Intelligent brushless motor and control method thereof

Abstract

The invention is suitable for the technical field of brushless motor control, and provides an intelligent brushless motor and a control method thereof, and the method is applied to a control system of the intelligent brushless motor. The system comprises a data acquisition module, a model construction module, a training tuning module, a model verification module, a parameter setting module, an operation application module, a wireless communication module, a memory, an alarm, a processing center and an intelligent mobile terminal. The data acquisition module, the model construction module, the training optimization module, the model verification module, the parameter setting module, the operation application module, the wireless communication module, the memory and the alarm are respectively connected with the processing center; the invention also provides an intelligent brushless motor. According to the method, the AI model is constructed and trained by collecting the daily motor operation quality data, and the operation control parameters are matched according to the model, so that various operation quality abnormalities are prevented, and accurate control of the motor is ensured.

Description

Technical Field

[0001] The present invention relates to the field of brushless DC motor control technology, and in particular to an intelligent brushless motor and a control method thereof in the field of artificial intelligence. Background Art

[0002] Brushless DC motors (Brushless DC motors) are a new type of motor characterized by compact size, high mass, high control precision, and strong anti-interference capabilities. They are widely used in industrial automation control, electronic communications, medical electronics, automotive manufacturing, aerospace, and other fields. Traditional manual input of control parameters cannot match the control accuracy of Brushless DC motors. Manual input also causes control time delays, reducing the efficiency of motor-driven equipment. Summary of the Invention

[0003] In order to address the deficiencies in the prior art, the purpose of the present invention is to provide a control method for an intelligent brushless motor. By setting various functional modules such as a data acquisition module, a model construction module, a training and tuning module, a model verification module, a parameter setting module, and an operation application module, daily motor operation quality data is collected to build an AI model and train it. The operation control parameters are matched according to the model to prevent various operation quality abnormalities and ensure precise control of the motor.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] A control method for an intelligent brushless motor is applied to a control system of an intelligent brushless motor. The system comprises a data acquisition module, a model construction module, a training and tuning module, a model verification module, a parameter setting module, an operation and application module, a wireless communication module, a memory, an alarm, a processing center, and an intelligent mobile terminal. The data acquisition module, model construction module, training and tuning module, model verification module, parameter setting module, operation and application module, wireless communication module, memory, and alarm are respectively connected to the processing center. The intelligent mobile terminal comprises a smart phone, a tablet computer, and an intelligent remote control, which are respectively connected to the wireless communication module within a wireless network or the Internet.

[0006] The present invention provides a control method for an intelligent brushless motor, comprising the following steps:

[0007] S10, after the motor is started, the data acquisition module obtains the motor's operating and state parameters, dynamic characteristics, environmental condition data, adversarial generation data, and data preprocessing for subsequent model construction, and passes it to the model construction module;

[0008] S20, the model building module performs key feature extraction, model design, integrated optimization, parameter initialization, dynamic optimization, model fusion and deployment based on the verified data to build a model for subsequent model training and verification, and passes it to the training and tuning module;

[0009] S30, the training and tuning module trains the model according to the extracted data and adjusts the model parameters to optimize the model performance and predetermine the DC brushless motor operation control parameters to ensure the accuracy of the motor operation control parameters, and transmits them to the model verification module;

[0010] S40, the model verification module performs model verification and evaluation based on the trained model, and adjusts and optimizes the model based on the verification results to improve the performance and accuracy of the DC brushless motor operation model, and transmits the results to the parameter setting module;

[0011] S50, the parameter setting module performs a motor operation control test based on the brushless motor control parameters predicted by the trained model to obtain the optimal DC brushless motor operation control parameters to ensure the accuracy of future operation control, and transmits them to the operation application module;

[0012] S60, the operation application module controls the operation of the drive device and performs quality inspection according to the set DC brushless motor operation control parameters to ensure the accuracy and stability of the motor's long-term operation control in the future, and transmits the results to the processing center;

[0013] S70. The processing center compares the motor operation quality information with the DC brushless motor operation quality standard stored in the memory: if the standard is met, the system notifies the user that the motor can continue to operate; if the standard is not met, the system transmits the information to an alarm and notifies the user to repair or replace the motor.

[0014] Further, the step S50 includes the following steps:

[0015] S51. The parameter optimization unit obtains dynamically adjusted PID parameters according to the real-time optimization calculation formula of the brushless DC motor PID parameters: "Kp(t) = Rl(e′, ∫e), Ki(t) = Nn(Eh, Tr), Kd(t) = Ga(de / dt, ωe), Kp(t) is the dynamic proportional coefficient, Rl is reinforcement learning, e′ is the speed error change, ∫e is the error accumulation, Ki(t) is the dynamic integral coefficient, Nn is the neural network, Eh is the historical error sequence, Tr is the torque pulsation rate, Kd(t) is the dynamic differential coefficient, Ga is the genetic algorithm optimized Kd parameter, de / dt is the error differential term, and ωe is the speed tracking error" to achieve PID parameter adaptive adjustment and transmit the obtained parameters to the energy efficiency optimization unit;

[0016] S52, the energy efficiency optimization unit calculates the motor torque pulsation suppression according to the formula "Tr=AI(sqrt[∑(Tk-T′)2 / N)] / Tn*100%≤5%, Tr is the torque pulsation rate, AI is the artificial intelligence algorithm, Tk is the instantaneous torque (N·m), T′ is the average torque (N·m), N is the number of samples, and Tn is the rated average torque (N·m). The torque pulsation rate and energy efficiency optimization constraints are obtained to improve the overall system efficiency and are transmitted to the back electromotive force unit;

[0017] S53. The back-EMF unit obtains the back-EMF AI compensation and performs field-oriented control according to the DC brushless motor back-EMF AI compensation calculation formula: "Ea' = k*N1*δ*D*L*Bo / a*ω+Ne(ω',I'), where Ea' is the estimated back-EMF value after compensation (V), k is the number of armature slots, N1 is the number of winding element turns, δ is the pole arc coefficient, D is the armature outer diameter (m), L is the core length (m), Bo is the air gap flux density (T), a is the number of armature parallel branches, ω is the actual angular velocity of the motor (rad / s), Ne is the neural network compensation term (V), ω' is the compensation speed (rad / s), and I' is the compensation current (A)" to balance torque output and energy consumption, and transmits the result to the operation test unit;

[0018] S54, the operation test unit performs an operation control test on the target drive device according to the DC brushless motor operation control parameters predetermined by the model and obtains motor operation quality information to ensure the accuracy of the motor control parameters, and transmits the information to the processing center;

[0019] S55. The processing center compares the motor operation quality information with the brushless motor operation quality standard stored in the memory. If the standard is met, the information is set as the brushless DC motor operation control parameter. If the standard is not met, the information is transmitted to the alarm device and the parameter adjustment is notified.

[0020] S56. The abnormality recognition unit matches the real-time acquired brushless motor operation control image with the corresponding image in the trained convolutional neural network model and confirms its abnormality category in order to adjust the operation test parameters.

[0021] The system described in the present invention also includes a computer-readable storage medium containing a memory; the memory stores a computer program, and when the functional modules execute the computer program, the steps of the control method of an intelligent brushless motor described in any one of the above are implemented; the computer-readable storage medium stores a computer program, and when the functional modules execute the computer program, the steps of the control method of an intelligent brushless motor described in any one of the above are implemented; and also includes an intelligent brushless motor control device, which is implemented using the control method of an intelligent brushless motor described in the above.

[0022] The present invention also provides an intelligent brushless motor, which is realized by using the above-mentioned control method for an intelligent brushless motor.

[0023] The beneficial effects of the present invention compared with the prior art are:

[0024] By setting up various functional modules, we collect daily motor operating quality data to build and train an AI model. Using the model's sophisticated algorithms, we match motor operating control parameters to the driven equipment's parameters. This prevents various motor operating quality anomalies and ensures consistent motor operation. (For more detailed information on the benefits, see the "Specific Implementation Methods" section below.) BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments or exemplary technical descriptions. The drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0026] Figure 1 Schematic diagram of the system module of the present invention;

[0027] Figure 2 This is a schematic diagram of a data acquisition module of the present invention;

[0028] Figure 3 This is a schematic diagram of the model building module of the present invention;

[0029] Figure 4 This is a schematic diagram of the training and tuning module of the present invention;

[0030] Figure 5 Schematic diagram of the model verification module of the present invention;

[0031] Figure 6 This is a schematic diagram of a parameter setting module of the present invention;

[0032] Figure 7 This is a schematic diagram of the operating application module of the present invention;

[0033] Figure 8 This is a schematic diagram of the method flow control program of the present invention;

[0034] Figure 9 This is a schematic diagram of step S10 in the method of the present invention;

[0035] Figure 10 Schematic diagram of step S20 in the method of the present invention;

[0036] Figure 11 Schematic diagram of step S30 in the method of the present invention;

[0037] Figure 12 Schematic diagram of step S40 in the method of the present invention;

[0038] Figure 13 Schematic diagram of step S50 in the method of the present invention;

[0039] Figure 14 Schematic diagram of step S60 in the method of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] The specific implementation of the present invention is described in detail below with reference to specific embodiments:

[0042] In order to make the technical problems, technical solutions and beneficial effects to be solved by this application more clearly understood, the application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application. It should be noted that when a module is referred to as being "set on" another module, it can be directly on the other module or indirectly on the other module. When a module is referred to as being "connected to" another module, it can be directly connected to the other module or indirectly connected to the other module. In the description of this application, "multiple" means two or more, unless otherwise clearly and specifically defined. "Several" means one or more, unless otherwise clearly and specifically defined.

[0043] In the description of this application, it should be noted that, unless otherwise expressly specified or limited, the terms "connected" and "connection" should be understood broadly. For example, it can mean a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection between two elements or the interaction between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood based on the specific circumstances. The terms "include," "comprising," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.

[0044] See also Figure 1As shown, the present invention provides a control method for an intelligent brushless motor, which is applied to a control system of an intelligent brushless motor. The system includes a data acquisition module, a model construction module, a training and tuning module, a model verification module, a parameter setting module, an operation application module, a wireless communication module, a memory, an alarm, a processing center, and an intelligent mobile terminal; the data acquisition module, the model construction module, the training and tuning module, the model verification module, the parameter setting module, the operation application module, the wireless communication module, the memory, and the alarm are respectively connected to the processing center; the intelligent mobile terminal includes a smart phone, a tablet computer, and an intelligent remote control, which are respectively connected to the wireless communication module within the range of a wireless network or the Internet.

[0045] The wireless communication module is provided with a wireless network unit, which is responsible for sending and receiving wireless signals and automatically networking with smart mobile terminals and various drive devices within an effective network range. The wireless signals include various IoT signals such as MQTT, CoAP, HTTP, REST API, Zagbee, LoRaWAN, NB-IoT, Bluetooth, or 5G, 4G, and Wi-Fi network signals.

[0046] The alarm compares the actual evaluation score of the motor model with the evaluation score standard of the motor model stored in the memory. If the standard is not met, the alarm automatically sounds an alarm and notifies to continue training; and compares the quality information of the motor operation with the operation quality standard of the brushless motor stored in the memory. If the standard is not met, the alarm automatically sounds an alarm and notifies to adjust the parameters; and compares the motor operation quality information with the operation quality standard of the brushless DC motor stored in the memory. If the standard is not met, the alarm automatically sounds an alarm and notifies to repair or replace.

[0047] The memory is responsible for storing information from the data acquisition module, model construction module, training and tuning module, model verification module, parameter setting module, operation application module, wireless communication module, and alarm, as well as the model evaluation criteria and the brushless motor control quality criteria.

[0048] The processing center is responsible for information transmission among various functional modules, wireless communication modules, alarms, and memories, and serves as the hub of the system. The processing center compares the actual evaluation score of the motor model with the evaluation score standard of the motor model stored in the memory: if the standard is met, the predetermined DC brushless motor operation control parameters are set; if the standard is not met, the parameters are transmitted to the alarm and a notification is given to continue training; and the processing center compares the motor operation quality information with the brushless motor operation quality standard stored in the memory: if the standard is met, the parameters are set as the DC brushless motor operation control parameters; if the standard is not met, the parameters are transmitted to the alarm and a notification is given to adjust the parameters; and the processing center compares the motor operation quality information with the DC brushless motor operation quality standard stored in the memory: if the standard is met, a notification is given that the motor can continue to operate; if the standard is not met, the information is transmitted to the alarm and a notification is given to repair or replacement.

[0049] See also Figure 2 As shown, the data acquisition module includes an information acquisition unit, a noise filtering unit, a data cleaning unit, a format conversion unit, a static normalization unit, a dynamic standard unit, a data enhancement unit, and a data encoding unit. It is responsible for acquiring the motor's operating and state parameters, dynamic characteristics, environmental operating condition data, adversarial generation data, and data preprocessing, and passing them to the model construction module.

[0050] Furthermore, the information acquisition unit synchronously collects motor operation data such as current and voltage, rotor position and speed, temperature, vibration and electromagnetic characteristics, and surrounding environment data through an acquisition device, and transmits the data to a noise filtering unit; the noise filtering unit obtains filtered motor data according to a sliding window mean filtering calculation formula "y(n)=y(n-1)+[x(n)-x(nM)] / M", and transmits the data to a data cleaning unit; the data cleaning unit locates missing data, fills missing values, corrects abnormal values, and eliminates duplicate values ​​on the filtered data to obtain cleaned motor data, and transmits the data to a format conversion unit; the format conversion unit unifies the current, voltage, speed, temperature, time, torque and power, and torque coefficient of the cleaned data into the international system of units according to a set formula, and transmits the data to a static normalization unit; the static normalization unit calculates the normalization value according to the static numerical normalization calculation formula "Z=[X-∑Xi / No] / sqrt[∑(Xi-μ) 2 / No]” to obtain static numerical standard data and pass it to the dynamic standard unit; the dynamic standard unit calculates the dynamic numerical normalization formula “Zn=[Xn-(μo+(Xn-Xo) / Ws)] / sqrt[σo 2 +((Xn-μn) 2 -(Xo-μo) 2) / Ws]” to obtain dynamic numerical standardized data and pass it to a data enhancement unit; the data enhancement unit performs noise injection, time series transformation, working condition simulation, and physical model driving on the standardized data to enhance data performance, and passes it to a data encoding unit; the data encoding unit converts the standardized overcurrent, overtemperature, overvoltage / short circuit and other fault types into one-hot encoding and constructs a multi-label classification input, and performs label encoding on the control mode.

[0051] It is further explained that the motor operation data includes: 1) operating status parameters: Clarke-Park transformation, bus current and other current data, bus voltage, phase voltage and other voltage data, electrical angle, mechanical speed and other rotor position and speed, winding temperature, ambient temperature and other temperature data, acceleration signal, spectrum characteristics and other vibration data, back electromotive force waveform, harmonic distortion rate and other electromagnetic characteristics; 2) environmental operating condition data: torque observation value, load mutation record and other load characteristics, humidity, electromagnetic interference intensity and other environmental parameters, historical operating condition data, adversarial generation data and other model auxiliary data; 3) image monitoring data: current harmonic spectrum diagram, back electromotive force waveform image and other current and voltage waveform images, winding and power device thermal imaging, radiator surface temperature field and other thermal imaging and temperature distribution diagrams, vibration spectrum diagram, noise time-frequency analysis diagram and other mechanical vibration and noise images, magnetic field oriented control FOC vector diagram, commutation logic timing diagram and other control algorithm state visualization data, high and low temperature cycle test curve, dust and waterproof test image and other environmental adaptability monitoring images.

[0052] See also Figure 3 As shown, the model building module includes a feature extraction unit, a voltage modeling unit, a power modeling unit, an electromagnetic torque unit, a coordinate transformation unit, a dynamic equation unit, a model standard unit, a discrete processing unit, and a noise modeling unit. The model is modeled based on key feature extraction, model design, integrated optimization, parameter initialization, dynamic optimization, model fusion and deployment according to the verified data, and the model is passed to the training and tuning module.

[0053] Furthermore, the feature extraction unit obtains the key features of the motor's electrical parameters, mechanical dynamics, control signals, temperature and faults, time-frequency domain, and dimensionality reduction fusion based on the pre-processed data, and transmits them to the voltage modeling unit; the voltage modeling unit obtains the circuit model based on the DC brushless motor equivalent input voltage calculation formula "Ue=Ro*i+Li*di / dt+E" and transmits it to the power modeling unit; the power modeling unit obtains the circuit model based on the motor angular acceleration calculation formula "α=[(Kt*iq)-Po / ω-(U*io-io 2 *Ri) / ω] / (∑M*r 2)” to obtain angular acceleration and transmit it to the electromagnetic torque unit; the electromagnetic torque unit calculates the electromagnetic torque of the brushless DC motor according to the formula “Te=J*α+Bi*ω+T L " obtain the motor dynamic torque balance equation and pass it to the coordinate transformation unit; the coordinate transformation unit obtains the two-phase static coordinate system according to the DC brushless motor Park transformation formula "id=2 / 3[ia*cosθ+ib*cos(θ-120°)+ic*cos(θ+120°)],iq=2 / 3[-ia*sinθ-ib*sin(θ-120°)-ic*sin(θ+120°)]" and passes it to the dynamic equation unit; the dynamic equation unit obtains the dynamic equation construction according to the DC brushless motor voltage balance equation "Ld*did / dt=Ud-Ra*id+ωe*Lq*iq,Lq*diq / dt=Uq-Ra*iq-ωe(Ld*id+ψf)" and passes it to the model standard unit; the model standard unit converts the nonlinear differential equation into the matrix standard state space form x'=Ax+Bu and passes it to the discrete processing unit; the discrete processing unit obtains the dynamic equation construction according to the DC brushless motor zero-order holder formula A discretized equation is obtained and passed to a noise modeling unit; the noise modeling unit obtains a noise model according to the DC brushless motor noise modeling calculation formula "x(k+1)=Φ*x(k)+Γ*u(k)+w(k), y(k)=C*x(k)+v(k)" to enhance the robustness of the model.

