Sensorless direct current brushless motor control system and method

Abstract

The invention relates to the technical field of motor control, and discloses a sensorless direct-current brushless motor control system which comprises a motor modeling module, a state estimation module, a deep learning prediction module, a filtering fusion module, a self-adaptive control module and a control execution module. The invention also discloses a sensorless direct current brushless motor control method, which comprises the following steps of: model establishment: constructing an electrical model and a mechanical model for describing the dynamic behavior of the motor so as to obtain a motor operation dynamic model suitable for control calculation; and state estimation: performing preliminary estimation on the rotor position and the rotor speed of the motor by utilizing the running dynamic model and combining current and voltage real-time signals acquired in the running process of the motor. According to the method, high-precision and high-stability control of the motor under different load and operation conditions is realized through adaptive control and rotor state estimation optimization technologies, real-time dynamic adjustment and commutation accuracy of the motor are ensured, and the overall control performance is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of motor control, and in particular to a sensorless brushless DC motor control system and method. Background Art

[0002] In modern industry and daily life, motors are core drive components, and their stability and precise control are crucial to the performance of equipment. With technological advancements, the requirements for motor control systems are becoming increasingly higher, especially in terms of high efficiency, stability, and precise control.

[0003] Existing motor control technologies mostly rely on control methods based on fixed-gain parameters, which can meet control requirements within a certain range. Fixed-gain control methods are relatively simple and easy to implement, and can provide relatively stable motor operation. In certain standardized and simple application scenarios, these technical solutions perform well, offering the advantages of low cost and ease of implementation. Furthermore, existing technologies also employ optimization algorithms for rotor state estimation, which can reduce estimation errors to a certain extent and improve the stability and reliability of motor operation.

[0004] However, existing technologies often lack real-time feedback and dynamic adjustment of the motor's operating state. Fixed-gain control cannot adapt to changes in load and operating conditions, resulting in unstable motor accuracy. While existing technologies can achieve certain optimizations in rotor state estimation, accumulated errors are unavoidable, particularly at high speeds or under sudden load changes. Furthermore, control of commutation timing is imprecise, failing to effectively adapt to real-time changes in the motor's operation. These deficiencies often lead to unresponsive systems and unstable performance in practical applications. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention provides a sensorless DC brushless motor control system and method, which solves the problem in the prior art that the motor control system cannot dynamically adjust the gain parameters in real time.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a sensorless DC brushless motor control system, comprising: A motor modeling module, which establishes a dynamic operation model of the motor based on a preset electrical model and a mechanical model, and outputs the dynamic operation model to the state estimation module; A state estimation module is used to receive the operation dynamic model output by the motor modeling module and the motor operation parameters collected in real time, make a preliminary estimate of the rotor position and rotor speed of the motor, and output the preliminary estimation results to the filtering and fusion module; The deep learning prediction module is used to receive the current and voltage signals of the motor during operation, and predict the rotor position and rotor speed of the motor based on the trained neural network model, generate the prediction results and output them to the filtering and fusion module; a filtering fusion module, configured to receive the preliminary estimation result output by the state estimation module and the prediction result output by the deep learning prediction module, fuse the two, and output an optimized rotor state estimation result to the adaptive control module; an adaptive control module, configured to receive the optimized rotor state estimation result output by the filtering and fusion module, dynamically adjust the gain parameters of the motor controller according to the error between the rotor state and the target state, and output the adjusted control parameters to the control execution module; The control execution module is used to receive the control parameters output by the adaptive control module and control the power-on logic and commutation timing of the motor.

[0007] Preferably, the motor modeling module includes: An electrical modeling unit, used to establish an electrical characteristic model of the stator side based on the inductance, resistance and electromotive force parameters of the motor; A mechanical modeling unit is used to establish a mechanical dynamic model describing the relationship between the rotor angular velocity and position based on the rotor inertia, load torque and damping coefficient of the motor; The model output unit is used to integrate the established electrical model and mechanical model into a joint dynamic model and transmit the model to the state estimation module.

[0008] Preferably, the state estimation module includes: Data acquisition unit, used to collect parameters such as current, voltage, temperature, etc. during motor operation in real time; an estimation processing unit, configured to estimate the current rotor position and rotor speed of the motor by combining the operating dynamic model with the collected data; The estimation result output unit is used to output the preliminary estimated rotor state to the filtering fusion module.

[0009] Preferably, the deep learning prediction module includes: A signal preprocessing unit is used to perform preprocessing operations such as normalization and filtering on the input current signal and voltage signal; The neural network inference unit is used to load and run the trained neural network model and output the corresponding rotor position and rotor speed prediction values; The prediction output unit is used to output the prediction value to the filtering fusion module.

[0010] Preferably, the filtering fusion module includes: An estimation result receiving unit, configured to receive the preliminary estimation result from the state estimation module and the prediction result from the deep learning prediction module; A fusion processing unit, configured to perform joint optimization processing on two sets of rotor state data based on an extended Kalman filter algorithm or other state fusion algorithms; The fusion result output unit is used to output the optimized rotor state estimation result to the adaptive control module.

[0011] Preferably, the adaptive control module includes: an error calculation unit, for calculating the error between the current rotor state and a preset target state; a parameter adjustment unit, configured to adaptively adjust proportional, integral, and differential parameters of the controller based on the error result; The control parameter output unit is used to transmit the adjusted controller parameters to the control execution module.

