Method and system for monitoring efficient heat dissipation of axial magnetic field motor winding

Abstract

The invention provides an efficient heat dissipation monitoring method and system for an axial magnetic field motor winding. The efficient heat dissipation monitoring method comprises the following steps that multiple sets of composite sensor units are arranged on the axial section and the radial end of the axial magnetic field motor winding in a honeycomb curved surface array mode; according to the invention, the honeycomb curved surface array is formed by not less than 24 groups of composite sensor units, multi-physical field parameters such as temperature and heat flux are collected in real time at high sampling frequency, the data quality is improved by combining an adaptive filtering algorithm, and the thermal state of the motor is accurately described; through a dynamic thermal resistance network model based on CFD simulation fitting and an adaptive Kalman filtering technology, high-precision real-time calculation of a three-dimensional temperature gradient field and equivalent thermal resistance is realized, and a high-refresh-rate visual cloud picture is generated; 12-dimensional characteristic parameters are extracted, double-layer LSTM network training is applied, and Dropout is combined to suppress overfitting, so that the fault recognition accuracy rate exceeds 98%, the recognition delay is smaller than 500 ms, and the defects of a traditional monitoring method in the aspects of data acquisition, model prediction and fault diagnosis are effectively overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal management of axial magnetic field motors, and in particular to a method and system for efficiently monitoring the heat dissipation of windings of an axial magnetic field motor. Background Art

[0002] Axial magnetic field motors are widely used in high-end equipment such as new energy vehicle drives, aerospace electric propulsion, and industrial robots due to their advantages such as compact axial dimensions, high power density, and excellent torque characteristics. However, their flat disc structure results in tight axial stacking of the windings and complex radial heat dissipation paths. In addition, the winding losses are concentrated during high power density operation, which can easily lead to local overheating problems, thereby accelerating the aging of the insulation layer and inducing faults such as inter-turn short circuits, seriously restricting the reliability and life of the motor. Therefore, it is of great significance to efficiently monitor the heat dissipation status of the axial magnetic field motor windings. However, there are still certain problems in the current monitoring of the heat dissipation of the windings of axial magnetic field motors: First, traditional monitoring methods mostly use single-point temperature sensors (such as thermocouples), which can only obtain local temperature data and cannot fully characterize the three-dimensional thermal field distribution and key parameters such as heat flux density and insulation stress. It is difficult to accurately locate hot spots and heat conduction bottlenecks. Second, existing thermal resistance network models are mostly constructed based on static parameters, without fully considering the dynamic coupling relationship between the cooling medium flow rate, density, and insulation material properties. This results in large prediction deviations when the operating conditions change, and the model cannot reflect the nonlinear evolution of the winding heat dissipation state in real time; Third, the characteristics of early faults such as insulation aging and cooling system blockage are hidden. Traditional methods that rely on threshold judgments have difficulty effectively extracting correlation characteristics between multi-dimensional parameters, which can easily lead to fault identification delays and fail to meet the high-reliability operation requirements of motors. Therefore, an efficient heat dissipation monitoring method and system for axial magnetic field motor windings are proposed. Summary of the Invention

[0003] In view of this, the embodiments of the present invention hope to provide an efficient heat dissipation monitoring method and system for axial magnetic field motor windings to solve or alleviate the technical problems existing in the prior art and at least provide a beneficial option.

[0004] To solve the above technical problems, a technical solution adopted in this application is: a method for efficiently monitoring heat dissipation of axial magnetic field motor windings, comprising the following steps: Step 1: Arrange multiple sets of composite sensor units in a honeycomb curved array on the axial cross section and radial end of the axial magnetic field motor winding; Step 2: Use a composite sensor unit to collect multi-dimensional data including winding temperature, heat flux density, insulation layer stress, and cooling medium flow rate in real time, with a sampling frequency of no less than 100 Hz; Step 3: The collected multi-dimensional data is transmitted to the edge computing processing unit, and noise is suppressed by a combination of adaptive median filtering and wavelet denoising algorithm; Step 4: pre-processing the filtered data based on a preset dynamic thermal resistance network model; Step 5: Use the adaptive Kalman filter algorithm to perform spatiotemporal fusion on the pre-processed multi-dimensional data, build a digital twin model of the winding thermal distribution, calculate the three-dimensional temperature gradient field and equivalent thermal resistance in real time, and generate heat dissipation status data; Step 6: Build and train an LSTM neural network model to analyze the heat dissipation status data, extract multi-dimensional feature parameters, and identify fault characteristics such as cooling system blockage, insulation layer aging, or winding turn short circuit; Step 7: Classify the fault level according to the identified fault characteristic information and trigger the three-level warning mechanism. At the same time, encrypt the fault characteristic information and upload it to the cloud diagnosis platform.

[0005] As a further preferred embodiment of the present technical solution, in step 4, the dynamic thermal resistance network model is used to characterize the coupling relationship between the winding heat conduction path and the cooling medium; The dynamic thermal resistance network model is: ; in, is the coupling correction function of the cooling medium flow rate, density and insulation dielectric constant, is the real-time total loss of the winding; The coupling correction function The results are obtained by fitting more than 2000 sets of CFD simulation data, specifically: ; in, is the reference flow rate of the cooling medium, is the base density, is the dielectric constant of insulation.

[0006] As a further preferred embodiment of the present technical solution, in step 1, the composite sensor units are at least 24 groups; the composite sensor units integrate micro-thermocouples, thin-film heat flux density sensors, fiber Bragg grating pressure sensors, and micro-ultrasonic Doppler flowmeters; the micro-thermocouples are densely distributed on the winding conductor layer with a spacing of 0.3 mm, with an accuracy of ±0.1°C and a response time of <50 ms; the thin-film heat flux density sensors are attached to the surface of the insulating layer, with a range of 0-800 W / m² and a resolution of 0.5 W / m²; The composite sensor unit adopts a polyimide-based flexible sensor array with a thickness of ≤0.5mm at the winding end. The flexible sensor array fits the spiral winding curved surface structure, the sensor spacing is ≤0.2mm, and the bending radius is ≥5mm.

[0007] As a further preferred embodiment of the present technical solution, in step 5, the step of constructing a digital twin model of the winding thermal distribution specifically includes: Divide the winding into three-dimensional grid cells and assign unique coordinates to each grid cell , and establish the heat conduction path association; Define the four-dimensional state vector: ; in, is the grid cell temperature, is the heat flux density, is the insulation stress, is the equivalent thermal resistance; Establish the state transfer equation: ; Among them, A is the heat transfer coefficient matrix, B is the cooling medium flow rate control matrix, is the real-time flow velocity input vector, is the process noise vector that obeys the normal distribution; Establish the measurement equation: ; in, is the sensor measurement vector, H is the spatial sensitivity matrix of the sensor array, is the measurement noise that obeys the normal distribution; The Kalman gain is calculated iteratively using the adaptive Kalman filter algorithm: ; in, is the forecast error covariance matrix, is the measurement noise covariance matrix; Based on the Kalman gain, the sensor measured value and the model predicted value are dynamically fused to update the state vector: ; At the same time, the error covariance matrix is ​​updated: ; in, is the identity matrix; Based on the updated state vector, a three-dimensional temperature gradient field cloud map and a heat flow vector distribution map are generated.

[0008] As a further preferred embodiment of the present technical solution, in step 6, the constructing and training of the LSTM neural network model specifically includes the following steps: Step 601: Acquire sample data including normal working conditions, cooling system blockage, insulation layer aging, and winding turn-to-turn short circuit; the number of sample data is ≥ 3000 groups; Step 602: normalize the sample data and extract multi-dimensional feature parameters as input vectors; Step 603: Build a two-layer LSTM network, use a dropout rate of 0.2 to suppress overfitting, minimize the cross entropy loss function using the Adam optimizer, and train with an early stopping mechanism until the validation set accuracy is greater than 98% and the loss value is less than 0.03; Step 604: Deploy the trained model to the edge computing unit to perform fault feature recognition on the real-time heat dissipation status data; The multidimensional characteristic parameters include temperature gradient, heat flux density change rate, insulation layer stress fluctuation amplitude, cooling medium flow rate deviation, equivalent thermal resistance change rate, temperature-heat flux coupling coefficient, stress-flow velocity correlation coefficient, three-dimensional temperature field entropy value, heat flux vector divergence, insulation dielectric constant drift, winding copper loss power and cooling system pressure drop.

