A method and instrument for evaluating the health status of new energy vehicle motors

Abstract

The present application belongs to the technical field of motor health evaluation and analysis, and particularly relates to a method and instrument for evaluating the health state of a motor of a new energy vehicle, comprising the following steps: collecting and synchronizing multi-modal motor signals, and generating a trigger event according to changes in the motor signals; constructing a four-dimensional tensor data set, fusing motor signal data through a graph neural network, and extracting cross-domain fault features based on the motor signal data; applying an improved wavelet packet decomposition to perform multi-resolution analysis on the cross-domain fault signals, and identifying implicit harmonics and transient impacts; extracting cross-domain fault features through an edge computing node, aggregating multi-device data to train a global diagnostic model using a federated learning framework, and a fault classification method under dynamic working conditions; and based on the global diagnostic model and the fault type under dynamic working conditions. The present application can realize multi-source synchronization of data in the evaluation of the health state of the motor, break through the limitations of single-modal data, and improve the accuracy of motor health evaluation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of motor health evaluation analysis, and particularly relates to a method and instrument for evaluating the health state of a new energy automobile motor. BACKGROUND

[0002] Motor health evaluation analysis is a method for evaluating the current state and predicting future performance of a motor by monitoring and analyzing its operating data, mainly involving data collection, data analysis, health evaluation, etc. of the motor. Motor health evaluation analysis has wide application in the fields of industry, transportation, household appliances, etc. and can reduce equipment failure, reduce maintenance cost, improve production efficiency and safety. In particular, the service life and safety of the new energy automobile motor play a crucial role in the vehicle operation process.

[0003] Problems existing in the prior art:

[0004] The evaluation method for the motor in the prior art mainly uses single data analysis, such as separate analysis based on vibration signals and current signals, which cannot achieve comprehensive analysis. Therefore, due to the small amount of data sources obtained during the analysis process, the precision of data analysis is low, and data island phenomenon also occurs. SUMMARY

[0005] The purpose of the present application is to provide a method and instrument for evaluating the health state of a new energy automobile motor, which can realize multi-source synchronization of data in the evaluation of the health state of the motor, break through the limitation of single modal data, and improve the precision of motor health evaluation.

[0006] The technical solution adopted by the present application is as follows:

[0007] A method for evaluating the health state of a new energy automobile motor, comprising the following steps:

[0008] Collecting and synchronizing multi-modal motor signals, and generating trigger events according to changes in the motor signals;

[0009] Constructing a four-dimensional tensor data set, fusing motor signal data through a graph neural network, and extracting cross-domain fault features based on the motor signal data;

[0010] Applying an improved wavelet packet decomposition to perform multi-resolution analysis on the cross-domain fault signals, and identifying implicit harmonics and transient impacts;

[0011] Extracting cross-domain fault features through an edge computing node, aggregating multi-device data to train a global diagnostic model using a federated learning framework, and a fault classification method under dynamic working conditions;

[0012] Evaluating the health state of the motor based on the global diagnostic model and the fault type under dynamic working conditions.

[0013] The trigger event generated by the motor signal change includes:

[0014] By establishing a multi-modal data fluctuation value model of the multi-modal operating state of the motor, comparing the abnormal change value of the multi-modal data fluctuation value signal change under the multi-modal actual operating state of the motor with the multi-modal data fluctuation value model, and comparing the fitting degree between the two signals, whether the event is triggered is determined.

[0015] The four-dimensional tensor data set includes a four-dimensional data set constructed from any of the multi-modal motor signals, and the four-dimensional data set includes X, Y, and Z axis data signals in the spatial dimension and time dimension data signals.

[0016] According to the collection of multi-modal motor signal data, a cross-domain fusion multi-physical scene correlation model is extracted and constructed.

[0017] The graph neural network fuses motor signal data, uses the nodes of the graph neural network to represent multi-modal data, and constructs a multi-modal data evaluation system through the graph neural network.

[0018] The improved wavelet packet decomposition includes:

[0019] Adaptive basis function selection: dynamically optimize the wavelet basis according to the signal frequency band energy distribution;

[0020] Noise suppression: combined with soft threshold processing to eliminate high frequency noise and retain implicit harmonic characteristics.

[0021] The federal learning framework runs a federal learning diagnosis method, adopts the federal learning diagnosis method, adapts the pre-trained model to different vehicle types and working conditions, reduces the dependence on labeled data, and encrypts the multi-modal data through the edge computing node.

[0022] The edge computing node generates a time-frequency domain feature vector through the improved wavelet packet decomposition, and uploads new fault analysis features using an incremental learning method.

[0023] An instrument for evaluating the health status of a new energy vehicle motor, the instrument comprising:

[0024] Multi-modal sensing module: the multi-modal sensing module includes at least one group of distributed sensors, including but not limited to: vibration sensors, temperature sensors, Hall sensors, optical encoders, motor state evaluation instruments;

[0025] Dynamic synchronization control module: The dynamic synchronization control module includes an event-triggered controller and a dynamic interpolation unit. The event-triggered controller dynamically adjusts the sampling strategy based on the multimodal motor signal mutation threshold; the dynamic interpolation unit optimizes the synchronization accuracy of the multimodal motor signal through cubic spline interpolation;

[0026] Multi-domain analysis module: This module uses a graph neural network processor to achieve cross-domain feature fusion between multimodal motor signals and an improved wavelet packet decomposition accelerator to support multi-resolution frequency domain analysis and implicit harmonic detection;

[0027] Federated Learning Diagnostic Module: This module is equipped with at least one edge computing node, which is used to extract local features and upload them to the cloud in encrypted form, achieving data synchronization and end-to-end computing collaboration. It also has a federated learning engine, which dynamically updates the global diagnostic model through online learning to adapt to changes in working conditions.

[0028] Motor status evaluator: The motor status evaluator is used to realize data acquisition of the multimodal sensor module, and is provided with a data display unit for realizing the display and control of the motor health status evaluation.

[0029] According to another aspect of an embodiment of the present invention, an electronic device is provided, comprising a memory and a processor; the memory is used to store a program; and the processor executes the program to implement any one of the aforementioned methods.

[0030] According to another aspect of the embodiments of the present invention, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, any one of the aforementioned methods is implemented.

[0031] According to yet another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, any one of the aforementioned methods is implemented.

[0032] The technical effects achieved by the present invention are:

[0033] The present invention, through the collection of multimodal motor signals, can realize the combined analysis and evaluation of multiple signals for subsequent data processing, which can improve the diversity of data analysis and evaluation. Through the diversity of data, the accuracy of data analysis can be achieved, and an evaluation system for multimodal data is constructed through a graph neural network. The four-dimensional tensor data set of any one of the vibration signals, current noise signals, temperature signals, magnetic signals, and photoelectric coding signals is mapped, and the cross-domain feature correlation strength is dynamically adjusted through the edge weight Pearson correlation coefficient to improve the correlation between data.

