New energy vehicle bearing defect detection method and device

Abstract

The present invention provides a new energy vehicle bearing defect detection method and device, the method comprising: obtaining discrete bearing vibration signals under different fault conditions; extracting features from the discrete bearing vibration signals to obtain a number of features; performing natural neighborhood hyperspherical oversampling on the several features to obtain a balanced data set; inputting the balanced data set into a bearing defect detection model for training, and the bearing defect detection model outputs predicted bearing defects. The method performs new energy bearing detection through a small sample of unbalanced data, ensuring that the new energy vehicle bearing detection is carried out efficiently and orderly.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and in particular relates to a method and device for detecting bearing defects of a new energy vehicle. Background Art

[0002] With the development of new energy vehicles, higher requirements are placed on the performance and reliability of bearings.

[0003] First of all, there are some specific factors in new energy vehicles that make their bearings prone to damage. For example, electric vehicles have higher speeds and operating temperatures, which makes some bearings prone to wear in such high-speed, high-temperature environments. In addition, the motor drive system of new energy vehicles will generate alternating currents, causing changes in the surrounding electromagnetic fields. This change in the electromagnetic field may cause electrical corrosion to the bearings, thereby affecting their service life and reliability. And the source car may generate vibrations and shocks during operation. These vibrations and shocks will increase the load on the bearings, easily causing indentations on the raceways or damaging the bearings. In addition, vibrations will also make the bearing seals easier to loosen, increasing the chances of dust particles and acidic and alkaline substances entering the bearings, thereby worsening the lubrication conditions of the bearings and accelerating their damage. These factors will cause bearings to be damaged to varying degrees.

[0004] Secondly, when the bearing is in operation in a new energy vehicle, cracks, surface damage, wear and other faults will cause elastic impact on the contact surface, resulting in the generation of soundprint signals and a large amount of fault information. Traditional bearing fault detection technology assumes sample balance by default during use. When the data is unbalanced, the expected risk minimization causes the model to pay more attention to large-class samples and ignore small-class samples with higher misclassification costs. At the same time, high-dimensional data usually has irrelevant and redundant features, which makes the algorithm complex and the classification effect poor. Data-driven deep learning methods require a large amount of data support, which is difficult to implement in the field of fault diagnosis. There is a serious problem of sample heterogeneity in the acquisition of industrial data: normal data is easy to obtain, but the cost of obtaining specific fault data is high. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a new energy vehicle bearing defect detection method and device, which performs new energy vehicle bearing detection through a small sample of unbalanced data, thereby ensuring that the new energy vehicle bearing detection is carried out efficiently and orderly.

[0006] In order to achieve the above-mentioned object, the present invention provides a new energy vehicle bearing defect detection method on the one hand, comprising:

[0007] Obtain discrete bearing vibration signals under different fault conditions;

[0008] Perform feature extraction on the discrete bearing vibration signal to obtain several features;

[0009] For these features, hyperspherical oversampling of natural neighborhood is performed to obtain a balanced data set;

[0010] The balanced data set is input into a bearing defect detection model for training, and the bearing defect detection model outputs predicted bearing defects.

[0011] In one embodiment, the discrete bearing vibration signal is subjected to feature extraction using empirical wavelet transform, comprising:

[0012] Performing signal preprocessing on the discrete bearing vibration signal;

[0013] The preprocessed discrete bearing vibration signal is decomposed into several IMF components through empirical mode decomposition.

[0014] Perform wavelet transform on each IMF component to obtain the frequency domain representation of each IMF component;

[0015] Wavelet coefficients of different frequency bands are extracted to obtain the features.

[0016] In one embodiment, all the maximum and minimum points of the preprocessed discrete bearing vibration signal are found, and the maximum envelope and the minimum envelope are fitted to the maximum and minimum points respectively through a cubic spline function;

[0017] Calculate the average value of the maximum envelope and the minimum envelope to obtain a mean value signal;

[0018] Based on the difference between the preprocessed discrete bearing vibration signal and the mean signal, an n-order IMF component is obtained;

[0019] The difference between the discrete bearing vibration signal and each order IMF component is used to obtain a disturbance signal corresponding to the order IMF component;

[0020] The discrete bearing vibration signal is decomposed into the sum of the weighted n-th order IMF component and the disturbance signal corresponding to the n-th order IMF component.

