A fault early warning method and system of an axial flux motor

Abstract

The application discloses a kind of fault early warning method and system of axial flux motor, it is related to motor fault early warning related technical field, including: axial flux motor is carried out the magnetic field disturbance analysis of multiple fault mode, deployment air gap magnetic field sensing network;In vehicle AI chip development fault decision component and with air gap magnetic field sensing network establishes communication interaction;With axial flux motor drive is carried out sensing collection, data back and trigger in-chip fault decision component, execute double threshold cascade determination, generate fault determination result;Send to control center, carry out fault early warning management to axial flux motor.Solve the technical problems that the prior art exists, it is difficult to carry out real-time, multiple fault mode cooperative early warning to axial flux motor in vehicle environment, leading to the technical problem that fault identification precision and operation reliability are insufficient, by combining air gap magnetic field sensing and edge side AI decision, reach the technical effect that the precision of fault identification early warning and operation reliability are improved.

Description

Technical Field

[0001] This invention relates to the technical field of motor fault early warning, specifically to a fault early warning method and system for an axial flux motor. Background Technology

[0002] The unique disc structure of axial flux motors, with the magnetic flux direction parallel to the motor shaft, offers advantages such as small axial dimensions, high torque density, and superior heat dissipation. However, their operational reliability and fault warning are becoming increasingly prominent issues. Due to their compact structure and complex electromagnetic coupling, these motors are prone to various fault modes during long-term operation, including partial demagnetization of permanent magnets, inter-turn short circuits in windings, air gap eccentricity, and bearing abnormalities. If these faults are not detected and addressed promptly, they can lead to decreased motor performance, reduced system efficiency, and even serious equipment damage and safety accidents. Currently, fault diagnosis for motors mainly includes vibration signal analysis, current characteristic analysis, temperature monitoring, and magnetic field detection. Among these, magnetic field detection directly reflects the internal electromagnetic state of the motor. The air gap magnetic field distribution of axial flux motors is extremely sensitive to faults, and different fault modes may lead to specific disturbance patterns in the magnetic field. However, existing fault warning systems often rely on offline analysis or costly external sensors, which are insufficient to meet the real-time, lightweight, and multi-fault collaborative warning requirements in automotive environments.

[0003] Therefore, current technologies face the challenge of providing real-time, multi-fault mode collaborative early warning for axial flux motors in vehicle environments, resulting in insufficient accuracy in fault identification and operational reliability. Summary of the Invention

[0004] This application provides a fault early warning method and system for axial flux motors, which solves the technical problem in the prior art that it is difficult to perform real-time, multi-fault mode collaborative early warning for axial flux motors in a vehicle environment, resulting in insufficient fault identification accuracy and operational reliability. By combining air gap magnetic field perception and edge-side AI decision-making, the technical effect of improving fault identification early warning accuracy and operational reliability is achieved.

[0005] This application provides a fault early warning method for an axial flux motor. The method includes: deploying an air gap magnetic field sensing network by performing magnetic field disturbance analysis on the axial flux motor using multiple fault modes; developing a fault decision component on an onboard AI chip and establishing communication interaction with the air gap magnetic field sensing network; as the axial flux motor is driven, the air gap magnetic field sensing network synchronously performs sensing and data acquisition, transmits data back, and triggers the fault decision component embedded in the chip to perform a dual-threshold cascaded judgment and generate a fault judgment result. Specifically, a fault existence check based on rotational symmetry breaking is performed through a first fault judgment threshold, and a second fault judgment threshold is optionally triggered to perform dynamic analysis of the attractor space; the fault judgment result is sent to a control platform for fault early warning management of the axial flux motor.

[0006] In one possible implementation, the air gap magnetic field sensing network is deployed, including: retrieving the operating records of the axial flux motor and mining the magnetic field disturbance path under multiple fault modes; extracting the magnetic field gradient abrupt change region and magnetic flux density sensitive region as candidate points for sensor deployment based on the magnetic field disturbance path; and deploying miniature high-bandwidth magnetoresistive sensors between the stator teeth of the motor based on the candidate points for sensor deployment to form the air gap magnetic field sensing network.

[0007] In a possible implementation, the fault early warning method for an axial flux motor includes: the air gap magnetic field sensing network comprises three layers, with a first core sensing deployment based on key nodes for fault feature propagation, a second auxiliary sensing deployment based on coverage of the main magnetic path, and a third verification sensing deployment based on symmetrical positions.

[0008] In a possible implementation, the construction of the attraction subspace in the fault decision component includes: determining the original magnetic field data under multiple healthy operating conditions based on the operation records, and reconstructing the high-dimensional phase space through time delay and embedding dimension; using the trajectory convergence of multiple normal operating condition classes in the high-dimensional phase space as the healthy attractor structure, and the trajectory convergence of multiple abnormal operating condition classes as the fault attractor structure; using the spatial distribution of the high-dimensional phase space as the basis, and the healthy attractor structure and the fault attractor structure as spatial elements, to construct the attraction subspace.

[0009] In a possible implementation, the fault early warning method for an axial flux motor includes: for a first abnormal operating condition class, determining a convergence kernel based on maximum probability density for trajectory points reconstructed from original magnetic field data in a high-dimensional phase space, and determining a geometric boundary based on a dense convergence pattern through clustering; determining a first fault attractor structure based on the convergence kernel and the geometric boundary, and adding a first dynamic identifier to the first fault attractor structure; wherein the first dynamic identifier includes geometric features, statistical features, and topological structure, wherein the geometric features include attractor complexity and initial value sensitivity, the statistical features include information generation rate and regression tendency, and the topological structure includes topological feature stability.

