgment. This creates a mutually reinforcing coupling problem. Therefore, without adding a high-resolution position sensor, air gap displacement sensor, and force sensor to the motor side, this invention simultaneously achieves stable electromagnetic torque, suppressed net unbalanced axial force, and high-precision position servo performance.

[0007] To achieve the above objectives, the present invention adopts the following technical solution.

[0008] This invention provides a high torque density double-sided air-gap axial brushless DC motor, comprising a shaft, a first rotor and a second rotor fixed to the shaft and rotating synchronously, and an intermediate stator located between the first rotor and the second rotor along their axial direction. The inner surfaces of the first rotor and the second rotor facing the intermediate stator are respectively provided with a first permanent magnet array and a second permanent magnet array arranged alternately with polarity along the circumference. A first air gap is formed between the first rotor and one end face of the intermediate stator, and a second air gap is formed between the second rotor and the other end face of the intermediate stator. The intermediate stator includes a plurality of stator teeth arranged along the circumference and a concentrated winding wound on each of the stator teeth. The concentrated winding is divided along the axial direction of the stator teeth into a first winding closer to the first rotor. The first winding segment and the second winding segment near the second rotor are connected by an insulating layer and led out independently. The first winding segments on each stator tooth are connected in a three-phase symmetrical pattern to form a first set of three-phase windings. The second winding segments on each stator tooth are connected in a circumferentially staggered pattern with one or more stator teeth offset and the coil polarities reversed to form a second set of three-phase windings. The resultant magnetomotive force axis of the second set of three-phase windings is spatially phase-shifted by a non-zero predetermined electrical angle from the resultant magnetomotive force axis of the first set of three-phase windings. The predetermined electrical angle is determined by the spatial phase difference between adjacent stator teeth, the number of staggered stator teeth, and the 180° electrical angle offset corresponding to the polarity reversal. This spatial phase shift can cancel out 5° of the magnetomotive force. The first and second sets of three-phase windings are star-connected, and their neutral points are isolated from each other and not led out. Therefore, there is no zero-sequence current path in either set of three-phase windings. Their phase lines are led out independently and used to connect to two independent three-phase full bridges. The number of turns of each first winding segment and each second winding segment is equal, and they are offset axially to the first air gap side and the second air gap side, respectively. In addition to linking the common axial main magnetic flux, the first set of three-phase windings and the second set of three-phase windings are also linked with edge magnetic flux components that enter from the side of the stator teeth and vary with the axial length of the first air gap and the second air gap, respectively. This transforms the originally unobservable double-sided air gap asymmetry into the difference in the amplitude of the fundamental back electromotive force of the two sets of three-phase windings, and provides an execution channel for applying differentiated armature reactions to the magnetic circuits on both sides.

[0009] As a preferred embodiment of the present invention, the number of stator teeth and the number of magnetic poles of the first permanent magnet array and the second permanent magnet array form a fractional-slot concentrated winding. The spatial phase difference between adjacent stator teeth is 150 electrical degrees, the number of staggered stator teeth is 1, and the predetermined electrical angle is 30 electrical degrees. Each phase of the first set of three-phase windings and the second set of three-phase windings is respectively composed of multiple first winding segments or multiple second winding segments connected in series. The composite magnetomotive force axis of the second set of three-phase windings lags behind or leads the composite magnetomotive force axis of the first set of three-phase windings by the predetermined electrical angle. The lag or lead is determined by the direction of circumferential staggering and is determined and stored by dragging and measuring the phase difference of the no-load back electromotive force of the two sets of three-phase windings after assembly, thereby avoiding the actual torque angle of the second set of three-phase windings deviating from the design value due to ambiguity in the description of the connection direction.

[0010] As a preferred embodiment of the present invention, the intermediate stator is a yokeless segmented armature structure, and each stator tooth is an axially penetrating fan-shaped column molded from a soft magnetic composite material bonded together with surface-insulated iron powder particles. Each stator tooth is embedded and fixed in a non-magnetic support along the circumference. The inner circumference of the non-magnetic support is provided with a positioning groove that matches the shape of the stator tooth and forms a partition between adjacent stator teeth. No parasitic magnetic circuit is formed between adjacent stator teeth. The magnetic flux in each stator tooth is closed only along the axial direction through the first air gap and the second air gap between the first rotor and the second rotor. This magnetic circuit feature increases the torque density while limiting the sensitivity of the edge magnetic flux to the asymmetry of the air gap.

[0011] As a preferred embodiment of the present invention, the end of the concentrated winding is folded inward and outward radially, and its axial end face does not exceed the end face of the stator teeth. An axial clearance is maintained between the end face of the stator teeth and the end of the concentrated winding, so as to reduce the copper loss ratio at the end and avoid the winding rubbing against the rotor disk. The pole arc coefficient of the first permanent magnet array and the second permanent magnet array is less than 1, and they are arranged in alternating polarity of fan-shaped permanent magnets, a Halebeck array, or a slanted pole arrangement combining radial segmentation and circumferential segmentation. The rotating shaft is supported by a pair of preloaded angular contact ball bearings, and the bearings bear the difference in axial magnetic pull generated by the first permanent magnet array and the second permanent magnet array.

[0012] This invention also provides a high-precision servo drive system for driving the aforementioned high torque density double-sided air-gap axial brushless DC motor, comprising an inverter power module, a current sampling module, a bus voltage and temperature acquisition module, a Hall signal acquisition module, and a main controller; the inverter power module includes two independent three-phase full-bridge circuits respectively connected to the first set of three-phase windings and the second set of three-phase windings; the current sampling module acquires the phase currents of the first set of three-phase windings and the second set of three-phase windings respectively; the Hall signal acquisition module acquires the phase currents fixed on the intermediate stator and spaced 120° apart. The discrete signals output by three switching Hall elements of the electrical angle; the main controller includes a coordinate transformation module, a high-resolution position estimation module, a dual air gap asymmetry identification module, a torque pulsation feedforward compensation module, a commutation angle and power distribution adaptive correction module, a three-closed-loop active disturbance rejection control module, and a space vector pulse width modulation output module; the dual air gap asymmetry identification module takes the phase current and the terminal voltage reconstructed by the duty cycle and DC bus voltage of the space vector pulse width modulation output module and corrected for dead zone voltage error as input, and estimates the extended back EMF amplitude of the two sets of three-phase windings respectively under the condition of taking into account the equivalent mutual inductance between the first set of three-phase windings and the second set of three-phase windings. Taking the rotor electrical angle as the phase reference, the amplitude of each extended back EMF is synchronously averaged for multiple electrical cycles according to the same electrical angle grid, and the ratio of the difference of the synchronous average values ​​of the two sets of three-phase windings to their sum is taken as the differential back EMF. The differential back electromotive force observation is measured and converted into a double-air gap asymmetry estimate characterizing the difference in axial length between the first and second air gaps according to a pre-calibrated decoupling sensitivity coefficient. The decoupling sensitivity coefficient is the calibration value of the first-order sensitivity of the edge magnetic flux component to the difference in axial length. The synchronous averaging enables the weak differential component to obtain a usable signal-to-noise ratio. The main controller sets an identification window during operation according to a set cycle. Within the identification window, the current commands of the first set of three-phase windings and the second set of three-phase windings are the same, that is, the current distribution weights of the two sets of three-phase windings are equal and the differential mode direct-axis current between them is zero. The synchronous averaging only accumulates within the identification window. The boundary of the identification window transitions with a slope, thereby making the mutual inductance, dead zone voltage error and device voltage drop common modulus for the two sets of three-phase windings and canceling them out in the differential operation, and cutting off the positive feedback path of the compensation current to the identification result.The adaptive correction module for commutation angle and power distribution allocates complementary quadrature-axis current weights, summing to 1, to the first and second sets of three-phase windings based on the estimated dual-air-gap asymmetry. This ensures that the three-phase winding adjacent to the air gap with the larger axial length receives a larger quadrature-axis current weight. It also superimposes differential-mode direct-axis currents proportional to and opposite in sign to both sets of three-phase windings, ensuring that the three-phase winding adjacent to the air gap with the smaller axial length receives a negative direct-axis current. This compresses the net unbalanced axial force acting on the bearing with almost no impact on the output torque. Simultaneously, it provides a commutation angle correction based on the estimated dual-air-gap asymmetry and uses this correction to provide the control electrical angle for coordinate transformation. The high-resolution position estimation module includes a vector tracking observer and a phase-locked loop, using the discrete signal to determine the 60°... The electrical angle sector serves as the absolute phase reference. The negative value of the direct-axis back EMF in the estimated coordinate system is constructed from the component of the extended back EMF in the stationary two-phase coordinate system and normalized by the back EMF amplitude to obtain the normalized position error signal. Then, the phase-locked loop containing a double-integral stage outputs continuous rotor position and speed estimates, ensuring that the error channel gain does not change with speed and flux amplitude and eliminating the inherent phase lag of the interpolation method. In the vector tracking observer, the permanent magnet flux parameters of the first and second three-phase windings are differentiated by performing a differential correction with opposite signs based on the product of the permanent magnet flux common-mode estimate, the double air gap asymmetry estimate, and the decoupling sensitivity coefficient. This ensures that the position error signal no longer contains the fixed bias and 2 caused by air gap asymmetry. The torque pulsation feedforward compensation module uses the estimated value of the dual air gap asymmetry as the amplitude basis and the estimated value of the rotor position as the phase reference to generate a periodic compensation current synchronized with the rotor position. This periodic compensation current is composed of a harmonic injection term given by the asymmetry model and a lookup value from an angle domain compensation table indexed by the rotor mechanical angle. The three-loop active disturbance rejection control module includes a position loop, a speed loop, a current loop, and an extended state observer. The extended state observer estimates and compensates for aperiodic disturbances to generate a basic quadrature-axis current command. This basic quadrature-axis current command, superimposed with the periodic compensation current, is weighted according to the quadrature-axis current and distributed as quadrature-axis current commands for each of the two sets of three-phase windings. The differential-mode direct-axis current is then superimposed, and the current loop separately adjusts the two sets of three-phase windings in a closed-loop manner.

[0013] As a preferred embodiment of the present invention, the dual-air-gap asymmetry identification module further includes a current harmonic identification channel. This channel takes the arithmetic mean of the cross-axis currents of the two sets of three-phase windings as input. Through adaptive harmonic extraction or a second-order generalized integrator, it separates the first and second electrical angle components, with the rotor position estimate as the phase reference. Similarly, it separates the first and second mechanical angle components according to the mechanical angle order. The separated first and second electrical angle components are then used as the basis for further analysis. The current harmonic observation is constructed by the ratio of the sum of the amplitudes of the secondary components to the DC component of the arithmetic mean. The sign of the current harmonic observation is taken from the sign of the differential back electromotive force observation. When the DC component is lower than the set minimum current threshold, the output of the current harmonic identification channel is invalidated and frozen. The differential back electromotive force observation and the current harmonic observation are fused according to weighted coefficients, and the estimated value of the dual air gap asymmetry is output after passing through a first-order low-pass filter or a Kalman filter with asymmetry degree and its rate of change. The weighted coefficients are adjusted online with the rotational speed and the common-mode component of the quadrature-axis current. When the rotational speed is lower than the set minimum identification speed, the differential back electromotive force observation is not involved in the fusion, so that the two channels complement each other under medium- and high-speed steady-state and low-speed dynamic conditions, and remain sensitive to the non-uniform air gap distribution along the circumference.

[0014] As a preferred embodiment of the present invention, the dual-air gap asymmetry identification module is provided with a compensation start threshold and an alarm threshold. The compensation start threshold is higher than the steady-state fluctuation of the estimated dual-air gap asymmetry and lower than the alarm threshold. When the absolute value of the estimated dual-air gap asymmetry is lower than the compensation start threshold for multiple consecutive update cycles, the harmonic injection term continuously decays to zero with the estimated dual-air gap asymmetry, and the angle domain compensation table is frozen and no longer updated to avoid continuous rewriting of the compensation table due to noise when there is no asymmetry. When the absolute value of the estimated dual-air gap asymmetry exceeds the compensation start threshold... At that time, the harmonic injection term changes continuously with the estimated value of the dual air gap asymmetry and the angle domain compensation table is updated. A hysteresis loop is provided at the compensation start threshold to avoid frequent start-stop. When the absolute value of the estimated value of the dual air gap asymmetry exceeds the alarm threshold, the main controller limits the torque command amplitude and the maximum speed according to the set slope, and reports the fault word and the estimated value of the dual air gap asymmetry through the communication module. The historical trend of the estimated value of the dual air gap asymmetry is stored in the non-volatile storage area for assembly quality assessment and bearing condition monitoring. When a step occurs within a set time, it is determined to be a mechanical shock or fastener loosening.

[0015] As a preferred embodiment of the present invention, the harmonic injection term includes electrical angle components of the 1st, 2nd, 6th, and 12th orders, and mechanical angle components of the 1st and 2nd orders. The amplitude of each component is a linear function of the estimated dual-air-gap asymmetry, and its constant term is the inherent pulsation amplitude under symmetrical air-gap conditions. Each component has a phase lead proportional to the rotational speed to compensate for the phase lag of the current loop. The torque gain used to convert torque to current for each component is updated online with the estimated common-mode value of the permanent magnet flux linkage. The angle domain compensation table discretizes a mechanical cycle into multiple equally spaced angle grids and linearly interpolates between the grids. The compensation amount at each grid point is iteratively updated according to the mechanical cycle number through a zero-phase low-pass filter with a forgetting factor and a learning gain with phase lead correction. Its error signal is taken from the periodic component of the total disturbance estimate of the extended state observer, the AC component of the common-mode component of the cross-axis current of the two sets of three-phase windings, or one of the angle domain waveforms of the rotational speed fluctuation, and normalized to the rated torque. The residual of the absorbed model injection amount is used to converge the inherent cogging pulsation that already exists in the symmetrical state. The iterative update is only performed within the steady-state window when the speed fluctuation, load change rate, and the change rate of the estimated value of the dual air gap asymmetry are all lower than the set value. When a speed command step or load change is detected, the update is frozen to avoid non-periodic disturbances being mistakenly learned into the periodic compensation table. The compensation amount of each grid point has an upper limit of amplitude and is written to the non-volatile storage area as the initial value of the next power-on iteration to shorten the number of mechanical cycles required for convergence. The periodic pulsation synchronized with the rotor position is handled by the harmonic injection term and the angle domain compensation table, while the non-periodic disturbance is handled by the extended state observer. The bandwidth of the extended state observer is between the bandwidth of the speed loop and the bandwidth of the current loop, so that it does not need to increase the bandwidth to track the periodic pulsation.

[0016] As a preferred embodiment of the present invention, the distribution gain of the quadrature-axis current weight is taken as the weighted value between a first gain aimed at equalizing the electromagnetic torque contributions of the two sets of three-phase windings and a second gain aimed at minimizing the net unbalanced axial force acting on the bearing. The first gain is taken as the decoupling sensitivity coefficient, and the second gain is greater than the first gain. The weighting coefficient is oriented towards the second gain under heavy load and low speed conditions, and towards the first gain under light load and high speed conditions, thereby making a trade-off between bearing life and efficiency temperature rise according to the operating conditions. The quadrature-axis current weight has upper and lower limits. The commutation angle correction is only used for the coordinate transformation of the currents of the two sets of three-phase windings and the inverse transformation of the voltage command, and is not used for the position feedback of the position loop, so as to avoid the commutation phase correction from polluting the position control accuracy. When the effective value of the current loop tracking error of any set of three-phase windings exceeds the set ratio of its command amplitude, or an open circuit or phase loss of the three-phase winding or a short circuit of its three-phase full bridge arm is detected, or the ratio of the equivalent flux linkage amplitudes of the two sets of three-phase windings deviates from 1 When the set ratio is exceeded, the three-phase winding or its power channel is determined to be faulty. The corresponding three-phase full bridge is blocked and the quadrature axis current weight of the remaining three-phase winding is reset to 1. At the same time, the upper limit of torque command is reduced, and the update of the dual air gap asymmetry estimation value is switched to use only the current harmonic identification channel to achieve single-sided fault-tolerant operation.

