Embedded system and method for self-tuning BLDC drive parameters for multiple device types

Abstract

The application discloses a multi-device type-oriented BLDC driving parameter self-tuning embedded system and method, and the method comprises the following steps: constructing a BLDC unified mathematical model comprising an electromagnetic torque equation, a back electromotive force equation and a mechanical motion equation, and establishing a nonlinear mapping relationship library of a motor type identification code and a key electrical parameter set; injecting a pre-defined multi-mode excitation signal sequence into a BLDC motor to be driven, and synchronously collecting dynamic response data, wherein the dynamic response data is used for characterizing the features of the key electrical parameter set; based on the collected dynamic response data, the equivalent inductance and the equivalent resistance in the key electrical parameter set are extracted by decoupling through a frequency domain impedance analysis method, and the back electromotive force coefficient and the rotational inertia in the key electrical parameter set are identified by using a time domain back electromotive force zero-crossing detection combined with a rotational speed differential operation. The application is based on a motor fingerprint feature library and a fuzzy adaptive architecture, realizes multi-model debugging-free plug-and-play, and maintains stable control performance under dynamic working conditions.

Description

Embedded System and Method for Self-Tuning BLDC Driver Parameters for Multiple Device Types Technical Field

[0001] This application relates to the field of embedded driver control technology, and in particular to an embedded system and method for self-tuning BLDC driver parameters for multiple device types. Background Technology

[0002] With the deep penetration of brushless DC motors into advanced manufacturing and intelligent equipment, their drive systems face increasingly complex and diversified challenges in application scenarios. From precision medical instruments to industrial robot joints, motors of different specifications and from different manufacturers exhibit a wide variety of electrical characteristics. Current drive technologies rely on pre-stored fixed parameters or manual parameter tuning based on experience, which exposes fundamental limitations when dealing with the need for rapid adaptation to multiple models.

[0003] Firstly, due to differences in electromagnetic design and manufacturing processes, different models of motors with the same power rating exhibit vastly different dynamic response characteristics during actual operation. Existing solutions require time-consuming and specialized debugging for each type of motor, which not only significantly extends the equipment deployment cycle but also makes it difficult to accurately capture the nonlinear coupling relationship between winding parameters and mechanical characteristics through manual parameter tuning. This leads to frequent loss of control phenomena such as torque oscillation and commutation step loss when switching motor models within the same series of drives, severely restricting the flexibility of multi-device collaborative operation systems.

[0004] Secondly, under complex operating conditions such as temperature fluctuations and sudden load changes, the equivalent parameters of the motor exhibit time-varying drift characteristics. Traditional fixed-parameter control strategies cannot track such changes in real time, leading to a chain reaction of problems such as increased speed response overshoot and deterioration of steady-state accuracy. Especially in long-term continuous operation scenarios, the cumulative errors caused by parameter mismatch often force the system to shut down and readjust, reducing equipment utilization and creating potential safety hazards, becoming a technical hurdle that is difficult to overcome for high-reliability applications. Summary of the Invention

[0005] To address the aforementioned problems, embodiments of the present invention provide an embedded method for self-tuning BLDC driver parameters for multiple device types, the method comprising:

[0006] A unified mathematical model of BLDC, including electromagnetic torque equation, back electromotive force equation and mechanical motion equation, is constructed, and a nonlinear mapping relationship library between motor type identification code and key electrical parameter set is established.

[0007] A predefined multi-mode excitation signal sequence is injected into the BLDC motor to be driven, and dynamic response data is collected synchronously. The dynamic response data is used to characterize the features of the key electrical parameter set.

[0008] Based on the collected dynamic response data, the equivalent inductance and equivalent resistance of the key electrical parameter set are extracted by frequency domain impedance analysis. At the same time, the back EMF coefficient and moment of inertia of the key electrical parameter set are identified by using time domain back EMF zero-crossing detection combined with rotational speed differential calculation.

[0009] The key electrical parameter set obtained by identification is input into the preset fuzzy adaptive tuner to generate the initial PID control parameter set. The control performance is verified by closed-loop speed response test. The PID control parameter set is dynamically optimized according to the speed overshoot and steady-state fluctuation rate.

[0010] The optimized PID control parameter set is bound to the motor type identification code, stored in the device parameter knowledge base of the embedded FLASH memory, and a credibility tag is attached.

[0011] Furthermore, methods for constructing nonlinear mapping relation libraries include:

[0012] The no-load starting current waveform of the motor is input into a waveform encoder to generate a feature vector; the feature vector is processed by a lightweight convolutional neural network; the output of the lightweight convolutional neural network is used as the initial value for the iteration of key electrical parameter identification.

