Method and system for evaluating fatigue life of pure electric vehicle motor

Abstract

The invention relates to the technical field of motor fatigue life evaluation, and discloses a pure electric vehicle motor fatigue life evaluation method and system, and the method comprises the steps: obtaining historical multi-physical field data and the residual life of a motor in the operation process of the motor of a pure electric vehicle, and carrying out the preprocessing; calculating a synergistic aging influence factor; constructing a data set according to the historical multi-physical field data, the collaborative aging influence factor and the residual life of the motor to train the probability deep learning model, and obtaining a motor fatigue life distribution prediction model; obtaining input data of a target pure electric vehicle, and obtaining a motor fatigue life distribution prediction result through the motor fatigue life distribution prediction model; and based on a Monte Carlo discarding method, sampling the motor fatigue life distribution prediction result for multiple times to obtain a motor fatigue life evaluation result. According to the scheme, accurate evaluation and uncertainty quantification of the fatigue life of the pure electric vehicle motor can be realized, and a decision basis with accuracy and reliability is provided for an active maintenance strategy.

Description

Technical Field

[0001] This invention belongs to the field of motor fatigue life assessment technology, and particularly relates to a method and system for assessing the fatigue life of a pure electric vehicle motor. Background Technology

[0002] As the core power component of pure electric vehicles, the motor's fatigue life directly affects the overall vehicle safety and reliability. Under complex multi-physics coupled conditions, motor materials degrade more rapidly, leading to failures such as insulation failure and bearing wear. Precise life assessment technology can identify potential risks in advance, avoid sudden downtime accidents, and reduce maintenance costs. It is a key technological support for ensuring the safe operation and extending the service life of electric vehicles.

[0003] However, existing technologies for traditional motor life assessment mainly rely on two types of methods:

[0004] Physical model-driven method: Based on material fatigue equations (such as the Coffin-Manson model) or thermal network models, the fatigue life is calculated analytically. This method heavily relies on simplifying assumptions, ignores the coupling effects between different physical fields, has fixed parameters, cannot adapt to dynamic operating conditions, and has high computational complexity, making it difficult to apply in real time.

[0005] Single-field data-driven approach: This method uses statistical features (such as RMS and kurtosis) of vibration signals or temperature data to train shallow machine learning models (such as SVM and random forest). This method only utilizes information from a single physical field and does not consider multi-field collaborative degradation mechanisms; the model output is a deterministic value, lacking uncertainty quantification, which causes the evaluation results to deviate from the actual degradation trajectory and cannot support preventive maintenance decisions with high reliability requirements; feature engineering relies on human experience, and its generalization ability is limited.

[0006] Therefore, there is an urgent need to develop a fatigue life assessment method and system for pure electric vehicle motors, which can accurately assess the fatigue life of pure electric vehicle motors and quantify their uncertainties, providing a decision-making basis with both accuracy and reliability for proactive maintenance strategies. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method and system for assessing the fatigue life of electric motors in pure electric vehicles. This method enables accurate assessment and uncertainty quantification of the fatigue life of electric motors in pure electric vehicles, providing a decision-making basis that is both accurate and reliable for proactive maintenance strategies.

[0008] This invention provides a method for assessing the fatigue life of a pure electric vehicle motor, the method comprising the following steps:

[0009] S1. Obtain historical multiphysics data and remaining lifespan of the motor during the operation of the pure electric vehicle, and preprocess the historical multiphysics data.

[0010] S2. Calculate the synergistic aging influencing factor based on the preprocessed historical multiphysics data;

[0011] S3. Construct a dataset based on historical multiphysics data, collaborative aging influencing factors, and remaining motor lifespan;

[0012] S4. Construct a probabilistic deep learning model, train the probabilistic deep learning model using the dataset, and obtain a motor fatigue life distribution prediction model.

[0013] S5. Obtain multi-physics field data of the motor of the target pure electric vehicle during operation, calculate the synergistic aging influencing factor, and input it into the motor fatigue life distribution prediction model to obtain the motor fatigue life distribution prediction result.

[0014] S6. Based on the Monte Carlo dropout method, the predicted results of motor fatigue life distribution are sampled multiple times to obtain the motor fatigue life assessment results.

[0015] Furthermore, in S1, the multiphysics data includes electromagnetic field data, temperature field data, and stress field data.

[0016] Furthermore, in S2, the synergistic aging influencing factors include electromagnetic-thermal coupling influencing factors, thermo-mechanical coupling influencing factors, and multi-field synergistic effect factors.

[0017] Furthermore, the electromagnetic-thermal coupling influence factor is calculated using the following formula:

[0018] ;

[0019] ;

[0020] ;

[0021] Among them, F em-th C represents the electromagnetic-thermal coupling influence factor. diss F represents the actual heat dissipation capacity of the system. cu F represents the copper loss factor. fe Represents the iron loss factor, I represents the effective value of the current, and R0 represents the phase resistance at the reference temperature. This indicates the temperature coefficient of copper resistance, where T represents the actual winding temperature and T0 represents the reference temperature. Indicates the hysteresis loss coefficient. This represents the eddy current loss coefficient. Represents the stray loss coefficient. B represents the frequency of the magnetic field. m Indicates the magnetic flux density amplitude, and x represents the Steinmetz coefficient. Indicates rated copper loss. This indicates the rated iron loss.