[0009] As a further preferred embodiment of the present technical solution, in step seven, the three-level warning mechanism includes a heat dissipation efficiency adjustment strategy linked with the motor cooling system and the control system.

[0010] As a further preferred embodiment of the present technical solution, in step three, the edge computing processing unit is equipped with an NPU neural network processor with a computing power ≥ 2TOPS, supports real-time parallel operation of the dynamic thermal resistance model and the LSTM algorithm, and generates a winding temperature cloud map with a refresh rate ≥ 100Hz.

[0011] To solve the above technical problems, another technical solution adopted in this application is: an axial magnetic field motor winding high-efficiency heat dissipation monitoring system, the system comprising: a sensor array deployment module, a multi-dimensional data acquisition module, a data transmission and filtering module, a dynamic thermal resistance modeling module, a digital twin construction module, an intelligent fault diagnosis module, and an early warning and control module; The sensor array deployment module is configured to arrange no less than 24 groups of composite sensor units in a honeycomb curved array on the axial cross section and radial end of the axial magnetic field motor winding; The multi-dimensional data acquisition module is configured to collect multi-dimensional data of the winding in real time through the composite sensor unit, with a sampling frequency of not less than 100 Hz; the multi-dimensional data includes the temperature, heat flux density, insulation layer stress and cooling medium flow rate of the winding; The data transmission and filtering module is configured to transmit the collected multi-dimensional data to the edge computing processing unit and perform noise suppression through a combination of adaptive median filtering and wavelet denoising algorithm; The dynamic thermal resistance modeling module is configured to process the filtered data based on a preset dynamic thermal resistance network model; The digital twin construction module is configured to perform spatiotemporal fusion of the pre-processed multi-dimensional data through an adaptive Kalman filter algorithm, construct a digital twin model of the winding thermal distribution, calculate the three-dimensional temperature gradient field and equivalent thermal resistance in real time, and generate heat dissipation status data; The intelligent fault diagnosis module is configured to build and train an LSTM neural network model, analyze the heat dissipation status data, extract multi-dimensional feature parameters, and identify fault characteristic information such as cooling system blockage, insulation layer aging, or winding turn short circuit; The early warning and control module is configured to classify the fault level according to the identified fault feature information and trigger a three-level early warning mechanism, while encrypting the fault feature information and uploading it to the cloud diagnosis platform.

[0012] As a further preferred embodiment of the present technical solution, the dynamic thermal resistance modeling module, the digital twin construction module and the intelligent fault diagnosis module are integrated into the edge computing processing unit.

[0013] As a further preferred embodiment of the present technical solution, the data transmission and filtering module adopts a dual redundant transmission architecture.

[0014] The embodiment of the present invention adopts the above technical solution, which has the following advantages: 1. The present invention uses no less than 24 groups of composite sensor units to form a honeycomb curved surface array, which collects temperature, heat flux and other multi-physical field parameters in real time at a high sampling frequency of ≥100Hz. It combines adaptive filtering algorithms to improve data quality, accurately depict the thermal state of the motor, and ensure accurate and reliable data. 2. The present invention uses a dynamic thermal resistance network model based on CFD simulation fitting and adaptive Kalman filtering technology to update parameters every 10ms, achieving high-precision real-time calculation of the three-dimensional temperature gradient field and equivalent thermal resistance, and generating a high-refresh-rate visual cloud map; 3. By extracting 12-dimensional feature parameters and using a two-layer LSTM network for training, combined with Dropout to suppress overfitting, the present invention can achieve a fault recognition accuracy of over 98% and a recognition delay of less than 500ms, effectively solving the shortcomings of traditional monitoring methods in data collection, model prediction and fault diagnosis.

[0015] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A schematic flow chart of a method for efficiently monitoring heat dissipation of an axial magnetic field motor winding according to the present invention; Figure 2 A schematic diagram of the process of constructing and training an LSTM neural network model according to the present invention; Figure 3 This is a schematic diagram of the functional modules of a high-efficiency heat dissipation monitoring system for axial magnetic field motor windings according to the present invention; Figure 4 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0019] It should be clear that the following embodiments of the present disclosure are described through specific concrete examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0020] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on this disclosure, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0021] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The illustrations only show components related to the present disclosure and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0022] Additionally, in the following description, specific details are provided to provide a thorough understanding of the examples. However, one skilled in the art will appreciate that the aspects described can be practiced without these specific details.

[0023] Example 1 Figure 1 It is a flow chart of an embodiment of the present invention, which is a method for efficiently monitoring heat dissipation of an axial magnetic field motor winding. It should be noted that if there are substantially the same results, the method of the present application is not based on Figure 1 The process sequence shown is limited. Figure 1-Figure 2 A method for efficiently monitoring heat dissipation of an axial magnetic field motor winding is shown, comprising the following steps: Step 1: Arrange multiple sets of composite sensor units in a honeycomb curved array on the axial cross section and radial end of the axial magnetic field motor winding; Specifically, a 3D scanner is used to obtain a precise model of the axial magnetic field motor winding (accuracy ±0.02mm), focusing on extracting curvature distribution data of the axial cross-section (z = constant plane) and the radial end (spiral surface). Finite element simulation (such as ANSYS Maxwell) is then used to analyze the heat flux distribution in the winding, identify high heat density areas (such as the end spiral inflection point and the interlayer insulation interface), and determine the key areas for sensor array layout. Then, the honeycomb array layout design is carried out, including the axial section sensor array arrangement and the radial end flexible sensor array arrangement; The specific arrangement of the axial cross-section sensor array is as follows: Take the motor axis as the origin O(0,0,0), the axial direction as the z-axis, select four axial sections z=z1, z=z2, z=z3, and z=z4 (with an interval of 5 mm), and arrange four groups of sensor units in each section; A regular hexagonal grid with a side length of 0.5 mm is used in each axial section. The center of the sensor unit is located at the grid vertex. A total of 4 layers of ring structures are arranged (radii are 2 mm, 4 mm, 6 mm, and 8 mm, respectively), with 4 groups in each layer, for a total of 16 groups; The sensor coordinates (x_i, y_i, z_i) are marked on the winding insulation layer by laser marking with an error of ≤0.1mm; The specific arrangement of the radial end flexible sensor array is as follows: Based on the spiral surface equation of the winding end (such as the Archimedean spiral ρ=kθ, k=0.5mm / rad), the sensor layout trajectory is generated using the Python NumPy library to ensure that the sensor spacing along the arc length of the surface is ≤0.2mm; Eight sets of composite sensor units were attached to a 0.5 mm thick polyimide flexible substrate and pre-bent to the target curvature (bending radius 5 mm) using a hot pressing process (temperature 150°C, pressure 0.5 MPa). Use vacuum adsorption tooling (adsorption force ≥ 10kPa) to attach the preformed flexible array to the curved surface of the winding end, ensuring that the sensor unit covers the area with the largest curvature (such as the wire end with a helix angle of 20°). Next, the axial cross-section sensor and the radial end flexible sensor are installed and fixed; The specific method of pasting the axial cross-section sensor is as follows: Clean the winding insulation surface with anhydrous ethanol and control the roughness to Ra≤1.6μm; Use a dispensing machine to evenly apply silver-filled epoxy resin (thermal conductivity 5 W / m·K, such as Master Bond EP21TDC) to a thickness of 0.05 mm ± 0.01 mm. Align the sensor unit with the marked coordinates and apply 0.1N pressure for 30 seconds to ensure a smooth fit without bubbles. Comb the lead wires along the winding axis, insulate them with polytetrafluoroethylene sleeves (0.3mm inner diameter), and fix them with epoxy resin every 10mm; The method for fixing the radial end flexible sensor array is as follows: Align the preformed flexible array to the curved surface of the winding end using a vision guidance system (accuracy ±0.05mm); Coat the edge of the flexible array with room temperature vulcanized silicone rubber (Shore hardness 40A, such as Dow Corning Sylgard184) to form a 0.2 mm thick protective layer. Curing conditions: room temperature for 24 hours or heating at 60°C for 2 hours; A 1mm radius arc transition is set at the lead outlet to avoid bending stress concentration and lead breakage; Finally, anti-interference and protection treatment is carried out, including the laying of electromagnetic shielding layer and environmental protection coating, as follows: Cover the sensor array surface with nickel-plated copper foil (5 μm thick) and connect it to the winding ground terminal through conductive glue (such as 3M Conductive Epoxy EC-3319). The ground resistance is less than 0.1Ω. Aluminum film (2 μm thickness) was sputtered on the back of the polyimide substrate to form a continuous shielding layer, which was grounded to the motor housing. The assembled sensor array was placed in a vacuum coating chamber, where a layer of parylene C (2 μm thick) was deposited at 0.1 Pa pressure and 150°C to provide moisture and oil resistance. In the cooling airflow channel area (such as the axial ventilation hole diameter ≥ 2mm), remove the sensor unit at the corresponding position and fill the hollow area with high-temperature resistant sealant (such as Loctite 598) to ensure that the airflow area loss is less than 5%.