[0034] The application solves the performance decline problem of a static graph model when the working condition changes by using a graph neural network to realize multi-physical field joint modeling through the dynamic design of nodes and edges, and through the real-time adjustment of edge weights, improves the dynamic adaptability, and further filters strong correlation edges through PCC, can focus on key fault paths (such as bearing wear -> vibration -> temperature rise -> magnetic field distortion), and improves the fault location accuracy.

[0035] The application can realize the same kind of learning or calculation method through online learning or transfer learning method under dynamic working condition, can be applied to the motor with the same data parameter, and can obtain whether the parameter appears abnormal fluctuation through the diagnostic model, if not, record the parameter law between the corresponding model and type, realize the data application under the multi-working condition state, and the system experience application, and further improve the result of the motor health state evaluation according to the big data, and improve the evaluation efficiency of the detection evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 It is a method flowchart of the application;

[0037] Figure 2 It is an internal structure schematic diagram of the instrument in the application;

[0038] Figure 3 It is a structure schematic diagram of the instrument in the application. DETAILED DESCRIPTION

[0039] In order to make the purpose and advantages of the application more clear and understandable, the application will be specifically described below in combination with examples. It should be understood that the following text is only used to describe one or several specific embodiments of the application, and does not strictly limit the specific protection scope requested by the application.

[0040] It should be noted that the terms "first", "second" and the like in the specification and claims of the application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0041] According to the embodiment of the present application, a method for evaluating the health state of a new energy vehicle motor is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0042] As shown in the accompanying drawings, a method for evaluating the health state of a new energy vehicle motor includes the following steps: Figure 1

[0043] S1, collecting and synchronizing multi-modal motor signals, and generating trigger events according to changes in motor signals;

[0044] S2, constructing a four-dimensional tensor dataset, fusing motor signal data through a graph neural network, and extracting cross-domain fault features based on motor signal data;

[0045] S3, applying an improved wavelet packet decomposition to perform multi-resolution analysis on cross-domain fault signals, and identifying implicit harmonics and transient impacts;

[0046] S4, extracting cross-domain fault features through an edge computing node, aggregating multi-device data to train a global diagnostic model using a federated learning framework, and a fault classification method under dynamic working conditions;

[0047] S5, evaluating the health state of the motor based on the global diagnostic model and the fault type under dynamic working conditions.

[0048] In the above step S1, the multi-modal motor signals include but are not limited to:

[0049] Vibration signal: the vibration signal is collected using a vibration sensor or the like, connected to a collection board through a BNC terminal, and synchronously collected at a sampling rate of 50 kHz, with 50000 sampling points per second per channel, so as to obtain motor vibration signal values, which support ICP excitation type sensors:

[0050] Current noise signal: the signal acquisition method uses a high-frequency current probe (such as a Rogowski coil or a Hall sensor, etc.), which is connected to the motor circuit through a current transformer or directly connected to the motor circuit, and collects high-frequency noise components (such as switching frequency harmonics) in the current spectrum;

[0051] Temperature signal: the signal acquisition method uses a digital temperature sensor or an infrared sensor, etc., and the use method includes embedding a thermocouple to monitor winding temperature or using a non-contact infrared thermal imager to scan the distribution of the motor surface temperature field;

[0052] Magnetic signal: the magnetic signal acquisition method uses a Hall sensor or a fluxgate sensor, etc., which can be installed outside the motor to monitor the change of magnetic field strength in real time, such as permanent magnet demagnetization monitoring, etc.​

[0053] Photoelectric coded signal: The photoelectric coded signal is obtained by using a photoelectric encoder to accurately count the motor rotation and calculate the motor rotation angle and rotation speed signals.

[0054] Among them, the multimodal motor signal includes any one or combination of the above-mentioned vibration signal, current noise signal, temperature signal, magnetic signal or photoelectric coding signal, as well as the operating environment of the motor signal during the acquisition process, such as idling state, installed state, simple transmission state, etc., which is used to realize fault assessment under various working conditions and fault assessment under dynamic working conditions, as well as data signals of the motor in different operating states. The data signals include but are not limited to signal data during instantaneous start, instantaneous stop, operation, acceleration, deceleration, start and stop.

[0055] In addition to the above-mentioned state types, multimodal motor signals also include any one or combination of vibration signals, current noise signals, temperature signals, magnetic signals or photoelectric coding signals for multi-dimensional framework acquisition, such as by using one or two or more sensors to collect motor signals.

[0056] By collecting the above-mentioned multimodal motor signals, a combination of multiple signals can be achieved for subsequent data processing, which can be used to improve the diversity of data analysis and evaluation. Through the diversity of data, the accuracy of data analysis can be achieved.

[0057] Furthermore, as an optional embodiment, the triggering event generated by the motor signal change includes:

[0058] By establishing a multimodal data fluctuation value model for the motor's multimodal operating state, the abnormal change value of the multimodal data fluctuation value model is compared with the multimodal data fluctuation value signal changes under the motor's actual multimodal operating state, and the degree of fit between the two signals is compared to determine whether an event is triggered.

[0059] Furthermore, the triggering events are captured by targeting the above-mentioned multimodal motor signal changes. The triggering conditions include establishing data fluctuation values ​​of various operating states of the motor, such as vibration and noise signals during instantaneous start, instantaneous stop, operation, acceleration, deceleration, start, and stop, and applying them to different signal types and different operating scenarios to form a standard signal model in a stable state. By comparing and evaluating the signal model formed when the motor is running, the degree of fit between the two signals is used to determine whether the condition is triggered, that is, by judging whether a point or a section of data in the data model of the motor during operation exceeds the threshold of the signal model in the stable state.

[0060] For example, the triggering condition is, for example, a sudden change in rotational speed Δrpm greater than 10%, or amplitude exceeding, temperature exceeding, magnetic field variable exceeding, current noise exceeding, volume exceeding, etc., and the event-triggered control technology is cited.

[0061] Further, the triggering condition is generated based on a preset threshold, and the signal sampling strategy is dynamically adjusted according to different motor types and operating states.

[0062] According to step S2, the four-dimensional tensor data set includes any one of the constructed four-dimensional data sets in the multi-modal motor signal, and the constructed four-dimensional tensor data set includes X, Y, Z axial data signals and time dimension data signals for independent signal types.

