Joint motor closed-loop control method with multiple error couplings

Abstract

This invention relates to the field of data processing technology and discloses a closed-loop control method for a joint motor with multi-error coupling. The method includes: acquiring multi-dimensional heterogeneous data within the joint and aligning it in the time domain to construct a system state observation vector; executing an energy integration algorithm based on the vector to calculate the liquefaction rate parameter reflecting the physical state of the phase change material; calculating the dynamic viscosity of the lubricating oil using temperature data and constructing a thermal-fluid coupling model to solve for the real-time physical backlash value affected by temperature and viscosity; mapping the real-time physical backlash value and the liquefaction rate parameter to a motion control algorithm to generate a current loop control command containing multi-source error compensation components; and generating a pulse width modulation signal to drive the joint motor according to the command. This invention achieves joint decoupling of thermodynamic state and mechanical error through energy state observation and thermal-fluid coupling modeling, solving the problems of blind spots in heat dissipation monitoring and thermal drift of errors, and improving position tracking accuracy and dynamic response performance.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a closed-loop control method for a joint motor with multiple error coupling. Background Technology

[0002] Currently, high-performance robots and exoskeletons place high demands on the power density and miniaturization of joint drive systems. Highly integrated joint modules encapsulate drive motors, precision reducers, and sensors within confined spaces, making it difficult to dissipate heat generated by electromagnetic losses and mechanical friction under high loads. In such confined spaces, maintaining high-precision motion control while ensuring system thermal safety through data processing has become a key issue in the field of precision electromechanical control.

[0003] For the control of the aforementioned joint systems, existing technologies typically employ sensor-based feedback control strategies. The controller monitors the temperature at various points in real time by reading the values ​​of thermistors embedded in the motor windings or the driver substrate. Once the monitored value exceeds a preset safety threshold, it triggers a shutdown or forced derating protection. In terms of motion control, existing solutions mostly use vector control algorithms and, based on fixed friction coefficients and gear backlash parameters calibrated under standard room temperature conditions, superimpose fixed compensation components into the control loop to correct mechanical transmission errors.

[0004] However, existing technologies have limitations under multi-physics coupled operating conditions. On the one hand, for compact joints using phase change materials for thermal buffering, the isothermal endothermic characteristics during the phase change process mean that temperature data cannot characterize the remaining heat capacity. Relying solely on temperature thresholds can easily lead to sudden thermal runaway under thermal saturation, lacking proactive awareness of the energy state. On the other hand, mechanical error parameters are not constant. Lubricating oil viscosity decreases non-linearly with increasing temperature, and thermal expansion of metal components and changes in oil film thickness can jointly cause drift in the actual gear backlash. Models based on fixed parameters cannot adapt to these dynamic time-varying characteristics, resulting in low-speed crawling vibrations or commutation position errors in the motor during temperature rise, making it difficult to meet the high-precision closed-loop control requirements under all operating conditions.

[0005] Therefore, the present invention provides a closed-loop control method for joint motors with multiple error coupling to overcome the shortcomings of the prior art. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a closed-loop control method for joint motors with multiple error coupling, which solves the problems of thermal management failure and decreased motion control accuracy caused by the unobservable phase change heat dissipation state and the dynamic drift of mechanical errors due to thermal flow coupling in existing high power density joints.

[0007] To achieve the above objectives, the present invention provides a closed-loop control method for a joint motor with multiple error coupling, comprising the following steps:

[0008] Collect data from inside the joint, read multi-dimensional sensor data through synchronous triggering and perform time-domain alignment to construct a system state observation vector;

[0009] Based on the system state observation vector, an energy integration algorithm is executed to calculate the liquefaction rate parameter, which reflects the physical state of the phase change material.

[0010] The dynamic viscosity of the lubricating oil is calculated using the temperature data in the system state observation vector. The friction model parameters are updated based on the dynamic viscosity value to calculate the friction feedforward compensation component. A thermal-fluid coupling model is constructed based on the dynamic viscosity value to calculate the real-time physical back clearance value.

[0011] The real-time physical backlash value, the friction feedforward compensation component, and the liquefaction rate parameter are mapped into the motion control algorithm to generate a current loop control command containing multi-source error compensation components.

[0012] The current loop control command generates a pulse width modulation signal to drive the joint motor.

[0013] By employing the above technical solution, and by using an energy integration algorithm to observe the latent heat state of the phase change material in real time, combined with a thermal-fluid coupling model to dynamically calculate the mechanical backlash and frictional characteristics affected by temperature and viscosity, this invention achieves joint decoupled control of the thermodynamic state and mechanical error of the joint motor. Therefore, it achieves the technical effect of balancing high dynamic response, thermal safety, and high-precision position tracking under conditions of limited heat dissipation in the compact joint space and variable operating conditions, solving the problems of traditional methods' inability to detect the phase change heat absorption process and their inability to adapt to the thermal drift of nonlinear mechanical errors.

[0014] Preferably, the process of acquiring internal joint data and performing time-domain alignment to construct a system state observation vector includes: synchronously reading the motor rotor angular position, load end angular position, load torque, and temperature data distributed at the joint heating parts; acquiring DC bus voltage and current data; calculating the motor angular velocity based on the differential value of the motor rotor angular position; performing digital filtering on the motor rotor angular position, motor angular velocity, load end angular position, load torque, temperature data, voltage data, and current data to remove high-frequency noise, and combining the processed data into the system state observation vector.

[0015] By adopting the above technical solution, strict synchronization of multidimensional heterogeneous physical quantities in the time dimension and uniformity in the spatial dimension are ensured. In particular, the introduction of motor angular velocity as a key observation state eliminates sensor quantization noise and provides a reliable data foundation for the accurate calculation of subsequent complex models.

[0016] Preferably, the process of calculating the liquefaction rate parameter reflecting the physical state of the phase change material includes: calculating the input electrical power based on the voltage data and the current data in the system state observation vector; calculating the output mechanical power based on the differential values ​​of the load torque and the load end angle position; and estimating the heat dissipation power based on the temperature data; calculating the net heat flow after subtracting the output mechanical power and the heat dissipation power from the input electrical power; accumulating and integrating the net heat flow in the discrete time domain to obtain the total net heat energy; and removing the sensible heat energy used for heating non-phase change components from the total net heat energy to obtain the latent heat energy; and normalizing the latent heat energy to the total latent heat capacity of the phase change material to obtain the liquefaction rate parameter.

