Method for detecting temperature of magnetic suspension motor rotor sheath

Abstract

The application discloses a kind of magnetic suspension motor rotor sheath temperature detection method, comprising: extracting the key structure of magnetic suspension motor, corresponding equivalent heat transfer model is constructed;In the preset time period after motor starts, the real-time loss parameter of each component is determined;Continuous multiple prediction values are arranged in time sequence, and form real-time temperature prediction value sequence;First time series prediction model and second time series prediction model are constructed, and time series temperature prediction model is constructed, for fusing two sub-model results;According to data set, the model is trained;Real-time temperature prediction value sequence is input into the time series temperature prediction model of training completion, and the temperature early prediction value sequence in next preset time period is obtained.The application generates real-time temperature sequence by constructing equivalent heat transfer model, and realizes future trend prediction by using fusion type time series model, solves the problem that rotor sheath cannot be directly temperature measured, and improves monitoring accuracy.

Description

Technical Field

[0001] This invention relates to the field of thermal management of magnetic levitation motors, and specifically to a method for detecting the temperature of the rotor sheath of a magnetic levitation motor. Background Technology

[0002] With the development of high-speed, high-efficiency, and high-reliability drive systems, magnetic levitation motors, due to their lack of mechanical friction, no lubrication required, low vibration and noise, and high-speed capability, are widely used in high-speed centrifugal compressors, flywheel energy storage, aerospace auxiliary power units, and precision manufacturing equipment. In these applications, the motor rotor is constantly rotating at high speed, making internal heat accumulation a particularly prominent issue. The rotor sheath, as a key structural component enclosing the permanent magnet, not only bears enormous centrifugal stress but also generates significant Joule heat due to eddy current losses induced by the air gap harmonic magnetic field within the sheath. If this heat cannot be dissipated in time, the sheath temperature will rise sharply, leading to irreversible demagnetization of the permanent magnet, decreased material strength, and even structural failure, severely impacting the safe operation and service life of the motor. Therefore, real-time and accurate monitoring of the rotor sheath temperature is a core technical requirement for ensuring the reliable operation of magnetic levitation motors.

[0003] However, due to the high-speed rotation of the rotor and its support by the magnetic levitation system, traditional contact temperature measurement methods (such as thermocouples and thermistors) are difficult to install on the rotor, and the wiring is difficult and susceptible to electromagnetic interference. There are also many types of non-contact temperature measurement, such as infrared temperature measurement. Patent CN113364221B discloses a motor rotor temperature detection system that uses an infrared temperature measurement module to detect the rotor temperature. However, due to limitations such as unstable emissivity of the sheath surface, obstructed view, and inaccurate positioning of high-temperature zones, the measurement accuracy and reliability for detecting the temperature of the rotor sheath of a magnetic levitation motor are relatively low.

[0004] Furthermore, the operating conditions of magnetic levitation motors are complex and variable, with frequent fluctuations in load, speed, and cooling conditions, resulting in highly nonlinear and time-varying thermal behavior. Simply relying on static thermal models or empirical formulas is insufficient to accurately reflect the actual temperature rise process. While existing technologies include temperature estimation methods based on equivalent thermal network models, these are mostly limited to steady-state or quasi-steady-state analysis, lacking the ability to dynamically predict future temperature trends and thus failing to meet the needs of intelligent early warning and proactive control.

[0005] In recent years, data-driven methods (such as neural networks and support vector machines) have been introduced into the field of motor temperature prediction. These methods estimate temperature by learning from historical operating data and possess strong nonlinear fitting capabilities. However, these methods rely on large amounts of labeled data, limiting their generalization ability and lacking physical interpretability. They are prone to prediction distortion when extrapolating operating conditions or when data is sparse. In contrast, physics-based equivalent heat transfer models can reflect the heat conduction, convection, and radiation relationships between various motor components, exhibiting good interpretability and cross-condition adaptability. However, their prediction accuracy still has room for improvement due to factors such as parameter perturbations and boundary condition uncertainties. Therefore, combining the prior knowledge of physical models with the self-learning capabilities of data-driven methods to construct a temperature monitoring system that combines accuracy and foresight has become a key direction for current technological development.

[0006] Against this backdrop, there is an urgent need for a comprehensive detection method that can overcome the limitations of traditional temperature measurement and achieve real-time estimation and future trend prediction of rotor sheath temperature. An ideal technical solution should be able to reconstruct the heat source input using measurable operating parameters without relying on rotor-side sensors, calculate the real-time temperature through a refined thermal network model, and further introduce a time-series prediction mechanism to predict future temperature rise trends in advance. Ultimately, this would form an integrated intelligent monitoring closed loop encompassing sensing, prediction, and early warning, providing reliable support for the safe operation and intelligent maintenance of magnetic levitation motors. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for detecting the temperature of the rotor sheath of a magnetic levitation motor. Under the condition that the temperature of the rotor sheath cannot be directly measured, the method can realize the real-time estimation of its temperature and accurately predict the future temperature change trend.

[0008] The objective of this invention is achieved through the following technical solution:

[0009] A method for detecting the temperature of the rotor sheath of a magnetic levitation motor, comprising:

[0010] The key structures of the magnetic levitation motor are extracted, and a corresponding equivalent heat transfer model is constructed. Within a preset time period after the motor starts, the real-time loss parameters of each component are determined based on the real-time operating parameters of the magnetic levitation motor. The real-time loss parameters are input into the equivalent heat transfer model to calculate the temperature values ​​of each part of the motor at each moment within the preset time period. The temperature value of the rotor sheath at each moment is used as the real-time temperature prediction value at that moment. Multiple real-time temperature prediction values ​​obtained continuously within the preset time period are arranged in chronological order to form a sequence of real-time temperature prediction values ​​for the rotor sheath.

[0011] Under various operating conditions, the operating parameters of the magnetic levitation motor and the corresponding temperature data of each component are obtained as the model training dataset. A first time-series prediction model and a second time-series prediction model are constructed, and a time-series temperature prediction model is constructed based on the first and second time-series prediction models. The time-series temperature prediction model is used to fuse the prediction results of the two sub-models. The first and second time-series prediction models are trained according to the model training dataset, and the fusion strategy of the time-series temperature prediction model is optimized based on the prediction outputs of the two models. The real-time temperature prediction value sequence is input into the trained time-series temperature prediction model to obtain the rotor sheath temperature advance prediction value sequence for the next preset time period. The temperature advance prediction value sequence represents the temperature change trend at multiple future moments.

[0012] The real-time temperature prediction sequence and the temperature advance prediction sequence are displayed on the central control screen, and an over-temperature warning is issued.

[0013] This invention achieves real-time estimation and future trend prediction of the rotor sheath temperature of a magnetic levitation motor through a collaborative mechanism that integrates physical modeling and data-driven approaches. First, an equivalent heat transfer model is constructed based on the motor structure. Real-time operating parameters (such as current, speed, and voltage) are used to calculate the real-time losses of each component (such as copper losses, iron losses, and sheath eddy current losses). These losses are input into the thermal model as heat sources, and the rotor sheath temperature is calculated time-by-time, forming a continuous sequence of real-time temperature predictions. This process eliminates the need for temperature sensors on the rotating rotor, achieving non-invasive temperature monitoring. Subsequently, a fusion-based time-series temperature prediction model is trained using historical operating data. This model consists of a first time-series prediction model (based on the output sequence of the physical model) and a second time-series prediction model (based on the data-driven model of operating parameters). An optimized fusion strategy balances physical laws and actual operating data characteristics. Inputting the real-time generated temperature prediction sequence into this model outputs a sequence of advance temperature predictions for the next time period, enabling accurate prediction of future temperature rise trends. Finally, the real-time and predicted temperature sequences will be displayed synchronously as curves on the central control screen. When the temperature exceeds the set threshold at any time, the system will automatically trigger an over-temperature warning.