[0012] Preferably, the control execution module includes: A control signal generating unit, configured to generate a commutation control signal and a PWM modulation signal according to received controller parameters; The drive instruction output unit is used to convert the control signal into a specific power-on logic control instruction for each phase winding of the motor, so as to realize low-speed sensorless starting and operation control of the motor.

[0013] Preferably, the model output unit includes: A model integration unit is used to perform time domain alignment and parameter matching on the modeling data generated by the electrical modeling unit and the mechanical modeling unit, and integrate them into a joint dynamic model; A structured output unit for formatting the joint dynamic model into a data structure adapted to the computational framework of the state estimation module; The dynamic parameter updating unit is used to update the modeling parameters according to real-time factors such as ambient temperature and motor aging during motor operation, and re-output the dynamic model.

[0014] Preferably, the signal preprocessing unit includes: The filtering and denoising unit is used to perform bandpass filtering, averaging filtering and other processing on the collected current and voltage signals to eliminate high-frequency noise and instantaneous interference components; The normalization processing unit is used to scale and offset the original signal according to the input specifications of the deep learning model to make it conform to the network input format; The data cache unit is used to cache and organize the continuous sampling data in time windows so that the neural network inference unit can perform batch input and feature extraction operations.

[0015] The present invention also provides a sensorless brushless DC motor control method, comprising the following steps: Model building: constructing electrical and mechanical models to describe the dynamic behavior of the motor, in order to obtain a dynamic model of the motor suitable for control calculations; State estimation, using the dynamic operation model and the real-time current and voltage signals collected during the operation of the motor to make a preliminary estimate of the rotor position and rotor speed of the motor; Predictive inference: The collected current and voltage signals are input into a trained deep learning model to predict the motor's current rotor position and speed. Fusion optimization: inputting the preliminary estimation result and the prediction result into the extended Kalman filter, and outputting the optimized rotor state estimation result through filter fusion processing; Control parameter adjustment, based on the optimized rotor state estimation results, compares the deviation between the current state and the preset target state, and then dynamically adjusts the gain parameters of the motor controller through an adaptive control strategy; Control execution, generating commutation control signals according to the adjusted control parameters, and then controlling the power-on logic of each phase of the motor to achieve stable startup and precise operation in the low-speed stage; Parameters are continuously optimized. The rotor position and speed errors during the motor control process are evaluated in real time based on the defined objective function, and the controller parameters are continuously optimized in combination with the optimization algorithm.

[0016] The present invention provides a sensorless brushless DC motor control system and method. It has the following beneficial effects: 1. This invention utilizes adaptive control technology to maintain efficient and stable motor operation under various operating conditions by adjusting the motor controller's gain parameters in real time. Compared to existing control schemes with fixed gain parameters, this invention solves the problem of unstable motor efficiency under varying loads and operating conditions, enabling precise dynamic adjustment.

[0017] 2. This invention achieves higher accuracy and response speed by optimizing rotor state estimation through a filter fusion module and adjusting control parameters in conjunction with an adaptive control module. Compared to traditional control schemes, this invention can more accurately estimate rotor state, significantly improving motor control accuracy and avoiding control instability caused by error accumulation.

[0018] 3. This invention precisely controls the motor's power-on logic and commutation timing by controlling the execution module, improving the motor's torque output and operational stability. Compared to conventional motor control schemes, this invention can better cope with transient changes during motor commutation in dynamically changing operating environments, avoiding performance fluctuations caused by improper commutation.

[0019] 4. This invention utilizes a control strategy based on real-time error adjustment, enabling the motor to flexibly adapt to varying load conditions, thereby improving the adaptability and stability of the motor control system. Compared to traditional hardware control methods, this invention solves the problems of system response delays and transitional jitter caused by changes in the motor's operating environment, thereby optimizing overall control performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the system construction of the present invention; Figure 2 This is a framework diagram of the motor modeling module of the present invention; Figure 3 This is a state estimation module framework diagram of the present invention; Figure 4 This is a framework diagram of the deep learning prediction module of the present invention; Figure 5 This is a framework diagram of the filtering fusion module of the present invention; Figure 6 This is a framework diagram of the adaptive control module of the present invention; Figure 7 This is a framework diagram of the control execution module of the present invention; Figure 8 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] Please see the attached Figure 1 -Attached Figure 7 , an embodiment of the present invention provides a sensorless DC brushless motor control system, comprising: A motor modeling module, which establishes a dynamic operation model of the motor based on a preset electrical model and a mechanical model, and outputs the dynamic operation model to the state estimation module; By establishing a dynamic motor model, we provide an accurate description of its dynamic behavior for the subsequent state estimation module. This modeling process includes two core components: electrical modeling and mechanical modeling. These two models are combined to form a complete dynamic motor model, ultimately providing the system with a basis for estimating parameters such as rotor position and rotor speed.

[0023] In this embodiment, the motor modeling module includes an electrical modeling unit, a mechanical modeling unit, and a model output unit. The electrical and mechanical modeling units describe the motor's behavior from two perspectives: electrical and mechanical. By modeling the motor's electrical characteristics and mechanical dynamics, combined with known physical laws, the system can form a comprehensive dynamic model of the motor. This model provides high-quality data support for subsequent state estimation and control modules.

[0024] First, the electrical modeling unit creates an electrical characteristic model based on the motor's electrical parameters. This model typically takes into account factors such as stator inductance, resistance, and back electromotive force. These parameters primarily influence the motor's current variation. Therefore, the electrical modeling unit describes the motor's electrical behavior using the following electrical model: ; in: is the stator voltage; is the stator inductance; is the stator current; is the stator resistance; is the back electromotive force; is the rate of change of current.