[0017] As a preferred embodiment of the present invention, the high-resolution position estimation module, each time the discrete signal undergoes a state reversal, subtracts the rotor position estimate from the sum of the theoretical boundary phase of the 60 electrical angle corresponding to the reversal and the Hall installation error correction amount of the sector to which the boundary belongs. The difference is multiplied by a fusion coefficient to incrementally correct the rotor position estimate, thereby eliminating integral drift without creating position steps. The fusion coefficient decreases as the back electromotive force amplitude increases. A single correction amount has an amplitude limit, and exceeding the limit results in an accumulated fault count. The Hall installation error correction amount is calculated by statistically analyzing the difference between the actual phase and the theoretical boundary phase of each sector's reversal boundary during steady-speed operation, and stored in a table. It is applied only once in the incremental correction to avoid repeated application of the same error. The main controller uses the absolute position feedback from the output of the reduction mechanism driven by the motor, calculated according to the transmission ratio, the power-on return-to-zero process, or the Hall reversal interval formed by the inherent differences of the magnetic poles of the first permanent magnet array. One of the three operations—sequence, pre-stored fingerprint template, and cyclic correlation—establishes the mechanical angle reference and full-turn count. The angle grid of the angle domain compensation table is aligned with the mechanical angle reference, so that the compensation table can be aligned with the actual mechanical angle across power-on cycles. The main controller also includes a permanent magnet flux linkage adaptive observer module, which takes the common-mode components of voltage and current of two sets of three-phase windings as input and the common-mode equivalent inductance as parameter. It outputs the common-mode estimate of the permanent magnet flux linkage online according to the model reference adaptive law. The common-mode estimate of the permanent magnet flux linkage is used to correct the model parameters of the vector tracking observer, the torque conversion gain of the periodic compensation current, and the input gain of the three-closed-loop active disturbance rejection control module. It also back-calculates the average temperature of the permanent magnet based on the remanent magnet temperature coefficient, so that the position estimation accuracy and torque accuracy do not systematically deteriorate during the temperature rise of the windings and permanent magnets.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The concentrated winding is divided axially into two winding segments, one near each side of the rotor, forming a 30° electrical-angle spatially phase-shifted, independently led-out double three-phase winding. The flux linkages of the two sets of three-phase windings each contain edge flux components directly related to the length of the adjacent air gap. The nearly unobservable bifacial air gap asymmetry, originally due to the air gaps being connected in series on the same main magnetic circuit, is thus transformed into a difference in the amplitude of the fundamental back electromotive force of the two sets of three-phase windings. The first-order sensitivity coefficient of this difference component is the same as the first-order sensitivity coefficient of the net unbalanced axial force; that is, the coefficient that determines whether the asymmetry can be observed also determines the magnitude of its harmfulness. This coefficient is approximately 0.09 in the yokeless segmented armature structure and approximately 0.21 in the stator-yoke structure. By combining synchronous averaging based on rotor electrical angle as phase reference, channel zero bias and gain consistency subtraction, and updating within the identification window where the differential excitation of the two windings is temporarily set to zero, dual air gap asymmetry can be obtained online without air gap displacement sensors. After weighted fusion of this differential channel with the identification channel based on the first and second harmonics of the quadrature axis current according to operating conditions, a signed estimate is provided under medium- and high-speed steady-state conditions, and it remains effective under low-speed and dynamic operating conditions. Based on random noise converted to 0.002, systematic residuals of the two channels converted to 0.005, and residuals of dead zone and device voltage drop after common-mode cancellation within the equal current window converted to 0.003, the steady-state fluctuation of the asymmetry estimate is better than 0.01, which is lower than the set compensation start threshold of 0.03.

[0020] Harmonic injection feedforward and angle domain iterative learning compensation, driven by this asymmetry, suppress torque pulsation under artificially set air gap asymmetry conditions. At an asymmetry of approximately 0.10, and under rated speed and rated torque conditions, the peak-to-peak value of torque pulsation decreased from 10.8% of the rated torque to 2.2%. Specifically, the amplitude of the first mechanical component decreased from 1.19% to 0.15%, the second mechanical component from 0.51% to 0.08%, the first electrical angle component from 0.90% to 0.12%, the second electrical angle component from 0.41% to 0.07%, and the sixth electrical angle component from 2.38% to 0.42%. Because the iterative learning converges in the angle domain for all periodic components synchronized with the rotor position, the 12th electrical angle cogging component, already present in the symmetrical state, also decreased from 1.40% to 0.35%. At an asymmetry of approximately 0.05, the peak-to-peak value of torque pulsation decreased from 8.3% to 2.0%. After applying differentiated quadrature-axis current weights to the two sets of three-phase windings according to the degree of asymmetry and superimposing differential-mode direct-axis currents, the net unbalanced axial force acting on the bearing decreases from approximately 55 N to approximately 20 N when the degree of asymmetry is approximately 0.10, and from approximately 33 N to approximately 13 N when the degree of asymmetry is approximately 0.05; in the yoke-deformed implementation with relatively independent magnetic circuits on both sides, it decreases from approximately 116 N to approximately 42 N at the same degree of asymmetry. The efficiency improvement is a minor effect and of limited magnitude: the overall efficiency at the rated operating point increases from 91.1% to 91.2%, with a reduction of approximately 0.15 W in harmonic copper losses, approximately 1.0 W in harmonic iron losses and permanent magnet eddy current losses, and approximately 0.2 W in bearing friction losses; at low-speed conditions with 40% of the rated speed and rated torque, it increases from 81.9% to 82.2%.

[0021] Using a low-resolution Hall signal as the absolute phase reference, a normalized position error is constructed by a vector tracking observer and a continuous position is output through a phase-locked loop containing a double-integration stage. Combined with soft fusion correction of the Hall flip boundary, position estimation is no longer limited by the inherent phase lag of the interpolation method. The position resolution is improved from 60 electrical degrees of the Hall signal itself to better than 0.1 electrical degrees. The steady-state position estimation error of the motor shaft at rated speed converges to within ±2.8 electrical degrees, i.e., a mechanical angle of approximately ±0.56°. The maximum instantaneous error during startup is reduced from approximately ±30 electrical degrees to approximately ±5 electrical degrees, without adding a high-resolution encoder or rotary transformer to the motor side. The positioning accuracy of the servo motor output also depends on the hysteresis of the reduction mechanism and the accuracy of the original absolute feedback at the output. In this embodiment, under the conditions of a reduction ratio of 50 and a reduction mechanism hysteresis of approximately 0.03°, the measured static positioning error at the output is no greater than 0.05°.

[0022] The asymmetry estimate is injected into the position observation model in a differentiated form according to the flux linkages of the two windings. The adaptive observation results of the permanent magnet flux linkage simultaneously correct the observation model, torque coefficient, and input gain of the active disturbance rejection control. This ensures that the position estimation accuracy does not significantly deteriorate under thermal equilibrium conditions with a winding temperature rise of approximately 80 K compared to the normal temperature conditions. As a control, when the above parameter adaptation and asymmetry injection are turned off and the fixed parameters at normal temperature are used, the position estimation error under the same conditions expands to approximately ±11 electrical degrees, and the peak-to-peak torque pulsation recovers to 5.8%. The position estimate, in turn, provides a phase reference for harmonic extraction, harmonic injection, and iterative learning, preventing the compensation current from being misjudged as a new asymmetry quantity due to phase errors. Setting the differentiated excitation of the two windings to zero within the identification window cuts off the contamination path of the compensation current through the mutual inductance of the two windings to the differential observation quantity, avoiding positive feedback in the asymmetry estimation. The extended state observer handles aperiodic disturbances, while the angle domain compensation handles periodic pulsations synchronized with the rotor position. This division of labor means that the bandwidth of the extended state observer does not need to be increased to track periodic pulsations. When the step load is 50% of the rated torque, the speed drop is reduced from 42 r / min to 14 r / min and the recovery time is shortened from 53 ms to 19 ms. At the same time, the effective value of the high-frequency chattering of the quadrature shaft current command is maintained within 1.5% of the rated current.

[0023] The aforementioned identification, compensation, and parameter adaptation are all implemented in software within the same main controller. Except for the independent lead interface of the dual three-phase windings set up by the motor body for load compensation, no additional sensor hardware is added. The mechanical angle reference is established by one of three factors: the original absolute position feedback at the servo output, the power-on return-to-zero process, or the Hall flip interval fingerprint formed by the inherent difference of the permanent magnet. Based on this, the angle domain compensation table can be aligned with the actual mechanical angle across power-on cycles. The calibration coefficients and compensation table are stored in the non-volatile memory area and used as the initial value for iteration on the next power-on. The number of mechanical cycles required for convergence is reduced from about 300 cycles to about 40 cycles. The long-term trend of asymmetry can be used for assembly quality assessment and bearing condition monitoring. Attached Figure Description

[0024] Figure 1 is a schematic diagram of the axial cross-sectional structure of the motor body in an embodiment of the present invention;

[0025] Figure 2 is a schematic block diagram of the servo drive system in an embodiment of the present invention;

[0026] Figure 3 is a signal flow diagram of the control method in an embodiment of the present invention;

[0027] Figure 4 is a schematic diagram illustrating the principle of online identification and collaborative compensation of dual air gap asymmetry in an embodiment of the present invention;

[0028] Figure 5 is a comparison curve of torque pulsation and position estimation error before and after compensation in an embodiment of the present invention;

[0029] Figure 6 is a schematic diagram showing the axial segmentation and phase band connection of the double three-phase winding in an embodiment of the present invention;

[0030] Figure 7 is a timing diagram of the identification window and compensation update in an embodiment of the present invention.

[0031] Reference numerals: 100, Motor body; 101, Intermediate stator; 101a, Stator teeth; 101b, Non-magnetic support; 101c, Concentrated winding; 101d, Dual three-phase winding; 102, First rotor; 102a, First permanent magnet array; 103, Second rotor; 103a, Second permanent magnet array; 104, First air gap; 105, Second air gap; 106, Shaft; 107, Bearing; 108, Temperature sensor; 200, Servo drive system; 201, Inverter power module; 202, Current sampling module; 203, Bus voltage and temperature acquisition module; 204, Hall signal acquisition module; 2 05. Coordinate Transformation Module; 206. High-Resolution Position Estimation Module; 2061. Vector Tracking Observer; 2062. Phase-Locked Loop; 207. Permanent Magnet Flux Adaptive Observer Module; 208. Dual Air Gap Asymmetry Identification Module; 209. Torque Pulse Feedforward Compensation Module; 210. Commutation Angle and Power Distribution Adaptive Correction Module; 211. Three-Closed-Loop Active Disturbance Rejection Control Module; 2111. Position Loop; 2112. Velocity Loop; 2113. Current Loop; 2114. Extended State Observer; 2115. Feedforward Link; 212. Space Vector Pulse Width Modulation Output Module; 213. Main Controller; 214. Communication Module. Detailed Implementation

[0032] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be noted that the following embodiments are only used to illustrate the technical solutions of the present invention. The slot-pole fits, geometric dimensions, electrical parameters, control cycles, and specific values ​​of each gain and threshold given are all exemplary. Solutions obtained by those skilled in the art based on the same technical concept and recalibrated according to actual motor parameters, as well as equivalent substitutions and local adjustments made on this basis, all fall within the protection scope of the present invention. For ease of comparison, all symbols herein adopt a uniform convention: For extreme logarithms, To calculate the equivalent moment of inertia of the motor shaft, This refers to the transmission ratio of the servo motor's reduction gear mechanism. For the resistance of each phase of a single three-phase winding, , It is a shaft inductor with a single set of three-phase windings. The equivalent mutual inductance between the two sets of three-phase windings. This represents the actual electrical angle of the rotor. Its estimated value, For the mechanical angle of the motor shaft, and For the actual and estimated electric angular velocity, the subscript is used. Indicates the first or second set of three-phase windings, subscript Indicates the discrete time sequence number. and These represent the harmonic orders in the electrical angular domain and the mechanical angular domain, respectively. This indicates the mechanical cycle number of the iterative learning process. Indicates the Hall sector number.

[0033] As shown in Figure 1, the motor body 100 in this embodiment is a dual-rotor intermediate stator axial flux brushless DC motor. Its main magnetic flux direction is along the motor axis, and the overall arrangement is axially sandwiched. From one end of the axial direction to the other, it consists of a first rotor 102, a first air gap 104, an intermediate stator 101, a second air gap 105, and a second rotor 103. The intermediate stator 101 is located in the middle of the motor axis and adopts a yokeless segmented armature structure, that is, the traditional stator yoke is eliminated, and magnetic flux transmission is carried out by several magnetically isolated independent iron core blocks. In this embodiment, the intermediate stator 101 is composed of 12 stator teeth 101a evenly arranged along the circumference. The stator teeth 101a are approximately fan-shaped cylinders that extend axially, with an axial length of 25 mm, an inner radius of 35 mm, an outer radius of 60 mm, and a duty cycle of approximately 72% along the circumferential direction. The stator tooth 101a is made of soft magnetic composite material through compression molding and heat treatment. The soft magnetic composite material is composed of iron powder particles with insulating surfaces bonded together. Its three-dimensional isotropic magnetic properties are suitable for carrying axial magnetic flux, and the insulating layer between the particles confines the eddy currents within the particle size. In this embodiment, the fundamental frequency does not exceed 150 Hz across the entire speed range, but the magnetic flux in the stator tooth 101a also carries high-frequency components caused by the switching frequency and its sidebands. Within this wide frequency range, the total iron loss of the soft magnetic composite material is lower than that of laminated silicon steel when magnetized axially. At the same time, compression molding allows the two end faces of the stator tooth 101a to remain parallel to the rotor disks on both sides with small dimensional and positional tolerances, providing a technological basis for controlling the initial consistency of the first air gap 104 and the second air gap 105.

[0034] Twelve stator teeth 101a are circumferentially embedded and fixed within a non-magnetic support 101b. The non-magnetic support 101b is integrally formed from cast aluminum alloy or glass fiber reinforced engineering plastic. Its inner circumference is provided with positioning grooves that match the shape of the stator teeth 101a, and it forms ribs between the stator teeth 101a, serving functions such as circumferential positioning, axial clamping, winding insulation, and heat dissipation channels. Since the non-magnetic support 101b is non-magnetic, no parasitic magnetic circuit is formed between adjacent stator teeth 101a. The magnetic flux of each stator tooth 101a can only close between the two rotors along the axial direction through the first air gap 104 and the second air gap 105. This magnetic circuit characteristic has a decisive influence on the observability analysis of the double air gap asymmetry in the following text. Each stator tooth 101a has a concentrated winding 101c wound on it. The concentrated winding 101c is wound with enameled copper wire and a polyimide composite film insulating skeleton. The wire gauge consists of two enameled wires with a diameter of 0.8 mm wound in parallel, with a slot fill factor of approximately 65%. The ends of the concentrated winding 101c are folded back radially inward and outward, and their axial end faces do not exceed the end faces of the stator teeth 101a. An axial clearance of not less than 0.3 mm is maintained between the end faces of the stator teeth 101a and the winding ends to ensure that there is no rubbing against the rotor disk surface within the 1.0 mm single-sided air gap. Since the ends only occupy space radially and not axially, the proportion of end copper loss to total copper loss is much lower than that of radial flux motors of the same capacity. This is one of the structural bases for achieving a higher torque density in this embodiment. The intermediate stator 101 adopts a 12-slot, 10-pole configuration. The winding coefficient of this combination is high, the cogging torque is small, and the spatial phase difference between adjacent stator teeth 101a is 150 electrical degrees.

[0035] As shown in Figure 6, the concentrated winding 101c is not wound continuously along the axial length of the stator tooth 101a, but is divided into two winding segments along the axial direction, namely the first winding segment near the first rotor 102 and the second winding segment near the second rotor 103. Each winding segment has 22 turns, with a 0.5 mm thick insulating layer between them and led out independently. The first winding segments on the 12 stator teeth 101a are connected in a three-phase symmetrical pattern. Each phase consists of 4 first winding segments connected in series, totaling 88 turns, forming the first set of three-phase windings. Since this embodiment uses a 12-slot, 10-pole configuration and the spatial phase difference between adjacent stator teeth 101a is 150 electrical degrees, the 4 first winding segments of the same phase are located on two pairs of circumferentially adjacent stator teeth 101a. Furthermore, the two winding segments on each pair of adjacent teeth have opposite polarities. For example, phase A1 occupies teeth 1, 2, 7, and 8, where teeth 1 and 8 are positive polarities and teeth 2 and 7 are negative polarities. Phase B1 occupies teeth 3, 4, 9, and 10, and phase C1... Phases occupying teeth 5, 6, 11, and 12 have a distribution coefficient of 0.966, a pitch coefficient of 0.966, and a winding coefficient of 0.933. If the phase bands are mistakenly connected according to the rule of taking one coil every three stator teeth, the combined back electromotive force vectors of the four winding segments in the same phase will differ by 90 electrical degrees and cancel each other out, resulting in a winding coefficient of zero. Therefore, the phase band arrangement must strictly follow the rule of pairing adjacent teeth. The second winding segments on the 12 stator teeth 101a are connected according to the rule of offsetting one stator tooth 101a along the circumference and simultaneously reversing, forming the second set of three-phase windings. Since the spatial phase difference between adjacent stator teeth 101a is 150 electrical degrees, and the reversal of coil polarity is equivalent to an additional phase shift of 180 electrical degrees, the superposition of the two causes the axis of the combined magnetomotive force of the second set of three-phase windings to shift by 330 electrical degrees relative to the first set of three-phase windings, i.e., lag by 30 electrical degrees. Both the first and second three-phase windings are star-connected, with the two neutral points isolated from each other and not led out to the outside. Therefore, there is no zero-sequence current path in either winding. Each three-phase winding has three phase lines, totaling six phase lines, which are connected to the two three-phase full-bridge windings of the inverter power module 201. Together, they form a double three-phase winding 101d. It should be noted that the circumferential direction of the "staggered stator tooth" determines the sign of the 30 electrical angle phase shift. If the staggered direction is reversed, the second three-phase winding will be 30 electrical angle ahead. In this embodiment, after assembly, the phase difference of the no-load back electromotive force of the two three-phase windings is measured by low-speed dragging. Based on its sign, the sign of the phase shift term in each transformation angle in the following text is determined at once and written into the non-volatile storage area of ​​the main controller 213, thereby avoiding the actual torque angle of the second three-phase winding deviating from 60 electrical angles due to ambiguity in the connection direction description.

