Brushless motor multi-mode control method based on humanoid robot effector

Abstract

The application discloses a brushless motor multi-mode control method based on a humanoid robot actuator and belongs to the technical field of humanoid robot actuator control, and solves the technical problems that the current characteristic extraction dimension of the existing method is single, abnormality determination lacks work condition adaptation, and waveform correction lacks self-adaptation. The method first collects and pre-processes three-phase current, rotor position and control mode signals, extracts current time-frequency domain core features and associates work condition information; then divides motor core control modes and typical load work conditions, constructs a multi-mode split-phase current reference feature library, and determines an abnormal phase of the current waveform in combination with real-time work conditions; finally, a deviation gradient and a correction amount are calculated, the abnormal phase is independently pre-corrected according to a self-adaptive coefficient, a cycle iteration correction process is realized, and after a termination condition is met, a conventional control mode is switched, the rationality of abnormality determination is improved, dynamic self-adaptive correction of the current waveform is completed, motor operation noise and power consumption are reduced, and the operation stability and robustness of the motor are improved.

Description

Technical Field

[0001] This invention belongs to the field of humanoid robot actuator control technology, specifically relating to a multi-mode control method for brushless motors based on humanoid robot actuators. Background Technology

[0002] The brushless motors of humanoid robot actuators have stringent requirements for operational stability, low noise, and low power consumption. Existing control methods have significant shortcomings in current signal feature extraction, anomaly detection for operating condition adaptation, and dynamic adaptive waveform correction, making it difficult to match the actual operating requirements of the motor. Therefore, this invention urgently needs to solve the following technical problems: The feature extraction of motor current signals is based on a single dimension, without being associated with operating condition information, making it difficult to comprehensively characterize the motor current operating status under different operating conditions and lacking effective data support. Without dividing the working conditions according to the actual scenario and establishing corresponding current benchmarks, the judgment standard is uniform, which is prone to misjudgment, cannot adapt to the differences in working conditions, and is difficult to accurately identify abnormal phases of current waveform; The current waveform correction parameters are fixed, there is no adaptive mechanism, it is easy to make improper corrections, there is no reasonable termination condition, and it is easy to interfere with the normal phase, increase energy consumption and noise, and affect the stability of motor operation. To address this, we propose a multi-mode control method for brushless motors based on humanoid robot actuators. Summary of the Invention

[0003] The purpose of this invention is to provide a multi-mode control method for brushless motors based on humanoid robot actuators, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a multi-mode control method for brushless motors based on humanoid robot actuators, comprising: Step 1: Periodically acquire and preprocess the three-phase current, rotor position, and control mode signals; extract window-level time-domain core features of the three-phase current by phase, and perform frequency domain analysis to extract high-frequency disturbance-related features; finally, organize the phase current feature vectors and associate them with the corresponding control mode and rotor speed signals. Step 2: Divide the core control modes of the brushless motor and their corresponding typical load conditions, collect three-phase current reference samples under each combination of normal operation, and extract window-level core current features by phase; determine the core reference value through statistical analysis, construct and integrate it into a multi-mode phase current reference feature library; identify the operating condition combination in real time, retrieve the corresponding core reference value, extract the actual features, calculate the normalized error feature value, and complete the determination of abnormal phases of current waveform. Step 3: Calculate the deviation gradient value of the abnormal phase of the current waveform, and calculate the corresponding current characteristic correction amount by combining the normalized error characteristic value. Perform pre-correction on the initial phase current control command of the next window and execute it. Iterate this process repeatedly. After the termination condition is met, switch to the normal control mode and terminate the recursive adaptive correction process.

[0005] Preferably, the specific process for extracting window-level time-domain core features of three-phase currents by phase is as follows: First, preset the fundamental period of the brushless motor current, then adapt and set the length of the sliding analysis time window and the sliding step size, and keep the timing consistency between the sliding step size and the motor current loop control cycle. The sampling frequency is determined based on the fundamental period, and the three-phase current, rotor position and actuator control mode signals of the brushless motor are periodically collected and pre-processed by low-pass filtering. For each sliding analysis time window, the preprocessed three-phase current signal is extracted independently into window-level time-domain core features for phases A, B, and C. The time-domain core features include: the effective value of the phase fundamental current, the phase total harmonic distortion rate, and the phase current ripple coefficient.

[0006] Preferably, the specific process of performing frequency domain analysis on the three-phase current to extract high-frequency disturbance-related features, and finally organizing the phase current feature vectors and associating the corresponding control modes with the rotor speed signals is as follows: The three-phase current signals after preprocessing within the sliding analysis time window are analyzed in frequency domain independently by phase. The signal sampling frequency is preset and the frequency domain spectrum of the corresponding phase is generated based on the fast Fourier transform. The spectrum data within the preset frequency analysis range is locked. The high-frequency harmonic components are determined according to the preset spectrum amplitude threshold. After smoothing the spectrum data, the random fluctuation components are determined according to the preset spectrum smoothing difference threshold. The high-frequency harmonic amplitude and the proportion of fluctuation components of the corresponding phase current are extracted. The core time-domain and frequency-domain characteristics of the three phases within the sliding analysis time window are organized to form the phase current characteristic vector, which is then synchronously associated with the control mode signal and rotor speed signal of the current window.

[0007] Preferably, the specific process of dividing the core control modes of the brushless motor and their corresponding typical load conditions, collecting three-phase current reference samples under normal operation in each combination, and extracting window-level core current characteristics by phase is as follows: For the core application scenarios of brushless motors for joint actuators in humanoid robots, the core control modes are divided, and all typical load conditions corresponding to each control mode are sorted out. For each control mode-load condition combination, three-phase current data of the brushless motor under normal and abnormal operation under that combination are collected in several independent sliding analysis time windows as the reference sample set for that combination. For the current data of each normal sliding analysis time window in the benchmark sample set, the core current features at the window level are extracted independently by phase. The extracted features include: effective value of phase fundamental current, phase total harmonic distortion rate, phase current ripple coefficient, phase high-frequency harmonic amplitude, and proportion of phase fluctuation components.

