Fault diagnosis method of electric drive system, fault diagnosis system and readable storage medium

Abstract

The invention discloses a fault diagnosis method and system of an electric drive system and a readable storage medium. The fault diagnosis method comprises the following steps: acquiring multi-dimensional data in real time; wherein the multi-dimensional data comprises at least two of water inlet temperature, controller temperature, voltage, current, rotating speed and torque; motor temperature prediction is carried out by using the target prediction model in combination with the multi-dimensional data, and corresponding multi-dimensional data segments when the predicted motor temperature is greater than a preset temperature threshold are obtained; performing feature engineering processing on the multi-dimensional data fragments by using a singular value decomposition technology to obtain a target data matrix; performing clustering analysis on the target data matrix by using a Gaussian mixture model to obtain a classification result; and obtaining a fault diagnosis result according to the classification result. In this way, the real-time performance, accuracy and reliability of fault detection of the electric drive system can be improved.

Description

Technical Field

[0001] This application relates to the technical field of fault diagnosis, and in particular to a fault diagnosis method, a fault diagnosis system, and a readable storage medium for an electric drive system. Background Art

[0002] With the rapid development of new energy vehicles, the reliability and safety of the electric drive system of new energy vehicles have become key factors to ensure the normal operation of vehicles. During actual driving, various abnormal operating conditions may occur in the electric drive system, and the most typical one is the cooling system failure, such as water leakage, oil leakage, etc. Usually, complex cooling systems are easily affected by various factors during operation, resulting in failures.

[0003] Related vehicle monitoring and maintenance means mainly rely on regular manual inspections and simple sensor alarms, and these methods have obvious limitations. Summary of the Invention

[0004] This application provides a fault diagnosis method, a fault diagnosis system, and a readable storage medium for an electric drive system, which can improve the real-time performance, accuracy, and reliability of electric drive system fault detection.

[0005] In a first aspect, this application provides a fault diagnosis method for an electric drive system, and the fault diagnosis method includes: obtaining multi-dimensional data in real time; where the multi-dimensional data includes at least two of the inlet temperature, controller temperature, voltage, current, rotational speed, and torque; using a target prediction model to combine the multi-dimensional data for motor temperature prediction, and obtaining a multi-dimensional data segment corresponding to when the predicted motor temperature is greater than a preset temperature threshold; using the singular value decomposition technique to perform feature engineering processing on the multi-dimensional data segment to obtain a target data matrix; using a Gaussian mixture model to perform clustering analysis on the target data matrix to obtain a classification result; and obtaining a fault diagnosis result according to the classification result.

[0006] Among them, using the singular value decomposition technique to perform feature engineering processing on the multi-dimensional data segment to obtain a target data matrix includes: performing SVD (Singular Value Decomposition) solution on the multi-dimensional data segment to obtain a left singular vector, a right singular vector, and a diagonal matrix containing singular values; obtaining a preset number of target singular values, and the left singular vector and right singular vector corresponding to each target singular value for matrix recombination to obtain a target data matrix; where the preset number of target singular values is greater than the remaining singular values among all singular values.

[0007] Among them, the Gaussian mixture model is used to perform clustering analysis on the target data matrix to obtain a classification result, including: performing clustering analysis on the target data matrix using the Gaussian mixture model and classifying the target data matrix into corresponding data clusters; and recording the category code of the data cluster as the classification result.

[0008] Among them, a fault diagnosis result is obtained according to the classification result, including: counting the number of multi-dimensional data segments corresponding to each category; according to the number of multi-dimensional data segments corresponding to each category, counting the proportion of the number corresponding to each category; and obtaining the fault diagnosis result according to the proportion of the number.

[0009] Among them, the categories include a normal category and a fault category; obtaining the fault diagnosis result according to the proportion of the number includes: in response to the proportion of the number of the normal category being greater than a first threshold, obtaining a first fault diagnosis result; the first fault diagnosis result indicates that the electric drive system is normal; in response to the proportion of the number of the fault category being greater than the first threshold, obtaining a second fault diagnosis result; the second fault diagnosis result indicates that the electric drive system has a fault; in response to the proportion of the number of the normal category and the fault category being less than the first threshold, obtaining a third fault diagnosis result; the first fault diagnosis result indicates that the electric drive system needs further diagnosis.

[0010] Among them, after obtaining the fault diagnosis result according to the classification result, it includes: determining the vehicles for which the fault diagnosis result indicates that the electric drive system has a fault; and performing a fault prompt according to the vehicle information.

[0011] Among them, obtaining the multi-dimensional data segment corresponding to the predicted motor temperature being greater than a preset temperature threshold includes: when the predicted motor temperature is greater than the preset temperature threshold, intercepting the multi-dimensional data of N time lengths after the multi-dimensional data used for prediction as the multi-dimensional data segment.

[0012] Among them, using the target prediction model to combine multi-dimensional data for motor temperature prediction includes: performing data cleaning on the multi-dimensional data; and using the target prediction model to combine the cleaned multi-dimensional data for motor temperature prediction.

[0013] In a second aspect, the present application provides a fault diagnosis system for an electric drive system. The fault diagnosis system includes: a communication interface, a memory, and a processor coupled to the memory and the communication interface. The memory stores at least one computer program. When at least one computer program is loaded and executed by the processor, it is used to implement the method provided in the first aspect.

[0014] In a third aspect, the present application provides a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement the method provided in the first aspect.

[0015] The beneficial effects of the present application are as follows: Different from the prior art, the fault diagnosis method, fault diagnosis system, and readable storage medium of the electric drive system provided by the present application obtain multi-dimensional data in real time to ensure the timeliness and accuracy of the data, and the multi-dimensional data can comprehensively reflect the operating state of the electric drive system. Furthermore, a target prediction model is used to combine the multi-dimensional data for motor temperature prediction, and the multi-dimensional data segment corresponding to the predicted motor temperature greater than the preset temperature threshold is obtained, so as to screen out in advance the real-time stream segments that can be used for fault diagnosis. Starting from the high-temperature data, the resolution ability of the fault diagnosis model for abnormal data patterns is further improved. The singular value decomposition technology is further used to perform feature engineering processing on the multi-dimensional data segment to obtain the target data matrix, extract key features from the multi-dimensional data segment, reorganize the data, and extract the important information that best reflects the change relationship between time and variables. It can combine the automatic feature selection method to reduce the subjectivity of feature selection, improve the objectivity and accuracy of feature extraction, and through an efficient feature extraction method, reorganize a new type of data that can better reflect the data change characteristics and differences, reduce the blindness of calculation, and improve the processing efficiency. Further, a Gaussian mixture model is used to perform clustering analysis on the target data matrix to obtain a classification result; the fault diagnosis result is obtained according to the classification result, which can improve the real-time performance, accuracy, and reliability of the electric drive system fault detection, thus ensuring the safe operation of new energy vehicles and improving the travel experience of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0017] Figure 1 is a flowchart of an embodiment of the fault diagnosis method of the electric drive system provided by the present application;