[0022] Furthermore, the thermo-mechanical coupling influence factor is calculated using the following formula:

[0023] ;

[0024] Among them, F th-st This represents the thermo-mechanical coupling influence factor, and E represents the elastic modulus. Indicates the coefficient of thermal expansion. This represents the temperature gradient, and v represents Poisson's ratio. Indicates yield strength.

[0025] Furthermore, the multi-field synergistic effect factor is calculated using the following formula:

[0026] ;

[0027] Among them, F multi Indicates a multi-field synergistic effect factor. , , Indicates the weighting coefficient. Indicates the maximum stress. Indicates the vibration amplitude. This indicates the permissible vibration amplitude.

[0028] Furthermore, in S4, probabilistic deep learning models include:

[0029] The system consists of a first input branch, a second input branch, a feature concatenation layer, a fully connected layer, a Dropout layer, and an output layer.

[0030] The first input branch includes a Transformer encoder and a multi-head attention mechanism for receiving multiphysics data.

[0031] The second input branch includes a fully connected layer and a Dropout layer, which are used to receive the co-aging influence factor;

[0032] The output layer contains two neurons, which output the predicted value of the motor fatigue life and the degree of uncertainty, respectively.

[0033] Furthermore, in S4, the loss function for the probabilistic deep learning model adopts the negative log-likelihood loss function, as expressed below:

[0034] ;

[0035] Where L represents the negative log-likelihood loss function, N represents the total number of training samples, i represents the i-th training sample, and y i μ represents the true remaining lifetime of the i-th training sample. iσ represents the mean of the remaining life distribution predicted by the model for the i-th training sample, i.e., the predicted fatigue life of the motor. i This represents the standard deviation of the remaining lifetime distribution predicted by the model for the i-th training sample, i.e., the degree of uncertainty.

[0036] Furthermore, in S6, the motor fatigue life distribution prediction results are sampled multiple times based on the Monte Carlo dropout method to obtain the motor fatigue life assessment results, including:

[0037] S61. Based on the Monte Carlo dropout method, the predicted results of motor fatigue life distribution are sampled n times to obtain n sets of motor fatigue life distribution sampling results.

[0038] S62. Calculate the mean value of the predicted motor fatigue life from the sampling results of n groups of motor fatigue life distribution, and use it as the predicted motor fatigue life result.

[0039] S63. Calculate the random uncertainty and cognitive uncertainty based on the sampling results of the fatigue life distribution of n groups of motors respectively;

[0040] S64. Calculate the confidence interval of the motor fatigue life prediction results based on accidental uncertainty and cognitive uncertainty;

[0041] S65. The final output includes the motor fatigue life prediction results, confidence interval, random uncertainty, and cognitive uncertainty, which are the motor fatigue life assessment results.

[0042] This invention also provides a fatigue life assessment system for a pure electric vehicle motor, used to perform the above-described fatigue life assessment method for a pure electric vehicle motor. The system includes the following modules:

[0043] The data acquisition module is used to acquire historical multiphysics field data and remaining lifespan of the motor during the operation of the pure electric vehicle, and to preprocess the historical multiphysics field data.

[0044] The co-aging characteristic calculation module is used to calculate the co-aging influencing factor based on the preprocessed historical multiphysics data.

[0045] The dataset building module is used to build datasets based on historical multiphysics data, co-aging influencing factors, and remaining motor life.

[0046] The model building module is used to build a probabilistic deep learning model, train the probabilistic deep learning model using a dataset, and obtain a motor fatigue life distribution prediction model.

[0047] The prediction module is used to acquire multi-physics field data during the operation of the motor of the target pure electric vehicle, calculate the synergistic aging influencing factor, and input it into the motor fatigue life distribution prediction model to obtain the motor fatigue life distribution prediction result.

[0048] The output module is used to sample the predicted fatigue life distribution of the motor multiple times based on the Monte Carlo dropout method to obtain the motor fatigue life assessment result.