[0017] By adopting the above technical solution, a latent heat observation mechanism based on energy conservation was established. The operation order of energy integration and sensible heat elimination was corrected. During the phase change plateau period when the temperature sensor reading remains constant, the remaining heat capacity of the phase change material can be accurately calculated by integrating the net heat flow, thus eliminating the blind spot of thermal management that relies solely on temperature monitoring.

[0018] Preferably, the process of calculating the dynamic viscosity value of the lubricating oil includes: extracting motor winding temperature data from the system state observation vector as the equivalent operating temperature of the lubricating oil; calling the stored lubricating oil viscosity-temperature characteristic equation, substituting the equivalent operating temperature into the lubricating oil viscosity-temperature characteristic equation, and solving for the dynamic viscosity value at the current moment through exponential operation.

[0019] By adopting the above technical solution, the slowly changing thermodynamic scalar is transformed into fluid dynamic parameters that directly affect the mechanical transmission characteristics, providing a physical benchmark for the real-time correction of the friction and backlash model.

[0020] Preferably, the process of calculating the real-time physical backlash value includes: establishing a dynamic backlash model that includes a thermal expansion term and an oil film thickness term; calculating the reduction in tooth flank clearance caused by thermal expansion of the metal gear based on the temperature data; calculating the equivalent clearance increase caused by the hydrodynamic lubricating oil film based on the dynamic viscosity value and the motor angular velocity in the system state observation vector; and superimposing the initial geometric backlash, the reduction in tooth flank clearance, and the equivalent clearance increase to calculate the real-time physical backlash value.

[0021] By adopting the above technical solution, the actual meshing clearance change under the interaction of thermal expansion effect and fluid lubrication effect is accurately quantified, which improves the accuracy of dead zone compensation in position control and avoids system oscillation caused by back clearance estimation deviation.

[0022] Preferably, the process of updating the friction model parameters based on the dynamic viscosity value to calculate the friction feedforward compensation component includes: correcting the Coulomb friction torque parameter, the maximum static friction torque parameter, and the viscous damping coefficient in the friction model according to the dynamic viscosity value; substituting the corrected Coulomb friction torque parameter, the maximum static friction torque parameter, the viscous damping coefficient, and the motor angular velocity in the system state observation vector into the friction model to calculate the estimated friction torque value at the current moment; and converting the estimated friction torque value into the friction feedforward compensation component.

[0023] By adopting the above technical solution, the nonlinear frictional characteristic drift caused by changes in lubricating oil viscosity is adapted, and the viscous resistance and static friction are offset by dynamic feedforward compensation, thereby improving the low-speed following performance of the system under different temperature conditions.

[0024] Preferably, the process of mapping the real-time physical backlash value to the motion control algorithm includes: executing dead zone inverse compensation logic in the position loop control algorithm; when the joint motor is detected to be in motion reversal state or zero speed holding state, determining the width of the compensation bias amount according to the real-time physical backlash value; and superimposing the compensation bias amount onto the position error signal to eliminate control oscillations caused by mechanical backlash.

[0025] By adopting the above technical solution, the dynamically changing mechanical gaps are virtually filled at the control algorithm level, eliminating the dead zone of the transmission chain and ensuring the smoothness and positional accuracy of the motor commutation process.

[0026] Preferably, the process of mapping the liquefaction rate parameter to the motion control algorithm includes: executing a gain scheduling algorithm in the speed loop control algorithm; monitoring whether the liquefaction rate parameter exceeds a stored safety threshold; when the liquefaction rate parameter exceeds the safety threshold, reducing the stiffness parameter of the speed loop proportional-integral controller according to the value of the liquefaction rate parameter to limit the heat generation rate of the system.

[0027] By adopting the above technical solution, flexible thermal management based on energy state is realized. When the phase change material is close to thermal saturation, heat generation is suppressed by reducing dynamic stiffness. Under the premise of ensuring thermal safety, the system shutdown caused by traditional overheat protection is avoided, and the continuous operation capability of the system is maintained.

[0028] Preferably, the process of generating a current loop control command containing multi-source error compensation components includes: calculating the speed loop output based on the position loop output after real-time physical backlash numerical compensation and the speed loop proportional-integral controller after adjusting the stiffness parameter by the liquefaction rate parameter; superimposing the speed loop output result with the friction feedforward compensation component; passing the superimposed signal through a notch filter to suppress mechanical resonance, and using the filtered output as the current loop control command. The process of generating a pulse width modulation signal to drive the joint motor includes: calculating the reference voltage in the rotating coordinate system based on the difference between the current loop control command and the current data in the system state observation vector using a current closed-loop regulation algorithm; performing an inverse coordinate transformation on the reference voltage to obtain the voltage space vector in the two-phase stationary coordinate system; calculating the three-phase bridge arm switch conduction time corresponding to the voltage space vector based on the volt-second balance principle, and generating a complementary pulse width modulation signal with inserted dead time to drive the inverter bridge.

[0029] By adopting the above technical solution, high-frequency response and precise execution of multi-source error compensation signals are achieved, effectively suppressing mechanical resonance and reducing current harmonic losses, thus ensuring the precise output of the motor's electromagnetic torque.

[0030] This invention provides a closed-loop control method for a joint motor with multiple error coupling. It has the following beneficial effects:

[0031] 1. This invention constructs a phase change latent heat state observation mechanism based on energy integration, and calculates the liquefaction rate parameter of the phase change material by utilizing the real-time balance of input electrical power, output mechanical power, and heat dissipation power. This effectively solves the technical problem of traditional temperature sensor monitoring failure caused by the isothermal endothermic phase of the phase change material. It can accurately quantify the remaining heat capacity of the system and dynamically adjust the speed loop stiffness accordingly, thereby preventing thermal runaway while maximizing the short-term overload capacity of the joints and avoiding system shutdown caused by traditional temperature protection mechanisms.