[0054] See also Figure 4 As shown, the training and tuning module includes a hardware acceleration unit, a supervised learning unit, a loss function unit, a reinforcement learning unit, and a parameter optimization unit. The model is trained based on the extracted data and the model parameters are adjusted to optimize the model performance to predetermine the DC brushless motor operation control parameters and pass them to the model verification module.

[0055] Furthermore, the hardware acceleration unit accelerates large-scale data training according to the model through hardware selection and configuration, data parallel processing optimization, edge-cloud collaborative architecture, performance tuning and energy efficiency management, and passes it to the supervised learning unit; the supervised learning unit uses simulation data to pre-train the model according to the model through physical guidance of neural network architecture, hybrid architecture, Xavier / He initialization, and transfer learning, and passes it to the loss function unit; the loss function unit calculates the loss function of the brushless DC motor according to the formula "Lt=∑(yi-y'i) 2 / N+λ1*∑max(0,|Ii|-Imax)+λ2*∑∥ea(t)-Ke*ω(t)∥2+λ3*∑∥y' t+1 -y't∥2+λ4*E x～D[∥xe*L(f(x),y)∥2]+∑(σt -2 *Lt+logσt) / T” to obtain the loss function value and pass it to the reinforcement learning unit; the reinforcement learning unit obtains the reward function value according to the DC brushless motor reward function calculation formula "fr=ε*(1-|ωr-ωo| / 0.1ωr)+β*(1-Pl / Pr)-γ-*∑max(0,|Ii|-Imax)" and strengthens learning, and passes it to the parameter optimization unit; the parameter optimization unit obtains the reward function value according to the DC brushless motor basic acquisition function calculation formula "fg=W1*EI(Ms(ωr,ω))+W2*Ei(Po / Pr)-γ*∑max(0,|Ii|-Imax)" and strengthens learning.

[0056] See also Figure 5 As shown, the model verification module includes a classification labeling unit, a learning update unit, a cost verification unit, and a training evaluation unit. It performs model verification and evaluation based on the trained model and adjusts and optimizes the model based on the verification results, and passes them to the parameter setting module;

[0057] Furthermore, the classification and labeling unit divides the data samples into corresponding categories according to the abnormal categories of the DC brushless motor operation quality, labels them, and transmits them to the learning and updating unit; the learning and updating unit determines the training model according to the gradient descent data, and inputs the DC brushless motor operation quality data in the training set into the model for simulation training, and transmits it to the cost verification unit; the cost verification unit obtains the cost function value after model training according to the cost function calculation formula "J(θ)=∑[1 / (1+σ(xi))]*[y*log(hθ(x))]" after model training, and transmits it to the training evaluation unit; the training evaluation unit obtains the F comprehensive evaluation score according to the model evaluation score calculation formula "F=2[(Tp+Tn) / (Tp+Fp+Fn+Tn)]*[Tp / (Tp+Fn)] / [(Tp+Tn) / (Tp+Fp+Fn+Tn)]+[Tp / (Tp+Fn)]" to ensure the comprehensive performance of the model in motor control, and transmits it to the processing center.

[0058] See also Figure 6 As shown, the parameter setting module includes a parameter optimization unit, an energy efficiency optimization unit, a back electromotive force unit, an operation test unit, and an abnormality identification unit. The motor operation control test is performed based on the brushless motor control parameters predicted by the trained model to obtain the optimal DC brushless motor operation control parameters, and the parameters are passed to the operation application module.

[0059] Furthermore, the parameter optimization unit obtains the dynamic adjustment PID parameters according to the real-time optimization calculation formula of the DC brushless motor PID parameters "Kp(t)=Rl(e′,∫e),Ki(t)=Nn(Eh,Tr),Kd(t)=Ga(de / dt,ωe)", and transmits them to the energy efficiency optimization unit; the energy efficiency optimization unit obtains the dynamic adjustment PID parameters according to the motor torque pulsation suppression calculation formula "Tr=AI(sqrt[∑(Tk-T′) 2 / N)] / Tn*100%≤5%” to obtain the torque pulsation rate and energy efficiency optimization constraint, and transmit it to the back electromotive force unit; the back electromotive force unit obtains back electromotive force AI compensation and magnetic field orientation control according to the DC brushless motor back electromotive force AI compensation calculation formula “Ea′=k*N1*ε*D*L*Bo / a*ω+Ne(ω′,I′)”, and transmits it to the operation test unit; the operation test unit performs an operation control test on the target drive device according to the DC brushless motor operation control parameters predetermined by the model and obtains motor operation quality information, and transmits it to the processing center; the abnormality recognition unit matches the brushless motor operation control image obtained in real time with the corresponding image in the trained convolutional neural network model and confirms its abnormality category to facilitate the adjustment of the operation test parameters.

[0060] See also Figure 7 As shown, the operation application module includes a parameter configuration unit, a commutation control unit, a closed-loop adjustment unit, a parameter optimization unit, and a quality detection unit. According to the set DC brushless motor operation control parameters, the drive device operation is controlled and quality detection is performed, and the results are transmitted to the processing center.

[0061] Furthermore, the parameter configuration unit is equipped with corresponding operation control parameters according to the load torque and moment of inertia input model of the driven device, identifies the initial position of the rotor, and transmits it to the commutation control unit; the commutation control unit realizes real-time position detection and commutation control through position signal processing, commutation logic execution, logic verification and debugging, parameter solidification and storage, and transmits it to the closed-loop adjustment unit; the closed-loop adjustment unit realizes PWM modulation and closed-loop adjustment through PWM modulation, closed-loop adjustment, and coordinated control, and transmits it to the parameter optimization unit; the parameter optimization unit realizes dynamic parameter optimization and protection through real-time parameter monitoring and adjustment, adaptive control, efficiency optimization, protection mechanism implementation, and fault recovery, and transmits it to the quality detection unit; the quality detection unit detects the static parameters, dynamic performance, environmental adaptability, and protection mechanism of the motor through the motor operation detection equipment.

[0062] System operation working principle:

[0063] When the motor starts, the data acquisition module first obtains the motor's operating and state parameters, dynamic characteristics, environmental conditions data, adversarial generation data and data preprocessing, and passes them to the model building module; secondly, the model building module performs key feature extraction, model design, integrated optimization, parameter initialization, dynamic optimization, model fusion and deployment based on the verified data to build a model, and passes it to the training and tuning module; again, the training and tuning module trains the model based on the extracted data and adjusts the model parameters to optimize the model performance and predetermine the DC brushless motor operating control parameters, and passes them to the model verification module; then the model verification module performs model verification and evaluation based on the trained model. The model is adjusted and optimized based on the verification results and passed to the parameter setting module; then the parameter setting module performs a motor operation control test based on the brushless motor control parameters predicted by the trained model to obtain the optimal DC brushless motor operation control parameters, and passes them to the operation application module; then the operation application module controls the operation of the drive device and performs quality inspection based on the set DC brushless motor operation control parameters, and passes it to the processing center; finally, the processing center compares the motor operation quality information with the DC brushless motor operation quality standard stored in the memory: if it meets the standard, it will notify that it can continue to operate; if it does not meet the standard, it will be passed to the alarm and notified for maintenance or replacement.

[0064] When operators or managers are outdoors or in other places, they can use smart mobile terminals to automatically connect to the wireless network or the Internet through the wireless communication module, realize the combination of smart mobile terminals, the Internet and Internet of Things technologies, and use the APP software on the smart mobile terminal or the remote control software on the computer to perform close or remote control of the brushless motor automated production. Managers can remotely control and monitor the operation quality of the DC brushless motor drive, view daily operation status, data changes, equipment activity logs, etc., realize remote connection and monitoring of the improvement of the operation quality of the DC brushless motor, improve control efficiency and reduce control costs.

[0065] See also Figure 8 As shown, the present invention provides a control method for an intelligent brushless motor, comprising the following steps:

[0066] S10, after the motor is started, the data acquisition module obtains the motor's operating and state parameters, dynamic characteristics, environmental condition data, adversarial generation data, and data preprocessing for subsequent model construction, and passes it to the model construction module;

[0067] See also Figure 9 As shown, the step S10 includes the following steps:

[0068] S11. The information acquisition unit synchronously collects motor operation data such as current and voltage, rotor position and speed, temperature, vibration and electromagnetic characteristics, and surrounding environment data through the acquisition device to ensure the consistency of timing, and transmits the data to the noise filtering unit;

[0069] Further explanation: according to the start-up command of the driven device, the three-phase current is first collected through the Hall sensor or shunt resistor for Clarke-Park transformation and current loop control in the field-oriented control (FOC), and the bus voltage is obtained through the voltage sensor. The system power state is monitored by the bus current, and the real-time energy efficiency is calculated in combination with the voltage data. The bus voltage is used to reflect the inverter power supply state for voltage feedforward compensation and power analysis, and the phase voltage supports the back electromotive force observer to realize sensorless control; secondly, the rotor electrical angle (θe) is obtained through the Hall sensor, encoder or BEMF observer for FOC field orientation and commutation control, and the rotor speed (ωm) is calculated through the encoder or speed observer for speed loop closed-loop control, and the winding temperature (T) is monitored in real time through the temperature sensor for resistance temperature drift compensation, and the obtained ambient temperature is used to assist in analyzing the influence of heat dissipation conditions on motor performance; secondly, the motor acceleration signal is collected through the vibration sensor for analyzing mechanical factors such as bearing wear and rotor imbalance. Faults are detected and spectral features such as high-frequency vibration components are extracted to detect magnetic saturation or electromagnetic interference anomalies. The back-electromotive force waveform is used for rotor position estimation and fault diagnosis in sensorless control. The quality of current / voltage signals is evaluated through harmonic distortion rate to locate inverter dead zone effects or switching device anomalies. Torque observation values ​​are calculated through mechanical power and speed to verify the output accuracy of the control algorithm. The dynamic response when the load is suddenly increased / decreased is collected by using load mutation records to optimize the anti-disturbance capability of the AI ​​model. Environmental parameters such as the ambient temperature and humidity of the motor are then obtained through sensors. Since humidity affects the insulation performance, it needs to be jointly analyzed with temperature data. The frequency distribution, amplitude and harmonic components of the electromagnetic interference signal are measured using a spectrum analyzer to evaluate the impact of external noise on the sensor signal. Finally, operating data under different temperatures, loads and speeds are obtained through the motor's historical operation database for reinforcement learning strategy training and model generalization optimization, as well as adversarial generated data. Extreme working conditions are simulated through GAN to enhance the control robustness of the AI ​​model.

[0070] S12. The noise filtering unit obtains filtered motor data according to the sliding window mean filtering calculation formula "y(n) = y(n-1) + [x(n) - x(nM)] / M, y(n) is the filtered output value at the current moment, y(n-1) is the filtered output value at the previous moment, x(n) is the original input value at the current moment, x(nM) is the input value at the oldest moment of the window, and M is the sliding window length (n≥M)" to filter out high-frequency noise and pass it to the data cleaning unit;

[0071] For further explanation, according to the program commands after information collection, when the window is in the sliding stage, the sliding window mean filtering calculation formula is used to calculate the sliding average of the three-phase current signal with a 10 ms window, reducing the amplitude of the noise above 20 kHz. For wavelet packet decomposition and reconstruction, the high-frequency noise band is separated through wavelet packet decomposition, and the fundamental wave signal is reconstructed to retain the effective control characteristics. For time-domain threshold truncation, the transient spikes exceeding the dynamic threshold in the voltage / current signal are replaced by the median value to eliminate the outliers caused by short circuits or open circuits; the y(n) is the smoothed signal, which is updated in real time through the algorithm; the y(n - 1) is the smoothed signal at the previous moment, which is automatically obtained from the database; the x(n) is the original signal collected by the sensor. The current signal is measured by a Hall sensor or a shunt resistor and converted into a digital signal through ADC. The voltage signal is measured by a voltage sensor; the x(n - M) is the data of the previous M sampling points before the current moment, which is obtained through each sensor; the M is the number of data points participating in the mean calculation, and the noise frequency covers the noise cycle through the window time. When the window is in the initial filling stage and not full, the formula "y[n] = 1 / (n + 1)∑x[k] (0 ≤ n < M)" is used for calculation. In the formula, n is the discrete time index at the current moment (starting from 0 and increasing), which is automatically generated by the controller sampling period, and x[k] is the original sampling value of the input signal at moment k, which is collected through the digital signal output by the Hall sensor, shunt resistor or ADC acquisition module; the noise includes high-frequency switching noise generated by the fast switching action of the PWM inverter, which appears as high-frequency spikes in the current / voltage waveform, and electromagnetic interference caused by the coupling capacitance and inductance between the motor controller and the power device, which forms common-mode noise superimposed on the DC signal. In order to suppress the filtering, an RC circuit is configured in the sensor signal chain to suppress the switching noise through the formula "fc = 1 / (2π*R*C), where fc is the cut-off frequency (Hz), R is the resistance value (Ω), and C is the capacitance value (F)", and a low-ESR (equivalent series resistance) electrolytic capacitor is selected and connected in parallel at the bus voltage terminal to absorb the high-frequency current ripple. At the same time, a high-frequency ceramic capacitor is deployed near the DC bus side of the inverter to reduce the impedance of the switching noise propagation path, and a star-shaped grounding design is adopted to avoid introducing high-frequency interference due to the ground wire loop; through PWM frequency adaptive adjustment, the PWM frequency is dynamically adjusted according to the load conditions to balance the noise suppression and switching loss, and dead-time compensation is performed to correct the voltage distortion caused by the inverter dead-time through the feed-forward compensation algorithm and reduce the high-frequency harmonic components.

[0072] S13. The data cleaning unit locates data missing, fills in missing values, corrects outliers, and removes duplicate values from the filtered data to obtain the cleaned motor data, ensuring the accuracy of the model input data and transmitting it to the format conversion unit;

[0073] Further explanation: according to the filtered program command, data missing location is first performed, and local data missing caused by sensor failure or communication interruption in key parameters such as three-phase current and bus voltage is detected through timing alignment, and missing timestamps are marked to avoid direct deletion and control timing interruption; secondly, missing value filling is performed, and linear interpolation or LSTM prediction model is used to repair short-term missing, such as when the current signal interruption is <100ms, it is reconstructed through adjacent data, and long-term missing such as temperature sensor failure is filled by historical working condition similarity, for example, it is replaced by the historical temperature average under the same load; then dynamic Threshold judgment: A dynamic threshold is set based on the motor operating status. A current mutation exceeding 3σ is judged as a short circuit or open circuit anomaly, and an excessive speed is marked as a sensor fault or load mutation. Then, abnormal value correction is performed. A sliding window median filter is used to replace abnormal points for transient spike noise such as electromagnetic interference, while a persistent abnormality such as Hall sensor offset triggers a hardware self-test program and isolates the fault channel. Finally, duplicate value deduplication is performed to identify and delete repeated sampling points such as two adjacent frames of data that are completely consistent due to data acquisition card cache overflow, and redundant data from multiple sensors in the same control cycle is merged, retaining the record with the earliest timestamp.

[0074] S14, the format conversion unit unifies the cleaning data into the international unit system according to the set formula for current, voltage, speed, temperature, time, torque and power, and torque coefficient to ensure the accuracy of subsequent data calculations, and transmits it to the static normalization unit;

[0075] Further explanation: According to the program command after cleaning, the speed unit conversion is first performed. The angular velocity ω output by the Hall sensor or encoder is unified into the mechanical speed υ of the international unit system through the rpm to rad / s conversion calculation formula "ω=2π*υ / 60, ω is the motor angular velocity (rad / s), υ is the motor mechanical speed (rpm)", and the mechanical speed υ obtained by the above Hall sensor, encoder or back electromotive force estimation is converted into the commutation frequency or electrical signal frequency f through the frequency and speed relationship calculation formula "f=p*υ / 60, f is the motor commutation frequency or electrical signal frequency (Hz), p is the number of motor poles, υ is the motor mechanical speed (rpm)", where p is the pairing of the permanent magnet poles. The number is used to connect the speed and the switching frequency of the electronic control system; secondly, the torque and power units are unified, and the motor torque and angular velocity are combined to calculate the output power through the mechanical power calculation formula "P=Te*ω, P is the motor mechanical power (W), Te is the motor output torque (Nm), ω is the motor angular velocity (rad / s)", which is used to verify the logical consistency of the data, and the motor torque coefficient is obtained through the torque coefficient conversion formula "Kt=Te / Iq, Kt is the motor torque coefficient (Nm / A), Te is the motor output torque (Nm), Iq is the quadrature axis current (A)", which supports the standardization of the input of the magnetic field oriented control algorithm; then the voltage and current dimensions are calibrated, and the back electromotive force calculation formula "Ea=K e*ω, Ea is the back electromotive force of the motor (V), Ke is the potential coefficient (V / (rad / s)), ω is the angular velocity of the motor (rad / s)” to obtain the back electromotive force of the motor, and according to the current dynamic range normalization formula "Inorm=(IA-Imin) / (Imax-Imin), Inorm is the normalized current value (A), IA is the real-time current measurement value of the motor (A), Imin is the minimum value of the current dynamic range (A), Imax is the maximum value of the current dynamic range (A)", the current signal is scaled to the [0,1] interval to adapt to the input requirements of the AI ​​model; then the temperature and time unit processing is performed, and the temperature sensor calibration formula "T=(Us-Uo ) / kt, T is the final temperature value output by the sensor (℃), Uo is the zero offset voltage (V), Us is the original output voltage signal of the temperature sensor (V), and kt is the calibration coefficient (℃ / V)". The voltage signal will be converted into a temperature value, and the timestamps will be aligned. The control cycle "to=1 / fc, to is the control cycle (s), fc is the control frequency (Hz)", and is synchronized with the sampling frequency to achieve unit and dimension unification; finally, a logic conflict detection is performed, and the current-torque relationship is verified first. If the measured torque Te=kt*iq deviates from the observed value by >5%, it is determined to be a data logic error, and the phase consistency of the back electromotive force waveform and the rotor position signal is checked to rule out the problem of encoder installation misalignment.