[0049] The embodiments of the present invention have the following technical effects:

[0050] This invention quantifies the interaction between electromagnetic, temperature, and stress fields by constructing electromagnetic-thermal coupling factors, thermo-mechanical coupling factors, and multi-field synergistic effect factors. Combining real-time sensor data with material physical properties, it transforms the multi-physics synergistic aging mechanism into a calculable dynamic index, addressing the problem of traditional methods neglecting cross-field coupling effects and significantly improving the characterization ability of lifetime degradation mechanisms. A dual-branch neural network structure is employed, with the Transformer branch extracting the long-term dependencies of multi-physics temporal features; the fully connected branch fuses the synergistic aging factor, and a Dropout layer is introduced to quantify cognitive uncertainty; the output layer synchronously generates the mean and variance, and the negative log-likelihood loss function is used to jointly optimize prediction accuracy and uncertainty calibration, enabling the model to simultaneously output lifetime point estimates and confidence levels, overcoming the limitations of traditional point prediction. Based on the Monte Carlo dropout method, the same sample is sampled multiple times, calculating random uncertainty to reflect inherent data noise, calculating cognitive uncertainty to reveal the reliability of model parameters, and generating confidence intervals to guide risk decision-making, clearly distinguishing the physical sources of uncertainty and providing multi-dimensional basis for maintenance strategies. Attached Figure Description

[0051] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating a method for assessing the fatigue life of a pure electric vehicle motor, as provided in an embodiment of the present invention.

[0053] Figure 2 This is a comparative diagram showing the fatigue life prediction results of pure electric vehicle motors under different input data provided in the embodiments of the present invention;

[0054] Figure 3 This is a schematic diagram comparing the fatigue life prediction results of the pure electric vehicle motor under the traditional solution provided in this embodiment of the invention;

[0055] Figure 4 This is a schematic diagram of the structure of a system for assessing the fatigue life of a pure electric vehicle motor provided in an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0057] This invention provides a method for assessing the fatigue life of a pure electric vehicle motor. Figure 1 This is a flowchart illustrating a method for assessing the fatigue life of a pure electric vehicle motor according to an embodiment of the present invention. (See attached diagram.) Figure 1 The method includes the following steps:

[0058] S1. Obtain historical multiphysics data and remaining lifespan of the motor during the operation of the pure electric vehicle, and preprocess the historical multiphysics data.

[0059] The multiphysics data includes electromagnetic field data, temperature field data, and stress field data.

[0060] Electromagnetic field data primarily reflects the dynamic changes in electrical parameters such as current, voltage, and magnetic field strength during motor operation. These parameters are directly related to the motor's output torque and power characteristics, and also determine the mechanism of internal losses. High-precision current sensors and voltage acquisition modules are used to record the time series of phase current and phase voltage in real time. Combined with motor structural parameters, key indicators such as air gap magnetic flux density distribution and magnetic field frequency are inverted, providing a foundation for subsequent loss calculations.

[0061] Temperature field data focuses on the temperature evolution of key motor components, particularly the stator windings, core, and permanent magnets. During operation, copper and iron losses are converted into heat, causing localized temperature rises. The temperature level and rate of change directly affect the aging rate of insulation materials and the risk of demagnetization of permanent magnets. By placing multiple temperature sensors inside the motor or combining them with thermal imaging technology to acquire surface temperature distribution, and using heat conduction models to calculate internal hotspot temperatures, precise monitoring of the temperature field is achieved.

[0062] Stress field data is used to characterize the stress state of a motor structure under the combined effects of electromagnetic force, thermal expansion, and mechanical vibration. The electromagnetic attraction between the stator and rotor, the thermal stress caused by temperature gradients, and the centrifugal force of rotating components all generate complex stress distributions within the material. Long-term alternating stress can easily lead to fatigue cracks. Stress amplitude, stress gradient, and vibration response information at key locations are obtained using strain gauges, accelerometers, or finite element simulation. These three types of physical field data are collected synchronously over time to ensure that the dynamic coupling relationships between the various physical quantities are preserved.

[0063] These data collectively constitute a multi-dimensional portrait of the motor's operating state, comprehensively reflecting the entire process of its internal energy conversion, heat accumulation, and mechanical deformation. Compared to evaluation methods that rely solely on a single type of signal, the joint analysis of electromagnetic, thermal, and mechanical field data significantly enhances the ability to characterize the complex aging mechanisms within the motor. Especially in typical pure electric vehicle usage scenarios such as variable operating conditions, frequent start-stop cycles, or high-load operation, the transient interactions of multiple physical fields are more intense, making it difficult to accurately capture the accumulation process of fatigue damage using only a single signal. By integrating this three types of physical field information, not only is the richness and representativeness of the data input improved, but comprehensive data support is also provided for the subsequent calculation of synergistic aging influencing factors, thus laying a solid foundation for building a high-precision, robust life prediction model.

[0064] S2. Calculate the synergistic aging influencing factor based on the preprocessed historical multiphysics data.

[0065] During motor operation, the various physical fields do not act independently, but interact through complex energy conversion and material response mechanisms, accelerating the degradation of material properties. In some embodiments, synergistic aging influencing factors include electromagnetic-thermal coupling influencing factors, thermo-mechanical coupling influencing factors, and multi-field synergistic effect factors.