[0014] As a preferred approach, the key structures of the magnetic levitation motor are extracted, and a corresponding equivalent heat transfer model is constructed, specifically including:

[0015] Based on the physical structure of the magnetic levitation motor, several modeling nodes for thermal analysis are constructed, wherein the modeling nodes include at least: stator core node, winding node, rotor sheath node, permanent magnet node, and bearing node;

[0016] The heat transfer relationship between the plurality of modeling nodes is determined, and the heat transfer value between two adjacent modeling nodes is represented by thermal resistance;

[0017] Based on the heat transfer relationship and the thermal resistance between each pair of modeling nodes, an equivalent heat transfer model of the magnetic levitation motor is constructed, wherein the heat source of each modeling node in the equivalent heat transfer model is represented by the node loss of the corresponding component.

[0018] As a preferred method, within a preset time period after the motor starts, the real-time loss parameters of each component are determined based on the real-time operating parameters of the magnetic levitation motor, specifically including:

[0019] Within the preset time period, the real-time operating parameters of the magnetic levitation motor at each moment are obtained, wherein the real-time operating parameters include at least: rotor speed, rotor torque, motor input current and friction torque;

[0020] Based on the real-time operating parameters, the real-time loss parameters of each component at each moment within the preset time period are calculated. The real-time loss parameters of each component are: stator core loss, rotor wind friction loss, winding copper loss, rotor sheath eddy current loss, permanent magnet eddy current loss, and bearing loss.

[0021] As a preferred embodiment, the eddy current loss of the rotor sheath is calculated using the following formula:

[0022] ;in, This refers to the eddy current loss in the rotor sheath. These are material constants; This represents the amplitude of the air gap magnetic flux density. The fundamental frequency of the stator current; The equivalent thickness of the rotor sheath; This represents the effective conductive area of ​​the rotor sheath.

[0023] As a preferred embodiment, in the equivalent heat transfer model, the radial thermal resistance between the rotor sheath node and the permanent magnet node is calculated using the following formula:

[0024] ;in, The radial thermal resistance between the rotor sheath and the permanent magnet; The radius of the permanent magnet body; The outer radius of the rotor sheath; This refers to the axial length of the rotor. is the thermal conductivity of the rotor sheath material.

[0025] As a preferred embodiment, after obtaining the real-time temperature prediction value of the rotor sheath, the method further includes: dynamically correcting the real-time temperature prediction value according to the Kalman filter algorithm to obtain the optimal estimated value of the rotor sheath temperature at time t.

[0026] The optimal temperature estimate is used as the current temperature state of the rotor sheath for subsequent temperature trend prediction or over-temperature warning judgment.

[0027] As a preferred method, the real-time temperature prediction sequence and the advance temperature prediction sequence are displayed as curves on the central control screen, and an over-temperature warning is issued, specifically including:

[0028] Within a preset time period after the motor starts, for each real-time temperature prediction value of the rotor sheath obtained, a corresponding data point is plotted on the temperature-time curve of the central control screen, and the curve display is updated in real time.

[0029] Once the complete real-time temperature prediction sequence is obtained, a future temperature trend curve in the form of a dashed line is plotted on the temperature-time curve based on the rotor sheath temperature prediction sequence for the next preset time period.

[0030] When the temperature value of any data point in the temperature-time curve is higher than the preset temperature threshold, an over-temperature warning message for the rotor sheath will be displayed on the central control screen; if the over-temperature data point is located on the future temperature trend curve, an early warning message will be issued.

[0031] As a preferred embodiment, the over-temperature warning information includes the specific value of the temperature exceeding the limit and the corresponding component location, wherein the component location is clearly marked as "rotor sleeve".

[0032] As a preferred embodiment, the first time-series prediction model is a time-series prediction model based on the output sequence of the equivalent heat transfer model, and the second time-series prediction model is a pure data-driven time-series prediction model based on the historical operating parameters of the magnetic levitation motor. The time-series temperature prediction model fuses the output results of the first time-series prediction model and the second time-series prediction model through weighted averaging, dynamic weight allocation, or machine learning fusion algorithms to generate the final temperature prediction value in advance.

[0033] As a preferred embodiment, the model training dataset includes operating parameters and corresponding temperature data under different operating conditions. The operating parameters include rotor speed, rotor torque, motor input current, friction torque, and cooling system operating parameters. The temperature data includes the equivalent temperature of the rotor sheath, which is calculated by the simulation model or obtained through indirect measurement methods.

[0034] The present invention has at least the following beneficial effects: By constructing an equivalent heat transfer model, combining real-time operating parameters to calculate the loss of the rotor sheath and generate a real-time temperature prediction value sequence, and then using a time-series temperature prediction model based on the fusion of physical output and data drive, the present invention can achieve accurate prediction of future temperature trends in advance, effectively solving the problem that the rotor sheath of the magnetic levitation motor cannot be directly measured. Combined with the curve display and over-temperature warning function on the central control screen, the accuracy, foresight and intelligence level of temperature monitoring are significantly improved. Attached Figure Description

[0035] To reveal the technical details of the embodiments of the present invention, the accompanying drawings involved in the embodiments will be briefly described below. It should be emphasized that these drawings only present several embodiments of the present invention and should not be considered as defining the scope of the invention. For those skilled in the art, other related drawings can still be derived based on these drawings without inventive effort.

[0036] Figure 1 This is a schematic flowchart of a method for detecting the temperature of the rotor sheath of a magnetic levitation motor, as described in an embodiment.

[0037] Figure 2 This is a flowchart of the TSMI monitoring process in the example. Detailed Implementation

[0038] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0039] In the following description, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. However, it should be understood that the present disclosure is not limited to the specific forms shown herein. Rather, it should be understood to encompass various variations, equivalents, and / or alternatives to the embodiments of the present disclosure. In illustrating the drawings, the same reference numerals will be used to denote similar components.

[0040] In the various embodiments of this disclosure, the terms "first," "second," "the first," or "the second" are intended to modify different components and not to indicate order and / or importance, nor do they constitute a limitation on the respective components. For example, a first user equipment and a second user equipment represent different user equipments, although they both fall under the category of user equipment. Similarly, a first component may be named a second component, and a second component may be named a first component, without changing their essential attributes within the scope of this disclosure.

[0041] In this disclosure, terminology is used to describe specific embodiments and does not constitute a limitation thereof. In this context, the use of the singular form also encompasses the plural form, unless otherwise expressly stated herein. In the course of description, terms such as “comprising” or “having” are intended to indicate the presence of features, quantities, steps, operations, structural components, parts, or combinations thereof, and do not preclude the possibility or addition of one or more other features, quantities, steps, operations, structural components, parts, or combinations thereof.

[0042] It should be clarified that while the following description provides detailed specific information to aid in a comprehensive understanding of the exemplary embodiments, those skilled in the art will recognize that the exemplary embodiments can be implemented even without these specific details. For example, the system may be illustrated using block diagrams to avoid excessive detail that could obscure the clarity of the example. In other cases, to maintain the clarity of the example, unnecessary details of well-known processes, structures, and techniques may be omitted.