[0018] Figure 2 is a flowchart of another embodiment of the fault diagnosis method of the electric drive system provided by the present application;

[0019] Figure 3 is a flowchart of another embodiment of the fault diagnosis method of the electric drive system provided by the present application;

[0020] Figure 4 is a structural diagram of an embodiment of the fault diagnosis system of the electric drive system provided by the present application;

[0021] Figure 5 is a structural diagram of an embodiment of the computer-readable storage medium provided by the present application. Detailed implementation manners

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. It can be understood that the specific embodiments described herein are only used to explain the present application, rather than limiting the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application rather than all the structures are shown in the drawings. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0023] Referring to "embodiments" herein means that specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the present application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0024] With the rapid development of new energy vehicles, the reliability and safety of the electric drive system of new energy vehicles have become key factors to ensure the normal operation of the vehicles. During actual driving, various abnormal operating conditions may occur in the electric drive system, and the most typical one is the cooling system failure, such as problems like water leakage and oil leakage. Usually, the complex cooling system is prone to be affected by various factors during operation, resulting in failures.

[0025] The related vehicle monitoring and maintenance means mainly rely on regular manual inspections and simple sensor alarms, and these methods have obvious limitations.

[0026] Based on this, the present application proposes to ensure the timeliness and accuracy of data by obtaining multi-dimensional data in real time, and that the multi-dimensional data can comprehensively reflect the operating state of the electric drive system. Furthermore, a target prediction model is used to combine the multi-dimensional data for motor temperature prediction, and the multi-dimensional data segment corresponding to the predicted motor temperature being greater than the preset temperature threshold is obtained. The real-time stream segments that can be used for fault diagnosis are screened in advance, starting from the high-temperature data to further improve the resolution ability of the fault diagnosis model for abnormal data pattern recognition. The singular value decomposition technique is further used to perform feature engineering processing on the multi-dimensional data segment to obtain a target data matrix, extract key features from the multi-dimensional data segment, reorganize the data and extract the important information that best reflects the change relationship between time and variables. The automatic feature selection method can be combined to reduce the subjectivity of feature selection, improve the objectivity and accuracy of feature extraction, and through an efficient feature extraction method, reorganize a new type of data that can better reflect the characteristics and differences of data changes, reduce the blindness of calculation, and improve the processing efficiency. Further, a Gaussian mixture model is used to perform clustering analysis on the target data matrix to obtain a classification result; the fault diagnosis result is obtained according to the classification result, which can improve the real-time performance, accuracy and reliability of the electric drive system fault detection, thereby ensuring the safe operation of new energy vehicles and improving the travel experience of users. For specific reference, see the following embodiments.

[0027] Refer to Figure 1 , Figure 1 is a schematic flowchart of an embodiment of a fault diagnosis method for an electric drive system provided by the present application. The fault diagnosis method includes:

[0028] Step 11: Obtain multi-dimensional data in real time; wherein, the multi-dimensional data includes at least two of inlet temperature, controller temperature, voltage, current, speed, and torque.

[0029] In some embodiments, the multi-dimensional data may include at least three of inlet temperature, controller temperature, voltage, current, speed, and torque.

[0030] In some embodiments, the multi-dimensional data may include at least four of inlet temperature, controller temperature, voltage, current, speed, and torque.

[0031] In some embodiments, the multi-dimensional data may include at least five of inlet temperature, controller temperature, voltage, current, speed, and torque.

[0032] In some embodiments, the multi-dimensional data may include inlet temperature, controller temperature, voltage, current, speed, and torque.

[0033] It can be understood that the inlet temperature, controller temperature, voltage, current, rotational speed, and torque are all related to the electric drive system. The electric drive system generally includes key components such as a motor, a controller, and a reducer. These components generate a large amount of heat during operation. Therefore, an efficient cooling system is the key to ensuring the normal operation of the electric drive system.

[0034] A considerable number of new energy vehicles adopt oil cooling technology to cool the electric drive system. Specifically, the oil fluid flows through the motor and other heat sources through a circulation system. After absorbing heat, it then exchanges heat with cooling water through a heat exchanger, thereby reducing the oil temperature. This cooling method can effectively control the temperature of the electric drive system and ensure its stable operation under various complex working conditions.

[0035] In some embodiments, multi-dimensional data can be associated according to the vehicle. For example, it is bound to the multi-dimensional data according to a unique identifier such as the serial number of each vehicle. That is, the present application can perform fault diagnosis on multiple vehicles simultaneously.

[0036] In some embodiments, the operating data of the electric drive system, including the inlet temperature of the electric drive, controller temperature, rotational speed, torque, current, voltage, etc., can be collected in real time through in-vehicle sensors (such as temperature sensors, torque sensors, etc.). For example, using Internet of Things (IoT) technology, the data is transmitted to a cloud server through a wireless communication module (such as 4G / 5G or Wi-Fi). The data collection frequency can be set according to actual needs to ensure real-time acquisition of vehicle operating data and provide accurate basic data for subsequent analysis.

[0037] Step 12: Use the target prediction model to combine multi-dimensional data for motor temperature prediction, and obtain the multi-dimensional data segment corresponding to when the predicted motor temperature is greater than the preset temperature threshold.

[0038] In some embodiments, the target prediction model can be obtained by pre-training with labeled multi-dimensional data. After training is completed, it can be used for motor temperature prediction of real-time multi-dimensional data.

[0039] In some embodiments, the multi-dimensional data can be data for a period of time. For example, data within 5 minutes, data within 10 minutes, etc. That is, in the case of real-time acquisition, it can form data on a continuous time series.

[0040] In some embodiments, before prediction, the multi-dimensional data is cleaned, and then the target prediction model is used to combine the cleaned multi-dimensional data for motor temperature prediction.

[0041] In some embodiments, when it is predicted that the motor temperature is greater than a preset temperature threshold, multi-dimensional data of N time lengths after the multi-dimensional data used for prediction is intercepted as a multi-dimensional data segment.

[0042] In some embodiments, it is judged whether the motor temperature exceeds a preset temperature threshold. If it exceeds, the next step of classification and summarization is entered; otherwise, it is considered that the current segment is not sufficient to be used as a segment for effectively distinguishing between a faulty and a normal electric drive mode, and it is discarded.

[0043] By means of step 12, real-time stream segments (multi-dimensional data segments) that can be used for fault diagnosis are screened in advance. Starting from the high-temperature data, the resolution ability of the fault diagnosis model for abnormal data patterns is further improved. Further, segments that are not sufficient to effectively distinguish between a faulty and a normal electric drive mode are preliminarily screened out, reducing unnecessary waste of computing resources.

[0044] Step 13: Use the singular value decomposition technique to perform feature engineering processing on the multi-dimensional data segment to obtain a target data matrix.