[0036] The aforementioned structural differentiation design brings three benefits: First, the two sets of three-phase windings are axially offset to one side of the first rotor 102 and the second rotor 103, respectively. Their magnetic flux contains edge flux components directly related to the length of adjacent air gaps, enabling the servo drive system 200 to extract the difference information of the length of the air gaps on both sides from pure electrical quantities, without the need to install displacement sensors in the air gaps. Second, the servo drive system 200 can apply current vectors with different amplitudes, phases, and even direct-axis components to the two sets of three-phase windings, thereby applying differentiated armature responses to the magnetic circuits on both sides, allowing the pure software compensation strategy to obtain an actual execution channel acting on the physical air gap asymmetry. Third, the two sets of three-phase windings are spatially phase-shifted by 30 electrical degrees, which can mutually cancel the 5th and 7th harmonics in the magnetomotive force, and support unilateral derating operation when one of them experiences an open-circuit or short-circuit fault.

[0037] Both the first rotor 102 and the second rotor 103 are disc-type structures, with a 6 mm thick low-carbon steel disc as the rotor yoke, located on opposite axial sides of the intermediate stator 101. The first permanent magnet array 102a and the second permanent magnet array 103a are respectively mounted on their inner disc surfaces facing the intermediate stator 101. In this embodiment, each of the first permanent magnet array 102a and the second permanent magnet array 103a consists of 10 fan-shaped neodymium iron boron permanent magnets arranged in alternating polarities along the circumference. The magnets are of grade N38SH and have room temperature remanence. Take 1.22 T, relative permeability Take 1.05 as the axial thickness of the permanent magnet. The diameter is 4 mm, and the pole arc coefficient is 0.85. To improve the sinusoidal strength of the air gap magnetic flux density, reduce leakage flux, and lower the risk of permanent magnet demagnetization, the first permanent magnet array 102a and the second permanent magnet array 103a can also adopt a Hellbeck array or a skewed pole arrangement combining radial and circumferential segments. A first air gap 104 is formed between the inner disk surface of the first rotor 102 and one end face of the intermediate stator 101, and a second air gap 105 is formed between the inner disk surface of the second rotor 103 and the other end face of the intermediate stator 101. These two constitute the double-sided air gap as referred to in this invention, and their axial lengths are respectively denoted as... and Under ideal assembly conditions mm. The dimensionless double-air-gap asymmetry is defined as...

[0038] (1)

[0039] When ideally symmetrical , This indicates that the first air gap 104 is greater than the second air gap 105, and its estimated value is denoted as [value missing]. The entire text refers to this as double air gap asymmetry. The first rotor 102 and the second rotor 103 are fixed to the same rotating shaft 106 via a hub with a stop and rotate synchronously. The rotating shaft 106 is supported by a pair of bearings 107 arranged at both ends of the housing. In this embodiment, the bearings 107 are a pair of preloaded angular contact ball bearings, which simultaneously bear the difference in axial magnetic pull generated by the first permanent magnet array 102a and the second permanent magnet array 103a. The temperature sensor 108 is a negative temperature coefficient thermistor, embedded between the root of a stator tooth 101a of the intermediate stator 101 and the end of the concentrated winding 101c and fixed with thermally conductive adhesive. It is used to provide a rough measurement of the winding temperature. Its measurement result is only used as auxiliary information for parameter calibration and rationality verification. Because its thermal time constant is much larger than the temperature change scale of the permanent magnet and there is an air gap thermal resistance between the measuring point and the permanent magnet, it does not participate in the main loop calculation of position estimation and flux linkage estimation.

[0040] In the yokeless segmented armature structure, the main magnetic flux originates from a pole of the first permanent magnet array 102a, passes through the first air gap 104, enters the stator tooth 101a, passes axially through the stator tooth 101a, passes through the second air gap 105, enters the opposite pole of the second permanent magnet array 103a, then passes through the rotor yoke of the second rotor 103, folds back circumferentially to the adjacent pole, and then passes in the opposite direction through the second air gap 105, the adjacent stator tooth 101a, and the first air gap 104 back to the rotor yoke of the first rotor 102, thus closing the circuit. Therefore, the first air gap 104, the second air gap 105, and the permanent magnets on both sides are connected in series in the main magnetic circuit. The peak value of the air gap magnetic flux density when the tooth surface faces the permanent magnet can be approximately expressed as...

[0041] (2)

[0042] Substituting the parameters of this embodiment into equation (2) yields the ideal symmetrical state. The peak magnetic flux density of the stator tooth 101a is approximately 1.35 T, which is lower than the saturation magnetic flux density of the soft magnetic composite material, allowing for necessary overload margin. Taking into account inter-pole leakage flux, tooth surface duty cycle, and Carter's coefficient, the average air gap magnetic flux density along the entire air gap working area is... T. All calculations involving axial force below will use the average magnetic flux density. The peak value of equation (2) Distinguishing between them. The axial magnetic pull exerted by a single-sided permanent magnet array on the corresponding rotor disk, calculated using the Maxwell tensor, is:

[0043] (3)

[0044] In the formula The effective working area of ​​a single-sided air gap is approximately [missing information - likely a measurement or value]. In this embodiment, the annular effective area corresponding to the stator tooth surface 101a is approximately [missing information - likely a measurement or value]. m², This represents the average magnetic flux density of the air gap on that side. is the vacuum permeability. Substituting the above value into equation (3), we can obtain that the axial magnetic pull force on one side under ideal symmetrical conditions is about 1260 N, and the direction is to pull the rotor on this side towards the middle stator 101.

[0045] The source of the net unbalanced axial force must be accurately defined here, as it simultaneously determines the object to be compensated for and the available observables in this invention. As can be seen from equation (2), in a yokeless segmented armature structure, the main magnetic flux depends only on… If the rotor assembly is translated along the axial direction or the intermediate stator 101 is offset along the axial direction, The main magnetic flux density in the air gaps on both sides remains constant and equal, and the axial forces on both sides calculated by equation (3) are also equal and cancel each other out. This property shows that the intermediate stator structure of the dual rotor under the ideal series magnetic circuit has the advantage of self-balancing axial force. It also shows that the net unbalanced axial force and the difference in back electromotive force between the two windings are not contributed by the main magnetic circuit, but by the secondary effects that disrupt the symmetry of the magnetic circuits on both sides. Specifically, these include the edge magnetic flux entering from the side of stator tooth 101a, the difference in the local saturation degree at the tooth tip, the residual parasitic magnetic circuit between adjacent stator teeth 101a, and the consistency deviation of the permanent magnet arrays on both sides in terms of remanence and thickness. The total influence of these secondary effects is summarized into a dimensionless decoupling sensitivity coefficient. The first-order expression for the net unbalanced axial force is:

[0046] (4)

[0047] In the formula This is caused by the inconsistency deviation of the permanent magnet arrays on both sides, and The irrelevant residual component, measured in this embodiment, is approximately 10 N. It is precisely in equation (13) that the difference in back electromotive force between the two windings is extracted. The same sensitivity coefficient has important engineering implications: it determines whether the double-air-gap asymmetry can be observed by electrical quantities, and also determines the magnitude of the net axial force caused by the asymmetry; the tighter the coupling of the two magnetic circuits, the more difficult the asymmetry is to observe, but the less harmful it is; conversely, the closer the two magnetic circuits are to being independent, the easier it is to observe, but the greater the net axial force that needs to be compensated. The yokeless segmented armature structure of this embodiment has been calibrated to obtain... Substituting into equation (4), the net unbalanced axial force corresponding to a unit asymmetry is approximately 453 N, i.e. Approximately 33 N Approximately 55 N; in the following embodiment with stator yoke deformation... The corresponding value is approximately 1057 N per unit of asymmetry. Compared to the axial force of approximately 1260 N on each side, the above net value is only a few percent of that. Any method that relies on calculating or measuring the axial forces on both sides separately and then subtracting them is unlikely to achieve usable accuracy. This is why this embodiment does not use a mechanical approach but instead uses electrical quantities to identify the asymmetry online.

[0048] Another consequence of asymmetrical double-sided air gaps is the simultaneous deterioration of torque pulsation and position estimation accuracy, with the two reinforcing each other. Regarding torque pulsation, its causes can be divided into three categories. The first category originates from the end-face runout and disc deformation of the rotor disc: in this case, the sum of the air gap lengths varies with the mechanical angle and radius, and the main magnetic flux is modulated by a component with a periodicity equal to the mechanical angle, manifesting as first and second mechanical pulsations in the electromagnetic torque. The first type corresponds to the 1 / 5 and 2 / 5 orders in the electrical angle domain. The second type originates from the unequal flux linkage amplitudes of the two sets of three-phase windings: on the one hand, the equivalent composite flux axis of the double three-phase winding 101d shifts relative to the ideal position, resulting in a deviation in the equivalent commutation phase; on the other hand, the 5th and 7th flux linkage harmonics, which originally canceled each other out by a 30-degree electrical angle spatial phase shift, no longer completely cancel out, introducing the 1st and 2nd orders and a residual 6th order component in the electrical angle domain, respectively. The third type is the inherent pulsation that already exists in the symmetrical state, mainly the 12th cogging component determined by the 12-slot 10-pole combination and the 6th order component caused by the non-sinusoidal flux linkage. The asymmetry of the air gap will increase its amplitude. Under the adjustment of the current loop 2113, the above-mentioned pulsation components are simultaneously mapped into harmonics of the same frequency in the phase current, which on the one hand increases the effective value of the winding current and harmonic iron loss, and on the other hand excites the mechanical structure to generate vibration and noise, and together with the net unbalanced axial force borne by the bearing 107, shortens the bearing life. In terms of position estimation, air gap asymmetry causes the permanent magnet flux linkages of the two sets of three-phase windings to become unequal. If the position observer still models with uniform flux linkage parameters, the phase of the back electromotive force output by the model will deviate from the actual phase as the asymmetry increases, thus increasing the position estimation error. This position estimation error, in turn, causes commutation phase deviation, further exacerbating torque ripple according to the second type of cause, and leading to misjudgments in asymmetry identification based on current harmonics. Therefore, compensation for double-sided air gap asymmetry and high-precision position servo control cannot be implemented in isolation; they must be designed collaboratively.

[0049] The main parameters of the prototype in this embodiment are as follows, which are used to explain the basis for the values ​​of each gain and threshold in the following text. The rated continuous operating point is 1000 r / min with a torque of 8 N·m, the short-time peak torque is 24 N·m, and the maximum speed under light load is 1800 r / min; the DC bus voltage is 100 V. The resistance per phase of a single three-phase winding is... The inductance is 0.15 Ω at 20 ℃, approximately 0.16 Ω at 40 ℃, and approximately 0.20 Ω at 100 ℃ (based on 88 turns per phase, an average turn length of approximately 100 mm, and a copper cross-section of 1.005 mm² for two 0.8 mm diameter enameled wires wound in parallel). Due to the use of a surface-mounted permanent magnet rotor, the direct-axis inductance and quadrature-axis inductance of a single three-phase winding are approximately equal. mH. Since the first winding segment and the second winding segment are wound on the same stator tooth 101a and linked to a shared main magnetic flux, there is a non-negligible mutual inductance between the two sets of three-phase windings. The measured equivalent mutual inductance in this embodiment is... mH corresponds to a coupling coefficient of approximately 0.75, and the corresponding common-mode equivalent inductance of the two windings is... mH, differential-mode equivalent inductance is mH. The permanent magnet flux linkage of a single three-phase winding at 20 ℃ is Wb, this value is determined by the polar area m², the peak air gap magnetic flux density of formula (2) is 0.966 T, the pole arc coefficient is 0.85, and the winding coefficient of 12 slots and 10 poles is 0.933. After adjusting for 88 turns per phase and taking into account about 8% leakage flux, the torque coefficient of a single three-phase winding is N·m / A (based on the peak value of the quadrature axis current); since equation (28) enables both sets of three-phase windings to operate simultaneously, the gain from the total quadrature axis current command to the electromagnetic torque is Based on this, the quadrature axis current command for each set of three-phase windings at the rated point is approximately 11.9 A, with a peak value (effective phase current) of 8.4 A, corresponding to a conductor current density of approximately 8.4 A / mm². The non-magnetic bracket 101b also serves as a heat dissipation channel and is verified according to the intermittent duty cycle of the servo motor. The equivalent moment of inertia converted to the motor shaft is... kg·m², transmission ratio of servo motor reduction mechanism Voltage margin is calculated using the following formula: the amplitude of the back electromotive force of a single unit at the rated point is... V, included V and The required phase voltage amplitude after V is approximately 26.5 V, while the phase voltage amplitude that can be output in the linear region of space vector pulse width modulation under a 100 V bus is [missing value]. The voltage rating is V, providing ample margin. At 1800 r / min, the back electromotive force amplitude is 42.4 V, and even at full torque operation, the phase voltage amplitude required does not exceed 46.2 V, allowing the maximum speed to be reached without field weakening. The line voltage amplitude at this speed is 73.4 V, which is 100 V lower than the bus voltage. Therefore, when power output is blocked, the two sets of anti-parallel diodes in the three-phase full-bridge circuit will not conduct, eliminating the risk of uncontrolled rectification raising the bus voltage. The 200 V withstand voltage of the selected power devices provides sufficient margin.

[0050] As shown in Figure 2, the servo drive system 200, which is matched with the motor body 100, uses the main controller 213 as its computing core. Its hardware channels include: an inverter power module 201, consisting of two independent three-phase full-bridge circuits, which drive the first and second three-phase windings of the dual three-phase windings 101d respectively; and a current sampling module 202, used to collect the phase currents of the two three-phase windings. , , , , , Bus voltage and temperature acquisition module 203 is used to acquire DC bus voltage. The power device case temperature and the winding temperature output by temperature sensor 108; Hall signal acquisition module 204, used to acquire discrete signals output by three switching Hall elements fixed on the intermediate stator 101. , , After the signals from the aforementioned hardware channels enter the main controller 213, they are processed sequentially by various functional modules implemented in pure software. These include a coordinate transformation module 205, a high-resolution position estimation module 206, a permanent magnet flux adaptive observer module 207, a dual air gap asymmetry identification module 208, a torque pulsation feedforward compensation module 209, a commutation angle and power distribution adaptive correction module 210, a three-loop active disturbance rejection control module 211, and a space vector pulse width modulation output module 212. The high-resolution position estimation module 206 internally includes a vector tracking observer 2061 and a phase-locked loop 2062. The three-loop active disturbance rejection control module 211 internally includes a position loop 2111, a velocity loop 2112, a current loop 2113, an extended state observer 2114, and a feedforward loop 2115. The communication module 214 provides a fieldbus interface for receiving servo position commands from the host computer. The system receives the original absolute position feedback reading from the servo motor output and transmits status information such as rotor position, speed, bus voltage, winding temperature, estimated permanent magnet temperature, estimated dual air gap asymmetry, and fault codes. Apart from the independent lead interface of the dual three-phase windings 101d on the motor body 100 for load compensation, the servo drive system 200 does not add a high-resolution position sensor to the motor side, nor does it add any air gap displacement or force sensors.

