Precise control method of eddy current permanent magnet speed regulator based on machine learning algorithm

Abstract

The application discloses a precise control method of a vortex permanent magnet speed regulator based on a machine learning algorithm, and relates to the technical field of permanent magnet transmission control.The method comprises the following steps: collecting multi-source operation data of the vortex permanent magnet speed regulator in real time, and calculating state parameters after preprocessing; forming an analysis period by a plurality of continuous time points, combining state parameters of historical analysis periods to train a machine learning algorithm model for deviation reason classification; performing spatial clustering on historical feature vectors according to deviation reasons, constructing feature clusters corresponding to different deviation reasons, and calculating spatial matching values of the current period feature vector and each feature cluster to determine the current deviation reason; and respectively executing differentiated regulation and control based on state parameter excess and regulation coefficient differences according to the identified deviation reason.The application effectively solves the problems of low control precision, response lag and poor stability of the vortex permanent magnet speed regulator under complex working conditions through multi-dimensional state representation, period feature clustering and reason-oriented precise control.

Description

Technical Field

[0001] This invention relates to the field of communication technology, specifically to a precise control method for eddy current permanent magnet speed controllers based on machine learning algorithms. Background Technology

[0002] Eddy current permanent magnet speed controllers typically utilize the relative motion between a permanent magnet and a conductor rotor to generate eddy currents in the conductor, forming a non-contact torque transmission through electromagnetic coupling, thereby regulating motor speed. They are commonly used in applications such as fans, pumps, conveying equipment, mining machinery, metallurgical equipment, and ventilation systems, and are particularly suitable for high-power, long-cycle, continuous operation requiring energy-saving speed regulation. However, in actual operation, existing eddy current permanent magnet speed controllers often face the problem of insufficient control accuracy. Especially under the combined effects of load fluctuations, ambient temperature changes, permanent magnet demagnetization, air gap misalignment, and mechanical vibration, the output motor speed easily deviates from the target motor speed, and this deviation is cumulative and time-varying, making it difficult to compensate for over a long period using fixed control parameters.

[0003] Existing control methods mostly employ empirical parameter setting, single feedback regulation, or simplified model control. Although they can achieve basic speed regulation within a certain range, they are prone to problems such as response lag, increased speed regulation deviation, decreased stability, and insufficient control accuracy when the operating conditions change significantly. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a precise control method for eddy current permanent magnet speed controllers based on machine learning algorithms, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a precise control method for an eddy current permanent magnet speed controller based on machine learning algorithms, comprising the following steps:

[0006] Step 1: During the operation of the eddy current permanent magnet speed controller, real-time operating data is collected. After preprocessing the operating data, full-state analysis is performed to obtain the state parameters.

[0007] Step 2: Construct an analysis period by using multiple consecutive sampling times, extract the mean and first-to-last difference of each state index within each analysis period, construct a period feature vector, and combine it with the historical period feature vectors and their corresponding deviation cause labels to train a machine learning classification model;

[0008] Step 3: Spatial clustering of the feature vectors of historical time periods according to the category of deviation cause to obtain the feature clusters corresponding to each deviation cause. Calculate the spatial matching value between the feature vector of the current analysis period and each feature cluster, and determine the current deviation cause through dynamic threshold. The deviation causes include load mutation type deviation, temperature rise demagnetization type deviation, air gap offset type deviation, insufficient cooling type deviation, and vibration eccentricity type deviation.

[0009] Step 4: Based on the identified cause of the deviation, implement the corresponding precise control strategy.

[0010] Furthermore, the operational data collected in step one includes at least: motor speed, instantaneous load torque, instantaneous input current, instantaneous output power, instantaneous permanent magnet temperature, instantaneous conductor disk temperature, instantaneous shell temperature, actual air gap distance, air gap eccentricity, actual cooling wind speed, actual cooling flow rate, actual shell heat dissipation temperature difference, instantaneous vibration amplitude, vibration change, and mechanical eccentricity; all sensors are synchronously triggered by a hardware clock, and the data undergoes resampling, interpolation, and normalization preprocessing.

[0011] Furthermore, the full-state analysis includes load state analysis, thermal state analysis, air gap state analysis, cooling state analysis, and mechanical state analysis, and the corresponding state parameters include load state index, thermal state index, air gap state index, cooling state index, and mechanical state index.

[0012] Furthermore, the load state analysis includes: obtaining the corresponding load reference torque, reference input current, and reference output power from a preset reference database based on the current motor speed; calculating the torque deviation term, current deviation term, and power deviation term respectively, and assigning weighting factors for weighted fusion to obtain the load state index, specifically including:

[0013] The torque deviation term is obtained by subtracting the reference load torque under the current motor speed condition from the instantaneous load torque and dividing the difference by the reference load torque; the current deviation term is obtained by subtracting the reference input current under the current motor speed condition from the instantaneous input current and dividing the difference by the reference input current; the power deviation term is obtained by subtracting the reference output power under the current motor speed condition from the instantaneous output power and dividing the difference by the reference output power. The torque deviation term characterizes the degree of deviation of the current load from the reference load under normal motor speed conditions, the current deviation term characterizes the degree of deviation of the input energy demand, and the power deviation term characterizes the degree of deviation of the output energy transfer capability. Weighting factors are assigned to the torque deviation term, current deviation term, and power deviation term, and they are weighted and fused according to the assigned weighting factors to obtain the load state index, with the sum of all weighting factors being one.

[0014] Further, the thermal state analysis includes: obtaining the corresponding permanent magnet reference temperature, conductor disk reference temperature, shell reference temperature, and allowable deviation values ​​for each temperature according to the equipment type; calculating the permanent magnet deviation term, conductor disk deviation term, and shell deviation term respectively, and assigning weighting factors for weighted fusion to obtain the thermal state index; specifically including:

[0015] The type of eddy current permanent magnet speed controller used in this embodiment is obtained and compared with all preset types to obtain the corresponding temperature reference parameters. The instantaneous permanent magnet temperature is subtracted from the permanent magnet reference temperature, and the non-negative part of the difference is divided by the allowable deviation of the permanent magnet temperature to obtain the permanent magnet deviation term. The instantaneous conductor disk temperature is subtracted from the conductor disk reference temperature, and the non-negative part of the difference is divided by the allowable deviation of the conductor disk temperature to obtain the conductor disk deviation term. The instantaneous housing temperature is subtracted from the housing reference temperature, and the non-negative part of the difference is divided by the allowable deviation of the housing temperature to obtain the housing deviation term. The permanent magnet deviation term is used to characterize the risk of thermal over-limit of the permanent magnet, the conductor disk deviation term is used to characterize the risk of temperature rise caused by eddy current loss, and the housing deviation term is used to characterize the risk of overall heat dissipation accumulation. A weighting factor is assigned to the permanent magnet deviation term, the conductor disk deviation term, and the housing deviation term respectively. The permanent magnet deviation term, the conductor disk deviation term, and the housing deviation term are weighted and fused according to the assigned weighting factors to obtain the thermal state index. The thermal state index is used to characterize the risk of temperature rise and demagnetization during the analysis period. The sum of the weighting factors is one.