[0043] like Figure 1 As shown, a method for detecting the temperature of the rotor sheath of a magnetic levitation motor includes:

[0044] The key structures of the magnetic levitation motor are extracted, and a corresponding equivalent heat transfer model is constructed. Within a preset time period after the motor starts, the real-time loss parameters of each component are determined based on the real-time operating parameters of the magnetic levitation motor. The real-time loss parameters are input into the equivalent heat transfer model to calculate the temperature values ​​of each part of the motor at each moment within the preset time period. The temperature value of the rotor sheath at each moment is used as the real-time temperature prediction value at that moment. Multiple real-time temperature prediction values ​​obtained continuously within the preset time period are arranged in chronological order to form a sequence of real-time temperature prediction values ​​for the rotor sheath.

[0045] Under various operating conditions, the operating parameters of the magnetic levitation motor and the corresponding temperature data of each component are obtained as the model training dataset. A first time-series prediction model and a second time-series prediction model are constructed, and a time-series temperature prediction model is constructed based on the first and second time-series prediction models. The time-series temperature prediction model is used to fuse the prediction results of the two sub-models. The first and second time-series prediction models are trained according to the model training dataset, and the fusion strategy of the time-series temperature prediction model is optimized based on the prediction outputs of the two models. The real-time temperature prediction value sequence is input into the trained time-series temperature prediction model to obtain the rotor sheath temperature advance prediction value sequence for the next preset time period. The temperature advance prediction value sequence represents the temperature change trend at multiple future moments.

[0046] The real-time temperature prediction sequence and the temperature advance prediction sequence are displayed as curves on the central control screen, and an over-temperature warning is issued.

[0047] This invention constructs an equivalent heat transfer model of a magnetic levitation motor, calculates the losses of each component by combining real-time acquired operating parameters, and inputs these losses as a heat source into the model. The temperature of the rotor sheath is then calculated time-by-time, forming a continuous real-time temperature prediction sequence, achieving non-contact temperature estimation without rotor-side sensors. A fusion-type time-series temperature prediction model is trained based on historical operating data. This model integrates a first time-series prediction model (learning the dynamic output of the physical model) and a second time-series prediction model (mining patterns in operating data), improving prediction accuracy through optimized fusion strategies. Inputting the real-time generated temperature sequence into this model yields the temperature change trend over a future period, forming a sequence of predicted values. Finally, the real-time and predicted temperatures are displayed synchronously as curves on the central control screen. An alarm is triggered once the temperature exceeds a set threshold. This embodiment not only overcomes the difficulty of measuring the temperature of rotating components but also achieves a leap from current state perception to future trend prediction, significantly improving the accuracy, foresight, and system intelligence of temperature monitoring.

[0048] In a preferred embodiment, the key structures of the magnetic levitation motor are extracted, and a corresponding equivalent heat transfer model is constructed, specifically including:

[0049] Based on the physical structure of the magnetic levitation motor, several modeling nodes for thermal analysis are constructed, wherein the modeling nodes include at least: stator core node, winding node, rotor sheath node, permanent magnet node, and bearing node;

[0050] The heat transfer relationship between the plurality of modeling nodes is determined, and the heat transfer value between two adjacent modeling nodes is represented by thermal resistance;

[0051] Based on the heat transfer relationship and the thermal resistance between each pair of modeling nodes, an equivalent heat transfer model of the magnetic levitation motor is constructed, wherein the heat source of each modeling node in the equivalent heat transfer model is represented by the node loss of the corresponding component.

[0052] In a magnetic levitation motor, various components generate heat during operation, which is transferred to each other through conduction and convection, ultimately affecting the temperature of critical parts such as the rotor sheath. To accurately estimate these temperatures, the motor's actual physical structure is first divided into several representative thermal analysis regions, called modeling nodes. These mainly include heat-generating points or key parts along the heat transfer path, such as the stator core, windings, rotor sheath, permanent magnets, and bearings. Each node represents a uniformly heated thermal unit, whose temperature variation is determined by the heat it generates and the heat exchange with adjacent nodes. The heat transfer path between nodes is described by thermal resistance; the greater the thermal resistance, the more difficult the heat transfer, just as resistance in a circuit impedes current. By analyzing the contact methods between components, the thermal conductivity of materials, and cooling conditions, the magnitude of the thermal resistance between adjacent nodes is determined. Then, these nodes and their inter-node thermal resistances are connected to form a network structure similar to a circuit, i.e., an equivalent heat transfer model. In this model, the heat generation of each node is represented by a heat source, which is the power loss of the corresponding component under its current operating state, such as copper losses in the windings, iron losses in the core, and eddy current losses in the rotor sheath. When the motor is running, these real-time calculated losses are input as heat sources to the corresponding nodes, and the model can simulate the distribution and flow of heat throughout the motor. This allows for the calculation of temperatures in locations that are difficult to measure directly, such as the rotor sheath, providing fundamental data for subsequent temperature monitoring and early warning.

[0053] In a preferred embodiment, within a preset time period after the motor starts, the real-time loss parameters of each component are determined based on the real-time operating parameters of the magnetic levitation motor, specifically including:

[0054] Within the preset time period, the real-time operating parameters of the magnetic levitation motor at each moment are obtained, wherein the real-time operating parameters include at least: rotor speed, rotor torque, motor input current and friction torque;

[0055] Based on the real-time operating parameters, the real-time loss parameters of each component at each moment within the preset time period are calculated. The real-time loss parameters of each component are: stator core loss, rotor wind friction loss, winding copper loss, rotor sheath eddy current loss, permanent magnet eddy current loss, and bearing loss.

[0056] During motor operation, the conversion between electrical and mechanical energy is not entirely efficient; some energy is lost as heat in various components, which is the root cause of temperature rise. To accurately estimate the temperature rise of critical components such as the rotor sheath, it is necessary to monitor the heating status of each component in real time. During a fixed period after motor startup, the system continuously collects real-time operating parameters, including rotor speed, torque, input current, and friction torque. These parameters reflect the motor's current load, motion state, and energy conversion efficiency. Based on this measurable operating data, combined with the physical characteristics and loss mechanisms of each motor component, various losses at each moment can be calculated. For example, the current in the windings generates copper losses, the stator core generates core losses under the influence of an alternating magnetic field, the high-speed rotating rotor generates wind friction losses due to friction with the surrounding air or gaseous medium, and the harmonic magnetic field in the air gap induces eddy currents in the conductive rotor sheath and permanent magnets, resulting in eddy current losses. Even though the bearings are magnetically levitated, they may still experience losses due to minor friction or electromagnetic disturbances. By calculating the losses from these different sources one by one and inputting them as heat sources into the equivalent heat transfer model, the heat distribution inside the motor can be accurately reflected, providing key input for subsequent accurate calculation of temperature changes in various parts.

[0057] In a preferred embodiment, the eddy current loss of the rotor sheath is calculated using the following formula:

[0058] ;in, The rotor sheath eddy current loss is expressed in watts (W). This is a material constant, related to the conductivity of the sheath material, and its unit is... ; The value represents the air gap magnetic flux density amplitude, in Tesla (T). The fundamental frequency of the stator current is expressed in Hertz (Hz). The equivalent thickness of the rotor sheath is expressed in meters (m). The effective conductive area of ​​the rotor sheath, expressed in square meters (m²). 2 ).

