An electric vehicle prime mover resonance and torque ripple optimization method

Abstract

This application provides a method for optimizing resonance and torque pulsation in an electric vehicle prime mover, comprising: S2, inputting real-time dynamic fluctuation generation indicators into a preset finite element analysis model to simulate the resonance frequency drift trend and calculate potential torque pulsation anomalies; S4, using acquired information on the uneven distribution of the electromagnetic field, combined with real-time sensing state data to generate a vicious cycle formation risk assessment score; S6, updating the torque control model by stabilizing the resonance frequency parameters to optimize the torque output curve to suppress pulsation anomalies; S7, after obtaining the optimized torque output curve, inputting a feedback loop mechanism to verify the material physical property response and determine whether the source of dynamic fluctuations has been eliminated; S8, if it is determined that the source of dynamic fluctuations has been eliminated, recording the adjustment log and updating the temperature load change database for subsequent simulation use. This invention achieves a significant improvement in the stability and efficiency of prime mover operation.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle technology, and in particular to a method for optimizing the resonance and torque pulsation of the prime mover in an electric vehicle. Background Technology

[0002] As the core of new energy vehicles, the performance of the drive system directly affects the vehicle's power, efficiency, and user experience. The prime mover, as a key component of the drive system, is crucial to the overall vehicle performance due to its operational stability.

[0003] However, existing technologies have significant shortcomings in addressing the dynamic problems of prime mover operation. Many solutions focus only on performance optimization under a single operating condition, neglecting the dynamic effects of multi-physics coupling in complex operating environments. This leads to instability in the system at different speeds or loads, affecting vehicle ride comfort and durability. During prime mover operation, the physical characteristics of components such as the rotor, stator, and bearings fluctuate dynamically due to changes in temperature and load. These fluctuations directly affect the system's resonance behavior. The core challenge lies in accurately understanding the mechanism by which these physical characteristic changes affect the resonance frequency. Dynamic changes in the material's elastic modulus and damping coefficient can cause resonance frequency drift, leading to abnormal torque pulsation. For example, at high speeds, the prime mover may experience abnormal vibrations due to the coupling between the resonance frequency and the operating frequency, increasing bearing wear and even causing noise problems. More complexly, this resonance frequency drift can be exacerbated by uneven electromagnetic field distribution, resulting in unstable torque output and affecting the smoothness of vehicle acceleration. These two factors are interconnected: changes in material physical properties cause resonance frequency drift, and frequency drift further exacerbates the dynamic imbalance between the electromagnetic field and torque, creating a vicious cycle.

[0004] Therefore, how to dynamically adjust the physical response characteristics of key components by sensing the operating status of the prime mover in real time, so as to stabilize the resonant frequency and optimize the torque output, has become the key issue of this study. Summary of the Invention

[0005] This invention provides a method for optimizing resonance and torque pulsation in the prime mover of an electric vehicle, the method comprising:

[0006] S1. Data on temperature load changes and material physical property parameters inside the prime mover are collected by sensors, and noise interference is filtered out to obtain real-time dynamic fluctuation generation indicators. S2. Based on the real-time dynamic fluctuation generation indicators, a preset finite element analysis model is input to simulate the resonant frequency drift trend and calculate potential torque pulsation anomalies. S3. If the simulated torque pulsation anomaly value exceeds a preset pulsation anomaly threshold, the vibration-increased bearing wear prediction module is activated to determine the electromagnetic field non-uniform distribution area. S4. Using the acquired electromagnetic field non-uniform distribution area information, combined with real-time sensing state data, a vicious cycle formation risk assessment score is generated. S5. When the vicious cycle formation risk assessment score is higher than the vicious cycle threshold, the damping coefficient of the prime mover components is adjusted to obtain stable resonant frequency parameters. S6. The torque control model is updated using the stable resonant frequency parameters to optimize the torque output curve to suppress pulsation anomalies. S7. After obtaining the optimized torque output curve, a feedback loop mechanism is input to verify the material physical property response and determine whether the source of dynamic fluctuation generation has been eliminated. S8. If the source of dynamic fluctuation generation is determined to have been eliminated, the adjustment log is recorded and the temperature load change database is updated for subsequent simulations.

[0007] Optionally, step S1 involves collecting data on internal temperature load changes and material physical property parameters of the prime mover using sensors, filtering out noise interference to obtain real-time dynamic fluctuation generation indicators, including:

[0008] Step S11: Collect temperature and load change data inside the prime mover through sensors, and use analog-to-digital conversion technology to obtain digital temperature and load signals;

[0009] Step S12: Filter out high-frequency noise from the digital temperature signal and load signal to obtain a denoised temperature signal and a denoised load signal.

[0010] Step S13: Extract time series features from the denoised temperature signal and the denoised load signal to obtain frequency domain feature data;

[0011] Step S14: If there are abnormal peaks in the frequency domain feature data, the abnormal fluctuations are judged by the preset fluctuation threshold to obtain the abnormal fluctuation identifier.

[0012] Step S15: Based on the abnormal fluctuation identifier and the material physical property parameters, determine the correlation model between the fluctuation and the material properties;

[0013] Step S16: Extract the dynamic fluctuation trend from the correlation model and generate a real-time fluctuation indicator;

[0014] Step S17: Update the preset fluctuation threshold using real-time fluctuation indicators.

[0015] Optionally, in step S13, time series features are extracted from the denoised temperature signal and the denoised load signal to obtain frequency domain feature data. The extraction of time series features from the denoised temperature signal and the denoised load signal includes calculating the mean and standard deviation of the denoised temperature signal and the denoised load signal.

[0016] Optionally, step S4, which uses the acquired information on the uneven distribution of the electromagnetic field and combines it with real-time sensing state data to generate a risk assessment score for a vicious cycle, includes:

[0017] Step S41: Obtain electromagnetic field distribution data and real-time sensing data, and generate the first dataset by weighted averaging and fusion.

[0018] Step S42: Extract the electromagnetic field distribution portion from the first dataset, identify non-uniform regions, and obtain regional feature data;

[0019] Step S43: Extract the real-time perception part from the first dataset, extract key state parameters, and generate a state feature set;

[0020] Step S44: If the key parameters in the state feature set exceed the preset state threshold, calculate the initial risk score.

[0021] Step S45: Generate a comprehensive risk assessment vector by combining the regional feature data with the initial risk score;

[0022] Step S46: Generate the final vicious cycle risk assessment score based on the comprehensive risk assessment vector;

[0023] Step S47: By comparing thresholds, the final vicious cycle risk assessment score is converted into a standardized risk level output.

[0024] Optionally, in step S44, if the key parameters in the state feature set exceed a preset state threshold, an initial risk score is calculated, including:

[0025] Calculate the initial risk score using the following formula:

[0026] Y = aX + b

[0027] Where Y is the initial risk score, X is the key parameter value, a is the slope, and b is the intercept.

[0028] Optionally, step S6, updating the torque control model by stabilizing the resonant frequency parameter and optimizing the torque output curve to suppress abnormal pulsation, includes:

[0029] Step S61: Obtain real-time torque output data and resonant frequency data from the sensor, and use fast Fourier transform on the torque output data to calculate the frequency response and obtain torque fluctuation characteristics.