[0016] Further, the air gap state analysis includes: obtaining the corresponding reference air gap, allowable air gap deviation, and allowable eccentricity according to the equipment type; calculating the air gap deviation term and air gap eccentricity term, and assigning weight factors for weighted fusion to obtain the air gap state index; specifically including:

[0017] The air gap deviation term is obtained by subtracting the reference air gap from the actual air gap distance and dividing the absolute value of the difference by the allowable air gap deviation value. The air gap eccentricity term is obtained by dividing the air gap eccentricity by the allowable eccentricity. A weighting factor is assigned to the air gap deviation term and the air gap eccentricity term respectively. The air gap deviation term and the air gap eccentricity term are weighted and fused according to the assigned weighting factors to obtain the air gap state index. The air gap state index is used to characterize the air gap offset state during the analysis period. The sum of the weighting factors is one.

[0018] Furthermore, the cooling state analysis includes: obtaining the corresponding cooling reference wind speed, cooling reference flow rate, casing reference heat dissipation temperature difference, and allowable deviations of each parameter according to the equipment type; calculating the wind speed deviation, flow rate deviation, and heat dissipation temperature difference deviation, and assigning weight factors for weighted fusion to obtain the cooling state index; specifically including:

[0019] Obtain the type identifier of the eddy current permanent magnet speed controller used in the embodiment, and compare it with all preset types to obtain the corresponding cooling reference parameters; subtract the actual cooling wind speed from the cooling reference wind speed, and divide the non-negative part by the allowable deviation of the cooling wind speed to obtain the wind speed deviation term; subtract the actual cooling flow rate from the cooling reference flow rate, and divide the non-negative part by the allowable deviation of the cooling flow rate to obtain the flow rate deviation term; subtract the reference heat dissipation temperature of the casing from the actual casing heat dissipation temperature difference, and divide the non-negative part by the allowable deviation of the casing heat dissipation temperature difference to obtain the heat dissipation deviation term; assign weight factors to the cooling wind speed deviation term, cooling flow rate deviation term, and heat dissipation temperature difference deviation term respectively, the sum of the weight factors is one, and perform weighted fusion of the cooling wind speed deviation term, cooling flow rate deviation term, and heat dissipation temperature difference deviation term according to the assigned weight factors to obtain the cooling state index, which represents the insufficient heat dissipation state during the analysis period.

[0020] Further, the mechanical state analysis includes: obtaining the corresponding vibration reference amplitude, vibration reference change, allowable vibration deviation, and allowable mechanical eccentricity according to the equipment type; calculating the amplitude deviation term, change deviation term, and mechanical eccentricity term, and assigning weighting factors for weighted fusion to obtain the mechanical state index; specifically including:

[0021] Obtain the type identifier of the eddy current permanent magnet speed controller used in this embodiment, and compare it with all preset types to obtain the corresponding mechanical reference parameters; subtract the vibration reference amplitude from the instantaneous vibration amplitude, and divide the absolute value of the difference by the allowable vibration deviation to obtain the amplitude deviation term; subtract the vibration reference change from the vibration change, and divide the absolute value of the difference by the allowable vibration deviation to obtain the change deviation term; divide the mechanical eccentricity by the allowable mechanical eccentricity to obtain the mechanical eccentricity term; assign weight factors to the amplitude deviation term, change deviation term, and mechanical eccentricity term respectively, the sum of the weight factors is one, and perform weighted fusion of the amplitude deviation term, change deviation term, and mechanical eccentricity term according to the assigned weight factors to obtain the mechanical state index; the mechanical state index characterizes the vibration eccentricity state during the analysis period.

[0022] Further, in step three, the historical time period feature vectors are spatially clustered according to the deviation cause category to obtain at least one cluster center under each deviation cause category; the L2 norm distance between the current time period feature vector and each cluster center is calculated, and the average value is taken to obtain the spatial distance with the feature cluster, and this spatial distance is mapped to a spatial matching value through a formula; a preset label threshold is set, and if the maximum spatial matching value is greater than or equal to the label threshold, the corresponding deviation cause is output as the identification result; specifically including:

[0023] Obtain the state parameters for all analysis periods during the historical operation of the eddy current permanent magnet speed governor, and record them as follows:

[0024]

[0025] A training sample library is constructed. Each sample in the training sample library includes at least one state parameter for an analysis period and its corresponding deviation cause label. The deviation cause label includes at least four types: load mutation type deviation, temperature rise demagnetization type deviation, air gap offset type deviation, vibration eccentricity type deviation, and insufficient cooling type deviation. Here, i represents the time index within the analysis period, n represents the total number of times within the analysis period, and j represents the index of the historical analysis period. To facilitate machine learning algorithm training, the state parameters of each historical analysis period are feature-encoded to form an analysis period feature vector. Specifically, the state parameter vectors for each time point within the analysis period can be expanded chronologically, and the statistical descriptive quantities of the analysis period can be extracted to form a sample input vector of uniform length. The analysis period feature vector is represented as follows: Z j Let represent the feature vector corresponding to the analysis period j, and Φ(·) represent the mapping function for feature encoding of the parameters of the analysis period. The mapping function Φ(·) includes, but is not limited to, one or more combinations of the following processing methods:

[0026] (1) Calculate the mean of the state index at each moment within the analysis period;

[0027] (2) Extract the maximum value of each state index within the analysis period;

[0028] (3) Calculate the standard deviation of each state index during the analysis period;

[0029] (4) Calculate the change in each state index between adjacent time points within the analysis period;

[0030] (5) Calculate the difference between the first and last moments of each state index within the analysis period;

[0031] In this example, the feature vector is constructed by calculating the mean of the state index at each moment within the analysis period and the difference between the first and last moments of each state index within the analysis period, i.e., (1) and (5) above. Thus, the feature vector can be expressed as:

[0032]

[0033] in , , , , ;

[0034] After obtaining the feature vectors of the historical samples for the analysis period, these vectors are input into a machine learning algorithm model for training to establish the classification relationship between the state parameters and deviation causes of the analysis period. The machine learning algorithm model includes, but is not limited to, any one of support vector machines, random forests, neural networks, and gradient boosting trees, or a fusion model composed of combinations of the above models. During training, the feature vectors of the historical analysis period are used as input, and the corresponding deviation cause labels are used as output to construct supervised learning sample pairs (Z). j y j ), y j The label represents the deviation reason label corresponding to the analysis period j.

[0035] Further, in step four, for load abrupt deviation, the average load state index during the current analysis period is subtracted from the load state threshold to obtain the load excess, which is multiplied by the load adjustment coefficient to obtain the air gap compensation amount. Then, the air gap adjustment direction is determined according to the motor speed deviation sign function, and the air gap control amount is synthesized to drive the actuator. For temperature rise demagnetization deviation, the average thermal state index is subtracted from the thermal state threshold to obtain the thermal excess, which is multiplied by the cooling adjustment coefficient and the thermal compensation adjustment coefficient to obtain the cooling enhancement amount and the air gap thermal compensation amount. The average actual cooling wind speed at each moment during the current analysis period is added to the cooling enhancement amount to obtain the target cooling wind speed. Then, the air gap thermal compensation amount is subtracted from the reference air gap to obtain the target air gap, and the cooling fan and actuator are driven according to the target cooling wind speed and the target air gap. For air gap offset deviation, the average air gap state index is subtracted from the air gap state threshold to obtain the air gap excess, which is multiplied by the air gap correction coefficient to obtain the air gap correction amount, and then... The air gap control quantity is synthesized based on the air gap deviation sign function. For vibration eccentricity type deviation, the mechanical excess is obtained by subtracting the mechanical state threshold from the mean mechanical state index. This excess is then multiplied by the speed limit coefficient and the eccentricity correction coefficient to obtain the speed limit adjustment and rotor alignment correction. The target speed adjustment is obtained by subtracting the speed limit adjustment from the mean actuator adjustment speed during the analysis period. The target mechanical eccentricity is obtained by subtracting the alignment correction from the mean mechanical eccentricity during the analysis period. The actuator adjustment speed and mechanical alignment are adjusted accordingly. For insufficient cooling type deviation, the cooling excess is obtained by subtracting the cooling state threshold from the mean cooling state index. This excess is then multiplied by the wind speed adjustment coefficient and the flow rate adjustment coefficient to obtain the cooling wind speed increment and the cooling flow rate increment. The target cooling wind speed is obtained by adding the cooling wind speed increment to the actual cooling wind speed during the analysis period. The target cooling flow rate is obtained by adding the cooling flow rate increment to the mean actual cooling flow rate during the analysis period. The cooling actuator is controlled accordingly.