[0059] During high-speed operation, the rotor sheath experiences eddy current losses due to the alternating magnetic field in the air gap, which is one of the main reasons for its temperature rise. When alternating current is applied to the stator windings, the resulting rotating magnetic field not only drives the rotor but also passes through the conductive rotor sheath. Because the magnetic field changes continuously over time, it induces circular eddy currents within the sheath. These currents, encountering resistance as they flow through the sheath material, generate heat, resulting in eddy current losses. The magnitude of these losses is closely related to several factors: the stronger and faster the air gap magnetic field changes (i.e., the higher the frequency), the more significant the induced eddy currents. Simultaneously, the thicker the sheath or the larger its conductive area, the wider the eddy current flow path, and the greater the losses. Furthermore, the higher the conductivity of the sheath material, the easier it is to generate current, and the greater the losses. Therefore, by comprehensively considering parameters such as air gap magnetic flux density, current frequency, sheath geometry, and material properties, a computational model reflecting these physical relationships can be established to accurately estimate the heating power of the sheath under different operating conditions. This loss value, as a key heat source input into the equivalent heat transfer model, provides an important basis for accurately predicting sheath temperature changes.

[0060] In a preferred embodiment, the radial thermal resistance between the rotor sheath node and the permanent magnet node in the equivalent heat transfer model is calculated using the following formula:

[0061] ;in, The radial thermal resistance between the rotor sheath and the permanent magnet is expressed in Kelvin per watt (K / W). The radius of the permanent magnet is expressed in meters (m). The outer radius of the rotor sheath is in meters (m). This is the axial length of the rotor, in meters (m). The value is the thermal conductivity of the rotor sheath material, expressed in watts per meter Kelvin (W / (m·K)).

[0062] During motor operation, the heat generated by the rotor sheath is transferred to the internal permanent magnets via radial heat transfer. The heat transfer capacity between the two directly affects the temperature rise and safety of the permanent magnets. Since the rotor sheath and permanent magnets have a cylindrical structure, heat is mainly transferred radially from the outside to the inside. The resistance of this heat transfer path is called radial thermal resistance. The magnitude of the thermal resistance depends on the thermal conductivity of the materials, the geometry of the heat transfer path, and its length. Specifically, the stronger the thermal conductivity of the sheath material, the easier it is for heat to transfer; conversely, the thicker the sheath, i.e., the greater the radial distance from the outer layer to the inner layer, the longer the path heat needs to traverse, making heat transfer more difficult; simultaneously, the longer the axial length of the rotor, the larger the cross-sectional area for heat transfer, making it easier for heat to pass through. Therefore, when constructing an equivalent heat transfer model, the radial thermal resistance between the rotor sheath and the permanent magnets can be calculated by analyzing their contact structure, material properties, and geometry. This thermal resistance value reflects the degree of obstruction in heat transfer between the two. Using it as a connection parameter in the model can more accurately simulate how the heat generated by the sheath is gradually conducted to the permanent magnet, thus providing a reliable basis for predicting the temperature distribution of the permanent magnet and the sheath itself.

[0063] In a preferred embodiment, after obtaining the real-time temperature prediction value of the rotor sheath, the method further includes: dynamically correcting the real-time temperature prediction value according to the Kalman filter algorithm to obtain the optimal estimated value of the rotor sheath temperature at time t.

[0064] The optimal temperature estimate is used as the current temperature state of the rotor sheath for subsequent temperature trend prediction or over-temperature warning judgment.

[0065] After calculating the real-time temperature prediction of the rotor sheath using an equivalent heat transfer model, the predicted value may deviate from the actual temperature rise due to factors such as model simplification, parameter errors, or fluctuations in operating conditions. To improve the accuracy and stability of temperature estimation, the system further introduces a Kalman filter algorithm to dynamically correct the prediction results. Kalman filtering is an advanced state estimation method that combines the model's predicted values ​​with the system's dynamic characteristics to automatically identify and suppress the effects of noise and errors. By analyzing the trend and uncertainty of temperature changes, it reasonably adjusts the predicted value output by the model at each moment, thereby obtaining an optimal estimate that is closer to the true state. This optimal estimate combines the regularity of the physical model with the accuracy of data processing, effectively improving the precision of temperature judgment. The corrected temperature value, as the current true state of the rotor sheath, is used for subsequent temperature trend prediction and over-temperature early warning judgment. This not only enhances the input reliability of the time-series prediction model but also improves the timeliness and accuracy of early warnings, making the entire temperature monitoring system more intelligent and reliable.

[0066] In a preferred embodiment, to prevent unreasonable jumps in the real-time temperature prediction value due to fluctuations in input parameters or calculation errors in the equivalent heat transfer model, this invention performs dynamic correction on the real-time temperature prediction value based on physical constraints. Since the rotor sheath has a certain thermal inertia, its temperature cannot change drastically in a very short time. Therefore, this invention introduces a maximum allowable temperature rise rate as a physical constraint condition to verify and correct the rationality of the real-time temperature prediction value at each moment.

[0067] Specifically, set For the current moment The real-time predicted temperature of the rotor sheath, calculated by the equivalent heat transfer model, is expressed in degrees Celsius (°C). For the previous moment The optimal estimated value of the rotor sheath temperature, in degrees Celsius (°C); The sampling time interval is in seconds (s), typically 1 second.

[0068] Calculate the rate of temperature rise at the current moment:

[0069] ;in, This is the rate of temperature rise, expressed in degrees Celsius per second (°C / s). If If the predicted value exceeds the physically reasonable range, it is considered to be outside the reasonable range and needs to be corrected. This is the preset maximum allowable temperature rise rate, expressed in °C / s. Its value is determined based on the heat capacity of the rotor sheath material and typical heat dissipation conditions. For titanium alloy sheaths, it can be taken as... .

[0070] The corrected optimal temperature estimate is:

[0071] ;in, This is the optimal estimate of the corrected rotor jacket temperature, in °C; sgn :Sign function, when Output 1 at time. The output is -1, indicating the direction of temperature rise or fall. If... Then take it directly:

[0072] ;

[0073] Corrected optimal temperature estimate The current temperature status is used to construct a real-time temperature prediction sequence, which is then input into a time-series temperature prediction model to predict future trends.

[0074] In a preferred embodiment, the real-time temperature prediction sequence and the advance temperature prediction sequence are displayed as curves on the central control screen, and an over-temperature warning is issued, specifically including:

[0075] Within a preset time period after the motor starts, for each real-time temperature prediction value of the rotor sheath obtained, a corresponding data point is plotted on the temperature-time curve of the central control screen, and the curve display is updated in real time.

[0076] Once the complete real-time temperature prediction sequence is obtained, a future temperature trend curve in the form of a dashed line is plotted on the temperature-time curve based on the rotor sheath temperature prediction sequence for the next preset time period.

[0077] When the temperature value of any data point in the temperature-time curve is higher than the preset temperature threshold, an over-temperature warning message for the rotor sheath will be displayed on the central control screen; if the over-temperature data point is located on the future temperature trend curve, an early warning message will be issued.

[0078] To facilitate operators' intuitive understanding of the rotor sheath's temperature status, the system displays temperature information in real-time as a dynamic curve on the central control screen. After the motor starts, as the equivalent thermal model continuously calculates the real-time temperature prediction value at each moment, the system sequentially plots these data points on a temperature-time coordinate system and connects them into a continuous curve, forming a visual trajectory of the current temperature rise process, enabling real-time tracking of temperature changes. After accumulating sufficient real-time data, the time-series prediction model outputs a sequence of predicted temperature values ​​for a future period. The system then extends a dashed line or a future trend curve marked with different colors at the end of the current curve, clearly showing the possible future temperature trend. This prediction curve not only reflects the trend but, more importantly, provides a preliminary judgment basis for risk warning. Once any current or future temperature point exceeds the set safety threshold, the system immediately pops up an over-temperature warning on the central control screen. If the over-temperature occurs on the future trend curve, it means that a fault has not yet occurred but is about to occur. The system will issue an early warning, reminding operators to adjust operating conditions or take protective measures in time, thereby realizing the transformation from post-event alarm to pre-event prevention, significantly improving the safety and intelligent management level of motor operation.