[0008] Preferably, the specific process of determining the core reference value through statistical analysis and constructing and integrating it into a multi-mode phase current reference feature library is as follows: For all normal sliding analysis time window reference samples under the same control mode and the same load condition, statistical analysis is performed on the core current characteristics of each window level according to phase. The average value of each characteristic in all reference samples is calculated as the core reference value of the corresponding phase, control mode and load condition. Based on the dispersion of the sample characteristic values, the normal fluctuation range of each characteristic around the core reference value is determined by standard deviation and used as an auxiliary constraint value. Based on the core reference values ​​and auxiliary constraint values ​​of each phase, a phase current reference characteristic table corresponding to the control mode-load condition is constructed; A multi-mode phase current reference feature library is formed by compiling a table of phase current reference features corresponding to all load conditions under the three control modes of position, speed, and torque. This library establishes a one-to-one correspondence between control mode, load condition, and phase feature. Each control mode corresponds to all load conditions to which it belongs, and each load condition corresponds to the core reference value and auxiliary constraint value of each of the three phases A, B, and C.

[0009] Preferably, the specific process of real-time identification of operating condition combinations, retrieval of corresponding core reference values, extraction of actual features, calculation of normalized error feature values, and completion of the determination of abnormal phases in the current waveform is as follows: The system identifies the control mode and actual operating load conditions of the brushless motor in the current sliding analysis time window in real time, and retrieves the core reference values ​​of phase A, phase B, and phase C that are precisely matched with the control mode and load conditions from the multi-mode phase current reference feature library. Based on the three-phase current data collected and preprocessed within the current sliding analysis time window, the core current characteristics at the window level are extracted independently for each phase to obtain the actual fundamental current effective value, actual total harmonic distortion rate, actual current ripple coefficient, actual high-frequency harmonic amplitude, and actual fluctuation component ratio for each phase. Calculate the normalized error characteristic value of each phase within the current sliding analysis time window according to phase, and quantify the degree of deviation between the actual core current characteristics and the core reference value. A preset threshold for the normalized error characteristic value is set for each phase. The normalized error characteristic value of each phase in the current window is compared with the corresponding threshold. If the normalized error characteristic value is not lower than the corresponding threshold, the phase is determined to be an abnormal current waveform phase and marked as a phase to be corrected. When the normalized error eigenvalue is lower than the corresponding threshold, the current waveform of the phase is determined to be normal and no subsequent gradient calculation and correction operation is performed.

[0010] Preferably, the specific process of calculating the deviation gradient value of the abnormal phase of the current waveform and combining it with the normalized error characteristic value to calculate the corresponding current characteristic correction amount is as follows: For each phase to be corrected, the finite difference method in the field of numerical differentiation is used to calculate the deviation gradient value of each anomalous phase; Based on the normalized error characteristic value and deviation gradient value of each abnormal phase in the current window, the current characteristic correction amount of each abnormal phase in the next sliding analysis time window is calculated independently by phase. The adaptive proportional coefficient and adaptive integral coefficient of each abnormal phase in the current characteristic correction are both monotonic adaptive functions of the normalized error characteristic value of the corresponding abnormal phase before the window. The upper and lower limits of the two types of coefficients and the threshold of the normalized error characteristic value are preset. The coefficients are adjusted by using a linear mapping rule. When the normalized error characteristic value increases, the two types of coefficients increase synchronously, and when it decreases, they decrease synchronously.

[0011] Preferably, the specific process for pre-correcting and executing the initial phase current control command for the next window is as follows: Before the current sliding analysis time window ends and the next sliding analysis time window begins, based on the abnormal phases identified in the current window and the corresponding calculated correction values ​​for the specific current characteristics of each abnormal phase, the initial phase-by-phase current control commands corresponding to each abnormality in the next window are pre-corrected. Specifically: First, retrieve the initial phase current control commands for each of the three phases in the next window. For abnormal phases, add the corresponding current characteristic correction amount to their initial commands to generate corrected commands. For normal phases, directly use the initial commands. Set a preset limit constraint on the corrected phase current control commands for all phases to limit them to the preset safe current command range. The final phase-by-phase current control command for each phase is input into the current loop controller of the brushless motor. The controller generates the corresponding drive signal for each phase, and drives the motor to run according to the corrected command in the next window, thereby realizing the pre-correction of the abnormal phase current waveform.

[0012] Preferably, this process is iterated repeatedly. After the termination condition is met, the system switches to the normal control mode. The specific process of terminating the recursive adaptive correction process is as follows: After the current sliding analysis time window finishes running, immediately collect the three-phase raw current data within the window, and sequentially perform the following operations: extract the core current characteristics of the phase window level, retrieve the core reference value of the phase that matches the current control mode and load condition, calculate the normalized error characteristic value of each phase, determine whether the current waveform of each phase is abnormal, and calculate the corresponding deviation gradient value for the phase marked as abnormal. If there is an abnormal phase of current waveform in the window, the current characteristic correction amount of the corresponding abnormal phase in the next sliding analysis time window is calculated and generated based on the deviation gradient value corresponding to each abnormality, and the pre-correction of the initial current control command of the next window is completed by phase. If all three phases within this window are determined to be without abnormality, then no current correction will be performed in the next window, and the standard current control command corresponding to the reference characteristic will be directly used. The process is iterated in cycles until the brushless motor exits the current control mode and stops running, or until the current waveforms of each phase are determined to be normal within multiple consecutive sliding analysis time windows. Then, the system automatically switches to the normal control mode, terminates the recursive adaptive correction process, and subsequently only uses the standard current control command corresponding to the reference feature.