[0025] Back EMF The expression is usually: ; in: is the back electromotive force; is the back electromotive force constant of the motor; is the motor rotor angular velocity.

[0026] The above electrical model can describe how the stator current is affected by the motor voltage and other electrical parameters, thus providing a basis for motor speed control.

[0027] Next, the mechanical modeling unit considers the physical properties of the motor rotor. Mechanical modeling focuses on describing the relationship between the rotor's angular velocity, rotor position, and load torque. Mechanical dynamics equations of the following form are typically used to describe the rotor's behavior: ; in: is the rate of change of rotational speed; is the rotor inertia; is the motor rotor angular velocity; is the torque generated by the motor; is the load torque; is the damping coefficient.

[0028] Using this equation, mechanical modeling can calculate the changing pattern of the rotor's angular velocity and accurately describe the rotor's dynamic behavior. Specifically, the effects of rotor inertia and load torque determine the rotor's response characteristics during motor control, while the damping coefficient further regulates the stability of the rotor's motion.

[0029] After combining the electrical and mechanical models, the model output unit integrates the two sub-models into a joint dynamic model. This joint model uses time-domain alignment and parameter matching to ensure seamless integration of electrical and mechanical characteristics. For example, the current and voltage signals in the electrical model may affect the rotor's angular velocity, which in turn affects the motor's back EMF, thus affecting the electrical characteristics. Therefore, the interaction between the two sub-models is fully accounted for in the joint model.

[0030] Electrical and mechanical modeling involve modeling the electrical and mechanical components of the motor system, using appropriate mathematical models to describe the motor's electrical and mechanical behavior, respectively. The electrical component uses parameters such as inductance and resistance to establish a circuit model, while the mechanical component uses factors such as rotor inertia and load to establish a dynamic model. To achieve seamless integration between the electrical and mechanical components, we employ a process of time domain alignment and parameter matching.

[0031] Circuit Model: Modeling the electrical components primarily relies on the circuit model. This model describes the motor's electrical behavior using electrical parameters such as inductance and resistance. Specifically, the circuit model focuses on the voltage and current in the motor armature, as well as the electromagnetic characteristics associated with the motor windings. It reflects the motor's electrical response under different operating conditions by establishing relationships between current, voltage, and electromagnetic fields. Parameters such as inductance and resistance describe the impedance characteristics of the motor circuit, while changes in voltage and current reveal how the motor exchanges energy with the external power supply and load.

[0032] Dynamic Model: The mechanical modeling is primarily performed using a dynamic model, which describes the motor's mechanical behavior. The dynamic model focuses on the rotor's inertia, the effects of the load, and the motion caused by the motor's torque. By using the rotor's inertia and load characteristics to establish dynamic relationships related to the motor's motion, the model can describe the motor's response to changes in torque and load. Furthermore, the dynamic model includes key variables such as speed and angular displacement, which, along with the electrical outputs such as voltage and current, determine the motor's overall behavior.

[0033] Specifically, the outputs of the electrical modeling component (such as voltage and current) are synchronized with the inputs of the mechanical modeling component (such as rotor speed and angular displacement). This process ensures physical consistency between the electrical and mechanical submodels by calculating the response of each submodel in the time domain, avoiding oversimplification or neglect of coupling effects between the two.

[0034] Regarding parameter matching, we adjust the interface parameters (such as torque and speed) between the electrical and mechanical components to ensure that the two models function correctly when the system is linked. This process not only ensures a natural connection between the electrical and mechanical components but also avoids deviations in the model output.

[0035] In one possible implementation, the joint dynamic model not only includes the effects of stator current and voltage but also incorporates the dynamic behavior of the rotor. This model provides an accurate description of the current motor state at every moment during operation. This state information is passed as input to the subsequent state estimation module for further real-time estimation.

[0036] Furthermore, in this embodiment, the output of the motor modeling module is not just a static mathematical model. It can be dynamically updated, adjusting model parameters in real time based on the operating environment. For example, motor performance varies under different temperatures or loads. Therefore, the model output unit adjusts relevant parameters in the electrical and mechanical models based on real-time collected environmental parameters (such as temperature and load) to ensure model accuracy.

[0037] Specifically, the resistance in the electrical model may change with temperature, and the inductance may also change with motor aging. These changes can affect motor performance. Therefore, the model output unit continuously updates the dynamic parameters during motor operation and transmits the updated dynamic model to the state estimation module.

[0038] In some embodiments, the electrical and mechanical models in the motor modeling module can be further optimized for different motor types (e.g., internal and external rotor motors). Different motor types exhibit different dynamic behaviors, so the model structure and parameters need to be adjusted based on the specific motor type during modeling.

[0039] A state estimation module is used to receive the operation dynamic model output by the motor modeling module and the motor operation parameters collected in real time, make a preliminary estimate of the rotor position and rotor speed of the motor, and output the preliminary estimation results to the filtering and fusion module; The state estimation module receives the dynamic operating model from the motor modeling module and, combined with real-time motor operating parameters, provides a preliminary estimate of the motor's rotor position and speed. This module's core function is to provide accurate rotor state information based on the motor's dynamic model and actual operating data. This process is crucial for sensorless control because it provides a preliminary rotor state estimate for the subsequent filter fusion module.

[0040] In this embodiment, the state estimation module includes a data acquisition unit, an estimation processing unit, and an estimation result output unit. Through the collaboration of these submodules, the state estimation module can obtain the motor's rotor position and rotor speed in real time through calculation and model inference without the need for additional sensors.