[0079] In a preferred embodiment, the over-temperature warning information includes the specific value of the temperature exceeding the limit and the corresponding component location, where the component location is clearly marked as "rotor sleeve". When the system detects that the rotor sleeve temperature exceeds a preset safety threshold, it will not only trigger an over-temperature warning but also clearly display the specific over-temperature information on the central control screen, helping operators quickly locate the problem. The warning message includes not only the specific value of the current or predicted temperature but also clearly marks the location of the over-temperature component as "rotor sleeve". This accurate information output avoids the limitation of traditional alarms that only indicate "motor overheating" without being able to identify the specific heat source, enabling maintenance personnel to immediately identify that the abnormality is due to the rotor sleeve temperature, rather than a problem with the stator or bearings. Since overheating of the rotor sleeve may cause demagnetization of permanent magnets or structural damage, clearly marking its location helps to take targeted countermeasures, such as reducing the speed, adjusting the load, or strengthening cooling. At the same time, by comparing the temperature value with the threshold, operators can also judge the severity of the over-temperature and make reasonable decisions on whether to immediately shut down the machine or continue to observe it. This design improves the accuracy and usability of early warning information, realizes the dual functions of fault location and risk classification, and effectively enhances the operability and operational safety of the system.

[0080] In a preferred embodiment, the first time-series prediction model is a time-series prediction model based on the output sequence of the equivalent heat transfer model, and the second time-series prediction model is a pure data-driven time-series prediction model based on the historical operating parameters of the magnetic levitation motor. The time-series temperature prediction model fuses the output results of the first time-series prediction model and the second time-series prediction model through weighted averaging, dynamic weight allocation, or machine learning fusion algorithms to generate the final temperature prediction value in advance.

[0081] To more accurately predict future temperature trends in the rotor sheath, the system employs a dual-model fusion strategy to enhance prediction reliability. One model, the first time-series prediction model, takes the real-time temperature prediction sequence output by the equivalent heat transfer model as input, learns the temporal evolution of the results from these physical models, and thus predicts the temperature trend over a future period. This model inherits the rationality of the physical mechanism and has good interpretability. The other model, the second time-series prediction model, does not rely on the physical model but directly starts from historical data accumulated during the long-term operation of the motor (such as current, speed, and torque). It uses a data-driven approach to uncover the implicit relationship between temperature and operating conditions, capturing complex nonlinear characteristics and environmental influences in actual operation. Since both models have their advantages—the former is stable but may be limited by model simplification, while the latter is flexible but depends on data quality—the system further constructs a fusion-type time-series temperature prediction model, combining the prediction results of the two sub-models. The fusion method can be a fixed weighted average, dynamically adjusting the weights according to the current operating conditions, or even using machine learning algorithms for intelligent decision-making, ultimately generating a more accurate temperature prediction value. This fusion mechanism of physical guidance and data correction leverages both the prior advantages of theoretical models and the feedback capabilities of actual data, significantly improving the accuracy and adaptability of predictions of future temperature rise trends.

[0082] To improve the accuracy of predicting the future temperature of the rotor sheath, two complementary time-series prediction models can be fused. Taking the operation of a high-speed magnetic levitation motor under variable load conditions as an example: the system first calculates the real-time predicted temperature of the rotor sheath per second over the past 10 minutes using an equivalent heat transfer model, forming a time series; simultaneously, motor operating parameters (such as speed, current, torque, etc.) within the same time period are collected as inputs to the data-driven model. The first time-series prediction model (such as an LSTM neural network) predicts the temperature change trend over the next 5 minutes based on the temperature series output by the above physical model, yielding the following result: (At the 5th minute). The second time-series prediction model (such as the GRU network) directly predicts the temperature at the same moment based on learning from historical operating parameters and measured temperature data. Because the physical model may be underestimated due to material parameter deviations, and the data model may fluctuate significantly due to noise interference, the system uses a weighted fusion method to generate the final predicted value.

[0083] ;

[0084] in, It is the predicted final temperature after fusion, in degrees Celsius (°C). It is the predicted temperature of the first time series prediction model (based on the output sequence of the physical model), in degrees Celsius (°C). It is the predicted temperature of the second time-series prediction model (a data-driven model based on historical operating parameters), in degrees Celsius (°C). These are the fusion weights, with values ​​ranging from [0, 1]. Under the current operating conditions, the system dynamically sets these weights based on the model's historical errors. This indicates a trend of greater trust in physical models.

[0085] Substituting the values, we get:

[0086]

[0087] The fusion result retains the stability of the physical model while incorporating the data model's ability to capture actual dynamics, making it closer to the true temperature rise trend than a single model. The system will then... This serves as a 5-minute advance temperature forecast, used for trend visualization and over-temperature warning assessment. The fusion weights are continuously optimized online. The model can adapt to different operating stages (such as startup, acceleration, and steady state), improving the accuracy of predictions.

[0088] In a preferred embodiment, the model training dataset includes operating parameters and corresponding temperature data under different operating conditions. The operating parameters include rotor speed, rotor torque, motor input current, friction torque, and cooling system operating parameters. The temperature data includes the equivalent temperature of the rotor sheath, which is calculated by the simulation model or obtained through indirect measurement methods.

[0089] To train a time-series model capable of accurately predicting rotor sheath temperature changes, the system requires a large amount of data reflecting motor operating characteristics as a learning foundation. This data is organized into a model training dataset, the core of which includes two parts: first, real-time operating parameters collected under various typical operating conditions (such as starting, acceleration, full load, variable frequency operation, and different cooling conditions); and second, the corresponding rotor sheath temperature data. The operating parameters include rotor speed, torque, input current, friction torque, and the operating status of the cooling system (such as cooling airflow velocity or coolant flow rate). These parameters directly affect the motor's internal losses and heat dissipation capacity, and are the main driving factors for temperature changes. Since the rotor sheath rotates at high speed, it is impossible to directly install sensors to measure its actual temperature. Therefore, the temperature data used is not a measured value, but rather an equivalent temperature calculated through a high-precision electromagnetic-thermal coupling simulation model, or a reference value obtained through indirect measurement methods (such as arranging temperature measurement points in close proximity and combining them with heat conduction back-calculation). Although these equivalent temperature data are not direct readings, they accurately reflect the temperature rise pattern of the sheath under different operating conditions. By training with a large number of such input parameter and output temperature sample pairs, the first and second time-series prediction models can learn the dynamic characteristics of temperature changes under complex operating conditions, thereby accurately predicting future temperature trends based on real-time inputs in actual operation, ensuring that the models have good generalization ability and engineering applicability.

[0090] In a preferred embodiment, training the time-series temperature prediction model includes: using Bayesian optimization or grid search methods to fine-tune the model's learning rate, time window length, and number of hidden layer neurons to minimize the mean square error between the predicted temperature and the actual temperature.