[0063] For example, the vibration signal four-dimensional tensor data set includes X, Y, Z axial data signals and time dimension data signals changing with time;

[0064] For example, the current noise signal four-dimensional tensor data set includes three-phase current spectrum components and current period sampling, such as fundamental amplitude, harmonic amplitude, switching frequency noise and time series;

[0065] For example, the temperature signal four-dimensional tensor data set includes temperature gradient and thermal imaging continuous frame timestamp of key parts of the motor, or temperature size, temperature range, temperature difference change amplitude and temperature change condition changing with time of independent temperature detection points;

[0066] For example, the magnetic signal four-dimensional tensor data set includes three-dimensional components of magnetic field strength (X-axis magnetic flux density, Y-axis magnetic flux density, Z-axis magnetic flux density) and vibration signal synchronous sampling time series;

[0067] For example, the photoelectric encoding signal four-dimensional tensor data set includes motor rotation angle and rotation speed, and the data obtained can realize cooperative monitoring and analysis with the vibration signal.

[0068] As an optional embodiment, the graph neural network fuses motor signal data, and uses the nodes of the graph neural network to represent multi-modal data, and constructs an evaluation system of multi-modal data through the graph neural network.

[0069] Further, the nodes of the graph neural network map X, Y, Z axial data signals and time dimension data signal features, and map according to any one of the four-dimensional tensor data sets of the vibration signal, the current noise signal, the temperature signal, the magnetic signal and the photoelectric encoding signal, and dynamically adjust the cross-domain feature correlation strength through the edge weight Pearson correlation coefficient.

[0070] Further, a cross-domain fusion multi-physical scene correlation model is extracted and constructed according to the set multi-modal motor signal data.

[0071] For example, based on the vibration signal, the data signals in the X, Y, and Z axes and the time dimension data signals that change over time are mapped, and the vibration signals of the motor X / Y / Z axes are collected through the vibration sensor;

[0072] Extract features:

[0073] Time domain characteristics: mean, variance, peak, impact pulse (such as transient impact of bearing failure).

[0074] Frequency domain features: Extract harmonic frequencies (such as electromagnetic resonance frequencies) and order amplitudes (such as speed-related orders) through FFT or improved wavelet packet decomposition.

[0075] If the time domain features are [mean, variance, peak] and the frequency domain features are [1-fold amplitude, 2-fold amplitude, ..., 5-fold amplitude], then:

[0076] Graph neural network node representation Each vibration node is represented as a vector Node vibration = [time domain features, frequency domain features] ∈ R d ;

[0077] Where d is the feature dimension (e.g. 5 dimensions in time domain + 10 dimensions in frequency domain → d = 15).

[0078] Temperature / Magnetic Field Node: Modeling thermal-magnetic coupling effects

[0079] Monitor the temperature gradient of the winding and bearing through temperature sensors, and collect the magnetic field strength and distribution through Hall sensors;

[0080] Depending on the coupling effects, the temperature increase may lead to demagnetization of the permanent magnets (reduction in magnetic field strength) or changes in the winding resistance (affecting the current characteristics).

[0081] Magnetic field distortion may be caused by mechanical wear (such as bearing eccentricity) and induce eddy current effects, further exacerbating the temperature increase.

[0082] The nodes of the graph neural network are represented as follows: Each temperature / magnetic field node is represented as a vector: Node temp / mag = [temperature gradient, magnetic field strength, coupling coefficient] ∈ R k ;

[0083] The coupling coefficient is calculated using physical models (such as the heat conduction equation) or data-driven methods.

[0084] Edge weights are dynamically adjusted using the Pearson correlation coefficient

[0085] Correlation calculation:

[0086] Edge weight measures the linear correlation between nodes through the Pearson Correlation Coefficient (PCC): ;

[0087] Among them, X, Y are the eigenvectors of the two nodes, Cov(X, Y) represents the covariance of X and Y, reflecting the direction and strength of the linear correlation between the two, σ is the standard deviation, It is the product of the standard deviations of X and Y and is used to eliminate the dimensionless effects of the eigenvectors of X and Y, making the results unitless and comparable.

[0088] By standardizing the covariance, we address the issue of covariance values ​​being affected by the fluctuations (standard deviation) of the variables. For example, if the standard deviations of X and Y are large, the covariance may be amplified, while the correlation coefficient can more accurately reflect the true relationship between the eigenvectors X and Y.

[0089] Dynamic adjustment mechanism: Recalculate the PCC between nodes based on the current working conditions (such as acceleration / climbing) and dynamically adjust the edge weights to achieve real-time updates.

[0090] Threshold screening: Only strongly correlated edges with PCC greater than 0.7 are retained to reduce redundant connections.

[0091] Graph Neural Network Edge Weight Design:

[0092] ;

[0093] in, Represents the edge weight between node i and node j, which is used to describe the strength of the association between the features of the two;

[0094] like If it is not equal to 0, then there is a direct connection between nodes i and j.

[0095] is the Pearson correlation coefficient;

[0096] |PCC| greater than 0.7 means that only strongly correlated edges with absolute values ​​greater than 0.7 are retained;

[0097] When |PCC| is less than or equal to 0.7, the edge weight is set to 0, indicating that the nodes are weakly related or unrelated, and the connection is disconnected to simplify the graph structure;

[0098] The above formula is used to capture cross-domain feature correlations (such as the strong correlation between vibration amplitude and magnetic field distortion).

[0099] Through cross-domain feature fusion and the use of graph neural networks, multi-physics field joint modeling is achieved through dynamic design of nodes and edges. The real-time adjustment of edge weights solves the performance degradation problem of static graph models when working conditions change, improves dynamic adaptability, and further screens strongly correlated edges through PCC, focusing on key fault paths (such as bearing wear → vibration → temperature rise → magnetic field distortion), thereby improving fault location accuracy.

[0100] For example, a motor experiences bearing wear:

[0101] Vibration node: The amplitude of the high-frequency impact pulse on the X / Y axis increases, and the characteristic frequency of the bearing fault (such as BPFI) appears in the frequency domain.

[0102] Temperature / Magnetic Field Node: The bearing temperature rise causes the permanent magnet to demagnetize and the magnetic field strength to decrease.

[0103] Edge weight adjustment: The PCC of vibration nodes and magnetic field nodes increased from 0.3 to 0.8, triggering edge weight enhancement. The graph neural network then associated fault paths. Through multimodal node representation and dynamic edge weight adjustment, it addressed the shortcomings of traditional methods in cross-domain feature fusion and adaptability to dynamic working conditions.