[0041] Typically, the state estimation module first receives the joint dynamic model output by the motor modeling module. The motor modeling module provides a model based on a combination of electrical and mechanical characteristics, encompassing both the electrical and mechanical behavior of the motor's rotor and stator. Specifically, the electrical model describes the relationship between stator voltage and current and inductance, resistance, and back EMF, while the mechanical model describes the dynamic relationship between rotor inertia, torque, and speed.

[0042] In one possible implementation, a state estimation module uses this joint dynamic model, combined with real-time motor operating data, to estimate the motor's rotor state. By monitoring parameters such as current, voltage, and temperature in real time and integrating them with the motor's dynamic operating model, the state estimation module can make a preliminary estimate of the motor's rotor position and speed.

[0043] Specifically, the data acquisition unit in the state estimation module first collects real-time data on motor parameters such as current, voltage, and temperature during operation. Current and voltage signals directly reflect the motor's electrical state, while temperature provides supplementary information on the motor's internal state. Temperature data is crucial to estimation accuracy, as motor performance can vary under high load or high temperature conditions.

[0044] As an option, the estimation processing unit combines the motor's operating dynamic model with the collected electrical data, and estimates the motor's rotor position and rotor speed through mathematical derivation and estimation methods (such as the Kalman filter algorithm). By using algorithms such as the Kalman filter, the system can effectively suppress the influence of noise and improve the accuracy of the estimation results. The basic idea of the Kalman filter algorithm is to update the estimated value in each time step by combining prediction and correction. Assume that the current estimated state is , the rotor is estimated using the dynamic model and corrected based on the measured values: ; in: is the state transfer matrix, which describes the dynamic characteristics of the system; Estimate the state of the previous step; is the damping coefficient; is the control input; is the process noise; is the estimated state at the current moment.

[0045] Through this method, the state estimation module can estimate the rotor position and speed in real time based on the dynamic model of the motor and the collected real-time data, and calculate the error value.

[0046] Specifically, during the estimation process, the motor's rotor position and speed are closely related to data such as the motor's operating current and torque. Therefore, the state estimation module continuously adjusts the estimation results using a dynamic model to eliminate external interference factors in the system. By using a Kalman filter, the system dynamically corrects errors during the estimation process, eliminating the impact of high-frequency noise and improving the accuracy of the estimation results.

[0047] In one possible implementation, the estimation processing unit further combines real-time environmental data, such as measured temperature, to dynamically update the motor's electrical and mechanical model parameters. Because the motor's operating state is affected by factors such as ambient temperature and load variations, the estimation module adjusts the model parameters in real time based on these changes, further improving the accuracy of the rotor state estimation.

[0048] The estimation result output unit is responsible for outputting the preliminary estimated rotor state information to the filter fusion module. This module further optimizes the rotor state estimate based on the estimation results, improving estimation accuracy through a fusion algorithm. The filter fusion module fuses the preliminary estimation results from the state estimation module with the output of the deep learning prediction module to ultimately obtain a more accurate rotor state estimate.

[0049] In some embodiments, the state estimation module can dynamically adjust the estimation strategy during motor operation. For example, under low load or low speed operation, the motor's electrical behavior is more complex, and the estimation model weights may need to be adjusted to suit the motor's performance under specific operating conditions. Furthermore, motor aging may affect estimation accuracy. Therefore, during long-term use, the state estimation module can combine historical motor operating data to make appropriate adjustments to the dynamic model.

[0050] The deep learning prediction module is used to receive the current and voltage signals of the motor during operation, and predict the rotor position and rotor speed of the motor based on the trained neural network model, generate the prediction results and output them to the filtering and fusion module; The deep learning prediction module primarily receives the motor's current and voltage signals during operation and, based on a trained neural network model, predicts the motor's rotor position and speed. This module's key task is to input these real-time electrical signals into the deep neural network for processing, generating predictions of rotor position and speed. These predictions are then output to the filtering and fusion module for further optimization and integration.

[0051] In this embodiment, the deep learning prediction module uses the reasoning capabilities of a neural network model to convert real-time data into precise rotor state information. In this way, the system overcomes the limitations of traditional sensor-based methods and improves the accuracy and stability of motor control.

[0052] In this embodiment, the deep learning prediction module includes a signal preprocessing unit, a neural network inference unit, and a prediction output unit. The signal preprocessing unit first processes the input current and voltage signals to ensure that they can effectively serve as input to the neural network model. The preprocessed signals are then fed into the neural network inference unit, which performs inference based on the trained deep learning model and outputs predictions for rotor position and speed. Finally, these predictions are fed into the filtering and fusion module for further fusion and optimization to produce the final rotor state estimation.

[0053] Generally speaking, the current and voltage signals received by the deep learning prediction module often contain noise or interference. Directly inputting them into the neural network model may reduce the accuracy of the prediction results. Therefore, the role of the signal preprocessing unit cannot be ignored. This unit usually includes two steps: filtering and normalization. Specifically, the filtering operation removes high-frequency noise and transient interference. Common filtering methods include bandpass filtering and low-pass filtering. Normalization is to scale the original signal according to the input requirements of the deep learning model, so that the signal amplitude fits the input range of the neural network.

[0054] In some embodiments, the signal preprocessing unit performs multiple windowing operations on the current and voltage signals to extract time-domain features from the signals. This process provides more stable and reliable input data for the neural network model, thereby improving the accuracy of the model's predictions.