[0091] To improve the prediction accuracy of time-series temperature prediction models in practical applications, fine-tuning of their internal parameters is necessary; this process is called model training optimization. After initial data preparation and model structure building, the system further employs intelligent tuning methods to automatically optimize several key hyperparameters, including: learning rate (controlling the speed of model learning), time window length (determining how much historical data the model references), and the number of hidden layer neurons (affecting model complexity and fitting ability). If these parameters are not set appropriately, the model may experience problems such as slow learning, large prediction bias, or overfitting. Therefore, the system introduces automated hyperparameter tuning techniques such as Bayesian optimization or grid search, repeatedly testing the effects of different parameter combinations on a large amount of historical data. These methods can systematically evaluate the model's prediction performance under each set of parameters, using the error between the predicted temperature and the actual equivalent temperature as the evaluation criterion. The goal is to minimize the overall error, especially reducing indicators such as mean squared error, which reflect average deviation. Through this optimization process, the model can find the parameter configuration that is most suitable for the current motor characteristics, thereby improving its ability to capture future temperature change trends and ensuring that it can output stable and accurate prediction results under various operating conditions, providing a reliable basis for subsequent early warning and decision-making.

[0092] We need to train a temperature prediction model for a magnetic levitation motor. We have a set of historical data showing operation at different speeds and loads, including recorded current-speed per second, cooling status, and corresponding equivalent rotor sheath temperatures. The model uses an LSTM neural network architecture, and its performance is affected by multiple hyperparameters. To find the optimal configuration, we use a grid search method to try different combinations within a limited range.

[0093] Candidate values ​​for three key parameters are set: learning rate (0.001 or 0.01; too small, learning is slow; too large, oscillations are likely); time window length (60 seconds or 120 seconds; determines whether the model references data from the past 1 or 2 minutes); and number of hidden layer neurons (32 or 64; more neurons result in stronger fitting but may lead to overfitting). Grid search will test these parameters one by one. There are several combinations. For example, the first combination has a learning rate of 0.001, a window length of 60 seconds, and 32 neurons. After training the model with these parameters, the model is asked to predict the temperature in the validation set, and the mean squared error (MSE) between the predicted value and the actual equivalent temperature is calculated. Then, the training and evaluation are repeated with a different set of parameters. After all the tests, it was found that when using a learning rate of 0.001, a time window of 120 seconds, and 64 neurons, the model had the smallest MSE, indicating that this combination provided the most accurate prediction. Therefore, the system selected this set as the final model parameters. Compared to manual trial and error, this method is more systematic and reliable, effectively improving the model's ability to predict future temperature trends and ensuring timely detection of abnormal temperature rises during actual operation.

[0094] In a preferred embodiment, the acquisition of the real-time temperature prediction value sequence adopts a sliding time window, and the step size of the sliding window is dynamically adjusted according to the temperature change rate of the rotor sheath.

[0095] During the continuous monitoring of rotor sheath temperature, the real-time calculated temperature values ​​need to be arranged into a time sequence for subsequent time-series prediction models. This process employs a sliding time window approach for data acquisition, retaining only the most recent temperature data at a time and continuously updating the oldest data over time, while maintaining a fixed length of the input sequence. This ensures that the prediction model always makes judgments based on the latest operating conditions. To adapt to the varying rates of temperature change under different operating conditions, the update step size of the sliding window is not fixed but dynamically adjusted according to the rate of change of the rotor sheath temperature. When the motor is in the startup or load surge phase, and the temperature rises rapidly, the system automatically shortens the sliding window step size and increases the data acquisition frequency, thereby capturing key changes during the temperature rise process more intensively. Conversely, during steady-state operation and when temperature changes are gradual, the step size is appropriately increased, reducing the update frequency and minimizing unnecessary computational burden. This dynamic adjustment mechanism ensures no distortion or missed detection during drastic temperature changes and conserves resources during stable phases, improving the system's response sensitivity and operating efficiency, and making temperature prediction more accurate and intelligent.

[0096] In a preferred embodiment, the real-time operating parameters also include the motor's input voltage and ambient temperature.

[0097] During motor operation, in addition to conventional parameters such as speed and current, input voltage and ambient temperature also significantly affect the temperature rise of the rotor sheath, and are therefore included in the real-time operating parameter acquisition scope. Changes in input voltage directly affect the electromagnetic field distribution and current response inside the motor, thereby altering the copper losses in the windings, the iron losses in the core, and the harmonic content of the air gap magnetic field, ultimately affecting the magnitude of eddy current losses in the rotor sheath. Excessive voltage or large voltage fluctuations may lead to magnetic circuit saturation, increasing additional losses and exacerbating sheath heating. Ambient temperature reflects the external heat dissipation conditions of the motor. If the ambient temperature rises, the overall cooling efficiency of the motor decreases, heat accumulation accelerates, and even under unchanged operating conditions, the sheath temperature may rise significantly. By acquiring these two parameters in real time, the system can more comprehensively understand the internal and external factors affecting temperature changes, and more accurately reflect actual thermal behavior in the equivalent heat transfer model and time-series prediction model. This multi-parameter collaborative analysis improves the accuracy of temperature estimation and trend prediction, especially significantly enhancing the adaptability and reliability of the monitoring system during operation under varying conditions or harsh environments.

[0098] In a preferred embodiment, when displaying the real-time temperature prediction sequence and the temperature advance prediction sequence on the central control screen, they can optionally be processed by moving average or low-pass filtering to improve the curve display effect.

[0099] When displaying real-time temperature predictions and future temperature trend curves on the central control screen, short-term fluctuations due to model calculations or data volatility can cause spikes or jumps in the temperature curve, affecting visual observation. To make the curve smoother and facilitate operators' intuitive judgment of temperature change trends, the system can selectively apply moving averages or low-pass filtering to the displayed data. This processing effectively reduces high-frequency noise and instantaneous fluctuations, preserving the overall temperature rise trend and making the curve more continuous and stable. It is important to note that this processing is only for interface display optimization and does not change the original prediction data or early warning judgment logic. In other words, over-temperature warnings still use the unsmoothed original values, ensuring the accuracy and timeliness of the alarm. This design, which provides smooth display and accurate judgment, improves the readability and aesthetics of the human-machine interface without affecting the reliability of the system's core functions, balancing user experience and operational safety.

[0100] In a preferred embodiment, the high-speed magnetic levitation motor is used in an industrial compressor system. Its rotor sheath is made of titanium alloy, making it impossible to install a temperature sensor. To achieve real-time monitoring and over-temperature warning of the sheath temperature, the method described in this invention is used for temperature detection. The system first establishes an equivalent heat transfer model based on the motor structure, dividing the rotor part into multiple thermal nodes, including the rotor sheath and permanent magnets, and determining the heat transfer path and thermal resistance between them. For example, heat is transferred between the sheath and the permanent magnet through radial heat conduction, and its thermal resistance is calculated based on the material's thermal conductivity, sheath thickness, and axial length. For instance, the system first establishes an equivalent heat transfer model based on the motor's physical structure, dividing the rotor region into three key thermal nodes: Node A: Rotor sheath (outer conductive structure, subjected to eddy current heating); Node B: Permanent magnet (internal magnetic material, susceptible to demagnetization at high temperatures); Node C: Rotor core / shaft (support structure, involved in heat conduction). These three nodes are connected by thermal resistance. There is a radial thermal resistance R1 between the sheath and the permanent magnet, determined by the sheath thickness, material thermal conductivity, and axial length; a radial thermal resistance R2 between the permanent magnet and the rotor core; and a convective thermal resistance R3 between the outer surface of the sheath and the cooling airflow, the magnitude of which depends on the cooling air velocity. The model resembles a thermal circuit, where the temperature change of each node is determined by the heat generated by itself (heat source) and the heat exchange with adjacent nodes. For example, when the motor is running, the air gap magnetic field induces eddy currents in the sheath, generating eddy current losses of 80W, which act as a heat source at node A; the permanent magnet also has 10W of eddy current losses at node B. The system automatically calculates how heat is transferred layer by layer based on the current heat generation of each component (e.g., the sheath generates 82 watts of heat) and the thermal resistance between them (e.g., the ease of heat transfer between the sheath and the permanent magnet). This results in the temperature change of the sheath, permanent magnet, and other parts every second, and the final output is the temperature value at each moment, achieving dynamic tracking of the entire process of rotor sheath temperature rise.

