Brushless motor control system and control method

Abstract

The invention relates to the field of brushless motors, and particularly discloses a brushless motor control system and a brushless motor control method, and the brushless motor control system uses a non-inductive FOC control mode to drive a brushless motor to operate in a no-load manner so as to obtain a first no-load rotor angle. Performing time-frequency analysis on sine wave feedback signals provided by two linear Hall sensors on a brushless motor driving board by adopting a signal processing technology based on deep learning, extracting time-frequency characteristic representation of the signals, and performing fine-grained global interactive coding on the time-frequency characteristics of the two signals; the method comprises the steps of identifying and strengthening key information interaction between signals to realize high-precision prediction of the angle of a second no-load rotor of the brushless motor, calculating an angle compensation value on this basis, and performing inductive FOC control based on the angle compensation value on the motor in the subsequent normal use process of the motor, so that the influence of signal noise and calculation errors can be effectively reduced, and the reliability of the motor is improved. The accuracy of motor rotor position estimation is improved, and the overall control precision of the motor is improved.

Description

Technical Field

[0001] This application relates to the field of brushless motor control, and more specifically, to a brushless motor control system and a control method. Background Art

[0002] Due to its advantages such as high efficiency, fast response, and high power density, brushless motors have been widely used in various application scenarios, such as office automation, smart home, medical devices, industrial automation, etc. To achieve precise control of brushless motors, common driving methods include sensorless FOC (Field Oriented Control) mode and sensor FOC mode.

[0003] The sensorless FOC control mode does not require an additional position sensor. It estimates the rotor position by detecting the back electromotive force of the motor, thereby achieving control of the motor. However, during the startup and low-speed operation stages, the back electromotive force signal of the motor is weak, and the sensorless FOC control mode often has difficulty accurately estimating the rotor position, resulting in insufficient control accuracy. The sensor FOC control mode directly measures the rotor position by using a position sensor, thereby providing more precise control. However, due to the installation error and device error brought by the sensor itself, they will be amplified as the motor speed increases, and the accuracy of the electrical angle obtained may be lower than that of the sensorless position observer at high speeds.

[0004] In this regard, the invention patent with the publication number CN118713532A discloses a DC brushless motor and its control method. When the DC brushless motor is in the factory mode, it uses the sensorless FOC control mode to start the DC brushless motor, obtains the first no-load rotor angle when the brushless motor is running without load, and at the same time calculates the second no-load rotor angle by using the sine wave feedback signals provided by two linear Hall sensors installed on the motor drive board. Thus, during the subsequent normal use of the motor, the angle compensation value is determined according to the difference between the first no-load rotor angle and the second no-load rotor angle, and the sensor FOC control mode is used to drive the brushless motor to run based on the angle compensation value.

[0005] When parsing and calculating the sine wave feedback signals provided by two linear Hall sensors in the above solution, the arctangent method, phase-locked loop method, and frequency doubling conversion technology are mainly used to determine the no-load rotor angle. However, there are certain computational complexities and error accumulation problems in this parsing and calculation method. For example, the arctangent method is very sensitive to the noise of the input signal. If the phase difference between the two sine wave signals is not strictly 90°, it will lead to a large error in the calculated angle. The phase-locked loop (PLL) technology needs to lock the phase of the feedback signal based on the reference signal and is easily affected by the frequency fluctuation or phase noise of the reference signal, thus generating errors. At the same time, if the frequency doubling conversion involves non-integer frequency doubling, it may further introduce rounding errors, resulting in inaccurate angle calculation.

[0006] Therefore, an optimized brushless motor control method and system are needed to solve the above technical problems. Summary of the Invention

[0007] To solve the above technical problems, this application is proposed. Embodiments of this application provide a brushless motor control system and control method. At the factory stage, the sensorless FOC control mode is used to drive the brushless motor to run no-load to obtain the first no-load rotor angle. Then, time-frequency analysis is performed on the sine wave feedback signals provided by two linear Hall sensors on the brushless motor drive board by using signal processing technology based on deep learning, and the time-frequency feature representation of the signals is extracted. By performing fine-grained global interaction encoding on the time-frequency features of the two sine wave feedback signals, the key information interaction between the signals is identified and strengthened, and the high-precision prediction of the second no-load rotor angle of the brushless motor is realized. On this basis, the angle compensation value is calculated, and during the subsequent normal use of the motor, sensor FOC control based on the angle compensation value is performed on it, which can effectively reduce the influence of signal noise and calculation errors, improve the accuracy of motor rotor position estimation, and thus improve the overall control accuracy of the motor.

[0008] According to one aspect of this application, a brushless motor control method is provided, which includes: at the factory stage, using the sensorless FOC control mode to drive the brushless motor to run no-load; obtaining the first no-load rotor angle of the motor rotor; respectively obtaining the first no-load sine wave feedback signal and the second no-load sine wave feedback signal generated by two linear Hall sensors installed on the drive board of the brushless motor; parsing and calculating the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain the second no-load rotor angle of the motor rotor; obtaining the angle compensation value according to the first no-load rotor angle and the second no-load rotor angle; at the start-up stage, based on the angle compensation value, using the sensor FOC control mode to drive the brushless motor to run.

[0009] In the above brushless motor control method, parsing and calculating the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain the second no-load rotor angle of the motor rotor includes: respectively performing time-frequency feature extraction on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a first no-load sine wave feedback signal time-frequency feature map and a second no-load sine wave feedback signal time-frequency feature map; performing global semantic interaction based on core joint feature guidance on the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map to obtain a first-second no-load feedback fine-grained joint semantic significantly fused coding feature map; and determining the second no-load rotor angle of the motor rotor based on the first-second no-load feedback fine-grained joint semantic significantly fused coding feature map.