[0030] Step S62: If the torque fluctuation characteristics exceed the preset torque fluctuation threshold, then apply K-means clustering to group abnormal peak values ​​in the frequency response data to determine the frequency range of the pulsation anomaly.

[0031] Step S63: Calculate the loss function based on the frequency range of the pulsation anomaly, update the resonant frequency parameters, and obtain the optimized parameter set;

[0032] Step S64: Adjust the torque control model using the optimized parameter set to generate a new torque output curve;

[0033] Step S65: Extract the smoothness index from the new torque output curve and determine whether it meets the preset smoothness requirements.

[0034] Step S66: If the smoothness index does not meet the requirements, the parameter set is iteratively updated to obtain the updated torque control model.

[0035] Step S67: Based on the updated torque control model, the output is adjusted using a PID controller to generate the final torque output curve and suppress abnormal pulsation.

[0036] Optionally, in step S61, real-time torque output data and resonant frequency data are acquired from the sensor, and the frequency response is calculated by using a fast Fourier transform on the torque output data to obtain torque fluctuation characteristics, wherein the torque fluctuation characteristics are the peak amplitude in the frequency response.

[0037] Optionally, in step S63, a loss function is calculated based on the frequency range of the pulsation anomaly, the resonant frequency parameters are updated, and an optimized parameter set is obtained. The loss function is the mean square error between the abnormal frequency and the resonant frequency parameters.

[0038] Optionally, step S8, if it is determined that the source of dynamic fluctuations has been eliminated, involves recording the adjustment log and updating the temperature load change database for subsequent simulation use, including:

[0039] Step S81: If a fluctuation signal is detected, determine the first dynamic fluctuation source from the fluctuation signal;

[0040] Step S82: By analyzing the first dynamic fluctuation source, obtain the first adjustment log and store it in the preset database;

[0041] Step S83: Extract the first adjustment log from the preset database and calculate the temperature load change rate to update the first temperature load change data;

[0042] Step S84: Based on the first temperature load change data, predict the first load change trend from the first temperature load change data;

[0043] Step S85: Calculate simulation analysis parameters based on the first load change trend;

[0044] Step S86: Using simulation analysis parameters, obtain the first dynamic fluctuation adjustment strategy from the simulation analysis parameters;

[0045] Step S87: Update the fluctuation monitoring configuration according to the first dynamic fluctuation adjustment strategy to improve the source detection accuracy.

[0046] Optionally, step S85, calculating simulation analysis parameters based on the first load change trend, includes: calculating the simulation analysis parameters using the following formula:

[0047] P=G×T 测 ,

[0048] Where P is the simulation analysis parameter, G is the slope of the load change per unit time, and T is the load change rate per unit time. 测 It predicts the length of the time period.

[0049] The technical solution provided by this invention has the following beneficial effects:

[0050] This invention discloses a method for optimizing resonance and torque pulsation in electric vehicle prime movers. Addressing the issues of resonance frequency drift and abnormal torque pulsation caused by temperature load variations and material physical properties during prime mover operation, the method collects internal temperature and load data using sensors, filters noise through signal processing, generates a real-time dynamic fluctuation index, and inputs it into a finite element analysis model to simulate the resonance frequency drift trend and calculate abnormal torque pulsation values. When the abnormal value exceeds a preset pulsation threshold, the invention activates a vibration and bearing wear prediction module, locates the electromagnetic field inhomogeneity region through time-series analysis, and generates a vicious cycle risk assessment score. If the score exceeds the vicious cycle threshold, the invention triggers a dynamic adjustment characteristic algorithm to adjust the component damping coefficient to stabilize the resonance frequency, optimize the torque output curve to suppress abnormal pulsation, and verify the material response through feedback loops to ensure the elimination of the fluctuation source. Simultaneously, the database is updated to optimize subsequent simulations. This invention, through multi-dimensional data fusion and dynamic adjustment, achieves a significant improvement in the stability and efficiency of prime mover operation. Attached Figure Description

[0051] Figure 1 This is a flowchart of a method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to the present invention.

[0052] Figure 2 This is a schematic diagram of a method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to the present invention.

[0053] Figure 3 This is another schematic diagram of the method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0055] like Figures 1-3 As shown, this invention provides a method for optimizing resonance and torque pulsation in the prime mover of an electric vehicle, which specifically includes:

[0056] S1 collects data on internal temperature load changes and material physical property parameters of the prime mover through sensors, and uses signal processing methods to filter out noise interference to obtain real-time dynamic fluctuation generation indicators.

[0057] Optionally, this step also includes:

[0058] Step S11: Collect temperature and load change data inside the prime mover through sensors, and use analog-to-digital conversion technology to obtain digital temperature and load signals.

[0059] Step S12: Apply wavelet transform from the PyWavelets library to the digitized temperature signal and load signal to filter out high-frequency noise and obtain the denoised temperature signal and denoised load signal.

[0060] Step S13: Extract time series features from the denoised temperature signal and the denoised load signal, including calculating the mean and standard deviation of the denoised temperature signal and the denoised load signal, and using the FFT function of NumPy to obtain frequency domain feature data.

[0061] Step S14: If there are abnormal peaks in the frequency domain feature data, the abnormal fluctuations are judged by the preset fluctuation threshold to obtain the abnormal fluctuation identifier.

[0062] Step S15: Based on the abnormal fluctuation identifier and combined with the material physical property parameters, support vector machine regression from the scikit-learn library is used to determine the correlation model between fluctuation and material properties.

[0063] Step S16: Extract the dynamic fluctuation trend from the correlation model, including calculating the slope of the dynamic fluctuation trend predicted by the correlation model and generating a real-time fluctuation index.

[0064] Step S17: Update the preset fluctuation threshold using a weighted average method based on the real-time fluctuation index.

[0065] Preferably, the optimized threshold parameters are obtained by using a calculation method in which the new fluctuation threshold is equal to the old fluctuation threshold multiplied by 0.8 of the fluctuation index plus a fixed offset of 0.2.

[0066] For example, in prime mover operation monitoring scenarios, sensors collect internal temperature and load data, demonstrating real-time performance and accuracy. Sensors are placed in key components of the prime mover, such as bearings or motor windings, to collect temperature signals (in degrees Celsius) and load signals (in Newtons). Analog-to-digital conversion technology converts the analog signals into digital signals. Assuming the temperature signal sampling frequency is 1000Hz and the load signal is 500Hz, high-resolution digital data is obtained. This conversion ensures data accuracy, lays the foundation for subsequent processing, and effectively avoids the vulnerability of analog signals to interference.

[0067] In one possible implementation, the AngriffSystem wavelet transform from the PyWavelets library is used for denoising. The db4 wavelet basis function is selected to decompose the signal to the third level and filter out high-frequency noise.

[0068] For example, raw temperature signal data may contain spikes caused by transient interference. Wavelet transform preserves the main trends and removes noise, resulting in a smooth, denoised temperature signal. Load signals undergo similar processing to filter out noise caused by mechanical vibrations. This denoising improves signal quality, provides reliable data for feature extraction, and enhances the accuracy of subsequent analysis.

[0069] Specifically, when extracting time series features, the mean and standard deviation of the denoised temperature signal and the denoised load signal are calculated.