[0055] Alternatively, the neural network inference unit utilizes a trained neural network model to process the preprocessed signals. This neural network model typically consists of multiple hidden layers, each trained to learn the complex nonlinear relationship between the motor state and the current and voltage signals. The neural network architecture can be a multilayer perceptron (MLP), convolutional neural network (CNN), or recurrent neural network (RNN). The specific network architecture chosen depends on the specific requirements of the motor system and the characteristics of the data.

[0056] In one possible implementation, the neural network's input layer receives processed current and voltage signals. After multiple layers of computation, the network generates predictions at the output layer, including rotor position and rotor speed. The predictions can be expressed as: ; in: is the predicted rotor state; is the stator current; is the stator voltage; is the mapping function of the neural network model.

[0057] Specifically, the neural network adjusts its output to match the motor's actual rotor state based on the weights and biases learned during training. This process enables the neural network to predict rotor position and speed, especially at low motor speeds or in sensorless conditions, ensuring system stability and high accuracy.

[0058] In some embodiments, the neural network inference unit can also use an incremental learning approach, continuously collecting new current and voltage data during system operation and updating the neural network model in real time. This approach helps improve the system's adaptability to motor operating conditions, especially when the motor load changes or environmental factors have a significant impact.

[0059] The prediction output unit's primary function is to pass the rotor position and speed predictions from the neural network model to the filter fusion module. Here, these predictions are combined with the initial estimates from the state estimation module to ensure a more accurate final rotor state estimate.

[0060] Specifically, the output form of the prediction results is: ; in: is the predicted value of the rotor state; is the rotor position estimate; is the rotor speed estimate.

[0061] After the result is sent to the filtering fusion module, the fusion module will use algorithms such as the extended Kalman filter (EKF) to combine the rotor state predicted by deep learning with the results of the estimation module, and finally output the optimized rotor state estimation result.

[0062] a filtering fusion module, configured to receive the preliminary estimation result output by the state estimation module and the prediction result output by the deep learning prediction module, fuse the two, and output an optimized rotor state estimation result to the adaptive control module; The filter fusion module receives the preliminary estimation results from the state estimation module and the rotor state prediction results from the deep learning prediction module, fuses them together, and outputs the optimized rotor state estimation results to the adaptive control module. The key to this module is that by effectively fusing data from these two different sources, it improves the accuracy of the rotor state estimation, ensuring that the control module receives more stable and accurate input, thereby achieving more efficient motor control.

[0063] In this embodiment, the filtering and fusion module includes an estimation result receiving unit, a fusion processing unit, and a fusion result output unit. Through the collaboration of these submodules, the filtering and fusion module can accurately fuse the preliminary estimation results output by the state estimation module with the prediction results output by the deep learning prediction module, thereby obtaining an optimized rotor state estimation result.

[0064] Typically, the state estimation module provides a preliminary estimate of the motor's rotor position and speed through dynamic extrapolation based on electrical and mechanical models. While this estimate accurately reflects the motor's operating state, it still contains some errors, particularly at low speeds and with large load variations, where the estimation accuracy may decrease. To address this, the deep learning prediction module leverages the reasoning capabilities of neural networks to generate a prediction of the rotor state based on real-time current and voltage signals. Because deep learning models excel at handling complex nonlinear relationships, they can further improve the estimation results.

[0065] Alternatively, the filter fusion module uses an extended Kalman filter (EKF) or other state fusion algorithm to perform a weighted fusion of the initial estimation results from the state estimation module and the output of the deep learning prediction module. The extended Kalman filter is a commonly used nonlinear filtering method that dynamically adjusts and corrects system errors during the motor state estimation process, achieving higher estimation accuracy.

[0066] Specifically, the fusion processing unit first receives two sets of inputs: one is the preliminary estimation result from the state estimation module, and the other is the rotor state prediction result from the deep learning prediction module. Assume that the rotor state estimation result after fusion is , the output of the deep learning prediction module is The fusion processing unit optimizes the two results based on the Kalman filter algorithm or other similar methods. For the Kalman filter, the system fusion processing formula is usually: ; in: is the rotor state estimation result after fusion; is the output of the state estimation module; is the output of the deep learning prediction module; is the Kalman gain, which represents the weighted ratio between the estimated result and the predicted result.

[0067] The difference between preliminary estimates and fused results: In Kalman filtering applications, preliminary estimates refer to preliminary predictions obtained using traditional state estimation methods based on the system's electrical and mechanical models and real-time data. This estimate may contain certain errors because it is calculated based on the model and does not fully reflect the system's true state. The fused rotor state estimate, on the other hand, is a more accurate state estimate obtained by further integrating other data sources (such as neural network predictions and sensor feedback) through the Kalman filter.

[0068] The derivation of the weighted fusion model can be inspired by the principle of Kalman Filter. Kalman Filter is a recursive estimation method based on minimizing the error covariance. Its core idea is to perform a weighted average of the predicted value and the measured value to obtain a more accurate estimate. The fusion gain is set to Used to determine the weight of predicted information and current estimated information.

[0069] For general estimation fusion problems, when there is an error between the estimated value and the predicted value of the system, the fusion process aims to obtain the optimal fusion result by weighing the error between the two. The principle is that the fusion gain Dynamically adjust according to the size of signal noise and changes in expected error to maximize the accuracy of the final estimate.

[0070] The key assumption in deriving the formula is that and The errors of each are independent and statistically distributed. Therefore, minimizing the error between the two requires weighted fusion to minimize the estimation error after fusion.