[0021] In one embodiment, natural neighbors are searched for the discrete bearing vibration signal after feature extraction to form a natural neighborhood;

[0022] According to the natural neighbor features, the discrete bearing vibration signals with different characteristics are divided into multiple parts including abnormal points, noise points, majority class safe points, minority class safe points and minority class boundary points;

[0023] Eliminate abnormal points and noise points from discrete bearing vibration signals;

[0024] A hypersphere is constructed for the minority class boundary points, and the small hyperspheres completely contained in the large hypersphere are merged to form a hypersphere set;

[0025] According to the radius of each hypersphere, a corresponding sampling ratio is adaptively allocated to each hypersphere, and random oversampling is performed only within the hypersphere to obtain a balanced data set of discrete bearing vibration signals.

[0026] In one embodiment, sample points in a plurality of features are selected;

[0027] Calculate the Euclidean distance between each sample point and other sample points to build a distance matrix;

[0028] According to the distance matrix, search for the natural neighbors of each sample point to form the natural neighborhood;

[0029] If a first sample point does not belong to the natural neighborhood, the first sample point is regarded as an outlier;

[0030] If a first sample point belongs to the natural neighborhood, and a second sample point belongs to a neighbor of the first sample point, and the second sample point is different from the first sample point after adding a label, then the first sample point is regarded as a noise point;

[0031] If a first sample point belongs to a neighbor of the natural neighborhood, and a second sample point belongs to the neighbor, and the second sample point is different from the first sample point after adding a label, then the first sample point is regarded as a minority class safe point;

[0032] The set of majority class sample points after removing outliers and noise points is recorded as majority class safe points;

[0033] Among the minority class sample points contained in the natural neighborhood, noise points and minority class safe points are removed, and the remaining minority class sample points are regarded as minority class boundary points.

[0034] In one embodiment, according to the total number of samples and the radius of each hypersphere, the corresponding sampling ratio in each hypersphere is calculated, and the number of samples in each hypersphere is determined;

[0035] The balanced data set is formed by generating a specified number of new sample points in each hypersphere through random sampling.

[0036] In one embodiment, the corresponding sampling ratio in each hypersphere is expressed as:

[0037] in, Indicates the adoption ratio, , represents the radius of the hypersphere, Indicates the total number of adoptions.

[0038] In one embodiment, it also includes:

[0039] Perform data cleaning and preprocessing on the balanced data set;

[0040] The balanced data set is divided into a training set, a validation set, and a test set, and the proportion of each category in the training set, the validation set, and the test set is the same as the proportion of each category in the balanced data set.

[0041] In one embodiment, a vibration sensor is installed on the outer ring of the bearing, and the vibration signal collected by the vibration sensor is amplified and processed by analog-to-digital conversion to obtain a discrete bearing vibration signal.

[0042] On the other hand, the present invention also provides a new energy vehicle bearing defect detection device, adopting the above-mentioned new energy vehicle bearing defect detection method, the device comprises:

[0043] Acquisition module, used to obtain discrete bearing vibration signals under different fault conditions;

[0044] A feature extraction module is used to extract features from the discrete bearing vibration signal to obtain a number of features;

[0045] A data set construction module is used to perform hyperspherical oversampling of natural neighborhoods on the features to obtain a balanced data set;

[0046] The model training module is used to input the balance data set into a bearing defect detection model for training, and the bearing defect detection model outputs the predicted bearing defect.

[0047] On the other hand, the present invention also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above-mentioned new energy vehicle bearing defect detection method are implemented and the same technical effect can be achieved.

[0048] It can be seen from the above scheme that the advantages of the present invention are:

[0049] The present invention provides a new energy vehicle bearing defect detection method, which obtains discrete bearing vibration signals under different fault conditions; then extracts features from the discrete bearing vibration signals to obtain a number of features; and performs natural neighborhood hyperspherical oversampling on the several features to obtain a balanced data set. Finally, the balanced data set is input into a bearing defect detection model for training, and the bearing defect detection model outputs predicted bearing defects. The method uses a small sample of unbalanced data to perform new energy bearing detection, ensuring that new energy vehicle bearing detection is carried out efficiently and orderly. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1The overall process diagram of the new energy vehicle bearing defect detection method provided by the embodiment of the invention is shown;

[0051] Figure 2 Shows Figure 1 Detailed flow chart of step S2 in FIG.

[0052] Figure 3 Shows Figure 1 Detailed flow chart of step S3 in FIG.

[0053] Figure 4 A schematic diagram showing five regions divided by hyperspherical oversampling of natural neighbors; Figure 5 The overall architecture diagram of the new energy vehicle bearing defect detection device provided by an embodiment of the present invention is shown.