[0070] For example, the mean of the denoised temperature signal over a certain period is 85.2°C, with a standard deviation of 2.3°C, reflecting temperature stability; the mean of the denoised load signal is 1200N, with a standard deviation of 50N, indicating load fluctuations. Using NumPy's `fft` function to perform a Fast Fourier Transform (FFT) yields frequency domain feature data, identifying an abnormal peak in the temperature signal at 10Hz and a spike in the load signal at 15Hz, indicating periodic abnormal fluctuations. This feature extraction helps identify potential faults and improves the predictability of equipment maintenance.

[0071] For example, if an abnormal peak value in the frequency domain feature data exceeds a preset fluctuation threshold (e.g., a temperature peak exceeding 5°C, or a load exceeding 100N), it is marked as an abnormal fluctuation. A correlation model is constructed using scikit-learn's support vector machine regression, combined with material physical property parameters (e.g., coefficient of thermal expansion, material fatigue limit), to analyze the relationship between fluctuations and material properties.

[0072] For example, the model reveals the correlation between high-temperature fluctuations and material thermal fatigue, predicting a decreasing trend in fatigue life. This correlation model provides a scientific basis for equipment health management and extends service life.

[0073] In one possible implementation, the dynamic fluctuation trend is extracted from the correlation model, and the slope of the dynamic fluctuation trend is calculated.

[0074] For example, a temperature fluctuation slope showing an increase of 0.5°C per hour indicates a potential overheating risk. Real-time fluctuation indicators are generated, and the fluctuation threshold is updated using a weighted averaging method.

[0075] For example, if the old fluctuation threshold was 5°C and the fluctuation index was 6°C, the new fluctuation threshold is calculated as 5 × 0.8 + 0.2 × 6 = 5.2°C. This dynamic adjustment makes the threshold more closely match actual operating conditions, improves the sensitivity and reliability of anomaly detection, and reduces false alarms or missed alarms.

[0076] It should be noted that the advantage of the above method lies in the end-to-end processing from data acquisition to threshold optimization, which enables precise monitoring of the prime mover's operating status.

[0077] For example, real-time fluctuation indicators can be used for predictive maintenance, detecting anomalies early and reducing downtime costs. The combination of wavelet transform and support vector machine regression enhances data analysis capabilities under complex operating conditions, providing technical support for the intelligent management of industrial equipment.

[0078] S2, based on the real-time dynamic fluctuations, generates an index input to a preset finite element analysis model, simulates the resonant frequency drift trend, and calculates potential torque pulsation anomalies.

[0079] Optionally, this step also includes:

[0080] Step S21: Obtain real-time dynamic fluctuation data. The vibration signal during equipment operation is collected by the sensor and converted into frequency domain data using the MATLAB fft function to obtain the dynamic fluctuation spectrum.

[0081] Step S22: Extract key frequency components from the dynamic fluctuation spectrum, decompose the signal using wavelet transform algorithm based on the key frequency components, detect potential resonant frequency anomalies, and obtain an abnormal frequency set.

[0082] Step S23: For the abnormal frequency set, if the frequency point exceeds the preset abnormal point threshold, the corresponding torque pulsation value is calculated by integrating the vibration signal using the Simpson rule to determine the pulsation abnormal value.

[0083] Step S24: Extract the maximum and minimum values ​​from the pulsation anomalies, combine them with key frequency components, and generate a continuous torque pulsation trend curve using a linear interpolation method to obtain torque pulsation trend prediction data.

[0084] Step S25: Extract significant change points from the torque pulsation trend prediction data. If the magnitude of the change point exceeds the preset trend threshold, generate an abnormal warning signal by predicting the torque pulsation trend through the ARIMA model to determine potential fault risks.

[0085] For example, in acquiring vibration signals during equipment operation, sensors are typically mounted on critical components of the prime mover, such as bearings or rotors, to capture high-precision vibration data. Accelerometers can be selected as sensors, with their sampling frequency usually set to 10kHz to ensure the capture of high-frequency vibration information. This acquisition method ensures signal integrity, providing a reliable data foundation for subsequent analysis.

[0086] Specifically, when using MATLAB's `fft` function to convert time-domain vibration signals into frequency-domain data, a spectrum diagram is obtained, containing multiple frequency components. Assuming the dominant frequency of the equipment during normal operation is 50Hz, spectrum analysis may reveal harmonic components at 100Hz and 150Hz. If an abnormally high amplitude frequency of 200Hz is found, it may indicate a potential mechanical imbalance or loosening problem. This spectrum analysis visually reflects the frequency distribution of the equipment's operating status.

[0087] In one embodiment, a wavelet transform algorithm based on spectral data can further decompose the signal and extract resonant frequency anomalies. Wavelet transform, through decomposition at different scales, can separate low-frequency trends and high-frequency details.

[0088] For example, if the frequency amplitude of 200Hz is found to be significantly higher than the normal range of 0.1mm / s, reaching 0.5mm / s, it can be marked as an abnormal frequency. This decomposition method can effectively isolate noise interference and accurately locate potential problem points.

[0089] For example, when determining the anomaly threshold for an abnormal frequency set, the preset anomaly threshold might be 0.3 mm / s. If the amplitude of a 200 Hz frequency exceeds this threshold, the torque ripple value is calculated using Simpson's rule integration. Assuming the integration result shows a torque ripple value of 5 Nm, far exceeding the normal range of 2 Nm, this indicates a significant ripple anomaly. This method, by quantifying the ripple value, provides an intuitive basis for assessing the degree of anomaly.

[0090] Specifically, the maximum and minimum values ​​are extracted from the pulsation anomalies, such as a maximum value of 5 Nm and a minimum value of 1 Nm. Combined with key frequency components, a torque pulsation trend curve is generated through linear interpolation.

[0091] For example, the interpolation results show that the torque gradually increased from 3 Nm to 5 Nm within 10 minutes, indicating that the abnormal torque pulsation trend is intensifying. This trend curve can clearly show the trajectory of changes in the equipment's condition.

[0092] In one embodiment, for significant changes in torque ripple trend prediction data, if the change exceeds a preset trend threshold of 0.5 Nm, torque ripple trend prediction is performed using the ARIMA model.

[0093] For example, model analysis might show that the torque could continue to rise to 6 Nm within the next 5 minutes, triggering an anomaly warning signal. This prediction can identify potential failure risks in advance, providing a basis for maintenance decisions. The warning signal can be directly transmitted to the equipment management system, prompting operators to check the bearings or lubrication status.

[0094] For example, the method described above achieves real-time monitoring of equipment operating status through a complete process from vibration signal acquisition to anomaly early warning. The implementation methods of each technical step are mutually supportive, from signal acquisition to spectrum analysis, and then to anomaly detection and trend prediction, with rigorous logic that collectively ensures accurate assessment of equipment status. This method is applicable to prime mover monitoring in a single scenario, maintaining consistency across the business domain.

[0095] S3. If the simulated abnormal torque pulsation value exceeds the preset abnormal pulsation threshold, the vibration increase bearing wear prediction module is activated to determine the region of uneven electromagnetic field distribution through time series data analysis.