[0066] In some embodiments, the electromagnetic-thermal coupling influence factor is calculated using the following formula:

[0067] ;

[0068] ;

[0069] ;

[0070] Among them, F em-th C represents the electromagnetic-thermal coupling influence factor. diss F represents the actual heat dissipation capacity of the system. cu F represents the copper loss factor. fe Represents the iron loss factor, I represents the effective value of the current, and R0 represents the phase resistance at the reference temperature. This indicates the temperature coefficient of copper resistance, where T represents the actual winding temperature and T0 represents the reference temperature. This represents the hysteresis loss coefficient, a constant related to the material properties of silicon steel sheets. The eddy current loss coefficient is a constant related to the material properties of silicon steel sheets and the thickness of the laminate. This represents the stray loss coefficient, a constant characterizing other additional iron losses. B represents the frequency of the magnetic field. m This represents the magnetic flux density amplitude, which is the peak value of the alternating magnetic flux density. x represents the Steinmetz coefficient, an empirical exponent, which defaults to 2. Indicates rated copper loss. The rated iron loss can be found directly in the motor's design documents, technical specifications, or type test reports.

[0071] The electromagnetic-thermal coupling influence factor characterizes how changes in current and magnetic field induce temperature rise through loss mechanisms, thereby affecting the thermal stability of materials. Copper and iron losses within the motor are the primary heat sources, their magnitude closely related to current intensity, magnetic flux density amplitude, and frequency. When the current increases or the magnetic field frequency rises, the heating of the winding resistance and the eddy current and hysteresis losses in the core increase significantly, leading to a rise in winding and core temperatures. This temperature increase, in turn, alters the resistivity and permeability of the material, further affecting electromagnetic performance and forming a positive feedback loop. This factor, by comprehensively considering parameters such as the effective value of the current, winding temperature, reference resistance and its temperature coefficient, and hysteresis and eddy current loss coefficients, establishes a quantitative index that dynamically reflects the interaction between electromagnetic and thermal processes.

[0072] In some embodiments, the formula for calculating the thermo-mechanical coupling influence factor is as follows:

[0073] ;

[0074] Among them, F th-st This represents the thermo-mechanical coupling influence factor, and E represents the elastic modulus. Indicates the coefficient of thermal expansion. This represents the temperature gradient, and v represents Poisson's ratio. Indicates yield strength.

[0075] The thermo-mechanical coupling influencing factor focuses on how temperature changes induce mechanical stress within the motor structure. Due to differences in the thermal expansion coefficients of materials in various motor components, the degree of expansion varies across different regions under a non-uniform temperature field, leading to thermal stress within the structure. Furthermore, the greater the temperature gradient, the more significant the thermal stress, especially during transient conditions such as startup and rapid acceleration. Simultaneously, the elastic modulus and yield strength of the materials change with temperature; at high temperatures, material stiffness decreases, and resistance to deformation weakens, further exacerbating the risk of structural fatigue. This factor constructs a coupled metric reflecting the relationship between thermal deformation and structural strength changes by introducing mechanical and thermal parameters such as elastic modulus, coefficient of thermal expansion, temperature gradient, Poisson's ratio, and yield strength.

[0076] In some embodiments, the multi-field synergistic effect factor is calculated using the following formula:

[0077] ;

[0078] Among them, F multi Indicates a multi-field synergistic effect factor. , , This represents the weighting coefficient, which is adjusted based on the actual payment request. The maximum stress can be represented by the finite element method, which simulates the stress distribution of the motor under different operating conditions to obtain the maximum stress value. Indicates the vibration amplitude. This indicates the permissible vibration amplitude.

[0079] The multi-field synergistic effect factor further integrates the combined effects of electromagnetic, thermal, and mechanical forces, building upon the first two types of coupling factors. Motor fatigue damage is often not caused by a single factor, but rather by the superposition and interaction of multiple physical fields in time and space. For example, high current induces high temperature, which in turn softens the material. The softened material is more prone to plastic deformation and crack propagation under electromagnetic forces and vibration loads. This factor introduces dynamic mechanical response parameters such as maximum stress, vibration amplitude, and their allowable thresholds, and combines them with the first two types of factors to construct a global synergistic effect index using a weighted fusion approach.

[0080] S3. Construct a dataset based on historical multiphysics data, synergistic aging influencing factors, and remaining motor life.

[0081] S4. Construct a probabilistic deep learning model, train the probabilistic deep learning model using the dataset, and obtain a motor fatigue life distribution prediction model.

[0082] In some embodiments, the probabilistic deep learning model includes:

[0083] The system consists of a first input branch, a second input branch, a feature concatenation layer, a fully connected layer, a Dropout layer, and an output layer.

[0084] The first input branch includes a Transformer encoder and a multi-head attention mechanism for receiving and processing multiphysics data. The Transformer encoder captures long-term dependencies in the data without relying on recurrent or convolutional operations. The multi-head attention mechanism further enhances the model's understanding of the interactions between different features. When multiphysics data is input into this branch, the Transformer encoder encodes it, extracting high-level feature representations, while the multi-head attention mechanism focuses on the most critical parts of the data, ensuring that the model can fully capture the complex relationships between various physics.