[0013] Compared with the prior art, the beneficial effects of the present invention are: (1) The brushless motor multi-mode control method based on humanoid robot actuator preprocesses the three-phase current and other signals by low-pass filtering, extracts the window-level time domain core features by phase, and simultaneously performs frequency domain analysis to extract high-frequency disturbance related features. Then, it organizes the phase current feature vectors and associates them with the control mode and rotor speed signal, realizing the time domain and frequency domain dual-dimensional full-band characterization of current features. It establishes the mapping relationship between motor operating conditions and current features, solves the problem of single dimension of current signal feature extraction and lack of association with operating condition information, and provides comprehensive and reliable core data support for subsequent current waveform anomaly judgment and correction, effectively improving the operating condition adaptability and operational robustness of the control method.

[0014] (2) The brushless motor multi-mode control method based on humanoid robot actuator divides the core control modes and corresponding typical load conditions of brushless motor by combining the actual application scenarios of humanoid robot joint actuator, collects current reference samples under each working condition combination and extracts core features, determines the core reference value and auxiliary constraint value through statistical analysis, constructs a multi-mode phase current reference feature library, identifies the current working condition of the motor in real time and retrieves the corresponding reference, calculates the normalized error feature value to complete the abnormal phase judgment, adapts to the differences in different operating conditions of the motor, effectively defines the normal fluctuation range of current characteristics, solves the problem of no working condition adaptation reference and unified judgment standard which is easy to misjudge, reduces invalid correction operations, improves the rationality of current waveform abnormal judgment, and realizes effective identification of abnormal phase.

[0015] (3) The brushless motor multi-mode control method based on humanoid robot actuator calculates the current characteristic correction by combining the deviation gradient value and the normalized error characteristic value, sets the adaptive proportional coefficient and integral coefficient that are dynamically adjusted with the error characteristic value, performs phase-by-phase independent pre-correction and amplitude limiting constraint on the abnormal phase of the next window, and designs a cyclic iterative correction process and dual termination conditions to realize the adaptive dynamic correction of the current waveform. It solves the problems of fixed correction parameters, easy improper correction, and easy interference with the normal phase. It avoids over-correction or under-correction, and does not cause unnecessary impact on the current control of the normal phase. It can also reduce the harmonic loss of the motor winding, reduce electromagnetic noise and mechanical vibration noise, reduce the useless power consumption during motor operation, and improve the stability and reliability of motor operation, which meets the low noise and low power consumption operation requirements of humanoid robot actuator. Attached Figure Description

[0016] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0017] The technical solutions of 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.

[0018] Example 1; Please see Figure 1 This invention provides a multi-mode control method for a brushless motor based on a humanoid robot actuator, comprising: Step 1: Periodically acquire and preprocess the three-phase current, rotor position, and control mode signals; extract window-level time-domain core features of the three-phase current by phase, and perform frequency-domain analysis to extract high-frequency disturbance-related features; finally, organize the phase-by-phase current feature vectors and associate them with the corresponding control mode and rotor speed signals. The specific process is as follows: The fundamental period of the brushless motor current is preset to T. Based on this, the window length and sliding step size of the sliding analysis time window are preset. The window length is adapted to the fundamental period T, and the sliding step size is kept in time sequence consistency with the motor current loop control cycle. Set a sliding analysis time window, determine the sampling frequency based on the fundamental period T, and the number of current sampling points within a single fundamental period T is N. Periodically collect the three-phase current signal, rotor position signal, and actuator control mode signal sent by the host computer of the brushless motor, and perform low-pass filtering preprocessing to obtain the preprocessed three-phase current signal, rotor position signal and control mode signal. For each sliding analysis time window, the preprocessed three-phase current signal is used to independently extract window-level time-domain core features for phases A, B, and C. Specifically, these include: The effective value of the phase fundamental current is calculated using the following formula: , in, This represents the effective value of the fundamental current of the i-th phase. N: Number of current sampling points within a single fundamental frequency period T; j: Sampling point index; This represents the instantaneous value of the i-th phase current at the j-th sampling point; Phase-specific total harmonic distortion (THD): The specific formula is as follows: , in, Let be the total harmonic distortion rate of the i-th phase current; i is the phase sequence index, with values ​​of 1, 2, and 3, which correspond to the three-phase currents of the brushless motor, namely phase A, phase B, and phase C, respectively. k is the harmonic order index, k=1 represents the fundamental frequency, and k≥2 represents each higher harmonic; n is the upper limit of the highest order of harmonic analysis; is the effective value of the kth harmonic component in the i-th phase current; Phase current ripple coefficient: The specific formula is as follows: , in, Let be the ripple coefficient of the i-th phase current; This represents the effective value of the i-th phase current. The average value of the i-th phase current; For the preprocessed three-phase current signals within the sliding analysis time window, frequency domain analysis is performed independently for each phase to extract high-frequency disturbance correlation features. The specific operation is as follows: The signal sampling frequency is preset, and the frequency domain conversion of the three-phase current signal within the sliding analysis time window is performed based on the fast Fourier transform to generate the frequency domain spectrum of the corresponding phase. The preset frequency analysis range is used to lock the spectral data within the preset frequency analysis range in the converted frequency domain spectrum; A preset spectral amplitude threshold is used to identify frequency components with spectral amplitudes greater than the preset spectral amplitude threshold within the locked range as high-frequency harmonic components. A preset spectrum smoothing difference threshold is used to smooth the spectrum data within the locked range, and components that deviate from the smoothing baseline and have a difference greater than the preset spectrum smoothing difference threshold are identified as random fluctuation components. For high-frequency harmonic components, the maximum value of the frequency domain amplitude of each component is taken to obtain the high-frequency harmonic amplitude of the corresponding phase current within the current window, which characterizes the intensity of high-frequency harmonic disturbance. For the random fluctuation component, calculate the ratio of its energy to the total energy of the corresponding phase current signal to obtain the proportion of the fluctuation component of the corresponding phase current within the current window. The calculation formula is as follows: , in, The proportion of the fluctuation component of the i-th phase current; This is the set of frequency indices corresponding to the random fluctuation component of the i-th phase current. Let f be the frequency domain amplitude of the f-th frequency component of the i-th phase current; This represents the total number of frequency points in the frequency domain analysis of the i-th phase current. Within the sliding analysis time window, the core time-domain and frequency-domain features (effective value of phase fundamental current, total harmonic distortion rate of phase, ripple coefficient of phase current, amplitude of high-frequency harmonics, and proportion of fluctuation components) extracted from phases A, B, and C are organized to form corresponding phase current feature vectors, which are then synchronously associated with the control mode signal and rotor speed signal of the current window.