[0045] In some embodiments, through the singular value decomposition (SVD) of the multi-dimensional data segment, the left singular vector U, the right singular vector V, and the middle sigma singular value matrix are extracted, and the matrix is reorganized to obtain a target data matrix, which can extract information that can better reflect the change relationship between time and feature items in the time series data.

[0046] Step 14: Use the Gaussian mixture model to perform clustering analysis on the target data matrix to obtain a classification result.

[0047] Step 15: Obtain a fault diagnosis result according to the classification result.

[0048] In some embodiments, according to the proportion of the number of classification results in different categories, by comparing the corresponding preset proportion threshold, the state of the vehicle electric drive system is comprehensively judged to be normal, faulty, or doubtful. That is, a corresponding fault diagnosis result can be formed.

[0049] In this embodiment, by obtaining multi-dimensional data in real time, the timeliness and accuracy of the data are ensured, and the multi-dimensional data can comprehensively reflect the operating state of the electric drive system. Furthermore, a target prediction model is used to combine the multi-dimensional data for motor temperature prediction, and the multi-dimensional data segment corresponding to the predicted motor temperature greater than the preset temperature threshold is obtained. The real-time stream segment that can be used for fault diagnosis is screened in advance, and starting from the high-temperature data, the resolution ability of the fault diagnosis model for abnormal data pattern recognition is further improved. Further, the singular value decomposition technique is used to perform feature engineering processing on the multi-dimensional data segment to obtain the target data matrix, extract key features from the multi-dimensional data segment, reorganize the data and extract the important information that best reflects the change relationship between time and variables. The automatic feature selection method can be combined to reduce the subjectivity of feature selection, improve the objectivity and accuracy of feature extraction, and through an efficient feature extraction method, reorganize a new type of data that can better reflect the characteristics and differences of data changes, reduce the blindness of calculation, and improve the processing efficiency. Further, a Gaussian mixture model is used to perform clustering analysis on the target data matrix to obtain a classification result; the fault diagnosis result is obtained according to the classification result, which can improve the real-time performance, accuracy and reliability of the electric drive system fault detection, thus ensuring the safe operation of new energy vehicles and improving the travel experience of users.

[0050] Refer to Figure 2 , Figure 2 is a schematic flowchart of another embodiment of the fault diagnosis method for an electric drive system provided by this application. The fault diagnosis method includes:

[0051] Step 21: Obtain multi-dimensional data in real time; among them, the multi-dimensional data includes at least two of the inlet temperature, controller temperature, voltage, current, speed and torque.

[0052] Step 22: Use the target prediction model to combine the multi-dimensional data for motor temperature prediction, and obtain the multi-dimensional data segment corresponding to the predicted motor temperature greater than the preset temperature threshold.

[0053] Step 23: Perform SVD solution on the multi-dimensional data segment to obtain the left singular vector, right singular vector and diagonal matrix containing singular values.

[0054] Step 24: Obtain a preset number of target singular values, and the left singular vector and right singular vector corresponding to each target singular value for matrix recombination to obtain the target data matrix; among them, the preset number of target singular values is greater than the remaining singular values among all singular values.

[0055] For example, select the first k largest singular values and the corresponding left and right singular vectors to construct a truncated Sigma' matrix (only retain the first k singular values). Then calculate U*Sigma'*V T, where T is the transpose of V. This operation generates a reconstructed low-dimensional data matrix (target data matrix), which contains the main information of the original data.

[0056] In this embodiment, by solving the singular values of multi-dimensional data segments, the information that can reflect the key change relationship between time and feature items is effectively extracted, reducing noise and improving the subsequent clustering effect.

[0057] In other embodiments, Singular Value Decomposition (SVD) is a powerful linear algebra tool for extracting important features in multi-dimensional datasets. SVD is used to extract key features from time series data. Some alternative or complementary methods can be used to enhance the feature engineering ability of SVD: For example, flexible combination of singular values: Conventional SVD methods may use a fixed k value (such as half of the number of feature items) to select the principal components. However, the selection of the k value can be determined based on cross-validation, cumulative explained variance, or model-based performance to achieve a better feature representation.

[0058] For example, multi-level SVD: For complex data structures, multi-level SVD can be considered, that is, applying SVD at different levels to capture the features of the data at different scales.

[0059] For example, combining other feature extraction techniques: In addition to SVD, wavelet transform, Fourier transform, or other signal processing techniques can also be considered to extract the features of the data from different perspectives.

[0060] For example, Dynamic Time Warping (DTW): For time series data, DTW is a method for measuring the similarity of time series and can be combined with SVD to improve the sensitivity to the dynamic changes of time series.

[0061] Step 25: Use the Gaussian mixture model to perform clustering analysis on the target data matrix to obtain the classification result.

[0062] In other embodiments, the Gaussian mixture model (GMM) is a probability-based clustering method that assumes that the data is generated by the mixture of multiple Gaussian distributions. Some clustering methods alternative to GMM can be used, and these methods also utilize the characteristics of Gaussian distributions:

[0063] Such as the Radial Basis Function Network (RBF network): The RBF network uses radial basis functions (usually Gaussian functions) as the activation function of the hidden layer, can capture the non-linear features of data, and is used for clustering tasks. The RBF network can adapt to different data distributions by adjusting the number of neurons in the hidden layer and the parameters of the Gaussian function.

[0064] Such as the autoencoder (AE): Autoencoders, especially Variational Autoencoders (VAEs), can use Gaussian noise as a regularization means or utilize the Gaussian distribution to model the latent representation of data during the encoding stage. Autoencoders can learn the compressed representation of data, and these representations can be used for subsequent clustering tasks.

[0065] Such as Deep Embedded Clustering (DEC): The DEC method combines the feature learning of deep learning and clustering algorithms. By using a deep neural network to learn the low-dimensional representation of data and then applying traditional clustering algorithms such as K-means, DEC can discover complex clustering structures in the data.

[0066] Such as neural network clustering with Gaussian distribution: Specific neural network layers can be designed to use the Gaussian distribution to model the similarity or distance between data points in the hidden layer or output layer. This method allows the network to learn the Gaussian representation of data while maintaining the flexibility and expressiveness of the neural network.

[0067] Step 26: Obtain the fault diagnosis result according to the classification result.