[0054] Among them, the symbols are:

[0055] 300-Bearing defect detection device;

[0056] 301-acquisition module;

[0057] 302-feature extraction module;

[0058] 303-dataset construction module;

[0059] 304-Model training module. DETAILED DESCRIPTION

[0060] In order to make the above features and effects of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings.

[0061] See also Figure 1-Figure 4 As shown, Figure 1 The overall process diagram of the new energy vehicle bearing defect detection method provided by one embodiment of the present invention is shown; Figure 2 Shows Figure 1 Detailed flow chart of step S2 in FIG. Figure 3 Shows Figure 1 Detailed flow chart of step S3 in FIG. Figure 4 A schematic diagram showing five regions divided by hyperspherical oversampling of natural neighbors.

[0062] See also Figure 1 As shown in, the new energy vehicle bearing defect detection method at least includes the following steps:

[0063] Step S1, obtaining discrete bearing vibration signals under different fault conditions.

[0064] Common rolling bearing fault diagnosis technologies are based on acoustic signals, vibration signals, and temperature signals, among which the majority are based on acquired vibration signals. The vibration signals generated by faulty rolling bearings during operation are characterized by nonlinearity and non-stationarity. When new energy vehicles are running, due to the structural characteristics of the bearings themselves, internal factors such as processing and assembly errors and faults that occur during operation, as well as external factors such as the movement and force of other parts on the transmission shaft, when the shaft runs at a certain speed and under a certain load, the vibration system composed of the bearing and the bearing seat or the housing is excited, causing the system to vibrate. In this regard, a vibration sensor is usually installed on the outer ring of the bearing, and the vibration sensor collects the bearing vibration signal, and then the bearing vibration signal is further amplified and analog-to-digital converted to obtain a discrete bearing vibration signal, so as to facilitate feature extraction of the discrete bearing vibration signal in subsequent steps.

[0065] Step S2: extract features from the discrete bearing vibration signal to obtain several features.

[0066] In one embodiment, the discrete bearing vibration signal is transformed by empirical wavelet transform to extract features, see Figure 2 As shown in Figure 2 The detailed flow diagram of step S2 is shown, and step S2 specifically includes:

[0067] First, the discrete bearing vibration signal is preprocessed (S21), that is, noise reduction is performed to improve signal quality and reduce the impact of noise on feature extraction.

[0068] Then, the preprocessed discrete bearing vibration signal is further decomposed into several IMF components (S22) through empirical mode decomposition. Specifically, the empirical mode decomposition method (EMD), based on instantaneous frequency and intrinsic mode function (IMF), can decompose the signal into several IMF components, each of which characterizes the local characteristics of the signal. Signal decomposition is based on the time scale characteristics of the data itself, without the need to pre-set any basis function, so it is adaptive. Constructing empirical mode functions based on common failure forms of rolling bearings such as "fracture, bonding, corrosion, fatigue, spalling, fatigue spalling", etc. can more effectively detect bearing problems.

[0069] In one embodiment, the decomposition process of step S22 is specifically as follows:

[0070] First, find all the maximum and minimum points of the preprocessed discrete bearing vibration signal x(t), and fit the maximum envelope and minimum envelope of these maximum and minimum points through cubic spline functions. It should be noted that all extreme points must be included in the upper and lower envelopes. Then, calculate the average value of the maximum envelope and the minimum envelope to obtain a mean signal, that is: ,in represents the maximum envelope, represents the minimum envelope, Represents the mean signal.

[0071] Then, based on the difference between the preprocessed discrete bearing vibration signal and the mean signal, the n-order IMF component is obtained. Specifically, the processed discrete bearing vibration signal x(t) is subtracted from the mean signal We get a new signal without low frequencies, namely: . Generally, this signal It is not a stationary signal and does not meet the two conditions defined by IMF. Repeat the above process. Assuming that after k times (k is generally less than 10), it meets the definition of IMF, then the first-order IMF component of the original discrete bearing vibration signal x(t) is expressed as: , repeat the process to get the second IMF component, and repeat this process until the nth order IMF component In addition, the difference between the discrete bearing vibration signal and each order IMF component is used to obtain a disturbance signal corresponding to the order IMF component. For example, the disturbance signal corresponding to the first-order IMF component is: .

[0072] Finally, the discrete bearing vibration signal is decomposed into the sum of the weighted n-order IMF component and the disturbance signal corresponding to the n-order IMF component, that is: .