[0013] Furthermore, the method for generating multi-mode excitation signal sequences includes:

[0014] The basic excitation template is loaded according to the motor type identification code; the gradient parameters of the basic excitation template are optimized using a swarm intelligence algorithm; and the optimized gradient parameters are injected into the dynamic response data acquisition stage.

[0015] Furthermore, frequency domain impedance analysis includes:

[0016] Wavelet packet decomposition is performed on the collected swept frequency response current; a neural differential equation solver is constructed based on the decomposition results; the neural differential equation solver outputs equivalent circuit parameters to the fuzzy adaptive tuner.

[0017] Furthermore, real-time temperature compensation methods include:

[0018] The temperature of the power module is collected as the input for parameter decoupling; the resistance drift compensation is generated through the heat transfer state space model; and the compensated resistance value is input into the optimization stage of the PID control parameter group.

[0019] Furthermore, the methods for generating credibility tags include:

[0020] Receive performance index data from the closed-loop speed response test verification output; process the performance index through a multi-criteria evaluation tree; set up a tag state machine based on the evaluation results and store it in the device parameter knowledge base.

[0021] Furthermore, the fault circuit interruption mechanism includes:

[0022] Monitor overcurrent and stall events in S4 closed-loop verification; trigger security parameter group switching and generate key parameter feature hashes; update stored trust labels according to event type.

[0023] Furthermore, the parameter iterative optimization function includes:

[0024] When the stored credibility label meets the conditions, the historical parameter set is read; an incremental excitation is generated based on the historical parameter set and injected into the dynamic response data acquisition stage; the moment of inertia is updated by decoupling and the PID integral term is readjusted.

[0025] Furthermore, the method also includes:

[0026] The optimized PID control parameter set is bound to the motor type identification code and stored in the device parameter knowledge base of the embedded FLASH memory. A confidence tag is attached. When the confidence tag is high confidence, a dynamic load disturbance sequence is injected and dynamic performance indicators are collected. If the dynamic performance indicators reach the limit condition, the confidence tag is downgraded to steady-state high confidence and the applicable operating condition range is marked. The marked confidence tag and the applicable operating condition range are associated and stored in the device parameter knowledge base.

[0027] An embedded system for self-tuning BLDC driver parameters for multiple device types, the system includes:

[0028] The core model building module constructs a unified mathematical model of BLDC that includes electromagnetic torque equation, back electromotive force equation and mechanical motion equation, and establishes a nonlinear mapping relationship library between motor type identification code and key electrical parameter set;

[0029] An excitation injection and acquisition module injects a predefined multi-mode excitation signal sequence into the BLDC motor to be driven and simultaneously acquires dynamic response data, which is used to characterize the features of a set of key electrical parameters.

[0030] The parameter decoupling identification module, based on the collected dynamic response data, decouples and extracts the equivalent inductance and equivalent resistance of the key electrical parameter set through frequency domain impedance analysis. At the same time, it uses time domain back EMF zero-crossing detection combined with rotational speed differential calculation to identify the back EMF coefficient and rotational inertia of the key electrical parameter set.

[0031] The parameter self-tuning module inputs the identified key electrical parameter set to a preset fuzzy adaptive tuner to generate an initial PID control parameter set. The control performance is verified through closed-loop speed response testing, and the PID control parameter set is dynamically optimized based on speed overshoot and steady-state fluctuation rate.

[0032] The knowledge base storage module binds the optimized PID control parameter group with the motor type identification code, stores it in the device parameter knowledge base of the embedded FLASH memory, and attaches a credibility tag.

[0033] The technical effects and advantages of the embedded method for self-tuning BLDC driver parameters for multiple device types provided by this invention are as follows:

[0034] This invention, based on a motor fingerprint feature library and a fuzzy adaptive architecture, enables plug-and-play operation across multiple motor models without debugging, while maintaining stable control performance under dynamic conditions. It automatically matches a nonlinear parameter mapping library to the motor's unique identifier, adapting to motors with different electromagnetic characteristics without manual intervention. Combining wavelet packet frequency domain analysis and time-domain inertia identification, it accurately constructs a dynamic motor model, avoiding torque fluctuations caused by parameter mismatches in traditional solutions. A fuzzy rule-driven PID parameter real-time tuning mechanism autonomously adjusts control parameters in scenarios of sudden load changes and temperature drift. Based on a heat transfer state-space model, it pre-compensates for winding resistance drift, significantly suppressing speed overshoot and steady-state fluctuations. A multi-criteria parameter reliability verification mechanism automatically eliminates abnormal identification results, triggering safety parameter circuit breaker switching. Lightweight neural network deployment achieves microsecond-level dynamic response, ensuring efficient execution of complex algorithms on resource-constrained controllers. Attached Figure Description

[0035] Figure 1 is a flowchart of the embedded method for self-tuning BLDC drive parameters for multiple device types in Embodiment 1.