[0051] The two sets of three-phase full-bridge inverter power modules 201 use silicon-based field-effect transistors with a withstand voltage of 200 V and an on-resistance of 8 mΩ. Silicon carbide devices can also be used to further reduce switching losses. The gates are driven by isolated drivers with Miller clamps, and the bridge arms have interlocked dead zones. Set to 500 ns, the switching frequency and the pulse width modulation carrier frequency All frequencies are 20kHz, corresponding to a control cycle of 50 μs. Each bridge arm is equipped with saturation voltage drop detection type short-circuit protection. The DC sides of the two sets of three-phase full bridges are connected in parallel to the same bus capacitor bank. The bus capacitor consists of a low-impedance electrolytic capacitor and a film capacitor connected in parallel to withstand the ripple current of the two sets of three-phase full bridges superimposed. The current sampling module 202 has a 0.5 mΩ shunt resistor connected in series in each of the three lower bridge arms of each set of three-phase full bridges. It fixedly collects two paths, phase a and phase b, which are sent to the 12-bit analog-to-digital converter built into the main controller 213 after differential amplification. The phase c current is collected by... Reconstruction; since both sets of three-phase windings are star-connected and neutral-point isolated, there is no zero-sequence path, so the reconstruction is strictly valid. Each set of windings occupies 2 sampling channels, for a total of 4 channels to obtain 6-phase current. The range is set to ±50 A, taking into account the quadrature-axis current of approximately 35.6 A at peak torque, with a quantization resolution of approximately 24.4 mA. The trigger time of the analog-to-digital conversion is synchronized with the trough point of the pulse width modulation carrier of this three-phase full-bridge. At this time, all lower bridge arms of this set are conducting, and the complete phase current flows through the shunt resistor. Since the two sets of carrier phases are staggered by 180° as described later, the trough points of the two sets of carriers do not coincide. Therefore, the current sampling module 202 triggers the two sets of three-phase windings independently, with a time interval of 25 μs between the two triggers; at rated speed, this time difference corresponds to approximately 0.75 electrical angles. When using the sampling results of the second set, the main controller 213 presses... After performing the corresponding phase shift correction, it is then used in the differential operation to eliminate the influence of inconsistent sampling times on the following equation (13). The bus voltage and temperature acquisition module 203 detects the bus voltage by resistive voltage division, and the sampling rate is synchronized with the current sampling; the winding temperature and shell temperature are acquired at a low rate of 100 Hz for over-temperature protection and parameter rationality verification.

[0052] Channel consistency calibration is performed in two steps, which is crucial for differential back EMF identification discussed later. Zero-point bias is calibrated during the power-on self-test phase under power output lockout, where the current in each phase is constant at zero. The zero point of each channel is obtained by averaging the data over several consecutive cycles. Gain consistency cannot be calibrated under unexcited conditions, so it is performed during the rotation phase of the power-on self-test: the main controller 213 alternately applies direct-axis current commands of the same amplitude to the two sets of three-phase windings, compares the readings of the two sampling channels with the bus current and input power for consistency verification, and calculates the relative gain error between the two sets of channels. Alternatively, during production, calibration can be performed unit-by-unit using an external reference current source and written to the non-volatile memory area. After online subtraction of the zero-point and relative gain corrections, the residual inconsistency between the two sets of sampling channels is better than 0.1%.

[0053] Three switch-type Hall elements are arranged circumferentially on the non-magnetic support 101b and aligned with the first permanent magnet array 102a on one side of the first rotor 102. Their spatial position makes... , , The states are 120 electrical degrees apart, resulting in six valid state combinations within each electrical cycle. The boundaries between adjacent states are known integer multiples of 60 electrical degrees. The Hall signal acquisition module 204 first performs hardware low-pass filtering on the three signals, then filters out jitter using a digital debounce window twice the control cycle. Simultaneously, the acquisition unit of the main controller 213 records the time and direction of each state flip, with a timestamp resolution on the order of 10 ns. The six state combinations uniquely determine the 60 electrical degree sector where the rotor's electrical angle is located, and the center of the sector is taken as the coarse position. The average electrical angular velocity is calculated by comparing the time interval between two adjacent flips with the known 60 electrical angle interval. Obviously, relying solely on the Hall signal acquisition module 204 can only obtain a position resolution of 60 electrical angles. During startup and extremely low speed phases, the instantaneous position error can reach a maximum of approximately ±30 electrical angles. Therefore, its output is only used as the absolute phase reference and initialization condition for the high-resolution position estimation module 206.

[0054] Since the Hall signal only provides electrical angle information modulo 2π, while To repeat the Hall state sequence 5 times within a mechanical cycle, a mechanical angle reference must be established; otherwise, position loop 2111 cannot achieve absolute positioning, and the angle domain compensation table in equation (25) cannot be aligned with the actual mechanical angle after a power failure and restart. This embodiment provides three alternative approaches that can be chosen or used in combination. First, the servo motor usually has existing absolute position feedback at the output end, such as a rotary potentiometer or a rotary variable differential transformer. Its reading is connected to the main controller 213 via communication module 214 or an independent acquisition channel. The mechanical angle and number of revolutions of the motor shaft can be determined by calculation, thus forming a dual-reference structure combining coarse and fine references with the high-resolution position estimation module 206. This feedback is an existing configuration of the servo system and is not a sensor added in this invention. Secondly, in the absence of output feedback, a zero-return process is executed upon power-up, driving the actuator to approach the mechanical limit or a known zero position at low speed before clearing the count. Thirdly, in the absence of output feedback and inconvenient zero-return, the Hall flip interval sequence fingerprint formed by the inherent slight differences in pole arc width and magnetization intensity of the 10 permanent magnets in the first permanent magnet array 102a is used. The currently measured interval sequence is cyclically correlated with the fingerprint template stored in the non-volatile storage area. The position of the correlation peak gives the current pole pair number. The determination can usually be completed within a few mechanical cycles, thus establishing a mechanical angle reference of modulus 2π without adding any sensors. The fingerprint template is built up using the average value of multiple mechanical cycles during factory calibration or first run, and is slowly updated during operation to adapt to changes in the permanent magnet temperature.

[0055] Coordinate transformation module 205 performs Clarke and Park transformations on the phase currents of the two sets of three-phase windings, respectively. (The last sentence appears to be incomplete and requires further context.) Taking a three-phase winding as an example, the Clarke transform, according to the equal amplitude convention, is as follows:

[0056] (5)

[0057] Parker transforms into

[0058] (6)

[0059] The first set of three-phase windings is taken The second set of three-phase windings takes This is used to compensate for the spatial phase shift of the second three-phase winding, which lags behind by 30 electrical degrees; if the second three-phase winding is found to be leading after assembly, then... The symbol is determined once as described above. The control power angle is given by the commutation angle and power distribution adaptive correction module 210. The coordinate transformation module 205 simultaneously uses the duty cycle actually issued by the space vector pulse width modulation output module 212 in the previous control cycle. , , With bus voltage Reconfiguration terminal voltage:

[0060] (7)

[0061] And superimposed voltage error correction caused by dead zone effect

[0062] (8)

[0063] Substituting 500 ns and 20 kHz into this embodiment, the single-phase correction is approximately 1.0 V. When the phase current is close to zero, the sign function is replaced by the sign of the current hysteresis or the current command to avoid chattering. The sum of equations (7) and (8) is obtained by successively transforming them through equations (5) and (6). , Therefore, no additional voltage sensor is needed to provide voltage information for back EMF and flux linkage estimation. It should be noted that the correction amount of equation (8) is much larger than the differential observation itself in the following text, and its accuracy seems to constitute a bottleneck; however, since the current commands of the two sets of three-phase windings in the identification window in the following text are exactly the same, the dead zone voltage error, device on-state voltage drop and shunt resistance error are common mode for the two sets of windings, and cancel each other out in the differential operation. The only thing that really affects the differential accuracy is the relative inconsistency between the two sets of channels, which has been reduced to less than 0.1% by the aforementioned consistency calibration.

[0064] The dual-air-gap asymmetry identification module 208 is a prerequisite for realizing online compensation of air-gap asymmetry in this embodiment. As shown in equation (2), the main magnetic flux in the yokeless segmented armature structure depends only on... Therefore, no matter how the three-phase combined electrical quantities are processed, the unequal distribution of the air gaps on both sides cannot be detected. This is the fundamental reason why conventional single-set three-phase winding schemes and conventional back EMF observation methods are almost invisible to the asymmetry of the double-sided air gaps. It is also the fundamental reason why this embodiment segments the concentrated winding 101c along the axial direction to form a double three-phase winding 101d. In addition to linking the common main magnetic flux, the axially offset first winding segment and second winding segment also link the edge magnetic flux entering from the side of the stator tooth 101a, respectively. The component of this edge magnetic flux decreases as the length of the adjacent air gap increases. It has a non-zero first-order sensitivity, the proportion of which is the decoupling sensitivity coefficient introduced in equation (4). Characterization.

[0065] Since the two sets of three-phase windings are wound on the same stator tooth 101a and share the same axial main magnetic circuit, their mutual inductance... Since the mutual inductance is of the same order of magnitude as the self-inductance, the back electromotive force estimation must include the mutual inductance term; otherwise, the differential current applied by the compensation itself will produce spurious differential components in the observations. After including the mutual inductance, the... The voltage equation of the three-phase winding in the synchronous rotating coordinate system is as follows:

[0066] (9)

[0067] In the formula This represents another three-phase winding. Based on this, an extended back electromotive force estimate is constructed.

[0068] (10)

[0069] (11)

[0070] The current differential term is negligible within the steady-state window, but is implemented using a band-limited differentiator during the dynamic process. The temperature coefficient of copper resistance is corrected according to the winding temperature. Equation (11) is the equivalent flux linkage amplitude obtained after normalization of rotational speed. It does not change with rotational speed, making it convenient to compare the differences between the two windings over a wide speed range. The equation (10) is... Further subtraction This yields the residual amount reflecting the model mismatch, which is used by the permanent magnet flux adaptive observer module 207 and the vector tracking observer 2061. It is not subtracted here to retain the complete back EMF amplitude information.

[0071] Considering Only on the order of 0.09 The relative difference component corresponding to the compensation start threshold is only about 0.5%, and the instantaneous value cannot obtain a usable signal-to-noise ratio. Therefore, the rotor electrical angle is first used as the phase reference for continuous... One electrical cycle The average value is calculated by accumulating the values ​​of grids with the same electrical angle.

[0072] (12)

[0073] Take 32-128, random noise according to Attenuation, in this embodiment, is set to 64, and the equivalent random noise is reduced to The value is approximately 0.002, and the cost is only a reduction in the identification bandwidth to a few Hz, which is perfectly compatible with the timescale of thermal deformation and assembly drift. The back EMF asymmetry observation is constructed using the synchronous average value of Equation (12):

[0074] (13)

[0075] Wherever back electromotive force differential calculations occur in the following text, the synchronous average value of equation (12) shall be used. For input. The first air gap 104 corresponding to the first set of three-phase windings is relatively large and the edge magnetic flux is relatively small, therefore Equation (13) is positive, and Consistent. When the intermediate stator 101 is replaced with a yoke structure with a stator yoke, the magnetic circuits on both sides are closed along the circumference through the stator yoke and are approximately independent of each other. The magnetic flux density of the air gap on one side degenerates into a form that is only related to the length of the air gap on that side. A first-order expansion of equation (13) yields...

[0076] (14)

[0077] Substituting the parameters into this embodiment yields That is, a yoke structure can achieve approximately twice the sensitivity; if we further neglect the magnetic reluctance of the permanent magnet and assume that the air gap permeability is inversely proportional to the air gap length, then... As it approaches 1, equation (13) degenerates into a pair. Direct measurement, at this time there is . The calibration was completed during the prototype assembly stage. The method involved adding adjusting shims with thicknesses of 0.025 mm, 0.05 mm, and 0.1 mm between the bearing housing and the end cover, artificially creating several sets of known... Record the corresponding values ​​at several speeds and load points. And perform least squares fitting, the result The zero-point bias is stored together with the non-volatile memory area of ​​the main controller 213.

[0078] To cut off the contamination path of the compensation current through the mutual inductance to the differential observations, the main controller 213 sets the identification window according to the timing shown in Figure 7. With a period of 1 second, approximately 20% of the time is dedicated to the identification window. Within this window, the current weights of the two sets of three-phase windings are equal, and the differential mode direct-axis current is zero. , The window boundary transitions with a ramp of not less than 20 ms to avoid torque step; the synchronous average of equation (12) accumulates only within the identification window, and accumulates until full. Output once after one electrical cycle At the rated speed, it takes about 3.8 s. The current commands of the two sets of three-phase windings within the window are exactly the same, so the mutual inductance terms in equation (10) are strictly equal for the two sets of windings and cancel each other out in the difference in equation (13). At the same time, the dead zone, device voltage drop and shunt resistance error are also canceled out as common modulus, leaving only the permanent magnet flux difference to be extracted by equation (13). If no dedicated window is set, it can also be directly calculated and deducted from the model of equation (10). The mutual inductance contribution caused by the weight difference, but this deducts the dependency. The accuracy is improved; this embodiment mainly uses the identification window and secondarily uses model subtraction, using both methods in combination. It should be emphasized that only the differential excitation between the two sets of windings is set to zero within the identification window, and harmonic injection and iterative learning compensation still operate normally, so the introduction of the window will not cause any interruption in compensation.

[0079] Considering that type (13) may still fail under low speed and light load conditions, and that the end face runout and disc deformation of the rotor disk will cause the air gap to be unevenly distributed along the circumference, the dual air gap asymmetry identification module 208 sets up an additional identification channel based on current harmonic characteristics. The input of this channel is the arithmetic mean of the cross-axis currents of the two sets of three-phase windings. Since torque ripple is mainly mapped to this common-mode component, each harmonic component is separated using an adaptive harmonic extraction method, and its recursive form is:

[0080] (15)

[0081] (16)

[0082] In the formula To extract the gain, its value is chosen such that the extraction time constant for each harmonic component is approximately 10 electrical cycles; the extraction of the mechanical subcomponent only requires... Change to As an equivalent replacement, a second-order generalized integrator can also be used to achieve the same function, and its state equation is:

[0083] (17)

[0084] In the formula The angular frequency of the harmonic to be extracted, and the damping coefficient. A value of 1.414 is chosen to balance selectivity and response speed. Based on this, the current harmonic characteristic quantity is constructed.

[0085] (18)

[0086] In the formula This represents the DC value of the common-mode component of the quadrature-axis current. To prevent the minimum current threshold from being divided by zero, this embodiment uses 10% of the rated current. When the output falls below this threshold, the channel output is deemed invalid and frozen. The proportionality coefficient determined by the gasket calibration is taken as 0.8 in this embodiment. Since the harmonic amplitude itself only reflects the severity of asymmetry and does not contain directional information, equation (18) borrows... The sign is determined The direction; in During the unavailable startup phase, the initial value is temporarily set to the symbol saved before the last shutdown, and will be corrected after entering the first identification window. The two channels thus complement each other: the differential back EMF channel has high accuracy and contains symbol information under medium- and high-speed steady-state conditions, while the current harmonic channel remains effective under low-speed and dynamic conditions and is more sensitive to non-uniform air gap distribution around the circumference.

[0087] The final dual-air-gap asymmetry is obtained by weighted fusion of two observations and then filtered:

[0088] (19)

[0089] (20)

[0090] Among them, the weighting coefficient The system is adjusted online according to operating conditions. In this embodiment, a value of 0.85 is used when the rotational speed is higher than 40% of the rated speed and the common-mode cross-axis current is higher than 30% of the rated value. A value of 0 is used when the rotational speed is lower than 15% of the minimum identification speed, i.e., the rated speed, to ensure that outdated systems that have stopped being updated are properly maintained. No longer participating in fusion, the intermediate section transitions using two-dimensional interpolation of rotational speed and current; low-pass coefficient Taking values ​​from 0.01 to 0.05, the equivalent time constant at a refresh rate of 500 Hz corresponds to 0.04 to 0.2 s, which is much larger than the current loop period but much smaller than the time scale of thermal deformation. As an equivalent replacement, the first-order low-pass filter in equation (20) can also be replaced by a Kalman filter with asymmetry and its rate of change in state, whose state equation and measurement equation are as follows:

[0091] (twenty one)

[0092] In the formula To identify the update cycle, process noise covariance The maximum rate of change of asymmetry caused by thermal deformation is set to approximately 0.01 / min. The noise covariance is then measured. The measured variance is set according to formula (19); this form can achieve more stable operation and smaller phase lag when the load and speed change frequently. Alternatively, a recursive least squares estimator with a forgetting factor can be used to directly identify the parameters of equations (13) and (18). In summary, The steady-state fluctuations consist of three parts: The random noise after synchronous averaging is approximately 0.002, the systematic residual after calibration between the two sets of sampling and driving channels is approximately 0.005, and the residual after common-mode cancellation of dead zone and device voltage drop within the equal current identification window is approximately 0.003, which is better than 0.01 in total.

[0093] The dual air gap asymmetry identification module 208 performs both judgment and protection functions. A compensation start threshold can be set. With alarm threshold In this embodiment, the values ​​are 0.03 and 0.15 respectively. The value of 0.03 is chosen to achieve a higher estimated noise level than the aforementioned 0.01 without being overly sluggish; this corresponds to a 0.06 mm difference in air gap between the two sides. Accordingly, the initial assembly stage requires adjusting the shims to ensure proper fit. The asymmetry can be reduced to within 0.01 mm, while the initial asymmetry, determined solely by the cumulative tolerance of the axial dimension chain of ±0.06 mm without selection, can reach 0.06 mm. Therefore, gasket selection is crucial. The necessary assembly process to distinguish between normal and abnormal processes. Take 0.15, at this time and The air gaps are 1.15 mm and 0.85 mm respectively. The smaller side still has 0.85 mm of air gap remaining, which is close to the risk boundary of rubbing due to the combined effects of thermal deformation and rotor disc deformation. When For several consecutive update cycles below When it is determined that the air gaps on both sides are basically symmetrical, the model injection term of equation (24) will change accordingly. The iterative learning table in equation (25) is continuously decayed to zero, while it is frozen and no longer updated to avoid continuously rewriting the compensation table due to noise when there is no asymmetry; when Exceed Collaborative compensation is initiated at the appropriate time, and the model-injected items follow accordingly. Continuous changes, iterative learning and recovery updates, and setting hysteresis near the threshold to avoid frequent start-stop; when Exceed At that time, the main controller 213 limits the torque command amplitude at a slope of 10% / s and limits the maximum speed to 60% of the rated speed. Simultaneously, it reports fault words and real-time information via the communication module 214. . The long-term trend is stored at a density of 1 point / h in the non-volatile storage area, which can serve as a basis for assembly quality assessment, bearing 107 wear monitoring, and predictive maintenance; if If a step change occurs within a short period of time, it is determined to be a mechanical impact or loose fasteners and should be reported immediately.