[0101] The temperature of the rotor sheath is dynamically calculated second-by-second using an equivalent heat transfer model, rather than relying on direct sensor measurement. Taking the 60th second of motor operation as an example, the system first calculates the eddy current loss caused by the air gap harmonic magnetic field induction in the rotor sheath at the current moment as 75W based on the real-time collected speed (25,000 rpm), current (100A), and cooling airflow (1.0 m³ / min), combined with the motor parameter model. This loss is used as the heat source input to the sheath node. The model divides the rotor structure into multiple thermal nodes. There is a radial heat conduction path between the sheath and the permanent magnet, with a thermal resistance of 0.4 kNW. Heat dissipation occurs through convection between the outer surface of the sheath and the cooling airflow, with a convection thermal resistance of 0.6 kNW. Based on the ratio of these two thermal resistances, the system determines the heat distribution trend: the smaller the thermal resistance, the easier the heat transfer, therefore more heat is preferentially transferred inward to the permanent magnet, and some heat is dissipated outward to the cooling airflow. Of the 75W of heat, 45W is transferred to the permanent magnet, 25W is carried away by the cooling air, and the remaining approximately 5W is temporarily accumulated inside the jacket due to heat transfer lag. Since the heat capacity of the jacket material (e.g., titanium alloy) is known (e.g., equivalent heat capacity of 7.1),... The model can further calculate that the 5W of heat will cause the sheath temperature to rise by approximately 0.7°C within 1 second. Combining this with the temperature value of 138.2°C at the 59th second, the final sheath temperature at the 60th second is 138.9°C. This calculation process is repeated every second during motor operation, using the temperature of the previous moment as the initial state and continuously updating to form a continuous, dynamic, real-time temperature prediction sequence, achieving high-precision, non-invasive tracking of the rotor sheath temperature rise process.

[0102] Once the motor starts, the control system collects real-time operating parameters every second, including speed, current, torque, input voltage, and cooling airflow, and calculates the real-time losses of each component: copper losses in the windings, iron losses in the stator core, and significant eddy current losses due to eddy currents induced by changes in the air gap magnetic field in the high-speed rotating rotor sheath. These losses are used as the heat source input in an equivalent heat transfer model. The model solves for the temperature response of each node every second, obtaining the rotor sheath temperature value for each second, such as 125.3°C in the 10th second, 126.1°C in the 11th second, and so on. The data for 120 consecutive seconds constitutes a real-time temperature prediction sequence, reflecting the current temperature rise process.

[0103] Meanwhile, the system also constructs two complementary time-series prediction models for future trend analysis. The first time-series prediction model (such as an LSTM network) is trained to learn the temperature sequence patterns output by the physical model mentioned above. It knows that during a similar load increase process, the sheath temperature typically rises by about 8°C per minute. The second time-series prediction model (another LSTM) does not rely on the physical model but learns directly from historical data. For example, in the past three months, when the engine speed accelerates from 10,000 rpm to 30,000 rpm with insufficient cooling, the sheath temperature often exceeds 180°C within 5 minutes. Both models are trained using historical operating parameters and corresponding simulated temperature data covering various operating conditions (start-stop, acceleration, full load, overload), and their prediction errors are minimized through Bayesian optimization.

[0104] Subsequently, the system constructs a time-series temperature prediction model based on these two models. This model does not simply replace one of them, but rather fuses their outputs through dynamic weighting. For example, during the current stable operation phase, the physical model is more stable, and its weight is set to 0.7; while under sudden load increases, the data-driven model is better able to capture abnormal temperature rises, and its weight is automatically adjusted to 0.6. When the latest real-time temperature prediction sequence (120 data points from the last 2 minutes) is input into this fused model, the system outputs a sequence of advance predictions for the rotor sheath temperature within the next 3 minutes. For example, it predicts that the temperature will reach 162°C after 1 minute, 175°C after 2 minutes, and 191°C after 3 minutes.

[0105] These data are plotted in real time on the central control screen: actual temperatures are represented by solid lines, and future forecasts are displayed using dashed lines. When the system determines that the predicted temperature will exceed the safety threshold of 190°C in 3 minutes, a red warning message immediately pops up: "Rotor sleeve over-temperature warning: Temperature expected to reach 191°C in 3 minutes," with the location marked as "rotor sleeve." Operators can then use this information to reduce the load or increase cooling in advance to prevent accidents.

[0106] This method achieves non-contact temperature estimation through physical modeling, and then achieves accurate trend prediction through dual-model fusion, truly realizing the leap from current perception to future prediction, and significantly improving the safety and intelligence level of magnetic levitation motor operation.

[0107] In a preferred embodiment, to achieve quantitative classification and forward-looking decision support for over-temperature early warning, this invention proposes a rotor sheath temperature safety margin index (TSM1) to comprehensively assess the margin of current and future temperature conditions relative to the material's safety limit, avoiding false alarms or missed alarms caused by single threshold alarms.

[0108] as follows:

[0109] Among them, TSMI For at any time Temperature safety margin index; This refers to the maximum allowable operating temperature of the rotor sheath material, expressed in degrees Celsius (°C). For titanium alloy sheaths, this is typically 300°C, while for high-strength stainless steel, it can be [missing value]. ; For at any time The predicted value of the rotor sheath temperature (which can be a real-time prediction or an advance prediction), in degrees Celsius (°C); The ambient temperature of the motor's operating environment, in degrees Celsius (°C), is collected in real time by a sensor.

[0110] Risk levels are determined based on TSMI values ​​(see [link]). Figure 2 ):

[0111] When TSMI > 0.6: the system is in a safe operating zone and displays green.

[0112] When 0.4 <TSMI The warning area, displayed in yellow, indicates "Temperature is high, please monitor the trend."

[0113] When 0.2 <TSMI Warning zone, triggering Level 1 alert: "Rotor sleeve temperature rises rapidly, it is recommended to reduce load," displayed in orange;

[0114] When 0 <TSMI Danger zone, triggering Level 2 warning: "Overheating expected, please prepare to reduce load or shut down," displaying a flashing red light.

[0115] When TSMI The limit has been exceeded. Emergency protection action will be triggered immediately, forcibly reducing power or shutting down the machine.

[0116] Furthermore, the system can also calculate the minimum safety margin for a future point in time:

[0117] ;in For the predicted time period (e.g., 5 minutes), This refers to the current moment. If TSMI This allows for the early issuance of long-term high-risk warnings. The index normalizes temperature conditions to the material's capability boundary, overcoming the problem of inconsistent sensitivity of a fixed threshold under different ambient temperatures. For example, in high-temperature environments during summer (… Even if the temperature of the sheath reaches Its safety margin is still better than that under low winter temperatures. Performance at the same temperature.