[0076] S15, the static normalization unit calculates the static numerical normalization formula "Z = [X-∑Xi / No] / sqrt[∑(Xi-μ) 2 / No], Z is the static standardized value, X is the static original data point, No is the number of dynamic samples, Xi is the variable in the i-th time period, and μ is the mean value of the variable. Static numerical standardized data is obtained to eliminate dimensional differences and is passed to the dynamic standard unit;

[0077] Further explanation: According to the program command after format conversion, the normalized value Z of the static numerical data is obtained by the static numerical normalization calculation formula. Generally, 68% of the data falls on Z∈[-1,1] and 95% of the data falls on Z∈[-2,2]. The static numerical normalization calculation formula is composed of "μ=∑Xi / N", "σ=sqrt[∑(Xi-μ) 2 / N]” is substituted into “Z = (X-μ) / σ” to obtain, where Z is the static normalized numerical scalar (i.e., the degree to which the original value deviates from the mean, measured as a multiple of the standard deviation), σ is the standard deviation (i.e., the degree of dispersion of the variable X, measuring the fluctuation range of the data); the original variable X is a continuous monitoring variable such as the phase current (A), voltage (V), and temperature (°C) during motor operation; the mean μ is the arithmetic mean of the variable X within a certain time period, reflecting the center position of the data; and N is the total number of samples in the static data set.

[0078] S16, dynamic standard unit is calculated according to the dynamic numerical normalization formula "Zn=[Xn-(μo+(Xn-Xo) / Ws)] / sqrt[σo 2 +((Xn-μn) 2 -(Xo-μo) 2 ) / Ws], Zn is the dynamic normalization value, Xn is the new data point, μo is the old mean, Ws is the window size, σo is the old standard deviation, Xo is the old data point to be replaced, and μn is the new mean. Dynamic numerical normalization data is obtained to eliminate the magnitude difference and is passed to the data enhancement unit;

[0079] Further explanation: According to the program command after static numerical conversion, the normalized value Zn of dynamic numerical data is obtained by the dynamic numerical normalization calculation formula, which is the value of the new data point Xn after dynamic normalization, reflecting the standard deviation multiple of its deviation from the current mean; the dynamic numerical normalization calculation formula is composed of "μn=μo+(Xn-Xo) / Ws", "σn=sqrt[σo 2 +((Xn-μn) 2 -(Xo-μo) 2) / Ws]” is substituted into “Zn=(X-μn) / σn” to obtain, where σn is the new standard deviation, that is, the standard deviation of the data window after the update, reflecting the degree of discreteness of the historical data; Xn is the real-time data such as current (A), speed (rpm), and temperature (℃) collected at the current moment; μo is the arithmetic mean of the data window before the update, calculated based on the first N-1 historical data; Xo is the earliest historical data point that needs to be removed when the new data Xn is added to the sliding window model; Ws is the total number of data points contained in the sliding window, which can be fixed or dynamically adjusted; σo is the standard deviation of the data window before the update, reflecting the degree of discreteness of the historical data; and μn is the updated window mean after adding Xn and removing Xo.

[0080] S17, the data enhancement unit performs noise injection, time series transformation, working condition simulation, and physical model driving on the standardized data to enhance data performance, thereby improving dynamic response speed and fault diagnosis capabilities, and transmits the data to the data encoding unit;

[0081] It is further explained that according to the program commands after data format conversion, the noise amplitude, load range and other enhancement parameters are first set according to the motor parameters such as rated current and temperature threshold, and the interpolation method or generative adversarial network (GAN) is used to expand the small sample fault data; then noise injection is performed, and for Gaussian noise, random noise with a mean of 0 is superimposed to simulate sensor error, and for pulse noise, instantaneous interference such as current spikes when the motor starts and stops is simulated to improve the stability of the current / voltage signal in a noisy environment and the robustness to real noise such as sensor error and electromagnetic interference; then time series transformation is performed, and the window is first sliced ​​to intercept the operating data of different time periods to simulate dynamic response, but for a single time series data, noise or time domain transformation is added, and for time domain scaling, the signal frequency is adjusted to simulate voltage fluctuations or load The model is then simulated to enhance its adaptability to sudden speed and load fluctuations. Operating condition simulations are then conducted, starting with combined enhancements. Multiple perturbations, such as injected noise and sudden load changes, are superimposed to simulate complex operating conditions. The torque-speed curve is adjusted to generate motor response data under different load conditions. Temperature perturbations are then applied to simulate the impact of ambient temperature on winding resistance to optimize temperature rise prediction and thermal protection strategies. Abnormal data, such as overcurrent, overtemperature, and sudden load changes, are generated to improve fault detection capabilities and reduce the risk of overfitting the model to specific operating conditions. Finally, a physical model is driven. Signal synthesis based on the FOC model is performed, utilizing field-oriented control (FOC) theory to generate a virtual current waveform. Back-electromotive force perturbations are then applied to simulate rotor position estimation errors during Hall effect sensor failures to improve the fault tolerance of the sensorless control algorithm. The accuracy of the pre- and post-enhancement models on the test set is compared, and enhancement parameters are adjusted. The enhancement strategy is then dynamically optimized using an online learning mechanism.

[0082] S18. The data encoding unit converts the standardized overcurrent, overtemperature, overvoltage / short circuit and other fault types into one-hot encoding and constructs a multi-label classification input, and performs label encoding on the control mode to provide structured input for the fault diagnosis algorithm.

[0083] Further explanation: According to the program command after data enhancement, the fault type and coding rules are first defined. First, the fault categories such as overcurrent, overtemperature, overvoltage, and short circuit that need to be coded are determined based on the common faults of the motor, and the unique hot coding mapping of normal state [0,0,0,0], overcurrent [1,0,0,0], overtemperature [0,1,0,0], overvoltage [0,0,1,0], and short circuit [0,0,0,1] is performed; secondly, real-time data collection and fault detection are carried out. First, the motor operation data is collected in real time through the current sensor and temperature sensor, and the fault judgment threshold is set. If it exceeds If the current exceeds 120% of the rated current, if it is over-temperature, the motor winding temperature exceeds the safety threshold, if it is over-voltage / short-circuit, the voltage or impedance is abnormal, and the circuit protection mechanism is combined; then the fault type is logically judged, and for a single fault scenario, the corresponding fault code is triggered by directly matching the threshold, and if multiple faults occur concurrently, they are processed according to priority, or the coding dimension is expanded; then the coding is implemented and output, first through the embedded system using the status register or array to dynamically generate a one-hot coding vector, and through the host computer / algorithm platform using Python and other languages ​​​​encapsulated as a function for the machine learning model to call.

[0084] S20, the model building module performs key feature extraction, model design, integrated optimization, parameter initialization, dynamic optimization, model fusion and deployment based on the verified data to build a model for subsequent model training and verification, and passes it to the training and tuning module;

[0085] See also Figure 10 As shown, the step S20 includes the following steps:

[0086] S21, the feature extraction unit obtains the key features of the motor's electrical parameters, mechanical dynamics, control signals, temperature and fault, time-frequency domain, and dimensionality reduction fusion based on the preprocessed data to improve the model training speed, and transmits them to the voltage modeling unit;

[0087] It is further explained that according to the program commands after data encoding, various key features of the motor operation data are extracted; the electrical parameter features include: the instantaneous value and effective value of the three-phase current for analyzing the load state and commutation logic, the current harmonic components for detecting commutation abnormalities through FFT analysis, the current imbalance that reflects the difference in the three-phase current amplitudes of winding faults or drive circuit abnormalities, and other current signal characteristics, as well as the back electromotive force amplitude and phase for rotor position estimation in sensorless control, and the bus voltage fluctuation characteristics that characterize the power supply stability and power device status. The mechanical dynamic features include: the real-time speed and its standard deviation that reflect the dynamic response when the load suddenly changes, the torque-current relationship slope that evaluates the matching degree between the motor output efficiency and the control strategy, and other speed and torque characteristics, as well as the high-frequency vibration spectrum for detecting mechanical wear or eccentricity faults, and the noise energy distribution that distinguishes electromagnetic noise from mechanical friction noise. The control signal features include: the PWM duty cycle dynamic range and harmonic distribution for evaluating switching loss and electromagnetic interference, the d-axis and Electronic commutation-related parameters such as the q-axis current component, as well as sensorless control features such as the back-electromotive force estimation error reflecting the rotor position detection accuracy and the sliding mode observer output signal for dynamic compensation of control delay; the thermodynamic and fault characterization features include: temperature-related features such as the winding temperature rise rate and steady-state temperature for predicting overload risks, the heat sink temperature gradient for evaluating the effectiveness of the cooling system, and fault diagnosis features such as overcurrent / overvoltage events for identifying short circuit or stall faults, and the insulation resistance change rate and leakage current amplitude for detecting insulation aging; the time series and frequency domain features include: time domain statistics such as mean / variance / kurtosis / peak factor for characterizing signal fluctuation characteristics, and frequency domain energy distributions such as the FFT main frequency energy ratio for abnormal frequency detection and the wavelet packet decomposition coefficient for extracting local features of non-stationary signals; the dimensionality reduction and fusion features include: feature dimensionality reduction methods such as principal component analysis to extract low-dimensional principal components that retain key electrical and mechanical information, linear discriminant analysis to optimize feature separability in classification tasks, and multi-sensor fusion features such as the covariance matrix of current, temperature, and vibration for constructing a multidimensional state monitoring model.

[0088] S22. The voltage modeling unit obtains a circuit model based on the brushless DC motor equivalent input voltage calculation formula: "Ue = Ro*i + Li*di / dt + E, where Ue is the equivalent input voltage (V), Ro is the equivalent internal resistance (Ω), i is the phase current (A), Li is the armature inductance (H), di / dt is the time derivative of the phase current (A / s), and E is the back electromotive force (V)." This provides core mathematical model support for the control algorithm and is passed to the power modeling unit.

[0089] Further explanation: According to the program command after data feature extraction, the neural network selection is first performed, first reinforcement learning (RL), designing the Actor-Critic network optimization control strategy u(k) to minimize the speed tracking error and energy consumption, and deep neural network (DNN), using multi-layer perceptron (MLP) or LSTM to compensate for nonlinear interference such as magnetic saturation and commutation torque fluctuation; then, according to the DC brushless motor equivalent input voltage calculation formula, a voltage model is established for the motor voltage operation data; the equivalent input voltage calculation formula is evolved from the armature circuit equation established according to Kirchhoff's law, where Ue is the equivalent voltage applied to both ends of the motor winding, driving current and overcoming back electromotive force; R is the equivalent resistance of the armature winding, including losses such as wire resistance and contact resistance, and is directly measured by an LCR meter (i.e., disconnecting the power supply and measuring the winding resistance with an LCR meter) or by a stall test (keeping the motor stationary ( ω=0), applying voltage V, measuring steady-state current I) and calculating R=V / i; the i is the real-time current flowing through the armature winding, which is proportional to the electromagnetic torque and is obtained by real-time measurement through a Hall current sensor or a sampling resistor; the "Li*di / dt" is the inductance voltage drop, which describes the voltage component caused by current change; the Li is the equivalent inductance of the armature winding, reflecting the hysteresis characteristics of current change, and is obtained by applying a step voltage through the step response method and fitting the current change curve with the time constant τ=Li / R to calculate Li=τR, or by inputting a sinusoidal voltage through the frequency domain analysis method to measure the current amplitude and phase difference and calculate the inductance; the E is the reverse voltage generated by the winding cutting the magnetic field when the motor rotates, which is proportional to the speed, and is obtained by driving the motor to a steady-state speed ω and measuring the open-circuit current e through a no-load experiment, or calculating the rotor position and speed through the Hall signal of the Hall sensor, and combining Ke to estimate the real-time back electromotive force.

[0090] S23, the power modeling unit calculates the motor angular acceleration according to the formula "α=[(Kt*iq)-Po / ω-(U*io-io 2 *Ri) / ω] / (∑m*r 2 ), α is the angular acceleration (rad / s 2 ), Kt is the motor torque coefficient (N·m / A), iq is the q-axis phase current (A), Po is the motor output power (W), ω is the angular velocity (rad / s), U is the no-load input voltage (V), io is the no-load input current (A), Ri is the winding resistance (Ω), m is the motor rotor mass (kg), and r is the motor rotor radius (m). The angular acceleration is obtained and a motion model is established to facilitate motor acceleration performance analysis and control, and is then transmitted to the electromagnetic torque unit;

[0091] Further explanation: after voltage modeling, the program command establishes a dynamic model for the motor motion data according to the motor torque balance equation; the motor angular acceleration calculation formula is "J = ∑m*r 2 ”“α=dω / dt”“Te=Kt*iq”“B=(U*II 2 *Ri) / ω 2 "T L =Po / ω” is substituted into the kinetic equation “J*dω / dt+B*ω=Te-T L "It is deduced from the electromagnetic torque, load torque and dynamic loss, providing a theoretical basis for the analysis and control of motor acceleration performance. J in the formula is the moment of inertia (kg·m 2 ), B is the friction coefficient (N·m·s / rad), ω is the angular velocity (rad / s), Te is the electromagnetic torque (N·m), T L is the load torque (N·m); Kt is the electromagnetic torque generated by the motor unit current, which is determined by the number of poles, the number of winding turns and the magnetic flux, and is obtained by measuring the steady-state current iq while keeping the motor running at no load; iq is the current component directly related to the torque in the magnetic field oriented control, and is obtained by three-phase current decoupling; Po is the mechanical effective power output by the motor, that is, the power to overcome the load torque, and is measured by a torque sensor and a tachometer; ω is the instantaneous rotation speed of the motor rotor, and the speed is measured by an encoder or a Hall sensor and then converted into rad / s; U is the input voltage of the motor when the load is disconnected, which is used to calculate the power loss, and is directly measured by a voltage sensor when running at no load; I is the input current of the motor when the load is disconnected, which is used to calculate the power loss, and is directly measured by a current sensor when running at no load; Ri is the equivalent resistance of the motor winding, and is obtained by detection by a resistance sensor; m and r are both obtained from the design data provided by the motor manufacturer.

[0092] S24, electromagnetic torque unit is calculated according to the electromagnetic torque formula of brushless DC motor "Te=J*α+Bi*ω+T L , Te is the electromagnetic torque (N·m), J is the moment of inertia (kg·m 2 ), α is the angular acceleration (rad / s 2 ), Bi is the friction coefficient (N·m·s / rad), ω is the angular velocity (rad / s), T L The dynamic torque balance equation of the motor is obtained for the load torque (N·m) to predict the torque-speed response characteristics during sudden load changes or speed regulation, and is passed to the coordinate transformation unit;

[0093] Further explanation: after dynamic modeling, the program command is used to substitute the data obtained from the above calculation into the brushless DC motor electromagnetic torque calculation formula to establish the motor dynamic torque balance equation; Te is the driving torque generated inside the motor, which is used to overcome the inertia, friction and load resistance of the mechanical system; J is the inertial resistance of the motor rotor and load to acceleration, which is related to the mass distribution and geometric shape, obtained through mechanical design manuals or dynamic response tests, and can also be obtained based on the rotor geometric size and material density formula "J = ∑m*r 2 , m is the rotor mass (kg), r is the rotor radius (m)” is determined by step response experiment; the change rate of the mechanical angular velocity α is calculated by the above step S23; Bi is the damping effect that reflects bearing friction, wind resistance, etc., which is proportional to the speed. The relationship between the torque and speed of the motor under no load is measured by no-load running experiment. When the armature inductance L=0, it is calculated by the formula "Bi=Te / ω"; T L The mechanical transmission resistance, gravity load and other resistance torques imposed by external loads are measured by torque sensors or by the power balance equation "T L =ω*Po, ω is the angular velocity, Po is the motor output power".

[0094] S25. The coordinate transformation unit obtains a two-phase stationary coordinate system (dq) based on the Park transformation formula of the brushless DC motor: "id = 2 / 3 [ia*cosθ + ib*cos(θ-120°) + ic*cos(θ+120°)], iq = 2 / 3 [-ia*sinθ - ib*sin(θ-120°) - ic*sin(θ+120°)], where id and iq are the direct-axis and quadrature-axis currents (A), ia, ib, and ic are the three-phase stator currents (A), and θ is the rotor electrical angle (rad). This system implements decoupling control of the magnetic field and torque, and transmits the result to the dynamic equation unit.

[0095] Further explanation, according to the program command after the electromagnetic torque, the coordinate system transformation (eliminating time-varying coupling) is performed according to the Park transformation formula of the DC brushless motor, that is, Clark transformation, mapping the current of the three-phase stationary coordinate system (abc) to the rotor synchronous rotating two-phase stationary coordinate system (dq), simplifying the variable dimension, that is, mapping the three-phase stator current ia, ib, ic to id (direct axis current) and iq (quadrature axis current) in the rotating coordinate system, eliminating the magnetic field coupling effect, realizing decoupling control, and eliminating the time-varying term in the electromagnetic equation; the id and iq are the two-phase currents in the rotating coordinate system; the ia, ib, ic are the stator currents in the stationary coordinate system, the two-phase current is measured by a Hall current sensor or a sampling resistor, and the third phase is obtained by calculating ic=-ia-ib according to Kirchhoff's law, θ is the pulse signal directly output by the photoelectric encoder or the Hall sensor, and the analytical position is obtained, which can also be estimated by a back electromotive force observer or a high-frequency injection method; and by "x=[id,iq,ω] T " is used to calculate the state variable, where T is the sampling period.