[0036] Figure 2 is a flowchart of the embedded method for self-tuning BLDC drive parameters for multiple device types in Embodiment 2;

[0037] Figure 3 is a schematic diagram of the connection of the embedded system for BLDC driver parameter self-tuning for multiple device types in Embodiment 3. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1

[0040] Please refer to Figure 1. An embodiment of the present invention provides an embedded method for self-tuning BLDC driver parameters for multiple device types. The method includes:

[0041] S1. Construct a unified mathematical model of BLDC that includes electromagnetic torque equation, back electromotive force equation and mechanical motion equation, and establish a nonlinear mapping relationship library between motor type identification code and key electrical parameter set;

[0042] S2. Inject a predefined multi-mode excitation signal sequence into the BLDC motor to be driven, and simultaneously collect dynamic response data. The dynamic response data is used to characterize the features of the key electrical parameter set.

[0043] S3. Based on the collected dynamic response data, the equivalent inductance and equivalent resistance of the key electrical parameter set are extracted by frequency domain impedance analysis. At the same time, the back EMF coefficient and moment of inertia of the key electrical parameter set are identified by using time domain back EMF zero-crossing detection combined with rotational speed differential calculation.

[0044] S4. Input the identified key electrical parameter set into the preset fuzzy adaptive tuner to generate the initial PID control parameter set. Verify the control performance through closed-loop speed response test. Dynamically optimize the PID control parameter set based on speed overshoot and steady-state fluctuation rate.

[0045] S5. Bind the optimized PID control parameter group with the motor type identification code, store it in the device parameter knowledge base of the embedded FLASH memory, and attach a credibility tag.

[0046] S1-S5 are completed automatically in a single embedded microcontroller without the need for external debugging equipment.

[0047] The set of key electrical parameters includes equivalent inductance, equivalent resistance, back electromotive force coefficient, and moment of inertia.

[0048] The dynamic response data includes three-phase current waveforms, rotor position signals, sweep frequency response current, and DC bus voltage.

[0049] The methods for constructing the nonlinear mapping relation library in S1 include:

[0050] Input the no-load starting current waveform of the motor into a waveform encoder to generate a feature vector;

[0051] When a new type of motor is first connected to the system, the controller controls the motor to perform a no-load start operation (i.e., the motor shaft is not connected to a mechanical load) and synchronously collects three-phase current waveforms (one of the dynamic response data). The current waveform during the start-up phase is input into a waveform encoder for feature extraction, including: the waveform encoder generates a compact feature vector by analyzing the geometric characteristics of the current waveform (such as peak amplitude, rising edge slope, oscillation frequency, etc.). For example, a certain type of motor may exhibit a typical "double-peak oscillation" current waveform when starting under no-load conditions, and the encoder will convert it into a 128-dimensional floating-point vector.

[0052] The feature vector is processed using a lightweight convolutional neural network;

[0053] This feature vector is then processed by a pre-built lightweight convolutional neural network (CNN). This CNN model is trained and optimized offline, and its network structure is designed for embedded resource constraints (e.g., containing only 3 convolutional layers and 1 fully connected layer), allowing it to run in real time on a microcontroller. The network output is directly associated with the estimated values ​​of a key electrical parameter set (equivalent inductance, equivalent resistance, back electromotive force coefficient, and moment of inertia). Taking equivalent inductance identification as an example: when the starting current characteristics of a flat motor are input, the CNN can output its initial estimated equivalent inductance (e.g., 0.85mH), while traditional methods require a complex iterative process.

[0054] The output of the lightweight convolutional neural network is used as the initial value for the S3 parameter identification iteration.

[0055] This estimated parameter set will serve as the initial values ​​for the frequency domain impedance analysis and time domain back EMF zero-crossing detection in step S3. Compared to random initialization, this method significantly improves the efficiency of parameter identification.

[0056] In the subsequent S3-S5 processes, the system performs refined parameter identification based on the initial value. The final optimized PID parameter set will be stored in the equipment parameter knowledge base along with the motor type identification code (such as "EC_Flat_48V"). When the same model of motor is connected again, the system can directly call the historical parameter set to achieve a rapid deployment effect of learning once and reusing multiple times.

[0057] The multi-mode excitation signal sequence in S2 is generated in the following way:

[0058] Load the basic excitation template according to the motor type identifier code defined in S1;

[0059] The gradient parameters of the basic excitation template are optimized using a swarm intelligence algorithm;

[0060] The optimized sequence is injected into the S2 response acquisition stage.