[0010] In a possible implementation, a fault decision component is developed, comprising: decomposing the three-dimensional spatial distribution of the air gap magnetic field of the axial flux motor into a combination of baseline patterns of different symmetry orders based on the rotational symmetry spatial characteristic pattern of the healthy operating condition, and constructing a three-dimensional rotational symmetry group; constructing a first fault judgment threshold based on the three-dimensional rotational symmetry group by quantifying the symmetry breaking entropy; constructing a second fault judgment threshold based on the attractor space; and constructing a fault decision component by cascading the first fault judgment threshold and the second fault judgment threshold, which is then embedded in an automotive AI chip.

[0011] In a possible implementation, data is transmitted back and the fault decision component embedded in the chip is triggered to perform a dual-threshold cascaded judgment and generate a fault judgment result. This includes: based on the air gap magnetic field sensing network, the air gap magnetic field signal is synchronously sensed and collected as the axial flux motor starts; based on the communication interaction between the air gap magnetic field sensing network and the vehicle-mounted AI chip, the fault decision component receives the air gap magnetic field signal, triggers the first fault judgment threshold, performs instantaneous spatial feature extraction on the air gap magnetic field signal, performs symmetry breaking entropy measurement and weighted summation under the baseline mode combination, and determines the first symmetry breaking entropy; the first symmetry breaking entropy is checked against a preset threshold, and if it is less than or equal to the preset entropy value, the fault judgment process is terminated and a standard condition operation instruction is generated.

[0012] In a possible implementation, the fault early warning method for an axial flux motor includes: triggering a second fault judgment threshold if the fault exceeds a preset threshold; performing time-series trajectory point conversion on the air gap magnetic field signal in a high-dimensional phase space, performing dynamic attraction analysis, and determining the fault judgment result, wherein the dynamic attractor analysis is determined at least based on the attractor membership degree and the attraction tendency vector.

[0013] In a possible implementation, the fault early warning method for an axial flux motor includes: through communication interaction between an onboard AI chip and a control platform, the control platform generates early warning information and an emergency plan based on the fault judgment result according to a multi-level early warning mode; performs onboard alarm management based on the early warning information; and performs braking control management of the axial flux motor according to the emergency plan.

[0014] This application also provides a fault early warning system for an axial flux motor. The system includes: a magnetic field disturbance analysis module, used to deploy an air gap magnetic field sensing network by performing magnetic field disturbance analysis on the axial flux motor for multiple fault modes; a communication interaction establishment module, used to develop a fault decision component in an on-board AI chip and establish communication interaction with the air gap magnetic field sensing network; a fault judgment result generation module, used to synchronously collect data from the air gap magnetic field sensing network as the axial flux motor is driven, transmit the data back, and trigger the fault decision component embedded in the chip to perform a dual-threshold cascaded judgment to generate a fault judgment result. The first fault judgment threshold is used to perform a fault existence check based on rotational symmetry breaking, and the second fault judgment threshold can be optionally triggered to perform dynamic analysis of the attractor space; and a fault early warning management module, used to send the fault judgment result to a control platform for fault early warning management of the axial flux motor.

[0015] This application proposes a fault early warning method and system for axial flux motors. The method involves performing multi-fault mode magnetic field disturbance analysis on the axial flux motor and deploying an air gap magnetic field sensing network. A fault decision component is developed on an onboard AI chip and establishes communication with the air gap magnetic field sensing network. Sensing data is collected during the axial flux motor drive, and the data is transmitted back to trigger the chip-embedded fault decision component, which performs a dual-threshold cascaded judgment to generate a fault judgment result. This result is then sent to the control center for fault early warning management of the axial flux motor. This solution addresses the technical problem in existing technologies where real-time, multi-fault mode collaborative early warning for axial flux motors in an onboard environment is difficult, leading to insufficient fault identification accuracy and operational reliability. By combining air gap magnetic field sensing with edge-side AI decision-making, the method achieves improved technical effects in terms of fault identification and early warning accuracy and operational reliability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a schematic flowchart of a fault early warning method for an axial flux motor provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of a fault warning system for an axial flux motor provided in an embodiment of this application.

[0019] Figure labeling: Magnetic field disturbance analysis module 10, communication interaction establishment module 20, fault judgment result generation module 30, fault early warning management module 40. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0021] This application provides a fault early warning method for an axial flux motor, such as... Figure 1 As shown, the method includes: Step S100: By performing multi-mode fault field disturbance analysis on the axial flux motor, an air gap magnetic field sensing network is deployed.

[0022] Preferably, a multi-mode fault field disturbance analysis is performed on the axial flux motor. Specifically, multiple fault modes of the axial flux motor are defined, including but not limited to local demagnetization of permanent magnets, inter-turn short circuits in windings, static / dynamic eccentricity of the air gap, and bearing damage. For each fault mode, the specific disturbance characteristics induced in the air gap magnetic field are analyzed through finite element electromagnetic simulation. The magnetic field disturbance caused by each fault mode is quantified into measurable changes in physical quantities, such as abnormal radial / tangential magnetic flux density amplitude at a specific spatial location, enhanced harmonic components, disruption of the spatial symmetry of the magnetic field distribution, and local abrupt changes in the magnetic field gradient. Identify the air gap space region most sensitive to the target fault group. For example, demagnetization of permanent magnets may cause a significant decrease in magnetic flux density at the corresponding circumferential position, while air gap eccentricity may cause periodic distortion of the magnetic field distribution at a specific spatial angle. Then, select magnetic sensors that are suitable for strong electromagnetic environments, have high bandwidth, and small size. Based on the specific axial depth and circumferential / radial position of the air gap space region on the stator or rotor side of the motor, arrange the sensors in an array to determine the air gap magnetic field sensing network. Ensure that it can cover all identified key monitoring areas and collect enough information to reconstruct the spatial distribution pattern of the air gap magnetic field or its characteristic modes.