[0071] This formula reflects how to combine the estimated value and the predicted value of the system to generate a more accurate fused state estimate. , which shows that the system dynamically adjusts the contribution of the two according to the deviation between the current estimate and the prediction. Specifically, when the difference between the predicted value and the estimated value is large, the fusion gain The weight of the predicted information will be increased; when the difference between the two is small, the weight of the estimated information will be increased.

[0072] The physical meaning of this approach can be explained as follows: by dynamically adjusting the weight between prediction and estimation, the advantages of both information sources can be comprehensively considered, uncertainty can be reduced, and system accuracy can be improved. For example, in a motor control system, current measurement estimates are subject to error, while predictions are extrapolated from historical data. Through weighted fusion, the system can suppress errors in both measurement and prediction while ensuring real-time performance and accuracy, thereby optimizing control results.

[0073] Fusion processing technology is widely used in various control systems and estimation systems, especially in methods such as Kalman filtering and the extended Kalman filter (EKF), which all use similar fusion principles. These methods combine measurements and predictions to optimize the estimated value, reducing the uncertainty caused by measurement and prediction errors.

[0074] Specifically, the Kalman gain The calculation is based on the system's current state estimation error and prediction error, ensuring the fusion result has high accuracy and stability. Through this weighted fusion process, the filter fusion module can fully consider the estimation and prediction results to obtain a more accurate rotor state estimate.

[0075] In one possible implementation, the filter fusion module can dynamically adjust the weighting strategy based on the system's operating conditions. For example, when the motor load is light or operation is stable, the prediction result can be weighted more heavily. However, when the load fluctuates significantly or the motor is unstable, the initial results of the state estimation module may be more reliable, so the weighting of the estimated result can be increased. This dynamic adjustment can be achieved by calculating the system's state error in real time, further improving estimation accuracy.

[0076] The fusion output unit is responsible for outputting the fused rotor state estimation results to the adaptive control module. This module serves as input to the subsequent control system, ensuring efficient and stable motor control during operation. Using the fused optimization results, the adaptive control module can precisely adjust the motor control strategy, providing more stable control performance, especially under conditions of fluctuating motor loads or significant external interference.

[0077] Furthermore, in some embodiments, the filter fusion module can also adaptively adjust based on real-time motor status. For example, if the motor's operating environment changes, the filter fusion module can automatically optimize the fusion algorithm based on the new data and adjust the fusion strategy for the prediction and estimation results to adapt to the new operating conditions.

[0078] an adaptive control module, configured to receive the optimized rotor state estimation result output by the filtering and fusion module, dynamically adjust the gain parameters of the motor controller according to the error between the rotor state and the target state, and output the adjusted control parameters to the control execution module; The adaptive control module receives the optimized rotor state estimation results from the filter fusion module and dynamically adjusts the motor controller gain parameters based on the error between the rotor state and the target state. By adjusting the control gain parameters in real time, the adaptive control module maintains efficient and stable motor operation under varying operating conditions, ensuring that motor performance accurately matches the target state. The adjusted control parameters are then output to the control execution module to implement the control strategy and ensure accurate and stable motor operation.

[0079] In this embodiment, the adaptive control module includes an error calculation unit, a gain adjustment unit, and a control output unit. The error calculation unit primarily calculates the error between the rotor state and the target state and provides a feedback signal to the gain adjustment unit. The gain adjustment unit dynamically adjusts the controller's gain parameters based on the error signal. Ultimately, the adjusted control parameters are transmitted to the control output unit, which in turn controls the execution module to actually control the motor.

[0080] In general, the workflow of the adaptive control module is to first receive the optimized rotor state estimation results from the filter fusion module. , the result usually includes key information such as the position and speed of the rotor. At the same time, the system will preset the target state , as the ideal state of the motor operation. The error calculation unit compares the rotor state and the target state and calculates the error between the two ,Right now: ; in, is the error between the rotor state and the target state; is the rotor state estimation result after fusion; The preset target state.

[0081] As an option, the gain adjustment unit is Dynamically adjust the gain parameters of the motor controller. Typically, the gain parameters of the motor controller include proportional gain , integral gain and differential gain These gain parameters determine the response speed and stability of the motor control system. In practical applications, the gain is usually adjusted according to the size and rate of change of the error. For example, when the error is large, it may be necessary to increase the proportional gain. To improve control accuracy; when the error changes more smoothly, it may be necessary to increase the differential gain To improve system stability.

[0082] Specifically, the gain adjustment unit is based on the error The size and change trend of can be adjusted by the following methods: Proportional gain adjustment: When the error is large, the proportional gain Will increase to speed up the system's response; Integral gain adjustment: When the error persists, the integral gain will increase, thereby reducing the steady-state error of the system; Differential gain adjustment: When the error changes rapidly, the differential gain Will increase to improve system stability.

[0083] In some embodiments, the gain adjustment unit uses a PID control algorithm (proportional-integral-derivative control algorithm) for dynamic adjustment. The controller output u(k)u(k)u(k) of the PID control algorithm is usually expressed as: ; in: is the control signal; is the error between the rotor state and the target state; is the proportional gain; is the integral gain; is the differential gain; is the sampling time interval; From time 0 to time The sum of the errors; is the rate of change of the error.

[0084] In one possible implementation, the PID controller's gain parameters are automatically optimized based on real-time operating data. Specifically, an adaptive algorithm can be used to adjust the PID controller's parameters to ensure optimal performance under varying operating conditions.