[0104] It should be further explained that, in addition to the above examples, graph neural networks can choose two processing modes during the data processing stage:

[0105] Mode 1: Processing single signal data through graph neural network;

[0106] Mode 2: Processing any two or more signal data through graph neural networks;

[0107] Among them, mode one is used to realize data collection and processing of simple evaluation status, such as it can be used in the production and delivery of a large number of motors, while mode two is used to evaluate the use of specific performance processes to improve the accuracy of health status assessment.

[0108] In step S3, the improved wavelet packet decomposition includes:

[0109] Adaptive basis function selection: Dynamically optimize the wavelet basis according to the signal frequency band energy distribution to ensure that the decomposition result is more in line with the actual signal characteristics;

[0110] Noise suppression: Combined with soft threshold processing to eliminate high-frequency noise and retain implicit harmonic features.

[0111] Furthermore, multi-resolution analysis can decompose the signal into multiple sub-bands, corresponding to different physical phenomena, such as transient impacts and electromagnetic harmonics. By analyzing the energy spectrum of each frequency band, implicit harmonics can be located, and time-frequency analysis can be used to extract transient impact characteristics, such as short-term high-frequency pulses caused by spindle wear.

[0112] Since motor fault signals come from changes in physical factors such as vibration, temperature, and magnetic fields, a single signal can be used to find a specific feature, but it is difficult to accurately locate and predict some hidden features. Therefore, multimodal signals are used for joint analysis, including but not limited to:

[0113] Vibration signal: capture mechanical shock and resonance;

[0114] Temperature signal: reflects the changing relationship between operating time and temperature, and reflects overheating or material aging;

[0115] Magnetic field signal: monitor demagnetization or current anomaly;

[0116] Photoelectric coding signal: monitor the motor rotation angle and speed;

[0117] By combining the above four signals and building a correlation model based on the timestamp, it can be used to infer motor health problems.

[0118] Furthermore, through improved wavelet packet decomposition, the above cross-domain signals can be unified into the same analysis framework to extract the fault characteristics of multi-physical field coupling.

[0119] For example: by identifying latent harmonics, obtaining vibration signals, detecting conditions such as unbalanced motor rotor operation or abnormal gear meshing, and generating low-frequency harmonic characteristics, and transient impact detection, it is used to diagnose high-frequency transient faults such as bearing cracks and winding short circuits. Through composite fault diagnosis, combined with multimodal signals such as vibration, temperature, and magnetic field, single faults can be distinguished from composite faults.

[0120] Furthermore, through improved wavelet packet decomposition, high-precision decomposition is achieved, as well as the capture of implicit harmonics and transient impacts. Noise drying is performed through soft threshold processing to improve the reliability of feature extraction. Through joint analysis of multimodal signals, signal diagnosis blind spots are avoided and the accuracy of diagnostic evaluation is improved.

[0121] According to steps S4 and S5, the federated learning framework runs the federated learning diagnostic method, which adapts the pre-trained model to different vehicle models and working conditions, reduces dependence on labeled data, and encrypts multimodal data through edge computing nodes.

[0122] Furthermore, the edge computing node generates time-frequency domain feature vectors through improved wavelet packet decomposition and uploads new fault analysis features using incremental learning methods.

[0123] Based on the above, the global diagnostic model includes multiple signals or motors with the same signal, or different signal types of the same motor.

[0124] Furthermore, under dynamic working conditions, the same learning or calculation method can be learned through online learning or transfer learning methods, which can be applied to motors with the same data parameters. The diagnostic model can be used to determine whether the parameters have abnormal fluctuations. If not, the parameter patterns between the corresponding models and types are recorded to realize data application under multiple working conditions and application of system experience, and further improve the rapid output of the results of motor health status assessment based on big data, thereby improving the evaluation efficiency of detection and evaluation.

[0125] Furthermore, local features are extracted through edge computing nodes, and a federated learning framework is used to aggregate multi-device data to train a global diagnostic model. The model parameters are then dynamically optimized through online learning to adapt to changes in working conditions.

[0126] According to another aspect of an embodiment of the present invention, an electronic device is provided. The electronic device includes a memory and a processor; the memory is used to store programs; and the processor executes the programs to implement any one of the aforementioned methods.

[0127] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The storage medium stores a computer program. When the computer program is executed by a processor, any one of the aforementioned methods is implemented.

[0128] According to yet another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, which implements any of the aforementioned methods when executed by a processor.

[0129] An instrument for evaluating the health status of a new energy vehicle motor, comprising:

[0130] Multimodal sensing module: The multimodal sensing module includes at least one group of distributed sensors, including but not limited to: vibration sensor, temperature sensor, Hall sensor, photoelectric encoding sensor, motor state evaluator;

[0131] Dynamic synchronization control module: The dynamic synchronization control module includes an event-triggered controller and a dynamic interpolation unit. The event-triggered controller dynamically adjusts the sampling strategy based on the multimodal motor signal mutation threshold; the dynamic interpolation unit optimizes the synchronization accuracy of the multimodal motor signal through cubic spline interpolation;

[0132] Multi-domain analysis module: This module uses a graph neural network processor to achieve cross-domain feature fusion between multimodal motor signals and an improved wavelet packet decomposition accelerator to support multi-resolution frequency domain analysis and implicit harmonic detection.

[0133] Federated Learning Diagnostic Module: This module is equipped with at least one edge computing node, which is used to extract local features and upload them to the cloud in encrypted form, enabling data synchronization and end-to-end computing collaboration. It also features a federated learning engine, which dynamically updates the global diagnostic model through online learning to adapt to changing operating conditions.

[0134] Motor condition evaluator: The motor condition evaluator is used to collect data from the multimodal sensor module and is equipped with a data display unit for displaying and controlling the motor health status.

[0135] Among them, the data display unit includes time-amplitude, time-frequency, time-order, speed-frequency, speed-order, order-amplitude, and the display of acquisition results and processing results.

[0136] In addition, please refer to Figure 2 and Figure 3 The hardware of the instrument consists of two parts: an acquisition board and a display board.

[0137] Among them, the acquisition board is responsible for signal synchronization and processing through the acquisition MCU, receives CAN bus messages through the communication controller FDCAN, decodes the CAN messages to obtain the motor speed signal, current signal and temperature signal, obtains the motor vibration signal from the voltage acquisition channel, and calculates the real-time frequency corresponding to the speed and vibration, thereby enabling correlation analysis in various algorithms. The analog-to-digital converter ADC converts the analog signal of the vibration sensor into a digital signal, the USB interface and the flexible circuit board FPC realize data transmission, and the connection with the hardware device is realized through USB, which is used to implement operations such as adjusting the instrument software configuration. The TFSD / ROM is used to store configuration parameters and firmware programs.