[0023] Furthermore, step S100 also includes retrieving the operating records of the axial flux motor and mining the magnetic field disturbance path under multiple fault modes; extracting the magnetic field gradient abrupt change region and magnetic flux density sensitive region as candidate points for sensor deployment based on the magnetic field disturbance path; and deploying miniature high-bandwidth magnetoresistive sensors between the stator teeth of the motor based on the candidate points for sensor deployment to form an air gap magnetic field sensing network.

[0024] Preferably, the operating records of the axial flux motor during its historical operation under different health conditions are retrieved through simulation, experiment, or pre-installed sensors to determine the spatiotemporal distribution data of the air gap magnetic field. The different health conditions include normal operating conditions and known fault conditions such as demagnetization, short circuit, and eccentricity. The magnetic field data corresponding to the multiple fault modes in the operating records are analyzed. By comparing with the normal magnetic field benchmark, the magnetic field anomalies caused by each fault development process are extracted and quantified. The trajectory of the magnetic field disturbance path in the motor air gap space with time or fault degree is determined to describe the spatial propagation direction and key influence area of ​​the fault characteristics. Then, based on the magnetic field disturbance path, the spatial gradient distribution of the air gap magnetic field is calculated. Abrupt changes in the magnetic field gradient value or the appearance of local extrema during specific faults are identified, typically corresponding to locations with significant fault characteristics and high signal change rates, making them sensitive to faults. Simultaneously, the magnetic flux density sensitive areas are analyzed under different fault modes, showing the largest relative change in magnetic flux density relative to the normal reference or exhibiting abnormal patterns such as depressions, bulges, or distortions. These abrupt magnetic gradient regions and magnetic flux density sensitive areas are spatially located and marked in the geometric model of the motor air gap, serving as candidate points for sensor deployment. Based on these candidate points and the actual physical structure of the motor stator, and while meeting requirements for installation space, mechanical safety, and electromagnetic shielding, miniature high-bandwidth magnetoresistive sensors, such as anisotropic magnetoresistive (AMR) or tunneling magnetoresistive (TMR) sensors, are deployed at specific locations between the stator teeth, such as tooth tips, tooth sides, or at specific depths in the tooth grooves. Finally, all deployed sensor nodes are wirelessly connected to form an air gap magnetic field sensing network for synchronous, multi-point, and direct measurement of the air gap magnetic field.

[0025] Furthermore, step S100 also includes the following: the air gap magnetic field sensing network comprises three layers, with the first core sensing deployment based on the fault feature propagation key nodes, the second auxiliary sensing deployment based on the coverage of the main magnetic path, and the third verification sensing deployment based on symmetrical positions.

[0026] Preferably, by analyzing electromagnetic field simulations or experimental data of multiple fault modes of the axial flux motor, the key locations of magnetic field disturbance characteristics within the air gap space of the motor when each fault occurs are determined, and miniature high-bandwidth magnetoresistive sensors are deployed at each location. Based on the electromagnetic design of the motor, the main magnetic path of the main magnetic flux in the air gap under normal healthy conditions is analyzed, which usually corresponds to the region with the minimum magnetoresistive resistance and high magnetic flux density. Sensors are deployed along the main magnetic path in sections or nodes not covered by the core sensor to continuously monitor the overall health status of the magnetic path and capture disturbances that may propagate along the magnetic path. Utilizing the rotational symmetry of the magnetic field distribution of the axial flux motor under healthy conditions, verification sensors are deployed at geometrically or electromagnetically symmetrical locations of the sensors deployed for the first core sensor and the second auxiliary sensor, for example, at locations separated by 180 electrical degrees or specific mechanical angles. These verification sensors are used to cross-verify the readings of the first and second layer sensors and directly quantify key fault indicators of rotational symmetry failure, thereby improving the robustness and reliability of fault diagnosis.

[0027] Step S200: Develop a fault decision component in the vehicle-mounted AI chip and establish communication interaction with the air gap magnetic field sensing network.

[0028] Preferably, based on the fault diagnosis requirements of axial flux motors, a fault decision component is developed in the vehicle-mounted AI chip. The fault decision component and the air gap magnetic field sensing network establish communication interaction through the vehicle-mounted domain controller or a dedicated communication bus. The fault decision component is used to receive raw or pre-processed data from the air gap magnetic field sensing network and output results such as fault existence judgment, fault type identification, and severity assessment. The air gap magnetic field sensing network synchronously collects data at a preset sampling rate and transmits it to the designated input buffer of the vehicle-mounted AI chip through a communication link in a periodic sending or event-triggered manner. The fault decision component reads the latest magnetic field data from the input buffer through an interrupt service routine or polling method, thereby ensuring real-time acquisition of data from the magnetic field sensor network.

[0029] Furthermore, step S200 also includes determining the original magnetic field data under multiple healthy operating conditions based on the operation record, reconstructing the high-dimensional phase space through time delay and embedding dimension; using the trajectory convergence of multiple normal operating condition classes in the high-dimensional phase space as the healthy attractor structure, and using the trajectory convergence of multiple abnormal operating condition classes as the fault attractor structure; using the spatial distribution of the high-dimensional phase space as the basis, and using the healthy attractor structure and the fault attractor structure as spatial elements, to form an attractor space.

[0030] Preferably, the original magnetic field data of the axial flux motor under various health conditions, collected by the air gap magnetic field sensing network, is extracted from historical operation records. A high-dimensional phase space is reconstructed through time delay and embedding dimension. Specifically, the magnetic field sensor signal that best represents the system state is selected as the observation time-series data. Time delay and false nearest neighbor are applied to calculate the time delay and embedding dimension. For the original one-dimensional time-series data, it is mapped to a high-dimensional phase space. As time progresses, these points form a trajectory in the high-dimensional phase space. The original magnetic field data under all multiple health conditions are reconstructed to obtain trajectories for different health conditions, which are then converged into health attractor structures and fault attractor structures, respectively representing the geometric structure of all healthy dynamics and all fault modes of the system. Then, the high-dimensional Euclidean space used to reconstruct the phase space is defined as the basis, and the health attractor structure and fault attractor structure are used as spatial elements to form an attractor space, where each element represents a complex geometry.