[0061] In step S2, the multi-mode excitation signal sequence is generated in the following way:

[0062] When the system detects a new type of motor being connected (by reading the motor nameplate QR code or by manual input by the user), it first extracts the motor type identification code (such as "EC_Disk_24V"). ​​The controller loads the corresponding basic excitation template from the device parameter knowledge base (the FLASH database established by S5) based on this identification code. This is a set of predefined PWM waveform pattern combinations, which includes excitation units with different duty cycles, frequencies and commutation timings. For example, for a disc motor, the basic excitation template may include two core modes: stepped boost start and sinusoidal sweep frequency.

[0063] To improve parameter identification accuracy, the system employs a swarm intelligence algorithm (Particle Swarm Optimization, PSO, is selected in this embodiment) to dynamically optimize the gradient parameters of the basic excitation template, specifically including:

[0064] Define the key change characteristics of the excitation signal, including:

[0065] Voltage ramp-up rate (e.g., from 5% to 95% of rated voltage in 0.1s); frequency switching step size (e.g., incremental interval from 50Hz to 200Hz); mode transition time.

[0066] The optimization process includes: constructing an objective function using the parameter estimates output by the CNN in S1; running PSO online in the embedded system to generate 20 sets of gradient parameter combinations; and evaluating the signal-to-noise ratio and spectral richness of the three-phase current waveform.

[0067] Real-time adjustment: When current sampling saturation is detected, the voltage ramp-up rate parameter is automatically reduced.

[0068] The optimized excitation sequence is injected into the motor through the three-phase inverter, and dynamic response data acquisition (three-phase current, rotor position and bus voltage) is triggered at the same time. This process is executed in a closed loop within the microcontroller. The final optimized excitation sequence will be bound to the motor type identification code and stored. When the same type of motor is connected again, the historical best sequence can be directly called to avoid repeated optimization calculations.

[0069] The frequency domain impedance analysis method in S3 includes:

[0070] Perform wavelet packet decomposition on the swept frequency response current acquired by S2;

[0071] A neural differential equation solver is constructed based on the decomposition results;

[0072] The neural differential equation solver outputs equivalent circuit parameters to the fuzzy adaptive tuner.

[0073] When performing the frequency domain impedance analysis in step S3, the system first preprocesses the three-phase current waveforms acquired in S2 under frequency sweep excitation. Taking a 12-pole BLDC motor as an example, when the microcontroller injects a linear frequency sweep voltage signal of 20Hz-2kHz, the stator current response data acquired synchronously is input into the wavelet packet decomposition module. Here, the db4 wavelet basis function is used to perform 3-level decomposition, dividing the current signal into 8 characteristic frequency bands in the frequency domain, effectively separating the winding fundamental response, switching noise and harmonic interference components.

[0074] The resulting frequency band energy feature matrix after separation is fed into a neural differential equation solver. This solver employs a lightweight, fully connected network structure (e.g., 8 nodes in the input layer, 5 nodes in the hidden layer, and 2 nodes in the output layer). Its core function is to establish a mapping model between the frequency domain response and the equivalent circuit parameters by learning the differential constraint relationships of the electromagnetic system. Specifically, this includes:

[0075] Input: Normalized energy values ​​for each frequency band, corresponding frequency points, and bus voltage;

[0076] Output: Equivalent inductance L eq (Unit: mH) and equivalent resistance R eq (Unit: Ω)

[0077] Training mechanism: The network is trained based on an offline-generated synthetic dataset (covering frequency response simulation data of various inductor and resistor combinations).

[0078] The equivalent circuit parameters output by the solver are then passed to the fuzzy adaptive tuner in S4, for example, when R is detected. eq =0.8Ω, L eq When the value is 2.1mH (example value), the fuzzy adaptive tuner will combine the rotational inertia identification results to automatically generate the initial value range of the proportional gain Kp, providing key electrical constraints for subsequent PID parameter optimization. The entire process runs in a closed loop within the microcontroller, without the need to export intermediate data for frequency domain analysis.

[0079] A fuzzy adaptive tuner is an intelligent controller that combines fuzzy logic control and adaptive control theory. It is specifically designed to automatically adjust and optimize the parameters of a PID controller—namely, the proportional gain, integral gain, and derivative gain—when the system model is inaccurate, nonlinear, or parameter changes occur. Its working principle includes:

[0080] S101. Map the precise input parameters (equivalent inductance L, equivalent resistance R, back EMF coefficient ke, moment of inertia J) to predefined fuzzy linguistic variables, such as "small inductance", "medium back EMF coefficient" and "large moment of inertia". Each linguistic variable is defined by a membership function, which indicates the degree to which the precise value belongs to a certain fuzzy concept.