[0024] Step 2: Use a composite sensor unit to collect multi-dimensional data including winding temperature, heat flux density, insulation layer stress, and cooling medium flow rate in real time, with a sampling frequency of no less than 100 Hz; Specifically: First, a composite sensor unit comprising a micro-thermocouple, a thin-film heat flux density sensor, a fiber Bragg grating pressure sensor, and a micro-ultrasonic Doppler flowmeter is integrated and arranged in a honeycomb curved array on the axial cross-section and radial end of the axial magnetic field motor winding; Then, multi-dimensional data on the winding temperature, heat flux density, insulation layer stress, and cooling medium flow rate are collected in real time. Micro-thermocouples are densely distributed on the winding conductor layer with a spacing of 0.3mm, with an accuracy of ±0.1°C and a response time of less than 50ms. Thin-film heat flux density sensors are attached to the surface of the insulation layer, with a range of 0-800W / m² and a resolution of 0.5W / m². The flexible sensor array has a sensor spacing of ≤0.2mm at the winding end, and a sampling frequency of no less than 100Hz. Finally, the collected data is subjected to noise suppression using a combination of adaptive median filtering and wavelet denoising algorithm and then transmitted to the edge computing processing unit.

[0025] Step 3: The collected multi-dimensional data is transmitted to the edge computing processing unit, and noise is suppressed by a combination of adaptive median filtering and wavelet denoising algorithm; Specifically, first, the collected multi-dimensional data is synchronously transmitted to the edge computing processing unit through a dual redundant transmission architecture (CAN bus and BLE 5.2), with a transmission rate of ≥2Mbps, a delay of ≤10ms, and a bit error rate of <10 -9 , and the data is encrypted in real time using the AES-256 encryption algorithm; Then, an adaptive median filter and wavelet denoising algorithm is used in the edge computing processing unit to suppress noise, as follows: First, the window size is dynamically adjusted using an adaptive median filtering algorithm (the window size can be adaptively switched from 3×3 to 7×7 pixels) to effectively suppress salt and pepper noise. The signal is then decomposed into three layers using the db4 wavelet basis function, and high-frequency coefficients are processed by threshold quantization to eliminate high-frequency vibration noise. Ultimately, the signal-to-noise ratio of signals such as temperature and heat flux density is improved by more than 20 dB, and the data efficiency is >99.5%, meeting the high-precision requirements of subsequent dynamic thermal resistance modeling.

[0026] Step 4: pre-processing the filtered data based on a preset dynamic thermal resistance network model; Specifically, first, the preset dynamic thermal resistance network model is loaded from the storage module of the edge computing processing unit. The model is fitted based on more than 2000 sets of CFD simulation data, including the cooling medium flow rate. ,density and dielectric constant The coupling correction function of : ; in, 、 is the reference flow rate and density, is the dielectric constant of insulation.

[0027] Then, the filtered multi-dimensional data (temperature, heat flux, real-time flow rate of cooling medium, etc.) is input into the dynamic thermal resistance network model and preprocessed through the following steps: The real-time flow rate ,density and the benchmark value 、 Normalization, calculation of coupling correction coefficient ; According to the dynamic thermal resistance network model: ; Solve the equivalent thermal resistance of each grid unit; where, is the coupling correction function, is the real-time total loss of the winding; Based on the real-time monitoring of the insulation dielectric constant drift, the coupling correction function parameters are updated every 10ms to ensure that the model's response delay to working condition changes is less than 20ms; The thermal resistance calculation results are mapped into a three-dimensional heat conduction matrix, and the heat conduction path association between each grid unit is established to provide basic matrix parameters for the subsequent Kalman filter time-space domain fusion.

[0028] Step 5: Use the adaptive Kalman filter algorithm to perform spatiotemporal fusion on the pre-processed multi-dimensional data, build a digital twin model of the winding thermal distribution, calculate the three-dimensional temperature gradient field and equivalent thermal resistance in real time, and generate heat dissipation status data; Specifically, first, the adaptive Kalman filter algorithm is initialized in the edge computing unit. Based on the winding three-dimensional grid division results (12 layers in the axial direction, 8 layers in the radial direction, and the minimum grid unit size of 0.5mm×0.5mm×0.5mm), each grid unit is assigned a unique coordinate. , and establish the heat conduction path association; Then, define the four-dimensional state vector: ; in, is the grid cell temperature, is the heat flux density, is the insulation stress, is the equivalent thermal resistance; Next, establish the state transfer equation: ; Where A is the heat transfer coefficient matrix obtained by finite element simulation, B is the cooling medium flow rate control matrix, is the real-time flow velocity input vector, is the process noise vector that follows a normal distribution (standard deviation ); Then, the measurement equation is established: ; in, is the sensor measurement vector, H is the spatial sensitivity matrix based on the physical position calibration of the sensor, is the measurement noise that follows the normal distribution (standard deviation ); Next, the Kalman gain is iteratively calculated using the adaptive Kalman filter algorithm: ; Where, is the forecast error covariance matrix, is the measurement noise covariance matrix; Finally, the state vector is updated based on the Kalman gain dynamic fusion of the sensor measured value and the model predicted value: ; At the same time, the error covariance matrix is ​​updated: ; in, is the identity matrix; Finally, based on the updated state vector, a three-dimensional temperature gradient field cloud map and heat flux vector distribution map with a resolution of 0.5mm are generated, the solution refresh rate is ≥100Hz, the temperature solution accuracy is ±1℃, the heat flux density solution error is ≤5%, and the heat dissipation status data (including the thermal parameters and spatiotemporal distribution characteristics of each grid unit) are stored in the edge computing unit cache.

[0029] Step 6: Build and train an LSTM neural network model to analyze the heat dissipation status data, extract multi-dimensional feature parameters, and identify fault characteristics such as cooling system blockage, insulation layer aging, or winding turn short circuit; Specifically, first, comprehensive heat dissipation status data is collected for scenarios such as normal operating conditions, cooling system blockage, insulation layer aging, and inter-turn short circuits. This includes multi-dimensional information such as three-dimensional temperature gradient fields, equivalent thermal resistance, heat flux density, and cooling medium flow rate. Abnormal data such as temperature rises exceeding 150°C and negative heat flux values ​​are eliminated using the 3σ principle. Short-term missing values ​​are filled using linear interpolation, and the Z-score standardization method is used to normalize all parameters to the [-1, 1] range to eliminate dimensional differences. Then, in the time domain, the temperature change rate (such as ΔT / Δt), heat flux fluctuation amplitude (Max-Min), and insulation stress mean are calculated. In the frequency domain, Fourier transform is used to obtain the dominant frequency of the temperature signal (cooling blockage often corresponds to a low frequency of 0.1-1 Hz) and the proportion of heat flux harmonic energy. In the statistical domain, correlations between parameters (such as the Pearson correlation coefficient between temperature and flow rate) and the three-dimensional temperature field entropy (reflecting the uniformity of heat distribution) are analyzed to form an input vector containing at least 12 dimensions of features. Next, a two-layer LSTM network architecture was built. The input layer matched the feature dimension (e.g., 12D). The first layer had 128 LSTM units to capture long-term temporal dependencies, and the second layer had 64 units with a dropout rate of 0.2 to prevent overfitting. The output layer used 4 neurons (corresponding to four states: normal, blocked, aged, and short-circuited) with a softmax activation function. The Adam optimizer was used to minimize the cross-entropy loss function, with an initial learning rate of 0.001 and a batch size of 32. Afterwards, the data is divided into training, validation, and test sets in a ratio of 7:1:2, and fed into the model for iterative training. The validation set is used to evaluate accuracy and loss every five training cycles. When the validation set accuracy stagnates or the loss rebounds, the early stopping mechanism is triggered and the learning rate is adjusted (for example, decayed to 0.0001) or the regularization parameter is increased until the validation set accuracy is greater than 98% and the loss is less than 0.03. Finally, the test set is used to calculate indicators such as precision, recall, and F1 value to verify the ability to identify various types of faults (for example, the recall rate for cooling blockages must be greater than 95%). If the recognition rate of a certain type of fault is lower than expected, the importance of the features is analyzed, and feature engineering is adjusted (such as increasing parameters such as the divergence of the heat flow vector) or the network structure is optimized (such as adding an attention layer to strengthen key features). The model is retrained until it is stable and finally deployed to the edge computing unit for real-time fault identification.