[0031] Furthermore, step S200 also includes, for the first abnormal operating condition class, determining the convergence kernel based on the maximum probability density for the trajectory points reconstructed based on the original magnetic field data in the high-dimensional phase space, and determining the geometric boundary based on the dense convergence pattern through clustering; determining the first fault attractor structure according to the convergence kernel and the geometric boundary, and adding a first dynamic identifier to the first fault attractor structure; wherein, the first dynamic identifier includes geometric features, statistical features and topological structure, wherein, the geometric features include attractor complexity and initial value sensitivity, the statistical features include information generation rate and regression tendency, and the topological structure includes topological feature stability.

[0032] Preferably, the first abnormal operating condition class is any one of multiple abnormal operating conditions. All historical original magnetic field time series data of the first abnormal operating condition class are reconstructed through time delay and embedding dimension and mapped to the same high-dimensional phase space to form a large number of trajectory points. For the trajectory points reconstructed based on the original magnetic field data in the high-dimensional phase space, kernel density estimation is used to calculate the spatial probability density distribution of the trajectory points. The region with the highest probability density is determined as the convergence kernel, which represents the most frequently accessed dynamic mode of the system under this fault state. Then, grid-based clustering or density clustering is applied to cluster all trajectory points in the phase space. Density-connected dense point sets are identified as the same cluster, and sparse noise points are distinguished. The outer contour of the main dense point clusters obtained by this clustering is extracted to form a geometric boundary based on the dense convergence morphology, which is used to define the main activity range of the fault dynamic in the phase space.

[0033] Preferably, the first fault attractor structure is determined based on the convergence kernel and geometric boundary, and a first dynamic identifier is added to the first fault attractor structure to accurately characterize and distinguish different attractor structures. The first dynamic identifier includes geometric features, statistical features, and topological structure. The geometric features include attractor complexity and initial value sensitivity. The attractor complexity is quantified using correlation dimension or fractal dimension to describe the spatial filling degree and self-similarity of the attractor structure, reflecting the complexity of the system dynamics. The initial value sensitivity is quantified using the maximum Lyapunov exponent to describe the exponential change in two infinitely close trajectories in phase space over time. The average rate of dispersion; statistical characteristics include the rate of information generation and regression tendency. The rate of information generation is quantified using Kolmogorov-entropy or its approximation to measure the average rate at which the system dynamically generates uncertainty. The regression tendency is measured by analyzing the quantitative characteristics of the trajectory point recurrence graph. It is used to describe the probability that the system state will return to a similar historical region, reflecting the periodicity or stability of the dynamics; topological structure includes topological feature stability, which refers to the robustness of the topological invariants of the attractor structure, such as the Betti number calculated based on continuous homology and persistent barcodes. It is used to describe the global topological features of attractors, such as connectivity, ring structure, and voids.

[0034] Furthermore, step S200 also includes: based on the three-dimensional spatial distribution of the air gap magnetic field of the axial flux motor, decomposing it into a combination of baseline patterns of different symmetry orders according to the rotational symmetry spatial characteristic pattern based on the healthy operating condition, and constructing a three-dimensional rotational symmetry group; constructing a first fault judgment threshold based on the three-dimensional rotational symmetry group by quantifying the symmetry breaking entropy; constructing a second fault judgment threshold according to the attractor space; and constructing a fault decision component by cascading the first fault judgment threshold and the second fault judgment threshold, which is embedded in the vehicle AI chip.

[0035] Preferably, under various healthy operating conditions of the motor, the three-dimensional spatial distribution data of the motor is obtained by using an air gap magnetic field sensing network, that is, the distribution of magnetic field strength / magnetic flux density in the axial, radial, and circumferential dimensions. Under fault-free conditions, the magnetic field distribution should have ideal rotational symmetry determined by the number of pole pairs and structure of the motor. The three-dimensional magnetic field spatial distribution data under healthy operating conditions is decomposed into a combination of baseline patterns of different symmetry orders based on the rotational symmetry spatial characteristic pattern. For example, spherical harmonic functions and cylindrical harmonic functions are used for decomposition. Each basis function pattern represents a specific spatial symmetry form, and then a three-dimensional rotational symmetry group is constructed to characterize the baseline pattern of ideal symmetry and all possible rotational operations that conform to the physics of the motor. This is used to accurately describe all the spatial symmetry rules that the air gap magnetic field distribution of the motor should follow under healthy conditions.

[0036] Preferably, the symmetry breaking entropy is used to measure the degree of deviation of the real-time measured three-dimensional magnetic field distribution from the healthy three-dimensional rotational symmetry group benchmark. Specifically, the real-time magnetic field data is also projected onto the same baseline pattern space, and the symmetry breaking entropy value is calculated by comparing the measured pattern coefficients with the healthy benchmark pattern coefficients to comprehensively reflect the severity and extent of deviations at all symmetry orders. The lower the symmetry breaking entropy value, the better the symmetry is maintained; the higher the symmetry breaking entropy value, the more severe and disordered the symmetry is broken. Based on a large amount of healthy and known minor anomaly data, a threshold for symmetry breaking entropy is statistically determined as the first fault judgment threshold to perform rapid and low-computational-cost anomaly screening.

[0037] Preferably, in the attractor space, quantitative criteria are defined to determine whether the real-time data trajectory belongs to a healthy or faulty attractor. These criteria include, for example, the Hausdorff distance or average minimum distance from the real-time trajectory to the convergence kernel of each attractor, and comparing the similarity between the dynamic identifier calculated from the real-time trajectory and the pre-stored identifiers of each attractor. This constructs a second fault judgment threshold for accurate fault type identification and classification after confirming anomalies. By cascading the first and second fault judgment thresholds, the symmetry breaking entropy of the real-time data is calculated. If it is lower than the first fault judgment threshold, the data is considered healthy, and the process terminates. If it is higher than the first fault judgment threshold, a second-level refined judgment is triggered, initiating a computationally complex attractor space analysis. The specific fault type is determined based on the second fault judgment threshold, ultimately forming a fault decision component. This component is embedded in the vehicle-mounted AI chip, completing local integration on the edge computing node of the electric off-road motorcycle.