[0068] In this embodiment, by obtaining multi-dimensional data in real time, the timeliness and accuracy of the data are ensured, and the multi-dimensional data can comprehensively reflect the operating state of the electric drive system. Furthermore, the target prediction model is used to combine the multi-dimensional data for motor temperature prediction, and the multi-dimensional data segment corresponding to the predicted motor temperature being greater than the preset temperature threshold is obtained. The real-time stream segment that can be used for fault diagnosis is screened in advance, and starting from the high-temperature data, the resolution ability of the fault diagnosis model for abnormal data pattern recognition is further improved. The singular value decomposition technology is further used to perform feature engineering processing on the multi-dimensional data segment to obtain the target data matrix, extract key features from the multi-dimensional data segment, reorganize the data and extract the important information that best reflects the change relationship between time and variables. The automatic feature selection method can be combined to reduce the subjectivity of feature selection, improve the objectivity and accuracy of feature extraction, and through an efficient feature extraction method, reorganize a new type of data that can better reflect the characteristics and differences of data changes, reduce the blindness of calculation, and improve the processing efficiency. Further, the Gaussian mixture model is used to perform clustering analysis on the target data matrix to obtain the classification result; the fault diagnosis result is obtained according to the classification result, which can improve the real-time performance, accuracy and reliability of the electric drive system fault detection, thereby ensuring the safe operation of new energy vehicles and improving the travel experience of users.

[0069] Refer to Figure 3 , Figure 3 is a schematic flowchart of another embodiment of the fault diagnosis method for an electric drive system provided by this application. The fault diagnosis method includes:

[0070] Step 31: Obtain multi-dimensional data in real time; among them, the multi-dimensional data includes at least two of the inlet temperature, controller temperature, voltage, current, speed, and torque.

[0071] Step 32: Use the target prediction model to combine the multi-dimensional data for motor temperature prediction, and obtain the multi-dimensional data segment corresponding to the predicted motor temperature being greater than the preset temperature threshold.

[0072] Step 33: Use the singular value decomposition technology to perform feature engineering processing on the multi-dimensional data segment to obtain the target data matrix.

[0073] Step 34: Use the Gaussian mixture model to perform clustering analysis on the target data matrix, and classify the target data matrix into the corresponding data clusters; and record the category code of the data cluster as the classification result.

[0074] Step 35: Count the number of multi-dimensional data segments corresponding to each category.

[0075] Step 36: According to the number of multi-dimensional data segments corresponding to each category, count the proportion corresponding to each category.

[0076] Step 37: Obtain the fault diagnosis result based on the quantity ratio.

[0077] In an application scenario, the categories include a normal category and a fault category.

[0078] In response to the quantity ratio of the normal category being greater than the first threshold, obtain the first fault diagnosis result; the first fault diagnosis result indicates that the electric drive system is normal. That is, if the statistical quantity of the categories judged as normal is more than that of the categories judged as faulty, it is considered that the electric drive system is normal.

[0079] Furthermore, by comparing the statistical quantities of the categories judged as normal and faulty, determine the state of the electric drive system. If the quantity of the normal category is in the majority, it is considered that the system is normal. Ensure that the vehicle electric drive system continues to operate in a normal state and avoid unnecessary maintenance operations.

[0080] In response to the quantity ratio of the fault category being greater than the first threshold, obtain the second fault diagnosis result; the second fault diagnosis result indicates that the electric drive system has a fault. That is, if the statistical quantity of the categories judged as faulty is more than that of the categories judged as normal, it is considered that the electric drive system has a fault.

[0081] Furthermore, by comparing the statistical quantities of the categories judged as the normal mode and the fault mode, determine the state of the electric drive system. If the quantity of the categories in the fault mode is in the majority, it is considered that the system has a fault, and the fault can be detected and processed in a timely manner to prevent the fault from deteriorating further.

[0082] In response to the quantity ratios of the normal category and the fault category being less than the first threshold, obtain the third fault diagnosis result; the first fault diagnosis result indicates that the electric drive system needs further diagnosis. That is, if the statistical quantities of the categories judged as the normal mode and the fault mode are equivalent, no conclusion is drawn temporarily.

[0083] Furthermore, by comparing the statistical quantities of the normal and fault categories, if the quantities of both are equivalent, it is considered that the data is insufficient to make a clear judgment, which can avoid misjudgment, and further cumulative data needs to be collected or manual inspection is required to confirm the system state.

[0084] In some embodiments, after obtaining the fault diagnosis result according to the classification result, determine the vehicles for which the fault diagnosis result indicates that the electric drive system has a fault; perform a fault prompt according to the vehicle information.

[0085] In an application scenario, the final diagnostic result is output based on the result judgment. For a single vehicle, statistics are made once a day. If the electric drive system is judged to be normal (the first fault diagnosis result), it is recorded as the "normal" state. For vehicles in the normal state, the database will not query and display other specific information of the vehicle by default, in order to save resources, focus on faulty vehicles, ensure that the electric drive system of the vehicle continues to operate in the normal state, and avoid unnecessary maintenance operations.

[0086] If the electric drive system is judged to be faulty (the second fault diagnosis result), it is recorded as the "fault" state. Through the result display interface or message push, relevant personnel are notified of which vehicles' electric drive systems have faults and need to be processed in a timely manner, so as to detect and handle faults in a timely manner and prevent the faults from deteriorating further.

[0087] If the electric drive system is judged to be doubtful (the third fault diagnosis result), it is recorded as the "doubtful" state. A separate pool for suspected faulty vehicles is set up to store the vehicle information belonging to the "doubtful" state. Through the result display interface or message push, relevant personnel are notified that the system state is doubtful and further inspection is required, which can avoid misjudgment and ensure a more accurate confirmation of the system state.

[0088] In an application scenario, during the daily driving process of new energy vehicles, the operating conditions of the electric drive system are in dynamic change. Sensors are mounted to collect the operating data of the electric drive system in real time, and the data is transmitted to the cloud server through a wireless communication module.

[0089] First, data acquisition and preprocessing: The multi-dimensional sensors of the vehicle collect the electric drive operation-related data once every fixed number of seconds as the data at a time point. Accumulating M finite fixed number of time points constitutes a time window. Since there are many vehicles driving on the road every day, every moment, the cloud real-time stream will continuously receive the operating data transmitted back by the vehicles. First, it is necessary to call the real-time temperature prediction model to predict the motor temperature corresponding to each window and obtain a value (predicted motor temperature).

[0090] Then temperature judgment is carried out: For the value predicted for each time window, judge whether its size exceeds the preset temperature threshold, so as to screen out the windows with temperature prediction values above the preset temperature threshold, and intercept the subsequent N (N < M) time points to form a new time window (multi-dimensional data segment) and enter the GMM classification model.

[0091] Then classification and summarization are carried out: Use the pre-trained GMM model to perform clustering analysis on each filtered new window, obtain the category judged for this window, and store the record.

[0092] For all the category results judged under all windows accumulated for a single vehicle in a single day, calculate the proportions of several special categories in the total number of windows to obtain the corresponding proportion values.

[0093] For all vehicles on the same day that meet the condition that the number of windows is greater than a certain preset number, calculate and record the above-mentioned proportion values.

[0094] Then conduct result judgment: Compare the proportion values of several special categories of a single vehicle with the corresponding preset thresholds. Based on the comparison results of each category, the vehicle can be judged as faulty or normal, and then count the number of vehicles judged as normal or faulty under these several special categories.