[0032] 2. This invention establishes a dynamic viscosity and back clearance model based on thermal-fluid coupling, which maps slowly changing thermodynamic temperature data into fluid dynamic viscosity values ​​that affect mechanical properties in real time. It also comprehensively considers the effects of metal thermal expansion and oil film thickness to calculate the real-time physical back clearance, thereby realizing the dynamic quantification of nonlinear mechanical errors. Combined with dead zone inverse compensation and friction feedforward strategies, it eliminates transmission clearance fluctuations and viscous resistance caused by changes in temperature and speed conditions, and improves the position tracking accuracy and low-speed stability of the joint motor in the entire temperature range.

[0033] 3. This invention integrates gain scheduling based on liquefaction rate, position compensation based on dynamic backlash, and friction feedforward components based on viscosity into the current loop control command, and combines this with a notch filter to suppress mechanical resonance. This achieves effective decoupling of thermodynamic constraints, fluid disturbances, and mechanical geometric errors, ensuring that the high-frequency pulse width modulation signal can drive the motor to generate electromagnetic torque that accurately cancels internal disturbances. It resolves the contradiction between high dynamic response and high-precision control in high-power-density joints under limited heat dissipation conditions within compact spaces. Attached Figure Description

[0034] Figure 1 A flowchart illustrating the multi-error-coupled joint motor closed-loop control method provided in an embodiment of the present invention;

[0035] Figure 2 This is a characteristic curve of dynamic back clearance changing with temperature calculated in a specific application embodiment of the present invention;

[0036] Figure 3 The graph shows a comparison of the position tracking error between the control method provided in this embodiment of the invention and the traditional PID control method under variable temperature conditions. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] See attached document Figure 1 This invention provides a closed-loop control method for a joint motor with multiple error coupling. This method operates in a high-performance digital signal processor or microcontroller integrated inside the joint, and achieves precise control of a high power density joint through a digital data processing flow.

[0039] The method first performs a synchronous acquisition and preprocessing step S100 of multidimensional heterogeneous data. The digital signal processor (DSP) synchronously triggers and reads data from various sensors within the joint via a parallel data bus interface. This data includes, but is not limited to, rotor position data fed back from the motor-end encoder, load position data fed back from the output-end encoder, high-frequency torque signals fed back from the integrated torque sensor, and temperature data fed back from multiple temperature sensors distributed on the motor coils, drive circuit, and housing surface. Simultaneously, the processor acquires voltage and current data from the DC bus via an analog-to-digital converter interface. The processor performs digital filtering on the acquired raw signals to remove high-frequency electromagnetic noise and performs time-domain alignment on data from different sampling frequencies to construct a system state vector reflecting the complete physical state of the joint system at the current moment.

[0040] Subsequently, the method proceeds to step S200, which involves observing the latent heat state of the phase change material based on energy integration. Addressing the physical characteristic of the phase change material used in the joint cooling system, where the temperature remains constant during the phase change process but its heat capacity is continuously consumed, the processor executes an energy integration algorithm. Based on the input electrical power data, the output mechanical power data, and the heat dissipation power data estimated based on the shell temperature, the processor calculates the net heat flow inside the joint at the current moment. By integrating this net heat flow over discrete time, the processor updates the liquefaction rate variable, which describes the physical state of the phase change material, in real time. This step, through computational means, solves the monitoring blind spot problem where relying solely on temperature sensor data cannot detect the remaining heat capacity of the phase change material, providing a thermodynamic state basis for subsequent control strategy adjustments.

[0041] Next, the method executes step S300, dynamic viscosity and backlash modeling based on thermal-fluid coupling. The processor uses the acquired real-time temperature data and calls the preset lubricating oil viscosity characteristic equation to calculate the dynamic viscosity value of the lubricating oil at the current temperature. Based on this, the processor further updates the mechanical error model parameters. On one hand, the processor uses the dynamic viscosity value to correct the viscous friction coefficient and Coulomb friction torque parameters in the friction model; on the other hand, the processor comprehensively considers the thermal expansion effect of the metal material and the change in lubricating oil film thickness with viscosity and rotational speed, dynamically calculating the real-time physical backlash value of the precision planetary reducer. This step maps thermodynamic and fluid dynamic data to error parameters of the mechanical transmission system, realizing cross-physics model reconstruction.

[0042] Next, the method performs multi-source error decoupling and composite control command generation steps S400. The processor applies the real-time physical backlash value calculated in step S300 to the position loop control algorithm, correcting the position error through dead-zone inverse compensation logic to eliminate control oscillations caused by reducer backlash. Simultaneously, based on the liquefaction rate variable calculated in step S200, the processor dynamically adjusts the proportional-integral parameters of the velocity loop through a gain scheduling algorithm to limit the dynamic stiffness of the system when the phase change material approaches thermal saturation. Furthermore, the processor calculates the friction feedforward compensation component based on the current motion state and, combined with the output of a notch filter designed for the mechanical resonant frequency, superimposes and synthesizes the above components into the final current loop control command.

[0043] Finally, the method executes the high-frequency drive signal generation and output step S500. The processor inputs the synthesized current loop control command to the space vector pulse width modulation algorithm module to calculate the corresponding three-phase bridge arm switch duty cycle. Utilizing the high switching frequency characteristics of the silicon carbide power devices in the drive circuit, the processor outputs a pulse width modulation signal at a high refresh rate, driving the motor coil to generate electromagnetic torque. This step ensures that the control strategy after multi-physics decoupling and compensation can be quickly and accurately converted into the actual physical action of the motor, thereby achieving high-precision, high-response motion control while ensuring a compact joint structure and heat dissipation safety.

[0044] See attached document Figure 1 The multi-error-coupled joint motor closed-loop control method provided by this invention first executes step S100, namely, synchronous acquisition and preprocessing of multi-dimensional heterogeneous data. In this step, a high-performance digital signal processor integrated inside the joint serves as the data processing core. Through direct memory access or a high-speed serial peripheral interface, it triggers the data conversion and transmission processes of each sensor in parallel to ensure strict alignment of different physical quantities in the time dimension. The processor first reads the absolute encoder data located at the motor rotor end and the absolute encoder data located at the reducer output end to obtain the motor rotor angular position and load end angular position, respectively. These two position data are the basis for subsequent calculation of transmission error and identification of backlash. To eliminate the quantization noise of the sensors themselves, the processor performs smoothing filtering on the raw position data and converts the discrete encoder count values ​​into standard radian units.