[0036] When implementing the above five types of control, the corresponding adjustment amount is first calculated, and then compared with the pre-calibrated corresponding dead zone to determine whether the adjustment amount falls into the dead zone, thereby deciding whether to implement the control.

[0037] The present invention has the following beneficial effects:

[0038] Through the synergistic effect of steps one through four, precise control of the eddy current permanent magnet speed controller is achieved under all operating conditions. Specifically: First, by deeply analyzing the operating data, five types of derived features are constructed: load state index, thermal state index, air gap state index, cooling state index, and mechanical state index. These features characterize the actual operating state of the equipment from multiple dimensions, including electromagnetic, thermal, mechanical, and heat dissipation aspects, providing rich and physically interpretable input for subsequent deviation cause identification. Second, step two uses analysis periods consisting of multiple consecutive moments as the basic unit, extracting the mean and first-to-last difference of the state index within each period as feature vectors. A machine learning algorithm is then used to establish a classification mapping between the feature vectors and deviation causes, overcoming the shortcomings of single-moment data being susceptible to noise interference and unable to reflect the state evolution trend, thus improving the robustness and accuracy of classification. Third, step three uses spatial clustering... By constructing feature clusters corresponding to different deviation causes and identifying deviation causes in the current analysis period through spatial matching value calculation, the normal fluctuations and true deviations can be effectively distinguished, avoiding false triggers and missed judgments. Finally, step four targets five typical deviation causes: load mutation, temperature rise demagnetization, air gap offset, vibration eccentricity, and insufficient cooling. Based on quantitative calculations such as state index excess, adjustment coefficient, and sign function, the identification results are directly converted into specific execution commands such as air gap adjustment, cooling enhancement, and speed limit, achieving precise control driven by cause and value. The above four steps are organically linked to jointly solve the problems of control lag, deviation accumulation, and poor stability caused by nonlinearity, time-varying nature, and multi-factor coupling in the existing technology, and improve the speed regulation accuracy, response speed, and long-term operational reliability of eddy current permanent magnet speed controllers under varying operating conditions. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0041] 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.

[0042] Application Scenarios: Existing eddy current permanent magnet speed controllers often suffer from insufficient control accuracy in practical applications. Especially under the combined effects of load fluctuations, ambient temperature changes, magnet demagnetization, air gap misalignment, and mechanical vibration, deviations easily occur between the output motor speed and the target motor speed, and these deviations are difficult to compensate for over a long period using fixed parameters. Because the electromagnetic coupling process of eddy current permanent magnet speed controllers exhibits significant nonlinearity, time-varying characteristics, and operating condition coupling, the controlled object is not only affected by the input motor speed and load but also by multiple coupled factors such as temperature rise, magnetic field distribution, material loss, and structural assembly errors. Existing control methods mostly rely on empirical parameter settings, single feedback adjustment, or simplified model control, which are difficult to accurately depict the real operating laws under complex working conditions. Therefore, they are prone to problems such as response lag, increased speed regulation deviation, and decreased stability under variable working conditions. To address the above technical problems, this invention uses multiple consecutive sampling moments to form an analysis period, constructs five types of state indices using the analysis period as the identification unit, trains a machine learning model based on historical analysis period samples, and then combines spatial clustering and minimum spatial distance matching to identify the deviation causes corresponding to the current analysis period, thereby achieving closed-loop precise control of the eddy current permanent magnet speed controller.

[0043] Please see Figure 1 This invention provides a technical solution: a precise control method for an eddy current permanent magnet speed controller based on machine learning algorithms, comprising the following steps:

[0044] Step 1: During the operation of the eddy current permanent magnet speed controller, sensors installed on the equipment synchronously collect real-time operating data. The collected data includes at least the following: motor speed, instantaneous load torque, instantaneous input current, instantaneous output power, instantaneous permanent magnet temperature, instantaneous conductor disk temperature, instantaneous shell temperature, actual air gap distance, air gap eccentricity, actual cooling wind speed, actual cooling flow rate, actual shell heat dissipation temperature difference, instantaneous vibration amplitude, vibration change, and mechanical eccentricity. Air gap eccentricity refers to the degree of uneven distribution of the actual gap between the permanent magnet rotor and the conductor rotor in the circumferential direction; actual shell heat dissipation temperature difference refers to the difference between the current shell temperature and the ambient temperature; vibration change refers to the difference in vibration amplitude between adjacent sampling times; and mechanical eccentricity refers to the offset between the rotor's rotation center and the ideal geometric center. All sensors are synchronously triggered via a hardware clock to ensure strict alignment of all physical quantities on the time axis.

[0045] For operational data with different sampling frequencies, all original sampled data are first resampled to an equally spaced time sequence using a unified timestamp. Missing values ​​are filled in using linear interpolation to ensure complete data at each time point. Then, the operational data is normalized to eliminate the influence of different physical quantity dimensions on subsequent calculations. For each physical quantity data point in the operational data, a minimum-maximum normalization method is used to map it to the interval [0, 1]. Let the minimum value of a certain physical quantity data point x within its historical normal operating range be... The maximum value is The normalized value X is then calculated using the following formula:

[0046]

[0047] Subsequent machine learning models are sensitive to the scale of input features. Normalization is used to avoid large-scale features dominating the model, ensuring a balanced contribution of each physical quantity to classification and similarity calculation. After preprocessing the operational data, further calculations and analyses are performed to obtain derived features reflecting the actual operating state of the governor, specifically including:

[0048] During the operation of the eddy current permanent magnet speed controller, the operating data is collected periodically in real time and preprocessed; then, an analysis period is formed by n collection times, and any time within the analysis period is indexed as i;

[0049] 1-1, Load Status Analysis: Extract the current motor speed and, based on the current motor speed, search or interpolate from a preset reference database to obtain the load reference torque, reference input current, and reference output power that match the motor speed conditions. The reference database can be constructed from bench test data, historical normal operation data, or calibration curves, and is preferably formed when the equipment is in a stable state without abnormal load disturbances, to characterize the normal load response law of the equipment under different motor speed conditions. Subtract the load reference torque under the current motor speed conditions from the instantaneous load torque, and divide the difference by the load reference torque to obtain the torque deviation term; subtract the reference input current under the current motor speed conditions from the instantaneous input current, and divide the difference by the reference input current to obtain the current deviation term; subtract the reference output power under the current motor speed conditions from the instantaneous output power. The load state index is obtained by dividing the difference by the reference output power. The torque deviation term characterizes the deviation of the current load from the reference load under normal motor speed conditions, the current deviation term characterizes the deviation of the input-side energy demand, and the power deviation term characterizes the deviation of the output-side energy transfer capability. Weighting factors are assigned to the torque, current, and power deviation terms, and they are weighted and fused according to these factors to obtain the load state index, with the sum of all weighting factors being one. The weighting factors are preferably determined based on the contribution of each item in the historical calibration samples to the accuracy of load change identification, or they can be adaptively obtained through machine learning training results to make the load state index more consistent with the abnormal load change patterns under actual operating conditions. The load state index characterizes whether abnormal load changes occur during the operation of the eddy current permanent magnet speed controller.