[0095] Then perform result output: Based on the number of normal and faulty vehicles counted under several special categories, select the state with a larger number as the final judgment result. If the number of normal and faulty vehicles is the same, it is judged that the current vehicle is in a state of suspected fault.

[0096] Furthermore, during the daily operation of the fault diagnosis model, perform temperature prediction and clustering analysis on each time window in the real-time stream data. The specific steps are as follows:

[0097] Temperature prediction: Use a time series prediction model to predict the motor temperature of each time window to screen out real-time stream segments suitable for fault diagnosis.

[0098] Clustering analysis: Perform clustering analysis on the data processed by SVD feature engineering using GMM to obtain the clustering results of each time window.

[0099] Statistics by day: Statistically analyze the clustering results of each vehicle every day and summarize them.

[0100] State determination: Obtain the state of the electric drive system of each vehicle through statistical analysis and judge it as faulty, continue to observe, or normal.

[0101] Regarding the detection effect of the model:

[0102] Fault: For electric drive system faults that have been manually confirmed to exist, the model can accurately identify and mark them.

[0103] Continue to observe: For those cases where the motor temperature performance is not severe enough but it is found through manual analysis that there is a high possibility of potential faults, mark them as "continue to observe" for further confirmation later.

[0104] Newly discovered fault: For those faults that have not been manually warned before, they can be detected in time and marked as "newly discovered fault" to remind relevant personnel to conduct warning analysis.

[0105] Normal state: For those vehicles operating normally, the model can accurately classify them as "normal".

[0106] Result statistics

[0107] The fault diagnosis model makes temperature predictions and clustering analyses for each window of real-time streaming data daily, statistically counts the clustering results of each vehicle by day, and then obtains the state of the electric drive system of each vehicle through statistical analysis.

[0108] That is, the technical solution of the present application can be summarized as having the following effects:

[0109] (1) Faulty vehicles can be effectively detected, and the accuracy shown daily currently can meet the pre-set fault diagnosis goal.

[0110] (2) Some vehicles with not-so-high motor temperature manifestations but already showing signs have been detected.

[0111] (3) The model can discover vehicles that have not been warned before.

[0112] (4) The error rate of the model detection is relatively low and has been statistically verified over a long time.

[0113] In an application scenario, the present application can be used for regular data analysis and maintenance. Suppose a new energy vehicle company conducts regular analysis on the operation data of all vehicles at the end of each month to evaluate the overall health status of the vehicles.

[0114] First, data acquisition and preprocessing: The system obtains the operation data of all vehicles in the past month from the cloud server and performs cleaning and preprocessing. The motor temperature is predicted through a time series prediction model to generate the predicted motor temperature values for each time window.

[0115] Then, temperature judgment is carried out: The predicted motor temperature data of each vehicle is judged one by one, and the vehicle windows with temperatures exceeding the preset temperature threshold are screened out.

[0116] Then, classification and summarization are carried out: Combining SVD feature engineering, GMM is used to perform clustering analysis on the screened vehicle window data to obtain the categories judged in this window and store the records. Taking days as the unit, for the category results judged in all windows accumulated by a single vehicle in a single month, the proportion of several special categories in the total number of windows is statistically calculated to obtain the corresponding proportion values. For all vehicles on a single day of the current month that meet the condition that the number of windows is greater than a certain preset number, the above-mentioned proportion values are calculated and recorded.

[0117] Then, the result judgment is carried out: for a single vehicle, compare the proportional values of several special categories in the GMM classification with the corresponding preset thresholds. From the comparison results of each category, the vehicle can be judged as faulty or normal. Then, count the number of vehicles judged as normal or faulty under these several special categories, and the statistics are also carried out on a daily basis within the same month.

[0118] Then, the result is output: the system generates a detailed report listing the status changes of each vehicle. Especially for the faulty vehicles, typically there is a typical change trend from normal to doubtful and then to faulty on a monthly basis. The report is sent to the relevant departments via email. For the vehicles judged as faulty, the system recommends further inspection and maintenance.

[0119] In some embodiments, the Gaussian mixture model (GMM) can obtain the optimal Gaussian mixture model through training and parameter selection. For example, through the training and parameter setting of the GMM model, the optimal number of components is determined. For example, initialize the parameters, the expectation step (E-step) and the maximization step (M-step), and set the corresponding convergence conditions.

[0120] From the perspective of machine learning, when building a clustering analysis model, it generally includes several basic steps such as data preprocessing, feature engineering, and model training. Specifically, merge the two waves of historical data of the initially extracted faulty and normal vehicle operations. At this time, the obtained data contains many driving segments. Divide each driving segment into time windows, and each window has a fixed and unified shape (all in matrix form), thus forming time series data. Then, use svd to solve the singular values of the matrix, and reorganize and transform the obtained vectors, and then integrate them into a dataset for GMM clustering training. The trained model is the final model for electric drive fault diagnosis.

[0121] Among them, the specific method and its advantages for building the electric drive fault diagnosis model are as follows:

[0122] 1. The construction method is as follows:

[0123] Data preprocessing: First, collect the historical data of the electric drive system operating in normal and faulty states, including but not limited to multi-variable time series data such as voltage, current, speed, and torque. Clean these data to remove noise and outliers to ensure data quality.

[0124] Feature Engineering: Next, apply the time window segmentation technique to the preprocessed data, that is, cut the continuous time series data into multiple time windows of fixed length. The data within each time window forms a matrix, representing a snapshot of the system state over a period of time. Then, use the SVD algorithm to perform dimensionality reduction on these matrices and extract the most representative feature vectors. SVD can effectively capture the main change trends in the data, while reducing the data dimension and computational complexity.

[0125] Model Training: Input the set of feature vectors obtained after SVD processing into the GMM for unsupervised learning. GMM can automatically discover the hidden patterns in the data and classify similar behaviors into the same cluster, thus achieving the classification of the operating states of the electric drive system. After training, the model can identify the differences between the normal operating state and various fault modes.

[0126] 2. The advantages and progressiveness of the model are as follows:

[0127] Compared with traditional fault diagnosis methods, the electric drive fault diagnosis model based on SVD feature engineering and GMM multivariate time series clustering proposed in the present invention has the following significant advantages:

[0128] Multivariate Comprehensive Analysis: Traditional methods often only focus on the changes in single or a few indicators, while this model can comprehensively consider the operating state of the electric drive system through multivariate time series analysis, improving the accuracy and reliability of fault detection.

[0129] Strong Adaptability: As a probability model, GMM can automatically adjust parameters to adapt to different data distributions, enabling the model to have good generalization ability and effectively identify even new fault data that has never been seen before (actual verification shows that although only 48 fault vehicle data were used during model training, a total of more than 170 fault vehicles on the road were detected, that is, the model can detect the vehicle fault data it has never seen).

[0130] Real-time Performance and Efficiency: Combined with big data processing technology, the model can quickly complete training and prediction on large-scale data sets and is suitable for real-time monitoring scenarios (in actual operation, the fault vehicles detected on the same day will be counted on a daily basis to achieve rapid analysis and early warning).