[0010] Among them, respectively performing time-frequency feature extraction on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a first no-load sine wave feedback signal time-frequency feature map and a second no-load sine wave feedback signal time-frequency feature map includes: respectively performing wavelet analysis on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a first no-load sine wave feedback signal time-frequency diagram and a second no-load sine wave feedback signal time-frequency diagram; using a neural network model to respectively extract the time-frequency features of the first no-load sine wave feedback signal time-frequency diagram and the second no-load sine wave feedback signal time-frequency diagram to obtain the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map.

[0011] Among them, using a neural network model to respectively extract the time-frequency features of the first no-load sine wave feedback signal time-frequency diagram and the second no-load sine wave feedback signal time-frequency diagram to obtain the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map includes: inputting the first no-load sine wave feedback signal time-frequency diagram and the second no-load sine wave feedback signal time-frequency diagram into a signal time-frequency feature extractor based on a dilated convolutional neural network model to obtain the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map.

[0012] Among them, performing global semantic interaction based on core joint feature guidance on the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map to obtain a first-second no-load feedback fine-grained joint semantic significantly fused coding feature map includes: Perform fine-grained common feature extraction on the time-frequency feature maps of the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a set of first-second no-load feedback fine-grained core joint feature vectors; Input the set of first-second no-load feedback fine-grained core joint feature vectors into a clustering network to obtain a first-second no-load feedback joint feature semantic clustering center vector; Based on the first-second no-load feedback joint feature semantic clustering center vector, guide the set of first-second no-load feedback fine-grained core joint feature vectors to perform fine-grained context semantic interaction to obtain the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map.

[0013] Among them, performing fine-grained common feature extraction on the time-frequency feature maps of the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a set of first-second no-load feedback fine-grained core joint feature vectors includes: Perform feature fine-grained decoupling on the time-frequency feature maps of the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a set of time-frequency feature vectors of the first no-load sine wave feedback signal and a set of time-frequency feature vectors of the second no-load sine wave feedback signal; Input each group of corresponding time-frequency feature vectors of the first no-load sine wave feedback signal and the second no-load sine wave feedback signal in the set of time-frequency feature vectors of the first no-load sine wave feedback signal and the set of time-frequency feature vectors of the second no-load sine wave feedback signal into a fine-grained common feature extraction network to obtain the set of first-second no-load feedback fine-grained core joint feature vectors.

[0014] Among them, inputting the set of first-second no-load feedback fine-grained core joint feature vectors into a clustering network to obtain a first-second no-load feedback joint feature semantic clustering center vector includes: Calculate the mean vector of the set of first-second no-load feedback fine-grained core joint feature vectors to obtain the first-second no-load feedback joint feature semantic clustering center vector.

[0015] Among them, based on the first-second no-load feedback joint feature semantic clustering center vector, guiding the set of first-second no-load feedback fine-grained core joint feature vectors to perform fine-grained context semantic interaction to obtain the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map includes: Construct a query vector and a value vector based on each of the first-second no-load feedback fine-grained core joint feature vectors in the set of the first-second no-load feedback fine-grained core joint feature vectors, and construct a key vector based on the first-second no-load feedback joint feature semantic clustering center vector, and input the query vector, the value vector, and the key vector into a global semantic interaction encoder based on a transformer structure to obtain a set of context first-second no-load feedback fine-grained core joint feature vectors; Perform feature shape reshaping on the set of the context first-second no-load feedback fine-grained core joint feature vectors to obtain the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map.

[0016] Among them, determining the second no-load rotor angle of the motor rotor based on the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map includes: Input the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map into a no-load rotor angle estimator based on a decoder to obtain the second no-load rotor angle of the motor rotor.

[0017] According to another aspect of the present application, a brushless motor control system is provided, which includes: a signal receiving module for receiving a first no-load sine wave feedback signal and a second no-load sine wave feedback signal generated by two linear Hall sensors installed on a drive board of a brushless motor; a time-frequency feature extraction module for respectively performing time-frequency feature extraction on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a first no-load sine wave feedback signal time-frequency feature map and a second no-load sine wave feedback signal time-frequency feature map; a semantic interaction fusion module for performing global semantic interaction guided by core joint features on the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map to obtain a first-second no-load feedback fine-grained joint semantic significant fusion coding feature map; a rotor angle determination module for determining the second no-load rotor angle of the motor rotor based on the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map.

[0018] The present application has at least the following technical effects: Compared with the prior art, for the brushless motor control system and control method provided by the present application, at the factory stage, the sensorless FOC control mode is used to drive the brushless motor to run no-load to obtain the first no-load rotor angle. Then, a signal processing technology based on deep learning is adopted to perform time-frequency analysis on the sine wave feedback signals provided by two linear Hall sensors on the brushless motor drive board, extract the time-frequency feature representation of the signals, and perform fine-grained global interaction coding on the time-frequency features of the two sine wave feedback signals to identify and strengthen the key information interaction between the signals, realize high-precision prediction of the second no-load rotor angle of the brushless motor, and calculate the angle compensation value based on this. During the subsequent normal use of the motor, sensor FOC control based on the angle compensation value is performed on it, which can effectively reduce the influence of signal noise and calculation errors, improve the accuracy of motor rotor position estimation, and further improve the overall control accuracy of the motor. Description of the Drawings

[0019] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 It is a flowchart of the brushless motor control method according to an embodiment of the present application.

[0021] Figure 2 It is a schematic diagram of data flow of the brushless motor control method according to an embodiment of the present application.

[0022] Figure 3 It is a flowchart of sub-step S1 of the brushless motor control method according to an embodiment of the present application.

[0023] Figure 4 It is a flowchart of sub-step S2 of the brushless motor control method according to an embodiment of the present application.