[0030] Step 7: Classify the fault level based on the identified fault characteristic information and trigger the three-level warning mechanism. At the same time, encrypt the fault characteristic information and upload it to the cloud diagnosis platform; Specifically, based on the fault feature information output by the LSTM neural network model, the fault level is divided according to preset rules: normal operating conditions (no feature matching), warning level (such as local heat flux density exceeds the threshold by 80% and insulation stress fluctuation is greater than 2MPa), alarm level (temperature gradient is greater than 5°C / mm and equivalent thermal resistance increases by 30%), and emergency shutdown level (inter-turn short circuit characteristics or complete blockage of the cooling system). Then, the three-level early warning mechanism is triggered; Early warning level: The edge computing unit sends a prompt message to the operation and maintenance terminal, and simultaneously marks abnormal areas in the winding temperature cloud map. For example, grid cells with heat flux density greater than 600W / m² are highlighted in yellow. Alarm level: Automatically adjust cooling system parameters, such as increasing the cooling medium flow rate by 20% or starting a backup circulation pump, and push detailed alarm reports containing three-dimensional temperature field data to the cloud platform; Emergency shutdown level: The motor power is cut off within 0.5 seconds, triggering the sound and light alarm device. At the same time, the full-dimensional data at the time of the fault (including the original sensor signal and the model prediction results) is encrypted and uploaded to the cloud for remote diagnosis by experts; Finally, the edge computing unit records the warning trigger time, fault characteristic parameters and control response effects in real time to form a traceable operation and maintenance log; the cloud platform performs cluster analysis based on similar fault data of multiple motors and automatically optimizes the fault recognition threshold of the LSTM model. For example, based on historical data, the stress fluctuation warning threshold for insulation aging is dynamically adjusted from 2MPa to 1.8MPa, thereby improving the sensitivity of early fault recognition.

[0031] In one embodiment, specifically, in step 4, a dynamic thermal resistance network model is used to characterize the coupling relationship between the winding heat conduction path and the cooling medium; The dynamic thermal resistance network model is: ; in, is the coupling correction function of the cooling medium flow rate, density and insulation dielectric constant, is the real-time total loss of the winding; The coupling correction function The results are obtained by fitting more than 2000 sets of CFD simulation data, specifically: ; in, is the reference flow rate of the cooling medium, is the base density, is the dielectric constant of insulation.

[0032] Specifically, take the heat dissipation monitoring of a large power transformer winding as an example. The transformer has a rated capacity of 100MVA, the winding is wound with copper wire, and the cooling medium during operation is transformer oil; First, deploy sensors and collect data as follows: Ten thermocouple sensors (type K thermocouple, accuracy ±0.5°C) are arranged at different locations inside the winding to measure the winding temperature. ; Two temperature sensors are arranged at the inlet and outlet of the transformer oil cooling circuit to measure the ambient temperature ; Install an electromagnetic flow sensor (model: LDG-100, accuracy ±1%) on the cooling oil pipeline to measure the cooling medium flow rate. ; Transformer oil density was measured using a vibrating tube densitometer (model: DMA 4100 M, accuracy ±0.001 g / cm³). ; The dielectric constant of the insulation was measured using a capacitive dielectric constant sensor (homemade, accuracy ±0.05) ; The real-time total loss of the winding is measured by a power analyzer (model: WT3000, accuracy ±0.1%) The data acquisition frequency is set to 10 Hz, and all sensor data are transmitted to the edge computing device through a data acquisition card (model: NI USB-6366); Then, determine the model parameters as follows: Determine the reference flow rate of the cooling medium based on the transformer design parameters and operating experience , baseline density ; When a set of real-time data is collected, such as , substitute the coupling correction function: ; Assume that at a certain moment, , the above values ​​and the calculated Substitute into the dynamic thermal resistance network model: ; Finally, the model is verified and optimized as follows: Use an infrared thermal imager (FLIR T1040, accuracy ±2%) to measure the winding temperature field and compare the theoretically calculated thermal resistance with the model-calculated thermal resistance. If the deviation exceeds 5%, analyze the cause. Regularly (after every 100 sets of data collection) re-fit and optimize the coupling correction function based on new CFD simulation data (obtained by changing operating conditions, environmental conditions, etc.) to ensure model accuracy; This dynamic thermal resistance network model can accurately monitor the heat conduction status of transformer windings in real time, providing strong data support for transformer operation and maintenance. For example, when the thermal resistance calculated by the model exceeds 10% of the normal operating range, an early warning signal is issued, prompting operation and maintenance personnel to check whether there are problems such as blockage and insufficient flow in the cooling system, effectively preventing equipment failures caused by winding overheating.

[0033] In one embodiment, specifically, in step 1, there are no less than 24 groups of composite sensor units; Specifically, at least 24 sets of composite sensor units need to be arranged in the monitoring area of ​​the axial magnetic field motor winding (axial section and radial end). This number is determined based on the spatial sampling theorem of three-dimensional thermal field solution to ensure that the sensor spacing is small enough (≤0.5mm) to avoid monitoring blind spots and meet the accuracy requirements of thermal field reconstruction (resolution ≤0.5mm); 24 sets of sensors can cover the main heat dissipation paths of the winding (such as interlayer insulation and end spiral surfaces), and realize intensive sampling of multi-physical fields of temperature, heat flow, stress, and flow velocity, providing sufficient data support for subsequent dynamic thermal resistance modeling and digital twins.

[0034] The composite sensor unit integrates a micro-thermocouple, a thin film heat flux density sensor, a fiber Bragg grating pressure sensor and a micro-ultrasonic Doppler flowmeter; Among them, the micro-thermocouple directly measures the temperature of the winding conductor with an accuracy of ±0.1°C and a response time of <50ms, and is used to capture rapid temperature rise anomalies (such as instantaneous heating caused by inter-turn short circuit); Thin-film heat flux density sensor: adheres to the surface of the insulation layer and measures the direction and intensity of heat flow through the insulation layer (range 0-800W / m², resolution 0.5W / m²), which can locate heat conduction bottlenecks (such as dead zones in the cooling medium flow); Fiber Bragg Grating pressure sensor: Based on the principle of fiber Bragg Grating wavelength drift, it measures the stress fluctuation of the insulation layer (accuracy ±0.5MPa) and is used to monitor stress concentration caused by aging of insulation materials; Miniature ultrasonic Doppler flowmeter: measures the cooling medium flow rate through the ultrasonic Doppler effect (accuracy ±2%), reflecting the cooling system flow status in real time (such as flow rate drop caused by pipe blockage); Multi-sensor integration avoids installation errors associated with discrete sensors, and synchronous data acquisition enables spatiotemporal correlation analysis between parameters (such as the lagged correlation between temperature and flow rate).

[0035] The micro thermocouples are densely distributed on the winding conductor layer with a spacing of 0.3mm, with an accuracy of ±0.1℃ and a response time of <50ms; Specifically, micro-thermocouples are distributed on the winding conductor layer with an extremely small spacing of 0.3mm, close to the physical size of the winding copper wire (usually the diameter of the enameled wire is 0.5-2mm), ensuring that the temperature measurement can reflect subtle changes in local hot spots (such as the local temperature rise caused by a single-turn short circuit). The accuracy of ±0.1°C can distinguish the temperature difference between normal operating conditions and early faults (for example, the initial temperature rise of insulation aging is only 1-2°C). The response time of <50ms meets real-time monitoring needs and is suitable for transient thermal analysis of high-speed motors (speed >10,000rpm).