[0085] The control output unit is responsible for outputting the control parameters adjusted by the gain adjustment unit to the control execution module. The control execution module adjusts the motor's operating state based on these control signals to ensure that the rotor position and speed are as close to the target state as possible, thereby improving the motor's operating accuracy and stability.

[0086] In some embodiments, to improve control accuracy and response speed, the adaptive control module can dynamically select different control strategies based on the motor's operating environment. For example, when the motor load undergoes significant changes, a more robust control strategy, such as fuzzy control or an optimization-based control algorithm, can be used to adapt to the motor's varying operating conditions.

[0087] A control execution module, configured to receive the control parameters output by the adaptive control module and control the power-on logic and commutation timing of the motor; The control execution module receives the control parameters output by the adaptive control module and, based on these parameters, precisely controls the motor's power-on logic and commutation timing. This module's control enables the system to accurately adjust the motor's operating state, ensuring it operates as intended, thus achieving efficient and stable motor control.

[0088] In this embodiment, the control execution module includes a power control unit, a commutation control unit, and an execution output unit. The power control unit is responsible for controlling the power supply status of the motor, ensuring that each motor winding is energized as needed. The commutation control unit determines the optimal commutation timing based on control parameters and accurately controls the commutation process. The execution output unit outputs the final execution signal based on the control parameters to control the operation of the motor.

[0089] Typically, the control execution module begins by receiving control parameters from the adaptive control module. These parameters typically include key information such as the current reference value and commutation angle. Based on this information, the control execution module first calculates the power supply logic for each motor winding. For example, motor power control requires adjusting the power supply to each phase winding based on the set current reference value to ensure appropriate drive torque under different operating conditions.

[0090] Alternatively, the power control unit uses PWM (Pulse Width Modulation) to precisely energize the motor windings. By regulating the current in each phase, the motor's torque output can be precisely controlled. Specifically, the PWM control signal can be expressed as follows: ; in: is the instantaneous voltage of a phase winding; is the maximum value of the voltage; the period of time for which the current is energized; is the total time of the PWM cycle.

[0091] Specifically, the control execution module adjusts the duty cycle of the PWM signal according to the control parameters output by the adaptive control module, so that the current passes through each winding of the motor to meet the motor torque and speed requirements.

[0092] In some embodiments, the commutation control unit controls the timing of motor commutation based on the commutation angle parameters provided by the adaptive control module. Commutation timing is typically determined based on information such as the motor's speed and rotor position to determine when to switch the current direction to maximize the motor's power output and operational stability. The core of commutation control lies in accurately determining the timing of current on and off, ensuring smooth current transitions during commutation and avoiding large current pulses or imbalances.

[0093] Specifically, the commutation angle control formula is usually expressed as: ; in: is the commutation angle; is the rotor mechanical angle; is the number of pole pairs of the motor.

[0094] The commutation control unit will calculate the commutation angle at each moment in real time and send out corresponding commutation instructions through control signals to ensure that the motor operates according to the set commutation timing.

[0095] In one possible implementation, the control execution module dynamically adjusts the power-on logic and commutation timing based on the motor's operating state. For example, under low load or low speed conditions, the commutation interval may need to be extended to reduce energy consumption; while under high load or high speed conditions, the commutation frequency may need to be increased to maintain efficient motor operation.

[0096] The execution output unit is responsible for outputting the control signals calculated by the power control unit and the commutation control unit to the motor driver. Based on these control signals, the motor driver adjusts the timing and current of each phase winding of the motor, thereby achieving precise torque control and speed regulation.

[0097] The sensorless BLDC motor control method described below and the sensorless BLDC motor control system described above may refer to each other.

[0098] Please see the attached Figure 8 The present invention also provides a sensorless DC brushless motor control method, comprising the following steps: S1. Model establishment: constructing electrical and mechanical models to describe the dynamic behavior of the motor, so as to obtain a motor operation dynamic model suitable for control calculations; S2, state estimation, uses the operation dynamic model and combines the real-time current and voltage signals collected during the operation of the motor to make a preliminary estimate of the rotor position and rotor speed of the motor; S3: Prediction and inference: Input the collected current and voltage signals into the trained deep learning model to predict the current rotor position and rotor speed of the motor; S4, fusion optimization, inputting the preliminary estimation result and the prediction result into the extended Kalman filter, and outputting the optimized rotor state estimation result through filter fusion processing; S5, control parameter adjustment, based on the optimized rotor state estimation result, compares the deviation between the current state and the preset target state, and then dynamically adjusts the gain parameters of the motor controller through the adaptive control strategy; S6, control execution, generates a commutation control signal according to the adjusted control parameters, and then controls the power-on logic of each phase of the motor to achieve stable startup and precise operation in the low-speed stage; S7, continuous parameter optimization, based on the defined objective function, the rotor position and speed errors in the motor control process are evaluated in real time, and the controller parameters are continuously optimized in combination with the optimization algorithm.

[0099] The method of this embodiment can be used to execute the above system embodiment. Its principles and technical effects are similar and will not be described in detail here.