[0050] 1-2, Thermal State Analysis: Different types of eddy current permanent magnet speed controllers have different temperature reference parameters, including the permanent magnet reference temperature, conductor disk reference temperature, and housing reference temperature, as well as the allowable temperature deviation values ​​for each of these three reference temperatures. Eddy current permanent magnet speed controllers of different models, power ratings, heat dissipation structures, and magnetic material configurations exhibit differences in heat capacity, thermal inertia, upper temperature limit, and temperature rise sensitivity. For example, the allowable temperature of the permanent magnet in high-power density models is usually lower than that in low-power models, and the housing temperature reference value for devices with enhanced heat dissipation structures differs from that of ordinary structures. Using a uniform temperature reference parameter would not accurately reflect the thermal safety boundary of the device itself, easily leading to misjudgments. Therefore, this embodiment pre-establishes different temperature reference parameter groups according to the equipment type to match the thermal state evaluation with the equipment's own thermal characteristics; it obtains the type of eddy current permanent magnet speed controller used in this embodiment and compares it with all preset types to obtain the corresponding temperature reference parameters; it subtracts the permanent magnet reference temperature from the instantaneous permanent magnet temperature, takes the non-negative part of the difference, and divides it by the allowable deviation of the permanent magnet temperature to obtain the permanent magnet deviation term; it subtracts the conductor disk reference temperature from the instantaneous conductor disk temperature, takes the non-negative part of the difference, and divides it by the allowable deviation of the conductor disk temperature to obtain the conductor disk deviation term; it subtracts the housing reference temperature from the instantaneous housing temperature, takes the non-negative part of the difference, and divides it by the allowable deviation of the housing temperature to obtain the housing deviation term; wherein the permanent magnet deviation term is used to characterize The thermal overshoot risk of permanent magnets is assessed. The conductor disk deviation term characterizes the temperature rise risk caused by eddy current losses, while the shell deviation term characterizes the overall heat accumulation risk. A weighting factor is assigned to each of the permanent magnet, conductor disk, and shell deviation terms. Based on these weighting factors, the permanent magnet, conductor disk, and shell deviation terms are weighted and fused to obtain the thermal state index, which characterizes the temperature rise demagnetization risk during the analysis period. The sum of the weighting factors is one. It should be noted that the permanent magnet temperature is directly related to the demagnetization risk and therefore should generally be assigned a higher weight. The conductor disk temperature reflects the internal eddy current losses and heat accumulation trend, playing a crucial role in characterizing thermal risk. The shell temperature reflects the external heat dissipation state and the overall degree of heat accumulation, and is a comprehensive thermal characterization quantity. The weighting factors are preferably determined based on the correlation strength between each temperature term and the temperature rise demagnetization event in historical thermal instability samples, or they can be automatically obtained through machine learning training to ensure that the thermal state index accurately characterizes the temperature rise demagnetization risk during the analysis period.

[0051] 1-3, Air Gap Status Analysis: Different types of eddy current permanent magnet speed controllers are pre-set to have corresponding reference air gaps, allowable air gap deviations, and allowable eccentricities. It should be noted that the standard air gaps of equipment with different structural dimensions, different magnetic pole arrangements, different conductor disk thicknesses, and different mechanical assembly tolerances are not the same, and the allowable air gap deviation ranges and eccentricity tolerances are also different. If a unified air gap standard is directly adopted without distinguishing the equipment type, it cannot accurately reflect the actual coupling boundary of each model of equipment, and it is easy to misjudge normal structural differences as air gap anomalies, or to mask the true offset in a broad threshold. Therefore, this embodiment presets corresponding air gap reference parameters according to the equipment type to ensure that the air gap evaluation is consistent with the actual mechanical structure. The air gap deviation term is obtained by subtracting the reference air gap from the actual air gap distance and dividing the absolute value of the difference by the allowable air gap deviation value. The air gap eccentricity term is obtained by dividing the air gap eccentricity by the allowable eccentricity. A weighting factor is assigned to each of the air gap deviation and eccentricity terms. Based on the assigned weighting factors, the air gap deviation and eccentricity terms are weighted and fused to obtain the air gap state index, which is used to characterize the air gap offset state during the analysis period. The sum of the weighting factors is one. The air gap deviation term directly determines the magnetic coupling distance and usually has the most direct impact on speed regulation performance, so its weight should be relatively high. The air gap eccentricity term reflects the uneven spatial distribution and structural deviation, and it has an auxiliary amplification effect on the coupling effect; therefore, its weight can be appropriately set according to the sensitivity of the equipment structure. The weighting factors are preferably obtained through historical speed regulation deviation samples, air gap calibration experiments, or machine learning feature contribution analysis to ensure that the air gap state index can accurately characterize the air gap offset state during the analysis period.

[0052] 1-4, Cooling Status Analysis: Different types of eddy current permanent magnet speed controllers have different cooling reference parameters. These parameters include at least the cooling reference wind speed, cooling reference flow rate, housing reference heat dissipation temperature difference, and the allowable deviation for each parameter. Because equipment with different power levels, heat dissipation structures, air duct arrangements, and cooling medium configurations has significant differences in heat exchange efficiency and heat dissipation capacity, using uniform cooling reference parameters would make it difficult to accurately reflect the heat dissipation boundary of each model under normal operating conditions, which could easily lead to misjudgment of insufficient cooling risk or excessive restrictions on high heat dissipation performance equipment. Therefore, this embodiment presets cooling reference parameters that match the thermal design capability of the equipment type to ensure that the cooling status evaluation is consistent with the actual heat dissipation conditions; obtains the type identifier of the eddy current permanent magnet speed controller used in this embodiment and compares it with all preset types to obtain the corresponding cooling reference parameters; subtracts the actual cooling wind speed from the cooling reference wind speed and divides the non-negative part by the allowable deviation of the cooling wind speed to obtain the wind speed deviation term; subtracts the actual cooling flow rate from the cooling reference flow rate and divides the non-negative part by the allowable deviation of the cooling flow rate to obtain the flow rate deviation term; subtracts the reference heat dissipation temperature of the casing from the actual casing heat dissipation temperature difference and divides the non-negative part by the allowable deviation of the casing heat dissipation temperature difference. The heat dissipation deviation term is obtained. Weighting factors are assigned to the cooling air velocity deviation term, cooling flow rate deviation term, and heat dissipation temperature difference deviation term, with the sum of these weighting factors being one. Based on the assigned weighting factors, the cooling air velocity deviation term, cooling flow rate deviation term, and heat dissipation temperature difference deviation term are weighted and fused to obtain the cooling state index. The cooling state index characterizes the insufficient heat dissipation state during the analysis period. It should be noted that the characterization capabilities of cooling air velocity, cooling flow rate, and shell heat dissipation temperature difference for insufficient heat dissipation are not entirely the same in different types of eddy current permanent magnet speed controllers. Air-cooled equipment is generally more sensitive to air velocity, liquid-cooled equipment is generally more sensitive to flow rate, while high-power-density equipment is more dependent on changes in shell heat dissipation temperature difference. Therefore, the weighting factors are preferably determined based on the equipment type, cooling structure form, and the feature contribution in historical thermal instability samples, or they can be adaptively obtained through thermal management experiments or machine learning training results to make the cooling state index more accurately characterize the insufficient cooling state during the analysis period.