[0036] The thin film heat flux density sensor is attached to the surface of the insulation layer, with a range of 0-800W / m² and a resolution of 0.5W / m²; Specifically, the sensor is directly attached to the outer surface of the insulation layer to avoid damaging the winding structure, while reducing the contact thermal resistance in the heat conduction path, ensuring that the measured value truly reflects the heat flux density of the insulation layer; the 0-800W / m² range covers the typical heat flux range of axial magnetic field motors (up to 600W / m² or more at high power density), and the 0.5W / m² resolution can detect tiny fluctuations in heat flux (such as a 5% change in cooling medium flow rate causing a heat flux density change of approximately 10W / m²), which is used for early identification of cooling system performance degradation.

[0037] The composite sensor unit uses a polyimide-based flexible sensor array with a thickness of ≤0.5mm at the winding end. The flexible sensor array fits the spiral winding surface structure, with a sensor spacing of ≤0.2mm and a bending radius of ≥5mm. Specifically, a polyimide (PI) substrate with a thickness of ≤0.5mm is used. It is resistant to high temperatures (long-term operating temperature >200°C) and highly flexible, allowing it to be bent along the spiral surface of the winding end (such as an Archimedean spiral), solving the problem that traditional rigid sensors cannot adapt to complex curved surfaces. The sensor spacing is ≤0.2mm, enabling dense sampling of the area with maximum end curvature (such as the inflection point of the output terminal). A bending radius of ≥5mm ensures that the flexible array does not break during installation and motor vibration (acceleration ≤10g), while also avoiding sensor performance degradation caused by excessive bending. The flexible array is fitted to the curved surface through a vacuum adsorption process, and the edges are encapsulated with silicone rubber (Shore hardness 40A), forming a monitoring layer integrated with the winding. The thickness is only 0.5mm, and the impact on the original heat dissipation path is negligible (thermal resistance increase is <3%).

[0038] In one embodiment, specifically, in step five, building a winding thermal distribution digital twin model specifically includes: Divide the winding into three-dimensional grid cells and assign unique coordinates to each grid cell , and establish the heat conduction path association; Define the four-dimensional state vector: ; in, is the grid cell temperature, is the heat flux density, is the insulation stress, is the equivalent thermal resistance; Establish the state transfer equation: ; Among them, A is the heat transfer coefficient matrix, B is the cooling medium flow rate control matrix, is the real-time flow velocity input vector, is the process noise vector that obeys the normal distribution; Establish the measurement equation: ; in, is the sensor measurement vector, H is the spatial sensitivity matrix of the sensor array, is the measurement noise that obeys the normal distribution; The Kalman gain is calculated iteratively using the adaptive Kalman filter algorithm: ; in, is the forecast error covariance matrix, is the measurement noise covariance matrix; Based on the Kalman gain, the sensor measured value and the model predicted value are dynamically fused to update the state vector: ; At the same time, the error covariance matrix is ​​updated: ; in, is the identity matrix; Based on the updated state vector, a three-dimensional temperature gradient field cloud map and a heat flow vector distribution map are generated; Specifically, assuming the power of the axial magnetic field motor is 150kW, the axial thickness of the winding is 12mm, and the radial outer diameter is 280mm; First, perform three-dimensional grid division and coordinate assignment; The specific method of mesh partitioning parameters is as follows: Axial (z-axis): 10 layers are equally spaced along the thickness direction of the winding (layer spacing is 1.2 mm), covering the full thickness of the winding (0-12 mm); Radial (x / y axis): With the motor axis as the origin, each axial section is divided into eight radius intervals (0-35mm, 35-70mm, ..., 245-280mm) according to polar coordinates. Each interval is divided into 12 sectors at 30° intervals, forming 10 × 8 × 12 = 960 three-dimensional grid cells; Minimum element size: The element near the axis (radius 0-35mm) is 1.2mm (axial) × 0.5mm (radial) × 0.5mm (circumferential), and the edge element is appropriately enlarged to 1.2mm × 2mm × 2mm to balance calculation accuracy and efficiency; The following is an example of coordinate assignment: The coordinates of a grid cell are defined as , corresponding to a radial radius of 70.7 mm (polar coordinates , ), axial middle layer; Then, define the four-dimensional state vector, establish the state transfer equation and measurement equation, as follows: Real-time acquisition of four-dimensional state vector: ; temperature (measured by micro-thermocouple); Heat flux (measured value of thin film sensor); Insulation stress (Fiber Bragg Grating Sensor Solution); Equivalent thermal resistance (Calculated value of dynamic thermal resistance model); Pre-calculate the heat transfer coefficient between adjacent grids through finite element simulation, such as the heat transfer coefficient between the current unit and the axial adjacent unit , radially adjacent units ; Cooling medium flow rate When the thermal resistance correction coefficient of the current unit is (calculated by coupling function of dynamic thermal resistance model); The sensitivity coefficient of the sensor to the current unit (temperature signal attenuation 5%), (heat flow signal attenuation 2%), reflecting the spatial deviation between the sensor position and the grid cell; Next, the adaptive Kalman filter iteration is performed. The specific process is as follows: Status prediction: ; Error covariance prediction: ; Process noise covariance matrix Take the empirical value, temperature noise , thermal flow noise ; Measurement vector: ; Measurement noise Obeying normal distribution, temperature noise , thermal flow noise ; Kalman gain calculation: ; Measurement noise covariance matrix Pick , corresponding to the measurement variance of temperature and heat flow; Best estimate: ; Error covariance update: ; Finally, the thermal distribution visualization and accuracy verification are performed as follows: The temperature values ​​of 960 grid cells are rendered as a color cloud map using an interpolation algorithm. Hot spots (e.g., T = 102°C at 180 mm radially and 8 mm axially) are highlighted in red. The temperature gradient resolution is 0.5°C / mm. Superimpose the heat flux vector on the spiral surface at the winding end (the arrow direction indicates the heat flux direction, and the length is proportional to the heat flux density) to intuitively display the convergence or divergence trend of the heat flux (for example, the divergence of the heat flux vector at the inflection point at the end is greater than 1.2W / m³); Comparing the model prediction value T = 102 ° C for the grid unit (x = 180 mm, y = 0, z = 8 mm) with the infrared measured value T = 100.5 ° C, the error is 1.5%, which meets the engineering accuracy requirement (≤ 3%). When the cooling medium velocity suddenly changes (from 2m / s to 3m / s), the model response delay is less than 20ms, and the deviation between the calculated heat flux value and the simulated value is less than 4%; Through high-density meshing, multi-physics field coupling modeling, and adaptive filtering algorithms, high-precision real-time reconstruction of the thermal distribution of axial magnetic field motor windings is achieved, providing an intuitive digital twin tool for heat dissipation efficiency optimization and early fault warning.

[0039] In one embodiment, specifically, in step six, building and training an LSTM neural network model specifically includes the following steps: Step 601: Acquire sample data including normal operating conditions, cooling system blockage, insulation layer aging, and winding turn-to-turn short circuit; the number of sample data is ≥ 3000 groups; Specifically, first, a 1:1 scale axial magnetic field motor test bench was built, equipped with a 150kW test motor, a closed-loop cooling system, and a programmable load simulation system; Then, by inserting flow-restricting orifices of different diameters (diameter ratios set to 100%, 70%, 50%, and 30%) into the cooling pipe to simulate cooling system blockage, the winding samples were placed in a 180°C high-temperature box for accelerated treatment for 0h, 100h, 200h, and 500h to simulate insulation layer aging. A local short-circuit fixture was used to create a short circuit between the winding turns (the number of short-circuited turns was 1-5 turns). Next, under normal motor operation, data was collected at different load rates (30%, 50%, 70%, and 100%). Furthermore, during the experiment, a composite sensor unit was used to collect multi-dimensional data such as temperature and heat flux at a frequency of 10 Hz. Finally, combined with historical operation and maintenance data, a total of 3,500 sets of sample data covering normal operating conditions, cooling system blockage, insulation layer aging, and winding inter-turn short circuit were obtained to ensure that the data quantity met the requirement of ≥3,000 sets.