[0085] The second input branch includes a fully connected layer and a Dropout layer, used to receive and process co-aging influencing factors. The fully connected layer performs linear transformations and non-linear activations on the input data to extract aging-related features. The Dropout layer is used for regularization, randomly dropping a portion of neurons to prevent overfitting and improve the model's generalization ability. When co-aging influencing factors are input into this branch, the fully connected layer performs preliminary processing to extract key features, while the Dropout layer introduces randomness during training to enhance the model's robustness. For example, the dropout probability of the Dropout layer can be set to 0.2-0.3.

[0086] The feature concatenation layer fuses the features extracted from the first and second input branches, integrating information from multiphysics data and co-aging influencing factors, and further enriching the model's expressive power through feature interaction. The data processed by the feature concatenation layer contains a more comprehensive and in-depth feature representation.

[0087] Fully connected layers and Dropout layers continue to process the concatenated features. Fully connected layers, through multi-level linear transformations and non-linear activations, progressively extract higher-level abstract features that more accurately reflect the relevant patterns of motor fatigue life. Dropout layers continue to play a role during training, randomly discarding some neurons to avoid overfitting and ensure the model maintains good predictive performance when faced with new data. For example, the dropout probability of the Dropout layer can be set to 0.2-0.3.

[0088] The output layer contains two neurons, which output the predicted motor fatigue life and the degree of uncertainty, respectively. The mean (μ) of the output distribution of the first neuron represents the most likely predicted motor fatigue life, and the variance (σ) of the output distribution of the second neuron represents the degree of uncertainty.2 The variance represents the perceived uncertainty of the model; a larger variance indicates greater uncertainty. These two neurons calculate the final prediction result through forward propagation. The fatigue life prediction directly reflects the expected service life of the motor under current operating conditions. The degree of uncertainty quantifies the reliability of the prediction result.

[0089] For example, the training process of the probabilistic deep learning model uses the Adam optimizer, with a learning rate of 0.001, a batch size of 32, and a training period of 100.

[0090] Furthermore, the loss function for the probabilistic deep learning model adopts the negative log-likelihood loss function, as expressed below:

[0091] ;

[0092] Where L represents the negative log-likelihood loss function, N represents the total number of training samples, i represents the i-th training sample, and y i μ represents the true remaining lifetime of the i-th training sample. i σ represents the mean of the remaining life distribution predicted by the model for the i-th training sample, i.e., the predicted fatigue life of the motor. i This represents the standard deviation of the remaining lifetime distribution predicted by the model for the i-th training sample, i.e., the degree of uncertainty.

[0093] The probabilistic deep learning model in this scheme uses the negative log-likelihood loss function as the optimization objective. This not only effectively measures the difference between the model's predicted values ​​and the true values, but also takes into account the degree of uncertainty of the prediction results, thereby improving the overall performance of the model.

[0094] In the formula, the first part This involves measuring the deviation between predicted and actual values. When the model's predicted value is close to the actual value, the loss value decreases; conversely, if the predicted value deviates significantly from the actual value, the loss value increases significantly. Furthermore, since the denominator is the square of the prediction standard deviation, when the model estimates a high level of uncertainty for a particular sample, even if the predicted value deviates significantly from the actual value, the loss value will not increase drastically, reflecting the model's tolerance for uncertainty.

[0095] In the formula, the second part This is a direct penalty term for the prediction standard deviation. By taking the logarithm of the prediction standard deviation and multiplying it by a constant coefficient, it aims to encourage the model to minimize the uncertainty of the prediction results while maintaining prediction accuracy. When the model estimates the uncertainty of a prediction result to be low, the loss value of this term will also decrease accordingly, thereby prompting the model to continuously optimize its prediction ability during training.

[0096] Using the negative log-likelihood loss function can comprehensively measure the accuracy and reliability of the model's prediction results, avoiding the bias that a single indicator may bring. Secondly, by introducing uncertainty estimation, the model can better cope with various complex situations and uncertainties in real-world applications, improving the robustness and credibility of the prediction results.

[0097] S5. Obtain multi-physics field data of the motor operation process of the target pure electric vehicle, calculate the synergistic aging influencing factor, and input it into the motor fatigue life distribution prediction model to obtain the motor fatigue life distribution prediction result.

[0098] The process of acquiring multiphysics field data during the operation of the target pure electric vehicle's motor, calculating the synergistic aging influencing factor, and inputting this data into the motor fatigue life distribution prediction model is the same as the training process and will not be repeated here. The motor fatigue life distribution prediction model can output the predicted motor fatigue life distribution results. .

[0099] S6. Based on the Monte Carlo dropout method, the predicted results of motor fatigue life distribution are sampled multiple times to obtain the motor fatigue life assessment results.

[0100] In some embodiments, S6 specifically includes the following sub-steps:

[0101] S61. Based on the Monte Carlo dropout method, the predicted results of motor fatigue life distribution are sampled n times to obtain n sets of motor fatigue life distribution sampling results.