[0096] S26. The dynamic equation unit obtains the dynamic equation according to the DC brushless motor voltage balance equation "Ld*did / dt=Ud-Ra*id+ωe*Lq*iq, Lq*diq / dt=Uq-Ra*iq-ωe(Ld*id+ψf), Ld and Lq are the direct-axis and quadrature-axis inductances (H), Ra is the armature resistance (Ω), Id and iq are the direct-axis and quadrature-axis currents (A), Ud and Uq are the direct-axis and quadrature-axis voltages (V), ωe is the electrical angular velocity, p is the number of motor pole pairs, and ψf is the permanent magnet flux (Wb)" to standardize the state space model and pass it to the model standard unit;

[0097] Further explanation: According to the program command after the coordinate system transformation, the electrical equation is first obtained according to the electrical equation formula of the brushless DC motor. It is a dynamic equation for deriving id and iq based on the electromagnetic relationship of the motor. The "d" in "diq / dt" in the formula is a differential operator, which represents the derivative of the current with respect to time, that is, "diq / dt" is the rate of change of id; the inductances Ld and Lq both reflect the energy storage capacity of the winding magnetic field, which are obtained by detecting with an inductance sensor, or by measuring the inductance value by applying an AC signal to the winding using an LCR meter method, or by calculating the inductance through a current response curve by applying a step voltage using a fixed rotor in a locked-rotor experiment, or by combining a high-frequency signal injection method or an extended Kalman filter (EKF) for real-time estimation; the Ra is the equivalent resistance of the winding, which is obtained by detecting with a resistance sensor, or by recursive least squares method ( RLS) is obtained by combining current-voltage data in real time, or by measuring the resistance of any two-phase windings with a multimeter and taking the average value; the shaft currents Id and iq are obtained by current sensor detection; the shaft voltages Ud and Uq are inverter outputs and are obtained by voltage sensor detection; the ωe is calculated by the formula "ωe=p*ω", and ω is obtained by encoder or Hall sensor detection; the p can be obtained by obtaining the motor model specification through the motor nameplate, or by disassembly measurement and counting the number of rotor permanent magnet poles to calculate p=number of poles / 2; the ψf equivalent magnetic flux is obtained by driving the motor to a constant speed through the no-load back electromotive force method, measuring the line voltage U and calculating ψf=U / ωe, or by a flux observer or finite element simulation modeling; then according to the electromagnetic induction equation "J*dω / dt=Te-Bi*ω-T L , J is the moment of inertia, Te is the electromagnetic torque, ω is the rotor speed, Bi is the friction coefficient, T L The mechanical equation for the load torque is obtained, which is the relationship between the speed ω and the torque Te. The meaning of each parameter in the formula and the data acquisition are the same as described above.

[0098] S27, the model standard unit converts the nonlinear differential equation into the following matrix A, B standard state space form x' = Ax + Bu to improve control robustness and subsequent discretization processing, and passes it to the discrete processing unit;

[0099]

[0100] Further explanation: According to the program command after the dynamic equation is constructed, the state space model is standardized according to the state matrix A and the input matrix B. First, linearization is performed. Assuming that Ld=Lq is a surface-mounted motor, the nonlinear terms are simplified. The state matrix A describes the internal dynamics of the system and is the dynamic coupling relationship between state variables. Its elements are determined by the motor parameters and operating status. The A11=-Ra / Ld is the direct-axis current attenuation rate (1 / s), which is determined by the armature resistance and the direct-axis inductance. In the formula, Ra is the armature resistance and Ld is the direct-axis inductance, which is obtained by measuring Ra with a bridge and measuring Ld with an LCR meter; the A12=ωLq / Ld is the cross-coupling effect of the quadrature-axis current on the direct-axis current (1 / s). Where ω is the current speed, Lq is the quadrature-axis inductance, which are measured or estimated in real time through dynamic parameters; A21 = -ωLd / Lq is the reverse coupling effect of the direct-axis current on the quadrature-axis current (1 / s), which is fed back through the inductance parameters and the speed; A22 = -Ra / Lq is the quadrature-axis current attenuation rate (1 / s), which is measured through static parameters; A23 = -ωψf / Lq is the back electromotive force effect of the speed on the quadrature-axis current (A / (rad / s)), where ψf is the permanent magnet flux, which is obtained by calculating "ψf = E0 / ω0" through the no-load back electromotive force test, where E0 is the no-load back electromotive force; A32 = 3Pψf / 2J is the torque driving ability of the quadrature-axis current to the speed (rad / (s 2 ·A)), where P is the number of pole pairs (an inherent parameter of the motor) and J is the moment of inertia (provided by flywheel experiments or manufacturers); A33 = -Bi / J is the speed attenuation caused by mechanical damping (1 / s), where Bi is the friction coefficient, which is fitted through step response or free deceleration experiments; the input matrix B reflects the influence of voltage on state variables, indicating the direct driving effect of input voltage on state variables, and is directly related to inverter voltages Ud and Uq; B11 = 1 / Ld is the driving efficiency of direct-axis voltage to direct-axis current (1 / H); increasing Ld will reduce the direct-axis current and response speed; B22 = 1 / Lq is the driving efficiency of quadrature-axis voltage to quadrature-axis current (1 / H); increasing Lq requires higher voltage for the motor to achieve fast torque control.

[0101] S28, discrete processing unit according to the DC brushless motor zero-order retainer formula " x(k+1) is the next state vector, e AT is the state transfer matrix (Φ), x(k) is the current state vector, is the input matrix integral term (Γ), B is the continuous input matrix, dτ is the time element, and u(k) is the input vector to obtain the discretized equation for subsequent noise modeling and parameter expansion, and pass it to the noise modeling unit;

[0102] Further explanation: according to the program command after model standardization, the continuous model is converted into a discrete form according to the DC brushless motor zero-order retainer formula to adapt to the digital controller and reinforcement learning or model predictive control (MPC), and the parameters are identified online, and the recursive least squares method (RLS) is used to update the time-varying parameters such as Ra and Ld in real time; the DC brushless motor zero-order retainer formula is composed of "Φ=e AT ” Substitute into "x(k+1)=Φ*x(k)+Γ*u(k)", where Φ is the state transfer matrix, Γ is the input matrix, T is the sampling period, and e AT is the dynamic evolution of the system within the sampling period T, e Aτ is a matrix exponential function, e is a matrix exponential function, B is the input matrix of the continuous system, τ is an integral variable, and dτ is an infinitesimal increment of the integral variable τ; the x(k+1) is the key dynamic variable of the motor at the next discrete sampling moment (k+1); the Φ is the state change when there is no input, and the natural response such as current decay and speed inertia within the time T; the x(k) is the key dynamic variable of the motor at the discrete moment k, and ia and ib are obtained by the current sensor, id and iq are calculated by Park transformation, and ω is obtained by the encoder / Hall sensor, or the state is estimated by the sliding mode observer or extended Kalman filter; the Γ is the excitation effect of the input signal on the state change, which quantifies the cumulative impact of signals such as input voltage on the state within T; the u(k) is the input vector at the discrete moment k, and Ud and Uq are calculated in real time by the PI controller, model predictive control or reinforcement learning strategy and obtained by driving the inverter through space vector modulation.

[0103] S29. The noise modeling unit obtains the noise model according to the DC brushless motor noise modeling calculation formula "x(k+1)=Φ*x(k)+Γ*u(k)+w(k), y(k)=C*x(k)+v(k), x(k+1) is the next discrete time state vector, Φ is the state transfer matrix, x(k) is the state variable vector, Γ is the input matrix, u(k) is the control input vector, w(k) is the process noise, y(k) is the output of the observation equation, C is the observation matrix, and v(k) is the observation noise" to enhance the robustness of the model.

[0104] Further explanation, according to the program command after the discretization equation, the noise model is first obtained according to the DC brushless motor noise modeling calculation formula to introduce process noise w(k) and observation noise v(k) to enhance the robustness of the model; the Φ is obtained by discretizing the continuous model of the motor and is calculated through the above step S28; the x(k) is the various dynamic characteristics of the motor, and the three-phase currents ia, ib, and ic are converted into id and iq through Park transformation, and the speed ω is measured by an encoder or back electromotive force method; the Γ reflects the excitation effect of the control voltage Ud and Uq on the state and is calculated through the above step S28; the u(k) is the inverter output voltage Ud, Uq or PWM modulation signal, which is obtained by voltage sensor detection; the w(k) is the unmodeled dynamic or external disturbance such as magnetic saturation and commutation interference, which is obtained by statistically analyzing the variance of the unmodeled dynamic through experimental data; the C is the current and speed of the state variable and the output quantity. The linear relationship is obtained through sensor detection; the v(k) is caused by the sensor error and is obtained by calibrating the sensor error; then parameter identification and expansion, first online parameter identification, taking the time-varying parameter (ψf) as the extended state quantity, real-time update through recursive least squares (RLS) or Kalman filtering, and multi-physics field coupling modeling, combining the magnetic circuit model and thermal model to enhance the model accuracy; then model implementation and verification, first Simulink modeling, building a modular model in Simulink, supporting code generation and SIL testing, and simulation verification, verifying the matching degree between the model prediction and the actual response through hardware-in-the-loop testing or measured data comparison; finally, intelligent control algorithm integration, first reinforcement learning (RL), optimizes the control strategy u(k) to minimize tracking error and energy consumption, such as combining adaptive fuzzy PID to achieve high-precision speed tracking, and neural network compensation, using deep network to compensate for unmodeled dynamics.

[0105] S30, the training and tuning module trains the model according to the extracted data and adjusts the model parameters to optimize the model performance and predetermine the DC brushless motor operation control parameters to ensure the accuracy of the motor operation control parameters, and transmits them to the model verification module;

[0106] See also Figure 11 As shown, the step S30 includes the following steps:

[0107] S31, the hardware acceleration unit accelerates large-scale data training based on the model through hardware selection and configuration, data parallel processing optimization, edge-cloud collaborative architecture, performance tuning and energy efficiency management to shorten the model training cycle and pass it to the supervised learning unit;

[0108] Further explanation: according to the program commands after the model is built, the hardware is selected and configured first. First, the GPU cluster is deployed, and high-performance GPUs such as NVIDIA A100 / A800 are used to accelerate the forward calculation and back propagation of the neural network. Parallel processing is achieved for large-scale multi-physics field current, speed timing and other simulation data to shorten the single training cycle, as well as edge computing device integration. In embedded scenarios, low-latency reasoning of lightweight models is achieved through NVIDIA Jetson series or Huawei Ascend chips, supporting real-time commutation control and dynamic compensation; then its data parallel processing is optimized, first the distributed training framework, based on the multi-GPU data parallel architecture of TensorFlow / PyTorch, and the Hall sensor signal, torque curve and other motor operation data are divided into multiple sub-batches for synchronous training to improve throughput, as well as heterogeneous computing acceleration, using CUDA or OpenCL technology to heterogeneously accelerate motor dynamics equations and neural network operations such as back electromotive force model, reducing communication delay by 60%; then the edge-cloud collaborative architecture, first cloud training and edge reasoning, The cloud uses concurrent training of over 10,000 intelligent agents to generate high-precision models, supporting dynamic physical simulations of multiple scenarios such as load mutations and temperature drifts, and deploying the quantized TinyML model to STM32 or Raspberry Pi on the edge to achieve microsecond-level commutation logic response; finally, performance tuning and energy efficiency management, first with mixed-precision training, using FP16 / INT8 quantization technology to compress the model parameter scale, taking into account both computing speed and energy consumption, adapting to low-power scenarios such as electric vehicle drive, as well as hardware instruction set optimization, deep adaptation to GPU / TPU instruction sets, and instruction-level acceleration of core control algorithm modules such as PID parameter optimization and error integral calculation, improving energy efficiency by more than 15%.

[0109] S32, the supervised learning unit pre-trains the model using simulation data by physically guiding the neural network architecture, adopting a hybrid architecture, Xavier / He initialization, and transfer learning according to the model, so as to initialize the network weights and pass them to the loss function unit;

[0110] Further explanation: according to the program command after hardware acceleration, first the physical embedding layer is added after the input layer. The parameters such as Rs and Ls interact with the dynamic signal through the weight matrix, forcing the model to learn the basic electromagnetic relationships such as Ohm's law V=IR+L*di / dt; secondly, the hybrid network structure is first the feedforward branch, which processes the static parameters and outputs the back electromotive force coefficient Ke and other latent variables related to the electromagnetic characteristics, and performs the timing branch, using LSTM or Transformer to process the dynamic signal and capture the timing characteristics such as the commutation transient; then Xavier / He initialization is selected according to the characteristics of the network activation function. Xavier initialization is used for the fully connected layer such as the electromagnetic parameter mapping layer, which is suitable for saturation activation functions such as Sigmoid / Tanh, balances the input and output variance, and reduces the risk of gradient disappearance. He initialization is used for the timing processing layer such as LSTM / CNN, which is designed for ReLU type activation functions. It compensates for the negative interval suppression effect of ReLU by amplifying the initial variance, improves the stability of the deep network, and physical guidance initialization, combined with the motor stator resistance R, The initial weight range is set for the inductance L and other intrinsic parameters. For example, the bias of the current prediction layer is initialized to the theoretical steady-state current value to accelerate the model's learning of electromagnetic characteristics. Then, transfer learning is performed. First, cross-scenario parameter migration is performed. The weights of some network layers, such as the back-electromotive force prediction module, are extracted from pre-trained models of other motor types, such as permanent magnet synchronous motors, to reduce the data requirements for BLDCM model training. For scenarios with large differences in parameters such as pole pairs p, a random initialization combined with fine-tuning strategy is adopted to retain the general feature extraction capability. Finally, simulation data pre-training is performed. Based on the motor operation simulation data generated by Simulink, the network base layer is pre-trained, and then the output layer is fine-tuned with real data, as well as dynamic characteristic adaptation. In the coarse initialization stage, Gaussian distribution is used to randomly initialize the weights of the entire network to quickly explore the parameter space. In the fine adjustment stage, zero bias initialization is used for the output layer related to the commutation logic, forcing the model to prioritize learning steady-state characteristics, and then gradually release dynamic characteristic learning. Special treatment is also performed on the timing-sensitive layer. A small value is set in the weight initialization of the forget gate of LSTM to avoid excessive forgetting of historical states in the initial stage, thereby improving the timing continuity of speed prediction.

[0111] S33, the loss function unit calculates the loss function of the brushless DC motor according to the formula "Lt=∑(yi-y'i) 2 / N+λ1*∑max(0,|Ii|-Imax)+λ2*∑∥ea(t)-Ke*ω(t)∥2+λ3*∑∥y' t+1 -y't∥2+λ4*E x～D [∥xe*L(f(x),y)∥2]+∑(σt -2*Lt+logσt) / To, Lt is the motor loss function, yi is the actual true value, y'i is the model prediction value, N is the number of samples, λ1 is the penalty weight coefficient, Ii is the actual phase current, Imax is the maximum allowable current of the motor, λ2 is the balance weight, ea(t) is the actual back electromotive force, Ke is the back electromotive force coefficient, ω(t) is the speed at time t, λ3 is the smoothness weight, y' t+1 is the model output at time t+1, y't is the model output at time t, λ4 is the adversarial training intensity coefficient, E x～D is the expected value of the input data, xe is the input data gradient, L(f(x), y) is the basic loss function between the model output and the true value, σt is the prediction variance at time t, Lt is the basic loss term, and To is the total step length of the time series. The loss function value is obtained to achieve high-precision and high-robustness DC brushless motor control and is passed to the reinforcement learning unit;

[0112] Further description, according to the program command after hardware acceleration, the loss function value is obtained according to the DC brushless motor loss function calculation formula, and the loss function value is obtained by "Lm=∑(yi-y'i) 2 / N""Lc=λ1*∑max(0,|Ii|-Imax)""Ls=λ3*∑∥y' t+1 -y't∥2"La=λ4*E x～D [∥xe*L(f(x),y)∥2]""Lu=∑(σt -2 *Lt+logσt) / T” is substituted into “Lt=Lm+Lc+Le+Ls+La+Lu” to obtain, where Lm is the mean square error (i.e., basic regression loss), Lc is the current limit penalty term (belongs to physical constraint loss), Le is the back electromotive force consistency loss (belongs to physical constraint loss), Ls is the commutation smoothness constraint (i.e., time sequence dynamic loss), La is the input disturbance robustness loss (i.e., adversarial training loss), and Lu is the variance weighted loss (i.e., uncertainty modeling loss); the yi is obtained from sensor or simulation data; the y'i is output through neural network inference; the N is determined by the size of the data set; the λ1 is optimized through grid search or dynamic adjustment strategy; the Ii is measured by the current sensor; the Imax is set according to the motor rated parameters or safety regulations; the λ2 is obtained through historical experience; the ea(t) is indirectly measured by the back electromotive force observer or sensor; the Ke is calculated by the motor electromagnetic parameters; the ω(t) is obtained through the encoder or Hall sensor; the λ3 is adjusted according to the commutation jitter tolerance; the y' t+1 , y't is output by the control signal generation module; the λ4 is adjusted according to the model robustness requirements, generally 0.05-0.2; the E x～DBy designing data in advance; the xe is calculated by back propagation; the σt is estimated by Monte Carlo Dropout sampling or Bayesian neural network; and the To is set according to the control frequency.

[0113] S34. The reinforcement learning unit obtains the reward function value and conducts reinforcement learning based on the DC brushless motor reward function calculation formula "fr=ε*(1-|ωr-ωo| / 0.1ωr)+β*(1-Pl / Pr)-γ-*∑max(0,|Ii|-Imax), where fr is the reward function value, ε is the speed tracking accuracy weight, ωr ​​is the target speed, ωo is the actual speed, β is the energy efficiency optimization weight, Pl is the actual power loss, Pr is the rated power, γ is the safety constraint weight, Ii is the actual phase current, and Imax is the maximum allowable current of the motor" to reduce the overshoot in the speed step response and shorten the steady-state time, and passes it to the parameter optimization unit;

[0114] It is further explained that according to the program command after the loss function is calculated, the reward function value is obtained according to the DC brushless motor reward function calculation formula, the model is deployed in the simulation environment, and the control strategy is optimized through trial and error. The reward function is designed as a weighted combination of speed tracking accuracy and energy consumption; the α, β, and γ are all obtained by dynamic adjustment algorithm optimization; the ωr is the expected speed value set for the control task, which is specified by the host computer or user input; the ω is measured by an encoder or Hall sensor; the Pl is the total loss during motor operation, including copper loss and iron loss, and the three-phase currents Ia, Ib, and Ic are measured by current sensors and calculated in combination with motor parameters (Pl = Ia 2 *Ra+Ib 2 *Rb+Ic 2 *Rc+Kh*ω 2 , Ra, Rb, Rc are phase resistances, and Kh is the iron loss coefficient); Pr is the upper limit of continuous operating power allowed by the motor design, which is measured by the motor nameplate parameters or specifications; Ii is measured by the current sensor; Imax is set according to the motor rated parameters or safety specifications.