[0138] The display board runs data processing, data analysis, and data evaluation systems through the display processor MCU. The uninterruptible power supply UPS ensures data integrity during power outages. The display screen displays analysis results such as time domain waveform, spectrum, order spectrum in real time, and enables data interaction with the instrument through the display screen. The TFSD / ROM is used to store diagnostic algorithms and historical data.

[0139] Furthermore, data collection is achieved through the acquisition MCU, and the collected data includes but is not limited to speed information, amplitude information, frequency information, etc.

[0140] For example: when a CAN message is detected, the speed information is immediately encoded. If the speed is not updated, the interpolation algorithm is run to maintain a 50kHz synchronization rate with an error of less than or equal to 40μs;

[0141] 50 vibration + speed data packets are transmitted to the processing board via USB at a 1ms period.

[0142] Furthermore, the operation graph neural network fuses vibration and speed data, extracts cross-domain fault features, and applies improved wavelet packet decomposition to identify latent harmonics and transient shocks.

[0143] One-click diagnosis integrates a federated learning model and aggregates multi-device data through edge computing nodes to train a global diagnostic model.

[0144] It should be noted that the above modules can be implemented through software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0145] As an optional embodiment, the time-amplitude, time-frequency, time-order, speed-frequency, speed-order, and order-amplitude displayed by the data display unit are all displayed through the corresponding algorithms through the acquisition results and processing results. The algorithms include: time-amplitude (amplitude); time-frequency; time-order; speed-frequency; speed-order; order-amplitude.

[0146] Once again, the acquisition board can synchronously collect CAN messages and vibration signals output by the automobile motor, decode the CAN messages to obtain the motor temperature signal, rotation angle and speed signal, motor current signal, etc., obtain the motor vibration signal from the voltage acquisition channel, and calculate the real-time frequency corresponding to the speed and vibration, thereby performing correlation analysis in various algorithms.

[0147] Acquisition board data:

[0148] 1. The default ADC sampling rate is 50KHz; the number of samples per second per channel is 50,000; the value obtained is the motor vibration signal value.

[0149] 2. The data obtained from the CAN channel message includes but is not limited to the photoelectric coding signal, i.e. the motor rotation angle and rotation speed signal, temperature signal, current signal, etc.

[0150] The relationship between each algorithm and data: In the current algorithm, (time-amplitude / time-frequency) requires the use of vibration signals for calculation, and (time-order / speed-frequency / speed-order / order-amplitude) requires the synchronous calculation of vibration signals and motor speed.

[0151] 1. Time-amplitude analysis

[0152] Time-amplitude analysis monitors the motor's health by observing how the amplitude of the motor's vibration signal changes over time. Common time-domain analysis methods, such as root mean square (RMS), peak value, and peak-to-peak value, can be used.

[0153] 1.1 Algorithm coupling method:

[0154] Input data: vibration signal (time domain data);

[0155] Analysis method: Calculate the amplitude characteristics of the vibration signal, such as root mean square (RMS), peak value, and peak-to-peak value, to characterize the health of the motor.

[0156] 1.2 Statistical characteristic trend analysis: including curves:

[0157] Red: Peak

[0158] Blue: Root Mean Square (RMS)

[0159] Green: Peak-to-Peak

[0160] Analysis rules:

[0161] RMS keeps rising Wear accumulation Peak surges Transient shock event Peak-to-peak fluctuation Assembly looseness

[0162] 1.3 Data Verification:

[0163] Health status RMS range (m / s2) Peak typical value Normal 0.2-0.5 Less than 2.0 Wear 0.5-1.2 2.0-3.5 Failure Greater than 1.5 Greater than 4.0

[0164] 1.4 Specific applications:

[0165] The following correlation analysis can be performed based on the vibration signal points obtained by the acquisition board at the current 50kHz frequency.

[0166] 1. Preliminary diagnosis:

[0167] Use a 0.1 second window (5000 points) to calculate RMS and peak values, balancing real-time performance and stability.

[0168] 2. Transient shock analysis:

[0169] Use a 0.01 second window (500 points) to detect short-term peaks and quickly locate abnormal events.

[0170] 3. Long-term trend monitoring:

[0171] The daily average RMS is calculated using a 10-second window (500,000 points) to observe the slow degradation process.

[0172] 1.5 Algorithm Principle:

[0173] (1) Root Mean Square (RMS)

[0174] Definition: The root mean square value of the vibration signal amplitude, reflecting the average energy of the signal.

[0175] formula: ;

[0176] in is a discrete sampling signal, and N is the number of sampling points.

[0177] (2) Peak

[0178] Definition: The maximum absolute value of a signal during the analysis period.

[0179] Formula: Peak=max .

[0180] (3) Peak-to-Peak

[0181] Definition: The difference between the maximum and minimum values ​​of a signal during the analysis period.

[0182] Formula: Peak-to-Peak = max ;

[0183] 2. Time-Frequency Analysis

[0184] Time-frequency analysis is mainly used to analyze non-stationary signals, such as instantaneous frequency changes during motor operation. Short-time Fourier transform (STFT) or wavelet transform (WT) can be used.

[0185] 2.1 Algorithm coupling method:

[0186] Input data: vibration signal (time domain data);

[0187] Analysis method: Apply short-time Fourier transform (STFT) or wavelet transform (WT) to analyze the characteristics of vibration signals in the time-frequency domain.

[0188] Short-time Fourier transform (STFT):

[0189] STFT is a method that divides a signal into multiple short time periods and applies Fourier transform to each time period. This method can obtain the frequency components of the signal at different time points.

[0190] Advantages: simple calculation and easy to implement.

[0191] Disadvantages: There is a trade-off between time resolution and frequency resolution, and it is impossible to obtain high time resolution and high frequency resolution at the same time.

[0192] Wavelet transform: Uses scalable and translatable wavelet functions to analyze signals, which can provide frequency information of signals at different time scales.

[0193] Advantages: Suitable for analyzing non-stationary signals and can provide better time and frequency resolution.

[0194] Disadvantages: The computational complexity is high, and the choice of wavelet basis has a great influence on the results.