[0131] Furthermore, in addition to fault diagnosis and prediction based on temperature, other methods can be explored to enhance the comprehensiveness and accuracy of fault diagnosis:

[0132] Vibration Analysis: By analyzing the vibration signals generated during motor operation, problems such as wear or imbalance of mechanical components can be detected.

[0133] Acoustic monitoring: Use acoustic sensors to capture the sound changes during the operation of the electric drive system, assist in judging whether there are abnormal noises or strange sounds, and prompt potential faults.

[0134] Electrical characteristic analysis: Deeply study the change rules of current and voltage waveforms, especially the non-linear characteristics, which helps to identify the aging or damage of electronic components.

[0135] Machine learning and deep learning: Combine more types of machine learning algorithms (such as support vector machines, random forests, etc.) or deep learning models (such as convolutional neural networks, recurrent neural networks, etc.) to further improve the intelligent level of fault diagnosis.

[0136] In summary, the electric drive fault diagnosis model based on SVD feature engineering and GMM multivariate time series clustering is not only innovative in technical implementation, but also shows strong performance advantages in practical applications. By introducing multiple data sources and technical means, this model can more comprehensively and accurately evaluate the health status of the electric drive system, providing strong support for preventive maintenance.

[0137] Furthermore, the vehicle network is used to send the real-time data such as voltage, current, speed, and torque collected by sensors during the operation of the electric drive system to the backend cloud platform. At the same moment, the cloud platform will receive the real-time data transmitted from different vehicles at the port. Therefore, corresponding real-time stream data will be accumulated over a period of time. In the fault diagnosis process, first, temperature prediction is performed on these data for each time window, and then the data that meets the temperature threshold requirements predicted is passed to the next level for SVD feature engineering and GMM clustering.

[0138] Specifically, a variety of high-precision sensors installed on the vehicle can continuously monitor and collect the key parameters of the electric drive system, such as voltage, current, speed, and torque information. These data are transmitted in real time to the backend cloud platform through the in-vehicle communication module with the help of a high-speed and stable wireless network connection. The cloud platform, as the core of data processing, uses a distributed computing framework to process the concurrent data streams from multiple vehicles. This process not only includes basic cleaning and formatting of the original data, but also involves complex time series analysis, such as executing a temperature prediction model within each time window to screen out data points that meet specific temperature thresholds. This step is crucial because abnormal high temperatures are often early indicators of potential faults in the electric drive system. Only when the data meets the preset conditions will it be further transmitted to the subsequent SVD (Singular Value Decomposition) feature extraction stage and GMM (Gaussian Mixture Model) clustering analysis link, thus providing accurate data support for fault mode recognition.

[0139] In addition, to ensure the efficiency and reliability of the entire data processing chain, a series of optimization measures are also required. For example, reducing network transmission load through data compression, accelerating data access speed using a caching mechanism, and implementing a fault tolerance strategy to ensure data integrity. In summary, the combined effect of this series of technical means has greatly improved the real-time performance and accuracy of electric drive system fault diagnosis.

[0140] Data cleaning and quality control: Implement a comprehensive data cleaning process to remove noise and outliers manually and in combination with automation (please specify), improve data quality, and lay a foundation for subsequent analysis.

[0141] This needs to be introduced in two parts:

[0142] The first part refers to the need to perform preliminary extraction on the historical vehicle operation data cached in the cloud platform before the training and shaping of the fault diagnosis model goes online. Since the samples of early fault vehicles were all discovered and extracted manually, this manual operation is essential in this step. By automation, it means that after loading the original operation data locally, necessary conditional screening is performed in combination with the program to meet the training needs.

[0143] The second part refers to when detecting based on real-time flow after the fault detection model goes online. At this time, the acquisition and screening of data are both based on a given policy and do not require manual intervention and processing.

[0144] Refer to Figure 4 , Figure 4 FIG. is a schematic structural diagram of an embodiment of a fault diagnosis system for an electric drive system provided by this application. The fault diagnosis system 40 of the electric drive system includes a communication interface 43, a memory 41, and a processor 42 coupled to the memory 41 and the communication interface 43. The memory 41 stores at least one computer program. When at least one computer program is loaded and executed by the processor 42, it is used to implement the following methods:

[0145] Obtain multi-dimensional data in real time; among them, the multi-dimensional data includes at least two of the inlet temperature, controller temperature, voltage, current, speed, and torque; use a target prediction model to combine the multi-dimensional data to predict the motor temperature, and obtain the multi-dimensional data segment corresponding to when the predicted motor temperature is greater than a preset temperature threshold; use singular value decomposition technology to perform feature engineering processing on the multi-dimensional data segment to obtain a target data matrix; use a Gaussian mixture model to perform clustering analysis on the target data matrix to obtain a classification result; obtain a fault diagnosis result according to the classification result.

[0146] In some embodiments, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the following method: perform SVD solution on the multi-dimensional data segment to obtain the left singular vector, the right singular vector, and the diagonal matrix containing singular values; obtain a preset number of target singular values, and the left singular vector and the right singular vector corresponding to each target singular value for matrix recombination to obtain the target data matrix; wherein, the preset number of target singular values is greater than the remaining singular values among all singular values.

[0147] In some embodiments, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the following method: perform clustering analysis on the target data matrix using the Gaussian mixture model, and classify the target data matrix into corresponding data clusters; and record the class code of the data cluster as the classification result.

[0148] In some embodiments, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the following method: count the number of multi-dimensional data segments corresponding to each category; according to the number of multi-dimensional data segments corresponding to each category, count the proportion corresponding to each category; obtain the fault diagnosis result according to the proportion.

[0149] In some embodiments, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the following method: in response to the proportion of the normal category being greater than the first threshold, obtain the first fault diagnosis result; the first fault diagnosis result indicates that the electric drive system is normal; in response to the proportion of the fault category being greater than the first threshold, obtain the second fault diagnosis result; the second fault diagnosis result indicates that the electric drive system has a fault; in response to the proportions of the normal category and the fault category being less than the first threshold, obtain the third fault diagnosis result; the first fault diagnosis result indicates that the electric drive system needs further diagnosis.

[0150] In some embodiments, after obtaining the fault diagnosis result according to the classification result, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the following method: determine the vehicles for which the fault diagnosis result indicates that the electric drive system has a fault; perform a fault prompt according to the vehicle information.

[0151] In some embodiments, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the following method: when the predicted motor temperature is greater than the preset temperature threshold, intercept the multi-dimensional data of N time lengths after the multi-dimensional data used for prediction as the multi-dimensional data segment.

[0152] In some embodiments, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the following method: perform data cleaning on the multi-dimensional data; use the target prediction model to combine the cleaned multi-dimensional data for motor temperature prediction.