[0115] S35. The parameter optimization unit obtains the reward function value and strengthens learning based on the basic acquisition function calculation formula of the brushless DC motor: "fg=W1*EI(Ms(ωr,ω))+W2*Ei(Po / / Pr)-γ*∑max(0,|Ii|-Imax), where fg is the basic acquisition function value, W1 is the weight coefficient of the speed tracking error term, W2 is the weight coefficient of the energy efficiency optimization term, EI is the expected improvement function of the error term, Ms is the speed error function, ωr is the target speed, ωo is the actual speed, Ei is the energy efficiency term function, Po is the actual power loss, Pr is the rated power, γ is the safety constraint weight coefficient, Ii is the actual phase current, and Imax is the maximum allowable current of the motor" to reduce the overshoot in the speed step response and shorten the steady-state time.

[0116] It is further explained that according to the program command after reinforcement learning, the basic acquisition function value is obtained according to the basic acquisition function calculation formula of the DC brushless motor, and the learning rate, batch size and other parameters are adjusted by Bayesian optimization or genetic algorithm to avoid local optimality; the basic acquisition function calculation formula of the DC brushless motor introduces multi-objective weights and safety penalty terms for the dynamic response and safety constraints of BLDC, and is transformed through fg=W1*EI(x)+W2*Ei(x)-γ*Pe(Imax); the W1 is the priority of adjusting the speed control accuracy in the overall optimization, which is determined by Bayesian optimization or dynamic adjustment strategy; the W2 is the optimization goal of balancing power loss and system efficiency, which is determined by Bayesian optimization or dynamic adjustment strategy; the EI is the potential optimization space for evaluating the speed error based on the Gaussian process, which is the power loss The energy consumption ratio is as follows: Ms is the deviation between the quantified target speed and the actual speed, which is calculated by the mean square error formula; Ei is the degree of deviation between the actual energy consumption and the rated value; ωr is the expected speed value set by the control task, which is specified by the control system host computer or user input; ω is measured by an encoder or Hall sensor; Po is the total loss during motor operation, which is calculated by the armature current and motor parameters; Pr is the upper limit of continuous operation power allowed by the motor design, which is determined by the motor nameplate parameters or specification sheet; γ is the intensity of the control current over-limit penalty term to prevent the motor from overheating or damage, which is determined by Bayesian optimization or dynamic adjustment strategy; Ii is collected in real time by a current sensor, or estimated based on a voltage-current model; Imax is set according to the motor rated parameters or safety specifications, or is determined by the motor thermal characteristics or driver protection threshold.

[0117] S40, the model verification module performs model verification and evaluation based on the trained model, and adjusts and optimizes the model based on the verification results to improve the performance and accuracy of the DC brushless motor operation model, and transmits the results to the parameter setting module;

[0118] See also Figure 12As shown, the step S40 includes the following steps:

[0119] S41. The classification and labeling unit classifies the data samples into corresponding categories according to the abnormal operating quality categories of the brushless DC motor and labels them for subsequent machine learning of the AI ​​model and transmits them to the learning update unit;

[0120] Further explanation, according to the abnormal phenomenon of DC brushless motor operation quality, classification is carried out: "dynamic response abnormality" is category 1, "control signal misjudgment" is category 2, "edge computing delay" is category 3, and "insufficient model generalization" is category 4; when category 1 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (1, 0, 0, 0), when category 2 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (0, 1, 0, 0), when category 3 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (0, 0, 1, 0), when category 4 is a positive sample, the remaining categories are negative samples, that is, the labeled data is (0, 0, 0, 1); the above four abnormal categories are relatively typical abnormal classifications for brushless motor design quality control, including but not limited to the above four categories, and one or more categories can also be re-divided into more categories, and a more detailed and in-depth division can be carried out according to the actual design quality effect. The above classification is not fixed. After each category change, the model is retrained to adapt to the subsequent actual operation of the motor. The operation quality abnormalities of the brushless DC motor include: 1) abnormal dynamic response: due to insufficient coverage of AI model training data, the parameter adaptability is poor, or due to the failure of online parameter identification algorithm, the resistance or inductance changes are not updated in time, resulting in speed tracking lag, large torque fluctuation, and stuttering or overshoot during acceleration / deceleration; 2) control signal misjudgment: due to the interference of sensor signals by inverter high-frequency noise, or due to multi-source data fusion abnormalities such as time synchronization between current and vibration signals, non-commanded reversal, sudden stop or jitter are caused; 3) edge computing delay: due to insufficient computing resources of lightweight models, the real-time control requirements cannot be met, or due to the interruption of edge-cloud collaborative communication, the strategy update is delayed, resulting in commutation delay, sudden increase in energy consumption and even triggering of overcurrent protection during high-speed operation; 4) insufficient model generalization: due to the lack of special working condition samples such as regenerative braking and low-voltage operation in the training data, or due to insufficient fusion of physical model and data drive, strategy conflict occurs, resulting in speed loss of control or efficiency drop in the weak magnetic control area.

[0121] S42, the learning update unit determines the training model according to the gradient descent data, and inputs the DC brushless motor operation quality data in the training set into the model for simulation training to ensure the accuracy of model recognition, and passes it to the cost verification unit;

[0122] It is further explained that according to the specific requirements of the motor for driving the equipment such as power, speed, time, etc., the motor operation abnormality prevention measures are obtained through model training and matching to prevent various motor operation quality abnormalities. The DC brushless motor operation quality abnormality prevention measures include: 1) Data and model optimization: first implement data enhancement, inject extreme working conditions such as load mutation and temperature rise limit through the simulation platform to improve the robustness of the model, and perform hybrid modeling combined with the physical feasibility of the physical information neural network (PINN) constraint control strategy; 2) Hardware and signal processing: first perform anti-interference design, add RC filter circuit or magnetic ring at the sensor signal acquisition end to suppress high-frequency noise, use shielded cable and optimize PCB layout to reduce electromagnetic interference, and implement redundancy mechanism, deploy dual-channel Hall sensor and local control instruction cache to avoid single point failure; 3 ) System real-time performance guarantee: First, perform model pruning, lightweight processing of the neural network to adapt to the computing power limitations of embedded platforms such as STM32, and optimize communication, using dual-mode communication such as CAN+Bluetooth to ensure real-time transmission of key instructions; after preventing various quality abnormalities, first perform simulation verification according to the program command, and compare the model prediction and actual response through Simulink hardware-in-the-loop (HIL) testing to evaluate the dynamic performance, and then embed deployment, and transplant the lightweight model to microcontrollers such as STM32 to achieve low-latency real-time control; then perform continuous learning and online updates, first online parameter identification, combined with recursive least squares (RLS) to dynamically update the motor parameters, adapt to the characteristic drift caused by temperature rise or aging, and perform edge-cloud collaboration, the cloud regularly retrains the model, and the edge adapts to the new working environment through incremental learning.

[0123] S43. The cost verification unit obtains the cost function value after model training according to the cost function calculation formula after model training: "J(θ)=∑[1 / (1+σ(xi))]*[y*log(hθ(x))], where J(θ) is the cost function after model training, σ(xi) is the estimated value of the signal-noise variance, y is the true label value, and hθ(x) is the model prediction probability" to improve the noise robustness of the model in industrial scenarios and pass it to the training evaluation unit;

[0124] It is further explained that the calculation formula of the cost function after training is a prediction for noise robustness correction. For electromagnetic interference in motor operation, a high-frequency noise filter weight is added to the loss function to reduce the impact of abnormal data; the θ is the model weight parameter, that is, the weight connecting each layer of neurons in the neural network, which is used to map input signals such as current and vibration to output predictions such as fault type and efficiency value. It is directly loaded from the solidified model without the need for real-time update; the σ(xi) is a quantitative indicator of the noise intensity of the input signal xi under specific working conditions. The variance is calculated using a sliding window by collecting the steady-state operation data of the motor, and the high-frequency noise power is extracted by FFT as a noise variance estimate, or Kalman filtering or adaptive filtering is used in embedded systems for real-time update, which is suitable for dynamic load scenarios; the y is the reference true value of the motor operation state, including the fault type code in the classification task, The efficiency and torque values ​​in the regression task are synchronously collected by high-precision torque sensors and encoders in a controllable environment. The artificial simulated fault labels are collected by offset Hall sensors, and labels are generated by expert knowledge combined with historical maintenance records and fault tree analysis (FTA). The hθ(x) is the predicted output of the neural network for the input x, that is, the confidence level (range [0,1]) that it belongs to a certain type of fault or state. The real-time sensor data x is input into the model through forward reasoning. The output layer uses the Sigmoid function to generate probability values. The current signal is scaled to the [-1,1] interval through normalization and noise reduction is performed by wavelet transform or moving average filtering. The xi is the single-sampled motor operation data, usually a multi-dimensional time series signal, obtained by current Hall sensor or MEMS vibration sensor detection, and multi-channel synchronous sampling is achieved through FPGA or high-precision ADC.

[0125] S43. The training evaluation unit obtains a comprehensive evaluation score F according to the model evaluation score calculation formula "F = 2[(Tp + Tn) / (Tp + Fp + Fn + Tn)] * [Tp / (Tp + Fn)] / [(Tp + Tn) / (Tp + Fp + Fn + Tn)] + [Tp / (Tp + Fn)], where F is the training model evaluation score, Tp is the number of true positive samples, Tn is the number of true negative samples, Fp is the number of actual false positive samples, and Fn is the number of false negative samples" to ensure the comprehensive performance of the model in motor control and transmit it to the processing center;

[0126] It is further explained that the model evaluation score calculation formula is obtained by substituting "Ac=(Tp+Tn) / (Tp+Fp+Fn+Tn)" and "Re=Tp / (Tp+Fn)" into "F=2(Ac*Re) / (Ac+Re)", where Ac is the precision and Re is the recall; after the model training is completed, a test is performed to verify the accuracy and reliability of the model. During the test, problems with the model may be found and optimized and adjusted; the training model evaluation score, namely the F value, is the harmonic mean of the precision and recall, which can evaluate the performance of the classification model. The value range is 0 to 1, where 1 is the best performance and 0 is the worst performance. The higher the F value, the better the prediction effect of the model, and vice versa. Therefore, if the F value is close to 1, it indicates that the model performs well while maintaining a balance between precision and recall; hyperparameters are adjusted according to the model's learning rate, regularization coefficient and other performance to optimize model performance.

[0127] S44. The processing center compares the actual evaluation score of the motor model with the evaluation score standard of the motor model stored in the memory: if it meets the standard, it is the predetermined DC brushless motor operation control parameter; if it does not meet the standard, it is transmitted to the alarm and notified to continue training.

[0128] It is further explained that the test evaluation standard of the training model for identifying abnormal operation quality of the brushless DC motor is best with a score greater than 0.8, which is determined by the following factors: 1) Control and drive system: improper controller parameter settings: AI model training data that is not adapted to extreme working conditions (such as sudden load changes, high temperature) leads to dynamic response lag or overshoot, and driver / electronic speed controller (ESC) failure: unstable output current or commutation logic error, causing stuttering and reversal abnormalities. In addition, commutation algorithm defects: insufficient accuracy of six-step commutation or sinusoidal wave control, resulting in current waveform distortion and torque pulsation; 2) Power supply and signal interference: power supply voltage fluctuation: unstable voltage or battery aging leads to increased speed tracking error, triggering stalling in extreme cases, and filter circuit failure: the inverter high-frequency noise is not effectively suppressed, interfering with the Hall sensor signal, causing commutation delay, and multi-source signal asynchrony: abnormal fusion of current, vibration and position sensor data, Leading to misjudgment of control instructions; 3) Machinery and load: bearing wear or eccentricity: increased mechanical friction causes excessive vibration (>0.05mm), decreased acceleration efficiency, and sudden load change or overload: exceeding the rated load capacity of the motor, causing speed loss of control or a sudden increase in energy consumption. In addition, the cooling system fails: the air duct is blocked or the fan stops, and the overheating protection shutdown is triggered when the temperature rise exceeds 150°C; 4) Sensor and feedback: Hall sensor failure: installation offset or dust accumulation causes rotor position detection error (>0.5°), abnormal commutation, and encoder signal distortion: high humidity environment causes corrosion or electrostatic discharge (ESD) damage, and feedback interruption; 5) Environment and maintenance: temperature and humidity exceed the limit: low temperature (<-20°C) grease solidifies or high temperature and high humidity accelerate bearing corrosion, and dust pollution: failure to clean regularly causes dust accumulation in the winding, increasing the risk of short circuit, and loose wiring / cold solder joints: poor contact of the power cord in a vibration scenario causes intermittent stalls. After training, the model is automatically matched and controlled during motor operation to avoid various operating quality abnormalities; if the model evaluation score does not meet the standard, it returns to step S42 to retrain and optimize and adjust the parameters until the standard is met.

[0129] S50, the parameter setting module performs a motor operation control test based on the brushless motor control parameters predicted by the trained model to obtain the optimal DC brushless motor operation control parameters to ensure the accuracy of future operation control, and transmits them to the operation application module;

[0130] See also Figure 13 As shown, the step S50 includes the following steps:

[0131] S51. The parameter optimization unit obtains dynamically adjusted PID parameters according to the real-time optimization calculation formula of the brushless DC motor PID parameters: "Kp(t) = Rl(e′, ∫e), Ki(t) = Nn(Eh, Tr), Kd(t) = Ga(de / dt, ωe), Kp(t) is the dynamic proportional coefficient, Rl is reinforcement learning, e′ is the speed error change, ∫e is the error accumulation, Ki(t) is the dynamic integral coefficient, Nn is the neural network, Eh is the historical error sequence, Tr is the torque pulsation rate, Kd(t) is the dynamic differential coefficient, Ga is the genetic algorithm optimized Kd parameter, de / dt is the error differential term, and ωe is the speed tracking error" to achieve PID parameter adaptive adjustment and transmit the obtained parameters to the energy efficiency optimization unit;

[0132] It is further explained that the dynamic adjustment PID parameters are obtained by the real-time optimization calculation formula of the motor PID parameters to achieve the minimization of the speed error e and torque pulsation Tr; the Kp(t) is the weight of the proportional term dynamically adjusted by reinforcement learning Rl according to the instantaneous change of the current speed error E′ and the historical error accumulation ∫e; the Rl is the weighted loss function of dynamically adjusting Kp to minimize E′ and ∫e; the e′ is the instantaneous change rate of the speed deviation at the current moment to reflect the dynamic response state of the system, and the deviation between the motor speed calculation and the target value is collected in real time by the Hall sensor / encoder; the ∫e is the integral value of the historical error to eliminate the steady-state error, and the error value is calculated by real-time accumulation in the microcontroller, and the integral limit is set to prevent saturation; the Ki(t) is the optimal integral term weight predicted by the neural network Nn based on the historical error sequence Eh and the torque pulsation rate Tr; the Nn is the correlation between Eh and Tr to predict the optimal Ki value; the Eh is the past N sampling cycles The error data set within is used to train the neural network, and a sliding window is used to store the latest N error data for time series feature extraction of the neural network; the Tr is the ratio of instantaneous torque to average torque, measured by a current sensor or a vibration sensor; the Kd(t) is the differential term weight optimized by the genetic algorithm Ga based on the error change rate de / dt and the speed tracking error ωe; the Ga uses de / dt and ωe as fitness functions to screen the optimal solution for suppressing oscillation; the de / dt is the real-time calculated value of the error change rate, calculated by the first-order difference method or extracted by Butterworth high-pass filtering; the ωe is the absolute deviation between the target speed and the actual speed, directly obtained by the Hall sensor / encoder; then the duty cycle is dynamically generated through the formula "Dc=AI(e(t),ωt,θf), Dc is the dynamic duty cycle, AI is the artificial intelligence algorithm" to generate the optimal PWM duty cycle based on the current speed error e(t), target speed ωt, and magnetic field oriented control angle θf.

[0133] S52, the energy efficiency optimization unit calculates the motor torque pulsation suppression according to the formula "Tr=AI(sqrt[∑(Tk-T′)2 / N)] / Tn*100%≤5%, Tr is the torque pulsation rate, AI is the artificial intelligence algorithm, Tk is the instantaneous torque (N·m), T′ is the average torque (N·m), N is the number of samples, and Tn is the rated average torque (N·m). The torque pulsation rate and energy efficiency optimization constraints are obtained to improve the overall system efficiency and are transmitted to the back electromotive force unit;

[0134] It is further explained that the torque pulsation rate is obtained by the motor torque pulsation suppression calculation formula, and the instantaneous torque fluctuation is analyzed in real time by AI to suppress the torque pulsation to within 5% of the rated value; the Tr reflects the instantaneous torque fluctuation amplitude; the AI ​​optimizes the current waveform, commutation strategy and magnetic field oriented control parameters in real time to suppress the pulsation caused by non-ideal back electromotive force and commutation delay; the Tk is the real-time torque value at a certain moment in the operation of the motor, which is indirectly calculated by a high-precision sensor or a current-back electromotive force model; the T′ reflects the stability of the torque output over a period of time, which is extracted by sliding average filtering of the Tk sequence; the N is the number of torque samples in the data window, which is dynamically adjusted according to the control period, generally 10-100 points; the Tn is the average torque under the rated working conditions marked on the motor nameplate, and the nameplate The calibration value or load test is obtained; then the comprehensive efficiency is obtained according to the energy efficiency optimization constraint formula "ηa=Po / (V*I-ΔPa)≥90%, ηa is the comprehensive efficiency, Po is the motor output power (W), U is the input voltage (V), I is the input current (A), ΔPa is the power loss after AI compensation (W)", and by compensating the switching loss ΔPa, the overall efficiency of the system is improved to more than 90%. In the formula, ηa is the total efficiency ≥90% after AI optimization to reduce energy consumption, Po is the effective power output of the motor shaft, U is the DC bus voltage of the drive motor, through real-time measurement or through power supply parameter calibration, I is the total current when the motor is running, including phase current and excitation components, collected by current sensor, ΔPa includes iron loss, copper loss, switching loss, etc., and is dynamically estimated and compensated by the AI ​​model.