[0024] Figure 5 It is a block diagram of the brushless motor control system according to an embodiment of the present application. Detailed Description of the Embodiments

[0025] As shown in the present application and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one", and / or "the" are not specifically singular and may also include the plural. Generally speaking, the terms "including" and "comprising" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0026] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules may be used and run on a user terminal and / or a server. The modules are merely illustrative, and different aspects of the system and method may use different modules.

[0027] Flowcharts are used in the present application to illustrate the operations performed by the system according to embodiments of the present application. It should be understood that the operations above or below do not necessarily have to be performed precisely in order. On the contrary, various steps may be processed in reverse order or simultaneously as needed. At the same time, other operations may also be added to these processes, or one or several operations may be removed from these processes.

[0028] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0029] The present application proposes an optimized brushless motor control method. At the factory stage, the sensorless FOC control mode is used to drive the brushless motor to run no-load to obtain the first no-load rotor angle. Then, a signal processing technology based on deep learning is adopted to perform time-frequency analysis on the sine wave feedback signals provided by two linear Hall sensors on the brushless motor drive board, extract the time-frequency feature representation of the signals, and perform fine-grained global interaction encoding on the time-frequency features of the two sine wave feedback signals to identify and strengthen the key information interaction between the signals, realize the high-precision prediction of the second no-load rotor angle of the brushless motor, and calculate the angle compensation value based on this. During the subsequent normal use of the motor, sensor FOC control based on the angle compensation value is performed, which can effectively reduce the influence of signal noise and calculation errors, improve the accuracy of motor rotor position estimation, and further improve the overall control accuracy of the motor.

[0030] Obtaining the first no-load rotor angle of the motor rotor is an important technical link in the field of motor control, which is related to the startup, positioning, and subsequent precise control of the motor. This process is usually carried out when the motor is not loaded with any load, aiming to determine the initial position of the motor rotor relative to the stator. Mastering this angle is crucial for implementing advanced control strategies such as sensorless vector control or direct torque control. The specific implementation of this process will be described in detail below: It should be clear that different types of motors (such as DC motors, asynchronous motors, synchronous motors, etc.) and different application scenarios may adopt different methods to obtain the first no-load rotor angle. However, in any case, a certain form of excitation signal is applied to the motor, and the rotor position information is inferred based on the motor response. Taking a permanent magnet synchronous motor as an example below, a common method based on pulse injection to obtain the first no-load rotor angle is introduced.

[0031] First, ensure that the motor is in a safe state, that is, not connected to any mechanical load, and the power supply has been correctly connected to the control system. In addition, the used controller also needs to be properly configured, including but not limited to setting the correct communication protocol, parameter adjustment, etc., to ensure that the subsequent operation steps can be accurately executed.

[0032] The pulse injection method is a technique widely used in induction motors and permanent magnet synchronous motors to estimate the rotor position. The basic principle of this method is to apply a short voltage or current pulse to the motor winding and then measure the resulting response. Due to the electromagnetic induction phenomenon inside the motor, when an external excitation is applied, the relative position between the stator and the rotor will affect the characteristics of the induced voltage waveform. By analyzing the changes in these waveforms, the current angle of the rotor can be indirectly deduced.

[0033] Amplitude: The size of the pulse should be large enough to cause a measurable change in the response, but not too large so as to cause unnecessary movement of the motor.

[0034] Duration: Too short may result in large signal noise and be difficult to distinguish; too long may cause the motor to actually rotate, affecting the accuracy of position estimation.

[0035] Frequency: Considering the electrical characteristics of the motor, reasonable selection of the pulse frequency is also important for improving the detection accuracy.

[0036] After determining the above parameters, then, the controller is used to apply pulses to one or more phases of the motor. At the same time, a high-precision data acquisition device is used to record the voltage or current changes on each phase. To reduce the influence of interference factors, it is best to conduct the experiment in a relatively stable environment, such as turning off the electronic devices that may generate electromagnetic interference around.

[0037] After collecting the raw data, the next step is to process and analyze it. This step mainly includes operations such as removing noise, smoothing the curve, and identifying key feature points. Then, using a mathematical model or physical law (such as the Lorentz force formula), combined with the design parameters of the motor, the information about the rotor position is extracted from the processed data.

[0038] Based on the above analysis results, the angle value of the rotor relative to a reference point (usually the axis of phase A) can be calculated through a specific algorithm. In practical applications, it is often necessary to repeat the above process multiple times and take the average value as the final result to improve the reliability of the measurement.

[0039] Through the above steps, the first no-load rotor angle of the motor rotor can be obtained more accurately.

[0040] At the same time, in order to achieve effective control of the brushless motor, it is usually necessary to use Hall sensors to detect the position information of the rotor. The Hall sensor can generate corresponding electrical signals according to the changes in the magnetic field. These signals are critical feedback information for the controller to determine when to switch the current direction in the stator winding to ensure smooth operation of the motor. In a linear Hall effect sensor, when the magnetic flux passes through its sensitive area, a voltage output proportional to it is generated. The following will introduce how to implement it to obtain these two sinusoidal feedback signals respectively: First of all, it should be clear that the so-called "no-load" state refers to the working condition when the motor is not loaded. In this state, the motor only overcomes its own friction and other internal losses to rotate, so the generated sinusoidal waveform is purer and easier to analyze and process. To obtain the first and second no-load sinusoidal wave feedback signals, the first step is to correctly install and connect the Hall sensor. Generally speaking, the Hall sensor is directly installed close to the motor rotor so that it can accurately sense the changing magnetic field generated by the rotor permanent magnet. Then, through appropriate circuit design, such as using a differential amplifier or a dedicated Hall IC, the original Hall element output is converted into a standard level signal that is easy to process later. It is worth noting that in actual applications, it may also be necessary to consider taking measures to suppress noise interference, such as using shielded cables, reasonably laying out PCB traces, etc., to improve signal quality.