[0096] Optionally, this step also includes:

[0097] Step S31: If the simulated torque pulsation value exceeds the preset pulsation abnormality threshold, then obtain the timing data from the motor operation data, use MATLAB's fft function to perform a fast Fourier transform to calculate the amplitude spectrum, identify the peak frequency, and determine the abnormal frequency range.

[0098] Step S32: Based on the abnormal frequency range, extract the field distribution data corresponding to the time step from the electromagnetic field simulation data, perform finite element analysis using ANSYS software to divide the mesh and solve the field equations, calculate the non-uniform region of the electromagnetic field, and obtain the field distribution deviation.

[0099] Step S33: If the field distribution deviation exceeds the predefined standard, extract bearing wear-related features from historical operating data, train and predict the bearing wear degree using the RandomForestRegressor model of scikit-learn, and obtain the wear prediction value.

[0100] Step S34: If the wear prediction value is higher than the safety threshold, the cumulative impact of vibration on the bearing is calculated using an integral formula based on the time series data and field distribution deviation to determine the high-risk area.

[0101] Preferably, the cumulative effect of vibration on the bearing is calculated using the following formula:

[0102] I=∫(v(t)·d(t))dt,

[0103] Where I is the cumulative effect, v(t) is the vibration velocity in the time series data, d(t) is the field distribution deviation, and t is the running time.

[0104] Step S35: Using data from high-risk areas, the NumPy average function is used to perform a weighted average fusion of simulated torque pulsation values ​​and predicted wear values ​​to generate optimized control parameters.

[0105] Step S36: Adjust the motor operation strategy according to the optimized control parameters, update the electromagnetic field distribution in real time, and reduce the impact of torque pulsation and vibration.

[0106] For example, in a motor operation monitoring scenario, if the simulated torque pulsation value exceeds a preset pulsation anomaly threshold, time-series data can be extracted from the motor operation data. Time-series data typically includes information such as rotational speed, current, and vibration velocity, spanning one hour with a sampling frequency of 1000Hz. Processing this data using MATLAB's `fft` function can generate an amplitude spectrum and identify the peak frequency.

[0107] For example, if a significant peak is found at 400Hz, exceeding the normal range of 300-350Hz, it indicates an abnormal frequency range. This frequency may be related to motor rotor imbalance or electromagnetic interference.

[0108] Specifically, when extracting field distribution data corresponding to the time step from electromagnetic field simulation data, ANSYS Maxwell software can be selected for analysis. Assuming a time step of 0.01 seconds, the mesh size can be set to 1 mm to ensure accuracy. After solving the field equations, the electromagnetic field inhomogeneity region was calculated, and it was found that the magnetic flux density deviation in a certain region reached 0.2T, exceeding the predefined standard of 0.15T. This deviation may originate from stator winding asymmetry or material defects.

[0109] It should be noted that field distribution deviation analysis helps to pinpoint the specific location of electromagnetic field anomalies, providing a basis for subsequent optimization.

[0110] In one possible implementation, when extracting bearing wear-related features from historical operating data, vibration acceleration, temperature, and lubricating oil particle count can be selected as features. Assume the historical data includes records from the past 30 days, with an average vibration acceleration of 2 m / s², an average temperature of 70°C, and a particle count of 5000 particles / cm³. These features are trained using the scikit-learn RandomForestRegressor model to predict the degree of bearing wear.

[0111] For example, the model outputs a wear prediction value of 0.8 mm, which exceeds the safety threshold of 0.5 mm, indicating that the bearing has a significant risk of wear.

[0112] For example, when calculating the cumulative effect of vibration on a bearing, the vibration velocity v(t) in the integral formula can be obtained from time-series data, assuming an average value of 0.1 m / s and a field distribution deviation d(t) of 0.2 T. The integral calculation reflects the interaction between vibration and electromagnetic field inhomogeneity, and the cumulative effect value I can be used to assess the bearing fatigue level. Assuming an I value of 500, it indicates that the high-risk area is concentrated in the outer ring of the bearing.

[0113] Specifically, when fusing simulated torque ripple values ​​and predicted wear values, a weighted average can be performed using NumPy's `average` function. Assuming a simulated torque ripple value of 10 N·m and a predicted wear value of 0.8 mm, with weighting coefficients of 0.6 and 0.4 respectively, optimized control parameters are generated. Based on these parameters, the motor operating strategy is adjusted, such as reducing the speed to 80% or optimizing current distribution, and the electromagnetic field distribution is updated in real time to reduce the magnetic flux density deviation to below 0.1 T, thereby reducing the impact of torque ripple and vibration. This method, through multi-dimensional data fusion, improves the stability and lifespan of motor operation.

[0114] S4 uses the acquired information on the uneven distribution of electromagnetic fields and combines it with real-time sensing state data to generate a risk assessment score for a vicious cycle.

[0115] Optionally, this step also includes:

[0116] Step S41: Obtain electromagnetic field distribution data and real-time sensing data, and generate the first dataset by weighted averaging and fusion.

[0117] Step S42: Extract the electromagnetic field distribution from the first dataset, use K-means clustering to identify uneven regions, and obtain regional feature data.

[0118] Step S43: Extract the real-time perception part from the first dataset, extract key state parameters through principal component analysis, and generate a state feature set.

[0119] Step S44: If the key parameters in the state feature set exceed the preset state threshold, then calculate the initial risk score through linear regression.

[0120] Preferably, the initial risk score is calculated using the following formula:

[0121] Y = aX + b

[0122] Where Y is the initial risk score, X is the key parameter value, a is the slope, and b is the intercept.

[0123] Step S45: Input the regional feature data and the initial risk score into the support vector machine algorithm to generate a comprehensive risk assessment vector.

[0124] Step S46: Based on the comprehensive risk assessment vector, the final vicious cycle risk assessment score is generated using the decision tree algorithm.

[0125] Step S47: By comparing thresholds, the final vicious cycle risk assessment score is converted into a standardized risk level output.

[0126] In one possible implementation, acquiring electromagnetic field distribution data and real-time sensing data are the core steps in motor operation analysis. Electromagnetic field distribution data typically comes from simulation software and reflects the spatial distribution of the magnetic field strength inside the motor; real-time sensing data is collected through sensors, including speed, temperature, vibration, etc. When fusing these two types of data, a weighted average method can be used to generate the first dataset.

[0127] For example, the weight of electromagnetic field distribution data is set to 0.6, and the weight of real-time sensing data is set to 0.4, to highlight the impact of electromagnetic fields on motor performance. Assuming a motor is running, the electromagnetic field distribution data contains 1000 sampling points, and the real-time sensing data contains 500 vibration and temperature data points. The first dataset generated after weighted averaging can contain a comprehensive feature vector for subsequent analysis.

[0128] For example, the electromagnetic field distribution is extracted from the first dataset, and K-means clustering is used to identify uneven regions. K-means clustering divides the electromagnetic field distribution into multiple regions by iteratively calculating the distance between data points and cluster centers.

[0129] For example, electromagnetic field data for a motor might be divided into three clusters: a high-intensity region, a medium-intensity region, and a low-intensity region. Each cluster generates regional characteristic data, such as the mean and variance of the magnetic field strength, reflecting the spatial characteristics of the non-uniform region. This characteristic data can be used to locate areas that may cause torque anomalies.