[0045] While acquiring position data, the processor reads the analog voltage signal from the integrated torque sensor via a high-precision analog-to-digital converter interface. Because the integrated joint contains high-frequency switching devices, the raw signal from the torque sensor is typically superimposed with electromagnetic interference noise. However, the periodic torque ripple generated by the precision planetary gearbox and motor body during operation contains important mechanical characteristic information and cannot be simply filtered out. Therefore, the processor performs a digital filtering algorithm with specific passband characteristics on the torque signal. This algorithm filters out random electromagnetic noise above a preset cutoff frequency while retaining the torque ripple components within a specific frequency range related to the number of motor pole pairs and the reduction ratio. The processed load torque data possesses both a high signal-to-noise ratio and retains the mechanical ripple characteristics required for subsequent precision control.

[0046] In addition, the processor sequentially acquires temperature sensor data distributed at key heat-generating components of the joint via a multiplexed analog-to-digital converter channel. Specifically, the processor reads the values ​​from the thermistors embedded in the motor stator windings, the temperature sensors attached to the servo driver power module substrate, and the temperature sensors located on the joint housing surface. Because the time constant of the thermodynamic process is relatively large, the temperature signals change relatively slowly. The processor uses low-pass filtering on these three sets of temperature data to eliminate transient spike interference in the measurement circuit, thereby obtaining stable data on the internal temperature of the motor windings, the temperature of the driver power module, and the surface temperature of the housing. This temperature data will be directly used for subsequent thermal-fluid coupling model calculations and phase transition state estimation.

[0047] To monitor the system's energy flow, the processor also needs to synchronously acquire DC bus voltage and current data. The processor obtains real-time bus voltage and current values ​​through a voltage sampling circuit and a current sampling resistor or Hall sensor in the drive circuit. Using the differential components of the acquired motor rotor angular position data, the processor calculates the motor's real-time angular velocity and performs a low-pass filter on the angular velocity signal to suppress high-frequency noise introduced by the differentiation operation. At this point, the processor has completed the acquisition and preprocessing of all raw physical data.

[0048] To facilitate unified invocation and computation by subsequent algorithm modules, the processor constructs a unified system state observation vector from the preprocessed and time-domain aligned physical quantities. This vector contains all the key variables describing the current motion, mechanical, and thermodynamic states of the joint system. (System time) State observation vector The definition is as follows:

[0049] ;

[0050] in, This indicates the angular position of the motor rotor after being acquired and converted by the encoder at the motor end; This indicates the load angle position after being acquired and converted by the encoder at the output end; This represents the motor angular velocity after differentiation and filtering. This represents the load torque after filtering. This indicates the internal temperature of the motor windings; This represents the temperature of the driver power module. This state observation vector... This data will be used as input data and passed to subsequent phase change latent heat state observation and heat flow coupling modeling steps.

[0051] See attached document Figure 1 After completing the data acquisition and preprocessing in step S100, the method provided by this invention proceeds to step S200, namely, the observation of the latent heat state of phase change based on energy integration. The core purpose of this step is to solve the problem of temperature detection blind spots in integrated joints that use phase change materials for heat dissipation. Because phase change materials continuously absorb a large amount of latent heat energy during physical phase changes (e.g., from solid to liquid), their own temperature remains almost constant near the phase change point. Under this physical characteristic, traditional overheat protection strategies relying solely on temperature sensor feedback will fail, as a lack of temperature increase does not necessarily indicate sufficient thermal capacity margin in the system. To accurately grasp the thermodynamic state of the joint, the processor runs an energy state observer algorithm to calculate and update the liquefaction rate parameter, which reflects the physical state of the phase change material, in real time.

[0052] The processor first calculates the net thermal power generated by the joint system at the current moment based on the law of conservation of energy. Using the DC bus voltage and current data acquired and synchronized in step S100, the processor calculates the total electrical power input to the joint. Simultaneously, the processor uses the load torque data fed back by the integrated torque sensor and the load angular velocity data fed back by the output encoder (obtained by differentiating the position data) to calculate the mechanical power output by the joint. Considering the conversion losses of the drive circuit and the copper and iron losses of the motor windings, the difference between the input electrical power and the output mechanical power is the power portion converted into heat energy. Based on this, the processor also needs to calculate the heat power naturally dissipated to the external environment through the joint shell. The processor calls the pre-stored thermal resistance model parameters and, based on the temperature difference between the real-time collected shell surface temperature and the preset ambient reference temperature, calculates the heat dissipation power at the current moment using the heat conduction formula. The heat power generated by the system minus the power dissipated to the environment is the heat flow injected into the joint.

[0053] The specific calculation process is executed by the digital signal processor in the discrete time domain. The processor first calculates the current sampling time according to the following formula. The balance between net heating power, mechanical output power, and heat dissipation power is denoted as... :

[0054] ;

[0055] in, Indicates the electrical efficiency coefficient of the drive system; and These represent the DC bus voltage and current at the current moment, respectively. This indicates the load torque at the current moment; This represents the load angular velocity at the current moment.

[0056] Next, the natural heat dissipation power of the processor computing system. This value is approximately equal to the surface temperature of the shell. With ambient temperature The difference divided by the system's equivalent thermal resistance After obtaining the net input heat flow, the processor needs to subtract the sensible heat energy used to raise the temperature of non-phase change components (such as metal stator cores, copper windings, etc.). The remaining energy is the latent heat energy absorbed by the phase change material and used for phase change. The processor calculates and subtracts the sensible heat power by monitoring the rate of change of winding temperature and combining it with the equivalent heat capacity constant of the non-phase change components.