[0195] 2.2 Short-time Fourier transform (STFT) time-frequency diagram

[0196] Horizontal axis: time (seconds)

[0197] Vertical axis: frequency (Hz)

[0198] Color Map: Energy Intensity (dB)

[0199] Typical abnormal characteristics:

[0200] High-frequency energy spread Bearing damage Harmonic component mutation Gearbox failure Energy band breakage Rotor eccentricity

[0201] 2.3 Wavelet transform time-frequency distribution

[0202] Horizontal axis: time (seconds)

[0203] Vertical axis: scale (corresponding frequency)

[0204] Color mapping: wavelet coefficient magnitude

[0205] Advantages and features:

[0206] High-frequency region transient spike Brush spark Low-frequency region sustained fluctuation Mechanical resonance Multi-scale correlation pattern Compound failure

[0207] Typical failure modes:

[0208] Failure type STFT feature Wavelet feature Bearing spalling High-frequency energy band spread High-frequency region intermittent spike Rotor imbalance Fundamental frequency energy periodic fluctuation Low-frequency sustained high energy Gear tooth breakage Sideband structure mutation Multi-scale impact correlation

[0209] 2.4 Specific applications:

[0210] Combined with the vibration signal points obtained at the current 50kHz frequency of the acquisition board, the following correlation analysis can be performed

[0211] (1) STFT high-frequency transient shock, calculated using 100-500 vibration signal data points;

[0212] (2) STFT low-frequency ramping component, calculated using 5000-50000 vibration signal data points;

[0213] (3) WT multi-scale transient and periodic, calculated using 500-10,000 vibration signal data points;

[0214] 2.5 Algorithm Principle:

[0215] (1) Short-time Fourier transform: By dividing the signal into multiple short time periods and performing Fourier transform on each time period, the signal characteristics can be located simultaneously in time and frequency.

[0216] Mathematical expression: STFT(t,f)

[0217] Where: x(τ) is the original signal, w(τ−t) is the sliding window function (such as Hamming window, Gaussian window) centered at time point t, It is a complex exponential function used to calculate the frequency component. t is the current time position, f is the frequency of the current analysis, and the integral result is the STFT value at time t and frequency f.

[0218] (2) Wavelet transform (WT), which uses scalable and translatable wavelet basis functions and adaptively adjusts the window length, using short windows (high time resolution) for high frequencies and long windows (high frequency resolution) for low frequencies.

[0219] Mathematical expression (continuous wavelet transform, CWT): CWT(a,b)= ;

[0220] Where: a is the scale parameter, which is inversely proportional to the frequency, b is the translation parameter, which represents the time position, f(t) is the input signal, ψ(t) is the wavelet mother function (such as Morlet, Daubechies wavelet), represents the complex conjugate of the wavelet mother function, is the normalization factor, keeping the energy unchanged.

[0221] 3. Time-Order Analysis

[0222] Time-order analysis uses a motor's rotational signal (typically related to speed) to analyze the motor's health. By extracting the order components of the rotor's rotational frequency, failure modes can be identified.

[0223] 3.1 Algorithm coupling method:

[0224] Input data: vibration signal, rotation speed;

[0225] Analysis method: Convert the vibration signal into order frequency through the rotation speed, and analyze the change of order over time.

[0226] 3.2 Order Spectrum

[0227] Horizontal axis: Order (integer multiples of speed frequency)

[0228] Vertical axis: Energy amplitude (dB)

[0229] Typical fault characteristics:

[0230] Order anomaly Rotor imbalance Order prominence Coupling misalignment Tooth number multiple order Gear wear

[0231] 3.3 Order Tracking

[0232] Horizontal axis: time (seconds)

[0233] Vertical axis: order amplitude

[0234] Curve Type:

[0235] Red: Target order (such as bearing characteristic order)

[0236] Dashed line: Speed ​​change curve Diagnostic rules:

[0237] Order amplitude rising with rotating speed Resonance risk Order amplitude mutation Transient shock event

[0238] Typical fault characteristics table

[0239] Failure type Characteristic order Time-domain performance Rotor imbalance 1st order Rotating speed synchronous vibration Gear tooth breakage Tooth number x rotating frequency order Modulation sideband Bearing outer ring damage (n·ball) / 2×(1-d / D·cosα) High-frequency impact sequence Motor eccentricity 2nd order Pole passing frequency modulation

[0240] Diagnose the following motor failure modes:

[0241] Demagnetization of permanent magnets: the energy of a specific electromagnetic force order increases

[0242] Bearing wear: Harmonic clusters appear at characteristic orders

[0243] Shaft crack: 1 / 2 order component abnormality

[0244] 3.4 Specific application: Combining the motor vibration signal and motor speed signal of the acquisition board, the following correlation analysis can be performed

[0245] 1. Motor fault diagnosis:

[0246] By converting the motor's vibration signal into order frequencies and observing how these frequencies change over time, we can identify vibration patterns caused by speed fluctuations. Different fault types manifest as changes in specific order frequency components, helping to determine the motor's health. For example, motor imbalance can produce specific order frequency components of its speed, while bearing failures can cause sudden changes in frequency components.

[0247] 2. Speed ​​comparison analysis:

[0248] Motor speed fluctuations can affect the order frequency distribution of the vibration signal during operation. By analyzing the synchronization between the vibration signal and speed changes, the motor's operational stability can be monitored in real time.

[0249] The order frequency during normal operation can be compared with the order frequency during abnormal operation. In this way, small changes in motor operation can be captured and potential faults can be warned in advance.

[0250] 3.5 Algorithm Principle:

[0251] Order: Order = vibration frequency / rotation frequency (fvibration / frotor).

[0252] 1st order: The rotor vibrates once per revolution (fundamental frequency), reflecting mass imbalance.

[0253] 2nd order: The rotor vibrates twice for every 1 revolution (may be caused by misalignment or bending).

[0254] Order (such as 0.5): caused by looseness or nonlinear vibration.

[0255] Order tracking technology

[0256] Angular domain resampling principle: uniformly sample the signal in time Convert to angle uniform sampling signal , eliminating the impact of speed changes on the spectrum.

[0257] Interpolation method: (Linear or cubic spline interpolation is commonly used)

[0258] 4. Speed-frequency analysis

[0259] Speed-frequency analysis typically involves correlating the speed of a motor with the frequency content of a signal. This analysis helps us identify vibration patterns caused by speed variations.

[0260] 4.1 Algorithm coupling method:

[0261] Input data: vibration signal, rotation speed;

[0262] Analysis method: Frequency analysis is performed based on the relationship between the rotational speed and the frequency components of the vibration signal.

[0263] The motor's speed and frequency are combined to identify vibration characteristics using spectrum analysis.