[0130] Specifically, the real-time sensing data is extracted from the first dataset, and principal component analysis (PCA) is used to extract key state parameters. PCA reduces the dimensionality of high-dimensional sensing data (such as vibration frequency, amplitude, and temperature) into a few key state parameters.

[0131] For example, vibration frequency and amplitude may be identified as the main components to generate a state feature set. If key parameters (such as vibration frequency) exceed a preset state threshold, such as 100Hz, it indicates that the motor may be in an abnormal state and further analysis is required.

[0132] In one embodiment, an initial risk score is calculated using linear regression. Assuming the key parameter is vibration frequency, the linear regression model is trained based on historical data, with a slope 'a' of 0.8 and an intercept 'b' of 5. In a particular run, the vibration frequency is 120Hz. Substituting this into the formula Y = 0.8 × 120 + 5, an initial risk score of 101 is obtained. This score reflects the potential risk of the abnormal state.

[0133] For example, regional feature data and initial risk scores are input into a support vector machine (SVM) algorithm to generate a comprehensive risk assessment vector. The SVM then uses classification or regression to integrate electromagnetic field inhomogeneity and operational anomalies, outputting the comprehensive risk assessment vector.

[0134] For example, during the operation of a motor, the comprehensive risk assessment vector might indicate that high-risk areas are concentrated at specific locations on the stator. Then, a decision tree algorithm is used to process the comprehensive risk assessment vector to generate the final vicious cycle risk assessment score.

[0135] For example, the decision tree outputs a final vicious cycle risk assessment score of 85 based on multiple dimensions of the comprehensive risk assessment vector, indicating a high risk.

[0136] Specifically, the final vicious cycle risk assessment score is converted into a standardized risk level by threshold comparison.

[0137] For example, a score of 80 or above indicates high risk, 60-80 indicates medium risk, and below 60 indicates low risk. A score of 85 corresponds to a high-risk level, and the output can be used to guide motor maintenance strategies, such as adjusting operating parameters or inspecting specific areas.

[0138] In one possible implementation, the advantage of the above method lies in its multi-level analysis, from data fusion to risk assessment, progressively focusing on potential problems in motor operation. The implementation of each technical topic is closely integrated, ensuring that the analysis results are accurate and actionable.

[0139] For example, K-means clustering can accurately locate regions of electromagnetic field anomalies, principal component analysis effectively extracts key operational states, and decision trees provide intuitive risk levels, offering a reliable basis for subsequent optimization. This multi-faceted analysis is logically rigorous and progressive, ensuring that the evaluation results are comprehensive and practical.

[0140] S5, when the risk assessment score for the formation of a vicious cycle is higher than the vicious cycle threshold, the dynamic adjustment characteristic algorithm is triggered to adjust the damping coefficient of the prime mover component to obtain stable resonance frequency parameters.

[0141] Optionally, this step also includes:

[0142] Step S51: If the risk assessment score exceeds the preset vicious cycle threshold, the real-time operating data of the prime mover component is obtained through the sensor to obtain the parameters of the first component.

[0143] Step S52: Based on the parameters of the first component, the support vector machine algorithm is used to classify the parameters of the first component and determine the risk level.

[0144] Step S53: If the risk level reaches the predefined high-risk state, the damping coefficient adjustment amount is calculated by gradient descent to obtain the first adjustment parameter.

[0145] Step S54: Based on the first adjustment parameter, update the damping coefficient of the prime mover component using a proportional-integral-derivative controller to obtain the second component parameter.

[0146] Step S55: Perform Fourier transform on the parameters of the second component using the MATLAB fft function to obtain the first resonant frequency parameters.

[0147] Step S56: Based on the first resonant frequency parameter, use the ARIMA model to detect frequency stability and determine the frequency stability status.

[0148] Step S57: If the frequency stability does not meet the preset stability standard, the damping coefficient is iteratively adjusted to obtain the second resonant frequency parameter.

[0149] For example, in scenarios involving electromagnetic field-related risk assessment and prime mover component control, when sensors acquire real-time operating data of prime mover components to generate first component parameters, data can be collected using high-precision magnetic field sensors and vibration sensors.

[0150] Specifically, the magnetic field sensor monitors the intensity and distribution of the electromagnetic field inside the prime mover, generating magnetic flux density values, such as those ranging from 0.5 Tesla to 1.2 Tesla; the vibration sensor records the vibration frequency and amplitude of the components, such as vibration frequencies ranging from 10 Hz to 50 Hz. These data together constitute the first component parameters, providing a basis for subsequent classification.

[0151] It should be noted that the selection of sensors should take into account their sensitivity and environmental adaptability to ensure the accuracy of the data.

[0152] In one embodiment, when using the support vector machine algorithm to classify the parameters of the first component, magnetic flux density and vibration frequency can be used as input features, and the pre-trained support vector machine model can classify low, medium and high risk levels based on historical data.

[0153] For example, when the magnetic flux density exceeds 1.0 Tesla and the vibration frequency is higher than 40 Hz, the model may output a high-risk level. The classification results can be displayed through a visual interface, making it easy for operators to quickly determine the risk status.

[0154] For example, when calculating the damping coefficient adjustment using gradient descent for high-risk conditions, optimization can be based on the difference between the vibration frequency and the expected stable frequency. Assuming the current vibration frequency is 45 Hz and the target stable frequency is 30 Hz, the gradient descent algorithm, through iterative calculation, determines that the damping coefficient needs to be increased by 0.2. This adjustment can effectively reduce the vibration amplitude of the component, thereby reducing the potential risk of resonance.

[0155] In one embodiment, the proportional-integral-derivative controller can achieve precise control through a real-time feedback mechanism when updating the damping coefficient.

[0156] For example, after receiving the first adjustment parameter of 0.2, the controller adjusts the damping coefficient from the initial value of 1.5 to 1.7, and ensures a smooth transition of the parameters through closed-loop control. The adjusted second component parameter, such as reducing the vibration frequency to 35 Hz, can significantly improve the stability of the system.

[0157] For example, when performing a Fourier transform on the parameters of the second component using MATLAB's `fft` function, the vibration signal can be decomposed into frequency components, and the first resonant frequency parameter, such as a dominant frequency of 32 Hz, can be extracted. This analysis can clearly reflect the working state of the prime mover and provide a basis for subsequent stability testing.

[0158] It should be noted that the application of Fourier transform requires ensuring that the signal sampling rate is high enough, such as 1000 Hz, to avoid frequency aliasing.

[0159] In one embodiment, when using the ARIMA model to detect frequency stability, future frequency trends can be predicted based on historical frequency data.

[0160] For example, the ARIMA model analyzes data from nearly 1000 time points to determine whether the 32 Hz dominant frequency fluctuates within ±5%. If the fluctuation exceeds this range, it is considered unstable, and the damping coefficient needs further adjustment. After iterative adjustments, the second resonant frequency parameter may drop to 30 Hz, approaching the target stable value.

[0161] For example, in iteratively adjusting the damping coefficient, it can be done through multiple small-step adjustments, such as increasing by 0.05 each time, until the frequency stabilizes. This method can gradually optimize system performance and avoid overresponse caused by large adjustments.

[0162] It should be noted that iterative adjustments must be combined with real-time monitoring data to ensure that each adjustment is based on the latest operating status, thereby improving the accuracy of regulation and the long-term stability of the system.