[0135] S53. The back-EMF unit obtains the back-EMF AI compensation and performs field-oriented control according to the DC brushless motor back-EMF AI compensation calculation formula: "Ea' = k*N1*ε*D*L*Bo / a*ω+Ne(ω',I'), where Ea' is the estimated back-EMF value after compensation (V), k is the number of armature slots, N1 is the number of winding element turns, ε is the pole arc coefficient, D is the armature outer diameter (m), L is the core length (m), Bo is the air gap flux density (T), a is the number of armature parallel branches, ω is the actual angular velocity of the motor (rad / s), Ne is the neural network compensation term (V), ω' is the compensated speed (rad / s), and I' is the compensation current (A)" to balance torque output and energy consumption, and transmits the result to the operation test unit;

[0136] Further explanation, the motor back electromotive force AI compensation calculation formula is obtained by substituting "Kb=k*N1*α*D*L*Bo / a" into "Ea′=Kb*ω+Nn(ω′,I′), and the neural network Nn is used to correct the back electromotive force coefficient Kb to compensate for the estimated deviation of the speed ω′ and the current I′, thereby obtaining the back electromotive force AI compensation, where Kb is the back electromotive force basic coefficient; the k is the number of rotor slots, which is determined by the motor design parameters in the manufacturer's specification or design drawings; the N1 is the number of windings of a single coil, which is obtained by the winding design parameters or by directly measuring the coil; the α is the ratio of the pole arc length to the pole pitch, which is determined by the motor magnetic circuit design, generally 0.6-0.9; the D is obtained by mechanical measurement or design drawings; the L is the direct measurement of the physical size of the iron core or the reference design parameters; the Bo is the magnetic induction intensity in the air gap, which is measured by a magnetic sensor or calculated based on the magnetic circuit design formula; the a is determined by the motor winding connection method and is obtained by the design parameters; the ω is passed The speed is measured in real time by an encoder or Hall sensor, or converted by the formula ω=2πn / 60; the Ne is generated by neural network training, and the input is the speed error and current error; the ω′ is the difference between the target speed and the actual speed, which is calculated in real time by the closed-loop control system; the I′ is the difference between the target current and the actual current, which is obtained by measuring the current sensor and comparing it with the set value; then the optimized current vector angle is obtained by the FOC current vector optimization calculation formula "θf=arctan(Iq / id)+Rl(T′,η′)", and the magnetic field oriented control is obtained by combining reinforcement learning Rl to optimize the direct-axis current id and quadrature-axis current iq vector angles to balance torque output and energy consumption, where θf is the optimized current vector angle, iq is the q-axis current, id is the d-axis current, Rl is the reinforcement learning compensation term, T′ is the target torque, η′ is the target efficiency threshold, arctan(iq / id) is the ratio of the q-axis current iq to the d-axis current id, and the corresponding inverse tangent angle value is inversely calculated.

[0137] S54, the operation test unit performs an operation control test on the target drive device according to the DC brushless motor operation control parameters predetermined by the model and obtains motor operation quality information to ensure the accuracy of the motor control parameters, and transmits the information to the processing center;

[0138] Further explanation: according to the control parameters and program control commands predetermined by the model, the initial position detection and commutation triggering are performed first. The rotor position is sensed first, and the rotor pole position signal is obtained in real time using a Hall sensor or a magnetic encoder. The accuracy must reach ±1° electrical angle. There is no sensor solution to estimate the rotor position through back electromotive force zero-crossing detection or high-frequency injection method, as well as commutation logic execution. First, the six-step commutation method is used. The winding power-on phase is switched every 60° electrical angle. The power tube on-off sequence is controlled by a three-phase inverter circuit to generate a rotating magnetic field to drive the rotor, and FOC control is performed. The magnetic field orientation technology is used to decompose the current into d-axis (excitation) and q-axis (torque) components, and continuous magnetic field control is achieved through SVPWM modulation; then Speed ​​and torque regulation, first of all, the speed regulation method is carried out, and the motor input power is changed by adjusting the output voltage of the driver by using voltage regulation, which is suitable for wide range speed regulation scenarios, and PWM modulation is carried out, and the PWM duty cycle is adjusted to change the average voltage to achieve high-precision speed control, and closed-loop control is implemented. First, the speed loop, the encoder or Hall signal feedback actual speed, and the PWM output is adjusted by the PID algorithm, with a response time of ≤5ms, and the torque loop is carried out. The q-axis current is monitored in real time based on the current sensor, compared with the target torque and dynamically compensated for load fluctuations; then the dynamic performance is optimized, first the magnetic field weakening control is carried out, and the d-axis negative current is injected under high-speed conditions to weaken the magnetic field to expand the speed range and meet the constraints: sqrt(id 2 2+iq 2 )≤Imax, and nonlinear compensation is performed. Temperature sensor data is introduced to correct for winding resistance temperature drift, improving current control accuracy. Feedforward compensation offsets torque ripple caused by back-EMF, reducing noise to below 45dB. Finally, the protection mechanism and fault handling begin with real-time monitoring and protection. Overcurrent protection monitors the instantaneous values ​​of the three-phase currents, triggering the hardware protection circuit to shut down the power transistors within 10μs. Overtemperature protection is also implemented. When the winding temperature exceeds 150°C, the operation is automatically derated and the carrier frequency is reduced to reduce switching losses. Fault diagnosis begins with phase loss detection, identifying open-circuit faults through current spectrum analysis with an accuracy of ≥98%. Bearing wear prediction, based on FFT analysis of vibration signals, provides early warning of mechanical failures. Static parameters, dynamic performance, environmental parameters, and fault diagnosis are simultaneously tested to obtain corresponding operating quality information to ensure the reliability and safety of motor operation control. During this process, the data calculated according to steps S51-S53 are compared and control parameters are adjusted to ensure optimal motor operating quality information.

[0139] S55. The processing center compares the motor operation quality information with the brushless motor operation quality standard stored in the memory. If the standard is met, the information is set as the brushless DC motor operation control parameter. If the standard is not met, the information is transmitted to the alarm device and the parameter adjustment is notified.

[0140] Further explanation: the operation control parameters of the brushless DC motor include: 1) electrical control parameters: (need to be dynamically adjusted according to load conditions) operating voltage range, (generally set to 1.3 times the rated current to prevent motor overload) stall current setting value, (determines the upper limit of motor power output, exceeding this value may damage winding insulation) maximum continuous current and other voltage and current parameters, as well as (no-load speed at 1V voltage) KV value definition, (1:30 wide range speed regulation is achieved through PWM modulation, and voltage fluctuation compensation needs to be controlled within ±10%) speed regulation range and other KV value and speed relationship; 2) key parameters of the control mode: (under the six-step commutation method, the power tube conduction angle is 240° electrical angle, the commutation period is determined by the Hall signal phase deviation commutation timing, (carrier frequency range is 10-20kHz, duty cycle adjustment directly affects the average voltage output) pulse width period and other square wave control parameters, as well as (maintaining id≈0 to achieve maximum torque / current ratio) d-axis / q-axis current components, (injecting d-axis negative current to expand the speed range, id 2 +iq 2 )≤Imax) weak magnetic control parameters and other FOC control parameters; 3) dynamic performance adjustment parameters: (measured by the motor drag method, calculated as Ke=U / n) back electromotive force constant, (phase resistance difference should be ≤±5%, phase inductance needs to be measured at PWM carrier frequency to match the control algorithm) phase resistance and inductance and other torque and efficiency optimization, and (determine the relationship between mechanical speed and electrical speed, identified by Hall signal or back electromotive force waveform) pole pair number, (affects magnetic field strength, matched with the number of stator slots to reduce Low torque pulsation) magnetic steel pole number and magnetic field parameters; 4) Environmental and protection parameters: (winding temperature rise ≤80K under rated load, ambient temperature 40℃, power device junction temperature ≤125℃) winding temperature rise limit, (surface temperature difference ≤15℃ under forced air cooling conditions to avoid local overheating and magnet demagnetization) heat sink temperature difference and other temperature and heat dissipation parameters, and (undervoltage threshold is set to 80% of rated voltage, overvoltage threshold is set to 120%) overvoltage / undervoltage protection, (no-load vibration ≤0.5mm / s 2 , full-load noise ≤ 55dB(A), high-frequency harmonics must be suppressed to below -30dBc) and other protection threshold parameters for vibration and noise limits. If the quality does not meet the standards (detailed testing standards are provided in subsequent step S70), the process proceeds to step S56 to identify the abnormality category, readjust the parameters, and improve the operating control parameters. If the quality meets the standards, the motor operating control parameters have passed the motor operation test to confirm the stability of its operating quality.

[0141] S56. The abnormality recognition unit matches the real-time acquired brushless motor operation control image with the corresponding image in the trained convolutional neural network model and confirms its abnormality category in order to adjust the operation test parameters.

[0142] It is further explained that according to the program command, the image is input into the model and the trained model is used to predict the new monitoring image, and the type of abnormality in the operation control quality of the DC brushless motor is judged according to the output result. If the model's judgment on the abnormality in the operation control quality of the brushless motor reaches a certain threshold, the early warning system is triggered and corresponding solutions are taken in time; the rectification measures for the abnormal operation control quality of the DC brushless motor include: 1) Dynamic response abnormality: Reinforcement learning is used to optimize the PID parameter weight distribution, reduce the overshoot to ≤3%, and upgrade the edge computing unit hardware to shorten the control cycle to ≤1ms; 2) Control signal misjudgment: Deploy a multi-source data fusion algorithm, combine the back electromotive force zero crossing point to correct the commutation phase deviation to ±1°, and optimize the ADC sampling circuit (add a second-order Butterworth filter), and the signal-to-noise ratio is improved to >60dB; 3) Edge computing delay: Upgrade the edge computing unit to Jetson AGX On the Orin platform, computing power is increased to 275TOPS, and model quantization compression (INT8 precision) is implemented, reducing the model size by 70%; 4) Insufficient model generalization: Transfer learning technology is used to reuse the underlying feature extraction layer of the pre-trained model, and adversarial samples are injected to enhance training, improving the model robustness to >99%; When the above problems occur in the brushless motor operation test, the corresponding corrective measures are taken, and the above step S56 is returned to re-debugging to ensure the accuracy of the brushless motor operation control quality to meet the motor requirements.

[0143] S60, the operation application module controls the operation of the drive device and performs quality inspection according to the set DC brushless motor operation control parameters to ensure the accuracy and stability of the motor's long-term operation control in the future, and transmits the results to the processing center;

[0144] See also Figure 14 As shown, the step S60 includes the following steps:

[0145] S61, the parameter configuration unit configures corresponding operation control parameters according to the load torque and moment of inertia input model of the driven device and identifies the initial position of the rotor to achieve reliable initialization of the motor control system and transmits it to the commutation control unit;

[0146] Further explanation: after obtaining the matching data corresponding to the driven device according to the motor startup program command, the motor hardware self-test is first performed. First, the sensor status is detected. The Hall sensor / back electromotive force detection circuit is used to detect whether the Hall signal level is stable after power-on to verify the phase alignment, and check whether the back electromotive force comparator threshold voltage triggers a valid zero-crossing signal. The power circuit integrity is verified by injecting low-voltage pulses to test the on-resistance of the three-phase full-bridge MOS tube to detect whether there is an open circuit / short circuit fault, and measure the bus voltage ripple to verify the effectiveness of the DC-DC power supply filter circuit. In addition, the signal link noise is tested. Acoustic test, use an oscilloscope to detect the signal-to-noise ratio of the current sampling signal, activate the second-order Butterworth filter algorithm when it exceeds the threshold, and verify the rising / falling edge slope of the PWM signal to ensure that the switching loss is controllable; then parameter loading and configuration, first the non-volatile storage parameter loading, read the preset parameters from the EEPROM, set the PWM frequency within 10-20kHz to control the switching loss and electromagnetic noise, set the rated current × 1.3 for the current limit, and the overcurrent protection threshold, and the PID parameters are dynamically adjusted according to the motor inertia, with closed-loop control stability adjustment, and dynamic parameter calibration, first the phase current Resistance / inductance measurement, using the bridge method to measure phase resistance and phase inductance, automatically correcting the current loop model, and comparing the stored parameters with the actual measured values. When the difference is greater than 10%, an alarm is triggered. Then there is the pole pair verification, by manually rotating the motor to detect the number of Hall signal jumps, or analyzing the number of back electromotive force waveform cycles to match the preset pole pair number; then the initial position identification is performed, first in the inductive control mode, injecting 6 groups of low-voltage pulses, determining the rotor static position by the current response difference, and using the minimum current peak principle to select the initial commutation phase to avoid startup shock, and in the inductive control mode, directly reading the Hall signal combination to match the phase of the six-step commutation method. The software interpolation algorithm is used to reconstruct the position information if the Hall signal is abnormal. Finally, the abnormality processing is performed. If no valid jump is detected for three consecutive PWM cycles for sensor signal loss, the system switches to the sensorless control mode and activates the fault-tolerant algorithm. If the parameter verification fails, the deviation between the measured value and the stored value is greater than 15%, and the startup is rejected and the fault indicator light is turned on. If the initial position identification timeout occurs, the positioning is not completed within 500ms after the pulse injection, and the self-test process is restarted and the injection voltage amplitude is reduced. In this way, the initialization and parameter configuration can be completed within 100ms to ensure the stable startup and operation of the motor control system.

[0147] S62, the commutation control unit realizes real-time position detection and commutation control through position signal processing, commutation logic execution, logic verification and debugging, parameter solidification and storage to meet the requirements of high dynamic load scenarios, and transmits it to the closed-loop regulation unit;

[0148] Further explanation: according to the program commands after system initialization and parameter configuration, the position signal processing is performed first, starting with the Hall signal analysis. The current sector position of the rotor is determined by the six combination signals output by the three Hall elements. Each sector corresponds to a 60° electrical angle, and hardware filtering is used to eliminate Hall signal jitter. Signal loss or abnormality is compensated by software interpolation, and non-sensing mode position detection is performed. First, the back electromotive force zero crossing detection is performed, and the terminal voltage of the non-conducting phase is compared with the virtual neutral point voltage. The zero crossing point is captured as the commutation trigger signal, and then the software filtering algorithm is used to perform sliding average filtering on the back electromotive force signal to suppress high Frequency noise interference, in addition, signal calibration and compensation are performed, first the Hall sensor phase offset calibration, by injecting test current to measure the deviation between the actual commutation point and the theoretical value, correcting the phase compensation table, then temperature drift compensation, dynamically adjusting the back electromotive force comparator threshold according to the NTC sensor data; then the commutation logic is executed, first realizing six-step commutation, triggering a commutation every 60° electrical angle, switching the conduction phase sequence through the inverter full-bridge circuit, and dynamic commutation optimization, first selecting the PWM modulation mode, the dual-chop mode PWM modulation of the upper and lower bridge arms at the same time to reduce switching losses, the single-chop mode only PWM modulation of the upper bridge arm, simplifying Control logic, then commutation advance angle compensation, dynamically adjust the commutation trigger time according to the speed to offset the MOS tube shutdown delay, and implement a fault tolerance mechanism. First, the Hall signal loss is processed. When no valid signal is detected for three consecutive PWM cycles, it switches to the back electromotive force zero-crossing detection mode, and then restarts when the commutation fails. When abnormal current fluctuations are detected, all MOS tubes are immediately turned off, and the commutation logic is reinitialized after a delay of 10ms; then logic verification and debugging, first the commutation accuracy test, verify the commutation angle error under no-load conditions, compare the Hall signal with the phase current phase through an oscilloscope, and monitor the commutation stability during the load mutation test. Qualitative analysis ensures that the speed fluctuation is less than ±2% and there is no torque pulsation, as well as dynamic parameter adjustment. The commutation frequency is adjusted according to the load inertia ratio. When the inertia ratio is less than 3, the commutation frequency is fixed, and when the inertia ratio is ≥3, the adaptive commutation frequency is enabled. Finally, the parameters are solidified and stored. The commutation phase compensation table is stored in EEPROM for call when the Hall signal is abnormal, the back electromotive force threshold curve is stored in FLASH for commutation triggering in the sensorless control mode, and the PWM duty cycle mapping table is stored in non-volatile RAM for real-time call of dynamic commutation optimization, thereby achieving 0.1ms position signal response and ±1° electrical angle commutation accuracy.