[0094] Torque ripple feedforward compensation module 209 The rotor position output from the high-resolution position estimation module 206 is used as input to implement periodic compensation that is strictly synchronized with the rotor position. The pulsating torque is modeled as...

[0095] (twenty two)

[0096] The first summation term is the electrical angle domain component. The corresponding equivalent flux axis offset and commutation phase deviation. Corresponding to the failure of flux linkage harmonic cancellation and non-sinusoidal flux linkage, This corresponds to the inherent cogging component of a 12-slot, 10-pole configuration; the latter summation term represents the mechanical angular domain component, corresponding to rotor disk end-face runout and disc deformation. Within the small asymmetry range of engineering interest, each magnitude can be approximated as a linear function of the asymmetry degree.

[0097] (twenty three)

[0098] In the formula and This represents the inherent pulsation amplitude under symmetrical conditions. and The asymmetric sensitivity coefficient, and This represents the corresponding phase. The coefficient obtained from the gasket calibration in this embodiment, expressed in N·m, is: , ; , ; , ; , ; , ; , The aforementioned coefficients and phases can be obtained through offline calibration or implicitly corrected through the iterative learning process described later; because Since there is an estimated residual of approximately 0.01, the model injection amount given by equation (23) will inevitably have a corresponding proportional deviation. This deviation is absorbed by the iterative learning term, and the two constitute the division of labor between the amplitude main body and the residual correction. The harmonic injection feedforward current based on the model is

[0099] (twenty four)

[0100] The denominator in the formula uses the torque gain when two sets of three-phase windings work together. This is consistent with the current distribution in equation (28); The common-mode value of the permanent magnet flux linkage, estimated online by the permanent magnet flux linkage adaptive observer module 207, is automatically factored into the compensation gain as its temperature changes. An additional phase lead proportional to the rotational speed is added to the phase of each cosine term to compensate for the phase lag of the current loop 2113. This lead is calculated based on the closed-loop bandwidth of the current loop and interpolated using a table.

[0101] Considering that the models upon which equations (22) to (24) depend cannot be completely accurate, the torque ripple feedforward compensation module 209 introduces an iterative learning stage in parallel with harmonic injection. Using rotor mechanical angles rather than time as independent variables, a mechanical cycle is discretized into... An equally spaced angular grid, i.e., an angular resolution of 1°, is due to Each electrical cycle corresponds to 72 grids, which can characterize up to the 36th electrical angle harmonics without distortion. This covers the 1st to 12th electrical angle components in equation (22), as well as the 1st and 2nd mechanical components. The absolute phase of the grid is determined by the aforementioned mechanical angle reference, thus allowing the compensation table to be aligned with the actual mechanical angle across electrical cycles. The compensation amount of each grid point is stored in a ring buffer, and the grids are interpolated linearly, with the compensation amount following the mechanical cycle number. Iterative updates are performed using the following formula:

[0102] (25)

[0103] In the formula The zero-phase low-pass filter with forgetting factor uses forward and backward filtering to eliminate phase distortion. Its cutoff frequency corresponds to the 36th electrical angle harmonic. The forgetting factor is 0.95 to 0.99. The closer the value is to 1, the smaller the final convergence residual and the weaker the robustness to model perturbation. Let be the learning gain including phase lead correction. To make equation (25) dimensionally closed and... The value of is independent of the source of the error signal; the error term The normalization is based on the rated torque, and the compensation is based on the rated current. Since it is a dimensionless quantity, this embodiment uses a constant gain of 0.2 to 0.4 and connects it in series with a first-order lead element. The three optional error sources are converted into normalized torque errors according to their respective physical gains: the total disturbance estimate is taken from the extended state observer 2114. When the periodic component is in the middle, multiply by Converted to N·m and divided by the rated torque; when taking the AC component of the common-mode quadrature-axis current, multiply by Conversion; when taking the angular domain waveform from the speed fluctuation, first perform angular domain differentiation and multiply by Conversion. Convergence condition is:

[0104] (26)

[0105] In the formula This is an equivalent discrete path from normalized quadrature-axis current command to normalized torque ripple, including the closed-loop transfer function of current loop 2113, the equivalent delay of pulse width modulation, and the mechanical object. The iterative learning stage only occurs when the speed fluctuation is less than 2% of the rated speed and the load change rate is less than the set value. The compensation table is updated within a steady-state window with a small rate of change. Once a step change in speed command or a sudden load change is detected, the update is frozen, and the converged compensation table is maintained. Iteration continues after steady-state recovery, thus preventing aperiodic disturbances from being mistakenly learned into the periodic compensation table. Simultaneously, the amplitude of the compensation table is set to an upper limit of 15% of the rated current to prevent divergence during the learning process. The compensation table is written to a non-volatile memory area during shutdown and used as the initial value for iteration upon the next power-on. In this embodiment, this measure reduces the number of mechanical cycles required for convergence from approximately 300 cycles to approximately 40 cycles.

[0106] The adaptive correction module 210 for commutation angle and power distribution is responsible for... This is transformed into differentiated excitation between two sets of three-phase windings to balance the output on both sides and suppress net unbalanced axial force. The current distribution weights on both sides are defined.

[0107] (27)

[0108] (28)

[0109] As can be seen from equation (27), when That is, when the first air gap 104 is larger and the equivalent flux linkage of the first three-phase winding is smaller, the first three-phase winding receives a larger current weight, while the second three-phase winding with a smaller air gap and larger flux linkage receives a correspondingly smaller current; when symmetrical. The quadrature axis current command for both sets of three-phase windings is equal to This is consistent with the gain from total command to torque mentioned earlier. Consistent. Assigned gain. The value of depends on the optimization objective. When the objective is to achieve a strict balance between the electromagnetic torque contributions on both sides, let And by performing a first-order expansion on the small quantity, we can obtain When the goal is to minimize the net unbalanced axial force, since the quadrature-axis current also generates armature reaction flux that crosses the air gap in the axial flux motor, the mean square value of the combined air gap flux density on that side is increased, and the axial force on that side is increased according to equation (3). Therefore, the current on the side with the smaller air gap should be further reduced, and the required gain is... Greater than This embodiment has been calibrated through experiments. When the goal is to minimize total copper loss, the optimal allocation should make the current weight proportional to the torque coefficient of each side, which requires increasing the current on the side with the smaller air gap, in the opposite direction to the previous two goals. This conflict between goals can be quantitatively determined: the additional copper loss caused by deviating from the optimal allocation is proportional to the square of the deviation, which is... The second-order minor component, while the first-order component of the torque pulsation and the net unbalanced axial force are both Since it is a first-order quantity, torque balance and axial force suppression should be prioritized within the asymmetry range of engineering concern. Therefore, [the following is taken as an example].

[0110] (29)

[0111] And the weighting coefficient Based on operating conditions: Under heavy load and low speed conditions, the axial load and life of bearing 107 are the main contradictions. Approaching 1; under light-load, high-speed operating conditions, efficiency and temperature rise are the main contradictions, taking... Approaching 0. Equation (27) also sets a weight limit, which is limited in this embodiment. and All are between 0.45 and 0.55; this range is consistent with... , The determined adjustment range is adapted to the actual adjustment range and is only used as a safety clamp to prevent severe current bias when parameters are abnormal, and not as a fault criterion.

[0112] To more directly compress the net unbalanced axial force with almost no impact on the output torque, the commutation angle and power distribution adaptive correction module 210 also superimposes a differential-mode direct-axis current between the two sets of three-phase windings:

[0113] (30)

[0114] The side with the smaller air gap is demagnetized while the side with the larger air gap is magnetized. Their contributions to the combined electromagnetic torque approximately cancel each other out, and the direction of adjusting the magnetic flux density of the air gaps on both sides is exactly opposite to the direction of the axial force imbalance. Equation (30) is most effective in implementations where the magnetic circuits on both sides are relatively independent. In yokeless segmented armature structures, it works by changing the edge magnetic flux and tooth end saturation on each side, with the ratio of their effects also varying. Characterization. Gain The value of is constrained by the current margin of the two sets of three-phase windings, the additional copper loss, and the risk of permanent magnet demagnetization. In this embodiment, 25% of the rated current corresponding to the unit asymmetry is taken, that is, in When a differential-mode direct-axis current of approximately 0.30 A is applied, it is estimated, according to equations (3) and (4), that approximately 20% of the net unbalanced axial force can be additionally offset beyond the effect of equation (27). It should be noted that the differential-mode direct-axis current, through the mutual inductance of the two windings, will introduce a current in equation (10) that is related to... The spurious difference components with the same sign are of similar magnitude to the true difference components. If left untreated, they will form a positive feedback loop in the asymmetry estimation. This embodiment uses the aforementioned identification window to... Furthermore, model subtraction was employed to sever this feedback path from a mechanistic perspective.

[0115] Air gap asymmetry causes an offset in the equivalent commutation phase, and the adaptive correction module 210 for commutation angle and power distribution provides a commutation angle correction accordingly.

[0116] (31)

[0117] in The sign and value are determined by calibration. In this embodiment, the value is taken as 5 electrical angles corresponding to the unit asymmetry. It should be noted that the installation position deviation of the Hall element is not compensated here, but is only included once in the Hall boundary correction of equation (36), so as to avoid the same error being applied repeatedly. Equation (31) gives It is only used for the Parker transformation and subsequent inverse Parker transformation in equation (6), and not for the position feedback of position loop 2111, so as to avoid the commutation phase correction from polluting the position control accuracy. The fault tolerance criterion of this module does not depend on whether the weight limit is touched, but is determined according to the following conditions: if the effective value of the tracking error of the current loop 2113 of a certain three-phase winding continuously exceeds 20% of its command amplitude, or the current sampling module 202 detects an open circuit and a phase loss, or the bridge arm short circuit protection is activated, or the ratio of the two sets of equivalent flux amplitudes calculated by equation (11) deviates from 1 by more than 20%, then the three-phase winding or its power channel is determined to be faulty, the corresponding three-phase full bridge is immediately blocked and the weight of the remaining three-phase winding is reset to 1, and the torque command upper limit is reduced to 40% of the peak torque and reported through the communication module 214 to achieve single-sided fault-tolerant operation; when operating on one side, the dual air gap asymmetry identification module 208 automatically switches to using only the current harmonic channel. Set to 0.

[0118] The high-resolution position estimation module 206 realizes position and speed estimation at the level of electrical angles based on the low-resolution Hall signal. Its estimation process consists of five stages: Hall coarse positioning and initialization, vector tracking observer 2061, phase-locked loop 2062, Hall boundary soft fusion correction, and parameter injection. During the power-on or start-up stage, the initial sector is determined by the 6-state combination given by the Hall signal acquisition module 204 and its center is taken as the initial position, thereby limiting the initial position error to within ±30 electrical angles and avoiding phase ambiguity and reversal of 180 electrical angles. Then, a short-term open-loop acceleration is performed according to the preset current amplitude and ramp frequency, or the torque command is applied according to the sector center position to make the rotor enter the closed loop. Both methods switch to full closed loop after the speed reaches the minimum identification speed. After the rotor starts to rotate, the extended back electromotive force component is obtained by the corresponding form of equation (10) in the stationary two-phase coordinate system.

[0119] (32)

[0120] The current differential term is implemented using a band-limited differentiator to suppress noise. The result for the second set of three-phase windings requires rotation first. The electrical angle is reduced to the stationary coordinate system of the first set of three-phase windings and then weighted and synthesized with the first set. The weights are determined according to the effective current values ​​and fault conditions of each winding. The synthesized result is denoted as... , The position error signal is taken as the negative value of the direct-axis back electromotive force in the estimated coordinate system, i.e.

[0121] (33)

[0122] In the formula To prevent division by zero of small positive numbers, the construction of equation (33) is based on the following: According to the Park transformation convention of equation (6), the unit vector of the rotor direct axis in the stationary coordinate system is The permanent magnet flux linkage vector is Taking the derivative with respect to time yields the back electromotive force vector. That is, located at the quadrature axis and leading the direct axis by 90 electrical degrees, which is consistent with the steady state in equation (10). and Consistent. Substituting the above components into the left side of equation (33) yields... It crosses zero at zero position error and has a positive slope, so it can be directly used as the phase detection output of a phase-locked loop. Divide by The normalization process ensures that the error channel gain remains constant regardless of speed and flux amplitude, thus achieving consistent closed-loop dynamics across the entire speed range. This is a key reason why the position estimation accuracy in this embodiment does not deteriorate with operating conditions. The normalized error is then fed into phase-locked loop 2062.

[0123] (34)

[0124] The open-loop transfer function of this phase-locked loop contains a double-integral element, exhibiting zero steady-state error for velocity step inputs (i.e., position ramp inputs), fundamentally overcoming the inherent phase lag of the average velocity method and interpolation method. The gain is designed according to the second-order canonical form.

[0125] (35)

[0126] Damping ratio Take 0.707, natural frequency according to Online tuning: Take 628 rad / s (i.e., 100 Hz) near the rated speed, corresponding to... , When the speed drops to the lowest identification speed The speed is linearly reduced to 150 rad / s to suppress back EMF estimation noise; when the speed increases to the maximum speed... Increased to 1200 rad / s and simultaneously checked that the phase margin is not less than 45°. Equation (34) is discretized using the backward Euler method, with an amplitude limit set for the integral element, and The integral is moduloed at 2π to avoid numerical overflow, and its mechanical cycle number is given by the aforementioned mechanical angle reference and full revolution count.

[0127] Since equation (34) contains an open-loop integral, a slow phase drift will inevitably occur during long-term operation. Therefore, whenever the Hall state flips, the absolute information that the rotor electrical angle is at the known 60° electrical angle boundary at that moment is used to... Correction is performed. To avoid step-like changes in position output that could excite mechanical vibration, a soft-blending method is used for correction:

[0128] (36)

[0129] In the formula This is the theoretical boundary phase corresponding to this flip. For the first The Hall installation error correction for each sector is calculated by the high-resolution position estimation module 206 by comparing the actual phase and theoretical phase of the six Hall flip boundaries during steady-speed operation, and then statistically analyzing each sector. This is achieved through a sliding average over hundreds of mechanical cycles and stored in a concurrent table, thereby simultaneously compensating for the unequal spacing error between the three Hall elements and the overall installation offset of the Hall elements relative to the first permanent magnet array 102a. This correction is applied only once at equation (36) and is not applied again in equation (31). Fusion coefficient The value is set to 0.02 to 0.2 and decreases as the back EMF amplitude increases. That is, the Hall boundary is trusted more at low speeds and the observer is trusted more at medium and high speeds. The single correction amount is set with an amplitude limit. When the limit is exceeded, it is judged as an abnormal Hall signal and the fault count is accumulated. If the limit is exceeded continuously, the Hall element is judged as a failure and the system is switched to pure observer operation and reported at the same time. The position between two flips is continuously interpolated by the vector tracking observer 2061 and the phase-locked loop 2062, so that the 6 discrete states in each electrical cycle are expanded to a continuous position with a resolution better than 0.1 electrical angle. Under the condition that the speed is extremely low or even stationary and the back EMF amplitude is insufficient to support equation (33), the position is mainly extrapolated by the Hall coarse position combined with the angle of the previous steady-state speed, and can be supplemented by a high-frequency voltage injection with an amplitude of 5% of the rated current and a frequency of 1 kHz to identify the position corresponding to the weak salient pole. As the back EMF amplitude increases, the weighting coefficient is adjusted according to the following formula. The transition to the observer output is continuous, with no hard switching of the estimation method throughout the process.