[0163] S6 updates the torque control model by stabilizing the resonant frequency parameter and optimizes the torque output curve to suppress abnormal pulsation.

[0164] Optionally, this step also includes:

[0165] Step S61: Obtain real-time torque output data and resonant frequency data from the sensor, calculate the frequency response using fast Fourier transform on the torque output data, and obtain the torque fluctuation characteristics, wherein the torque fluctuation characteristics are the peak amplitude in the frequency response.

[0166] Step S62: If the torque fluctuation characteristics exceed the preset torque fluctuation threshold, then apply K-means clustering to group abnormal peak values ​​in the frequency response data to determine the frequency range of the pulsation anomaly.

[0167] Step S63: Based on the frequency range of the pulsation anomaly, the gradient descent algorithm is used to calculate the loss function, which is the mean square error between the abnormal frequency and the resonant frequency parameter. The resonant frequency parameter is then updated to obtain the optimized parameter set.

[0168] Step S64: Adjust the torque control model by optimizing the parameter set, solve the model differential equation using the Runge-Kutta method, and generate a new torque output curve.

[0169] Step S65: Extract the smoothness index from the new torque output curve, where the smoothness index is the standard deviation of the torque output curve, and determine whether it meets the preset smoothness requirements.

[0170] In step S66, if the smoothness index does not meet the requirements, the parameter set is iteratively updated based on the standard deviation of the torque output curve to obtain the updated torque control model.

[0171] Step S67: Based on the updated torque control model, the output is adjusted using a PID controller to generate the final torque output curve and suppress abnormal pulsation.

[0172] For example, in the field of prime mover operation monitoring, acquiring real-time torque output data and resonance frequency data from sensors is a crucial step. Sensors are typically mounted on the prime mover's output shaft to collect torque and vibration signals in real time.

[0173] For example, suppose a prime mover is running and a sensor collects torque data 1000 times per second, obtaining a torque value sequence such as 500 Nm, 502 Nm, 498 Nm, etc., while simultaneously collecting vibration frequency data. By processing the torque data using a Fast Fourier Transform, the time-domain signal can be converted into the frequency domain, yielding the frequency response curve.

[0174] For example, the frequency response shows a peak amplitude of 10 Nm at 50 Hz, exceeding the preset torque fluctuation threshold of 8 Nm, indicating a risk of torque fluctuation. This analysis method, through frequency domain feature extraction, helps to accurately locate abnormal fluctuations.

[0175] Specifically, applying K-means clustering to frequency response data can identify abnormal peaks.

[0176] For example, frequency response data is divided into three categories. The clustering results show peak anomalies at 50Hz and 75Hz, determining the frequency range of pulsation anomalies to be 45-80Hz. The advantage of K-means clustering is its ability to automate grouping, reduce manual intervention, and improve anomaly detection efficiency. Based on this frequency range, a gradient descent algorithm is used to calculate the loss function.

[0177] For example, assuming the loss function is defined as the mean square error between the anomalous frequency and the target resonant frequency, with an initial resonant frequency of 60Hz and an anomalous frequency of 50Hz, the resonant frequency parameter is updated to 58Hz through iterative optimization. This method uses quantization error to guide parameter adjustment, ensuring the accuracy of the optimization direction.

[0178] In one embodiment, the torque control model is adjusted by optimizing the parameter set, and the Runge-Kutta method is used to solve the model differential equations.

[0179] For example, the model describes the change in torque output over time, with an initial torque of 500 Nm and a target of stable output. Using the Runge-Kutta method, a new torque output curve is generated, showing that the torque fluctuation range decreases from ±5 Nm to ±2 Nm. The smoothness index is calculated using standard deviation; for instance, the new curve's standard deviation is 1.5 Nm, lower than the preset standard deviation threshold of 2 Nm, indicating that the smoothness requirement is met. This method improves control accuracy and ensures output stability through numerical solution.

[0180] For example, if the smoothness index does not meet the standard and the standard deviation is 2.5 Nm, the parameter set is updated iteratively based on the feedback.

[0181] For example, the proportional gain in the control model is increased from 1.0 to 1.2, and the torque output curve is recalculated until the standard deviation drops to 1.8 Nm. This iterative feedback mechanism gradually approaches the target state by dynamically adjusting the parameters. Finally, based on the updated torque control model, a PID controller is used to regulate the output.

[0182] For example, the PID controller adjusts the torque to 498 Nm based on the error signal, generating a smooth final torque output curve and successfully suppressing abnormal pulsation. This method significantly improves the system's dynamic response capability through real-time adjustment.

[0183] S7. After obtaining the optimized torque output curve, the input feedback loop mechanism verifies the material's physical property response and determines whether the source of dynamic fluctuations has been eliminated.

[0184] Optionally, this step also includes:

[0185] Step S71: Collect torque data through sensors to obtain an initial torque dataset.

[0186] Step S72: Generate the torque output curve from the initial torque dataset.

[0187] Step S73: Use Python's SciPy library to perform median filtering and low-pass filtering on the initial torque dataset to obtain the optimized torque dataset.

[0188] Step S74: Extract material physical property response data from the optimized torque dataset to obtain the physical property response dataset.

[0189] Step S75: If the fluctuation amplitude in the physical characteristic response dataset is lower than the preset fluctuation amplitude threshold, it is determined that the source of dynamic fluctuation has been eliminated, and the fluctuation elimination state is obtained.

[0190] Step S76: Based on the fluctuation elimination state, extract the peak value and frequency features from the torque output curve to obtain the fluctuation feature dataset.

[0191] Step S77: Classify the fluctuation feature dataset using RandomForestClassifier from the Scikit-learn library to determine the stability of the fluctuation source and obtain the stability classification result.

[0192] Step S78: Based on the stability classification results, the parameters of the extraction process are adjusted using a PID controller to obtain optimized extraction parameters.

[0193] For example, in the process of acquiring torque data from sensors, the sensors are typically mounted on the output shaft of rotating machinery to record torque changes in real time. Suppose that an industrial motor is running, and the sensor acquires data at a sampling rate of 100Hz, generating an initial torque dataset containing 5000 data points. This data reflects the torque fluctuations of the motor under different loads and may contain noise or abnormal peaks. Accurate sensor calibration must be ensured during acquisition to avoid data distortion. This method effectively captures the dynamic characteristics of torque, providing a reliable foundation for subsequent analysis.

[0194] Specifically, the process of generating torque output curves from initial torque datasets can be achieved by using data visualization tools to convert time-series data into curves.

[0195] For example, the torque values ​​in the dataset fluctuate between 0 and 100 Nm, and the curve shows periodic oscillations with peaks occurring at specific time points. When generating the curve, it is necessary to ensure that the time axis is aligned to avoid data offset. This type of curve intuitively reflects the dynamic changes in torque, facilitating subsequent filtering processing.

[0196] In one embodiment, when using the SciPy library for median filtering and low-pass filtering, the median filtering window can be set to 5 data points, and the cutoff frequency of the low-pass filtering can be set to 10Hz.

[0197] For example, the initial torque data contains high-frequency noise, manifesting as small random fluctuations. Median filtering can effectively remove isolated outliers, such as a sudden change in data to 150 Nm, while low-pass filtering can smooth high-frequency interference and preserve the main fluctuation trend. The optimized torque dataset has a smoother waveform and significantly reduced noise, which helps to extract accurate physical characteristics.