[0019] It should be noted that by adapting the sliding analysis time window to the fundamental period and keeping the window sliding step size consistent with the motor current loop control cycle, coupled with the low-pass filtering signal preprocessing method, redundant noise in the signal is effectively suppressed, ensuring the accuracy and timing synchronization of the three-phase current, rotor position and control mode signal acquisition, and laying a stable and reliable data foundation for subsequent feature extraction work. By independently extracting the effective value of the phase fundamental current, the phase total harmonic distortion rate, and the phase current ripple coefficient, etc., window-level time-domain core features can be extracted, and the time-domain waveform characteristics of the current of each phase of the brushless motor can be comprehensively characterized from multiple dimensions such as amplitude accuracy, harmonic distortion degree, and ripple level, so as to objectively quantify the time-domain operating state of the current. A phase-independent frequency domain analysis is performed on the preprocessed current signal. By defining high-frequency harmonic components and random fluctuation components through multiple thresholds and extracting corresponding features, the limitations of single time domain analysis are overcome. This can effectively capture the degrading effects of high-frequency disturbances and random noise on the current waveform, and achieve full-band, multi-dimensional characterization of the current waveform of each phase of the motor. By integrating the core time-domain and frequency-domain features of each phase into a phase-by-phase current feature vector and synchronously associating it with the corresponding control mode signal and rotor speed signal, a mapping relationship between motor operating conditions and current characteristics is effectively established. This provides comprehensive and reliable core data support for subsequent construction of a multi-mode phase-by-phase current reference feature library, real-time identification of operating condition combinations, determination of abnormal phases in current waveforms, deviation gradient calculation, and current feature correction. This enhances the operating condition adaptability and operational robustness of the entire control method and lays a solid data foundation for the targeted implementation of subsequent control strategies.