[0038] In step S300, driven by the axial flux motor, the air gap magnetic field sensing network synchronously performs sensing and data acquisition, transmits data back, and triggers the fault decision component embedded in the chip to perform dual-threshold cascade judgment and generate a fault judgment result. The first fault judgment threshold performs a fault existence check based on rotational symmetry breaking, and the second fault judgment threshold is optionally triggered to perform dynamic analysis of the attractor space.

[0039] Preferably, during the operation of the axial flux motor, all sensors in the magnetic field sensing network deployed in its air gap simultaneously begin to continuously measure the magnetic field at their location, obtaining a spatially synchronized and time-aligned three-dimensional magnetic field sequence. The data acquisition unit in the sensing network collects and packages the measurement values ​​of each sensor node and sends them to the vehicle network via the vehicle communication link. The data is then transmitted to the designated input buffer of the vehicle AI chip and triggers the fault decision component embedded in the chip. A dual-threshold cascaded judgment is then executed. Specifically, the processing component corresponding to the first fault judgment threshold is called, and based on the three-dimensional rotational symmetry group established under healthy operating conditions, symmetry analysis is performed on the real-time three-dimensional magnetic field data. Specifically, its symmetry breaking entropy is calculated, and the calculated symmetry breaking entropy is compared with the preset first fault judgment threshold. If the symmetry... If the symmetry breaking entropy is less than or equal to the first fault judgment threshold, it indicates that the magnetic field symmetry is good and no significant abnormality is detected. The judgment result of "standard operation" or "no fault" is output. If the symmetry breaking entropy is greater than the first fault judgment threshold, it indicates that the magnetic field symmetry is broken and there is a clear abnormality. The judgment result of "abnormality exists" is output and the second-level judgment process is triggered. This includes reconstructing the real-time magnetic field signal through phase space and mapping it to a pre-constructed attractor space to form a dynamic trajectory. In this space, the dynamic characteristics of this real-time trajectory are analyzed, and its spatial relationship or feature similarity with healthy attractor structures and various faulty attractor structures are calculated. This determines the most likely fault type corresponding to the abnormality and may assess its severity, such as local demagnetization of permanent magnets. Finally, a fault judgment result containing specific fault category information is determined.

[0040] Furthermore, step S300 also includes: according to the air gap magnetic field sensing network, as the axial flux motor starts, synchronous sensing and acquisition of air gap magnetic field signals are determined; according to the communication interaction between the air gap magnetic field sensing network and the vehicle-mounted AI chip, the fault decision component receives the air gap magnetic field signal, triggers the first fault judgment threshold, performs instantaneous spatial feature extraction on the air gap magnetic field signal, performs symmetry breaking entropy measurement and weighted summation under the baseline mode combination, and determines the first symmetry breaking entropy; the first symmetry breaking entropy is checked against a preset threshold, and if it is less than or equal to the preset entropy value, the fault judgment process is terminated and a standard condition operation command is generated.

[0041] Preferably, the start command or operating status of the axial flux motor is sent as a synchronous trigger signal to the air gap magnetic field sensing network. All sensor nodes in the air gap magnetic field sensing network start sampling the magnetic field physical quantities at their respective locations at the same moment they receive the trigger signal, determine the air gap magnetic field signal that completely represents the three-dimensional spatial distribution of the air gap magnetic field at that moment, and send it to the vehicle-mounted AI chip through the communication link. The fault decision component of the vehicle-mounted AI chip receives the air gap magnetic field signal, immediately triggers the first fault judgment threshold to perform the first-level judgment, extracts the instantaneous spatial features of the air gap magnetic field signal, projects the measured spatial distribution data onto the baseline pattern learned in advance from the health data, and obtains multiple measured coefficients, which represent the strength or weight of the corresponding baseline pattern components in the measured magnetic field distribution, so as to reflect the spatial distribution pattern of the magnetic field. The baseline pattern is a standard spatial morphological basis function obtained by intrinsic mode decomposition or spherical harmonic decomposition under the health state. Each pattern corresponds to a specific symmetry order and spatial frequency.

[0042] Preferably, the measured coefficients are compared with the ideal coefficients under a healthy baseline. For each symmetry order, the deviation between the measured coefficients and the ideal coefficients is calculated to determine the symmetry breaking entropy. Then, based on prior knowledge or sensitivity analysis, different weights are assigned to the baseline pattern breaking of different orders. The deviations of all patterns are weighted and summed to obtain a comprehensive first symmetry breaking entropy, which is used to quantify the overall disorder of the current magnetic field spatial distribution relative to the ideal healthy symmetry state. The first symmetry breaking entropy is compared with a preset threshold, which is determined based on a large amount of health data statistics and represents the upper limit of the allowable normal fluctuation. If the first symmetry breaking entropy is less than or equal to the preset threshold, it is determined that the current magnetic field distribution does not show significant symmetry breaking and belongs to the normal fluctuation range. The fault judgment process is terminated and a standard condition operation instruction is generated, indicating that "the current motor magnetic field state is normal and no alarm or further processing is required." If the first symmetry breaking entropy is greater than the preset threshold, it is determined that a significant abnormality has occurred. The process will continue and automatically trigger and enter the execution process of the second fault judgment threshold.

[0043] Furthermore, step S300 also includes: if the value is greater than a preset threshold, triggering a second fault judgment threshold; performing time-series trajectory point conversion in high-dimensional phase space on the air gap magnetic field signal, performing dynamic attraction analysis, and determining the fault judgment result, wherein the dynamic attractor analysis is determined at least based on attractor membership degree and attraction tendency vector.