[0198] For example, when extracting material physical property response data, one can focus on the relationship between torque data and material stress and strain. Assuming the motor shaft is made of steel, the physical property response dataset may reflect the elastic deformation of the material under different torques.

[0199] For example, when the torque is 50 Nm, the strain data remains stable at 0.01%, while when the torque increases to 80 Nm, the strain increases to 0.015%. By analyzing these data, it can be determined whether the material is approaching its fatigue limit. This extraction method can accurately identify the dynamic changes in material properties.

[0200] Specifically, to determine whether the source of dynamic fluctuations has been eliminated, a fluctuation amplitude threshold, such as 0.5 Nm, needs to be set. If the fluctuation amplitude of the optimized dataset drops to 0.3 Nm, it indicates that noise and abnormal fluctuations have been effectively filtered out, achieving fluctuation elimination. This determination method can quickly confirm data quality and provide a reliable basis for subsequent feature extraction.

[0201] In one embodiment, when extracting peak and frequency features from the torque output curve, the maximum value and periodicity of the curve can be identified.

[0202] For example, the curve shows three peaks per second, with a peak amplitude of 70 Nm and a frequency of 3 Hz. These characteristics reflect the motor's operating pattern, and the algorithm must be sensitive to periodic signals during extraction. This method can accurately capture fluctuation characteristics, providing data support for classification.

[0203] For example, when using RandomForestClassifier to classify fluctuation features, the feature dataset can be divided into two categories: stable and unstable. Assuming the training data contains 1000 features, the model uses features such as peak amplitude and frequency to determine whether the fluctuation is caused by mechanical loosening.

[0204] For example, fluctuations with frequencies above 5 Hz may be associated with bearing failure. The classification results can guide subsequent parameter adjustments and enhance system stability.

[0205] Specifically, when using a PID controller to adjust the extracted parameters, the control gain can be optimized based on the classification results.

[0206] For example, if the stability classification indicates that the fluctuation originates from changes in external load, the PID controller can quickly stabilize the output by adjusting the proportional gain to 2.0 and the integral time to 0.5 seconds. This method can dynamically adapt to different operating conditions, ensuring a smooth and reliable torque output curve.

[0207] S8. If it is determined that the source of dynamic fluctuations has been eliminated, the adjustment log is recorded and the temperature load change database is updated for subsequent simulation.

[0208] Optionally, this step also includes:

[0209] Step S81: If a wave signal is detected, a fast Fourier transform is used to determine the first dynamic wave source from the wave signal.

[0210] Step S82: By analyzing the first dynamic fluctuation source, obtain the first adjustment log and store it in the preset database.

[0211] Step S83: Extract the first adjustment log from the preset database and calculate the temperature load change rate to update the first temperature load change data.

[0212] Step S84: Based on the first temperature load change data, use a linear regression model to predict the first load change trend from the first temperature load change data.

[0213] Step S85: Calculate simulation analysis parameters based on the first load change trend.

[0214] Preferably, the simulation analysis parameters are calculated using the following formula:

[0215] P=G×T 测 ,

[0216] Where P is the simulation analysis parameter, G is the slope of the load change per unit time, and T is the load change rate per unit time. 测 It predicts the length of the time period.

[0217] Step S86: Using simulation analysis parameters, perform Monte Carlo simulation to obtain the first dynamic fluctuation adjustment strategy from the simulation analysis parameters.

[0218] Step S87: Update the fluctuation monitoring configuration according to the first dynamic fluctuation adjustment strategy to improve the source detection accuracy.

[0219] For example, in the scenario of testing physical properties of materials, the detection of fluctuation signals is a crucial step. Fluctuation signals are typically acquired in real time by sensors and contain physical quantities such as torque, vibration, or temperature. Fast Fourier Transform (FFT) is an analytical method that converts time-domain signals into frequency-domain signals to identify the main frequency components in the fluctuation signal. Assuming the sensor acquires a torque fluctuation signal, FFT can decompose the signal into multiple frequency components, determining the frequency with the highest amplitude as the primary source of dynamic fluctuation.

[0220] For example, if the analysis reveals that the 10Hz frequency component has the highest amplitude, it can be identified as the primary source of dynamic fluctuations. This method clearly separates different fluctuation sources through frequency domain analysis, facilitating subsequent precise processing.

[0221] Specifically, the generation and storage of the first regulation log requires an efficient data management mechanism. The first regulation log records information such as the frequency, amplitude, and timestamp of the first dynamic fluctuation source. Assuming the first dynamic fluctuation source is 10Hz and has an amplitude of 5N·m, the log will record this data and store it in a pre-defined database, such as MySQL or a NoSQL database. During storage, it is necessary to ensure that the data format is consistent, such as JSON format, to facilitate subsequent extraction and analysis.

[0222] Preferably, the database design should support high-concurrency writes to handle the high-frequency data collected in real time. This log storage method provides a reliable data foundation for subsequent analysis.

[0223] In one embodiment, calculating the rate of change of temperature load is a key step after retrieving the first adjustment log from a preset database. The rate of change of temperature load reflects the thermal response characteristics of the material under dynamic conditions.

[0224] For example, assuming the log shows the temperature rising from 20°C to 25°C over a period of 10 seconds, the rate of change of temperature load is 0.5°C / s. Based on this data, a linear regression model can predict the temperature load trend over a future period.

[0225] Specifically, if a regression model predicts that the temperature will rise by 3°C in the next 60 seconds, the trend of load changes can be inferred. This prediction method derives future trends from historical data, providing a basis for dynamic adjustments.

[0226] For example, simulation analysis parameters are calculated based on the trend slope and time factor. Assuming the linear regression model calculates a slope of 0.05°C / s and a prediction time period of 100 seconds, the simulation analysis parameter P = 0.05 × 100 = 5. This parameter quantifies the degree of impact of load changes, providing input for subsequent simulations. Monte Carlo simulations utilize this parameter to simulate fluctuation adjustment strategies through multiple random samplings.

[0227] For example, the simulation might generate 1000 different temperature load scenarios, analyze the changing trends of the fluctuation sources under each scenario, and ultimately determine the optimal adjustment strategy, such as adjusting the sensor sampling frequency to 50Hz to improve detection accuracy. This method enhances the robustness of the strategy through statistical analysis.

[0228] Specifically, updating the fluctuation monitoring configuration is a closed-loop step in the entire process. Based on Monte Carlo simulation results, the adjustment strategy may include increasing sensor sensitivity or changing the sampling interval.

[0229] For example, if high-frequency fluctuations are easily missed in the simulation, the sampling frequency can be increased from 50Hz to 100Hz.

[0230] Preferably, the updated configuration needs to be verified in real time to ensure improved detection accuracy. This closed-loop adjustment mechanism ensures the accuracy and stability of fluctuation source detection through continuous configuration optimization.

[0231] In one embodiment, the above-mentioned steps support each other, forming a complete system for analyzing and adjusting the source of fluctuations. From identifying the source using Fast Fourier Transform, to log storage and retrieval, to temperature load trend prediction, simulation analysis parameter calculation, and Monte Carlo simulation, the system ultimately optimizes performance by updating monitoring configurations. This systematic approach ensures logical rigor from data acquisition to strategy execution, providing efficient support for the detection of material physical properties.