[0102] For example, n sets of motor fatigue life distribution sampling results are obtained: {(μ1,,σ1) 2 ),(μ2,σ2 2 ),...,(μ n ,σ n 2 )}.

[0103] S62. Calculate the mean value of the predicted motor fatigue life from the sampling results of n groups of motor fatigue life distribution, and use it as the predicted motor fatigue life.

[0104] ;

[0105] in, This represents the predicted fatigue life of the motor, where j represents the sampling result of the fatigue life distribution of the j-th group of motors. This represents the predicted fatigue life value of the motor in the sampling results of the motor fatigue life distribution in the j-th group.

[0106] S63. Calculate the random uncertainty and cognitive uncertainty based on the sampling results of the fatigue life distribution of n groups of motors.

[0107] In some embodiments, the formula for calculating random uncertainty is as follows:

[0108] ;

[0109] The formula for calculating cognitive uncertainty is as follows:

[0110] ;

[0111] in, Indicates randomness and uncertainty. This indicates cognitive uncertainty. This represents the degree of uncertainty in the sampling results of the motor fatigue life distribution in the j-th group. Random uncertainty stems from inherent noise in the data, such as sensor measurement noise and random fluctuations in operating conditions. It is an inherent property of the data and cannot be reduced by adding more data. Cognitive uncertainty arises from the model's own ignorance or uncertainty. This is because the training data is limited, and the model cannot learn all possible operating conditions. It is an inherent property of the model and can be reduced by adding more relevant training data.

[0112] S64. Calculate the confidence interval of the motor fatigue life prediction results based on accidental uncertainty and cognitive uncertainty.

[0113] In some embodiments, a 95% confidence interval may be used:

[0114] ;

[0115] ;

[0116] in, This represents the 95% confidence interval for the predicted fatigue life of the motor. This indicates total uncertainty.

[0117] S65. The final output includes the motor fatigue life prediction results, confidence interval, random uncertainty, and cognitive uncertainty, which are the motor fatigue life assessment results.

[0118] By providing a comprehensive output incorporating four-dimensional information, predictions are transformed from simple numbers into a set of information supporting risk-based decision-making. Confidence intervals provide a safe boundary for maintenance, avoiding both the waste of resources caused by premature maintenance and the sudden failures resulting from over-reliance on point estimates.

[0119] By separating random uncertainty from cognitive uncertainty, we can guide subsequent actions and provide accurate decision-making basis:

[0120] These two types of uncertainty indicate completely different directions for optimization and coping strategies.

[0121] High cognitive uncertainty indicates insufficient knowledge within the model itself. For example, cognitive uncertainty increases significantly when the motor operates under extreme conditions never covered in the training data (such as a combination of extremely high torque and extremely low ambient temperature). When such a situation is observed in the motor fatigue life assessment results, the system can label the sample for focused attention or include it in subsequent data collection plans for iterative model training, guiding model improvement.

[0122] High random uncertainty indicates that the data itself is noisy or that the phenomenon has strong inherent randomness. For example, even under common operating conditions, predictions will have inherent, unavoidable fluctuations due to sensor measurement noise or instantaneous fluctuations in motor load. When this is shown in motor fatigue life assessment results, it means that even if the model is good enough, predictions for such inputs inherently have a large range of fluctuations. In this case, maintenance strategies should be adjusted; for example, a wider buffer zone should be reserved when developing maintenance plans.

[0123] This type of analysis allows for targeted optimization of models and datasets, rather than blindly adjusting parameters. If only a general uncertainty is given, it becomes impossible to determine whether to trust the model (increase data) or accept reality (expand the safety margin), leading to a vague course of action.

[0124] Figure 2 This is a comparative diagram showing the fatigue life prediction results of pure electric vehicle motors under different input data provided in this embodiment of the invention. See also... Figure 2 Comparing the two prediction curves, it can be seen that the prediction result combining multiphysics data and the synergistic aging influencing factor (red dashed line) shows a higher degree of agreement with the actual remaining life of the motor (blue solid line) across the entire operating mileage range. In contrast, the prediction result using only multiphysics data (green dotted line) shows a significant deviation starting from the middle of operation, and this deviation gradually widens with increasing mileage. This indicates that the synergistic aging influencing factor effectively captures the accelerating effect of multiphysics coupling on motor aging and is a key factor in improving prediction accuracy. While the prediction method using only multiphysics data can roughly reflect the lifespan decline trend, it cannot accurately describe the changes in the aging rate, especially exhibiting systematic errors after high mileage. The present invention, by introducing the synergistic aging influencing factor and fusing it with monitoring data, fully utilizes data information and introduces physical constraints, achieving accurate modeling of the motor degradation process, significantly reducing prediction errors, and making the prediction results more consistent with the actual physical process development.