[0153] In some embodiments, when at least one computer program is loaded and executed by the processor 42, it is further used to implement the methods mentioned in any of the above embodiments.

[0154] In some embodiments, the communication interface 43 is mainly coupled to the communication interface on the vehicle to receive multi-dimensional data collected by sensors.

[0155] Refer to Figure 5 , Figure 5 FIG. is a schematic structural diagram of an embodiment of a computer-readable storage medium provided by the present application. The computer-readable storage medium 50 stores computer-executable instructions 51, and when the computer-executable instructions 51 are executed by a processor, they are used to implement the following methods:

[0156] Obtain multi-dimensional data in real time; wherein, the multi-dimensional data includes at least two of the inlet temperature, controller temperature, voltage, current, rotational speed, and torque; use the target prediction model to combine the multi-dimensional data to predict the motor temperature, and obtain the multi-dimensional data segment corresponding to when the predicted motor temperature is greater than the preset temperature threshold; use the singular value decomposition technology to perform feature engineering processing on the multi-dimensional data segment to obtain the target data matrix; use the Gaussian mixture model to perform clustering analysis on the target data matrix to obtain a classification result; obtain a fault diagnosis result according to the classification result.

[0157] In some embodiments, when the computer-executable instructions 51 are executed by a processor, they are used to implement the following methods: perform SVD solution on the multi-dimensional data segment to obtain the left singular vector, right singular vector, and diagonal matrix containing singular values; obtain a preset number of target singular values, and the left singular vector and right singular vector corresponding to each target singular value for matrix recombination to obtain the target data matrix; wherein, the preset number of target singular values is greater than the remaining singular values among all singular values.

[0158] In some embodiments, when the computer-executable instructions 51 are executed by a processor, they are used to implement the following methods: use the Gaussian mixture model to perform clustering analysis on the target data matrix, and classify the target data matrix into corresponding data clusters; and record the category code of the data cluster as the classification result.

[0159] In some embodiments, when the computer-executable instructions 51 are executed by a processor, they are used to implement the following methods: count the number of multi-dimensional data segments corresponding to each category; according to the number of multi-dimensional data segments corresponding to each category, count the proportion corresponding to each category; obtain a fault diagnosis result according to the proportion.

[0160] In some embodiments, when the computer-executable instruction 51 is executed by a processor, it is used to implement the following method: obtaining a first fault diagnosis result in response to the proportion of the normal category being greater than a first threshold; the first fault diagnosis result indicates that the electric drive system is normal; obtaining a second fault diagnosis result in response to the proportion of the fault category being greater than the first threshold; the second fault diagnosis result indicates that the electric drive system has a fault; obtaining a third fault diagnosis result in response to the proportions of the normal category and the fault category being less than the first threshold; the first fault diagnosis result indicates that the electric drive system needs further diagnosis.

[0161] In some embodiments, after obtaining the fault diagnosis result according to the classification result, when the computer-executable instruction 51 is executed by a processor, it is used to implement the following method: determining the vehicles for which the fault diagnosis result indicates that the electric drive system has a fault; performing a fault prompt according to the vehicle information.

[0162] In some embodiments, when the computer-executable instruction 51 is executed by a processor, it is used to implement the following method: when predicting that the motor temperature is greater than a preset temperature threshold, intercepting the multi-dimensional data of N time lengths after the multi-dimensional data used for prediction as a multi-dimensional data segment.

[0163] In some embodiments, when the computer-executable instruction 51 is executed by a processor, it is used to implement the following method: performing data cleaning on the multi-dimensional data; using a target prediction model to combine the cleaned multi-dimensional data for motor temperature prediction.

[0164] In some embodiments, when the computer-executable instruction 51 is executed by a processor, it is used to implement the methods mentioned in any of the above embodiments.

[0165] Combined with the above technical solutions, the present application significantly improves the fault diagnosis and early warning capabilities of the electric drive system through a series of innovative technical means. First, the present application adopts big data technology and Internet of Things technology to achieve real-time collection and preprocessing of vehicle operation data. This not only ensures the timeliness and accuracy of the data, but also can comprehensively reflect the multi-dimensional operation status of the electric drive system. Through the real-time flow motor temperature prediction model, the present application can screen out the real-time flow segments that may have problems in advance, thereby improving the efficiency and accuracy of subsequent fault diagnosis.

[0166] Secondly, the present application introduces the singular value decomposition (SVD) feature engineering technology to extract key features from multi-dimensional time series data and generate a reduced-dimensional data representation through a matrix recombination method. This method not only reduces noise, but also improves the clustering effect of the GMM classifier. Compared with traditional statistical analysis methods (such as mean, variance, etc.), SVD feature engineering can more effectively capture the complex patterns in multi-variable time series data, thereby providing high-quality input data for subsequent fault diagnosis.

[0167] In addition, this application uses the Gaussian Mixture Model (GMM) for clustering analysis and determines the optimal number of components through the Bayesian Information Criterion (BIC). GMM can handle non-linear and high-dimensional data and effectively identify complex abnormal patterns. This GMM-based clustering analysis method not only improves the robustness of the algorithm in processing complex data but also reduces the computational complexity and improves the real-time performance of clustering analysis through optimized algorithms and parallel computing technologies.

[0168] Finally, this application constructs an efficient fault diagnosis model. This model can effectively identify and detect the driving segment data of faulty vehicles based on the SVD feature engineering and the multivariate time series clustering results of GMM. Through a large amount of data training and verification, this model has good generalization ability and reduces the overfitting phenomenon. At the same time, with the help of this model, the system can perform real-time analysis on the vehicle operation data transmitted daily and remind users or maintenance personnel to handle it in time when abnormal data is found, preventing the further deterioration of faults.

[0169] In summary, through real-time data collection and preprocessing, efficient feature extraction, stable clustering analysis, and a powerful fault diagnosis model, this application significantly improves the real-time performance, accuracy, and reliability of electric drive system fault detection. These innovations not only solve some problems existing in related technologies but also provide a strong guarantee for the safe operation of new energy vehicles and improve the travel experience of users.

[0170] Finally, corresponding explanations are given for the nouns appearing in this application:

[0171] 1. Fault Diagnosis:

[0172] Definition: Fault diagnosis refers to the process of monitoring and analyzing the operating state of a device or system to detect and identify faults. In an electric drive system, fault diagnosis usually includes multiple steps such as data collection, preprocessing, feature extraction, and classification.

[0173] Application scenario: By analyzing the real-time data of the electric drive system, potential faults can be discovered in a timely manner to ensure the normal operation of the system.

[0174] 2. Singular Value Decomposition (SVD):

[0175] Definition: Singular value decomposition is a linear algebra method that decomposes a matrix into the product of three matrices. Given an m×n matrix A, it can be decomposed as:

[0176] A = U∑V T .

[0177] Among them, U is an m×m unitary matrix, Σ is an m×n diagonal matrix containing singular values σ i , V T is an n×n unitary matrix.