[0053] Mechanical condition analysis: Different types of eddy current permanent magnet speed controllers have different mechanical reference parameters. The mechanical reference parameters include at least the vibration reference amplitude, vibration reference change, allowable vibration deviation, and allowable mechanical eccentricity. Equipment with different structural dimensions, rotor mass distributions, bearing configurations, and installation precisions will have different normal vibration levels and allowable eccentricity ranges. If a uniform mechanical reference value is used, it will be difficult to accurately reflect the normal mechanical condition of each type of equipment, and structural differences may be misjudged as mechanical abnormalities, or the true eccentricity may be masked under a broad threshold. Therefore, this embodiment establishes mechanical reference parameters that match the structural characteristics of the equipment type to ensure that the mechanical condition evaluation conforms to the actual operating conditions; obtains the type identifier of the eddy current permanent magnet speed controller used in this embodiment and compares it with all preset types to obtain the corresponding mechanical reference parameters; subtracts the vibration reference amplitude from the instantaneous vibration amplitude and divides the absolute value of the difference by the allowable vibration deviation to obtain the amplitude deviation term; subtracts the vibration reference change from the vibration change and divides the absolute value of the difference by the allowable vibration deviation to obtain the change deviation term; divides the mechanical eccentricity by the allowable mechanical eccentricity to obtain the mechanical eccentricity term; assigns weight factors to the amplitude deviation term, change deviation term, and mechanical eccentricity term respectively, the sum of the weight factors is one, and performs weighted fusion of the amplitude deviation term, change deviation term, and mechanical eccentricity term according to the assigned weight factors to obtain the mechanical condition index; the mechanical condition index characterizes the vibration eccentricity state during the analysis period; it should be noted that the vibration amplitude usually directly reflects the mechanical energy disturbance, the vibration change reflects whether there is a sudden deterioration in the mechanical condition, and the mechanical eccentricity reflects the degree of structural misalignment or shaft offset. For different installation structures and different load conditions, these three types of parameters have different sensitivities to mechanical anomalies. Therefore, the weighting factors are preferably determined based on historical vibration anomaly samples, equipment structure, and mechanical failure mode recognition results. Alternatively, they can be determined adaptively through machine learning training to make the mechanical state index more accurately represent the vibration eccentricity state during the analysis period.

[0054] Therefore, the state parameters of the eddy current permanent magnet speed governor at various moments during operation can be obtained, including the load state index, thermal state index, air gap state index, cooling state index, and mechanical state index, and denoted as F. i R i Q i C i and M i ;

[0055] Step 2: Obtain the state parameters for all analysis periods during the historical operation of the eddy current permanent magnet speed governor, and record them as follows:

[0056]

[0057] A training sample library is constructed. Each sample in the training sample library includes at least one state parameter for an analysis period and its corresponding deviation cause label. The deviation cause label includes at least four types: load mutation type deviation, temperature rise demagnetization type deviation, air gap offset type deviation, vibration eccentricity type deviation, and insufficient cooling type deviation. Here, i represents the time index within the analysis period, n represents the total number of times within the analysis period, and j represents the index of the historical analysis period. To facilitate machine learning algorithm training, the state parameters of each historical analysis period are feature-encoded to form an analysis period feature vector. Specifically, the state parameter vectors for each time point within the analysis period can be expanded chronologically, and the statistical descriptive quantities of the analysis period can be extracted to form a sample input vector of uniform length. The analysis period feature vector is represented as follows: Z j Let represent the feature vector corresponding to the analysis period j, and Φ(·) represent the mapping function for feature encoding of the parameters of the analysis period. The mapping function Φ(·) includes, but is not limited to, one or more combinations of the following processing methods:

[0058] (1) Calculate the mean of the state index at each moment within the analysis period;

[0059] (2) Extract the maximum value of each state index within the analysis period;

[0060] (3) Calculate the standard deviation of each state index during the analysis period;

[0061] (4) Calculate the change in each state index between adjacent time points within the analysis period;

[0062] (5) Calculate the difference between the first and last moments of each state index within the analysis period;

[0063] In this example, the feature vector is constructed by calculating the mean of the state index at each moment within the analysis period and the difference between the first and last moments of each state index within the analysis period, i.e., (1) and (5) above. Thus, the feature vector can be expressed as:

[0064]

[0065] in , , , , ;

[0066] After obtaining the feature vectors of the historical samples for the analysis period, these vectors are input into a machine learning algorithm model for training to establish the classification relationship between the state parameters and deviation causes of the analysis period. The machine learning algorithm model includes, but is not limited to, any one of support vector machines, random forests, neural networks, and gradient boosting trees, or a fusion model composed of combinations of the above models. During training, the feature vectors of the historical analysis period are used as input, and the corresponding deviation cause labels are used as output to construct supervised learning sample pairs (Z). j y j ), y j The label represents the deviation cause label corresponding to the analysis period j. Through training, the machine learning algorithm model learns the distribution pattern of different deviation causes in the feature space of the analysis period, thereby forming a preliminary classification ability for the deviation causes of the current analysis period. Deviation causes usually have continuous evolution characteristics. Compared with the state parameters at a single moment, the samples of the analysis period can more accurately reflect the trajectory of state change, which is conducive to improving classification accuracy and robustness.

[0067] Step 3: After the machine learning model has been trained, the feature vectors of the historical analysis period are spatially clustered according to the category of deviation cause, constructing feature clusters corresponding to different deviation causes. Specifically, feature vectors belonging to the same deviation cause category in the historical analysis period are input into the clustering algorithm for clustering processing. Scattered feature vectors are removed to obtain at least one cluster center under that deviation cause category. The cluster center is used to characterize the typical distribution position of this type of deviation cause in the feature space. The same deviation cause is usually not concentrated at a single point in the feature space, but forms a sample cluster with a certain discrete range. The cluster center can more accurately reflect the typical characteristics of this type of deviation pattern, thereby improving the stability of matching and identification. The feature vector corresponding to the analysis period of the eddy current permanent magnet speed controller during the current operation is extracted and used as the matching vector. The spatial matching value between the matching vector and all feature clusters is calculated. The larger the spatial matching value, the closer the matching vector is to the feature cluster, that is, the closer the pattern of the deviation cause is. The specific process includes:

[0068] After obtaining the cluster centers corresponding to each deviation cause, the spatial distance between the matching vector and the cluster centers of each feature cluster is calculated using the L2 norm. Then, the average of these spatial distances is used to calculate the spatial distance between the matching vector and the feature clusters. Finally, the spatial distance between the matching vector and the feature clusters is mapped to a spatial matching value η, calculated using the following formula:

[0069]

[0070] Where d is the spatial distance between the matching vector and the feature cluster;

[0071] In another embodiment, the spatial matching value η can also be calculated using an exponential decay form, as shown in the following formula:

[0072]

[0073] Where λ is the attenuation coefficient, the larger λ is, the more obvious the attenuation effect of distance on the matching value; λ ranges from 2 to 5; in this embodiment, λ is taken as 1; usually, its specific value is determined based on the recognition effect of historical samples;