[0020] Step Two: Divide the brushless motor's core control modes and corresponding typical load conditions; collect three-phase current reference samples under each combination of normal operation; extract window-level core current features by phase; determine the core reference value through statistical analysis; construct and integrate a multi-mode phase-by-phase current reference feature library; identify the operating condition combination in real time, retrieve the corresponding core reference value, extract actual features, calculate normalized error feature values, and complete the determination of abnormal phases in the current waveform. The specific process is as follows: For the core application scenarios of brushless motors in humanoid robot joint actuators, core control modes are defined. For each control mode, all typical load conditions are identified to achieve full coverage of the actual operating scenarios of brushless motors under each control mode. Among them, the control modes correspond one-to-one with the control commands issued by the host computer, and the load conditions are divided based on the real-time operating status of the motor and the load level. For each control mode-load condition combination, three-phase current data of the brushless motor under normal and abnormal operation in that combination are collected in several independent sliding analysis time windows as the reference sample set for that combination. For each normal sliding analysis time window current data in the benchmark sample set, window-level core current features are extracted independently by phase. The extracted features include: effective value of phase fundamental current, phase total harmonic distortion rate, phase current ripple coefficient, phase high-frequency harmonic amplitude, and phase fluctuation component proportion. For all normal sliding analysis time window reference samples under the same control mode and the same load condition, statistical analysis was performed on the core current characteristics of each window level by phase: Calculate the average value of each feature across all benchmark samples, and use this average value as the core benchmark value of the feature under the corresponding phase, control mode, and load condition to characterize the typical level of the feature under normal operating conditions. Based on the dispersion of sample feature values, the normal fluctuation range of each feature around its core benchmark value is determined by standard deviation quantization. This normal fluctuation range is used as the auxiliary constraint value of the corresponding feature. This processing method is a conventional method in the field of signal feature analysis, which can effectively define the normal fluctuation range of features and avoid abnormal misjudgments caused by small fluctuations. Based on the core reference values ​​and auxiliary constraint values ​​calculated for each phase, a phase-by-phase current reference characteristic table corresponding to the load condition under this control mode is constructed. The characteristic table independently records the average effective value of the fundamental current, the normal fluctuation range of the effective value of the fundamental current, the average total harmonic distortion rate, the normal fluctuation range of the total harmonic distortion rate, the average current ripple coefficient, the normal fluctuation range of the current ripple coefficient, the average high-frequency harmonic amplitude, the normal fluctuation range of the high-frequency harmonic amplitude, the average fluctuation component ratio, and the normal fluctuation range of the fluctuation component ratio under the corresponding load condition. The phase current reference feature tables corresponding to all load conditions under the three control modes of position, speed and torque are summarized and integrated to form a multi-mode phase current reference feature library. This feature library establishes a one-to-one correspondence between control mode, load condition and phase feature. Each control mode corresponds to all load conditions to which it belongs, and each load condition corresponds to the core reference value and auxiliary constraint value of each of the three phases A, B and C. The system identifies the control mode and actual operating load conditions of the brushless motor in the current sliding analysis time window in real time, and retrieves the core reference values ​​of phase A, phase B, and phase C that are precisely matched with the control mode and load conditions from the multi-mode phase current reference feature library. Based on the three-phase current data collected and preprocessed within the current sliding analysis time window, the core current characteristics at the window level are extracted independently for each phase to obtain the actual fundamental current effective value, actual total harmonic distortion rate, actual current ripple coefficient, actual high-frequency harmonic amplitude, and actual fluctuation component ratio for each phase. The normalized error characteristic value of each phase within the current sliding analysis time window is calculated according to phase to quantify the deviation between the actual core current characteristics and the core reference value (all parameters involved in the formula have been normalized and dimensionless). The calculation formula is as follows: ; in, Let be the normalized error eigenvalue of the i-th phase within the sliding analysis time window; This represents the effective value of the fundamental current of the i-th phase during the sliding analysis time window. It is the core reference value of the fundamental current RMS value of the i-th phase current under the corresponding mode and operating condition; Let be the actual total harmonic distortion rate of the i-th phase current within the sliding analysis time window; It is the core benchmark value for the effective value of the total harmonic distortion rate of the i-th phase current under the corresponding mode and operating conditions; The actual high-frequency harmonic amplitude of the i-th phase current within the sliding analysis time window; It is the core reference value for the effective value of the high-frequency harmonic amplitude of the i-th phase current under the corresponding mode and operating condition; The actual current ripple coefficient of the i-th phase current during the sliding analysis time window; The core reference value for the current ripple coefficient of the i-th phase current under the corresponding mode and operating condition. This represents the proportion of the actual fluctuation component of the i-th phase current within the sliding analysis time window; This is the core reference value for the proportion of the fluctuation component of the i-th phase current under the corresponding mode and operating condition; a1, a2, a3, a4, and a5 are preset weighting coefficients, and a1+a2+a3+a4+a5=1; A preset threshold for normalized error eigenvalues ​​is established for each phase. The normalized error eigenvalues ​​of each phase in the current window are compared with the corresponding thresholds to determine current waveform anomalies. If the normalized error characteristic value of the i-th phase is greater than or equal to the corresponding preset threshold, then the phase is determined to be an abnormal current waveform phase and marked as a phase to be corrected. If the normalized error characteristic value of the i-th phase is less than the corresponding preset threshold, then the current waveform of that phase is determined to be normal, and no subsequent gradient calculation and correction operation is performed.

[0021] It should be noted that by dividing the core control modes based on the core application scenarios of the brushless motor of the humanoid robot joint actuator and sorting out the typical load conditions under each control mode, the actual operating scenarios of the motor are fully covered, so that the establishment of subsequent benchmark features can closely match the actual application requirements of the motor and make the benchmark more scenario-adaptable. For each control mode-load condition combination, current reference samples under normal operating conditions are collected and core current characteristics are extracted by phase. This provides basic data that fits the actual operating state of the motor for subsequent statistical analysis, so that the determination of the core reference value has real data support and avoids the disconnect between the reference and the actual operating conditions. By determining the core benchmark values ​​of each feature through statistical analysis and using the normal fluctuation range of the feature based on the standard deviation as an auxiliary constraint value, the normal fluctuation range of the feature can be effectively defined, avoiding misjudgment of anomalies caused by minor fluctuations, improving the rationality and accuracy of subsequent current waveform anomaly judgment, and reducing invalid correction operations. A multi-modal phase current reference feature library is constructed and a correspondence between control mode, load condition and phase feature is established, so that the current features of the motor under different operating conditions have corresponding references, providing a clear and systematic basis for reference retrieval after real-time condition identification, and enabling the reference retrieval to match the current operating condition of the motor. By identifying the current operating condition of the motor in real time and retrieving the corresponding phase core reference value, and combining the actual extracted current characteristics to calculate the normalized error feature value, and then judging the abnormal phase of the current waveform based on the threshold, the abnormal judgment process can adapt to the differences in the operating conditions of the motor, making the judgment result more consistent with the actual operating state of the motor. At the same time, it clarifies the target for subsequent targeted current waveform correction operations and lays the foundation for the orderly implementation of the correction process.