[0040] Step 602: normalize the sample data and extract multi-dimensional feature parameters as input vectors; Specifically, we first analyzed the 3,500 sets of sample data obtained. We used the Z-score standardization method to normalize the original characteristics of the data, such as temperature, heat flux, insulation stress, cooling medium flow rate, and equivalent thermal resistance, to a standard distribution with a mean of 0 and a standard deviation of 1. We also used the min-max scaling method to scale the heat flux data to the [0, 1] interval. Then, based on the normalized data, multi-dimensional characteristic parameters are extracted: the rate of change and fluctuation amplitude of temperature and heat flux are calculated in the time domain; FFT transformation of signals such as flow velocity is performed in the frequency domain to obtain the main frequency and its energy proportion; and the skewness and kurtosis of insulation stress are calculated in the statistical domain. Finally, a sliding window of length 50 and a step size of 5 are used to construct the processed multi-dimensional feature data into an input vector of dimension [50, 12], providing standardized and characterized data samples for subsequent LSTM neural network model training.

[0041] Step 603: Build a two-layer LSTM network, use a dropout rate of 0.2 to suppress overfitting, minimize the cross entropy loss function using the Adam optimizer, and train with an early stopping mechanism until the validation set accuracy is greater than 98% and the loss value is less than 0.03; Specifically, we first constructed a two-layer LSTM network using Python's TensorFlow framework. The input layer received time series data of dimension [50, 12]. The first LSTM layer had 128 units and sequence return enabled, followed by a BatchNormalization layer and a Dropout layer with a dropout rate of 0.2. The second LSTM layer had 64 units and sequence return disabled, and was also connected to a BatchNormalization layer and a Dropout layer to prevent overfitting. Then, the output layer first passes through a fully connected layer of 16 neurons with ReLU activation function, and then connects to a fully connected layer of 4 neurons with Softmax activation function for four-class classification task; Next, configure the Adam optimizer, set the initial learning rate to 0.001, select sparse_categorical_crossentropy as the loss function, and set accuracy, recall, and precision as evaluation metrics; Afterwards, the dataset was split into training, validation, and test sets in a 7:1:2 ratio, and the batch_size was set to 32, and model training began. Early stopping was enabled during training. If the validation set loss did not decrease for 10 consecutive epochs, the learning rate was decayed to 0.1 times the original value. If there was no improvement for 20 consecutive epochs, training was stopped. Finally, after 45 epochs of training, the model achieved an accuracy of 98.2% and a loss of 0.028 on the validation set, meeting the requirements of a validation set accuracy greater than 98% and a loss less than 0.03.

[0042] Step 604: Deploy the trained model to the edge computing unit to perform fault feature recognition on the real-time heat dissipation status data; Specifically, the trained LSTM neural network model was first quantized, converting 32-bit floating-point (FP32) parameters into 8-bit integers (INT8), reducing the model size from 2.4MB to 0.6MB while maintaining a 97.8% recognition accuracy. The quantized model is then deployed to an industrial-grade edge computing unit (ARM Cortex-A53 quad-core processor, 1.4GHz, 2GB RAM), where model inference is implemented using the TensorFlow Lite framework, with a single-sample processing latency of <8ms. Next, a real-time data processing process was developed on the edge unit: sensor data was collected at a frequency of 10 Hz, an input vector was constructed every 50 time steps (i.e., 5 seconds), and continuous data processing was achieved using a sliding window (step size 5). 12-dimensional features such as the temperature gradient change rate and heat flux density fluctuation amplitude were calculated in real time from the raw data, and normalized using the normalization parameters used in the training phase. The model then infers the input vector and outputs the probability distribution of four types of faults (normal, blocked, aging, and short circuit). When the probability of a particular fault type exceeds 90%, a warning of the corresponding level is triggered. Simultaneously, the edge unit packages the inference results with the original data and uploads them to the cloud platform via the MQTT protocol. The upload frequency is dynamically adjusted based on the fault level (1 minute / time for normal status, 10 seconds / time for warning status, and real-time upload for alarm status). Finally, a local caching mechanism is established on the edge unit to store the last seven days of inference results and raw data, supporting offline analysis and historical traceability to ensure data is not lost during network outages. The entire deployment solution achieves an end-to-end processing time of less than 500ms, from data acquisition to fault warning, meeting the real-time monitoring requirements of axial-field motors.

[0043] Multidimensional characteristic parameters include temperature gradient, heat flux density change rate, insulation layer stress fluctuation amplitude, cooling medium flow rate deviation, equivalent thermal resistance change rate, temperature-heat flux coupling coefficient, stress-flow velocity correlation coefficient, three-dimensional temperature field entropy value, heat flux vector divergence, insulation dielectric constant drift, winding copper loss power and cooling system pressure drop; The temperature gradient refers to the rate of change of temperature along the spatial direction inside the motor, usually in units of °C / mm. In an axial magnetic field motor, the temperature gradient distribution at different locations of the winding is relatively uniform under normal operating conditions (for example, 0.5-1.0 °C / mm). The heat flux rate of change is the change in heat flux per unit time, measured in W / m²·s. It reflects the dynamic characteristics of the heat transfer process and is closely related to the efficiency of the cooling system. The insulation layer stress fluctuation amplitude refers to the peak value of the electric field stress borne by the insulation material per unit time, and the unit is V / m. Under normal operating conditions, the insulation stress fluctuation amplitude is small and relatively stable (for example, <10kV / m); Cooling medium flow rate deviation is the difference between the measured flow rate and the rated flow rate, and is expressed in m / s. It directly reflects the operating status of the cooling system and is an important indicator for determining whether the cooling system is working properly. The equivalent thermal resistance change rate refers to the rate of change of the motor's equivalent thermal resistance over time, measured in K / W·s. The equivalent thermal resistance reflects the heat transfer resistance from the heat source to the cooling medium, and its rate of change is closely related to insulation aging and cooling system performance degradation. The temperature-heat flow coupling coefficient is a dimensionless parameter that describes the temporal and spatial correlation between the temperature field and the heat flow field and is obtained by calculating the cross-correlation function between the two. The stress-velocity correlation coefficient is the Pearson correlation coefficient between insulation stress and cooling medium velocity, reflecting the degree of linear correlation between the two; Heat flux vector divergence is a physical quantity that describes the local characteristics of a heat flux field. It is mathematically defined as the divergence of the heat flux vector field (▽·q), with units of W / m³. It represents the net outflow of heat flux per unit volume and reflects the intensity of the heat flux's convergence or divergence. The dielectric constant drift of insulation refers to the percentage change of the dielectric constant of the insulation material relative to the initial value, which is a direct indicator for evaluating the degree of insulation aging. Winding copper loss power refers to the Joule heat power generated when current passes through the winding resistance. The calculation formula is P=I²R, and the unit is W; The pressure drop of a cooling system refers to the pressure difference between the inlet and outlet of the cooling medium, measured in Pa. It reflects the energy consumed by the cooling medium to overcome resistance during flow and is closely related to pipe resistance, flow velocity, and flow rate.

[0044] In one embodiment, specifically, in step seven, the three-level warning mechanism includes a heat dissipation efficiency adjustment strategy linked with the motor cooling system and the control system; Specifically, when a single parameter exceeds a threshold but does not continue to deteriorate, such as a temperature gradient temporarily exceeding 2°C / mm and lasting less than 5 minutes, a first-level warning is triggered, and the system initiates cooling system pre-adjustment, appropriately increasing the cooling medium flow rate by 10%-15%, while sending a warning message to the operation and maintenance personnel. If multiple parameters are abnormal at the same time, such as a temperature gradient greater than 3°C / mm and a heat flux density change rate greater than 100W / m²·s, indicating a developing fault, a second-level warning is triggered, and the control system automatically increases the cooling medium flow rate by 20%-30%, starts the backup cooling circulation pump, and pushes a detailed fault analysis report to the operation and maintenance platform. Once a key parameter exceeds the danger threshold, such as a turn-to-turn short circuit causing the winding copper power consumption to surge by more than 15%, a third-level warning is triggered, and the system immediately cuts off the motor power supply and simultaneously starts all cooling equipment to dissipate heat at full capacity, minimizing the impact of the fault, thereby achieving fault classification management and dynamic optimization of heat dissipation efficiency.