[0149] S63, the closed-loop regulation unit realizes PWM modulation and closed-loop regulation through PWM modulation, closed-loop regulation, and coordinated control to meet the high performance requirements of industrial equipment and precision instruments, and transmits the results to the parameter optimization unit;

[0150] Further explanation: According to the program command after real-time position detection and commutation control, PWM modulation is first performed. First, the modulation mode is selected. If it is square wave modulation, the conduction phase sequence is switched based on the Hall signal, and commutation is performed every 60° electrical angle. The H-PWM / L-PWM dual-chop mode is used to reduce switching losses, and the PWM duty cycle transition algorithm is used to reduce torque pulsation and noise during commutation. If it is sinusoidal wave modulation, the Hall signal is first interpolated and expanded into continuous angle information to generate a sinusoidal PWM waveform to reduce torque fluctuations. The sinusoidal voltage signal is converted into a PWM pulse sequence through the equivalent area principle, and the duty cycle and frequency are set. The duty cycle is generated first, and the current loop outputs the duty cycle instruction, and the equivalent full-bridge circuit is used to achieve Voltage regulation, and dynamically limit the duty cycle change rate to avoid MOS tube overvoltage caused by current mutation, and perform frequency optimization, select PWM frequency according to motor inductance parameters, balance switching loss and electromagnetic noise, and add RC absorption circuit for high-frequency application to suppress voltage spikes, and configure dead time, set the delay time for switching between upper and lower bridge arms to avoid direct short circuit, and dynamically adjust the dead time, extend it at low speed, shorten it at high speed to reduce switching loss; then closed-loop regulation, first speed loop control, perform error calculation, measure the actual speed through photoelectric encoder or Hall signal, compare it with the target value to generate error signal, the error signal is output as current reference value through PID operation, and anti-saturation processing is carried out, and integral separation algorithm is implemented. When the speed error When the difference is greater than 10%, the integral term is suspended to prevent overshoot, and the speed filter is performed, and the feedback signal is subjected to sliding average filtering to suppress jitter, as well as current loop control. First, current sampling and feedback are performed. A low-side current sampling resistor is used, and the phase current is obtained through a differential amplifier. The real-time monitoring current peak is triggered when it exceeds 1.3 times the rated value. Overcurrent protection is triggered and closed-loop regulation logic is implemented. The current loop response time is less than 100μs. The duty cycle is adjusted through the PI controller, and feedforward compensation is used. The current change trend is predicted according to the load mutation, and the duty cycle is adjusted in advance. In addition, dynamic parameter adjustment is performed. First, temperature compensation is performed. The current limit is corrected according to the NTC sensor data. The PWM frequency is automatically reduced at high temperatures to reduce switching losses. Then load adaptation and identification are implemented. Load inertia ratio, dynamically adjust the speed loop bandwidth, reduce the bandwidth to 50Hz when the inertia ratio is greater than 3, improve stability, and implement weak magnetic control. When the speed reaches 80% of the base speed, reduce the back electromotive force compensation voltage to expand the speed regulation range; then perform collaborative control, when the initial position is uncertain, open loop start is performed, 6 groups of low-voltage pulses are injected to position the rotor, and the fixed duty cycle is used to achieve soft start. When steady-state speed regulation / dynamic load is required, closed loop operation is performed, dual-loop collaborative control, and the duty cycle and commutation phase are adjusted in real time. When the overcurrent / overtemperature protection is triggered, the fault is recovered, and after restarting, derating operation is adopted to gradually recover to the target speed. In this way, the speed control accuracy of ±1% and the dynamic response of 10ms can be achieved, meeting the high performance requirements of industrial equipment and precision instruments.

[0151] S64, the parameter optimization unit realizes dynamic parameter optimization and protection through real-time parameter monitoring and adjustment, adaptive control, efficiency optimization, protection mechanism implementation, and fault recovery to ensure the success rate of fault protection response and transmit it to the quality inspection unit;

[0152] Further explanation: according to the program command after PWM modulation and closed-loop regulation, first of all, real-time parameter monitoring and adjustment are carried out, the real-time speed is detected by Hall sensor or back electromotive force, and the PID parameters are dynamically adjusted in combination with the current sampling data to achieve fast response and steady-state accuracy of speed closed-loop control, and the fuzzy logic algorithm is used to automatically correct the current loop proportional gain according to the load mutation amplitude to suppress torque pulsation; secondly, adaptive control is carried out, first the weak magnetic control optimization, when the motor speed exceeds 80% of the base speed, the back electromotive force compensation voltage is gradually reduced, the speed regulation range is expanded to 150% of the rated speed, and the commutation Phase compensation dynamically adjusts the phase offset of the Hall signal according to the temperature sensor data to ensure commutation accuracy in high and low temperature environments; then efficiency optimization is carried out, and the PWM duty cycle is adjusted in real time according to the bus voltage fluctuation to keep the copper loss and iron loss ratio in the range of 1:1.2~1.5, thereby improving the overall efficiency of the system and automatically switching to the sine wave modulation mode under light load conditions to reduce electromagnetic noise and core loss; then a protection mechanism is implemented, first of all, overcurrent protection, using dual threshold detection, implementing instantaneous overcurrent triggering hardware protection circuit, shutting down all MOS tubes within 5μs, and continuously overloaded, reducing PWM duty cycle The ratio is increased to 50% to activate the forced cooling mode, and then the temperature protection is implemented, and the graded temperature control protection is implemented. When the temperature is 70-85℃, the power is limited. When the temperature is 85-100℃, it switches to the frequency reduction mode. When it is greater than 100℃, it stops immediately and locks the fault code. As for the voltage abnormality protection, the bus voltage detection is implemented. When the undervoltage protection is activated, the motor is prohibited from starting and the alarm indicator is lit. When the overvoltage protection is activated, the braking resistor is activated to consume excess energy. In addition, short circuit and stall protection are performed, and the current slope detection method is used. When the phase current rise rate is greater than 10A / ms, it is determined to be a short circuit fault and cut off within 3ms. When the speed remains below 5% of the rated value and the current exceeds 80% of the rated value for more than 2 seconds, a stall detection is performed and a reverse pulse is applied to release the mechanical jam. Finally, fault recovery begins with a self-recovery mechanism. Once the overtemperature fault is resolved, output power is gradually restored at a rate of 10% / min to prevent secondary surges. When the voltage anomaly is restored, the bus capacitor is pre-charged three times before restarting the motor. Parameter learning is also implemented, recording historical fault data for optimizing protection thresholds. Each power-up automatically verifies the stored parameters against the measured values. If the deviation exceeds ±5%, the factory default values ​​are applied. This ensures a speed control accuracy of ±0.8% within an ambient temperature range of -25°C to 85°C and a 99.9% fault protection response success rate.

[0153] S65. The quality inspection unit uses motor operation inspection equipment to detect the static parameters, dynamic performance, environmental adaptability, and protection mechanism of the motor to ensure that the motor operation quality meets the standards and is stable.

[0154] Further explanation: after the motor starts the program command, the static parameter detection is carried out first. The winding parameters are measured first. The resistance and inductance of the three-phase winding are measured by the automatic bridge device. The resistance deviation is ≤±5%, and the inductance difference is ≤3%. The integrated megohmmeter is used to automatically detect the insulation resistance. When an abnormality occurs, an alarm signal is triggered. The Hall sensor is automatically calibrated. The oscilloscope is used to link with the servo motor to automatically rotate the rotor and verify the Hall signal phase difference; then the dynamic performance is automatically tested. First, the no-load / load response test is performed, the no-load start is automatically performed, the starting current peak and speed stability are monitored, and a 50% step load is simulated by the electric load table to record the speed recovery time and torque fluctuation, as well as the commutation logic. The high-speed data acquisition card is used to synchronously capture the MOS drive signal and the Hall phase to verify that the commutation delay is ≤5μs, and the d is automatically detected in the FOC mode. Zero deviation of shaft current id; then automatic evaluation of environmental adaptability, first automatic temperature rise monitoring, using infrared thermal imager and NTC sensor linkage, real-time recording of winding temperature rise after 2 hours of full load operation, and using temperature control box to simulate high temperature environment, automatic detection of torque attenuation rate, and automatic analysis of vibration and noise, using three-axis accelerometer to automatically collect vibration data, radial vibration amplitude ≤2.8mm / s, and using sound level meter and microphone array to automatically evaluate noise at a distance of 1 meter; finally, automatic verification of protection mechanism, first fault simulation and protection triggering, automatic injection of 120% overcurrent signal, verification of hardware protection circuit response time ≤1μs, and simulation of single-phase circuit breaker condition, detection of remaining two-phase output torque ≥70% rated value, and fault-tolerant control switching test, using Hall signal simulator to automatically disconnect, detection of back electromotive force zero crossing detection switching time ≤10ms.

[0155] S70. The processing center compares the motor operation quality information with the DC brushless motor operation quality standard stored in the memory: if the standard is met, the system notifies the user that the motor can continue to operate; if the standard is not met, the system transmits the information to an alarm and notifies the user to repair or replace the motor.

[0156] Further explanation: the operation quality standards of the brushless DC motor include: 1) static parameters: the three-phase winding resistance deviation is ≤±5%, the inductance difference is ≤3%, and the insulation resistance is ≥100MΩ detected by the bridge method or multimeter to ensure that there is no leakage risk; the Hall sensor phase error is ≤±5° ​​and the back electromotive force zero-crossing detection delay compensation is ≤30° electrical angle by oscilloscope calibration; 2) dynamic performance: no-load speed deviation is ≤±2% rated value, recovery time is ≤50ms when the load suddenly changes, and the torque fluctuation rate under rated load is verified by the torque sensor to be ≤±3%; the six-step commutation logic commutation delay is ≤5μs, FOC Under control, the d-axis current deviation is ≤2%, and the weak magnetic control extends the speed to 150% of the rated value, and the efficiency attenuation must be maintained at ≤10%; 3) Materials and processes: The stator core is made of high-purity silicon steel sheets, the permanent magnets are made of neodymium iron boron rare earth materials, the winding copper wire purity is ≥99.9%, and the solder joints have no cold solder joints or oxidation marks; the axial / radial swing is ≤0.05mm, the shell has no deformation or cracks, and the bearing vibration amplitude is ≤2.8mm / s; 4) Environmental adaptability standards: Continuous full-load operation for 2 hours, the winding temperature rise is ≤65K, and the torque attenuation is ≤10% in a high temperature environment (+85℃); electromagnetic compatibility complies with GB / T 17626 standard, with voltage fluctuation resistance of ±15%, and insulation resistance ≥50MΩ in a hot and humid environment; 5) Protection mechanism standard: overcurrent protection response time ≤1μs, overvoltage / undervoltage protection trigger threshold error ≤±5%, when the Hall signal fails, the back electromotive force detection switching time ≤10ms; when a single-phase circuit is broken, the remaining two-phase output torque ≥70% of the rated value, after the stall protection is triggered, the motor automatically reduces power and the winding temperature ≤130℃; 6) Life and reliability: 72 hours of continuous operation at full load, performance degradation The following requirements must be met: 1) a reduction of ≤2%; 2) insulation resistance ≥50MΩ after 100 high and low temperature cycles; 3) mean time between failures ≥50,000 hours as verified by an accelerated life test; 4) control system: speed loop control accuracy ≤±1%, position loop positioning error ≤±0.1°, PID parameter auto-tuning overshoot ≤3%, and adjustment time ≤50ms; 5) support for CAN / RS485 communication protocols, data transmission bit error rate ≤0.001%, and parameter storage EEPROM read / write cycles ≥1 million times, with data retention ≥10 years. If the operating quality meets the requirements, the brushless DC motor can meet the requirements for efficient, stable, and long-life operation in industries such as industry, automobiles, and household appliances, with an overall pass rate ≥99.5%. If the requirements are not met, the process returns to step S60 for repair.

[0157] It is further explained that the above steps are displayed in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this document, there is no strict order restriction for the execution of these steps. These steps can also be executed in other orders, and some steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0158] Further explanation: The present invention is described according to the content implemented by the software program of the control system of the intelligent brushless motor. The control method for each intelligent brushless motor is divided into several modules or units to implement the software program instructions generated by each step. The software program instructions include the control method of the intelligent brushless motor described above.

[0159] The system described in the present invention also includes a computer-readable storage medium containing a memory; the memory stores a computer program, and when the functional modules execute the computer program, the steps of the control method of an intelligent brushless motor described in any one of the above are implemented; the computer-readable storage medium stores a computer program, and when the functional modules execute the computer program, the steps of the control method of an intelligent brushless motor described in any one of the above are implemented; and also includes an intelligent brushless motor control device, which is implemented using the control method of an intelligent brushless motor described in the above.

[0160] It is further explained that the computer-readable storage medium includes a memory for storing non-volatile software programs, non-volatile computer-executable programs and modules, such as the programs and instructions corresponding to the control method of an intelligent brushless motor in the present invention, including information transmission instructions of each module; the memory executes various functional applications and data processing of each module by running the stored non-volatile software programs and instructions, thereby realizing the control method of an intelligent brushless motor in the above embodiment; one or more units are stored in the memory, and when executed by the one or more modules, a control method of an intelligent brushless motor in any of the above embodiments is executed; the computer-readable storage medium stores computer-executable instructions, which are executed by one or more modules, and can also be a control method of an intelligent brushless motor in any of the above process embodiments.

[0161] It is further explained that the computer program is divided into multiple modules / units, which are stored in the memory and executed by each module / unit to complete the present invention; the multiple modules / units can be a series of computer program instructions that can perform specific functions, and the instructions are used to describe the execution process of the computer program in each module / unit.

[0162] It is further explained that the control device is composed of the above-mentioned associated monitoring equipment or devices. According to the needs of motorless application production, it can be welded or fixed together with the DC brushless motor, or it can be made into a unit device of each functional module belonging to each step or the entire control device. During installation, each control device is connected to the brushless motor production equipment through a wireless connection; the structures of these devices are not described in detail here.

[0163] The present invention also provides an intelligent brushless motor, which is realized by using the above-mentioned control method for an intelligent brushless motor.

[0164] It is obvious to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, from all perspectives, the embodiments should be considered as illustrative and non-restrictive. The scope of the present application is defined by the appended claims rather than the foregoing description, and it is intended that all changes that fall within the meaning and range of equivalents of the claims be encompassed within the present application.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents, and all of these should be included in the scope of protection of the present application.

Claims

1. A control method for an intelligent brushless motor, characterized in that : The method comprises the following steps: S10, after the motor is started, the data acquisition module obtains the motor's operating and state parameters, dynamic characteristics, environmental condition data, adversarial generation data, and data preprocessing for subsequent model construction, and passes it to the model construction module; S20, the model building module performs key feature extraction, model design, integrated optimization, parameter initialization, dynamic optimization, model fusion and deployment based on the verified data to build a model for subsequent model training and verification, and passes it to the training and tuning module; S30, the training and tuning module trains the model according to the extracted data and adjusts the model parameters to optimize the model performance and predetermine the DC brushless motor operation control parameters to ensure the accuracy of the motor operation control parameters, and transmits them to the model verification module; S40, the model verification module performs model verification and evaluation based on the trained model, and adjusts and optimizes the model based on the verification results to improve the performance and accuracy of the DC brushless motor operation model, and transmits the results to the parameter setting module; S50, the parameter setting module performs a motor operation control test based on the brushless motor control parameters predicted by the trained model to obtain the optimal DC brushless motor operation control parameters to ensure the accuracy of future operation control, and transmits them to the operation application module; S60, the operation application module controls the operation of the drive device and performs quality inspection according to the set DC brushless motor operation control parameters to ensure the accuracy and stability of the motor's long-term operation control in the future, and transmits the results to the processing center; S70. The processing center compares the motor operation quality information with the DC brushless motor operation quality standard stored in the memory: if the standard is met, the system notifies the user that the motor can continue to operate; if the standard is not met, the system transmits the information to an alarm and notifies the user to repair or replace the motor.

2. The control method of an intelligent brushless motor according to claim 1 is characterized in that The method is applied to a control system of an intelligent brushless motor, the system comprising a data acquisition module, a model construction module, a training and tuning module, a model verification module, a parameter setting module, an operation and application module, a wireless communication module, a memory, an alarm, a processing center, and an intelligent mobile terminal; the data acquisition module, the model construction module, the training and tuning module, the model verification module, the parameter setting module, the operation and application module, the wireless communication module, the memory, and the alarm are respectively connected to the processing center; the intelligent mobile terminal comprises a smart phone, a tablet computer, and an intelligent remote control, which are respectively connected to the wireless communication module within a wireless network or the Internet; the wireless communication module is provided with a wireless network unit, which is responsible for sending and receiving wireless signals, and is automatically networked with the intelligent mobile terminal and various drive devices within an effective network range; The alarm device compares the actual evaluation score of the motor model with the evaluation score standard of the motor model stored in the memory, and automatically sounds an alarm and notifies to continue training if the standard is not met; and compares the quality information of the motor operation with the operation quality standard of the brushless motor stored in the memory, and automatically sounds an alarm and notifies to adjust the parameters if the standard is not met; and compares the motor operation quality information with the operation quality standard of the brushless DC motor stored in the memory, and automatically sounds an alarm and notifies to repair or replace the motor if the standard is not met; The memory is responsible for storing information of the data acquisition module, model construction module, training and tuning module, model verification module, parameter setting module, operation application module, wireless communication module, and alarm, as well as the model evaluation criteria and the brushless motor control quality criteria; The processing center is responsible for information transmission among the functional modules, wireless communication modules, alarms, and memories, and serves as the hub of the system. The processing center compares the actual evaluation score of the motor model with the evaluation score standard of the motor model stored in the memory: if the standard is met, the predetermined DC brushless motor operation control parameters are set; if the standard is not met, the parameters are transmitted to the alarm and a notification is given to continue training; the processing center also compares the motor operation quality information with the brushless motor operation quality standard stored in the memory: if the standard is met, the parameters are set as the DC brushless motor operation control parameters; if the standard is not met, the parameters are transmitted to the alarm and a notification is given to adjust the parameters; the processing center also compares the motor operation quality information with the brushless motor operation quality standard stored in the memory: if the standard is met, the notification is given that the motor can continue to operate; if the standard is not met, the notification is given to the alarm and a notification is given to repair or replacement; The data acquisition module includes an information acquisition unit, a noise filtering unit, a data cleaning unit, a format conversion unit, a static normalization unit, a dynamic standard unit, a data enhancement unit, and a data encoding unit. It is responsible for acquiring the motor's operating and state parameters, dynamic characteristics, environmental working condition data, adversarial generation data, and data preprocessing, and passing them to the model construction module; The model building module includes a feature extraction unit, a voltage modeling unit, a power modeling unit, an electromagnetic torque unit, a coordinate transformation unit, a dynamic equation unit, a model standard unit, a discrete processing unit, and a noise modeling unit. It performs key feature extraction, model design, integrated optimization, parameter initialization, dynamic optimization, model fusion and deployment based on the verified data to build a model, and then passes it to the training and tuning module; The training and tuning module includes a hardware acceleration unit, a supervised learning unit, a loss function unit, a reinforcement learning unit, and a parameter optimization unit. The model is trained based on the extracted data and the model parameters are adjusted to optimize the model performance and preset the DC brushless motor operation control parameters, which are then passed to the model verification module. The model verification module includes a classification marking unit, a learning update unit, a cost verification unit, and a training evaluation unit. It performs model verification and evaluation based on the trained model and adjusts and optimizes the model based on the verification results, and passes them to the parameter setting module. The parameter setting module includes a parameter optimization unit, an energy efficiency optimization unit, a back electromotive force unit, an operation test unit, and an abnormality identification unit. The motor operation control test is performed based on the brushless motor control parameters predicted by the trained model to obtain the optimal DC brushless motor operation control parameters, and the parameters are passed to the operation application module. The operation application module includes a parameter configuration unit, a commutation control unit, a closed-loop adjustment unit, a parameter optimization unit, and a quality detection unit. It controls the operation of the drive device and performs quality detection according to the set DC brushless motor operation control parameters, and transmits the results to the processing center.