[0022] Step 3: Calculate the deviation gradient value of the abnormal phase of the current waveform, and calculate the corresponding current characteristic correction amount by combining it with the normalized error characteristic value. Perform pre-correction on the initial phase current control command for the next window and execute it. Iterate this process repeatedly. After the termination condition is met, switch to the normal control mode and terminate the recursive adaptive correction process. The specific process is as follows: For each phase to be corrected, the finite difference method, a common technique in numerical differentiation, is used to calculate the deviation gradient value of each anomalous phase. This value characterizes the rate of change of the normalized error eigenvalue between the current sliding analysis time window and the previous sliding analysis time window (all parameters involved in the formula have been normalized and dimensionless). The specific formula is as follows: , in, Let be the deviation gradient value of the i-th abnormal phase within the current sliding analysis time window; The normalized error eigenvalue of the i-th phase in the previous sliding analysis time window is taken as 0 for the first calculation, corresponding to the core benchmark value. This is the time interval between two adjacent sliding analysis time windows, i.e., the window sliding step size; Based on the normalized error characteristic values ​​and deviation gradient values ​​of each abnormal phase in the current sliding analysis time window, the current characteristic correction amount for the next sliding analysis time window corresponding to each abnormality is calculated independently for each phase using an adaptive proportional-integral correction algorithm (all parameters involved in the formula have been normalized and dimensionless): , in, This is the correction amount for the current characteristic of the i-th phase anomaly relative to the next sliding analysis time window; Let i be the initial time of the current sliding analysis time window. At the beginning of the next sliding analysis time window The integral term of the normalized error eigenvalue; and Both are adaptive scaling coefficients, defined as the monotonic adaptive function of the normalized error eigenvalue of the i-th corresponding front window, and are obtained as follows: Preset ratio lower limit Upper limit Lower limit of integral coefficient Upper limit and the threshold of normalized error eigenvalues ; Using linear mapping rules: , Under this rule, when the normalized error characteristics When the proportional coefficient increases, the integral coefficient increases simultaneously, enhancing the waveform correction strength; when the normalized error eigenvalue... When the coefficient is reduced, the proportional coefficient and the integral coefficient decrease simultaneously to avoid current distortion caused by overcorrection. Before the current sliding analysis time window ends and the next sliding analysis time window begins, based on the abnormal phases identified in the current window and the corresponding calculated correction values ​​for the specific current characteristics of each abnormal phase, the initial phase current control commands corresponding to each abnormality in the next sliding analysis time window are pre-corrected, as follows: Based on the current control mode and load conditions, retrieve the initial phase current control commands (i.e., uncorrected reference current commands) for each of the three phases A, B, and C within the next sliding analysis time window to ensure that the commands correspond one-to-one with each phase. For each phase marked as an abnormal phase in the current window, the corresponding current characteristic correction is added to the initial phase-separated current control command for the current phase in the next window, generating the corrected phase-separated current control command for the current phase in the next window. Normal phases directly use the initial command without modification. When adding the current characteristic correction, the current characteristic correction is first converted from the dimensionless deviation value to a current quantity with the same dimension as the initial current command through a preset dimension mapping coefficient. The modified (or initial) phase current control commands for all phases are constrained according to preset limits to limit the commands within the preset safe current command range (to avoid over-correction or abnormal commands causing motor overcurrent or instability). The final phase current control command for each phase is input to the current loop controller of the brushless motor. The current loop controller generates the corresponding PWM drive signal for each phase, and drives the motor to run according to the corrected command in the next sliding analysis time window, thereby realizing the pre-correction of abnormal phase current waveform (eliminating hysteresis deviation). After the current sliding analysis time window finishes running, immediately collect the three-phase raw current data within the window and perform the following operations in sequence: extract the phase window-level core current characteristics from the data in the window, retrieve the phase core reference value that matches the current control mode and load condition, calculate the normalized error characteristic value of each phase, determine whether the current waveform of each phase is abnormal, and calculate the corresponding deviation gradient value for the phase marked as abnormal. If there is an abnormal current waveform phase in the window, the current characteristic correction amount of the corresponding abnormal phase in the next sliding analysis time window is calculated and generated based on the deviation gradient value corresponding to each abnormality, and the pre-correction of the initial current control command of the next window is completed by phase; if the three phases A, B and C in the window are all determined to be without abnormality, the current correction is not performed in the next sliding analysis time window, and the standard current control command corresponding to the reference characteristic is directly used. The process is iterated in cycles until the brushless motor exits the current control mode and stops running, or until the current waveforms of each phase are determined to be normal within multiple consecutive sliding analysis time windows. Then, the system automatically switches to the normal control mode, terminates the recursive adaptive correction process, and subsequently only uses the standard current control command corresponding to the reference feature.

[0023] It should be noted that, for the phase to be corrected, the deviation gradient value is calculated, and this value is combined with the normalized error characteristic value to calculate the current characteristic correction amount. This allows the calculation of the correction amount to not only refer to the actual value of the error, but also reflect the rate of change of the error, making the determination of the correction amount more in line with the actual changes of the current waveform anomaly, and making the subsequent correction operation more reasonable. By setting adaptive proportional coefficients and adaptive integral coefficients, and using linear mapping rules to adjust the coefficients according to the changes of the normalized error characteristic value, the coefficients are increased synchronously when the error characteristic value increases to strengthen the correction intensity, and decreased synchronously when the error characteristic value decreases. This can adapt to different error levels for targeted adjustments, effectively correcting current waveform anomalies while avoiding over-correction that causes current distortion, thus balancing the correction effect and the stability of current operation. When pre-correcting the initial phase current control command for the next window, the correction amount is only superimposed on the abnormal phase, while the normal phase directly uses the initial command, realizing independent correction by phase. This makes the correction operation more targeted and avoids unnecessary impact on the current control of the normal phase. It avoids current redundancy fluctuations caused by indiscriminate correction, reduces additional harmonic losses in the motor windings caused by current fluctuations, and reduces electromagnetic noise caused by harmonics. At the same time, the corrected command is subject to preset amplitude constraints, limiting it to the safe current command range. This can effectively avoid motor overcurrent and instability problems caused by overcorrection or abnormal commands, ensuring the safe operation of the motor. Furthermore, completing the pre-correction before window switching can effectively eliminate lag deviations, improve the timeliness of the correction operation, make the current waveform closer to the ideal state, reduce internal energy loss of the motor caused by current control lag, and further reduce useless power consumption. The design employs an iterative correction process. After each sliding analysis time window, current data is collected, features are extracted, anomalies are identified, and subsequent correction operations are performed. This allows the monitoring and correction of the current waveform to dynamically follow the real-time operating status of the motor, adapting to continuous changes in motor operating conditions. This ensures the correction work is continuous and dynamic, promptly addressing current waveform anomalies that occur during operation. Continuous waveform correction effectively optimizes the motor's electromagnetic characteristics, reduces mechanical vibration caused by current waveform distortion, thereby lowering vibration noise during motor operation. Simultaneously, it improves the efficiency of converting electrical energy to mechanical energy, reduces energy loss during conversion, and achieves reasonable control of motor power consumption, meeting the low-power operation requirements of humanoid robot joint actuators. The system is equipped with dual termination conditions. When the motor exits the current control mode and stops running, or when the current waveforms of each phase are normal within multiple consecutive sliding analysis time windows, the recursive adaptive correction process is automatically terminated and the system switches to the normal control mode. This avoids ineffective cyclic correction after the motor has stabilized, reduces unnecessary calculation and operation losses of the controller, and improves the overall efficiency of the control method. It also allows the motor to run in the normal mode when the waveform is stable and there are no abnormalities, further ensuring the smoothness of motor operation and maintaining a low-noise and low-power operating state. Overall, this process, from calculating the deviation gradient and determining the correction amount to executing the pre-correction, iterative loop, and conditional termination, forms a complete recursive adaptive correction closed loop. This loop can effectively correct abnormal current waveforms in a targeted and dynamic manner, improve the current operating state of the brushless motor, enhance the stability and reliability of motor operation, and effectively reduce noise and power consumption during motor operation, thus better meeting the actual application needs of humanoid robot joint actuators.