[0081] S102. Set the fuzzy rule base. The fuzzy rule base is the core knowledge base of the fuzzy adaptive tuner. It contains a series of IF-THEN rules based on expert experience or system knowledge. The IF-THEN rules describe what kind of PID parameter combination (THEN part) should correspond to different combinations of motor parameters (IF part).

[0082] Example rule 1:

[0083] IF (Inductance L is large) AND (Moment of inertia J is medium) THEN (Proportional gain K) p It should be medium to high), (integral gain K) i It should be low), (differential gain K) d It should be (in Chinese).

[0084] Example rule 2:

[0085] IF (resistance R is large) AND (back electromotive force Ke is small) THEN (K) p It should be high), (K) p It should be high), (K) i It should be medium to high), (K) d It should be small.

[0086] S103. Evaluate all rules related to the current input state (fuzzy parameters). For each rule, calculate the activation strength of the IF part, typically using an AND operation (min) or an OR operation (max). For example, for example rule 1, the activation strength = min(L_large, J_medium). Based on the activation strength, crop or scale the output fuzzy set (K_min) corresponding to the THEN part. p Medium to high, K i Low, K d In China, common methods include minimum operation or product operation; the output fuzzy set of all activated rules (i.e., the cropped or scaled K) is used. p K i K d The fuzzy sets are combined to form each output variable (K). p K i K d The total output fuzzy set is often aggregated using the maximum value operation.

[0087] S104. Convert the total output fuzzy set (a fuzzy, irregularly shaped region) obtained in step S103 back to a precise value as the initial suggested value for the PID parameters. Common methods include: centroid method, maximum value averaging method, and maximum value median method.

[0088] S104. By combining the initial suggested value with the moment of inertia J and other factors (such as the desired dynamic response speed and disturbance rejection requirements), the fuzzy adaptive tuner will define a reasonable range around the initial suggested value. This range serves as a key electrical constraint and is passed to the subsequent PID parameter optimization algorithm.

[0089] The real-time temperature compensation methods for S3 include:

[0090] The temperature of the power module is collected as the input for S3 parameter decoupling;

[0091] The resistance drift compensation is generated using a heat transfer state-space model.

[0092] The compensated resistance value is input into the S4 control parameter optimization stage.

[0093] During the parameter identification process in step S3, the system simultaneously performs real-time temperature compensation to eliminate the impact of power device heating on electrical parameters. Specific operations include:

[0094] Temperature signals from the power module (e.g., IGBT substrate temperature) are synchronously acquired via the microcontroller's analog input channel. These temperature signals, along with dynamic response data (three-phase current, rotor position, DC bus voltage), are acquired on the same time base. The acquired temperature data is input into the heat transfer state-space model, which describes the process of heat transfer from the power module to the motor windings, considering heat capacity and thermal resistance effects (e.g., junction-to-casing thermal resistance). Its output is an estimate of the winding temperature rise. The state-space model is essentially a mathematical abstraction of the thermal network, dynamically updating the temperature of each node through state equations. The dynamic equations employ a third-order model (junction → substrate → winding), including:

[0095] ;

[0096] In the formula, For state vectors, The updated state vector, For power device junction temperature, This refers to the temperature of the substrate or casing. For the motor winding temperature, This represents the total power device losses. The thermal resistance of the junction to the substrate. The thermal resistance from the substrate to the winding. , and The heat capacity of the junction, substrate and winding.

[0097] Extract from the state vector Calculate the difference between the ambient temperature and the winding temperature, which is the estimated value of the winding temperature rise.

[0098] Based on this winding temperature rise estimate, the system calculates the resistance drift compensation. For example, when the IGBT substrate temperature is detected to rise from room temperature (25°C) to operating temperature (70°C), the heat transfer state-space model may estimate the winding temperature rise to be 45°C. Then, based on the temperature coefficient of resistance of the conductor material (e.g., copper wire approximately 0.00393 / °C), the resistance compensation coefficient α is calculated (α≈1+0.00393*45≈1.177). At this point, if the equivalent resistance R calculated by the frequency domain impedance analysis method is 0.82Ω, then the compensated resistance value R... compThat is: R comp =α*R≈0.82Ω*1.177≈0.96Ω.

[0099] The equivalent resistance value R after compensation comp Together with the equivalent inductance L obtained from frequency domain analysis, and the back EMF coefficient Ke and moment of inertia J identified through time-domain back EMF zero-crossing detection and speed micro-resolution, these constitute the temperature-corrected set of key electrical parameters. This complete parameter set is then passed to the fuzzy adaptive tuner in step S4 to generate the initial PID control parameter set. The accurate correction of the resistance value ensures the rationality of the current loop gain calculation and avoids the increase in resistance caused by temperature rise being misidentified as other electrical characteristic changes, thereby improving the accuracy of subsequent PID parameter tuning and the stability of the control system. Finally, when the optimized PID parameter set is stored in the equipment parameter knowledge base along with the motor type identification code, its corresponding confidence label is also improved due to the introduction of temperature compensation.