[0232] The above description is merely a specific implementation of this specification. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the scope of protection of this specification is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this specification, and these modifications or substitutions should all be covered within the scope of protection of this specification.

Claims

1. A method for optimizing resonance and torque pulsation in the prime mover of an electric vehicle, characterized in that, The method includes: S1. Data on temperature load changes and material physical property parameters inside the prime mover are collected by sensors, and noise interference is filtered out to obtain real-time dynamic fluctuation generation indicators. S2. Based on the real-time dynamic fluctuation generation indicators, a preset finite element analysis model is input to simulate the resonant frequency drift trend and calculate potential torque pulsation anomalies. S3. If the simulated torque pulsation anomaly value exceeds a preset pulsation anomaly threshold, the vibration-increased bearing wear prediction module is activated to determine the electromagnetic field non-uniform distribution area. S4. Using the acquired electromagnetic field non-uniform distribution area information, combined with real-time sensing state data, a vicious cycle formation risk assessment score is generated. S5. When the vicious cycle formation risk assessment score is higher than the vicious cycle threshold, the damping coefficient of the prime mover components is adjusted to obtain stable resonant frequency parameters. S6. The torque control model is updated using the stable resonant frequency parameters to optimize the torque output curve to suppress pulsation anomalies. S7. After obtaining the optimized torque output curve, a feedback loop mechanism is input to verify the material physical property response and determine whether the source of dynamic fluctuation generation has been eliminated. S8. If the source of dynamic fluctuation generation is determined to have been eliminated, the adjustment log is recorded and the temperature load change database is updated for subsequent simulations.

2. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 1, characterized in that, Step S1 involves collecting data on temperature load changes and material physical property parameters inside the prime mover using sensors, filtering out noise interference to obtain real-time dynamic fluctuation indicators, including: Step S11: Collect temperature and load change data inside the prime mover through sensors, and use analog-to-digital conversion technology to obtain digital temperature and load signals; Step S12: Filter out high-frequency noise from the digital temperature signal and load signal to obtain a denoised temperature signal and a denoised load signal. Step S13: Extract time series features from the denoised temperature signal and the denoised load signal to obtain frequency domain feature data; Step S14: If there are abnormal peaks in the frequency domain feature data, the abnormal fluctuations are judged by the preset fluctuation threshold to obtain the abnormal fluctuation identifier. Step S15: Based on the abnormal fluctuation identifier and the material physical property parameters, determine the correlation model between the fluctuation and the material properties; Step S16: Extract the dynamic fluctuation trend from the correlation model and generate a real-time fluctuation indicator; Step S17: Update the preset fluctuation threshold using real-time fluctuation indicators.

3. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 2, characterized in that, In step S13, time series features are extracted from the denoised temperature signal and the denoised load signal to obtain frequency domain feature data. Extracting time series features from the denoised temperature signal and the denoised load signal includes calculating the mean and standard deviation of the denoised temperature signal and the denoised load signal.

4. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 1, characterized in that, Step S4 involves using the acquired information on the uneven distribution of the electromagnetic field, combined with real-time sensing state data, to generate a risk assessment score for a vicious cycle, including: Step S41: Obtain electromagnetic field distribution data and real-time sensing data, and generate the first dataset by weighted averaging and fusion. Step S42: Extract the electromagnetic field distribution portion from the first dataset, identify non-uniform regions, and obtain regional feature data; Step S43: Extract the real-time perception part from the first dataset, extract key state parameters, and generate a state feature set; Step S44: If the key parameters in the state feature set exceed the preset state threshold, calculate the initial risk score. Step S45: Generate a comprehensive risk assessment vector by combining the regional feature data with the initial risk score; Step S46: Generate the final vicious cycle risk assessment score based on the comprehensive risk assessment vector; Step S47: By comparing thresholds, the final vicious cycle risk assessment score is converted into a standardized risk level output.

5. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 4, characterized in that, In step S44, if the key parameters in the state feature set exceed a preset state threshold, an initial risk score is calculated, including: Calculate the initial risk score using the following formula: Y = aX + b Where Y is the initial risk score, X is the key parameter value, a is the slope, and b is the intercept.

6. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 1, characterized in that, Step S6, which updates the torque control model by stabilizing the resonant frequency parameters and optimizes the torque output curve to suppress abnormal pulsations, includes: Step S61: Obtain real-time torque output data and resonant frequency data from the sensor, and use fast Fourier transform on the torque output data to calculate the frequency response and obtain torque fluctuation characteristics. Step S62: If the torque fluctuation characteristics exceed the preset torque fluctuation threshold, then apply K-means clustering to group abnormal peak values ​​in the frequency response data to determine the frequency range of the pulsation anomaly. Step S63: Calculate the loss function based on the frequency range of the pulsation anomaly, update the resonant frequency parameters, and obtain the optimized parameter set; Step S64: Adjust the torque control model using the optimized parameter set to generate a new torque output curve; Step S65: Extract the smoothness index from the new torque output curve and determine whether it meets the preset smoothness requirements. Step S66: If the smoothness index does not meet the requirements, the parameter set is iteratively updated to obtain the updated torque control model. Step S67: Based on the updated torque control model, the output is adjusted using a PID controller to generate the final torque output curve and suppress abnormal pulsation.

7. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 6, characterized in that, In step S61, real-time torque output data and resonant frequency data are acquired from the sensor, and the frequency response is calculated by using fast Fourier transform on the torque output data to obtain torque fluctuation characteristics, which are the peak amplitudes in the frequency response.

8. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 7, characterized in that, In step S63, based on the frequency range of the pulsation anomaly, a loss function is calculated, the resonant frequency parameters are updated, and an optimized parameter set is obtained. The loss function is the mean square error between the abnormal frequency and the resonant frequency parameters.

9. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 1, characterized in that, Step S8, if it is determined that the source of dynamic fluctuations has been eliminated, then the adjustment log is recorded and the temperature load change database is updated for subsequent simulation use, including: Step S81: If a fluctuation signal is detected, determine the first dynamic fluctuation source from the fluctuation signal; Step S82: By analyzing the first dynamic fluctuation source, obtain the first adjustment log and store it in the preset database; Step S83: Extract the first adjustment log from the preset database and calculate the temperature load change rate to update the first temperature load change data; Step S84: Based on the first temperature load change data, predict the first load change trend from the first temperature load change data; Step S85: Calculate simulation analysis parameters based on the first load change trend; Step S86: Using simulation analysis parameters, obtain the first dynamic fluctuation adjustment strategy from the simulation analysis parameters; Step S87: Update the fluctuation monitoring configuration according to the first dynamic fluctuation adjustment strategy to improve the source detection accuracy.

10. The method for optimizing resonance and torque pulsation of an electric vehicle prime mover according to claim 9, characterized in that, Step S85, calculating simulation analysis parameters based on the first load change trend, includes: calculating simulation analysis parameters using the following formula: P=G×T 测 , Where P is the simulation analysis parameter, G is the slope of the load change per unit time, and T is the load change rate per unit time. 测 It predicts the length of the time period.

Images