[0024] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-mode control method for brushless motors based on humanoid robot actuators, characterized in that, include: Step 1: Periodically acquire and preprocess the three-phase current, rotor position, and control mode signals; extract window-level time-domain core features of the three-phase current by phase, and perform frequency domain analysis to extract high-frequency disturbance-related features; finally, organize the phase current feature vectors and associate them with the corresponding control mode and rotor speed signals. Step 2: Divide the core control modes of the brushless motor and their corresponding typical load conditions, collect three-phase current reference samples of normal operation under each combination, and extract window-level core current characteristics by phase. Core benchmark values ​​were determined through statistical analysis, and a multi-mode phase current benchmark feature library was constructed and integrated. Real-time identification of operating condition combinations retrieves corresponding core reference values, extracts actual features to calculate normalized error feature values, and completes the determination of abnormal phases in the current waveform; Step 3: Calculate the deviation gradient value of the abnormal phase of the current waveform, and calculate the corresponding current characteristic correction amount by combining the normalized error characteristic value. Perform pre-correction on the initial phase current control command of the next window and execute it. Iterate this process repeatedly. After the termination condition is met, switch to the normal control mode and terminate the recursive adaptive correction process.

2. The brushless motor multi-mode control method based on a humanoid robot actuator according to claim 1, characterized in that: The specific process for extracting window-level time-domain core features of three-phase currents by phase is as follows: First, preset the fundamental period of the brushless motor current, then adapt and set the length of the sliding analysis time window and the sliding step size, and keep the timing consistency between the sliding step size and the motor current loop control cycle. The sampling frequency is determined based on the fundamental period, and the three-phase current, rotor position and actuator control mode signals of the brushless motor are periodically collected and pre-processed by low-pass filtering. For each sliding analysis time window, the preprocessed three-phase current signal is extracted independently into window-level time-domain core features for phases A, B, and C. The time-domain core features include: the effective value of the phase fundamental current, the phase total harmonic distortion rate, and the phase current ripple coefficient.

3. The brushless motor multi-mode control method based on a humanoid robot actuator according to claim 2, characterized in that: The specific process of performing frequency domain analysis on the three-phase current to extract high-frequency disturbance-related features, and finally organizing the phase current feature vectors and correlating the corresponding control modes with the rotor speed signals is as follows: The three-phase current signals after preprocessing within the sliding analysis time window are analyzed in frequency domain independently by phase. The signal sampling frequency is preset and the frequency domain spectrum of the corresponding phase is generated based on the fast Fourier transform. The spectrum data within the preset frequency analysis range is locked. The high-frequency harmonic components are determined according to the preset spectrum amplitude threshold. After smoothing the spectrum data, the random fluctuation components are determined according to the preset spectrum smoothing difference threshold. The high-frequency harmonic amplitude and the proportion of fluctuation components of the corresponding phase current are extracted. The core time-domain and frequency-domain characteristics of the three phases within the sliding analysis time window are organized to form the phase current characteristic vector, which is then synchronously associated with the control mode signal and rotor speed signal of the current window.

4. The brushless motor multi-mode control method based on a humanoid robot actuator according to claim 3, characterized in that: The specific process of classifying the core control modes of the brushless motor and their corresponding typical load conditions, collecting three-phase current reference samples under normal operation in each combination, and extracting window-level core current characteristics by phase is as follows: For the core application scenarios of brushless motors for joint actuators in humanoid robots, the core control modes are divided, and all typical load conditions corresponding to each control mode are sorted out. For each control mode-load condition combination, three-phase current data of the brushless motor under normal and abnormal operation under that combination are collected in several independent sliding analysis time windows as the reference sample set for that combination. For the current data of each normal sliding analysis time window in the benchmark sample set, the core current features at the window level are extracted independently by phase. The extracted features include: effective value of phase fundamental current, phase total harmonic distortion rate, phase current ripple coefficient, phase high-frequency harmonic amplitude, and proportion of phase fluctuation components.