[0125] Figure 3 This is a comparative diagram of the fatigue life prediction results of pure electric vehicle motors under the present invention and the traditional solution, as shown in the embodiment of the invention. Figure 3The predicted curve (red dashed line) of this solution shows a high degree of agreement with the actual remaining life curve of the motor (blue solid line) throughout the entire operating mileage. In contrast, while the predicted curve of the traditional LSTM model (yellow dotted line) can roughly follow the degradation trend, it exhibits significant deviations at several stages (e.g., in the middle of operation). The traditional LSTM model relies solely on the time-series characteristics of historical monitoring data and fails to delve into the physical mechanisms of motor aging. Its prediction bias indicates that purely data-driven methods struggle to accurately capture the complex impact of electromagnetic-thermal-mechanical multi-field coupling on the aging process. The prediction results of the traditional LSTM model are relatively volatile and may have significant errors at key decision points (such as when the remaining life drops to near the threshold), which introduces uncertainty risks to predictive maintenance decisions. In contrast, the prediction results of this invention not only have higher overall accuracy but also exhibit a smoother and more stable trajectory, providing a more reliable and credible basis for maintenance planning and significantly reducing the operational risks caused by misjudgments.

[0126] This invention quantifies the interaction between electromagnetic, temperature, and stress fields by constructing electromagnetic-thermal coupling factors, thermo-mechanical coupling factors, and multi-field synergistic effect factors. Combining real-time sensor data with material physical properties, it transforms the multi-physics synergistic aging mechanism into a calculable dynamic index, addressing the problem of traditional methods neglecting cross-field coupling effects and significantly improving the characterization ability of lifetime degradation mechanisms. A dual-branch neural network structure is employed, with the Transformer branch extracting the long-term dependencies of multi-physics temporal features; the fully connected branch fuses the synergistic aging factor, and a Dropout layer is introduced to quantify cognitive uncertainty; the output layer synchronously generates the mean and variance, and the negative log-likelihood loss function is used to jointly optimize prediction accuracy and uncertainty calibration, enabling the model to simultaneously output lifetime point estimates and confidence levels, overcoming the limitations of traditional point prediction. Based on the Monte Carlo dropout method, the same sample is sampled multiple times, calculating random uncertainty to reflect inherent data noise, calculating cognitive uncertainty to reveal the reliability of model parameters, and generating confidence intervals to guide risk decision-making, clearly distinguishing the physical sources of uncertainty and providing multi-dimensional basis for maintenance strategies.

[0127] This invention also provides a fatigue life assessment system for a pure electric vehicle motor, used to perform the aforementioned fatigue life assessment method for a pure electric vehicle motor. Figure 4 This is a schematic diagram of the structure of a fatigue life assessment system for a pure electric vehicle motor provided in an embodiment of the present invention. See also: Figure 4 The system includes the following modules:

[0128] The data acquisition module is used to acquire historical multiphysics field data and remaining lifespan of the motor during the operation of the pure electric vehicle, and to preprocess the historical multiphysics field data.

[0129] The co-aging characteristic calculation module is used to calculate the co-aging influencing factor based on the preprocessed historical multiphysics data.

[0130] The dataset building module is used to build datasets based on historical multiphysics data, co-aging influencing factors, and remaining motor life.

[0131] The model building module is used to build a probabilistic deep learning model, train the probabilistic deep learning model using a dataset, and obtain a motor fatigue life distribution prediction model.

[0132] The prediction module is used to acquire multi-physics field data during the operation of the motor of the target pure electric vehicle, calculate the synergistic aging influencing factor, and input it into the motor fatigue life distribution prediction model to obtain the motor fatigue life distribution prediction result.

[0133] The output module is used to sample the predicted fatigue life distribution of the motor multiple times based on the Monte Carlo dropout method to obtain the motor fatigue life assessment result.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the fatigue life of a pure electric vehicle motor, characterized in that, The method includes the following steps: S1. Obtain historical multiphysics field data and remaining lifespan of the motor during the operation of the pure electric vehicle, and preprocess the historical multiphysics field data. S2. Calculate the synergistic aging influencing factor based on the preprocessed historical multiphysics data; S3. Construct a dataset based on the historical multiphysics data, the synergistic aging influencing factor, and the remaining lifespan of the motor; S4. Construct a probabilistic deep learning model, and train the probabilistic deep learning model using the dataset to obtain a motor fatigue life distribution prediction model. S5. Obtain multi-physics field data of the motor of the target pure electric vehicle during operation, calculate the synergistic aging influencing factor, and input it into the motor fatigue life distribution prediction model to obtain the motor fatigue life distribution prediction result. S6. Based on the Monte Carlo dropout method, the predicted fatigue life distribution of the motor is sampled multiple times to obtain the motor fatigue life assessment result.

2. The fatigue life assessment method for a pure electric vehicle motor according to claim 1, characterized in that, In S1, the multiphysics data includes electromagnetic field data, temperature field data, and stress field data.

3. The fatigue life assessment method for a pure electric vehicle motor according to claim 2, characterized in that, In S2, the synergistic aging influencing factors include electromagnetic-thermal coupling influencing factors, thermo-mechanical coupling influencing factors, and multi-field synergistic effect factors.