[0057] To quantify the current state of the phase change material, the processor introduces a dimensionless state variable, namely the liquefaction rate parameter, denoted as […]. The value of this variable is strictly limited to between 0 and 1, where 0 represents that the phase change material is completely in a solid state (initial state), and 1 represents that the phase change material is completely in a liquid state (thermally saturated state). The processor uses a discrete-time integral algorithm to accumulate the net heat energy in each control cycle and normalize it to the total latent heat capacity of the phase change material. Current time... liquefaction rate The update algorithm is shown in the following formula:

[0058] ;

[0059] in, This indicates the liquefaction rate at the previous moment; Indicates the sampling period of the control system; This represents the total latent heat energy constant required for the phase change material filling the joint to undergo a complete phase change; This indicates the heat dissipation power at the current moment; This indicates the equivalent specific heat capacity of the non-phase change metal components inside the joint; and These represent the winding temperatures at the current and previous moments, respectively.

[0060] Using the algorithm described above, the processor can sensitively detect the consumption of heat capacity margin during the phase transition plateau period, before the temperature sensor reading shows a significant increase. When the calculated liquefaction rate... As the temperature approaches the threshold of 1, it indicates that the phase change material is about to lose its isothermal heat absorption capacity, and the system is about to face the risk of a temperature surge. This liquefaction rate parameter... As a key intermediate state variable, it will be passed to the subsequent control command generation steps in real time to guide the dynamic adjustment of the speed loop gain, thereby intervening in advance before physical temperature runaway and realizing advanced thermal management and control based on energy state.

[0061] See attached document Figure 1 After completing step S200, which involves observing the latent heat state of phase change based on energy integration, the method provided by this invention continues to execute step S300, namely, dynamic viscosity and back clearance modeling based on thermal-fluid coupling. The main purpose of this step is to utilize the powerful computing capabilities of a digital signal processor to map the collected thermodynamic state data into fluid dynamic parameters and further reconstruct the dynamic model of the mechanical transmission system, thereby achieving real-time quantification and correction of nonlinear mechanical errors.

[0062] During the operation of a kinetic motor, the viscosity characteristics of the lubricating oil inside the reducer have a decisive impact on transmission efficiency, frictional torque, and gear meshing. However, the viscosity of the lubricating oil is not a constant value, but exhibits a significant nonlinear variation with temperature. To accurately obtain the current lubrication state, the processor first reads the filtered motor winding temperature data from step S100 and calculates the dynamic viscosity of the lubricating oil based on this temperature data. The processor internally stores the parameters of the Vogel-Fulcher-Tammann equation describing the viscosity-temperature characteristics of a specific type of lubricating oil. The processor substitutes the real-time temperature into the equation and solves for the dynamic viscosity value at the current moment through exponential operations. This process transforms the originally slowly changing thermodynamic scalar into a fluid dynamic parameter that directly affects mechanical properties, providing a physical benchmark for subsequent friction and backlash model corrections.

[0063] The specific dynamic viscosity calculation process is as follows: The processor calls the floating-point unit to calculate the dynamic viscosity of the lubricating oil at the current moment according to the following formula. :

[0064] ;

[0065] in, This indicates the current motor winding temperature, which is considered the equivalent operating temperature of the lubricating oil inside the reducer. , and All are constants, representing the correlation coefficient, viscosity-temperature coefficient, and viscosity-corrected temperature constant related to the material properties of lubricating oil, respectively.

[0066] After obtaining the real-time dynamic viscosity value, the processor then reconstructs and updates the friction torque model. Traditional control methods typically use a fixed Stribeck friction model, which cannot adapt to the changes in friction characteristics throughout the entire process from low-temperature start-up to high-temperature operation. In this step, the processor constructs a dynamic friction model whose parameters vary with viscosity. Specifically, the processor assumes a positive correlation between Coulomb friction torque and maximum static friction torque and the viscosity of the lubricating oil; that is, the higher the viscosity, the greater the shear resistance under boundary lubrication conditions. Simultaneously, the viscous friction coefficient is also directly proportional to the dynamic viscosity of the fluid. The processor then updates the model based on the current viscosity value. The system updates the Coulomb friction torque parameters, maximum static friction torque parameters, and viscous damping coefficient in real time, and substitutes these updated parameters into the Stribeck model to calculate the accurate friction torque estimate at the current rotational speed.

[0067] Corrected total friction torque The processor calculates this using the following formula:

[0068] ;

[0069] in, Indicates the angular velocity of the motor; This represents Stribeck's velocity constant; Represents a function; This represents the basic viscous damping coefficient. In the formula... and Let Coulomb friction torque and static friction torque be represented as changes in viscosity, respectively. Both are modeled as changes in viscosity. The linear function of the friction torque is calculated from this. It will be used as a feedforward compensation to counteract nonlinear frictional disturbances within the system.

[0070] Besides frictional characteristics, gear backlash in precision planetary gear reducers is also a key nonlinear factor affecting position control accuracy. Traditional control schemes often treat backlash as a fixed value, but in high-power-density joints, the backlash value is significantly drifted due to the dual modulation of thermodynamic and hydrodynamic effects. In this step, the processor establishes a dynamic backlash model that includes thermal expansion and oil film thickness terms. On the one hand, as temperature increases, the metal gears undergo thermal expansion, leading to an increase in tooth profile size and consequently a decrease in tooth flank clearance. On the other hand, as rotational speed and viscosity increase, the thickness of the elastohydrodynamic lubricating oil film formed between the tooth surfaces increases; this oil film wedging effect increases the equivalent meshing clearance. The processor integrates these physical mechanisms to calculate the current equivalent backlash value in real time.

[0071] Dynamic back gap The calculation formula is as follows:

[0072] ;

[0073] in, Indicates at reference temperature The initial geometric back clearance constant was determined below; This represents the coefficient of thermal expansion, which incorporates the properties of the gear material. Indicates the coefficient of fluid dynamic lubricating oil film thickness; This represents the absolute value of the motor's angular velocity. Using this formula, the processor can accurately quantify the actual backlash of the reducer under different temperature and speed conditions. The calculated dynamic backlash value... It will be passed to the subsequent position loop control algorithm to perform dead zone inverse compensation for position error, thereby eliminating system oscillation or steady-state error caused by backlash changes.