[0044] Preferably, when the first symmetry breaking entropy is greater than a preset threshold, the second fault judgment threshold is triggered, and the fault decision component switches from the first-level spatial pattern analysis to the second-level temporal dynamic analysis, that is, to convert the air gap magnetic field signal into temporal trajectory points in a high-dimensional phase space. Specifically, from the current moment and the historical cache, a time series magnetic field signal of a key sensor in the air gap magnetic field sensing network is extracted. Using the same time delay and embedding dimension parameters as in the modeling stage, the phase space of the time series signal is reconstructed, including converting it into trajectory points in a high-dimensional phase space. The points are then connected in chronological order to form a dynamic trajectory in the high-dimensional phase space. Then, dynamic attraction analysis is performed, that is, to determine the attractor structure that exhibits the strongest dynamic affinity in the attractor space. The dynamic attractor analysis is at least based on the attractor membership degree and the attraction tendency vector.

[0045] Preferably, the attractor membership degree represents the degree to which the real-time trajectory belongs to or is close to a known attractor structure, and is a value between 0 and 1. The higher the attractor membership degree value, the higher the membership degree. It is determined by calculating the reciprocal of the average distance from the real-time trajectory point to the geometric boundary point of the attractor structure, or by calculating the probability of the trajectory point falling into the probability density distribution region of the attractor through kernel density estimation, or by comparing the similarity between the real-time trajectory and the dynamic identification feature vector of the attractor structure. The dimension of the attraction tendency vector is equal to the number of known attractor categories, and each component value of the vector represents the strength or probability of the real-time trajectory's dynamic evolution tending towards the corresponding attractor. It is determined by analyzing the instantaneous motion direction or short-term evolution trend of the real-time trajectory in phase space.

[0046] Preferably, a comprehensive decision is made by combining the attractor membership degree and the attraction tendency vector to generate a fault judgment result. If the membership degree to the faulty attractor is the highest and exceeds a certain confidence threshold, and the faulty I component of the attraction tendency vector is also the strongest, then it is judged as faulty I. If the membership degree to the healthy attractor is still high, but the attraction tendency vector shows that it is moving away from the healthy attractor and tending towards a faulty attractor, then it may be judged as an early fault or a developing fault, and the trend is warned. The final fault judgment result is output, which includes specific fault information, such as the fault type being local demagnetization of the permanent magnet, confidence degree 92%, healthy state, but a trend toward dynamic eccentricity is detected.

[0047] Step S400: Send the fault judgment result to the control center to perform fault early warning management for the axial flux motor.

[0048] Step S400 further includes, through communication and interaction between the vehicle-mounted AI chip and the control center, generating early warning information and emergency plans based on the fault judgment results in the control center according to the multi-level early warning mode; performing vehicle alarm management based on the early warning information; and performing braking control management of the axial flux motor according to the emergency plan.

[0049] Preferably, the fault diagnosis results are sent from the onboard AI chip to the control platform, such as the central domain controller or vehicle controller, through communication and interaction between the onboard AI chip and the control platform. After receiving the fault diagnosis results, the control platform makes higher-level safety and control decisions. Specifically, it invokes a multi-level warning mode, mapping the fault diagnosis results to different warning levels based on the fault type and diagnostic confidence / severity. For example, a level 1 warning corresponds to minor anomalies or early trends, such as a slight exceedance of the symmetry entropy threshold, but not confirmed by attractor analysis, in which case a log is recorded to increase the monitoring frequency; a level 2 warning corresponds to a confirmed non-serious fault, such as mild demagnetization, in which case a driver alarm message is generated and suggestions are made. Power-limited operation; Level 3 warning corresponds to serious or rapidly developing faults, such as severe short circuits, which may lead to demagnetization chain reactions, generating the highest level alarm and preparing to execute emergency control; then, based on the fault judgment result, warning information is generated, such as "slight deterioration of motor performance, please schedule an inspection" under Level 2 warning, and "motor fault! Safety-limited operation is about to be implemented" under Level 3 warning. At the same time, multiple control command sequences for vehicle actuators are generated based on the warning information, i.e., emergency plans. For example, the emergency plan for Level 3 warning may include "Command 1 requests the drive system to reduce torque to 50%; Command 2 requests to switch to limp mode if the fault continues within 3 seconds; Command 3 prepares to trigger safety braking logic". The control center sends the generated warning information to the in-vehicle infotainment system or instrument panel. Depending on the urgency of the information, it issues warnings to the driver using icons, text, sound, and voice, completing the warning at the human-machine interaction level. Finally, based on the emergency plan, it manages the braking of the axial flux motor. That is, the control center sends the control commands in the emergency plan to the actuator controller through the real-time control bus. The control commands of the motor controller may include torque reduction commands, speed limit commands, and shutdown request commands, while the control commands of the vehicle controller may include commands to request coordinated braking to achieve smooth deceleration or safe stopping of the vehicle.

[0050] In the above text, refer to Figure 1 A fault early warning method for an axial flux motor according to an embodiment of the present invention is described in detail. Next, reference will be made to... Figure 2 A fault early warning system for an axial flux motor according to an embodiment of the present invention is described.

[0051] A fault early warning system for an axial flux motor according to an embodiment of the present invention addresses the technical problem in the prior art where it is difficult to provide real-time, multi-fault mode collaborative early warning for axial flux motors in an on-board environment, resulting in insufficient accuracy in fault identification and operational reliability. By combining air gap magnetic field sensing with edge-side AI decision-making, the system achieves the technical effect of improving the accuracy of fault identification and early warning and operational reliability. Figure 2As shown, a fault early warning system for an axial flux motor includes: a magnetic field disturbance analysis module 10, a communication interaction establishment module 20, a fault judgment result generation module 30, and a fault early warning management module 40.