[0130] The parameter injection process demonstrates the synergy between the two inventions. The permanent magnet flux linkage parameters used in the Vector Tracker 2061 are not fixed constants; their common-mode reference value... The permanent magnet flux adaptive observer module 207 provides online parameters to track temperature drift, while the model parameters for the two sets of three-phase windings are output by the dual air gap asymmetry identification module 208. Make differentiated adjustments:

[0131] (37)

[0132] Substituting the parameters corrected by equation (37) into equations (9), (10), and (32), the extended back electromotive force estimates of the two sets of three-phase windings are matched with their actual flux linkages. The position error signal of equation (33) no longer includes the fixed bias and second-order fluctuations caused by air gap asymmetry. Therefore, the position estimation accuracy is not affected by air gap asymmetry and temperature rise. The inductance parameters of the two sets of three-phase windings can also be differentiated according to the form of equation (37), and the correction coefficients are determined by calibration. Conversely, the high-precision position and speed output by equations (33) and (34) serve as the harmonic extraction phase reference for equations (15) to (18), the harmonic injection phase reference for equation (24), and the angle grid index for equation (25), ensuring that the compensation current is strictly in phase with the actual pulsating torque. If there is an error in the phase reference, the injected compensation current will not only fail to cancel the pulsation, but will also be identified by the current harmonic channel as a new asymmetric quantity, causing mutual excitation between identification and compensation. Therefore, this bidirectional information exchange between the two invention points is a necessary condition for the stable operation of this embodiment, see Figure 4.

[0133] The permanent magnet flux adaptive observer module 207 uses a model reference adaptive structure to estimate the common-mode value of the permanent magnet flux of a single set of three-phase windings online. Its input is taken as the common-mode components of two sets of three-phase windings, i.e. , , , The corresponding equivalent inductance is taken as the common-mode value. and Since the differential-mode direct-axis currents in equation (30) have opposite signs, their common-mode value is always zero. Therefore, the input of this module is not affected by differential-mode excitation. The estimated This is the symmetrical reference value expanded to both sides of equation (37), and also the flux linkage parameter used in equations (24), (44), and (46). The voltage equation, composed of measured voltage and current, is used as the reference model, and the current equation containing the parameters to be estimated is used as the adjustable model:

[0134] (38)

[0135] Under steady-state conditions, a coarse estimate of the flux linkage can be directly obtained from the quadrature-axis voltage equation.

[0136] (39)

[0137] Equation (39) is extremely sensitive to noise and resistance mismatch at low speeds because the denominator approaches zero. Therefore, it is only used for checking the initial power-on value and its rationality. In practice, a proportional-integral adaptive law based on the Popov hyperstability criterion is used:

[0138] (40)

[0139] in and For adaptive gain, its value is chosen to ensure that the equivalent bandwidth of the flux linkage estimation is 2–5 Hz, which is sufficient to track the rate of change of the permanent magnet temperature while being far below the bandwidth of the current loop to avoid interaction with it. To mitigate the estimation bias caused by the mismatch of the quadrature-axis inductor, the permanent magnet flux linkage and the quadrature-axis inductor are jointly estimated within a medium-to-high-speed steady-state window. The update law of the quadrature-axis inductor is...

[0140] (41)

[0141] Its adaptive bandwidth is less than 1 / 5 of the estimated flux linkage bandwidth, allowing the two parameters to be decoupled and updated in the frequency domain. Furthermore, considering that the contributions of permanent magnet flux linkage and direct-axis inductance to the same residual in the quadrature-axis voltage equation are difficult to distinguish, a common-mode negative direct-axis current with an amplitude of 10%–15% of the rated current is periodically injected within the identification window, and two sets of steady-state equations are collected before and after the injection.

[0142] (42)

[0143] The permanent magnet flux linkage and the direct-axis inductance can be solved separately. The first equation of equation (42) is based on the premise that the common-mode direct-axis current is zero before injection, which is precisely the premise of the identification window. This is ensured by both the common-mode direct-axis command being zero; the acquisition of both sets of data must be completed within the same window, and the period is required to be... and The change must not exceed 1%; otherwise, the result is invalid. The obtained results are used to correct the convergence point of the online estimation. The average temperature of the permanent magnet can be inferred from the estimated magnetic flux linkage.

[0144] (43)

[0145] in Reference temperature The permanent magnet chain below, The remanence temperature coefficient of the permanent magnet is given. For neodymium iron boron permanent magnets, this coefficient can be between -0.08% / ℃ and -0.12% / ℃ depending on the grade. In this embodiment, -0.12% / ℃ is taken as a conservative value. In practice, the permanent magnet grade should be calibrated one by one in a constant temperature chamber. Otherwise, the temperature back-calculation of equation (43) will have a systematic deviation. Under thermal equilibrium steady state, the temperature rise of the permanent magnet is close to that of the winding. Therefore, when the winding temperature rises by 80 K, the permanent magnet flux linkage drops from about 0.045 Wb to 0.041 Wb. If no compensation is made, the torque coefficient will deviate by about 9.6%, and the position observation model will also be mismatched. During the transient process, the temperature rise of the two is not equal. At this time, the estimated value of equation (43) should be used instead of the reading of temperature sensor 108. The permanent magnet temperature obtained by equation (43) is compared with the winding temperature measured by temperature sensor 108. If the deviation between the two exceeds the set range, it is determined that the flux estimation is abnormal or the temperature sensor 108 is malfunctioning. The flux estimation is then switched to the open-loop value of the winding temperature table and reported through the communication module 214. When the permanent magnet temperature estimation value is close to the highest working temperature of the selected grade, the main controller 213 limits the current amplitude according to the preset curve to prevent irreversible demagnetization.

[0146] The three-loop active disturbance rejection control module 211 constitutes the main control loop of the servo system. Its outer loop is the position loop 2111, the middle loop is the speed loop 2112, and the inner loop is the current loop 2113. Together with the extended state observer 2114 and the feedforward link 2115, it forms the active disturbance rejection structure. The communication module 214 issues position commands. The angle command from the servo motor output is, after being limited and smoothed by a 7th-order polynomial trajectory, converted into a motor shaft mechanical angle command according to the transmission ratio. Correspondingly, the velocity feedforward and acceleration feedforward provided by the feedforward link 2115 are also based on the motor shaft and are on the same scale as the mechanical quantities of the motor shaft tracked by the extended state observer 2114:

[0147] (44)

[0148] In the formula The gain of the acceleration feedforward is calculated to include the equivalent moment of inertia of the motor shaft after the load is converted according to the transmission ratio. Online updates are implemented to maintain feedforward accuracy during temperature rise. The path from the quadrature-axis current to the motor shaft mechanical angle position is treated as a second-order object, and a third-order extended state observer 2114 is constructed:

[0149] (45)

[0150] In the formula Combined with the mechanical angle reference and the full circle count, it is expanded into a continuous absolute mechanical angle, state Tracking location, Tracking rotation speed, The total disturbance estimate is denoted as . To distinguish it from the back electromotive force sign of equation (10), the input gain is

[0151] (46)

[0152] Substituting the parameters into this embodiment yields And its follow Online updates ensure that the disturbance compensation amount does not exhibit systematic deviations during temperature rise. The observer gain is configured according to a single bandwidth parameter.

[0153] (47)

[0154] This embodiment takes rad / s, or 238.7 Hz, corresponds to three gains of 4500, ... and Equation (45) is discretized at a rate of 4 kHz using the forward Euler method. When the gain is configured according to equation (47), the characteristic polynomial of the forward Euler discretization is: Its stability boundary is 0.375 is an engineering value that balances noise amplification and phase margin, and is not the limit required for stability. An amplitude limit is set, with its upper limit corresponding to the peak torque, to prevent disturbance estimation divergence in case of position sensing abnormalities or mechanical jamming. The control law adopts a combination of state error feedback and disturbance compensation. The basic command is first given by the active disturbance rejection structure, and then periodic compensation is superimposed.

[0155] (48)

[0156] (49)

[0157] The position loop gain is calculated based on the closed-loop bandwidth. Configured as , ,Pick rad / s, which is 30 Hz, with a damping ratio of 1, yields , The disturbance compensation term in equation (48) makes the controlled object approximately a pure second-order integral element after compensation, thus making the gain configuration of position loop 2111 independent of the precise knowledge of load inertia and friction characteristics. Equation (49) is the final total quadrature-axis current command, which is then distributed to the two sets of three-phase windings according to equation (28). There are no other forms of quadrature-axis command synthesis formulas in the entire text. This demonstrates the division of labor and complementarity between the two inventions in disturbance handling: the extended state observer 2114 is responsible for estimating and compensating for non-periodic or broadband disturbances such as frictional changes, load abrupt changes, inertial perturbations, and residual parameter mismatches, while the periodic pulsations that are strictly synchronized with the rotor position are handled by the torque pulsation feedforward compensation module 209 in the angular domain; if this division of labor is not made, the bandwidth of the extended state observer 2114 must be increased to several thousand rad / s in order to track the periodic disturbances from the first mechanical angle to the twelfth electrical angle. The resulting high-frequency noise amplification will exacerbate current command jitter, increase power device losses, and make it more prone to instability when there is quantization noise in the position estimation. The two share the same set of position and speed estimations and the same set of flux linkage parameters, thus maintaining consistency within the same model framework. The speed loop 2112 degenerates into the differential feedback term in equation (48) in pure position servo mode, and is implemented as a proportional-integral regulator with anti-integral saturation when independent speed limiting or speed control mode is required. Its gain is configured according to a closed-loop bandwidth of 150 Hz.

[0158] The current loop 2113 provides independent closed-loop control for two sets of three-phase windings. Each set includes two proportional-integral controllers (one direct-axis and one quadrature-axis) with superimposed cross-coupling terms for feedforward decoupling. The feedforward quantity is... and The flux linkage term is taken according to equation (37). Since there is strong mutual inductance between the two sets of three-phase windings, the equivalent inductance of their common-mode and differential-mode channels differs by nearly an order of magnitude. Therefore, the current loop 2113 adopts a common-mode and differential-mode decoupling tuning method: first, the current commands and feedback of the two sets are transformed into common-mode and differential-mode quantities, and the common-mode channel is tuned according to the equivalent inductance. mH setting, differential channel according to The mH setting is adjusted and then converted back into two sets of voltage commands. The closed-loop bandwidth is 1.5 kHz. rad / s, tuned using the zero-pole placement method, common-mode channel proportional gain V / A, differential channel V / A, the integral gain of both channels is V / A. V / (A·s); If common-mode and differential-mode decoupling is not performed and the two sets are tuned separately, then it must be done according to Gain is taken to ensure the stability of the differential-mode channel, at the cost of a significant reduction in the bandwidth of the common-mode channel. The regulator output is circularly limited by the bus voltage and anti-integral saturation is achieved by inverse calculation. During the limiting, priority is given to ensuring the realization of the direct-axis current command. The common-mode command of the direct-axis current in the surface-mounted permanent magnet rotor structure is 0, and the differential-mode direct-axis current of equation (30) is superimposed on the common-mode command. As can be seen from the voltage margin calculation above, this embodiment still has a voltage margin at the highest speed and full torque. Therefore, under normal operating conditions, there is no need for field weakening. The common-mode negative direct-axis current is given according to the margin only when the bus voltage drops. The bandwidth of each ring is configured according to the principle of decreasing from the inside to the outside: the current ring 2113 is about 1.5 kHz, the velocity ring 2112 is about 150 Hz, and the position ring 2111 is about 30 Hz. The bandwidth ratios of adjacent rings are 10 times and 5 times, respectively. The bandwidth of the extended state observer 2114 is 238.7 Hz, which is between the 150 Hz of the velocity ring 2112 and the 1.5 kHz of the current ring 2113, which meets the bandwidth requirements of the active disturbance rejection structure for the observer.

[0159] The space vector pulse width modulation output module 212 receives two sets of three-phase winding voltage commands output by the current loop 2113, respectively... and Perform inverse Parker transform and inverse Clarke transform to obtain their respective static two-phase voltage commands; determine the 60° sector where the voltage vector is located based on this, calculate the three-phase duty cycle of the seven-segment symmetrical pulse width modulation according to the action time of two adjacent effective vectors and two zero vectors, and use the minimum amplitude error method to perform circular limiting for the over-modulation region. In order to reduce DC bus current ripple and common-mode interference, the carrier phases of the two sets of three-phase full bridges are staggered by 180°, and the effective value of the measured bus capacitor ripple current in this embodiment is reduced by about 30%; as mentioned above, this staggering makes the carrier valley points of the two sets not coincide, so the current sampling is triggered independently for the two sets. Each bridge arm implements dead zone compensation and minimum pulse width limitation based on the current direction. The minimum pulse width is 1.5 μs, and narrow pulses smaller than this value are omitted and compensated back in the next control cycle by voltage error accumulation; the actual duty cycle after compensation is sent back to the coordinate transformation module 205 for the terminal voltage reconstruction of equations (7) and (8) so that the back electromotive force estimation does not produce system deviation due to the dead zone effect.

[0160] The main controller 213 can be a single-chip floating-point digital signal processor, a high-performance microcontroller, or a combination of a digital signal processor and a field-programmable gate array. In this embodiment, a dedicated motor control processor with a main frequency of 200 MHz and a hardware floating-point unit and a trigonometric function acceleration unit is used. The above operations are scheduled in a hierarchical timing manner. The following operations are performed in the 20 kHz carrier interrupt layer: current sampling and zero bias gain correction, coordinate transformation and voltage reconstruction of equations (5) to (8), back electromotive force estimation of equations (10) and (32), position estimation of equations (33) and (34), calculation of model harmonic injection term of equation (24) and table lookup interpolation of compensation table of equation (25), instruction synthesis of equation (49) and allocation on both sides of equation (28), common-mode and differential-mode decoupling adjustment of current loop 2113, and space vector pulse width modulation output. The calculation time of this layer is about 22 μs, accounting for 44% of the control cycle. It is necessary to place the value of Equation (24) and the compensation table lookup in this layer because the frequency of the 12th component of the electrical angle at the rated speed has reached 1 kHz. If it is updated with the position loop at 1 kHz, it will be slightly insufficient per cycle, and it cannot be guaranteed that the injected current is in phase with the actual pulsation. The speed loop 2112, the extended state observer 2114 of Equation (45) and the harmonic extraction of Equations (15) to (18) are performed in the 4 kHz layer. The position loop 2111, the permanent magnet flux adaptive observer module 207 of Equations (38) to (43) and the feedforward calculation of Equation (44) are performed in the 1 kHz layer. The double air gap asymmetry identification of Equations (12), (19) to (21), the online correction of the fitting coefficient of Equation (23) and the weight, differential mode direct axis current and commutation angle correction of Equations (27) to (31) are performed in the 500 Hz layer. The iterative learning table update in Equation (25) is based on the mechanical cycle, updating the entire table once per mechanical cycle within the steady-state window. The opening and closing of the identification window is managed by a state machine at a 500 Hz layer with a period of 1 s. If a fixed-point implementation is used, the scale of each intermediate quantity is taken as an integer power of 2 according to the largest physical quantity. The back electromotive force component should be selected with a higher scale separately to avoid quantization loss due to its small value. The compensation table, the historical trend of asymmetry, the Hall installation error correction amount and fingerprint template, the channel zero bias and relative gain, and each calibration coefficient are all stored in a non-volatile storage area, which is not lost when power is off, and a check word is set to prevent data corruption.

[0161] Communication module 214 organizes communication using two types of messages: periodic process data and non-periodic parameter access. Its upload rate depends on the selected protocol. Periodic uplink messages contain rotor mechanical angular position, speed, RMS current values ​​of the two three-phase windings, bus voltage, winding temperature, estimated permanent magnet temperature, estimated dual-air gap asymmetry, torque ripple indicators, and fault words, totaling approximately 24 bytes. Downlink messages contain position commands, speed limits, torque limits, and operating modes, totaling approximately 10 bytes. When using EtherCAT or 5 Mbit / s CAN FD, the above process quantities are uploaded in full at 1 kHz, with a bus load of approximately 10%. When using the classic 500 kbit / s CAN, the upload rate drops to 200 Hz, and the process quantities are divided into two polling groups, with a bus load of approximately 20%. When using 115.2 kbit / s RS485, the upload rate drops to 100 Hz, and only 8 bytes of position, speed, asymmetry, and fault words are uploaded. The torque ripple index is directly calculated from the existing quantity in the controller, without the need for a torque sensor: multiply the amplitude of each harmonic extracted by equation (16) by The index is obtained by converting the torque to its root mean square and normalizing it to the rated torque. Non-periodic messages are used to read and write calibration coefficients, compensation tables, and thresholds, and support uploading the entire compensation table to the host computer for offline analysis. When the dual air gap asymmetry exceeds the alarm threshold, the permanent magnet temperature exceeds the limit, a set of three-phase windings fails, or the Hall signal is abnormal, the corresponding fault word is set to 1, along with a snapshot of the time of occurrence and the operating status at that time.