[0074] This yields the spatial matching values ​​between the matching vector and the feature clusters under all deviation cause categories. The deviation cause corresponding to the largest spatial matching value is selected as the preliminary label for the matching vector. A preset label threshold is set, determined statistically based on historical sample data: specifically, the distribution of spatial matching values ​​between historical normal operation analysis periods and each deviation cause feature cluster is statistically analyzed, and the upper bound of the spatial matching values ​​under normal operation is selected as the label threshold, or the boundary point between the matching values ​​of deviation samples and normal samples is selected as the label threshold. If the largest spatial matching value is greater than or equal to the label threshold, it indicates that the matching vector has a high similarity to the feature cluster of that deviation cause, suggesting that the current... If the state change pattern during the analysis period is consistent with the historical sample distribution corresponding to the deviation cause, and the pre-defined label is valid, then the pre-defined label is used as the deviation cause corresponding to the current operating period of the eddy current permanent magnet speed controller as the identification result and output. If the maximum spatial matching value is less than the label threshold, it means that the similarity between the feature vector of the current analysis period and all feature clusters of deviation causes is insufficient, indicating that the current state change has not reached any feature pattern of deviation cause, the current operating state is within the normal fluctuation range, and the eddy current permanent magnet speed controller does not have obvious deviation. Then, return to step one, continue to obtain data for the next analysis period and re-analyze to maintain continuous monitoring of the equipment's operating status.

[0075] Step four: Extract the causes of deviations in the current operating period of the eddy current permanent magnet speed controller and execute corresponding adjustments, specifically including:

[0076] Based on the mechanical clearance, static friction, and positioning accuracy of the actuator, an air gap adjustment dead zone δ1 is calibrated through field testing. The unit of the air gap adjustment dead zone δ1 is millimeters, which represents the minimum displacement change that the actuator can reliably respond to. The typical value range is 0.02 to 0.05 mm.

[0077] Based on the minimum controllable speed step size of the fan motor, the duct resistance characteristics, and the measurement resolution of the wind speed sensor, the wind speed adjustment dead zone δ2 is calibrated through on-site step response testing. The unit of wind speed adjustment dead zone δ2 is revolutions per minute or meters per second. It represents the minimum speed change or outlet wind speed change that the cooling fan can reliably respond to. The typical value range is: 30 to 100 r / min for speed-controlled fans; 0.2 to 0.5 m / s for direct wind speed control fans.

[0078] Based on the minimum controllable speed step size of the cooling pump, the pipeline resistance characteristics, and the measurement resolution of the flow sensor, a flow regulation dead zone δ3 is calibrated through on-site flow step response test. The unit of flow regulation dead zone δ3 is liters per minute, which represents the minimum flow change that the cooling system can reliably respond to. The typical value range is 2 to 10 L / min.

[0079] Based on the speed regulation resolution of the actuator driver, the friction characteristics of the mechanical transmission system, and the dynamic accuracy of the speed feedback sensor, a speed adjustment dead zone δ4 is calibrated through speed response tests under no-load and loaded conditions on site. The unit of speed adjustment dead zone δ4 is millimeters per second (mm / s), which represents the minimum speed change that the actuator can stably output under a given speed command. The typical value range is 0.5 to 2.0 mm / s.

[0080] Based on the adjustment accuracy of the rotor alignment device, the resolution of the mechanical eccentricity measurement sensor, and the shaft assembly tolerance, an eccentricity adjustment dead zone δ5 is calibrated using on-site static alignment calibration and dynamic eccentricity monitoring data. The unit of the eccentricity adjustment dead zone δ5 is millimeters, which represents the minimum eccentricity change that the rotor alignment mechanism can reliably correct, with a typical value range of 0.05 to 0.15 mm.

[0081] The system has preset thresholds for load state, thermal state, air gap state, mechanical state, and cooling state. The load state threshold indicates the level at which the load state index is considered to be abnormally deviated from the load; the thermal state threshold indicates the level at which the thermal state index is considered to be significantly at risk of temperature rise and demagnetization; the air gap state threshold indicates the level at which the air gap state index is considered to be abnormally offset from the air gap; the mechanical state threshold indicates the level at which the mechanical state index is considered to be abnormally eccentric; and the cooling state threshold indicates the level at which the cooling state index is considered to be insufficient cooling capacity. Each state index represents the degree of deviation of the original operating parameters from the reference benchmark. The threshold is the dividing line between normal and abnormal states. Only when the state index exceeds the corresponding threshold is it indicated that the deviation of this type of state has reached the level that requires identification and intervention, thereby realizing the graded judgment and precise control of the operating state of the eddy current permanent magnet speed governor.

[0082] If the deviation cause for the current analysis period is identified as a load abrupt change type deviation, it indicates that the load state index has increased significantly during the analysis period, and the load deviation is mainly caused by abnormal fluctuations in torque, current, or output power. The overload is obtained by subtracting the load state threshold from the average load state index during the current analysis period, and then multiplying the non-negative part by the load adjustment coefficient to obtain the air gap compensation amount Δg. The load adjustment coefficient is used to convert the overload into the corresponding air gap adjustment amplitude. The motor speed deviation e is obtained by subtracting the average actual output motor speed at each moment during the analysis period from the target motor speed. The air gap control amount is then calculated using the following formula. :

[0083]

[0084] Where g0 is the reference air gap, and sgn(e) is the sign function of motor speed deviation. When e > 0, it means that the actual output motor speed is lower than the target motor speed. At this time, the actuator drives the actual air gap to decrease, so as to enhance the magnetic coupling strength and increase the output torque. When e < 0, it means that the output motor speed is higher than the target motor speed. At this time, the actuator drives the actual air gap to increase, so as to weaken the magnetic coupling strength and suppress the output overshoot. Thus, the air gap adjustment direction matches the current motor speed deviation direction, so as to offset the influence of sudden load changes on the output motor speed.

[0085] The air gap adjustment is obtained by subtracting the current actual air gap from the air gap control value. If |air gap adjustment value| ≤ air gap adjustment dead zone δ1, the air gap adjustment value is considered to fall into the dead zone, and no displacement command is sent to the actuator to avoid actuator oscillation and wear caused by small repeated adjustments. At the same time, the current motor speed deviation e is included in the integral hold, and the unified output adjustment is only given when the accumulated deviation exceeds the dead zone equivalent threshold. If |air gap adjustment value| > air gap adjustment dead zone δ1, the actuator is driven in the original direction and amplitude, and the integral hold is cleared to zero.

[0086] If the deviation cause corresponding to the current analysis period is identified as a temperature rise demagnetization type deviation, it indicates that the thermal state index continues to rise during the analysis period, and the temperature deviation of the permanent magnet, conductor disk, or shell has exceeded the normal thermal boundary, indicating that the magnetic properties of the permanent magnet material are attenuated. Specifically, the average thermal state index is subtracted from the thermal state threshold, and the non-negative part is multiplied by the cooling adjustment coefficient and the thermal compensation adjustment coefficient to obtain the cooling enhancement amount and the air gap thermal compensation amount, respectively. The cooling adjustment coefficient is used to convert the excess heat into the cooling wind speed increment, and the thermal compensation adjustment coefficient is used to convert the excess heat into the air gap thermal compensation amount. The average actual cooling wind speed at each moment during the current analysis period is added to the cooling enhancement amount to obtain the target cooling wind speed. Then, the reference air gap is subtracted from the air gap thermal compensation amount to obtain the target air gap. The cooling fan is driven to increase the motor speed according to the target cooling wind speed and the target air gap to enhance the heat dissipation capacity. At the same time, the air gap adjustment actuator is driven to reduce the actual air gap according to the target air gap, thereby enhancing the magnetic coupling strength and compensating for the magnetic property attenuation caused by the temperature rise, so that the output torque of the speed controller is restored to the normal range.