[0178] Application scenario: In the fault diagnosis of the electric drive system, SVD is used to extract key features from time series data and perform dimensionality reduction.

[0179] 3. Gaussian Mixture Model (GMM):

[0180] Definition: The Gaussian mixture model is a probability model used to represent data composed of multiple Gaussian distributions (normal distributions). Given a data set X = {x1, x2,..., xN}, GMM can be expressed as:

[0181]

[0182] Among them, π i is the weight of the i-th Gaussian distribution, μ i is the mean vector, Σ i is the covariance matrix, and K is the number of Gaussian distributions.

[0183] Application scenario: In the fault diagnosis of the electric drive system, GMM is used to perform clustering analysis on the data processed by SVD, and identify fault segments by clustering the proportion of different categories.

[0184] 4. Bayesian Information Criterion (BIC):

[0185] Definition: The Bayesian information criterion is a statistical method for model selection. BIC evaluates the goodness of a model by balancing the goodness of fit and complexity of the model. BIC is defined as:

[0186]

[0187] Among them, is the maximum likelihood estimate of the model, k is the degrees of freedom of the model, and n is the number of samples.

[0188] Application scenario: In the fault diagnosis of the electric drive system, BIC is used to determine the optimal number of components of the GMM model.

[0189] 5. Dynamic Time Warping (DTW):

[0190] Definition: Dynamic Time Warping (DTW) is a method for measuring the similarity between two sequences, even if the two sequences have different lengths on the time axis. DTW calculates the similarity by finding the optimal matching path between the two sequences.

[0191] Application scenario: In the fault diagnosis of electric drive systems, DTW is used to compare the similarity between different time series data and identify abnormal patterns.

[0192] 6. Radial Basis Function Kernel (RBF):

[0193] Definition: The Radial Basis Function Kernel is a commonly used kernel function for mapping data into a high-dimensional space. The RBF kernel is defined as:

[0194]

[0195] where ||x - y|| 2 is the square of the Euclidean distance between two points, and σ is the kernel width parameter.

[0196] Application scenario: In the fault diagnosis of electric drive systems, the RBF kernel is used for classification tasks such as Support Vector Machines (SVM) to improve the non-linear classification ability of the model.

[0197] 7. Autoencoder (AE):

[0198] Definition: An autoencoder is an unsupervised learning method for learning an efficient encoding of data. An autoencoder consists of an encoder and a decoder, and the goal is to minimize the reconstruction error.

[0199] Application scenario: In the fault diagnosis of electric drive systems, autoencoders are used for feature extraction and noise reduction to improve the quality of data.

[0200] 8. Deep Embedded Clustering (DEC):

[0201] Definition: Deep Embedded Clustering is a deep learning-based clustering method that performs clustering by learning the embedded representation of data. The goal of DEC is to minimize the distance between the embedded representation and the cluster centers.

[0202] Application scenario: In the fault diagnosis of electric drive systems, DEC is used for clustering analysis of high-dimensional data to identify different fault patterns.

[0203] In several embodiments provided by the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0204] If the integrated units in the above-mentioned other embodiments are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0205] The above are only the embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A fault diagnosis method for an electric drive system, characterized in that: The fault diagnosis method comprises: Acquire multi-dimensional data in real time; wherein the multi-dimensional data includes at least two of water inlet temperature, controller temperature, voltage, current, speed and torque; Predicting the motor temperature by using the target prediction model in combination with the multi-dimensional data, and obtaining a multi-dimensional data segment corresponding to when the predicted motor temperature is greater than a preset temperature threshold; Performing feature engineering processing on the multi-dimensional data fragments using singular value decomposition technology to obtain a target data matrix; Performing cluster analysis on the target data matrix using a Gaussian mixture model to obtain a classification result; The fault diagnosis result is obtained according to the classification result.

2. The fault diagnosis method according to claim 1, characterized in that: The method of performing feature engineering processing on the multi-dimensional data segments using singular value decomposition technology to obtain a target data matrix includes: Performing SVD on the multi-dimensional data segment to obtain left singular vectors, right singular vectors, and a diagonal matrix containing singular values; A preset number of target singular values ​​and a left singular vector and a right singular vector corresponding to each target singular value are obtained to perform matrix reorganization to obtain the target data matrix; wherein the preset number of target singular values ​​are greater than the remaining singular values ​​among all the singular values.

3. The fault diagnosis method according to claim 1, characterized in that: The Gaussian mixture model is used to perform cluster analysis on the target data matrix to obtain a classification result, including: Performing cluster analysis on the target data matrix using a Gaussian mixture model, and classifying the target data matrix into corresponding data clusters; And the category code of the data cluster is recorded as the classification result.

4. The fault diagnosis method according to claim 1, characterized in that: Obtaining a fault diagnosis result according to the classification result includes: Count the number of multi-dimensional data fragments corresponding to each category; According to the number of multi-dimensional data fragments corresponding to each category, the proportion of the corresponding number of each category is counted; A fault diagnosis result is obtained according to the quantity ratio.

5. The fault diagnosis method according to claim 4, characterized in that: The categories include normal categories and fault categories; the fault diagnosis results obtained according to the quantity proportions include: In response to the proportion of the number of the normal categories being greater than a first threshold, a first fault diagnosis result is obtained; the first fault diagnosis result indicates that the electric drive system is normal; In response to the proportion of the number of the fault categories being greater than the first threshold, obtaining a second fault diagnosis result; the second fault diagnosis result indicates a fault in the electric drive system; In response to the fact that the proportion of the number of the normal category and the fault category is less than the first threshold, a third fault diagnosis result is obtained; the first fault diagnosis result indicates that the electric drive system requires further diagnosis.

6. The fault diagnosis method according to claim 1, characterized in that: After the fault diagnosis result is obtained according to the classification result, the method includes: determining a vehicle for which a fault diagnosis result indicates a fault in the electric drive system; Provide fault prompts based on vehicle information.

7. The fault diagnosis method according to claim 1, characterized in that: The obtaining of the multi-dimensional data segment corresponding to when the predicted motor temperature is greater than a preset temperature threshold comprises: When the predicted motor temperature is greater than a preset temperature threshold, multi-dimensional data of N time lengths after the multi-dimensional data used for prediction is intercepted as the multi-dimensional data segment.

8. The fault diagnosis method according to claim 1, characterized in that: The method of using the target prediction model in combination with the multi-dimensional data to predict the motor temperature includes: Performing data cleaning on the multi-dimensional data; The target prediction model is used in combination with the multi-dimensional data after cleaning to predict the motor temperature.

9. A fault diagnosis system for an electric drive system, characterized in that: The fault diagnosis system includes: a communication interface, a memory, and a processor coupled to the memory and the communication interface, wherein the memory stores at least one computer program, and when the at least one computer program is loaded and executed by the processor, it is used to implement the method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method according to any one of claims 1 to 8.

Images