[0074] See attached document Figure 1 After calculating the dynamic backlash and dynamic friction torque using a thermal-fluid coupling model, and obtaining the liquefaction rate of the phase change material using an energy integral observer, the method provided by this invention proceeds to step S400, namely, multi-source error decoupling and composite control command generation. In this step, the digital signal processor maps the aforementioned intermediate state parameters across physical fields to various stages of motion control, and synthesizes the final Q-axis current reference command for driving the motor through multi-dimensional decoupling and compensation strategies. This command not only includes the tracking requirements of the target trajectory, but also includes compensation components for mechanical nonlinear errors and adaptive adjustments to thermodynamic boundary conditions.

[0075] The processor first executes a dead-zone inverse compensation algorithm based on dynamic backlash in the position control loop. Because the precision reducer has physical tooth backlash, and this backlash width, as described in step S300, dynamically changes with temperature and lubrication conditions, traditional linear controllers generate unavoidable steady-state errors or zero-crossing oscillations at the moment of motor commutation. The processor reads the real-time dynamic backlash value output in step S300 and substitutes it into the preset dead-zone inverse model. When the system detects an impending change in motion direction or a zero-speed holding state, the processor superimposes a compensation bias in the opposite direction onto the position error signal based on the current dynamic backlash width. This operation virtually fills the mechanical backlash at the control algorithm level, causing the motor to move before the actual gears engage, thus eliminating the dead-zone in the mechanical transmission chain. The compensated equivalent position error signal is input to the position loop proportional controller to generate a reference speed command.

[0076] Subsequently, the processor executes a thermally adaptive gain scheduling algorithm based on liquefaction rate in the speed control loop. To address the heat dissipation bottleneck of high-power-density joints within a compact space, the processor uses the liquefaction rate of the phase change material calculated in step S200 as a scheduling variable to dynamically adjust the stiffness parameters of the proportional-integral controller in the speed loop. The processor internally stores a gain decay curve related to the liquefaction rate. When the liquefaction rate is below a preset safety threshold, it indicates that the phase change material is in a solid state or at a low degree of liquefaction, possessing sufficient latent heat absorption capacity. At this time, the processor maintains a high speed loop gain to ensure the high dynamic response performance of the joint. When the liquefaction rate exceeds the safety threshold and approaches saturation, the processor linearly reduces the proportional and integral coefficients of the speed loop based on the liquefaction rate value. This flexible derating strategy based on energy state can maximize the release of the instantaneous overload capacity of the joint while ensuring that the system does not experience thermal runaway, avoiding the sudden stop or hard cut-off problems caused by traditional temperature protection mechanisms.

[0077] After generating the basic speed closed-loop control torque, the processor further introduces feedforward compensation and filtering to synthesize the final current command. The processor reads the dynamic friction torque value calculated in step S300 based on the VFT equation and Stribeck model, divides it by the motor's torque constant, and converts it into an equivalent friction feedforward current component. This component is directly superimposed on the speed loop output to counteract the viscous resistance of the lubricating oil and the static friction of the mechanical system before the motor generates macroscopic motion, thereby improving the following accuracy in the low-speed domain. Simultaneously, to suppress specific frequency mechanical resonances caused by the compact structure of the integrated joint, the processor passes the superimposed control signal through a second-order infinite impulse response notch filter. The center frequency of this filter is configured as the structural resonant frequency of the joint system, effectively attenuating the resonance-induced components in the control signal.

[0078] In summary, the processor combines the thermally adaptive speed loop output, dynamic friction feedforward, and vibration suppression filtering to synthesize the final Q-axis current reference command according to the following formula. :

[0079] ;

[0080] in, The transfer function operator representing a notch filter; Indicates the change with liquefaction rate The varying stiffness attenuation coefficient has a value range of (0,1]. The operational functions of the speed loop proportional-integral controller; and These represent the reference angular velocity and the feedback motor angular velocity, respectively. This represents the torque constant of the motor; This represents the dynamic friction torque calculated in step S300; This represents the position loop dead zone compensation current term calculated based on dynamic backlash. The current command generated by this formula... It achieves comprehensive decoupling and control of thermodynamic constraints, fluid dynamic disturbances, and mechanical geometric errors.

[0081] See attached document Figure 1 After calculating and generating the final Q-axis current reference command in step S400, the method provided by this invention executes step S500, namely, high-frequency drive signal generation and output. This step is the execution end of the entire control logic, and its main function is to convert the calculation results in the digital domain into physical voltage waveforms that can directly drive the stator windings of the motor. The high-performance digital signal processor uses a built-in current loop controller and space vector pulse width modulation module to convert the decoupled current command into a series of high-frequency switching pulse signals, which are then applied to the motor power stage through the drive circuit.

[0082] The processor first executes the current closed-loop regulation algorithm. The processor reads the Q-axis current reference command, which includes multi-source error compensation components, output in step S400, and simultaneously sets the D-axis current reference command to zero (for surface-mount permanent magnet synchronous motors) or a specific value calculated based on the maximum torque-to-current ratio curve (for embedded permanent magnet synchronous motors). The processor calculates the difference between the reference current and the real-time feedback current acquired and transformed in step S100, i.e., the current error. This current error is then input to the discrete proportional-integral controller. Based on the motor's inductance and resistance parameters, the processor calculates the required D-axis and Q-axis reference voltages for the current control cycle. To improve dynamic response performance, the processor also introduces a cross-coupling decoupling term during voltage calculation to eliminate mutual voltage coupling interference between the D-axis and Q-axis inductors during motor rotation.

[0083] After obtaining the reference voltage in the rotating coordinate system, the processor performs an inverse coordinate transformation. Using real-time rotor position angle data, the processor maps the D-axis and Q-axis reference voltages to a two-phase stationary coordinate system, generating... Shaft reference voltage component and Shaft reference voltage component These two orthogonal voltage components combine to form a rotating voltage space vector, the magnitude and phase of which determine the electromagnetic output state of the motor at the next moment. The processor, based on... shaft and The voltage component of the axis is used to determine the sector position of the target voltage space vector, and the duration of action of the vector on two adjacent basic non-zero voltage vectors and zero voltage vector is calculated.