[0087] The actual air gap adjustment is obtained by subtracting the target air gap from the current actual air gap. If |actual air gap adjustment| ≤ air gap adjustment dead zone δ1, air gap adjustment is temporarily suspended to avoid frequent actions caused by minor thermal drift. Simultaneously, the thermal state index is accumulated to the thermal compensation integrator. Compensation is performed all at once when the air gap adjustment corresponding to the accumulated amount exceeds the air gap adjustment dead zone δ1. For cooling fan speed adjustment, the fan speed adjustment is obtained by subtracting the target cooling fan speed from the current actual cooling fan speed. If |fan speed adjustment| ≤ fan speed adjustment dead zone δ2, the fan speed is not adjusted, and only the heat accumulation trend is recorded. If |fan speed adjustment| > fan speed adjustment dead zone δ2, the fan speed is not adjusted.

[0088] If the deviation cause corresponding to the current analysis period is identified as an air gap offset type deviation, it indicates that the air gap state index has increased significantly during this analysis period. This indicates that the actual air gap between the permanent magnet rotor and the conductor rotor deviates from the reference air gap, or the eccentricity exceeds the allowable range, leading to abnormal changes in the magnetic coupling strength. Subtracting the air gap state threshold from the mean air gap state index yields the air gap excess, and multiplying its non-negative part by the air gap correction coefficient yields the air gap correction amount ΔK. The air gap correction coefficient is used to convert the air gap state deviation into the actual displacement adjustment amount. Subtracting the reference air gap from the mean actual air gap during the current analysis period yields the air gap deviation q, which is then used to calculate the air gap control amount according to the following formula. :

[0089]

[0090] Where sgn(q) is the air gap deviation sign function. When the air gap deviation > 0, it means that the actual air gap is greater than the reference air gap. At this time, the actuator drives the actual air gap to decrease. When the air gap deviation < 0, it means that the actual air gap is less than the reference air gap. At this time, the actuator drives the actual air gap to increase, thereby causing the actual air gap to converge towards the reference air gap and restoring the magnetic coupling stability.

[0091] The air gap adjustment is obtained by subtracting the current actual air gap from the air gap control value. If |air gap adjustment value| ≤ air gap adjustment dead zone δ1, the air gap adjustment is temporarily suspended to avoid limit cycle oscillation caused by mechanical backlash or measurement noise. At the same time, the dead zone timer is started. If the air gap deviation q persists and |air gap adjustment value| is less than or equal to the air gap adjustment dead zone δ1 for 10 consecutive analysis cycles, it is determined that the actuator is stuck or the air gap sensor zero point drifts, and a maintenance warning is issued. If |air gap adjustment value| > air gap adjustment dead zone δ1, the adjustment is performed normally, and the dead zone timer is cleared.

[0092] If the deviation cause for the current analysis period is identified as a vibration-eccentric type deviation, it indicates that the mechanical condition index has been continuously increasing during this analysis period, representing mechanical anomalies such as enhanced vibration, shaft eccentricity, component loosening, or transient impact. Subtracting the mechanical condition threshold from the mean mechanical condition index yields the mechanical excess. Multiplying the non-negative portion by the speed limit coefficient and eccentricity correction coefficient yields the adjustment speed limit and centering correction amount. The speed limit coefficient is used to convert the mechanical excess into a reduction in the actuator's adjustment speed, thus limiting the amplitude of control actions under vibration conditions and preventing further amplification of mechanical vibration due to rapid adjustments. The eccentricity correction coefficient... This is used to convert mechanical excess into rotor alignment correction to reduce the disturbance of shaft eccentricity on air gap distribution and magnetic coupling state; the average adjustment speed of the actuator during the analysis period is subtracted from the adjustment speed limit to obtain the target adjustment speed, which is set as the mechanism adjustment speed to reduce the control action amplitude under vibration; at the same time, the average mechanical eccentricity during the analysis period is subtracted from the alignment correction to obtain the target mechanical eccentricity to reduce the disturbance of shaft eccentricity on air gap distribution and magnetic coupling state; if the vibration excess is large, the controller can further reduce the output speed adjustment rate to allow the mechanical state to recover to stability before performing a larger air gap adjustment.

[0093] For controlling the speed adjustment of the actuator: the speed adjustment amount is obtained by subtracting the current speed adjustment of the actuator from the target speed adjustment amount. If |speed adjustment amount| ≤ speed adjustment dead zone δ4, it is considered that the speed adjustment amount falls into the dead zone, and the current speed adjustment is maintained unchanged to avoid frequent reversing of the drive due to small speed adjustments; if |speed adjustment amount| > speed adjustment dead zone δ4, the actuator is driven according to the target speed adjustment.

[0094] For rotor alignment correction: subtract the current mechanical eccentricity from the target mechanical eccentricity to obtain the eccentricity adjustment amount. If |eccentricity adjustment amount| ≤ eccentricity adjustment dead zone δ5, then no alignment adjustment is performed, and only the eccentricity trend is recorded; if |eccentricity adjustment amount| > eccentricity adjustment dead zone δ5, then the alignment mechanism is driven to make corrections.

[0095] If the deviation cause for the current analysis period is identified as insufficient cooling, it indicates that the cooling state index has increased during this analysis period. This indicates insufficient cooling airflow, decreased cooling flow, or increased temperature difference in the casing, resulting in heat accumulation due to the inability to dissipate heat in a timely manner. Specifically, the cooling state index is subtracted from the cooling state threshold to obtain the cooling excess. The non-negative part of this excess is multiplied by the airflow adjustment coefficient and the flow rate adjustment coefficient to obtain the cooling airflow increment and cooling flow rate increment, respectively. The airflow adjustment coefficient is used to convert the cooling excess into an increase in fan motor speed or cooling airflow to enhance heat dissipation capacity. The flow rate adjustment coefficient is used to convert the cooling excess into an increase in cooling flow rate to increase the heat removal capacity of the cooling medium. The actual cooling airflow during the analysis period is then added to the cooling airflow increment to obtain the target cooling airflow. Simultaneously, the average actual cooling flow rate during the analysis period is added to the cooling flow rate increment to obtain the target cooling flow rate. The cooling actuator of the eddy current permanent magnet speed controller is controlled according to the target cooling airflow and target cooling flow rate to reduce the heating rate, prevent further temperature accumulation, and bring the cooling state back to the allowable range.

[0096] The target cooling airflow rate is subtracted from the current actual cooling airflow rate to obtain the airflow rate adjustment amount. If |airflow rate adjustment amount| ≤ airflow rate adjustment dead zone δ2, the fan speed is not adjusted, and only the heat accumulation trend is recorded; if |airflow rate adjustment amount| > airflow rate adjustment dead zone δ2, the fan speed is not adjusted.

[0097] The flow rate adjustment amount is obtained by subtracting the current actual cooling flow rate from the target cooling flow rate. If |flow rate adjustment amount| ≤ flow rate adjustment dead zone δ3, the cooling pump or valve is not adjusted; if |flow rate adjustment amount| > flow rate adjustment dead zone δ3, the flow rate adjustment is not performed.