4. The fatigue life assessment method for a pure electric vehicle motor according to claim 3, characterized in that, The electromagnetic-thermal coupling influence factor is calculated using the following formula: ； ； ； Among them, F em-th C represents the electromagnetic-thermal coupling influence factor. diss F represents the actual heat dissipation capacity of the system. cu F represents the copper loss factor. fe Represents the iron loss factor, I represents the effective value of the current, and R0 represents the phase resistance at the reference temperature. This indicates the temperature coefficient of copper resistance, where T represents the actual winding temperature and T0 represents the reference temperature. Indicates the hysteresis loss coefficient. This represents the eddy current loss coefficient. Represents the stray loss coefficient. B represents the frequency of the magnetic field. m Indicates the magnetic flux density amplitude, and x represents the Steinmetz coefficient. Indicates rated copper loss. This indicates the rated iron loss.

5. The fatigue life assessment method for a pure electric vehicle motor according to claim 4, characterized in that, The thermo-mechanical coupling influence factor is calculated using the following formula: ； Among them, F th-st This represents the thermo-mechanical coupling influence factor, and E represents the elastic modulus. Indicates the coefficient of thermal expansion. This represents the temperature gradient, and v represents Poisson's ratio. Indicates yield strength.

6. The fatigue life assessment method for a pure electric vehicle motor according to claim 5, characterized in that, The multi-field synergistic effect factor is calculated using the following formula: ； Among them, F multi Indicates a multi-field synergistic effect factor. , , Indicates the weighting coefficient. Indicates the maximum stress. Indicates the vibration amplitude. This indicates the permissible vibration amplitude.

7. The fatigue life assessment method for a pure electric vehicle motor according to claim 1, characterized in that, In S4, the probabilistic deep learning model includes: The system consists of a first input branch, a second input branch, a feature concatenation layer, a fully connected layer, a Dropout layer, and an output layer. The first input branch includes a Transformer encoder and a multi-head attention mechanism for receiving the multiphysics data. The second input branch includes a fully connected layer and a Dropout layer for receiving the collaborative aging influence factor; The output layer contains two neurons, which output the predicted value of motor fatigue life and the degree of uncertainty, respectively.

8. The fatigue life assessment method for a pure electric vehicle motor according to claim 7, characterized in that, In step S4, the loss function of the probabilistic deep learning model adopts the negative log-likelihood loss function, as expressed below: ； Where L represents the negative log-likelihood loss function, N represents the total number of training samples, i represents the i-th training sample, and y i μ represents the true remaining lifetime of the i-th training sample. i σ represents the mean of the remaining life distribution predicted by the model for the i-th training sample, i.e., the predicted fatigue life of the motor. i This represents the standard deviation of the remaining lifetime distribution predicted by the model for the i-th training sample, i.e., the degree of uncertainty.

9. The fatigue life assessment method for a pure electric vehicle motor according to claim 1, characterized in that, In step S6, the predicted fatigue life distribution of the motor is sampled multiple times based on the Monte Carlo dropout method to obtain the motor fatigue life assessment results, including: S61. Based on the Monte Carlo dropout method, the predicted motor fatigue life distribution is sampled n times to obtain n sets of motor fatigue life distribution sampling results. S62. Calculate the mean value of the predicted motor fatigue life from the sampling results of n groups of motor fatigue life distribution, and use it as the predicted motor fatigue life result. S63. Calculate the random uncertainty and cognitive uncertainty based on the sampling results of the fatigue life distribution of n groups of motors respectively; S64. Calculate the confidence interval of the motor fatigue life prediction results based on accidental uncertainty and cognitive uncertainty; S65. The final output includes the motor fatigue life prediction results, confidence interval, random uncertainty, and cognitive uncertainty, which are the motor fatigue life assessment results.

10. A fatigue life assessment system for a pure electric vehicle motor, used to execute the fatigue life assessment method for a pure electric vehicle motor according to any one of claims 1-9, characterized in that, The system includes the following modules: The data acquisition module is used to acquire historical multiphysics field data and remaining lifespan of the motor during the operation of the pure electric vehicle, and to preprocess the historical multiphysics field data. The co-aging characteristic calculation module is used to calculate the co-aging influencing factor based on the preprocessed historical multiphysics data. A dataset construction module is used to construct a dataset based on the historical multiphysics data, the collaborative aging influencing factor, and the remaining lifespan of the motor. The model building module is used to build a probabilistic deep learning model, and to train the probabilistic deep learning model using a dataset to obtain a motor fatigue life distribution prediction model. The prediction module is used to acquire multi-physics field data during the operation of the motor of the target pure electric vehicle, calculate the synergistic aging influencing factor, and input it into the motor fatigue life distribution prediction model to obtain the motor fatigue life distribution prediction result. The output module is used to sample the predicted fatigue life distribution of the motor multiple times based on the Monte Carlo dropout method to obtain the motor fatigue life assessment result.

Images