[0264] 4.2 Speed-Frequency Waterfall Plot

[0265] Horizontal axis: frequency (Hz)

[0266] Vertical axis: speed (RPM)

[0267] Color Mapping: Vibration Energy (dB)

[0268] Key diagnostic features:

[0269] Diagonal energy band Rotating speed related resonance Vertical line Fixed frequency interference (such as electromagnetic noise) Energy cloud cluster Wideband random vibration

[0270] 4.3 Envelope Analysis Spectrum

[0271] Horizontal axis: modulation frequency (Hz)

[0272] Vertical axis: Envelope energy

[0273] Typical applications:

[0274] Bearing fault characteristic frequency identification

[0275] Gear local damage detection

[0276] Early crack diagnosis

[0277] Typical fault characteristics correspondence table

[0278] Failure type Rotating speed-frequency feature Diagnosis rule Rotor bar breakage 2×pole passing frequency spike at specific rotating speed Amplitude rising with rotating speed square Bearing outer ring damage Rotating speed modulation sideband appearing in high-frequency band Sideband interval = failure characteristic frequency Air gap eccentricity Rotating speed related electromagnetic force harmonic Odd multiple pole passing frequency component Coupling misalignment 2×rotating frequency component linearly rising with rotating speed Phase stable, axial vibration dominant

[0279] Identify the following abnormal operating conditions:

[0280] Critical speed: energy increases sharply in a specific speed range

[0281] Resonance: Energy accumulation at the same frequency at multiple speed points

[0282] Loosening: Broadband random vibration energy boost

[0283] 4.5 Specific applications:

[0284] Combining the motor vibration signal and motor speed signal of the acquisition board, the following correlation analysis can be performed

[0285] (1) The frequency components of the vibration signal are obtained through FFT (Fast Fourier Transform) analysis; FFT is used to convert the time domain signal to the frequency domain, so that the frequency components can be extracted.

[0286] (2) The motor speed data is coupled with the frequency components in (1); the corresponding speed-related frequency components are calculated according to the motor speed; which frequency components are caused by the speed can be identified, thereby revealing potential resonance phenomena or other faults.

[0287] 4.6 Algorithm Principle:

[0288] (1) Time-frequency conversion and speed synchronization

[0289] Core formula: Short-time Fourier transform (STFT) vibration signal time-frequency analysis formula:

[0290] Where x(τ) is the vibration signal and w(τ) is the window function (such as Hanning window). The frequency axis is converted to the order axis using the speed signal RPM(t):

[0291] (The order indicates the relationship between the vibration frequency and the fundamental frequency of the speed)

[0292] (2) Speed-frequency matrix construction

[0293] Synchronous averaging formula:

[0294] Divide the rotation speed into constant intervals (e.g., every 10 RPM) and perform spectrum averaging on the vibration signal in each interval:

[0295] ;

[0296] Generate a 3D matrix , used to identify critical speed resonance peaks;

[0297] 5. Speed-order analysis

[0298] Speed-order analysis is based on the order relationship between the motor's speed and its frequency, and is mainly used to diagnose the motor's operating status at different speeds.

[0299] 5.1 Algorithm coupling method:

[0300] Input data: vibration signal, rotation speed;

[0301] Analysis method: Analyze the change in order through the change in speed to evaluate the status of the motor.

[0302] Based on the relationship between motor speed and order, mechanical faults can be effectively identified through order analysis.

[0303] 5.2 Speed-order matrix diagram

[0304] Horizontal axis: Order (integer multiples of speed frequency)

[0305] Vertical axis: speed range (RPM)

[0306] Color Mapping: Vibration Energy (dB)

[0307] Diagnostic features:

[0308] Diagonal energy band Rotating speed linearly related failure (such as imbalance) Vertical energy band Fixed order problem (such as gear meshing order) Energy hot spot Resonance risk area

[0309] 5.3 Campbell-Order Composite Diagram

[0310] Horizontal axis: speed (RPM)

[0311] Vertical axis: order

[0312] Color Map: Energy Intensity (dB)

[0313] Overlay Display:

[0314] Red contour line Structure resonance frequency curve Dotted line mark Design critical rotating speed

[0315] Typical fault characteristics table

[0316] Failure type Rotating speed-order feature Diagnosis rule Bearing ball damage Specific order (such as BPFO) high energy at full rotating speed Order energy independent of rotating speed Gear wear Tooth number x rotating frequency order energy linearly rising with rotating speed Order slope = tooth number / 60 Shaft bending 1st order energy surging at critical rotating speed area Phase difference 90° Electromagnetic imbalance Multiple order energy anomaly Specific electromagnetic order (such as 48th order) prominence

[0317] Diagnose the following typical faults of the motor:

[0318] Permanent magnet falls off: electromagnetic force order energy changes suddenly

[0319] Cooling failure: high temperature causes order energy drift

[0320] Assembly error: Abnormal phase of specific order

[0321] 5.4 Specific applications:

[0322] The following correlation analysis can be performed by combining the motor vibration signal and the motor speed signal of the acquisition board:

[0323] Order frequency: Calculates the order of each frequency component based on the rotational speed. Order is a multiple of the rotational speed frequency. Determine the type of motor fault by comparing the vibration characteristics of different orders with empirical data on fault modes.

[0324] 5.5 Algorithm Principle:

[0325] (1) Order calculation formula: ;

[0326] in: : frequency of the nth order (Hz), n: order, : The motor speed frequency (revolutions per minute, RPM) is converted to frequency (Hz) for analysis.

[0327] (2) Speed ​​and frequency calculation:

[0328] Motor speed

[0329] Motor speed Convert to frequency (Hz), the following formula can be used: ;

[0330] (3) Order analysis of vibration signals: The vibration signal v(t) is converted from the time domain to the frequency domain, usually by using the fast Fourier transform (FFT) to obtain the spectrum of the vibration signal: ;

[0331] in, is the frequency spectrum of the vibration signal, is the frequency; then by comparing it with the order frequency, the vibration peaks corresponding to different orders are extracted.

[0332] 6. Order-Amplitude Analysis

[0333] Order-amplitude analysis, which combines the relationship between order and vibration amplitude, is often used to detect failure characteristics during rotor or motor operation. By analyzing the order, the vibration amplitude of each order of the motor can be evaluated, thereby performing health diagnosis.

[0334] 6.1 Algorithm coupling method:

[0335] Input data: vibration signal, rotation speed;

[0336] Analysis method: Motor health analysis is performed through the relationship between the order and the amplitude of the vibration signal.

[0337] Perform Fourier transform or other spectrum analysis methods on the signal to extract the amplitude of each order to help determine the operating status of the motor.