[0098] All the above-mentioned control methods follow the same control logic: first, the excess is calculated based on the mean and threshold of the corresponding state index during the current analysis period; then, the excess is converted into a specific control quantity through the corresponding adjustment coefficient; finally, the control quantity is applied to the actuator displacement, actuator speed, cooling fan speed, cooling flow rate, or output load limit command to achieve targeted and precise control for different causes of deviation. The dead zone measures in all control strategies are designed to avoid ineffective adjustment and oscillation caused by minor disturbances, sensor noise, or mechanical clearance, thereby improving the stability of the system and the lifespan of the actuator. At the same time, through dead zone integral accumulation and timeout judgment mechanisms, it is ensured that the real deviation will not be permanently ignored, balancing control accuracy and execution efficiency.

[0099] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention 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 invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A precise control method for an eddy current permanent magnet speed controller based on machine learning algorithms, comprising: characterized in that, Includes the following steps: Step 1: During the operation of the eddy current permanent magnet speed controller, real-time operating data is collected. After preprocessing the operating data, full-state analysis is performed to obtain the state parameters. Step 2: Construct an analysis period by using multiple consecutive sampling times, extract the mean and first-to-last difference of each state index within each analysis period, construct a period feature vector, and combine it with the historical period feature vectors and their corresponding deviation cause labels to train a machine learning classification model; Step 3: Spatial clustering of the feature vectors of historical time periods according to the category of deviation cause to obtain the feature clusters corresponding to each deviation cause. Calculate the spatial matching value between the feature vector of the current analysis period and each feature cluster, and determine the current deviation cause through dynamic threshold. The deviation causes include load mutation type deviation, temperature rise demagnetization type deviation, air gap offset type deviation, insufficient cooling type deviation, and vibration eccentricity type deviation. Step 4: Based on the identified cause of the deviation, implement the corresponding precise control strategy.

2. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 1, characterized in that, The operational data collected in step one includes at least: motor speed, instantaneous load torque, instantaneous input current, instantaneous output power, instantaneous permanent magnet temperature, instantaneous conductor disk temperature, instantaneous shell temperature, actual air gap distance, air gap eccentricity, actual cooling wind speed, actual cooling flow rate, actual shell heat dissipation temperature difference, instantaneous vibration amplitude, vibration change, and mechanical eccentricity. All sensors are triggered synchronously by a hardware clock, and the data undergoes resampling, interpolation, and normalization preprocessing.

3. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 2, characterized in that, The full-state analysis includes load state analysis, thermal state analysis, air gap state analysis, cooling state analysis, and mechanical state analysis. The corresponding state parameters include load state index, thermal state index, air gap state index, cooling state index, and mechanical state index.

4. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 3, characterized in that, The load state analysis includes: obtaining the corresponding load reference torque, reference input current and reference output power from a preset reference database based on the current motor speed; calculating the torque deviation, current deviation and power deviation terms respectively, and assigning weight factors for weighted fusion to obtain the load state index.

5. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 4, characterized in that, The thermal state analysis includes: obtaining the corresponding permanent magnet reference temperature, conductor disk reference temperature, shell reference temperature, and allowable deviation values ​​for each temperature according to the equipment type; calculating the permanent magnet deviation term, conductor disk deviation term, and shell deviation term respectively, and assigning weight factors for weighted fusion to obtain the thermal state index.

6. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 5, characterized in that, The air gap state analysis includes: obtaining the corresponding reference air gap, allowable air gap deviation, and allowable eccentricity according to the equipment type; calculating the air gap deviation term and air gap eccentricity term, and assigning weight factors for weighted fusion to obtain the air gap state index.

7. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 6, characterized in that, The cooling status analysis includes: obtaining the corresponding cooling reference wind speed, cooling reference flow rate, shell reference heat dissipation temperature difference, and allowable deviation of each parameter according to the equipment type; calculating the wind speed deviation, flow rate deviation, and heat dissipation temperature difference deviation, and assigning weight factors for weighted fusion to obtain the cooling status index.

8. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 7, characterized in that, The mechanical state analysis includes: obtaining the corresponding vibration reference amplitude, vibration reference change, allowable vibration deviation, and allowable mechanical eccentricity according to the equipment type; calculating the amplitude deviation term, change deviation term, and mechanical eccentricity term, and assigning weight factors for weighted fusion to obtain the mechanical state index.

9. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 8, characterized in that, In step three, the historical time period feature vectors are spatially clustered according to the deviation cause category to obtain at least one cluster center under each deviation cause category; the L2 distance between the current time period feature vector and each cluster center is calculated, and the average value is taken to obtain the spatial distance with the feature cluster, and the spatial distance is mapped to a spatial matching value through a formula; If a maximum spatial matching value is greater than or equal to the label threshold, the corresponding deviation reason will be output as the recognition result.

10. The precise control method for an eddy current permanent magnet speed controller based on a machine learning algorithm according to claim 9, characterized in that, In step four, for load abrupt deviation, the load excess is obtained by subtracting the load state threshold from the average load state index during the current analysis period, multiplied by the load adjustment coefficient to obtain the air gap compensation amount, and then the air gap adjustment direction is determined according to the motor speed deviation sign function to synthesize the air gap control quantity to drive the actuator; for temperature rise demagnetization deviation, the thermal excess is obtained by subtracting the thermal state threshold from the average thermal state index, multiplied by the cooling adjustment coefficient and the thermal compensation adjustment coefficient to obtain the cooling enhancement amount and the air gap thermal compensation amount, respectively, and the target cooling wind speed is obtained by adding the average actual cooling wind speed at each moment during the current analysis period to the cooling enhancement amount, and then the target air gap is obtained by subtracting the air gap thermal compensation amount from the reference air gap, and the cooling fan and actuator are driven according to the target cooling wind speed and the target air gap; for air gap offset deviation, the air gap excess is obtained by subtracting the air gap state threshold from the average air gap state index, multiplied by the air gap correction coefficient to obtain the air gap correction amount, and then the air gap is adjusted according to the air gap adjustment coefficient. The deviation sign function synthesizes the air gap control quantity. For vibration eccentricity type deviation, the mechanical excess is obtained by subtracting the mechanical state threshold from the mean mechanical state index. This excess is then multiplied by the speed limit coefficient and the eccentricity correction coefficient to obtain the speed limit adjustment and rotor alignment correction. The target speed adjustment is obtained by subtracting the speed limit adjustment from the mean actuator adjustment speed during the analysis period. The target mechanical eccentricity is obtained by subtracting the alignment correction from the mean mechanical eccentricity during the analysis period. The actuator adjustment speed and mechanical alignment are adjusted accordingly. For insufficient cooling type deviation, the cooling excess is obtained by subtracting the cooling state threshold from the mean cooling state index. This excess is then multiplied by the wind speed adjustment coefficient and the flow rate adjustment coefficient to obtain the cooling wind speed increment and the cooling flow rate increment. The target cooling wind speed is obtained by adding the cooling wind speed increment to the actual cooling wind speed during the analysis period. The target cooling flow rate is obtained by adding the cooling flow rate increment to the mean actual cooling flow rate during the analysis period. The cooling actuator is controlled accordingly. When implementing the above five types of control, the corresponding adjustment amount is first calculated, and then compared with the pre-calibrated corresponding dead zone to determine whether the adjustment amount falls into the dead zone, thereby deciding whether to implement the control.

Images