[0045] In one embodiment, specifically, in step three, the edge computing processing unit is equipped with an NPU neural network processor with a computing power of ≥2TOPS, supports real-time parallel operation of the dynamic thermal resistance model and the LSTM algorithm, and generates a winding temperature cloud map with a refresh rate of ≥100Hz; Specifically, the edge computing processing unit is equipped with an NPU neural network processor with a computing power of no less than 2TOPS (2 trillion operations per second). This powerful computing power foundation provides the system with efficient data processing capabilities; with the help of the parallel computing characteristics of the NPU, it can simultaneously support the real-time operation of the dynamic thermal resistance model and the LSTM algorithm, and realize rapid analysis and processing of the heat dissipation status data; the dynamic thermal resistance model can quickly calculate and update the thermal resistance status of the motor winding based on the real-time collected heat dissipation data; the LSTM algorithm conducts in-depth data mining to identify potential fault characteristics; at the same time, the edge computing processing unit can generate a winding temperature cloud map at a refresh rate of no less than 100Hz, that is, at a frequency of at least 100 times per second, the temperature distribution of the motor winding is presented intuitively and dynamically. Operation and maintenance personnel can quickly grasp the temperature change trend inside the motor through the visual interface, and promptly discover abnormal hotspots, providing strong guarantees for the safe and stable operation of the motor.

[0046] In summary, the embodiment of the present application provides an efficient heat dissipation monitoring method for axial magnetic field motor windings. By arranging a composite sensor array at key positions of the winding to collect multi-dimensional data, after filtering, modeling and twin construction by an edge computing unit, an LSTM network is used to extract 12-dimensional features to identify faults (the accuracy of the verification set is >98%), and the heat dissipation efficiency is adjusted based on a three-level early warning linkage cooling system (the flow rate is increased by 10%-30%, etc.), realizing full-process intelligent monitoring from data collection to fault response, and improving the accuracy of heat dissipation monitoring and fault handling efficiency.

[0047] Example 2 Figure 3 This is a functional module diagram of a wireless charging communication control system according to an embodiment of the present application. Figure 3 As shown, an efficient heat dissipation monitoring system for axial magnetic field motor windings includes: a sensor array deployment module, a multi-dimensional data acquisition module, a data transmission and filtering module, a dynamic thermal resistance modeling module, a digital twin construction module, an intelligent fault diagnosis module, and an early warning and control module; A sensor array deployment module is configured to arrange no fewer than 24 groups of composite sensor units in a honeycomb curved array on the axial cross section and radial end of the axial magnetic field motor winding; A multi-dimensional data acquisition module is configured to collect multi-dimensional data of the winding in real time through a composite sensor unit with a sampling frequency of not less than 100 Hz; the multi-dimensional data includes winding temperature, heat flux density, insulation layer stress, and cooling medium flow rate; A data transmission and filtering module is configured to transmit the collected multi-dimensional data to the edge computing processing unit and perform noise suppression through a combination of adaptive median filtering and wavelet denoising algorithm; a dynamic thermal resistance modeling module configured to process the filtered data based on a preset dynamic thermal resistance network model; The digital twin construction module is configured to perform spatiotemporal fusion of pre-processed multi-dimensional data using an adaptive Kalman filter algorithm to construct a digital twin model of the winding thermal distribution, calculate the three-dimensional temperature gradient field and equivalent thermal resistance in real time, and generate heat dissipation status data; An intelligent fault diagnosis module is configured to build and train an LSTM neural network model, analyze heat dissipation status data, extract multi-dimensional feature parameters, and identify fault characteristics such as cooling system blockage, insulation layer aging, or inter-turn short circuits in the winding; The early warning and control module is configured to classify fault levels according to the identified fault feature information and trigger a three-level early warning mechanism, while encrypting the fault feature information and uploading it to the cloud diagnosis platform.

[0048] In one embodiment, specifically, the dynamic thermal resistance modeling module, the digital twin construction module and the intelligent fault diagnosis module are integrated into the edge computing processing unit. The edge computing processing unit is equipped with an NPU neural network processor with a computing power ≥ 2TOPS, which supports the real-time parallel operation of the dynamic thermal resistance model and the LSTM algorithm, and the data processing delay is less than 50ms.

[0049] In one embodiment, specifically, the data transmission and filtering module adopts a dual redundant transmission architecture.

[0050] In summary, the embodiment of the present application provides an efficient heat dissipation monitoring system for axial magnetic field motor windings. 24 or more composite sensor units are arranged at key positions of the windings through a sensor array deployment module. The multi-dimensional data acquisition module collects temperature, heat flux density and other data at a frequency of 100 Hz or more, and transmits the data to the edge computing unit through a dual redundant architecture. After noise is suppressed by adaptive filtering, fault identification (verification set accuracy > 98%) is achieved using a dynamic thermal resistance modeling module, a digital twin construction module (integrated adaptive Kalman filtering) and an intelligent fault diagnosis module (LSTM network extracts multi-dimensional features). Finally, the warning and control module triggers a three-level warning and links the heat dissipation adjustment. The edge computing unit is equipped with an NPU with a computing power of 2TOPS or more to support parallel operation of the model (delay < 50ms), forming an intelligent, high-precision heat dissipation monitoring and fault handling closed loop.

[0051] For other details about the technical solutions for implementing each module in the above-mentioned embodiment of an efficient heat dissipation monitoring system for an axial magnetic field motor winding, please refer to the description of the above-mentioned embodiment of an efficient heat dissipation monitoring method for an axial magnetic field motor winding, which will not be repeated here.

[0052] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similarities between the various embodiments can be referred to in conjunction with each other. For system-related embodiments, since they are generally similar to method-related embodiments, their description is relatively simple. For relevant details, refer to the description of the method-related embodiments.

[0053] Example 3 like Figure 4 The present invention provides a schematic structural diagram of an electronic device according to an embodiment of the present invention, which is suitable for implementing the electronic device according to an embodiment of the present invention. Figure 4 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0054] like Figure 4 As shown, an electronic device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) or programs loaded from a storage device into a random access memory (RAM). The RAM also stores various programs and data required for the operation of the electronic device. The processor, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0055] Typically, the following devices can be connected to the I / O interface: input devices such as sensors or visual information acquisition devices; output devices such as display screens; storage devices such as tapes and hard disks; and communication devices. The communication device allows the electronic device to communicate with other devices (such as edge computing devices) wirelessly or by wire to exchange data. Figure 4 The electronic device is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0056] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of an axial magnetic field motor winding efficient heat dissipation monitoring method of an embodiment of the present disclosure are performed.

[0057] For detailed description of this embodiment, please refer to the corresponding description in the aforementioned embodiments, which will not be repeated here.

[0058] Example 4 An embodiment of the present application also provides a computer-readable storage medium, which can be set in a server to store at least one instruction or at least one program related to implementing a flight control method based on a deep learning model in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement a flight control method based on a deep learning model provided in the above method embodiment.

[0059] Optionally, in this embodiment, the storage medium may be located in at least one of a plurality of network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0060] An embodiment of the present invention further provides a computer program product or computer program, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a flight control method based on a deep learning model provided in any of the aforementioned optional embodiments.

[0061] It should be noted that the order of the embodiments of the present application described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. The above description is of specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or advantageous.

[0062] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device, equipment, and storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0063] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk, etc.

[0064] The basic principles of the present disclosure have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this disclosure are merely illustrative and not restrictive, and should not be construed as necessarily possessed by each embodiment of the present disclosure. Furthermore, the specific details disclosed above are provided for illustrative purposes and to facilitate understanding, rather than as limitations. These details do not limit the present disclosure to necessarily being implemented using these specific details.

[0065] In the present disclosure, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The block diagrams of the devices, devices, equipment, and systems involved in the present disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "including," "comprising," "having," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0066] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0067] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0068] Various changes, substitutions, and modifications may be made to the technology described herein without departing from the teachings defined by the appended claims. Moreover, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of things, means, methods, and actions described above. Currently existing or later developed processes, machines, manufactures, compositions of things, means, methods, or actions that perform substantially the same function or achieve substantially the same results as the corresponding aspects described herein may be utilized. Accordingly, the appended claims include within their scope such processes, machines, manufactures, compositions of things, means, methods, or actions.