[0041] Next, we enter the signal acquisition stage. For brushless motor control systems, a microcontroller unit (MCU) is usually equipped, which is responsible for receiving data from each Hall sensor and performing necessary calculation tasks. In this process, the MCU periodically reads the state of each Hall sensor, and then determines the exact position of the rotor based on the combined state of all sensors at the current moment. Since there is a fixed phase difference between the Hall sensors, when the motor is running, each sensor generates a time-varying voltage signal that is close to a sine wave. In particular, if two specific Hall sensors are of interest, two independent but interrelated sinusoidal signals can be obtained. For example, if the Hall sensors of phase A and phase B are selected as the observation objects, then as the rotor rotates, the phase A sensor will experience a process from low level to high level to low level before the phase B sensor; and vice versa. In this way, the estimation accuracy of the rotor position can be further refined by comparing the phase relationship between the two signals.

[0042] In order to achieve the above functions, the design at the software level is also crucial. A common practice is to write a dedicated interrupt service program on the MCU, trigger an interrupt whenever the state of the Hall sensor changes, and then update the relevant variable values in the interrupt service routine. In addition, a timer interrupt can be set to check the sensor status regularly, so that the motor status can be continuously monitored even when no external events occur. In this way, not only can the operating status of the motor be tracked in real time, but also basic data support can be provided for subsequent speed regulation, position control and other functions.

[0043] In summary, through the collaborative work of hardware configuration, signal processing, software programming and other aspects, the first no-load sinusoidal wave feedback signal and the second no-load sinusoidal wave feedback signal generated by the two-way linear Hall sensors installed on the brushless motor driver board are obtained, thereby providing key data for achieving precise control of the brushless motor.

[0044] Further, Figure 1 Flow chart of a brushless motor control method according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the brushless motor control method according to an embodiment of the present application. Figure 1 and Figure 2As shown, the brushless motor control method includes the steps of: S1, performing time-frequency feature extraction on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal respectively to obtain a time-frequency feature map of the first no-load sine wave feedback signal and a time-frequency feature map of the second no-load sine wave feedback signal; S2, performing global semantic interaction based on core joint feature guidance on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a first-second no-load feedback fine-grained joint semantic significant fusion coding feature map; S3, based on the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map, determining the second no-load rotor angle of the motor rotor.

[0045] In the above brushless motor control method, in step S1, time-frequency feature extraction is performed on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal respectively to obtain a time-frequency feature map of the first no-load sine wave feedback signal and a time-frequency feature map of the second no-load sine wave feedback signal. Among them, Figure 3 is a flowchart of sub-step S1 of the brushless motor control method according to an embodiment of the present application. As Figure 3 shown, step S1 includes the steps of: S11, performing wavelet analysis on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal respectively to obtain a time-frequency diagram of the first no-load sine wave feedback signal and a time-frequency diagram of the second no-load sine wave feedback signal; S12, using a neural network model to extract the time-frequency features of the time-frequency diagram of the first no-load sine wave feedback signal and the time-frequency diagram of the second no-load sine wave feedback signal respectively to obtain the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal.

[0046] Specifically, in step S11, wavelet analysis is performed on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal respectively to obtain the time-frequency diagram of the first no-load sine wave feedback signal and the time-frequency diagram of the second no-load sine wave feedback signal. Among them, wavelet transform is an effective time-frequency analysis tool that can decompose a signal into components of different scales and positions, obtain the wavelet coefficients of the signal at different scales, thereby reflecting the local characteristics of the signal at different frequencies and time points, and at the same time capturing the long-term trend and short-term rapid changes of the signal. Based on this, in this application, wavelet analysis is performed on the sine wave feedback signal, and the time-frequency diagram is drawn on the time-frequency plane according to the calculated wavelet coefficients, which helps to more intuitively understand the characteristics of the sine wave feedback signal changing with time and frequency, more accurately identify various frequency components in the signal, and excavate the high-frequency detail characteristics related to the position of the motor rotor, thereby improving the estimation accuracy of the rotor position. In addition, wavelet transform also has good denoising performance, and the noise components in the signal can be removed by selecting an appropriate threshold, thereby improving the accuracy of extracting the time-frequency characteristics of the signal in subsequent processing steps.

[0047] Specifically, step S12 further includes: inputting the time-frequency diagram of the first no-load sine wave feedback signal and the time-frequency diagram of the second no-load sine wave feedback signal into a signal time-frequency feature extractor based on a dilated convolutional neural network model to obtain the time-frequency feature diagram of the first no-load sine wave feedback signal and the time-frequency feature diagram of the second no-load sine wave feedback signal. Among them, in order to further capture the time-frequency change pattern of the sine wave feedback signal, this application uses a dilated convolutional neural network model to construct a signal time-frequency feature extractor, processes the time-frequency diagram of the first no-load sine wave feedback signal and the time-frequency diagram of the second no-load sine wave feedback signal respectively, and extracts highly representative time-frequency feature information from them. Among them, dilated convolution expands the receptive field by inserting zero values between the convolutional kernels without increasing the number of parameters or computational cost, enabling the model to capture context information in a larger range while maintaining a low computational complexity, thereby being able to effectively identify the periodic change pattern of the signal and enhancing the perception ability of the long-term change trend of the signal time-frequency characteristics.