[0162] The power-on process of the servo drive system 200 is as follows: First, the power channel self-test and the zero-bias calibration of the current sampling channel are performed. Then, the calibration coefficients, fingerprint templates and compensation tables in the non-volatile storage area are read. Subsequently, the Hall signal acquisition module 204 determines the initial sector and completes the initial position estimation. At the same time, the mechanical angle reference and full-cycle count are established by one of the three: output absolute feedback, homing process or fingerprint correlation. Then, a self-calibration is performed. The self-calibration runs at 300 r / min, which is 30% of the rated speed, for no less than 8 mechanical cycles. This speed and number of cycles simultaneously meet the requirements of the differential back EMF channel for the minimum identification speed and the equation (12) in the equation. The system requires a certain number of cumulative electrical cycles. During this period, the relative gain calibration of the two sampling channels, the statistics of Hall effect installation error correction, and the consistency verification of the initial value of the permanent magnet flux linkage and the reconstructed terminal voltage are completed. The initial identification of the dual air gap asymmetry is also performed. After successful self-verification, the system enters normal servo mode, and each functional module is put into operation according to its operating condition gating conditions. If the self-verification detects that the dual air gap asymmetry has exceeded the alarm threshold, the system will not enter normal servo mode but will directly report the issue to avoid operating with a fault.

[0163] Referring to Figures 3 and 4, the complete signal chain in this embodiment operates in a closed loop within each control cycle. The current sampling module 202 synchronously collects the a-phase and b-phase currents of the two sets of three-phase windings at their respective carrier valley points and reconstructs the c-phase. The Hall signal acquisition module 204 collects three discrete Hall signals and records the flip-off time. The bus voltage and temperature acquisition module 203 collects the bus voltage and winding temperature. The communication module 214 receives the position command from the host computer and outputs the absolute feedback. The coordinate transformation module 205 transforms the two sets of phase currents to their respective synchronous rotating coordinate systems according to equations (5) to (8) and reconstructs the terminal voltage. Based on this, the mutual inductance is included in the mutual inductance to estimate the extended back electromotive force amplitude and equivalent flux linkage amplitude of the two sets of three-phase windings and is synchronously averaged according to equations (9) to (12). The permanent magnet flux linkage adaptive observer module 207 updates the permanent magnet flux linkage and quadrature axis inductance with common modulus according to equations (38) to (42) and reverse-calculates the permanent magnet temperature according to equation (43). The vector tracking observer 2061 in the high-resolution position estimation module 206 The normalized position error is constructed according to equations (32) and (33). The phase-locked loop 2062 outputs high-resolution position and rotation speed according to equations (34) and (35), and corrects the integral drift in a soft fusion manner according to equation (36) when the Hall state flips. Its model parameters are jointly corrected by the permanent magnet flux estimate and the asymmetry in equation (37). The dual air gap asymmetry identification module 208 updates the differential back electromotive force observation according to equation (13) in the identification window, and outputs the current harmonic observation according to equation (18) after fusion filtering according to equations (19) to (21). And perform threshold judgment; torque ripple feedforward compensation module 209 based on The model harmonic injection current is generated according to equations (22) to (24) based on the rotor position, and the angle domain compensation table is iteratively updated according to equation (25) within the steady-state window; the commutation angle and power distribution adaptive correction module 210 is based on... The weights on both sides, differential mode direct-axis current and commutation angle correction are calculated according to equations (27) to (31) to give the control power angle; in the three-loop active disturbance rejection control module 211, the position command is compared with the position estimate after the transmission ratio is converted. The basic cross-axis command is generated by the position loop 2111, the extended state observer 2114 and the feedforward and control law of equations (44) and (48). After periodic compensation is superimposed according to equation (49), it is allocated into two sets of three-phase winding cross-axis commands according to equation (28) and the common mode and differential mode direct-axis commands are superimposed. The current loop 2113 is closed-loop adjusted in the common mode and differential mode decoupling mode; the space vector pulse width modulation output module 212 performs inverse transformation with the control power angle and generates two sets of three-phase pulse width modulation waves to drive the two sets of three-phase full bridges of the inverter power module 201. The current output from the inverter power module 201 generates a rotating magnetomotive force in the dual three-phase winding 101d. This force interacts with the first permanent magnet array 102a and the second permanent magnet array 103a via the first air gap 104 and the second air gap 105, respectively, forming an axial main magnetic flux in the stator teeth 101a of the intermediate stator 101 and generating a stable electromagnetic torque, driving the shaft 106 to output power. The differentiated current distribution on both sides balances the output while compressing the net unbalanced axial force acting on the bearing 107. The actual response of the motor is fed back to the start of the current cycle via the current sampling module 202 and the Hall signal acquisition module 204. The position observation model and power allocation are continuously corrected. The position estimate continuously provides a phase reference for harmonic compensation and iterative learning. The permanent magnet flux estimate continuously corrects the torque model, observation model and input gain of active disturbance rejection control. This constitutes a closed-loop coordinated process of sampling, estimation, identification, compensation, control, output to resampling. The identification window is used to decouple the compensation action and the identification action in time to avoid mutual excitation between the two.

[0164] To verify the effectiveness of the above implementation method, a comparative test was conducted on the prototype in this embodiment. Two asymmetric states were artificially created by adding adjusting shims with thicknesses of 0.05 mm and 0.1 mm, respectively, between the bearing housing and the end cover. The measured asymmetry of the dual air gap was approximately 0.05 and 0.10, respectively. The statistical caliber of torque pulsation is uniformly defined as follows: each component is given as a percentage of its amplitude relative to the rated torque, and the peak-to-peak value is the peak-to-peak value of the waveform synthesized by combining the actual phases of each component. Therefore, the peak-to-peak value is approximately 1.6 times the sum of the amplitudes of each component, rather than their algebraic sum. As shown in Figure 5, under the conditions of rated speed, rated torque, and asymmetry of approximately 0.10, when only conventional three-loop vector control is used and the current is equally distributed between the two sets of three-phase windings, the amplitudes of the mechanical first, mechanical second, electrical angle first, electrical angle second, electrical angle sixth, and electrical angle twelfth components are 1.19%, 0.51%, 0.90%, 0.41%, 2.38%, and 1.40%, respectively, with a combined peak-to-peak value of 10.8%. The net unbalanced axial force acting on bearing 107, as measured by the force measuring ring, is approximately 55 N. After activating the dual air gap asymmetry identification module 208, the torque pulsation feedforward compensation module 209, and the commutation angle and power distribution adaptive correction module 210, The steady-state fluctuation is less than 0.01, and the above six components decrease to 0.15%, 0.08%, 0.12%, 0.07%, 0.42%, and 0.35% respectively, the composite peak-to-peak value decreases to 2.2%, and the net unbalanced axial force decreases to about 20 N. Among them, the 12th cogging component and part of the 6th component of the electrical angle already exist in the symmetrical state and are not caused by the air gap asymmetry. However, since the iterative learning of Equation (25) converges on all periodic components synchronized with the rotor position in the angular domain, this part of the inherent pulsation is also suppressed. Under the condition of an asymmetry of about 0.05, the composite peak-to-peak value decreases from 8.3% to 2.0%, and the net unbalanced axial force decreases from about 33 N to about 13 N. As a control, under the normal assembly state where the asymmetry is less than 0.01 after the gasket is selected, the composite peak-to-peak value is 6.2% without compensation and 1.9% after compensation is enabled.

[0165] The improvement in efficiency is limited and considered a minor effect. At a winding temperature of 40°C, the output power at the rated operating point is 837.8 W, copper loss is 67.7 W, iron loss is approximately 8 W, mechanical loss is approximately 6 W, and the overall efficiency is 91.1%. After enabling compensation, the effective value of harmonic current decreases from approximately 5% of the fundamental frequency to 1.5%, harmonic copper loss decreases by approximately 0.15 W, iron loss and permanent magnet eddy current loss related to harmonic magnetomotive force decrease by approximately 1.0 W, and the bearing friction loss corresponding to a 35 N reduction in net unbalanced axial force decreases by approximately 0.2 W, totaling approximately 1.35 W, increasing the overall efficiency to 91.2%. Under low-speed conditions (400 r / min, 40% of the rated speed) and rated torque, the output power is 335.1 W, copper loss remains at 67.7 W, iron loss is approximately 3.5 W, mechanical loss is approximately 3 W, and the overall efficiency is 81.9%. Under these conditions, the proportion of harmonic current to the fundamental frequency is relatively high; after compensation, the total loss is reduced by approximately 1.6 W, increasing the efficiency to 82.2%. The above data indicates that the efficiency gain of this implementation method mainly comes from the reduction in harmonic losses and bearing friction losses, which are only a few tenths of a percentage point and should not be compared with improvements in torque pulsation and net axial force.

[0166] Regarding position estimation, when only the Hall signal acquisition module 204 is used, the position resolution is 60 electrical degrees, and the instantaneous error during startup is approximately ±30 electrical degrees. After enabling the high-resolution position estimation module 206 and the permanent magnet flux adaptive observer module 207, the steady-state position estimation error of the motor shaft at rated speed converges to within ±2.8 electrical degrees, corresponding to a mechanical angle of approximately ±0.56°, and the maximum instantaneous error during startup is reduced to approximately ±5 electrical degrees. Based on a transmission ratio of 50, the motor shaft estimation error corresponds to approximately ±0.011° at the output end. However, the actual positioning accuracy at the output end also depends on the hysteresis of the reduction mechanism and the accuracy of the absolute feedback at the output end. In this embodiment, the hysteresis of the reduction mechanism is approximately 0.03°, and the measured static positioning error at the output end is no greater than 0.05°. That is, the accuracy at the output end is dominated by the transmission chain rather than by the estimation accuracy on the motor side. The improvement in the estimation accuracy on the motor side is mainly reflected in the reduction of commutation phase error and low-frequency torque pulsation, as well as the improvement of the motor shaft positioning capability under semi-closed-loop conditions without output feedback. Under thermal equilibrium conditions where the winding temperature rises to approximately 80 K, the position estimation error does not significantly worsen compared to the normal temperature conditions due to the online tracking of the permanent magnet flux linkage by equation (40) and the differentiated correction of the two winding models by equation (37). In contrast, when the adaptive permanent magnet flux linkage and asymmetry injection are turned off and the fixed parameters at normal temperature are used, the position estimation error under the same conditions expands to approximately ±11 electrical angles, and the peak-to-peak value of the torque pulsation synthesis rises to 5.8%. In terms of dynamic performance, under a step load of 50% of the rated torque, the speed drop decreases from 42 r / min when using a conventional proportional-integral speed loop to 14 r / min, and the recovery time is shortened from 53 ms to 19 ms, demonstrating the suppression effect of the extended state observer 2114 on aperiodic disturbances. At the same time, since the periodic pulsation has been compensated by the angle domain, the bandwidth of the extended state observer 2114 does not need to be increased to track the pulsation, and the effective value of the high-frequency chattering of the quadrature axis current command remains within 1.5% of the rated current. Furthermore, the differential-mode direct-axis current gain is artificially increased. In a control trial where the recognition window was closed and the efficiency was doubled, The signal monotonically rose to the alarm threshold within tens of seconds, verifying the existence of positive feedback caused by mutual inductance and the blocking effect of the identification window on this feedback.

[0167] Without changing the above core technical solution, this embodiment can also be modified in several ways. In addition to adopting a segmented armature structure without a yoke, the intermediate stator 101 can also adopt a yoke structure with a stator yoke. In this case, the magnetic circuits on both sides are closed along the circumference through the stator yoke and are approximately independent of each other. The sensitivity coefficient in equation (13) can reach 0.21 according to equation (14). The signal-to-noise ratio of the differential back electromotive force channel, the axial force adjustment capability of equation (30) and the net axial force level characterized by equation (4) are increased synchronously. In this modified embodiment, the net unbalanced axial force when the asymmetry is about 0.10 is reduced from about 116 N to about 42 N. The synchronous average number of cycles can be reduced from 64 to 16 and the identification bandwidth is correspondingly increased. The cost is that the torque density is slightly reduced. In addition to adopting soft magnetic composite material, the stator teeth 101a can also adopt silicon steel strip or amorphous alloy strip wound radially in the lamination direction. In this case, the iron loss model should be modified accordingly. In addition to 12 slots and 10 poles, the slot-pole combination can also adopt fractional slot concentrated winding combinations such as 18 slots and 16 poles, 24 slots and 20 poles, and 24 slots and 22 poles. The spatial phase shift of the double three-phase winding 101d can also be 20 electrical angles or other values ​​that can cancel specific harmonics. Accordingly, the phase shift in equation (6) and the inverse transformation, the set of harmonic orders to be compensated in equation (22), and the number of angle grids in equation (25) should be adjusted synchronously. In addition to 2 segments, the axial segmentation of the concentrated winding 101c can also be divided into 3 or more segments. The middle segment bears the main output force, while the two end segments are used for differential detection and axial force adjustment, thereby improving the detection sensitivity without sacrificing the output force. At this time, the mutual inductance between the two end segments and the middle segment must also be included in the model in the form of equation (9).

[0168] If the target application is extremely cost-sensitive, the double three-phase winding 101d can be simplified to a single three-phase winding. In this case, the differential excitation on both sides of equations (27) to (30) is cancelled, and only the commutation angle correction of equation (31) and the harmonic compensation of equations (22) to (25) are retained. The identification of the double air gap asymmetry depends only on the current harmonic channel. Since the differential back electromotive force channel does not exist, equation (18) cannot be obtained from the current harmonic channel. Since symbols are borrowed, a separate symbol method must be defined. This embodiment provides a method for this variation involving small-scale trials: alternating applications within a steady-state window. and tentative phase angle offset, Take electrical angles of 1 to 2, compare the sum of the first and second harmonic amplitudes of the electrical angles extracted by equation (16) under the two biases, and take the direction that reduces the sum as the direction of electrical angle. The symbol is determined; this determination is executed every few seconds and outputs the majority vote of multiple consecutive results. Once the symbol is determined, it is written to the non-volatile storage area and used in subsequent operations. It is only re-determined when the harmonic amplitude increases abnormally. This deformation loses the ability to actively adjust for net unbalanced axial forces and the ability to tolerate one-sided faults, but still improves the accuracy of torque pulsation and position estimation. Conversely, for high-reliability aerospace servos, the dual three-phase winding 101d should be retained and can be expanded to 3 or 4 sets of multiple three-phase windings. Each set of windings is powered by an independent power channel and an independent sampling channel, so that even if any channel fails, it still maintains an output of not less than 50% of the rated value. At the same time, the rotor axial oscillation that may be induced when a three-phase winding experiences a phase-to-phase short circuit should be included in the fault criterion. That is, after a short circuit is detected and the winding is disconnected, monitoring should continue. The correlation between the fluctuation amplitude and the speed fluctuation is such that once a coupling characteristic related to the mechanical resonance frequency appears, the output will be immediately drated or blocked.

[0169] There are also several equivalent substitutions at the algorithm and implementation level. The first-order low-pass filter of Equation (20) can be replaced by the Kalman filter of Equation (21) or the recursive least squares estimator with forgetting factor; the iterative learning link of Equation (25) can be replaced by a quasi-resonant regulator group or a repetitive controller tuned according to the mechanical first, second and electrical first, second, sixth and twelfth frequencies respectively. The three are equivalent in suppressing periodic pulsations. Iterative learning is implemented in the angle domain and is not sensitive to speed changes. The quasi-resonant regulator is easy to tune in the frequency domain and prove its stability. The repetitive controller is simple to implement but is more sensitive to speed fluctuations. In addition to the proportional integral type, the phase-locked loop 2062 can also adopt an orthogonal signal generation structure based on the second-order generalized integrator, a nonlinear tracking differentiator or a sliding mode observer. The construction method of its position error signal is consistent with that of Equation (33), that is, taking the negative value of the direct-axis back electromotive force in the estimated coordinate system; the permanent magnet flux adaptive observer module 207 In addition to the model reference adaptive structure, extended Kalman filtering or least squares parameter identification with a forgetting factor can also be used; in addition to the three methods mentioned above, a single-turn index signal can be added to the back of the rotating shaft 106 or the first rotor 102 to establish the mechanical angle reference. Although this adds a switching element, fingerprint-related calculations can be omitted. The current loop 2113, coordinate transformation module 205 and space vector pulse width modulation output module 212 can be placed in a field programmable gate array to shorten the control delay to a few μs, while the dual air gap asymmetry identification module 208, permanent magnet flux linkage adaptive observer module 207 and iterative learning loop can be executed in a digital signal processor or embedded processor core; alternatively, The long-term trend is uploaded to the host computer or cloud via communication module 214, where the host computer performs trend analysis and lifespan prediction, while only the threshold judgment is retained locally. At the application level, although this implementation uses electric servo motors and servo actuators as typical examples, it is directly applicable to robot joint modules, direct-drive joints of collaborative robotic arms, hub drives, control surface drives of electric vertical take-off and landing aircraft, and various applications using double-sided air-gap axial flux motors with high requirements for torque stability, positioning accuracy, and bearing life. In these applications, the source of position commands, transmission ratio, and specific values ​​of each threshold can be adjusted accordingly, while the collaborative mechanism of identification, compensation, and parameter adaptation remains unchanged.