[0069] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0070] The above description has been provided for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for efficiently monitoring heat dissipation of axial magnetic field motor windings, characterized in that: The following steps are involved: A plurality of composite sensor units are arranged in a honeycomb curved array on the axial cross section and radial end of the axial magnetic field motor winding; The composite sensor unit collects multi-dimensional data including winding temperature, heat flux density, insulation layer stress and cooling medium flow rate in real time, with a sampling frequency of no less than 100Hz; The collected multi-dimensional data is transmitted to the edge computing processing unit, and noise is suppressed through a combination of adaptive median filtering and wavelet denoising algorithm; Preprocessing the filtered data based on a preset dynamic thermal resistance network model; The pre-processed multi-dimensional data is fused in time and space through the adaptive Kalman filter algorithm to build a digital twin model of the winding thermal distribution, calculate the three-dimensional temperature gradient field and equivalent thermal resistance in real time, and generate heat dissipation status data; Build and train an LSTM neural network model to analyze heat dissipation status data, extract multi-dimensional feature parameters, and identify fault characteristics such as cooling system blockage, insulation layer aging, or inter-turn short circuits in windings. The fault level is divided according to the identified fault characteristic information and a three-level warning mechanism is triggered. At the same time, the fault characteristic information is encrypted and uploaded to the cloud diagnosis platform.

2. The method for efficiently monitoring heat dissipation of an axial magnetic field motor winding according to claim 1, characterized in that: The dynamic thermal resistance network model is used to characterize the coupling relationship between the winding heat conduction path and the cooling medium; The dynamic thermal resistance network model is: ； in, is the coupling correction function of the cooling medium flow rate, density and insulation dielectric constant, is the real-time total loss of the winding; The coupling correction function The results are obtained by fitting more than 2000 sets of CFD simulation data, specifically: ； in, is the reference flow rate of the cooling medium, is the base density, is the dielectric constant of insulation.

3. The method for efficiently monitoring heat dissipation of an axial magnetic field motor winding according to claim 1, characterized in that: The composite sensor units are composed of no fewer than 24 groups; the composite sensor units integrate micro-thermocouples, thin-film heat flux density sensors, fiber Bragg grating pressure sensors, and micro-ultrasonic Doppler flowmeters; the micro-thermocouples are densely distributed on the winding conductor layer with a spacing of 0.3 mm, with an accuracy of ±0.1°C and a response time of <50 ms; the thin-film heat flux density sensors are attached to the surface of the insulation layer, with a range of 0-800 W / m² and a resolution of 0.5 W / m²; The composite sensor unit adopts a polyimide-based flexible sensor array with a thickness of ≤0.5mm at the winding end. The flexible sensor array fits the spiral winding curved surface structure, the sensor spacing is ≤0.2mm, and the bending radius is ≥5mm.

4. The method for monitoring the efficient heat dissipation of an axial magnetic field motor winding according to claim 1, characterized in that: The construction of the winding thermal distribution digital twin model specifically includes: Divide the winding into three-dimensional grid cells and assign unique coordinates to each grid cell , and establish the heat conduction path association; Define the four-dimensional state vector: ； in, is the grid cell temperature, is the heat flux density, is the insulation stress, is the equivalent thermal resistance; Establish the state transfer equation: ； Among them, A is the heat transfer coefficient matrix, B is the cooling medium flow rate control matrix, is the real-time flow velocity input vector, is the process noise vector that obeys the normal distribution; Establish the measurement equation: ； in, is the sensor measurement vector, H is the spatial sensitivity matrix of the sensor array, is the measurement noise that obeys the normal distribution; The Kalman gain is calculated iteratively using the adaptive Kalman filter algorithm: ； in, is the forecast error covariance matrix, is the measurement noise covariance matrix; Based on the Kalman gain, the sensor measured value and the model predicted value are dynamically fused to update the state vector: ； At the same time, update the error covariance matrix: ； in, is the identity matrix; Based on the updated state vector, a three-dimensional temperature gradient field cloud map and a heat flow vector distribution map are generated.

5. The method for efficiently monitoring heat dissipation of an axial magnetic field motor winding according to claim 1, characterized in that: The construction and training of the LSTM neural network model specifically includes the following steps: Obtain sample data including normal operating conditions, cooling system blockage, insulation layer aging, and winding turn-to-turn short circuit; the number of sample data is ≥ 3000 groups; Normalize the sample data and extract multi-dimensional feature parameters as input vectors; A two-layer LSTM network was constructed, with a dropout rate of 0.2 to suppress overfitting. The Adam optimizer was used to minimize the cross-entropy loss function, and the network was trained with early stopping until the validation set accuracy was >98% and the loss was <0.

03. Deploy the trained model to the edge computing unit to identify fault characteristics based on real-time heat dissipation status data; The multidimensional characteristic parameters include temperature gradient, heat flux density change rate, insulation layer stress fluctuation amplitude, cooling medium flow rate deviation, equivalent thermal resistance change rate, temperature-heat flux coupling coefficient, stress-flow velocity correlation coefficient, three-dimensional temperature field entropy value, heat flux vector divergence, insulation dielectric constant drift, winding copper loss power and cooling system pressure drop.

6. The method for efficiently monitoring heat dissipation of an axial magnetic field motor winding according to claim 1, characterized in that: The three-level early warning mechanism includes a heat dissipation efficiency adjustment strategy linked with the motor cooling system and the control system.

7. The method for monitoring high-efficiency heat dissipation of axial magnetic field motor windings according to claim 1, characterized in that: The edge computing processing unit is equipped with an NPU neural network processor with a computing power ≥ 2TOPS, supports real-time parallel operation of the dynamic thermal resistance model and the LSTM algorithm, and generates a winding temperature cloud map with a refresh rate ≥ 100Hz.

8. An axial magnetic field motor winding high-efficiency heat dissipation monitoring system, applied to an axial magnetic field motor winding high-efficiency heat dissipation monitoring method according to any one of claims 1 to 7, characterized in that: The system includes: a sensor array deployment module, a multi-dimensional data acquisition module, a data transmission and filtering module, a dynamic thermal resistance modeling module, a digital twin construction module, an intelligent fault diagnosis module, and an early warning and control module; The sensor array deployment module is configured to arrange no less than 24 groups of composite sensor units in a honeycomb curved array on the axial cross section and radial end of the axial magnetic field motor winding; The multi-dimensional data acquisition module is configured to collect multi-dimensional data of the winding in real time through the composite sensor unit, with a sampling frequency of not less than 100 Hz; the multi-dimensional data includes the temperature, heat flux density, insulation layer stress and cooling medium flow rate of the winding; The data transmission and filtering module is configured to transmit the collected multi-dimensional data to the edge computing processing unit and perform noise suppression through a combination of adaptive median filtering and wavelet denoising algorithm; The dynamic thermal resistance modeling module is configured to process the filtered data based on a preset dynamic thermal resistance network model; The digital twin construction module is configured to perform spatiotemporal fusion of the pre-processed multi-dimensional data through an adaptive Kalman filter algorithm, construct a digital twin model of the winding thermal distribution, calculate the three-dimensional temperature gradient field and equivalent thermal resistance in real time, and generate heat dissipation status data; The intelligent fault diagnosis module is configured to build and train an LSTM neural network model, analyze the heat dissipation status data, extract multi-dimensional feature parameters, and identify fault characteristic information such as cooling system blockage, insulation layer aging, or winding turn short circuit; The early warning and control module is configured to classify the fault level according to the identified fault feature information and trigger a three-level early warning mechanism, and at the same time encrypt the fault feature information and upload it to the cloud diagnosis platform.

9. The high-efficiency heat dissipation monitoring system for axial magnetic field motor windings according to claim 8, characterized in that: The dynamic thermal resistance modeling module, digital twin construction module and intelligent fault diagnosis module are integrated into the edge computing processing unit.

10. The high-efficiency heat dissipation monitoring system for axial magnetic field motor windings according to claim 8, characterized in that: The data transmission and filtering module adopts a dual redundant transmission architecture.

Images