[0048] In the above brushless motor control method, in step S2, global semantic interaction based on core joint feature guidance is performed on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a first-second no-load feedback fine-grained joint semantic significantly fused coding feature map. Among them, two linear Hall sensors of the brushless motor respectively provide two independent sine wave feedback signals. Due to the different installation positions of the sensors, there are differences in the magnetic field changes detected by the two. Therefore, in order to effectively utilize the phase difference and amplitude difference between the two sine wave feedback signals and extract the complementary information between the signals, the present application further performs fine-grained global interaction coding on the two sine wave feedback signals to identify and strengthen the key information interaction between the signals, so as to achieve high-precision prediction of the rotor position. Among them, Figure 4 is a flowchart of sub-step S2 of the brushless motor control method according to an embodiment of the present application. As Figure 4 shown, step S2 includes steps: S21, performing fine-grained common feature extraction on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a set of first-second no-load feedback fine-grained core joint feature vectors; S22, inputting the set of first-second no-load feedback fine-grained core joint feature vectors into a clustering network to obtain a first-second no-load feedback joint feature semantic clustering center vector; S23, based on the first-second no-load feedback joint feature semantic clustering center vector, guiding the set of first-second no-load feedback fine-grained core joint feature vectors to perform fine-grained context semantic interaction to obtain the first-second no-load feedback fine-grained joint semantic significantly fused coding feature map.

[0049] Specifically, step S21 includes: performing feature fine-grained decoupling on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a set of time-frequency feature vectors of the first no-load sine wave feedback signal and a set of time-frequency feature vectors of the second no-load sine wave feedback signal, which is expressed by the formula: Among them, represents the time-frequency feature map of the first no-load sine wave feedback signal, represents the time-frequency feature map of the second no-load sine wave feedback signal, represents feature decoupling, 、 、 and respectively represent the first, second, th, and th time-frequency feature vectors of the first no-load sine wave feedback signal in the set of time-frequency feature vectors of the first no-load sine wave feedback signal, , , and respectively represent the first, second, th, and th second no-load sine wave feedback signal time-frequency feature vectors in the set of second no-load sine wave feedback signal time-frequency feature vectors.

[0050] That is, by performing fine-grained decoupling processing on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal, both are decomposed into more detailed time-frequency feature units to more effectively identify and distinguish the local detail information in the time-frequency feature map.

[0051] Specifically, the step S21 further includes: inputting each group of corresponding first no-load sine wave feedback signal time-frequency feature vectors and second no-load sine wave feedback signal time-frequency feature vectors in the set of first no-load sine wave feedback signal time-frequency feature vectors and the set of second no-load sine wave feedback signal time-frequency feature vectors into a fine-grained common feature extraction network to obtain the set of first-second no-load feedback fine-grained core joint feature vectors, which is expressed by the formula: where, represents dot product, represents dot addition, represents dot subtraction, represents concatenation, and respectively represent the weight matrix and the bias term, represents the hyperbolic tangent function, represents the th first-second no-load feedback fine-grained core joint feature vector in the set of first-second no-load feedback fine-grained core joint feature vectors.

[0052] Specifically, by using a fine-grained common feature extraction network, through multi-level feature interaction and transformation, to identify and extract the common feature patterns between the two time-frequency feature sets, refine the internal connection between the two, so as to reduce the interference of inter-domain differences on the subsequent feature context interaction analysis, thereby generating the set of first-second no-load feedback fine-grained core joint feature vectors.

[0053] Specifically, the step S22 further includes: calculating the mean vector of the set of first-second no-load feedback fine-grained core joint feature vectors to obtain the first-second no-load feedback joint feature semantic clustering center vector, which is expressed by the formula: where, Represents the first - second no - load feedback combined feature semantic clustering center vector.

[0054] That is, using the clustering network to globally aggregate the set of the first - second no - load feedback fine - grained core combined feature vectors, generating the first - second no - load feedback fine - grained combined feature semantic clustering center vector, and using this as the global context information to guide the subsequent global context interaction of the combined features, thereby strengthening the consistency and semantic richness of the feature representation.

[0055] Specifically, step S23 includes: constructing a query vector and a value vector based on each of the first - second no - load feedback fine - grained core combined feature vectors in the set of the first - second no - load feedback fine - grained core combined feature vectors, and constructing a key vector based on the first - second no - load feedback combined feature semantic clustering center vector, and inputting the query vector, the value vector, and the key vector into the global semantic interaction encoder based on the Transformer structure to obtain the set of context first - second no - load feedback fine - grained core combined feature vectors, which is expressed by the formula: Where, Represents the first - second no - load feedback combined feature semantic clustering center vector, , And Represent the query embedding matrix, the key embedding matrix, and the value embedding matrix respectively, And Represent the query vector and the value vector corresponding to the th first - second no - load feedback fine - grained core combined feature vector respectively, Represents the key vector, Represents the length of the key vector, Represents the matrix multiplication operation, Represents the transpose of the vector, Represents the softmax function, Represents the set of context first - second no - load feedback fine - grained core combined feature vectors, , , And Represent the first, the second, the th, and the th context first - second no - load feedback fine - grained core combined feature vectors in the set of the context first - second no - load feedback fine - grained core combined feature vectors respectively.

[0056] That is, a query vector and a value vector are constructed based on the first and second no-load feedback fine-grained core joint feature vectors. At the same time, a key vector is constructed using the first and second no-load feedback fine-grained joint feature semantic clustering center vectors. A transformer structure is used to perform global context interaction encoding for each local fine-grained joint feature. Through the action of the self-attention mechanism, local joint information exchange within the global scope is realized, which helps to adjust its own representation according to the global context information and improve the accuracy of the expressed features.

[0057] Specifically, step S23 further includes: reshaping the feature shape of the set of the context first and second no-load feedback fine-grained core joint feature vectors to obtain the first and second no-load feedback fine-grained joint semantic significant fusion encoded feature map, which is expressed by the formula: Wherein, represents feature shape reshaping, represents the first and second no-load feedback fine-grained joint semantic significant fusion encoded feature map. That is, a feature shape reshaping operation is performed on the generated set of the context first and second no-load feedback fine-grained core joint feature vectors to restore the original feature structure and generate the first and second no-load feedback fine-grained joint semantic significant fusion encoded feature map.