[0100] The credibility tag generation methods in S5 include:

[0101] Receive performance metrics data output from the S4 closed-loop verification;

[0102] The performance metrics are processed using a multi-criteria evaluation tree;

[0103] The tag state machine is set up based on the evaluation results and stored in the S5 device parameter knowledge base.

[0104] During the parameter knowledge base storage process in step S5, the system dynamically generates credibility labels through a multi-criteria evaluation tree. The specific process includes: after the fuzzy adaptive tuner in S4 outputs the optimized PID parameter group, the system immediately performs verification operation in closed-loop control mode (e.g., the motor accelerates to the rated speed of 3000 rpm under no-load and suddenly adds 50% load). The performance index data collected during this process (including the custom torque fluctuation coefficient, speed overshoot, and steady-state tracking error) is transmitted to the evaluation module in real time.

[0105] The evaluation module uses a tree-like decision structure to process performance index data, including:

[0106] Primary node verification of basic stability (e.g., speed overshoot <5%)

[0107] Secondary node dynamic response analysis (e.g., torque fluctuation coefficient ≤ 0.03 N·m);

[0108] The final node is used to comprehensively evaluate steady-state accuracy (e.g., position tracking error <0.1°).

[0109] Based on the above evaluation results, the tag state machine is updated, including:

[0110] If all indicators meet the criteria (e.g., overshoot 2.3% / volatility coefficient 0.028 N·m / error 0.08°), the state machine switches to a high-confidence state;

[0111] If a single indicator slightly exceeds the limit (such as a fluctuation coefficient of 0.035 N·m), it will be downgraded to a pending verification status.

[0112] When a key indicator is abnormal (such as overshoot > 10%), a failure state is triggered and parameter backtracking is initiated.

[0113] Finally, the motor type identification code, optimized PID parameter group, and corresponding confidence label (e.g., BLDC-12P-48V: K) are included. p =0.85, K i =12, K d =0.02|Confidence: High) is associated and stored in the device parameter knowledge base. When the system calls this parameter group in the future, the confidence label will be used as the priority sorting basis. For example, when matching the same motor model, the parameter set with high confidence will be loaded first to improve control robustness.

[0114] The fault circuit interruption mechanism of S4 includes:

[0115] Monitor overcurrent and stall events during S4 closed-loop verification;

[0116] Trigger a switch in the security parameter group and generate a key parameter feature hash;

[0117] Update the trustworthiness label of S5 storage based on the event type.

[0118] During the closed-loop verification in step S4, the system synchronously activates the fault circuit interruption mechanism, which continuously monitors abnormal events in the control loop, including:

[0119] By analyzing the defined dynamic response data (three-phase current, rotor position) in real time, overcurrent (e.g., instantaneous phase current > 30A) or stall (e.g., speed < 5rpm within 500ms) events can be identified.

[0120] If an overcurrent event is detected (such as a peak current of 35A in phase U), immediately disconnect the currently verified PID parameter group and switch to the pre-stored safety parameter group (such as a conservative parameter that reduces the current loop proportional gain by 60%).

[0121] While triggering the switch, extract the key parameter feature hash of the current operating condition, that is, combine and encode the equivalent resistance, inductance and back EMF coefficient at this moment. For example, equivalent resistance = 0.95Ω | inductance = 1.2mH | back EMF coefficient = 0.05V / rpm, that is, hash value 0x8E3A.

[0122] The incident handling results will be fed back to the S5 device parameter knowledge base, including:

[0123] If the circuit breaker is triggered due to overcurrent or stall (e.g., a stall event lasting 800ms), the system will automatically downgrade the confidence label of the corresponding PID parameter group to failure.

[0124] The feature hash value is stored in association with the event type (overcurrent, stalled rotor, and voltage anomaly) and timestamp.

[0125] When the same motor model calls this parameter group again, the system prioritizes comparing the key parameter feature hash (such as the current operating condition hash value 0x8E3A matching the historical failure record) and directly skips this parameter group to prevent repeated failures.

[0126] The parameter iterative optimization function includes:

[0127] When the credibility tags stored in S5 meet the conditions, the historical parameter set is read.

[0128] Incremental excitation is generated based on historical parameter sets and injected into the S2 acquisition stage;

[0129] The moment of inertia is updated by decoupling in S3 and the PID integral term in S4 is readjusted.