[0084] The processor calculates the duty cycle of the space vector pulse width modulation signal based on the volt-second balance principle. Specifically, the processor needs to calculate the duty cycle of the pulse width modulation signal within the pulse width modulation period. Within a reference time window, the target reference voltage vector is synthesized. This synthesis process controls the space vectors of two adjacent basic voltages generated by the inverter. , and zero voltage vector The duration of action is determined by the duration of each voltage vector. , and The following linear combination equations must be satisfied:

[0085] ;

[0086] in, Indicates by and Synthesized target reference voltage vector; Indicates the switching cycle; and This represents two adjacent non-zero fundamental voltage space vectors of the sector containing the target vector; and This indicates the duration of conduction of the corresponding fundamental vector; Represents the zero voltage vector; This represents the duration of the zero vector. The processor solves this system of equations to obtain the on-time of each phase and converts it into the corresponding comparator register count value.

[0087] When generating specific gate drive signals, the processor utilizes its internal advanced control timer module to generate three-phase symmetrical center-aligned pulse width modulation (PWM) waveforms. To prevent a short circuit caused by the simultaneous conduction of the upper and lower power switches on the same bridge arm of the inverter, the processor inserts a dead time between the complementary PWM signals. The length of this dead time is strictly set according to the switching characteristic parameters of the power devices. The generated six PWM logic signals are transmitted to the gate drive circuit, which amplifies and levels the signals, ultimately controlling the on / off state of the power stage inverter bridge.

[0088] In this embodiment, the power stage inverter bridge uses silicon carbide metal-oxide-semiconductor field-effect transistors (MOSFETs) as the core switching device. Silicon carbide devices possess extremely low switching losses and excellent thermal conductivity, allowing the processor to set the switching frequency at high frequencies above tens of kilohertz. This high switching frequency characteristic significantly expands the control bandwidth of the current loop, enabling rapid response to the current command containing high-frequency compensation components generated in step S400. Through high-frequency PWM modulation, the current waveform in the motor windings is closer to an ideal sine wave, reducing current harmonic components. This not only reduces electromagnetic heating of the motor but also ensures that the joint motor can generate precise and smooth electromagnetic torque when performing dynamic backlash compensation and friction torque compensation, ultimately achieving high-precision closed-loop control of the joint motion trajectory.

[0089] Specific application examples:

[0090] See attached document Figure 2 In this embodiment, an integrated joint module with a rated power of 400W is selected as the controlled object, and the joint contains a paraffin-based composite phase change material with a phase change temperature of 45℃. System sampling period. The time was set to 100 μs. The processor loaded the following key physical parameters: lubricating oil VFT equation coefficients. =0.05 Pa·s, =800K, =100K; the reducer is at the reference temperature. Initial geometric back clearance at 20℃ The coefficient of thermal expansion of the gear material is set to 35 μrad. Set to 0.6 μrad / ℃; hydrodynamic lubricant film thickness coefficient It is set to 0.006 rad / (Pa·s·rad / s).

[0091] The system operates until the motor winding temperature reaches... The temperature rises to 55°C and the angular velocity At a time when the velocity is 100 rad / s, the processor first calculates the dynamic viscosity of the lubricating oil. ≈8.71 Pa·s. The dynamic back clearance was then calculated using the formula. =35-0.6×(55-20)+0.006×8.71×100. Among them, the thermal expansion effect causes the back clearance to decrease by 21μrad, while the oil film effect provides support of about 5.23μrad. The final calculated real-time back clearance is about 19.23μrad.

[0092] like Figure 2 As shown, the chart background uses different shades of gray to visually divide two stages with vastly different physical characteristics:

[0093] The darker gray background on the left corresponds to the low-temperature region (approximately 10-40℃). In this region, the lubricating oil viscosity is extremely high, the oil film support effect is significant, and the curve exhibits a steep non-linear descent characteristic, with the back clearance value rapidly decreasing from over 80 μrad at 10℃.

[0094] The lighter gray background on the right corresponds to the high-temperature region (approximately 40-80℃). Within this region, viscosity changes more slowly, the thermal expansion effect of the metallic material dominates, and the back gap decreases almost linearly with temperature. The processor, based on this full-temperature-range model, distinguishes the current temperature region in real time and outputs precise compensation parameters, avoiding control errors caused by a single model.

[0095] Simultaneously, the processor performs energy integration observation in step S200. When the liquefaction rate λ reaches 0.8, the stiffness decay logic in step S400 is triggered to actively reduce the velocity loop gain to prevent thermal runaway.

[0096] Experimental verification and effect comparison:

[0097] See attached document Figure 3 The experiment was set up so that the joint motor could perform a 2Hz sinusoidal trajectory tracking task, and the temperature would rise linearly from 20℃ to 70℃ within 10 seconds. Figure 3 The vertical axis is displayed in scientific notation (×10). -3 (rad), that is, a scale of 0.5000 represents 0.0005 rad.

[0098] The diagram clearly identifies the comparison objects: the traditional PID control method (gray dashed line) and the multi-error coupling control method of this invention (black solid line).

[0099] The two horizontal dashed lines in the figure are marked as error control zones, and their corresponding vertical axis scale is accurate to ±0.5000×10. -3 rad (i.e. ±0.0005 rad).

[0100] Observing the gray dashed line representing the traditional method, as time progresses (temperature increases), its fluctuation amplitude gradually exceeds the error control band, especially with a sharp peak appearing at the zero-crossing point of the waveform. The traditional method exhibits zero-crossing oscillation divergence: due to the lack of dynamic compensation for thermal and fluid coupling backlash, the traditional controller cannot suppress mechanical oscillations caused by sudden gap changes, and the maximum error peak value approaches 3.0000 × 10⁻⁶. -3 .

[0101] Conversely, the solid black line representing the method of this invention is always confined within the range of the two auxiliary lines of the error control band, indicating that even under high temperature and variable friction conditions, the system can still maintain micron-level high-precision tracking, verifying the effectiveness of the multi-error coupling control model.