[0058] Through the above method, deep interaction analysis of the two-way no-load sine wave feedback signals can be realized at the fine-grained level, which helps to improve the feature sensitivity and decoding accuracy of the rotor position information.

[0059] In the above brushless motor control method, in step S3, based on the first and second no-load feedback fine-grained joint semantic significant fusion encoded feature map, the second no-load rotor angle of the motor rotor is determined. In a specific example of the present application, step S3 further includes: inputting the first and second no-load feedback fine-grained joint semantic significant fusion encoded feature map into a no-load rotor angle estimator based on a decoder to obtain the second no-load rotor angle of the motor rotor. Wherein, the decoder extracts and learns the spatio-temporal correlation interaction features of the two-way no-load sine wave signals layer by layer through multi-level feature learning of the first and second no-load feedback fine-grained joint semantic significant fusion encoded feature map, and decodes and estimates the rotor angle accordingly.

[0060] A decoder is a technology for processing data after complex transformations, which is designed to recover valuable information from the coded features after complex processing, namely, the second no-load rotor angle of the motor rotor. In the technical solution of the present application, the decoder adopts a multi-layer nonlinear transformation method to achieve in-depth mining of the signal time-frequency interaction features in the first-second no-load feedback fine-grained joint semantic saliency fusion coding feature map. Specifically, the decoder is composed of multiple neural network layers, each of which has a specific function and is nonlinearly transformed through activation functions such as ReLU, Sigmoid or Tanh to enhance the nonlinear ability of feature expression. In addition, in each layer of the decoder, the network weights are gradually optimized through forward propagation and back propagation algorithms to minimize the error between the estimated rotor angle and the actual rotor angle, so that the decoder gradually learns how to accurately extract useful information from the given coding features and more accurately estimate the angle of the motor rotor. This not only helps to improve the control accuracy of the brushless motor, but also maintains good robustness and stability under interference or changing operating conditions. Finally, the output result of the decoder, i.e., the second no-load rotor angle of the motor rotor, can provide an important basis for the subsequent motor control strategy, ensuring that the motor operates efficiently and safely in an expected manner.

[0061] When the first no-loaded sine wave feedback signal time-frequency feature graph and the second no-loaded sine wave feedback signal time-frequency feature graph respectively represent the image semantic coding features of the first no-loaded sine wave feedback signal time-frequency graph and the second no-loaded sine wave feedback signal time-frequency graph determined by wavelet analysis of the first no-loaded sine wave feedback signal and the second no-loaded sine wave feedback signal, the first-second no-loaded feedback fine-grained joint semantically significant fusion coding feature graph will also have the joint interactive distribution diversity of heterogeneous sinusoidal wave feedback information based on their respective feature fine-grained cores under the cross-feature domain feature distribution. Therefore, when the first-second no-loaded feedback fine-grained joint semantically significant fusion coding feature graph is decoded by a decoder, it will affect the accuracy of the decoding result.

[0062] Preferably, in the process of inputting the first-second no-load feedback fine-grained joint semantically significant fusion coding feature map into a decoder-based no-load rotor angle estimator to obtain a second no-load rotor angle of the motor rotor, the first-second no-load feedback fine-grained joint semantically significant fusion coding feature map is subjected to feature fine-grained optimization, and the process includes: Expanding the first-second no-load feedback fine-grained joint semantically significant fusion coding feature map into a first-second no-load feedback fine-grained joint semantically significant fusion coding feature vector; Perform a correlation analysis based on linear operations on the eigenvalues at any two positions in the first-second no-load feedback fine-grained joint semantic significant fusion coding feature vector to obtain the first no-load feedback fine-grained joint semantic significant fusion coding heterogeneous correlation matrix and the second no-load feedback fine-grained joint semantic significant fusion coding heterogeneous correlation matrix, expressed as: Wherein, represents the first-second no-load feedback fine-grained joint semantic significant fusion coding feature vector, and respectively represent the eigenvalues at the th and th positions in the first-second no-load feedback fine-grained joint semantic significant fusion coding feature vector, represents the first weight hyperparameter, represents the second weight hyperparameter, represents the third weight hyperparameter, represents the fourth weight hyperparameter, represents the eigenvalue at the position in the first no-load feedback fine-grained joint semantic significant fusion coding heterogeneous correlation matrix, represents the eigenvalue at the position in the second no-load feedback fine-grained joint semantic significant fusion coding heterogeneous correlation matrix; Using the first no-load feedback fine-grained joint semantic significant fusion coding heterogeneous correlation matrix and the second no-load feedback fine-grained joint semantic significant fusion coding heterogeneous correlation matrix as the target feature domain, perform vector mapping modulation on the first-second no-load feedback fine-grained joint semantic significant fusion coding feature vector to obtain the first no-load feedback fine-grained joint semantic significant fusion coding gain compensation characterization coding vector and the second no-load feedback fine-grained joint semantic significant fusion coding gain compensation characterization coding vector, expressed as: Wherein, represents matrix multiplication, represents the natural exponential function, represents the first no-load feedback fine-grained joint semantic significant fusion coding gain compensation characterization coding vector, represents the second no-load feedback fine-grained joint semantic significant fusion coding gain compensation characterization coding vector; Calculate the F-norm of the first no-load feedback fine-grained joint semantic significant fusion encoded heterogeneous association matrix and the second no-load feedback fine-grained joint semantic significant fusion encoded heterogeneous association matrix, and regard it as the gain space significant adjustment factor to perform significance optimization on the first no-load feedback fine-grained joint semantic significant fusion encoded gain compensation characterization encoding vector and the second no-load feedback fine-grained joint semantic significant fusion encoded gain compensation characterization encoding vector to obtain the first no-load feedback fine-grained joint semantic significant fusion encoded discriminative gain vector and the second no-load feedback fine-grained joint semantic significant fusion encoded discriminative gain vector, expressed as: where, represents the F-norm of the matrix, represents the first gain space significant adjustment factor, represents the second gain space significant adjustment factor, represents element-wise multiplication, represents the first no-load feedback fine-grained joint semantic significant fusion encoded discriminative gain vector, represents the second no-load feedback fine-grained joint semantic significant fusion encoded discriminative gain vector; Fuse the first no-load feedback fine-grained joint semantic significant fusion encoded discriminative gain vector and the second no-load feedback fine-grained joint semantic significant fusion encoded discriminative gain vector to obtain the optimized first-second no-load feedback fine-grained joint semantic significant fusion encoded feature vector, expressed as: where, represents the fifth weight hyperparameter, represents the sixth weight hyperparameter, represents vector addition, represents the optimized first-second no-load feedback fine-grained joint semantic significant fusion encoded feature vector.