[0338] 6.2 Order Amplitude Distribution Histogram

[0339] Horizontal Axis: Order (integer multiple of rotational speed frequency)

[0340] Vertical Axis: Amplitude (g)

[0341] Key Diagnostic Features:

[0342] 1st order prominence Rotor dynamic imbalance Tooth number order high amplitude Gear meshing anomaly Fractional order existing Bearing damage feature

[0343] 6.3 Order Amplitude Trend Curve

[0344] Horizontal Axis: Time (seconds)

[0345] Vertical Axis: Order Amplitude (dB)

[0346] Curve Type:

[0347] Red: Target Diagnostic Order (such as bearing characteristic order)

[0348] Gray: Background Noise Level

[0349] Analysis Rules:

[0350] Continuous over threshold Progressive failure Sudden spike Transient shock event

[0351] Typical Fault Feature Table

[0352] Failure type Order-amplitude feature Diagnosis threshold Rotor imbalance 1st order amplitude greater than 2 mm / s ISO 10816-3 Class I Bearing inner ring damage Characteristic order + 3rd harmonic greater than background noise 10 dB Envelope spectrum kurtosis greater than 3.5 Gear eccentricity 1×tooth number order amplitude surging 50% Phase difference greater than 30° Electromagnetic interference Multiple order amplitude anomaly Current harmonic correlation greater than 0.8

[0353] Identify Abnormal Working Conditions:

[0354] Mechanical Looseness: Random fluctuations in multiple order amplitudes

[0355] Resonance Precursor: Non-linear growth of specific order amplitude with speed

[0356] Early Wear: Appearance of sub-harmonics of characteristic order

[0357] 6.4 Specific Applications:

[0358] Combining motor vibration signals and motor speed signals collected by the acquisition board, the following correlation analysis can be performed

[0359] (1) Order Analysis:

[0360] Perform FFT conversion on the vibration signal to convert the time domain signal to the frequency domain signal.

[0361] According to the motor speed, calculate the corresponding frequency of each order.

[0362] (2) Amplitude Comparison: Extract the amplitude of each order corresponding frequency, identify abnormal fluctuations by comparing the amplitude changes of each order. The order with larger amplitude is usually associated with some mechanical faults (such as imbalance, looseness, eccentricity, etc.).

[0363] 6.5 Algorithm principle:

[0364] (1) Relationship between order and rotating speed: ;

[0365] Wherein, forder is the frequency corresponding to the order, RPM is the rotating speed of the motor, and n is the order;

[0366] (2) Amplitude extraction formula: ;

[0367] Wherein, Aorder is the amplitude of the nth order, and X(forder) is the amplitude of the frequency corresponding to the order in the frequency spectrum;

[0368] The above is only the preferred embodiment of the present application, it should be noted that for those skilled in the art, without departing from the principles of the present application, can make a number of improvements and refinements, these improvements and refinements should also be considered as the protection scope of the present application. The structure, device and operation method not specifically described and explained in the present application, such as no special description and limitation, are implemented according to the conventional means in the art.

Claims

1. A method for evaluating the health status of a new energy vehicle motor, characterized in that: The steps include: Acquire and synchronize multi-modal motor signals, and generate trigger events based on motor signal changes; Construct a four-dimensional tensor dataset, fuse motor signal data through a graph neural network, and extract cross-domain fault features based on the motor signal data; Apply improved wavelet packet decomposition to perform multi-resolution analysis on cross-domain fault signals to identify latent harmonics and transient impacts; Extract cross-domain fault features through edge computing nodes, use a federated learning framework to aggregate multi-device data to train a global diagnostic model, and implement fault classification methods under dynamic working conditions; Evaluate the motor health status based on the global diagnostic model and fault types under dynamic conditions.

2. A method for evaluating the health status of a new energy vehicle motor according to claim 1, characterized in that: The triggering events generated by the motor signal change include: By establishing a multimodal data fluctuation value model of the motor's multimodal operating state, the abnormal change value of the multimodal data fluctuation value model is compared with the change of the multimodal data fluctuation value signal under the actual multimodal operating state of the motor, and the degree of fit between the two signals is compared to determine whether an event is triggered.

3. The method for evaluating the health status of a new energy vehicle motor according to claim 1, characterized in that: The four-dimensional tensor data set includes a four-dimensional data set constructed from any one of the multi-modal motor signals, and the four-dimensional data set includes X, Y, and Z axial data signals in the spatial dimension and a time dimension data signal; A multi-physical scene association model with cross-domain fusion is extracted and constructed based on the collected multi-modal motor signal data.

4. The method for evaluating the health status of a new energy vehicle motor according to claim 1, characterized in that: The graph neural network fuses motor signal data, uses the nodes of the graph neural network to represent multimodal data, and constructs an evaluation system for multimodal data through the graph neural network.

5. The method for evaluating the health status of a new energy vehicle motor according to claim 1, characterized in that: The improved wavelet packet decomposition comprises: Adaptive basis function selection: Dynamically optimize the wavelet basis according to the signal band energy distribution; Noise suppression: Combined with soft threshold processing to eliminate high-frequency noise and retain implicit harmonic features.

6. The method for evaluating the health status of a new energy vehicle motor according to claim 1, characterized in that: The federated learning framework runs a federated learning diagnostic method, which adapts the pre-trained model to different vehicle models and working conditions, reduces dependence on labeled data, and encrypts multimodal data through edge computing nodes.

7. The method for evaluating the health status of a new energy vehicle motor according to claim 1, characterized in that: The edge computing node generates time-frequency domain feature vectors through improved wavelet packet decomposition and uploads new fault analysis features using an incremental learning method.

8. An instrument for evaluating the health status of a new energy vehicle motor, using the method according to any one of claims 1 to 7, characterized in that: The instrument comprises: Multimodal sensing module: The multimodal sensing module includes at least one group of distributed sensors, wherein the distributed sensors include vibration sensors, temperature sensors, Hall sensors, photoelectric encoders or motor state evaluators; Dynamic synchronization control module: The dynamic synchronization control module includes an event-triggered controller and a dynamic interpolation unit. The event-triggered controller dynamically adjusts the sampling strategy based on the multimodal motor signal mutation threshold; the dynamic interpolation unit optimizes the synchronization accuracy of the multimodal motor signal through cubic spline interpolation; Multi-domain analysis module: This module uses a graph neural network processor to achieve cross-domain feature fusion between multimodal motor signals and an improved wavelet packet decomposition accelerator to support multi-resolution frequency domain analysis and implicit harmonic detection; Federated Learning Diagnostic Module: This module is equipped with at least one edge computing node, which is used to extract local features and upload them to the cloud in encrypted form, achieving data synchronization and end-to-end computing collaboration. It also has a federated learning engine, which dynamically updates the global diagnostic model through online learning to adapt to changes in working conditions. Motor status evaluator: The motor status evaluator is used to realize data acquisition of the multimodal sensor module, and is provided with a data display unit for realizing the display and control of the motor health status evaluation.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images