[0130] During the continuous operation of the parameter knowledge base, when multiple credibility tags for a certain motor model are detected to be in a state to be verified, the system automatically triggers a parameter iterative optimization loop. Taking a certain brushless DC motor (identification code BLDC-48V-200W) as an example:

[0131] The system reads the three most recent historical parameter sets for this model from S5 and analyzes their differences (e.g., the integral term K). i (fluctuating within the range of 12 to 18)

[0132] Based on the historical parameter dispersion, a controlled incremental excitation is injected during the S2 dynamic response acquisition stage, that is, a step disturbance of ±4% of the rated torque is superimposed on the torque command (the safety boundary is constrained by the fuse mechanism to ensure that the current does not exceed the safety threshold).

[0133] The step response waveform is analyzed by S3 time-frequency domain hybrid decoupling, and the rotational inertia identification value is updated (e.g., corrected from the initial 0.0015 kg·m² to 0.0018 kg·m²).

[0134] The PID integral gain of S4 is automatically readjusted based on changes in moment of inertia, including:

[0135] In the formula, For the new integral gain, For the original integral gain, For the updated moment of inertia, For the original moment of inertia, the original parameter set β has Ki=15 → new value = 15× If the torque fluctuation coefficient decreases from 0.028 N·m to 0.025 N·m (a decrease of >10%), the new parameter set's reliability label is upgraded to high reliability; otherwise, it reverts to the historically optimal parameter set α, and iterative optimization for this model is paused. The effective parameter set formed in this process will be continuously accumulated in the equipment parameter knowledge base, driving the autonomous evolution of control parameters.

[0136] Example 2

[0137] As shown in Figure 2, this embodiment further improves upon the design of Embodiment 1. The difference is that, in actual operation, Embodiment 1 revealed insufficient adaptability of the reliability label to transient conditions, leading to instability risks in the high-reliability parameter group during dynamic load switching. Therefore, the embedded method for self-tuning BLDC drive parameters for multiple device types further includes:

[0138] S5: Bind the optimized PID control parameter group with the motor type identification code, store it in the device parameter knowledge base of the embedded FLASH memory, and attach a confidence tag. When the confidence tag is high confidence, inject a dynamic load disturbance sequence and collect dynamic performance indicators. If the dynamic performance indicators reach the limit condition, downgrade the confidence tag to steady-state high confidence and mark the applicable operating condition range. Associate the marked confidence tag with the applicable operating condition range and store it in the device parameter knowledge base.

[0139] Dynamic performance indicators include speed fluctuation rate and torque response delay time; speed fluctuation rate is the maximum relative deviation of speed from steady state value when load changes suddenly, and torque response delay is the lag time from torque command change to actual torque reaching the target; over-limit conditions include speed fluctuation rate exceeding the first threshold (preset) and torque response delay exceeding the second threshold (preset). If either of the above two over-limit conditions is met, the credibility label will be downgraded.

[0140] Example 3

[0141] As shown in Figure 3, based on the same inventive concept as the BLDC driver parameter self-tuning embedded method for multiple device types in the aforementioned embodiments, this application provides a BLDC driver parameter self-tuning embedded system for multiple device types. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0142] The core model building module constructs a unified mathematical model of BLDC that includes electromagnetic torque equation, back electromotive force equation and mechanical motion equation, and establishes a nonlinear mapping relationship library between motor type identification code and key electrical parameter set;

[0143] The excitation injection and acquisition module injects a predefined multi-mode excitation signal sequence into the BLDC motor to be driven and simultaneously acquires dynamic response data, which is used to characterize the features of the key electrical parameter set.

[0144] The parameter decoupling identification module, based on the collected dynamic response data, decouples and extracts the equivalent inductance and equivalent resistance of the key electrical parameter set through frequency domain impedance analysis. At the same time, it uses time domain back EMF zero-crossing detection combined with rotational speed differential calculation to identify the back EMF coefficient and moment of inertia of the key electrical parameter set.

[0145] The parameter self-tuning module inputs the identified key electrical parameter set into a preset fuzzy adaptive tuner to generate an initial PID control parameter set. The control performance is verified through closed-loop speed response testing, and the PID control parameter set is dynamically optimized based on speed overshoot and steady-state fluctuation rate.

[0146] The knowledge base storage module binds the optimized PID control parameter group with the motor type identification code, stores it in the device parameter knowledge base of the embedded FLASH memory, and adds a credibility tag.

[0147] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0148] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present application, based on the technical solution and concept of the present application, should be covered within the scope of protection of the present application.