[0063] Finally, input the optimized first-second no-load feedback fine-grained joint semantic significant fusion encoded feature vector into the decoder-based no-load rotor angle estimator to obtain the second no-load rotor angle of the motor rotor.

[0064] Accordingly, by improving the projection matrix tuning mechanism of the first and second no-load feedback fine-grained joint semantic significant fusion coding feature vectors obtained by unfolding the first and second no-load feedback fine-grained joint semantic significant fusion coding feature maps in the heterogeneous attribute field, a progressive optimization framework for the hierarchical random topological structure in the unannotated homomorphic representation extraction process is constructed. Furthermore, by applying the elastic gain adjustment strategy, the interference of the skewed data attenuation factor on the feature representation is effectively eliminated, and at the same time, the discriminative target improvement mechanism under the dimensionality reduction constraint is integrated, significantly enhancing the identification efficiency of the attribute saliency weighting metric in the heterogeneous space. In this way, the accuracy of the second no-load rotor angle of the motor rotor obtained by the first and second no-load feedback fine-grained joint semantic significant fusion coding feature maps through the decoder-based no-load rotor angle estimator is improved.

[0065] Finally, the present application determines the angle compensation value based on the difference between the first no-load rotor angle and the second no-load rotor angle. During the subsequent motor startup, this angle compensation value is applied to the sensorless FOC control algorithm to drive the brushless motor to operate, so as to achieve precise control of the brushless motor.

[0066] In summary, the brushless motor control method based on the embodiments of the present application is elucidated. At the factory stage, it uses the sensorless FOC control mode to drive the brushless motor to run no-load to obtain the first no-load rotor angle. Then, it uses signal processing techniques based on deep learning to perform time-frequency analysis on the sine wave feedback signals provided by two linear Hall sensors on the brushless motor drive board, extracts the time-frequency feature representations of the signals, and performs fine-grained global interaction coding on the time-frequency features of the two sine wave feedback signals to identify and strengthen the key information interaction between the signals, achieve high-precision prediction of the second no-load rotor angle of the brushless motor, and calculate the angle compensation value based on this. During the subsequent normal use of the motor, sensorless FOC control based on the angle compensation value is performed on it, which can effectively reduce the influence of signal noise and calculation errors, improve the accuracy of motor rotor position estimation, and further improve the overall control accuracy of the motor.

[0067] Furthermore, a brushless motor control system is also provided. Figure 5 It is a block diagram of the brushless motor control system according to the embodiments of the present application. As Figure 5As shown, the brushless motor control system 100 according to an embodiment of the present application includes: a signal receiving module 110, configured to receive a first no-load sine wave feedback signal and a second no-load sine wave feedback signal generated by two linear Hall sensors installed on a drive board of a brushless motor; a time-frequency feature extraction module 120, configured to perform time-frequency feature extraction on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal respectively to obtain a time-frequency feature map of the first no-load sine wave feedback signal and a time-frequency feature map of the second no-load sine wave feedback signal; a semantic interaction fusion module 130, configured to perform global semantic interaction based on core joint feature guidance on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a first-second no-load feedback fine-grained joint semantic significant fusion coding feature map; a rotor angle determination module 140, configured to determine a second no-load rotor angle of the motor rotor based on the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map.

[0068] The specific operations of each module in the above brushless motor control system have been described in detail above with reference to Figures 1 to 4 the description of the brushless motor control method, and thus, the repeated description thereof will be omitted.

[0069] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, advantages, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, advantages, effects, etc. are essential for each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and easy understanding, rather than limitations. The above details do not limit the present invention to necessarily adopt the above specific details to implement.

[0070] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0071] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any regard, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Accordingly, all changes that fall within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims concerned.

[0072] In addition, it is obvious that the term "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. The multiple elements recited in the system claims can also be implemented by one element through software or hardware.

[0073] Finally, it should be noted that the above description has been given for purposes of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A brushless motor control method, comprising: At the factory stage, use the sensorless FOC control mode to drive the brushless motor to run no-load; obtain the first no-load rotor angle of the motor rotor; Respectively obtain the first no-load sine wave feedback signal and the second no-load sine wave feedback signal generated by two linear Hall sensors installed on the drive board of the brushless motor; analyze and calculate the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain the second no-load rotor angle of the motor rotor; obtain the angle compensation value according to the first no-load rotor angle and the second no-load rotor angle; at the start-up stage, based on the angle compensation value, use the sensor FOC control mode to drive the brushless motor to run. It is characterized in that analyzing and calculating the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain the second no-load rotor angle of the motor rotor includes: Respectively perform time-frequency feature extraction on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a first no-load sine wave feedback signal time-frequency feature map and a second no-load sine wave feedback signal time-frequency feature map; Perform global semantic interaction based on the core joint feature guidance on the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map to obtain a first-second no-load feedback fine-grained joint semantic significantly fused coding feature map; Based on the first-second no-load feedback fine-grained joint semantic significantly fused coding feature map, determine the second no-load rotor angle of the motor rotor.