[0170] The above description is merely a preferred embodiment of the present invention and does not limit the scope of patent protection of the present invention. Any equivalent structural transformations or algorithmic substitutions made based on the description and drawings of the present invention, or any direct or indirect application of the above technical solutions to other related technical fields, are similarly included within the scope of protection of the present invention.

Claims

1. A high torque density double-sided air gap axial brushless DC motor, comprising a shaft (106), a first rotor (102) and a second rotor (103) fixedly connected to the shaft (106) and rotating synchronously, and an intermediate stator (101) located axially between the first rotor (102) and the second rotor (103); the inner disks of the first rotor (102) and the second rotor (103) facing the intermediate stator (101) are respectively provided with first permanent magnet arrays with alternating polarities along the circumference. The first rotor (102a) and the second permanent magnet array (103a) form a first air gap (104) between the first rotor (102) and one end face of the intermediate stator (101), and the second rotor (103) forms a second air gap (105) between the second end face of the intermediate stator (101); the intermediate stator (101) includes a plurality of stator teeth (101a) arranged circumferentially and concentrated windings (101c) respectively wound on each of the stator teeth (101a); characterized in that: The concentrated winding (101c) is divided into a first winding segment near the first rotor (102) and a second winding segment near the second rotor (103) along the axial direction of the stator teeth (101a). An insulating layer is provided between the first winding segment and the second winding segment and they are led out independently. The first winding segments on each stator tooth (101a) are connected in a three-phase symmetrical pattern to form a first set of three-phase windings. The second winding segments on each stator tooth (101a) are connected in a circumferentially staggered pattern with one or more stator teeth (101a) and the coil polarities are reversed to form a second set of three-phase windings. The composite magnetomotive force axis of the second set of three-phase windings is spatially phase-shifted by a non-zero predetermined electrical angle with the composite magnetomotive force axis of the first set of three-phase windings. The predetermined electrical angle is determined by the spatial phase difference between adjacent stator teeth (101a), the number of staggered stator teeth (101a), and the 180-degree electrical angle offset corresponding to the polarity reversal. The first set of three-phase windings and the second set of three-phase windings are connected in a star configuration. Their neutral points are isolated from each other and are not brought out. Their phase lines are brought out independently and used to connect to two independent three-phase full bridges respectively. The number of turns of each first winding segment and each second winding segment are equal, and they are respectively offset in the axial direction to one side of the first air gap (104) and one side of the second air gap (105), so that in addition to the common axial main magnetic flux, the first set of three-phase windings and the second set of three-phase windings are also respectively linked with edge magnetic flux components that enter from the side of the stator teeth (101a) and vary with the axial length of the first air gap (104) and the second air gap (105).

2. The high torque density double-sided air gap axial brushless DC motor according to claim 1, characterized in that: The number of stator teeth (101a) and the number of magnetic poles of the first permanent magnet array (102a) and the second permanent magnet array (103a) form a fractional slot concentrated winding. The spatial phase difference between adjacent stator teeth (101a) is 150 electrical angles. The number of staggered stator teeth (101a) is 1. The predetermined electrical angle is 30 electrical angles. Each phase of the first set of three-phase windings and the second set of three-phase windings is composed of multiple first winding segments or multiple second winding segments connected in series; The composite magnetomotive force axis of the second set of three-phase windings lags behind or leads the composite magnetomotive force axis of the first set of three-phase windings by a predetermined electrical angle. The lag or lead is determined by the direction offset along the circumference and is determined and stored by dragging and measuring the phase difference of the no-load back electromotive force of the two sets of three-phase windings after assembly.

3. The high torque density double-sided air gap axial brushless DC motor according to claim 1, characterized in that: The intermediate stator (101) is a segmented armature structure without a yoke. Each stator tooth (101a) is an axially penetrating fan-shaped column and is molded from a soft magnetic composite material bonded together with iron powder particles with an insulating surface. Each stator tooth (101a) is embedded and fixed in a non-magnetic support (101b) along the circumference. The inner circumference of the non-magnetic support (101b) is provided with a positioning groove that matches the shape of the stator tooth (101a) and forms a partition between adjacent stator teeth (101a). There is no parasitic magnetic circuit between adjacent stator teeth (101a). The magnetic flux in each stator tooth (101a) is closed only along the axial direction between the first rotor (102) and the second rotor (103) through the first air gap (104) and the second air gap (105).

4. The high torque density double-sided air gap axial brushless DC motor according to claim 1, characterized in that: The end of the concentrated winding (101c) is folded inward and outward in the radial direction, and its axial end face does not exceed the end face of the stator tooth (101a), and an axial clearance is maintained between the end face of the stator tooth (101a) and the end of the concentrated winding (101c). The first permanent magnet array (102a) and the second permanent magnet array (103a) have a pole arc coefficient of less than 1, and are arranged in a fan-shaped permanent magnet alternating polarity bonding arrangement, a Heilbeck array, or a slanted pole arrangement combining radial segmentation and circumferential segmentation. The rotating shaft (106) is supported by a pair of preloaded angular contact ball bearings (107), which bear the difference in axial magnetic pull generated by the first permanent magnet array (102a) and the second permanent magnet array (103a).

5. A high-precision servo drive system for driving a high torque density double-sided air-gap axial brushless DC motor as described in any one of claims 1 to 4, comprising an inverter power module (201), a current sampling module (202), a bus voltage and temperature acquisition module (203), a Hall signal acquisition module (204), and a main controller (213); the inverter power module (201) comprises two independent three-phase full-bridges respectively connected to the first set of three-phase windings and the second set of three-phase windings; the current sampling module (202) acquires the phase currents of the first set of three-phase windings and the second set of three-phase windings respectively; the Hall signal acquisition module (204) acquires the discrete signals output by three switching Hall elements fixed on the intermediate stator (101) and spaced 120 electrical degrees apart; characterized in that: The main controller (213) includes a coordinate transformation module (205), a high-resolution position estimation module (206), a dual air gap asymmetry identification module (208), a torque pulsation feedforward compensation module (209), a commutation angle and power distribution adaptive correction module (210), a three-closed-loop active disturbance rejection control module (211), and a space vector pulse width modulation output module (212). The dual-air gap asymmetry identification module (208) takes the phase current and the terminal voltage reconstructed by the duty cycle of the space vector pulse width modulation output module (212) and the DC bus voltage and corrected by the dead zone voltage error as input. Under the condition of taking into account the equivalent mutual inductance between the first set of three-phase windings and the second set of three-phase windings, it estimates the amplitude of the extended back electromotive force of the two sets of three-phase windings respectively. Taking the rotor electrical angle as the phase reference, the amplitude of each extended back electromotive force is synchronously averaged for multiple electrical cycles according to the same electrical angle grid. The ratio of the difference of the synchronous average values ​​of the two sets of three-phase windings to their sum is taken as the differential back electromotive force observation. The differential back electromotive force observation is converted into a dual-air gap asymmetry estimate characterizing the difference in axial length between the first air gap (104) and the second air gap (105) according to the pre-calibrated decoupling sensitivity coefficient. The decoupling sensitivity coefficient is the calibration value of the first-order sensitivity of the edge magnetic flux component to the difference in axial length. The main controller (213) sets an identification window during operation according to a set cycle. Within the identification window, the current commands of the first set of three-phase windings and the second set of three-phase windings are the same, that is, the current distribution weights of the two sets of three-phase windings are equal and the differential mode direct axis current between them is zero. The synchronous average is accumulated only within the identification window, and the boundary of the identification window is transitioned by a slope. The commutation angle and power distribution adaptive correction module (210) allocates complementary quadrature axis current weights to the first set of three-phase windings and the second set of three-phase windings according to the estimated value of the dual air gap asymmetry, so that the set of three-phase windings adjacent to the air gap with a larger axial length obtains a larger quadrature axis current weight. It also superimposes differential mode direct axis currents proportional to the estimated value of the dual air gap asymmetry and opposite to each other to the two sets of three-phase windings, so that the set of three-phase windings adjacent to the air gap with a smaller axial length obtains a negative direct axis current. At the same time, it gives the commutation angle correction amount according to the estimated value of the dual air gap asymmetry and gives the control power angle used for coordinate transformation accordingly. The high-resolution position estimation module (206) includes a vector tracking observer (2061) and a phase-locked loop (2062). Using the 60° electrical angle sector determined by the discrete signal as the absolute phase reference, the negative value of the direct-axis back electromotive force in the estimated coordinate system is constructed from the components of the extended back electromotive force in the stationary two-phase coordinate system, and normalized by the back electromotive force amplitude to obtain a normalized position error signal. The phase-locked loop (2062), containing a double-integral element, then outputs continuous rotor position and speed estimates. The permanent magnet flux linkage parameters of the first and second three-phase windings in the vector tracking observer (2061) are differentiated by performing a differential correction based on the product of the permanent magnet flux linkage common-mode estimate, the double air gap asymmetry estimate, and the decoupling sensitivity coefficient, with opposite signs. The torque pulsation feedforward compensation module (209) uses the estimated value of the dual air gap asymmetry as the amplitude basis and the estimated value of the rotor position as the phase reference to generate a periodic compensation current synchronized with the rotor position. The periodic compensation current is composed of the harmonic injection term given by the asymmetry model and the lookup value of the angle domain compensation table indexed by the rotor mechanical angle. The three-loop active disturbance rejection control module (211) includes a position loop (2111), a speed loop (2112), a current loop (2113), and an extended state observer (2114). The extended state observer (2114) estimates and compensates for non-periodic disturbances to generate a basic quadrature-axis current command. The basic quadrature-axis current command is superimposed with the periodic compensation current and distributed according to the quadrature-axis current weight to the quadrature-axis current commands of the two sets of three-phase windings respectively. The differential-mode direct-axis current is superimposed and the current loop (2113) adjusts the two sets of three-phase windings in closed loop respectively.

6. The high-precision servo drive system according to claim 5, characterized in that: The dual-air-gap asymmetry identification module (208) is also equipped with a current harmonic identification channel. The current harmonic identification channel takes the arithmetic mean of the cross-axis currents of the two sets of three-phase windings as input, and separates the first-order electrical angle component and the second-order electrical angle component with the rotor position estimate as the phase reference through adaptive harmonic extraction or a second-order generalized integrator. The first-order mechanical angle component and the second-order mechanical angle component are separated in the same way according to the mechanical angle order. The separated first-order electrical angle component and the second-order electrical angle component are used as the basis for the identification. The current harmonic observation is constructed by the ratio of the sum of the amplitudes of the secondary components to the DC component of the arithmetic mean. The sign of the current harmonic observation is taken from the sign of the differential back electromotive force observation. When the DC component is lower than the set minimum current threshold, the output of the current harmonic identification channel is judged to be invalid and frozen. The differential back electromotive force observation and the current harmonic observation are fused according to the weighting coefficients, and the estimated value of the dual air gap asymmetry is output after passing through a first-order low-pass filter or a Kalman filter with the asymmetry degree and its rate of change. The weighting coefficients are adjusted online with the rotational speed and the common-mode component of the quadrature-axis current, and the differential back electromotive force observation is not involved in the fusion when the rotational speed is lower than the set minimum identification speed.

7. The high-precision servo drive system according to claim 6, characterized in that: The dual-air gap asymmetry identification module (208) is provided with a compensation start threshold and an alarm threshold. The compensation start threshold is higher than the steady-state fluctuation of the dual-air gap asymmetry estimate and lower than the alarm threshold. When the absolute value of the estimated dual-air gap asymmetry is lower than the compensation start threshold for multiple consecutive update cycles, the harmonic injection term decays to zero as the estimated dual-air gap asymmetry decreases, and the angle domain compensation table is frozen and no longer updated. When the absolute value of the estimated value of the dual air gap asymmetry exceeds the compensation start threshold, the harmonic injection term changes continuously with the estimated value of the dual air gap asymmetry and the angle domain compensation table is updated. A hysteresis loop is provided at the compensation start threshold. When the absolute value of the estimated value of the dual air gap asymmetry exceeds the alarm threshold, the main controller (213) limits the torque command amplitude and the maximum speed according to the set slope, and at the same time reports the fault word and the estimated value of the dual air gap asymmetry through the communication module (214). The historical trend of the estimated dual-air gap asymmetry is stored in a non-volatile storage area, and when a step occurs within a set time, it is determined to be a mechanical impact or a loose fastener.

8. The high-precision servo drive system according to claim 5, characterized in that: The harmonic injection term includes electrical angle components of the 1st, 2nd, 6th and 12th order and mechanical angle components of the 1st and 2nd order. The amplitude of each component is taken as a linear function of the estimated value of the dual air gap asymmetry. Its constant term is the inherent pulsation amplitude under the symmetrical state of the two air gaps. The phase addition of each component is a lead amount proportional to the rotational speed. The torque gain used to convert the torque to current for each component is updated online with the estimated value of the permanent magnet flux common mode. The angle domain compensation table discretizes a mechanical cycle into multiple equally spaced angle grids and linearly interpolates between the grids. The compensation amount of each grid point is iteratively updated according to the mechanical cycle number by a zero-phase low-pass filter with a forgetting factor and a learning gain with phase lead correction. Its error signal is taken from the periodic component in the total disturbance estimate of the extended state observer (2114), the AC component of the common-mode component of the cross-axis current of the two sets of three-phase windings, or one of the angle domain waveforms of the speed fluctuation and normalized to the rated torque. The iterative update is only performed within a steady-state window where the rate of change of speed fluctuation, load change rate and the rate of change of the estimated value of dual air gap asymmetry are all lower than the set value. The update is frozen when a speed command step or load change is detected. The compensation amount of each grid point has an upper limit of amplitude and is written to the non-volatile storage area as the initial value of the next power-on when the machine is stopped. The periodic pulsations synchronized with the rotor position are handled by the harmonic injection term and the angle domain compensation table, while the non-periodic disturbances are handled by the extended state observer (2114), and the bandwidth of the extended state observer (2114) is between the bandwidth of the velocity loop (2112) and the bandwidth of the current loop (2113).

9. The high-precision servo drive system according to claim 6, characterized in that: The weighting gain of the cross-axis current is taken as the weighted value between a first gain aimed at equalizing the electromagnetic torque contributions of the two sets of three-phase windings and a second gain aimed at minimizing the net unbalanced axial force acting on the bearing (107). The first gain is taken as the decoupling sensitivity coefficient, and the second gain is greater than the first gain. Its weighting coefficient is oriented towards the second gain under heavy load and low speed conditions and towards the first gain under light load and high speed conditions. The cross-axis current weight is subject to upper and lower limits. The commutation angle correction is only used for the coordinate transformation of the current of the two sets of three-phase windings and the inverse transformation of the voltage command, and is not used for the position feedback of the position loop (2111). When the effective value of the tracking error of the current loop (2113) of any three-phase winding exceeds the set ratio of its command amplitude, or when an open circuit or phase loss of the three-phase winding or a short circuit of the three-phase full bridge arm is detected, or when the ratio of the equivalent flux amplitude of the two three-phase windings deviates from 1 by more than the set ratio, it is determined that the three-phase winding or its power channel is faulty, the corresponding three-phase full bridge is blocked, and the quadrature axis current weight of the remaining three-phase winding is reset to 1. At the same time, the upper limit of the torque command is reduced, and the update of the dual air gap asymmetry estimation value is switched to using only the current harmonic identification channel.

10. The high-precision servo drive system according to claim 5, characterized in that: The high-resolution position estimation module (206) subtracts the rotor position estimate from the sum of the theoretical boundary phase of the 60 electrical angle corresponding to the reversal and the Hall installation error correction amount of the sector to which the boundary belongs. The difference is multiplied by the fusion coefficient to perform incremental correction on the rotor position estimate. The fusion coefficient decreases as the back electromotive force amplitude increases. The single correction amount is limited by the amplitude and the cumulative fault count is accumulated when the limit is exceeded. The Hall installation error correction amount is calculated and stored in a table by sector by the difference between the actual phase and the theoretical boundary phase of each sector during steady-speed operation, and is applied only once in the incremental correction. The main controller (213) establishes a mechanical angle reference and full circle count based on one of the following three methods: absolute position feedback from the output end of the reduction mechanism driven by the motor, calculated according to the transmission ratio, power-on return to zero process, or Hall flip interval sequence formed by the inherent difference of each magnetic pole of the first permanent magnet array (102a) and cyclic correlation operation of the pre-stored fingerprint template. The angle grid of the angle domain compensation table is aligned with the mechanical angle reference. The main controller (213) is also equipped with a permanent magnet flux adaptive observer module (207), which takes the common-mode components of voltage and current of two sets of three-phase windings as input and the common-mode equivalent inductance as parameter, and outputs the common-mode estimate of the permanent magnet flux online according to the model reference adaptive law. The common-mode estimate of the permanent magnet flux is used to correct the model parameters of the vector tracking observer (2061), the torque conversion gain of the periodic compensation current and the input gain of the three closed-loop active disturbance rejection control module (211), and back-calculates the average temperature of the permanent magnet according to the remanent magnet temperature coefficient of the permanent magnet.

Images