[0100] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. Sensorless DC brushless motor control system, characterized in that, include: A motor modeling module, which establishes a dynamic operation model of the motor based on a preset electrical model and a mechanical model, and outputs the dynamic operation model to the state estimation module; A state estimation module is used to receive the operation dynamic model output by the motor modeling module and the motor operation parameters collected in real time, make a preliminary estimate of the rotor position and rotor speed of the motor, and output the preliminary estimation results to the filtering and fusion module; The deep learning prediction module is used to receive the current and voltage signals of the motor during operation, and predict the rotor position and rotor speed of the motor based on the trained neural network model, generate the prediction results and output them to the filtering and fusion module; a filtering fusion module, configured to receive the preliminary estimation result output by the state estimation module and the prediction result output by the deep learning prediction module, fuse the two, and output an optimized rotor state estimation result to the adaptive control module; an adaptive control module, configured to receive the optimized rotor state estimation result output by the filtering and fusion module, dynamically adjust the gain parameters of the motor controller according to the error between the rotor state and the target state, and output the adjusted control parameters to the control execution module; The control execution module is used to receive the control parameters output by the adaptive control module and control the power-on logic and commutation timing of the motor.

2. The sensorless DC brushless motor control system according to claim 1, characterized in that: The motor modeling module includes: An electrical modeling unit, used to establish an electrical characteristic model of the stator side based on the inductance, resistance and electromotive force parameters of the motor; A mechanical modeling unit is used to establish a mechanical dynamic model describing the relationship between the rotor angular velocity and position based on the rotor inertia, load torque and damping coefficient of the motor; The model output unit is used to integrate the established electrical model and mechanical model into a joint dynamic model and transmit the model to the state estimation module.

3. The sensorless DC brushless motor control system according to claim 1, characterized in that: The state estimation module includes: Data acquisition unit, used to collect the current, voltage and temperature parameters of the motor in real time during operation; an estimation processing unit, configured to estimate the current rotor position and rotor speed of the motor by combining the operating dynamic model with the collected data; The estimation result output unit is used to output the preliminary estimated rotor state to the filtering fusion module.

4. The sensorless DC brushless motor control system according to claim 1, characterized in that: The deep learning prediction module includes: A signal preprocessing unit, used for performing preprocessing operations such as normalization and filtering on the input current signal and voltage signal; The neural network inference unit is used to load and run the trained neural network model and output the corresponding rotor position and rotor speed prediction values; The prediction output unit is used to output the prediction value to the filtering fusion module.

5. The sensorless DC brushless motor control system according to claim 1, characterized in that: The filtering fusion module includes: An estimation result receiving unit, configured to receive the preliminary estimation result from the state estimation module and the prediction result from the deep learning prediction module; A fusion processing unit, configured to perform joint optimization processing on two sets of rotor state data based on an extended Kalman filter algorithm or other state fusion algorithms; The fusion result output unit is used to output the optimized rotor state estimation result to the adaptive control module.

6. The sensorless DC brushless motor control system according to claim 1, characterized in that: The adaptive control module includes: an error calculation unit, for calculating the error between the current rotor state and a preset target state; a parameter adjustment unit, configured to adaptively adjust proportional, integral, and differential parameters of the controller based on the error result; The control parameter output unit is used to transmit the adjusted controller parameters to the control execution module.

7. The sensorless DC brushless motor control system according to claim 1, characterized in that: The control execution module includes: A control signal generating unit, configured to generate a commutation control signal and a PWM modulation signal according to received controller parameters; The drive instruction output unit is used to convert the control signal into a specific power-on logic control instruction for each phase winding of the motor, so as to realize low-speed sensorless starting and operation control of the motor.

8. The sensorless DC brushless motor control system according to claim 2, characterized in that: The model output unit includes: A model integration unit is used to perform time domain alignment and parameter matching on the modeling data generated by the electrical modeling unit and the mechanical modeling unit, and integrate them into a joint dynamic model; A structured output unit for formatting the joint dynamic model into a data structure adapted to the computational framework of the state estimation module; The dynamic parameter updating unit is used to update the modeling parameters according to the real-time factors of the ambient temperature and the motor aging degree during the operation of the motor, and re-output the dynamic model.

9. The sensorless DC brushless motor control system according to claim 3, characterized in that: The signal preprocessing unit includes: The filtering and denoising unit is used to perform band-pass filtering and average filtering on the collected current and voltage signals to eliminate high-frequency noise and instantaneous interference components; The normalization processing unit is used to scale and offset the original signal according to the input specifications of the deep learning model to make it conform to the network input format; The data cache unit is used to cache and organize the continuous sampling data in time windows so that the neural network inference unit can perform batch input and feature extraction operations.

10. A sensorless brushless DC motor control method, according to the sensorless brushless DC motor control system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Model building: constructing electrical and mechanical models to describe the dynamic behavior of the motor, in order to obtain a dynamic model of the motor suitable for control calculations; State estimation, using the dynamic operation model and the real-time current and voltage signals collected during the operation of the motor to make a preliminary estimate of the rotor position and rotor speed of the motor; Predictive inference: The collected current and voltage signals are input into a trained deep learning model to predict the motor's current rotor position and speed. Fusion optimization: inputting the preliminary estimation result and the prediction result into the extended Kalman filter, and outputting the optimized rotor state estimation result through filter fusion processing; Control parameter adjustment, based on the optimized rotor state estimation results, compares the deviation between the current state and the preset target state, and then dynamically adjusts the gain parameters of the motor controller through an adaptive control strategy; Control execution, generating commutation control signals according to the adjusted control parameters, and then controlling the power-on logic of each phase of the motor to achieve stable startup and precise operation in the low-speed stage; Parameters are continuously optimized. The rotor position and speed errors during the motor control process are evaluated in real time based on the defined objective function, and the controller parameters are continuously optimized in combination with the optimization algorithm.

Images