[0052] The magnetic field disturbance analysis module 10 is used to deploy an air gap magnetic field sensing network by performing multi-fault mode magnetic field disturbance analysis on the axial flux motor; the communication interaction establishment module 20 is used to develop a fault decision component on the vehicle-mounted AI chip and establish communication interaction with the air gap magnetic field sensing network; the fault judgment result generation module 30 is used to synchronously collect data with the air gap magnetic field sensing network as the axial flux motor is driven, transmit the data back and trigger the fault decision component embedded in the chip to perform dual-threshold cascade judgment and generate a fault judgment result. Among them, the first fault judgment threshold is used to perform a fault existence check based on rotational symmetry breaking, and the second fault judgment threshold can be optionally triggered to perform dynamic analysis of the attractor space; the fault early warning management module 40 is used to send the fault judgment result to the control center to perform fault early warning management for the axial flux motor.

[0053] The specific configuration of the magnetic field disturbance analysis module 10 will be described in detail below. The magnetic field disturbance analysis module 10 further includes: retrieving the operating records of the axial flux motor to mine the magnetic field disturbance path under multiple fault modes; extracting the magnetic field gradient abrupt change region and magnetic flux density sensitive region as candidate points for sensor deployment based on the magnetic field disturbance path; and deploying miniature high-bandwidth magnetoresistive sensors between the stator teeth of the motor based on the candidate points for sensor deployment to form an air gap magnetic field sensing network.

[0054] The specific configuration of the magnetic field disturbance analysis module 10 will be described in detail below. The magnetic field disturbance analysis module 10 further includes: the air gap magnetic field sensing network comprises three layers, with the first core sensing deployment based on key nodes for fault feature propagation, the second auxiliary sensing deployment based on coverage of the main magnetic path, and the third verification sensing deployment based on symmetrical positions.

[0055] The specific configuration of the communication interaction establishment module 20 will be described in detail below. The communication interaction establishment module 20 further includes: determining the original magnetic field data under multiple healthy operating conditions based on the operation record; reconstructing a high-dimensional phase space through time delay and embedding dimension; using the trajectory convergence of multiple normal operating condition classes in the high-dimensional phase space as a healthy attractor structure, and the trajectory convergence of multiple abnormal operating condition classes as a fault attractor structure; and constructing an attractor space based on the spatial distribution of the high-dimensional phase space and using the healthy attractor structure and the fault attractor structure as spatial elements.

[0056] The specific configuration of the communication interaction establishment module 20 will be described in detail below. The communication interaction establishment module 20 further includes: for the first abnormal operating condition class, determining the convergence kernel based on the maximum probability density for the trajectory points reconstructed from the original magnetic field data in the high-dimensional phase space, and determining the geometric boundary based on the dense convergence pattern through clustering; determining the first fault attractor structure according to the convergence kernel and the geometric boundary, and adding a first dynamic identifier to the first fault attractor structure; wherein, the first dynamic identifier includes geometric features, statistical features, and topological structure, wherein the geometric features include attractor complexity and initial value sensitivity, the statistical features include information generation rate and regression tendency, and the topological structure includes topological feature stability.

[0057] The specific configuration of the communication interaction establishment module 20 will be described in detail below. The communication interaction establishment module 20 further includes: based on the three-dimensional spatial distribution of the air gap magnetic field of the axial flux motor, decomposing it into a combination of baseline patterns of different symmetry orders according to the rotational symmetry spatial characteristic pattern based on healthy operating conditions, and constructing a three-dimensional rotational symmetry group; constructing a first fault judgment threshold based on the three-dimensional rotational symmetry group by quantifying the symmetry breaking entropy; constructing a second fault judgment threshold based on the attractor subspace; and constructing a fault decision component by cascading the first fault judgment threshold and the second fault judgment threshold, which is embedded in the vehicle-mounted AI chip.

[0058] The specific configuration of the fault judgment result generation module 30 will be described in detail below. The fault judgment result generation module 30 further includes: based on the air gap magnetic field sensing network, synchronously sensing and acquiring the air gap magnetic field signal as the axial flux motor starts; based on the communication interaction between the air gap magnetic field sensing network and the vehicle-mounted AI chip, the fault decision component receives the air gap magnetic field signal, triggers a first fault judgment threshold, performs instantaneous spatial feature extraction on the air gap magnetic field signal, performs symmetry breaking entropy measurement and weighted summation under baseline mode combination, and determines the first symmetry breaking entropy; the first symmetry breaking entropy is checked against a preset threshold; if it is less than or equal to the preset entropy value, the fault judgment process is terminated and a standard condition operation command is generated.

[0059] The specific configuration of the fault judgment result generation module 30 will be described in detail below. The fault judgment result generation module 30 further includes: if the value is greater than a preset threshold, triggering a second fault judgment threshold; performing time-series trajectory point conversion in high-dimensional phase space on the air gap magnetic field signal, performing dynamic attraction analysis, and determining the fault judgment result, wherein the dynamic attractor analysis is determined at least based on the attractor membership degree and the attraction tendency vector.

[0060] The specific configuration of the fault warning management module 40 will be described in detail below. The fault warning management module 40 further includes: through communication and interaction between the vehicle-mounted AI chip and the control center, the control center generates warning information and emergency plans based on the fault judgment results according to the multi-level warning mode; performs vehicle alarm management based on the warning information; and performs braking control management of the axial flux motor according to the emergency plan.

[0061] The fault early warning system for an axial flux motor provided in this embodiment of the invention can execute the fault early warning method for an axial flux motor provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A fault early warning method for an axial flux motor, characterized in that, The method includes: By performing multi-mode fault field disturbance analysis on axial flux motors, an air gap magnetic field sensing network is deployed. A fault decision component is developed for the vehicle-mounted AI chip, and communication interaction is established with the air gap magnetic field sensing network. Driven by the axial flux motor, the air gap magnetic field sensing network synchronously performs sensing and data acquisition, transmits data back and triggers the fault decision component embedded in the chip to perform dual-threshold cascade judgment and generate fault judgment results. Among them, the first fault judgment threshold performs a fault existence check based on rotational symmetry breaking, and the second fault judgment threshold is optionally triggered to perform dynamic analysis of the attractor space. The fault judgment results are sent to the control center for fault early warning management of the axial flux motor.