2. The brushless motor control method according to claim 1, characterized in that Respectively perform time-frequency feature extraction on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a first no-load sine wave feedback signal time-frequency feature map and a second no-load sine wave feedback signal time-frequency feature map, including: Respectively perform wavelet analysis on the first no-load sine wave feedback signal and the second no-load sine wave feedback signal to obtain a first no-load sine wave feedback signal time-frequency diagram and a second no-load sine wave feedback signal time-frequency diagram; Use a neural network model to respectively extract the time-frequency features of the first no-load sine wave feedback signal time-frequency diagram and the second no-load sine wave feedback signal time-frequency diagram to obtain the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map.

3. The brushless motor control method according to claim 2, wherein Use a neural network model to respectively extract the time-frequency features of the first no-load sine wave feedback signal time-frequency diagram and the second no-load sine wave feedback signal time-frequency diagram to obtain the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map, including: Input the first no-load sine wave feedback signal time-frequency diagram and the second no-load sine wave feedback signal time-frequency diagram into a signal time-frequency feature extractor based on a dilated convolutional neural network model to obtain the first no-load sine wave feedback signal time-frequency feature map and the second no-load sine wave feedback signal time-frequency feature map.

4. The brushless motor control method according to claim 3, wherein Perform global semantic interaction based on core joint feature guidance on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a first-second no-load feedback fine-grained joint semantic significant fusion coding feature map, including: Extract fine-grained common features from the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a set of first-second no-load feedback fine-grained core joint feature vectors; Input the set of first-second no-load feedback fine-grained core joint feature vectors into a clustering network to obtain a first-second no-load feedback joint feature semantic clustering center vector; Based on the first-second no-load feedback joint feature semantic clustering center vector, guide the set of first-second no-load feedback fine-grained core joint feature vectors to perform fine-grained context semantic interaction to obtain the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map.

5. The brushless motor control method according to claim 4, characterized in that, Extract fine-grained common features from the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a set of first-second no-load feedback fine-grained core joint feature vectors, including: Perform feature fine-grained decoupling on the time-frequency feature map of the first no-load sine wave feedback signal and the time-frequency feature map of the second no-load sine wave feedback signal to obtain a set of time-frequency feature vectors of the first no-load sine wave feedback signal and a set of time-frequency feature vectors of the second no-load sine wave feedback signal; Input each pair of corresponding time-frequency feature vectors of the first no-load sine wave feedback signal and the second no-load sine wave feedback signal in the set of time-frequency feature vectors of the first no-load sine wave feedback signal and the set of time-frequency feature vectors of the second no-load sine wave feedback signal into a fine-grained common feature extraction network to obtain the set of first-second no-load feedback fine-grained core joint feature vectors.

6. The brushless motor control method according to claim 5, wherein Input the set of first-second no-load feedback fine-grained core joint feature vectors into a clustering network to obtain a first-second no-load feedback joint feature semantic clustering center vector, including: Calculate the mean vector of the set of first-second no-load feedback fine-grained core joint feature vectors to obtain the first-second no-load feedback joint feature semantic clustering center vector.

7. The brushless motor control method according to claim 6, wherein Based on the first-second no-load feedback joint feature semantic clustering center vector, guide the set of first-second no-load feedback fine-grained core joint feature vectors to perform fine-grained context semantic interaction to obtain the first-second no-load feedback fine-grained joint semantic significant fusion coding feature map, including: Construct query vectors and value vectors based on each first-second no-load feedback fine-grained core joint feature vector in the set of first-second no-load feedback fine-grained core joint feature vectors, and construct key vectors based on the first-second no-load feedback joint feature semantic clustering center vector, and input the query vectors, the value vectors and the key vectors into a global semantic interaction encoder based on the transformer structure to obtain a set of context first-second no-load feedback fine-grained core joint feature vectors; Reshape the set of the first - second no - load feedback fine - grained core joint feature vectors of the context to obtain the first - second no - load feedback fine - grained joint semantic significant fusion encoded feature map.

8. The brushless motor control method according to claim 7, wherein, Based on the first - second no - load feedback fine - grained joint semantic significant fusion encoded feature map, determining the second no - load rotor angle of the motor rotor includes: Inputting the first - second no - load feedback fine - grained joint semantic significant fusion encoded feature map into a no - load rotor angle estimator based on a decoder to obtain the second no - load rotor angle of the motor rotor.

9. A brushless motor control system, characterized in that, Including: A signal receiving module, configured to receive a first no - load sine wave feedback signal and a second no - load sine wave feedback signal generated by two linear Hall sensors installed on a drive board of a brushless motor; A time - frequency feature extraction module, configured to perform time - frequency feature extraction on the first no - load sine wave feedback signal and the second no - load sine wave feedback signal respectively to obtain a first no - load sine wave feedback signal time - frequency feature map and a second no - load sine wave feedback signal time - frequency feature map; A semantic interaction fusion module, configured to perform global semantic interaction guided by core joint features on the first no - load sine wave feedback signal time - frequency feature map and the second no - load sine wave feedback signal time - frequency feature map to obtain a first - second no - load feedback fine - grained joint semantic significant fusion encoded feature map.

10. The brushless motor control system according to claim 9, characterized in that, Further including: A rotor angle determination module, configured to determine the second no - load rotor angle of the motor rotor based on the first - second no - load feedback fine - grained joint semantic significant fusion encoded feature map.

Images