Claims

1. An embedded method for self-tuning BLDC drive parameters for multiple device types, characterized in that, The methods include: A unified mathematical model of BLDC, including electromagnetic torque equation, back electromotive force equation and mechanical motion equation, is constructed, and a nonlinear mapping relationship library between motor type identification code and key electrical parameter set is established. A predefined multi-mode excitation signal sequence is injected into the BLDC motor to be driven, and dynamic response data is collected synchronously. The dynamic response data is used to characterize the features of the key electrical parameter set. Based on the collected dynamic response data, the equivalent inductance and equivalent resistance of the key electrical parameter set are extracted by frequency domain impedance analysis. At the same time, the back EMF coefficient and moment of inertia of the key electrical parameter set are identified by using time domain back EMF zero-crossing detection combined with speed differential operation. The identified key electrical parameter set is input into a preset fuzzy adaptive tuner to generate an initial PID control parameter set. The control performance is verified by closed-loop speed response test. The PID control parameter set is dynamically optimized according to speed overshoot and steady-state fluctuation rate. The optimized PID control parameter set is bound to the motor type identification code, stored in the device parameter knowledge base of the embedded FLASH memory, and a credibility tag is attached.

2. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The method for constructing a nonlinear mapping relationship library includes: inputting the no-load starting current waveform of the motor into a waveform encoder to generate a feature vector; processing the feature vector through a lightweight convolutional neural network; and using the output of the lightweight convolutional neural network as the iterative initial value for identifying key electrical parameters.

3. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The method for generating multi-mode excitation signal sequences includes: loading a basic excitation template based on the motor type identification code; optimizing the gradient parameters of the basic excitation template using a swarm intelligence algorithm; and injecting the optimized gradient parameters into the dynamic response data acquisition stage.

4. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The frequency domain impedance analysis method includes: performing wavelet packet decomposition on the acquired swept frequency response current; constructing a neural differential equation solver based on the decomposition results; and outputting equivalent circuit parameters from the neural differential equation solver to a fuzzy adaptive tuner.

5. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The real-time temperature compensation method includes: collecting the power module temperature as the input for parameter decoupling; generating resistance drift compensation through a heat transfer state-space model; and inputting the compensated resistance value into the optimization stage of the PID control parameter group.

6. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The credibility label generation method includes: receiving performance index data output from closed-loop speed response test verification; processing the performance index through a multi-criteria evaluation tree; setting a label state machine based on the evaluation results and storing it in the device parameter knowledge base.

7. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The fault circuit interruption mechanism includes: monitoring overcurrent and stall events in the S4 closed-loop verification; triggering the switching of security parameter groups and generating key parameter feature hashes; and updating the stored trust labels according to the event type.

8. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The parameter iteration optimization function includes: reading the historical parameter set when the stored credibility label meets the conditions; generating incremental excitation based on the historical parameter set and injecting it into the dynamic response data acquisition stage; updating the moment of inertia and readjusting the PID integral term by decoupling.

9. The embedded method for self-tuning BLDC drive parameters for multiple device types according to claim 1, characterized in that, The method also includes: binding the optimized PID control parameter group with the motor type identification code, storing it in the device parameter knowledge base of the embedded FLASH memory, and attaching a confidence tag. When the confidence tag is high confidence, a dynamic load disturbance sequence is injected and dynamic performance indicators are collected. If the dynamic performance indicators reach the limit condition, the confidence tag is downgraded to steady-state high confidence and the applicable operating condition range is marked. The marked confidence tag and the applicable operating condition range are associated and stored in the device parameter knowledge base.

10. An embedded system for self-tuning BLDC drive parameters for multiple device types, characterized in that, The system includes: a core model building module, which constructs a unified mathematical model of BLDC including electromagnetic torque equation, back electromotive force equation, and mechanical motion equation, and establishes a nonlinear mapping relationship library between motor type identification code and key electrical parameter set; an excitation injection and acquisition module, which injects a predefined multi-mode excitation signal sequence into the BLDC motor to be driven and simultaneously acquires dynamic response data, which is used to characterize the features of the key electrical parameter set; and a parameter decoupling and identification module, which, based on the acquired dynamic response data, decouples and extracts the equivalent inductance and equivalent current inductance from the key electrical parameter set through frequency domain impedance analysis. The system employs an effective resistance and utilizes time-domain back EMF zero-crossing detection combined with speed differential calculation to identify the back EMF coefficient and moment of inertia in the key electrical parameter set. A parameter self-tuning module inputs the identified key electrical parameter set into a preset fuzzy adaptive tuner to generate an initial PID control parameter set. Control performance is verified through closed-loop speed response testing, and the PID control parameter set is dynamically optimized based on speed overshoot and steady-state fluctuation rate. A knowledge base storage module binds the optimized PID control parameter set with the motor type identifier code, stores it in the embedded FLASH memory's device parameter knowledge base, and adds a credibility tag.

Images