3. The control method of an intelligent brushless motor according to claim 1 is characterized in that : Described step S10 comprises the following steps: S11. The information acquisition unit synchronously collects motor operation data such as current and voltage, rotor position and speed, temperature, vibration and electromagnetic characteristics, and surrounding environment data through the acquisition device to ensure the consistency of timing, and transmits the data to the noise filtering unit; S12. The noise filtering unit obtains the filtered motor data according to the sliding window mean filtering calculation formula "y(n) = y(n-1) + [x(n) - x(nM)] / M, where y(n) is the filtered output value at the current moment, y(n-1) is the filtered output value at the previous moment, x(n) is the original input value at the current moment, x(nM) is the input value at the oldest moment of the window, and M is the sliding window length (n≥M)" to filter out high-frequency noise and pass it to the data cleaning unit; S13, the data cleaning unit performs data missing location, missing value filling, outlier correction, and duplicate value removal on the filtered data to obtain cleaned motor data to ensure the accuracy of the model input data, and transmits it to the format conversion unit; S14, the format conversion unit unifies the cleaning data into the international unit system according to the set formula for current, voltage, speed, temperature, time, torque and power, and torque coefficient to ensure the accuracy of subsequent data calculations, and transmits it to the static normalization unit; S15. The static normalization unit obtains static numerical normalized data according to the static numerical normalization calculation formula "Z = [X-∑Xi / No] / sqrt[∑(Xi-μ)2 / No], where Z is the static normalized value, X is the static original data point, No is the number of dynamic samples, Xi is the variable in the i-th time period, and μ is the mean value of the variable" to eliminate dimensional differences and pass it to the dynamic standard unit; S16. The dynamic standard unit obtains dynamic numerical normalized data according to the dynamic numerical normalization calculation formula "Zn=[Xn-(μo+(Xn-Xo) / Ws)] / sqrt[σo2+((Xn-μn)2-(Xo-μo)2) / Ws], where Zn is the dynamic normalization value, X is the new data point, μo is the old mean, Ws is the window size, σo is the old standard deviation, Xo is the replaced old data point, and μn is the new mean" to eliminate magnitude differences and pass it to the data enhancement unit; S17, the data enhancement unit performs noise injection, time series transformation, working condition simulation, and physical model driving on the standardized data to enhance data performance, thereby improving dynamic response speed and fault diagnosis capabilities, and transmits the data to the data encoding unit; S18. The data encoding unit converts the standardized overcurrent, overtemperature, overvoltage / short circuit and other fault types into one-hot encoding and constructs a multi-label classification input, and performs label encoding on the control mode to provide structured input for the fault diagnosis algorithm.

4. The control method of an intelligent brushless motor according to claim 1 is characterized in that : The step S20 comprises the following steps: S21, the feature extraction unit obtains the key features of the motor's electrical parameters, mechanical dynamics, control signals, temperature and fault, time-frequency domain, and dimensionality reduction fusion based on the preprocessed data to improve the model training speed, and transmits them to the voltage modeling unit; S22. The voltage modeling unit obtains a circuit model based on the brushless DC motor equivalent input voltage calculation formula: "Ue = Ro*i + Li*di / dt + E, where Ue is the equivalent input voltage (V), Ro is the equivalent internal resistance (Ω), i is the phase current (A), Li is the armature inductance (H), di / dt is the time derivative of the phase current (A / s), and E is the back electromotive force (V)." This provides core mathematical model support for the control algorithm and is passed to the power modeling unit. S23, power modeling unit calculates the motor angular acceleration according to the formula "α=[(Kt*iq)-Po / ω-(U*io-io2*Ri) / ω] / (∑m*r 2 ), α is the angular acceleration (rad / s 2 ), Kt is the motor torque coefficient (N·m / A), iq is the q-axis phase current (A), Po is the motor output power (W), ω is the angular velocity (rad / s), U is the no-load input voltage (V), io is the no-load input current (A), Ri is the winding resistance (Ω), m is the motor rotor mass (kg), and r is the motor rotor radius (m). The angular acceleration is obtained and a motion model is established to facilitate motor acceleration performance analysis and control, and is then transmitted to the electromagnetic torque unit; S24, electromagnetic torque unit is calculated according to the electromagnetic torque formula of brushless DC motor "Te=J*α+Bi*ω+T L , Te is the electromagnetic torque (N·m), J is the moment of inertia (kg·m 2 ), α is the angular acceleration (rad / s 2 ), Bi is the friction coefficient (N·m·s / rad), ω is the angular velocity (rad / s), T L The dynamic torque balance equation of the motor is obtained for the load torque (N·m) to predict the torque-speed response characteristics during sudden load changes or speed regulation, and is passed to the coordinate transformation unit; S25, coordinate transformation unit according to the DC brushless motor Park transformation formula "id=2 / 3[ia*cosθ+ib*cos(θ-120°)+ic*cos(θ+120°)],iq=2 / 3[-ia*sinθ-ib*sin(θ-120°)-ic*sin(θ+120°)],id,iq are the direct-axis and quadrature-axis currents (A),ia,ib,ic are the three-phase stator currents (A), andθ is the rotor electrical angle (rad). The two-phase stationary coordinate system (dq) is obtained to achieve decoupling control of magnetic field and torque, and is transferred to the dynamic equation unit. S26. The dynamic equation unit obtains the dynamic equation construction based on the DC brushless motor voltage balance equation "Ld*did / dt=Ud-Ra*id+ωe*Lq*iq, Lq*diq / dt=Uq-Ra*iq-ωe(Ld*id+ψf), Ld and Lq are the direct-axis and quadrature-axis inductances (H), Ra is the armature resistance (Ω), id and iq are the direct-axis and quadrature-axis currents (A), Ud and Uq are the direct-axis and quadrature-axis voltages (V), ωe is the electrical angular velocity, p is the number of motor pole pairs, and ψf is the permanent magnet flux (Wb)" to facilitate the subsequent state space model standardization and transfer to the model standard unit; S27, the model standard unit converts the nonlinear differential equation into the following matrix A, B standard state space form x' = Ax + Bu to improve control robustness and subsequent discretization processing, and passes it to the discrete processing unit; S28, discrete processing unit according to the DC brushless motor zero-order holder formula x(k+1) is the next state vector, e AT is the state transfer matrix (Φ), x(k) is the current state vector, is the input matrix integral term (Γ), B is the continuous input matrix, dτ is the time element, and u(k) is the input vector. The discretized equation is obtained for subsequent noise modeling and parameter expansion, and is passed to the noise modeling unit. S29. The noise modeling unit obtains the noise model according to the DC brushless motor noise modeling calculation formula "x(k+1)=Φ*x(k)+Γ*u(k)+w(k), y(k)=C*x(k)+v(k), x(k+1) is the next discrete time state vector, Φ is the state transfer matrix, x(k) is the state variable vector, Γ is the input matrix, u(k) is the control input vector, w(k) is the process noise, y(k) is the output of the observation equation, C is the observation matrix, and v(k) is the observation noise" to enhance the model robustness.

5. The control method of an intelligent brushless motor according to claim 1 is characterized in that : The step S30 comprises the following steps: S31, the hardware acceleration unit accelerates large-scale data training based on the model through hardware selection and configuration, data parallel processing optimization, edge-cloud collaborative architecture, performance tuning and energy efficiency management to shorten the model training cycle and pass it to the supervised learning unit; S32, the supervised learning unit pre-trains the model using simulation data by physically guiding the neural network architecture, adopting a hybrid architecture, Xavier / He initialization, and transfer learning according to the model, so as to initialize the network weights and pass them to the loss function unit; S33, loss function unit calculates the loss function of the brushless DC motor according to the formula "Lt=∑(yi-y'i)2 / N+λ1*∑max(0,|Ii|-Imax)+λ2*∑∥ea(t)-Ke*ω(t)∥2+λ3*∑∥y' t+1 -y't∥2+λ4*E x～D [∥xe*L(f(x),y)∥2]+∑(σt -2 *Lt+logσt) / To, Lt is the motor loss function, yi is the actual true value, y'i is the model prediction value, N is the number of samples, λ1 is the penalty weight coefficient, Ii is the actual phase current, Imax is the maximum allowable current of the motor, λ2 is the balance weight, ea(t) is the actual back electromotive force, Ke is the back electromotive force coefficient, ω(t) is the speed at time t, λ3 is the smoothness weight, y' t+1 is the model output at time t+1, y't is the model output at time t, λ4 is the adversarial training intensity coefficient, E x～D is the expected value of the input data, xe is the input data gradient, L(f(x), y) is the basic loss function between the model output and the true value, σt is the prediction variance at time t, Lt is the basic loss term, and To is the total step length of the time series. The loss function value is obtained to achieve high-precision and high-robustness DC brushless motor control and is passed to the reinforcement learning unit; S34. The reinforcement learning unit obtains the reward function value and conducts reinforcement learning based on the DC brushless motor reward function calculation formula "fr=ε*(1-|ωr-ωo| / 0.1ωr)+β*(1-Pl / Pr)-γ-*∑max(0,|Ii|-Imax), where fr is the reward function value, ε is the speed tracking accuracy weight, ωr ​​is the target speed, ωo is the actual speed, β is the energy efficiency optimization weight, Pl is the actual power loss, Pr is the rated power, γ is the safety constraint weight, Ii is the actual phase current, and Imax is the maximum allowable motor current." This is to reduce the overshoot in the speed step response and shorten the steady-state time, and pass it to the parameter optimization unit; S35. The parameter optimization unit obtains the reward function value and strengthens learning based on the DC brushless motor basic acquisition function calculation formula "fg=W1*EI(Ms(ωr,ω))+W2*Ei(Po / Pr)-γ*∑max(0,|Ii|-Imax), where fg is the basic acquisition function value, W1 is the weight coefficient of the speed tracking error term, W2 is the weight coefficient of the energy efficiency optimization term, EI is the expected improvement function of the error term, Ms is the speed error function, ωr is the target speed, ωo is the actual speed, Ei is the energy efficiency term function, Po is the actual power loss, Pr is the rated power, γ is the safety constraint weight coefficient, Ii is the actual phase current, and Imax is the maximum allowable current of the motor" to reduce the overshoot in the speed step response and shorten the steady-state time.

6. The control method of an intelligent brushless motor according to claim 1 is characterized in that : The step S40 comprises the following steps: S41. The classification and labeling unit classifies the data samples into corresponding categories according to the abnormal operating quality categories of the brushless DC motor and labels them for subsequent machine learning of the AI ​​model and transmits them to the learning update unit; S42, the learning update unit determines the training model according to the gradient descent data, and inputs the DC brushless motor operation quality data in the training set into the model for simulation training to ensure the accuracy of model recognition, and passes it to the cost verification unit; S43, the cost verification unit calculates the cost function formula "J(θ)=∑ [1 / (1+σ(xi))]*[y*log(hθ(x))], where J(θ) is the cost function after model training, σ(xi) is the estimated value of the signal-noise variance, y is the true label value, and hθ(x) is the model prediction probability. The cost function value after model training is obtained to improve the noise robustness of the model in industrial scenarios and is passed to the training and evaluation unit. S44. The training evaluation unit calculates the model evaluation score according to the model evaluation score calculation formula "F = 2[(Tp + Tn) / (Tp + Fp + Fn + Tn)] * [Tp / (Tp + Fn)] / [(Tp + Tn) / (Tp + Fp + Fn + Tn)] + [Tp / (Tp + Fn)], where F is the training model evaluation score, Tp is the number of true positive samples, Tn is the number of true negative samples, Fp is the number of actual false positive samples, and Fn is the number of false negative samples" to obtain a comprehensive evaluation score F to ensure the comprehensive performance of the model in motor control and transmit it to the processing center; S45. The processing center compares the actual evaluation score of the motor model with the evaluation score standard of the motor model stored in the memory: if it meets the standard, it is the predetermined DC brushless motor operation control parameter; if it does not meet the standard, it is transmitted to the alarm and notified to continue training.

7. The control method of an intelligent brushless motor according to claim 1 is characterized in that : The step S50 comprises the following steps: S51, parameter optimization unit according to the brushless DC motor PID parameter real-time optimization calculation formula "Kp(t)=Rl(e′,∫ e), Ki(t)=Nn(Eh,Tr), Kd(t)=Ga(de / dt,ωe), Kp(t) is the dynamic proportional coefficient, Rl is reinforcement learning, e′ is the speed error change, ∫e is the error accumulation, Ki(t) is the dynamic integral coefficient, Nn is the neural network, Eh is the historical error sequence, Tr is the torque pulsation rate, Kd(t) is the dynamic differential coefficient, Ga is the genetic algorithm optimized Kd parameter, de / dt is the error differential term, ωe is the speed tracking error" to obtain dynamic adjustment of PID parameters to achieve PID parameter adaptive adjustment and pass it to the energy efficiency optimization unit; S52. The energy efficiency optimization unit obtains the torque pulsation rate and energy efficiency optimization constraints based on the motor torque pulsation suppression calculation formula "Tr = AI (sqrt [∑ (Tk - T')2 / N)] / Tn * 100% ≤ 5%, Tr is the torque pulsation rate, AI is the artificial intelligence algorithm, Tk is the instantaneous torque (N m), T' is the average torque (N m), N is the number of samples, and Tn is the rated average torque (N m)" to improve the overall system efficiency and transmit it to the back electromotive force unit; S53. The back-EMF unit obtains the back-EMF AI compensation and performs field-oriented control according to the DC brushless motor back-EMF AI compensation calculation formula: "Ea′=k*N1*δ*D*L*Bo / a*ω+Ne(ω′,I′), where Ea′ is the estimated back-EMF value after compensation (V), k is the number of armature slots, N1 is the number of winding element turns, δ is the pole arc coefficient, D is the armature outer diameter (m), L is the core length (m), Bo is the air gap flux density (T), a is the number of armature parallel branches, ω is the actual angular velocity of the motor (rad / s), Ne is the neural network compensation term (V), ω′ is the compensated speed (rad / s), and I′ is the compensation current (A)" to balance torque output and energy consumption, and transmits the result to the operation test unit. S54, the operation test unit performs an operation control test on the target drive device according to the DC brushless motor operation control parameters predetermined by the model and obtains motor operation quality information to ensure the accuracy of the motor control parameters, and transmits the information to the processing center; S55. The processing center compares the motor operation quality information with the brushless motor operation quality standard stored in the memory. If the standard is met, the information is set as the brushless DC motor operation control parameter. If the standard is not met, the information is transmitted to the alarm device and the parameter adjustment is notified. S56. The abnormality recognition unit matches the real-time acquired brushless motor operation control image with the corresponding image in the trained convolutional neural network model and confirms its abnormality category in order to adjust the operation test parameters.

8. The control method of an intelligent brushless motor according to claim 1 is characterized in that : The step S60 comprises the following steps: S61, the parameter configuration unit configures corresponding operation control parameters according to the load torque and moment of inertia input model of the driven device and identifies the initial position of the rotor to achieve reliable initialization of the motor control system and transmits it to the commutation control unit; S62, the commutation control unit realizes real-time position detection and commutation control through position signal processing, commutation logic execution, logic verification and debugging, parameter solidification and storage to meet the requirements of high dynamic load scenarios, and transmits it to the closed-loop regulation unit; S63, the closed-loop regulation unit realizes PWM modulation and closed-loop regulation through PWM modulation, closed-loop regulation, and coordinated control to meet the high performance requirements of industrial equipment and precision instruments, and transmits the results to the parameter optimization unit; S64, the parameter optimization unit realizes dynamic parameter optimization and protection through real-time parameter monitoring and adjustment, adaptive control, efficiency optimization, protection mechanism implementation, and fault recovery to ensure the success rate of fault protection response and transmit it to the quality inspection unit; S65. The quality inspection unit uses motor operation inspection equipment to detect the static parameters, dynamic performance, environmental adaptability, and protection mechanism of the motor to ensure that the motor operation quality meets the standards and is stable.

9. A control method for an intelligent brushless motor according to claims 1-8, characterized in that : The system also includes a computer-readable storage medium containing a memory; the memory stores a computer program, and when the functional modules execute the computer program, the steps of the control method of an intelligent brushless motor described in any one of claims 1 to 8 are implemented; the computer-readable storage medium stores a computer program, and when the functional modules execute the computer program, the steps of the control method of an intelligent brushless motor described in any one of claims 1 to 8 are implemented; and also includes an intelligent brushless motor control device, which is implemented by the control method of an intelligent brushless motor described in any one of claims 1 to 8.

10. An intelligent brushless motor, characterized in that : It is realized by adopting the control method of an intelligent brushless motor described in any one of claims 1-8 above.

Images