5. The brushless motor multi-mode control method based on a humanoid robot actuator according to claim 4, characterized in that: The specific process of determining the core benchmark values ​​through statistical analysis and constructing and integrating them into a multi-mode phase current benchmark feature library is as follows: For all normal sliding analysis time window reference samples under the same control mode and the same load condition, statistical analysis is performed on the core current characteristics of each window level according to phase. The average value of each characteristic in all reference samples is calculated as the core reference value of the corresponding phase, control mode and load condition. Based on the dispersion of the sample characteristic values, the normal fluctuation range of each characteristic around the core reference value is determined by standard deviation and used as an auxiliary constraint value. Based on the core reference values ​​and auxiliary constraint values ​​of each phase, a phase current reference characteristic table corresponding to the control mode-load condition is constructed; A multi-mode phase current reference feature library is formed by compiling a table of phase current reference features corresponding to all load conditions under the three control modes of position, speed, and torque. This library establishes a one-to-one correspondence between control mode, load condition, and phase feature. Each control mode corresponds to all load conditions to which it belongs, and each load condition corresponds to the core reference value and auxiliary constraint value of each of the three phases A, B, and C.

6. The brushless motor multi-mode control method based on a humanoid robot actuator according to claim 5, characterized in that: The specific process of identifying abnormal phases in the current waveform by real-time identification of operating condition combinations, retrieving corresponding core reference values, extracting actual features, calculating normalized error feature values, and completing the determination is as follows: The system identifies the control mode and actual operating load conditions of the brushless motor in the current sliding analysis time window in real time, and retrieves the core reference values ​​of phase A, phase B, and phase C that are precisely matched with the control mode and load conditions from the multi-mode phase current reference feature library. Based on the three-phase current data collected and preprocessed within the current sliding analysis time window, the core current characteristics at the window level are extracted independently for each phase to obtain the actual fundamental current effective value, actual total harmonic distortion rate, actual current ripple coefficient, actual high-frequency harmonic amplitude, and actual fluctuation component ratio for each phase. Calculate the normalized error characteristic value of each phase within the current sliding analysis time window according to phase, and quantify the degree of deviation between the actual core current characteristics and the core reference value. A preset threshold for the normalized error characteristic value is set for each phase. The normalized error characteristic value of each phase in the current window is compared with the corresponding threshold. If the normalized error characteristic value is not lower than the corresponding threshold, the phase is determined to be an abnormal current waveform phase and marked as a phase to be corrected. When the normalized error eigenvalue is lower than the corresponding threshold, the current waveform of the phase is determined to be normal and no subsequent gradient calculation and correction operation is performed.

7. The brushless motor multi-mode control method based on a humanoid robot actuator according to claim 6, characterized in that: The specific process for calculating the deviation gradient value of the abnormal phase of the current waveform and combining it with the normalized error characteristic value to calculate the corresponding current characteristic correction is as follows: For each phase to be corrected, the finite difference method in the field of numerical differentiation is used to calculate the deviation gradient value of each anomalous phase; Based on the normalized error characteristic value and deviation gradient value of each abnormal phase in the current window, the current characteristic correction amount of each abnormal phase in the next sliding analysis time window is calculated independently by phase. The adaptive proportional coefficient and adaptive integral coefficient of each abnormal phase in the current characteristic correction are both monotonic adaptive functions of the normalized error characteristic value of the corresponding abnormal phase before the window. The upper and lower limits of the two types of coefficients and the threshold of the normalized error characteristic value are preset. The coefficients are adjusted by using a linear mapping rule. When the normalized error characteristic value increases, the two types of coefficients increase synchronously, and when it decreases, they decrease synchronously.

8. The brushless motor multi-mode control method based on a humanoid robot actuator according to claim 7, characterized in that: The specific process of pre-correcting and executing the initial phase current control command for the next window is as follows: Before the current sliding analysis time window ends and the next sliding analysis time window begins, based on the abnormal phases identified in the current window and the corresponding calculated correction values ​​for the specific current characteristics of each abnormal phase, the initial phase-by-phase current control commands corresponding to each abnormality in the next window are pre-corrected. Specifically: First, retrieve the initial phase current control commands for each of the three phases in the next window. For abnormal phases, add the corresponding current characteristic correction to their initial commands to generate corrected commands. For normal phases, directly use the initial commands. The modified phase current control commands for all phases are subject to preset limit constraints, limiting them to the preset safe current command range. The final phase-by-phase current control command for each phase is input into the current loop controller of the brushless motor. The controller generates the corresponding drive signal for each phase, and drives the motor to run according to the corrected command in the next window, thereby realizing the pre-correction of the abnormal phase current waveform.

9. The multi-mode control method for brushless motors based on humanoid robot actuators according to claim 8, characterized in that: This process is iterated repeatedly. Once the termination condition is met, the system switches to the normal control mode. The specific process of terminating the recursive adaptive correction process is as follows: After the current sliding analysis time window finishes running, immediately collect the three-phase raw current data within the window, and sequentially perform the following operations: extract the core current characteristics of the phase window level, retrieve the core reference value of the phase that matches the current control mode and load condition, calculate the normalized error characteristic value of each phase, determine whether the current waveform of each phase is abnormal, and calculate the corresponding deviation gradient value for the phase marked as abnormal. If there is an abnormal phase of current waveform in the window, the current characteristic correction amount of the corresponding abnormal phase in the next sliding analysis time window is calculated and generated based on the deviation gradient value corresponding to each abnormality, and the pre-correction of the initial current control command of the next window is completed by phase. If all three phases within this window are determined to be without abnormality, then no current correction will be performed in the next window, and the standard current control command corresponding to the reference characteristic will be directly used. The process is iterated in cycles until the brushless motor exits the current control mode and stops running, or until the current waveforms of each phase are determined to be normal within multiple consecutive sliding analysis time windows. Then, the system automatically switches to the normal control mode, terminates the recursive adaptive correction process, and subsequently only uses the standard current control command corresponding to the reference feature.

Images