2. The fault early warning method for an axial flux motor as described in claim 1, characterized in that, Deploying an air gap magnetic field sensing network includes: Retrieve the operating records of the axial flux motor to explore the magnetic field disturbance path under multiple fault modes; Based on the magnetic field disturbance path, the magnetic field gradient abrupt region and the magnetic flux density sensitive region are extracted as candidate points for sensor deployment. Based on the aforementioned candidate points for sensor deployment, miniature high-bandwidth magnetoresistive sensors are deployed between the stator teeth of the motor to form an air gap magnetic field sensing network.

3. The fault early warning method for an axial flux motor as described in claim 2, characterized in that, The air gap magnetic field sensing network comprises three layers: a first core sensing deployment based on key nodes for fault feature propagation, a second auxiliary sensing deployment based on coverage of the main magnetic path, and a third verification sensing deployment based on symmetrical locations.

4. The fault early warning method for an axial flux motor as described in claim 2, characterized in that, In developing the fault decision component, the construction of the attraction subspace includes: Based on the operation records, the original magnetic field data under various health conditions are determined, and the high-dimensional phase space is reconstructed by time delay and embedding dimension. The trajectory convergence of the multivariate normal operating condition class in the high-dimensional phase space is taken as the health attractor structure, and the trajectory convergence of the multivariate abnormal operating condition class is taken as the fault attractor structure. Based on the spatial distribution of the high-dimensional phase space, and using the healthy attractor structure and the fault attractor structure as spatial elements, an attractor space is constructed.

5. A fault early warning method for an axial flux motor as described in claim 4, characterized in that, For the first abnormal operating condition, the convergence kernel based on the maximum probability density is determined for the trajectory points reconstructed based on the original magnetic field data in the high-dimensional phase space, and the geometric boundary based on the dense convergence pattern is determined by clustering. The first fault attractor structure is determined based on the convergence kernel and geometric boundary, and a first dynamic identifier is added to the first fault attractor structure. The first dynamic identifier includes geometric features, statistical features, and topological structure. The geometric features include attractor complexity and initial value sensitivity, the statistical features include information generation rate and regression tendency, and the topological structure includes topological feature stability.

6. The fault early warning method for an axial flux motor as described in claim 5, characterized in that, Develop fault decision-making components, including: Based on the three-dimensional spatial distribution of the air gap magnetic field of the axial flux motor, the rotational symmetry spatial characteristic mode based on the healthy working condition is decomposed into a combination of baseline modes of different symmetry orders to construct a three-dimensional rotational symmetry group. By quantifying the symmetry breaking entropy, a first fault judgment threshold based on the three-dimensional rotational symmetry group is constructed. Based on the attractor subspace, a second fault judgment threshold is constructed; By cascading the first fault judgment threshold and the second fault judgment threshold, a fault decision component is formed and embedded in the vehicle AI chip.

7. The fault early warning method for an axial flux motor as described in claim 1, characterized in that, Data is transmitted back and the chip's embedded fault decision component is triggered to perform a dual-threshold cascaded judgment, generating a fault judgment result, including: According to the air gap magnetic field sensing network, the air gap magnetic field signal is synchronously sensed and collected to determine the axial flux motor as it starts. Based on the communication interaction between the air gap magnetic field sensing network and the vehicle-mounted AI chip, the fault decision component receives the air gap magnetic field signal, triggers the first fault judgment threshold, extracts instantaneous spatial features of the air gap magnetic field signal, performs symmetry breaking entropy measurement and weighted summation under the baseline mode combination, and determines the first symmetry breaking entropy. The first symmetry breaking entropy is compared with the preset threshold. If it is less than or equal to the preset entropy value, the fault judgment process is terminated and a standard condition operation instruction is generated.

8. A fault early warning method for an axial flux motor as described in claim 7, characterized in that, If the value exceeds the preset threshold, the second fault judgment threshold is triggered; The air gap magnetic field signal is converted into time-series trajectory points in high-dimensional phase space, and dynamic attraction analysis is performed to determine the fault judgment result. The dynamic attractor analysis is based at least on attractor membership degree and attraction tendency vector.

9. A fault early warning method for an axial flux motor as described in claim 1, characterized in that, Through communication and interaction between the vehicle-mounted AI chip and the control center, the control center generates early warning information and emergency plans based on the fault judgment results according to the multi-level early warning mode. Vehicle alarm management is performed based on the aforementioned warning information, and braking control management is performed on the axial flux motor according to the aforementioned emergency plan.

10. A fault early warning system for an axial flux motor, characterized in that, The system is used to implement a fault early warning method for an axial flux motor according to any one of claims 1 to 9, the system comprising: The magnetic field disturbance analysis module is used to deploy an air gap magnetic field sensing network by performing multi-fault mode magnetic field disturbance analysis on axial flux motors. A communication interaction establishment module is used to develop a fault decision component in the vehicle-mounted AI chip and establish communication interaction with the air gap magnetic field sensing network. The fault judgment result generation module is used to synchronously collect data from the air gap magnetic field sensing network driven by the axial flux motor, transmit the data back and trigger the fault decision component embedded in the chip to perform dual-threshold cascade judgment and generate fault judgment results. The first fault judgment threshold is used to perform a fault existence check based on rotational symmetry breaking, and the second fault judgment threshold can be optionally triggered to perform dynamic analysis of the attractor space. The fault early warning management module is used to send the fault judgment results to the control center to perform fault early warning management for the axial flux motor.

Images