Claims

1. A closed-loop control method for a joint motor with multiple error couplings, characterized in that, Includes the following steps: Collect data from inside the joint, read multi-dimensional sensor data through synchronous triggering and perform temporal alignment to construct a system state observation vector; Based on the system state observation vector, an energy integration algorithm is executed to calculate the liquefaction rate parameter, which reflects the physical state of the phase change material. The dynamic viscosity of the lubricating oil is calculated using the temperature data in the system state observation vector. The friction model parameters are updated based on the dynamic viscosity value to calculate the friction feedforward compensation component. A thermal-fluid coupling model is constructed based on the dynamic viscosity value to calculate the real-time physical back clearance value. The real-time physical backlash value, the friction feedforward compensation component, and the liquefaction rate parameter are mapped into the motion control algorithm to generate a current loop control command containing multi-source error compensation components. The current loop control command generates a pulse width modulation signal to drive the joint motor.

2. The closed-loop control method for a joint motor with multiple error coupling according to claim 1, characterized in that, The process of acquiring internal joint data, performing temporal alignment, and constructing a system state observation vector includes: Simultaneously read the motor rotor angular position, load end angular position, load torque, and temperature data distributed at the joint heating parts; Collect voltage and current data from the DC bus; Calculate the motor angular velocity based on the differential value of the motor rotor angular position; The motor rotor angular position, motor angular velocity, load end angular position, load torque, temperature data, voltage data, and current data are digitally filtered to remove high-frequency noise, and the processed data are combined into the system state observation vector.

3. The closed-loop control method for a joint motor with multiple error coupling according to claim 2, characterized in that, The process of calculating the liquefaction rate parameter, which reflects the physical state of the phase change material, includes: The input electrical power is calculated based on the voltage and current data in the system state observation vector; the output mechanical power is calculated based on the differential values ​​of the load torque and the load end angle position; and the heat dissipation power is estimated based on the temperature data. Calculate the net heat flow after subtracting the output mechanical power and the heat dissipation power from the input electrical power; The net heat flow is accumulated and integrated in the discrete time domain to obtain the total net heat energy, and the sensible heat energy used for heating non-phase change components is removed from the total net heat energy to obtain the latent heat energy. The latent heat energy is normalized to the total latent heat capacity of the phase change material to obtain the liquefaction rate parameter.

4. The closed-loop control method for a joint motor with multiple error coupling according to claim 1, characterized in that, The process of calculating the dynamic viscosity value of the lubricating oil includes: The motor winding temperature data is extracted from the system state observation vector as the equivalent operating temperature of the lubricating oil; The stored viscosity-temperature characteristic equation of lubricating oil is invoked, and the equivalent working temperature is substituted into the viscosity-temperature characteristic equation of lubricating oil. The dynamic viscosity value at the current moment is obtained by solving the equation through exponential operation.

5. The closed-loop control method for a joint motor with multiple error coupling according to claim 1, characterized in that, The process of calculating the real-time physical backlash value includes: A dynamic back clearance model incorporating thermal expansion and oil film thickness terms is established. Calculate the reduction in tooth backlash due to thermal expansion of the metal gear based on the temperature data; The equivalent clearance increase caused by the hydrodynamic lubricating oil film is calculated based on the dynamic viscosity value and the motor angular velocity in the system state observation vector. The real-time physical backlash value is calculated by superimposing the initial geometric backlash, the reduction in tooth flank clearance, and the increase in equivalent clearance.

6. The closed-loop control method for a joint motor with multiple error coupling according to claim 2, characterized in that, The process of updating the friction model parameters based on the dynamic viscosity value to calculate the friction feedforward compensation component includes: Based on the dynamic viscosity value, the Coulomb friction torque parameter, the maximum static friction torque parameter, and the viscous damping coefficient in the friction model are corrected. Substitute the corrected Coulomb friction torque parameters, the maximum static friction torque parameters, the viscous damping coefficient, and the motor angular velocity in the system state observation vector into the friction model to calculate the estimated friction torque value at the current moment. The estimated friction torque is converted into the friction feedforward compensation component.

7. The closed-loop control method for a joint motor with multiple error coupling according to claim 1, characterized in that, The process of mapping the real-time physical backlash value to the motion control algorithm includes: Execute dead-time inverse compensation logic in the position loop control algorithm; When the joint motor is detected to be in a motion reversal state or a zero-speed holding state, the width of the compensation offset is determined according to the real-time physical backlash value. The compensation bias is superimposed on the position error signal to eliminate control oscillations caused by mechanical backlash.

8. The closed-loop control method for a joint motor with multiple error coupling according to claim 1, characterized in that, The process of mapping the liquefaction rate parameter to the motion control algorithm includes: Execute the gain scheduling algorithm within the speed loop control algorithm; Monitor whether the liquefaction rate parameter exceeds the stored safety threshold; When the liquefaction rate parameter exceeds the safety threshold, the stiffness parameter of the speed loop proportional-integral controller is reduced according to the value of the liquefaction rate parameter to limit the heat generation rate of the system.

9. The closed-loop control method for a joint motor with multiple error coupling according to claim 1, characterized in that, The process of generating current loop control commands that include multi-source error compensation components includes: Based on the position loop output after real-time physical backlash numerical compensation, and the velocity loop proportional-integral controller after adjusting the stiffness parameter by the liquefaction rate parameter, the velocity loop output result is calculated. The output of the velocity loop is superimposed with the friction feedforward compensation component; The superimposed signal is passed through a notch filter to suppress mechanical resonance, and the filtered output is used as the current loop control command.

10. The closed-loop control method for a joint motor with multiple error coupling according to claim 1, characterized in that, The process of generating a pulse width modulation signal to drive the joint motor includes: Based on the difference between the current loop control command and the current data in the system state observation vector, the reference voltage in the rotating coordinate system is calculated using the current closed-loop regulation algorithm. Perform an inverse coordinate transformation on the reference voltage to obtain the voltage space vector in the two-phase stationary coordinate system; The on-time of the three-phase bridge arm switches corresponding to the voltage space vector is calculated based on the volt-second balance principle, and a complementary pulse width modulation signal with inserted dead time is generated to drive the inverter bridge.

Images