[0118] In summary, this invention provides a non-invasive temperature detection method for the rotor sheath of a magnetic levitation motor. By integrating physical modeling and data-driven technologies, a closed-loop monitoring system is constructed, encompassing real-time estimation, trend prediction, and intelligent early warning. First, an equivalent heat transfer model is established based on the key structure of the motor. Combined with real-time operating parameters, the losses of each component are dynamically calculated and used as the heat source input model to solve for the rotor sheath temperature time-by-time, forming a continuous sequence of real-time temperature prediction values. This achieves high-precision temperature estimation without the need for sensors on the rotating side. Furthermore, Kalman filtering and physical constraint mechanisms are introduced to dynamically correct the predicted values, improving the stability and reliability of the state estimation. Based on this, a fusion-type time-series temperature prediction model is constructed, consisting of a first time-series prediction model (based on the output of the physical model) and a second time-series prediction model (based on a data-driven model of historical operating parameters). The prediction performance is optimized through weighted fusion or machine learning strategies, enabling accurate early prediction of future temperature rise trends. The system displays real-time and predicted temperatures as curves on the central control screen, and combines over-temperature threshold judgment and a tiered early warning mechanism to support pre-emptive warnings and proactive intervention. In particular, by introducing the Temperature Safety Margin Index (TSMI), it achieves a normalized risk assessment that considers ambient temperature and material limits, improving the scientific rigor and adaptability of the early warning system. This invention effectively solves the problems of difficult temperature measurement of high-speed rotating components, strong nonlinear thermal behavior, and traditional alarm lag. It combines physical interpretability with data self-adaptability, significantly improving the temperature monitoring accuracy, foresight, and operational safety of magnetic levitation motors under complex operating conditions, providing key technical support for the intelligent operation and maintenance of high-end equipment.

[0119] Although preferred embodiments of the present invention have been described in detail, those skilled in the art, upon understanding the basic innovative concept, can still make various modifications and extensions to it. Therefore, the appended claims are intended to cover the preferred embodiments and all equivalent variations, alternatives, and improvements falling within the technical essence and protection scope of the present invention. The above description is merely illustrative and does not constitute a limitation of the present invention. Any modifications, variations, equivalent substitutions, or improvements made within the technical scope defined by the spirit and claims of the present invention should be considered as falling within the protection scope of the present invention.

Claims

1. A method for detecting the temperature of the rotor sheath of a magnetic levitation motor, characterized in that, include: The key structures of the magnetic levitation motor are extracted, and the corresponding equivalent heat transfer model is constructed. During a preset time period after the motor starts, the real-time loss parameters of each component are determined based on the real-time operating parameters of the magnetic levitation motor. The real-time loss parameters are input into the equivalent heat transfer model to calculate the temperature values ​​of each part of the motor at each moment within the preset time period. The temperature value of the rotor sheath at each moment is used as the real-time temperature prediction value at that moment. The multiple real-time temperature prediction values ​​obtained continuously within the preset time period are arranged in chronological order to form a sequence of real-time temperature prediction values ​​for the rotor sheath. Under various operating conditions, the operating parameters of the magnetic levitation motor and the corresponding temperature data of each component are obtained as the model training dataset. A first time-series prediction model and a second time-series prediction model are constructed, and a time-series temperature prediction model is constructed based on the first and second time-series prediction models. The time-series temperature prediction model is used to fuse the prediction results of the two sub-models. The first and second time-series prediction models are trained according to the model training dataset, and the fusion strategy of the time-series temperature prediction model is optimized based on the prediction outputs of the two models. The real-time temperature prediction value sequence is input into the trained time-series temperature prediction model to obtain the rotor sheath temperature advance prediction value sequence for the next preset time period. The temperature advance prediction value sequence represents the temperature change trend at multiple future moments. It also includes: dynamic correction of real-time temperature predictions based on physical constraints, resulting in the optimal temperature estimate after correction. ； in, This is the optimal estimate of the corrected rotor jacket temperature; For the current moment The real-time predicted temperature of the rotor sheath calculated by the equivalent heat transfer model; The preset maximum allowable temperature rise rate; sgn represents the sampling time interval. For a sign function, when Output 1 at time. The output is -1 to indicate the direction of temperature rise or fall; For the rate of temperature rise, if Then take directly ; Corrected optimal estimate of rotor jacket temperature As the current temperature state, it is used to construct a sequence of real-time temperature prediction values; The real-time temperature prediction sequence and the advance temperature prediction sequence are displayed on the central control screen, and an over-temperature warning is issued; The key structures of the magnetic levitation motor are extracted, and the corresponding equivalent heat transfer model is constructed. In the equivalent heat transfer model, the radial thermal resistance between the rotor sheath node and the permanent magnet node is calculated using the following formula: ； in, The radial thermal resistance between the rotor sheath and the permanent magnet; The radius of the permanent magnet body; The outer radius of the rotor sheath; This refers to the axial length of the rotor. The thermal conductivity of the rotor sheath material; Within a preset time period after the motor starts, the real-time loss parameters of each component are determined based on the real-time operating parameters of the magnetic levitation motor, specifically including: Within a preset time period, the real-time operating parameters of the magnetic levitation motor at various moments are acquired. The real-time operating parameters include at least: rotor speed, rotor torque, motor input current, and friction torque. Based on the real-time operating parameters, the real-time loss parameters of each component at each moment within the preset time period are calculated. The real-time loss parameters of each component are: stator core loss, rotor wind friction loss, winding copper loss, rotor sheath eddy current loss, permanent magnet eddy current loss, and bearing loss. The rotor sheath eddy current loss is calculated using the following formula: ； in, This refers to the eddy current loss in the rotor sheath. These are material constants; This represents the amplitude of the air gap magnetic flux density. The fundamental frequency of the stator current; The equivalent thickness of the rotor sheath; This represents the effective conductive area of ​​the rotor sheath.

2. The method for detecting the temperature of the rotor sheath of a magnetic levitation motor according to claim 1, characterized in that, The real-time temperature prediction sequence and the advance temperature prediction sequence are displayed as curves on the central control screen, and an over-temperature warning is issued, specifically including: Within a preset time period after the motor starts, for each real-time temperature prediction value of the rotor sheath obtained, a corresponding data point is plotted on the temperature-time curve of the central control screen, and the curve display is updated in real time. Once the complete real-time temperature prediction sequence is obtained, a future temperature trend curve in the form of a dashed line is plotted on the temperature-time curve based on the rotor sheath temperature prediction sequence for the next preset time period. When the temperature value of any data point in the temperature-time curve is higher than the preset temperature threshold, an over-temperature warning message for the rotor sheath will be displayed on the central control screen; if the over-temperature data point is located on the future temperature trend curve, an early warning message will be issued.

3. The method for detecting the temperature of the rotor sheath of a magnetic levitation motor according to claim 2, characterized in that, The over-temperature warning information includes the specific value of the temperature exceeding the limit and the corresponding component location, wherein the component location is clearly marked as "rotor sleeve".

4. The method for detecting the temperature of the rotor sheath of a magnetic levitation motor according to claim 1, characterized in that, The first time-series prediction model is a time-series prediction model based on the output sequence of the equivalent heat transfer model, and the second time-series prediction model is a pure data-driven time-series prediction model based on the historical operating parameters of the magnetic levitation motor. The time-series temperature prediction model integrates the output results of the first time-series prediction model and the second time-series prediction model through weighted averaging, dynamic weight allocation or machine learning fusion algorithms to generate the final temperature prediction value in advance.

5. The method for detecting the temperature of the rotor sheath of a magnetic levitation motor according to claim 1, characterized in that, The model training dataset includes operating parameters and corresponding temperature data under different operating conditions. The operating parameters include rotor speed, rotor torque, motor input current, friction torque, and cooling system operating parameters. The temperature data includes the equivalent temperature of the rotor sheath, which is calculated by the simulation model or obtained through indirect measurement methods.

Images