[0073] In actual situations, the mean of the upper and lower envelopes cannot be zero. Usually, when the following formula is satisfied, the mean of the envelope is considered to satisfy the IMF mean zero condition: It is most reasonable to stop screening when the standard deviation SD value is between 0.2 and 0.3.

[0074] According to the above theoretical knowledge, the EMD of the obtained vibration signal is constructed to find the EMD empirical function of different faults.

[0075] In this embodiment, after the discrete bearing vibration signal is decomposed into several IMF components, each IMF component is further subjected to wavelet transform to obtain the frequency domain representation of each IMF component (S23). Specifically, in the empirical wavelet transform, an adaptive method is usually used to determine the wavelet basis function, that is, the wavelet basis function corresponding to each IMF. Different wavelet basis functions, such as Morlet wavelet, Mexican Hat wavelet, etc., can be used to capture different frequency and time characteristics of the signal. Since the vibration signal generated by the bearing damage of new energy vehicles is unbalanced, the wavelet basis functions selected in this embodiment are the Mexican Hat (mexh) wavelet function and the Daubechies (dbN) wavelet.

[0076] Among them, the Mexican Hat (mexh) wavelet function is also called the Mexican Hat wavelet or Ricker wavelet, which is a wavelet obtained by the second-order derivative of the Gaussian function. It is a symmetrical, non-compactly supported non-orthogonal wavelet with a spatial feature close to human vision. The Mexican Hat (mexh) wavelet mother function formula selected in this embodiment is as follows:

[0077] It is a Gaussian function The negative value of the second derivative of . The derived wavelet function is Among them, t is a parameter variable, and a and b are constants.

[0078] The Daubechies (dbN) wavelet function has good local properties and multi-resolution analysis capabilities, and has been widely used in signal processing, image processing, data compression, etc. Its basis function is composed of a set of orthogonal polynomials and a set of filters, which can decompose the signal into sub-signals of different scales, thereby realizing multi-resolution analysis.

[0079] To this end, in this embodiment, the determined Mexican Hat (mexh) wavelet function and Daubechies (dbN) wavelet are applied to each IMF component for wavelet transform. Wavelet transform decomposes each IMF component into a local spectrum on the frequency-time plane, reflecting the energy distribution of the vibration signal at different frequencies and time scales.

[0080] After the discrete bearing vibration signal x(t) undergoes a series of changes from step S21 to step S23, the wavelet coefficients of different frequency bands are extracted, that is, the features are obtained (S24). These wavelet coefficients contain localized feature information of the signal, such as frequency, phase, and amplitude, which are crucial for further analysis and processing of the signal.

[0081] Step S3: Perform hyperspherical oversampling of the natural neighborhood of the features to obtain a balanced data set.

[0082] In one embodiment, for step S3, refer to Figure 3 As shown in Figure 3 The detailed flow chart of step S3 is shown. First, for the discrete bearing vibration signal after feature extraction, natural neighbors are searched to form a natural neighborhood (S31). Specifically, in this step, a distance matrix can be constructed by selecting sample points in several features and calculating the Euclidean distance between each sample point and other sample points. And according to the distance matrix, the natural neighbors of each sample point are searched to form a stable natural neighborhood. Among them, the formula for calculating the Euclidean distance is as follows: in, is the Euclidean distance between the sample point (x2, y2) and the sample point (x1, y1), Represents the Euclidean distance from the sample point (x, y) to the origin.

[0083] Then, according to the natural neighbor feature, the discrete bearing vibration signals with different features are divided into multiple parts (S32) including abnormal points, noise points, majority class safety points, minority class safety points and minority class boundary points, see Figure 4 shown.

[0084] Specifically, when the data set forms a stable natural neighborhood, if a first sample point If it does not belong to the natural neighborhood NSS, the first sample point is regarded as an outlier, that is, .

[0085] After removing the outliers, if the first sample point Belongs to this natural neighborhood, for the first sample point Neighbors , Indicates the distance from the first sample point If there is a second sample point Belongs to the neighbors of the first sample point , and the second sample point is different from the first sample point after adding the label, that is, , , labels function represents the function of adding labels to each data point in the data set, then the first sample point Considered as noise points, that is: .

[0086] If the first sample point Belongs to this natural neighborhood of Neighbors ,Right now of Neighbors , and a second sample point belongs to the neighbor, and the second sample point is different from the first sample point after adding the label, that is, , , then the first sample point Considered as the minority class safe point D(safe_min).

[0087] In addition, the set of majority class sample points after removing outliers and noise points is recorded as majority class safe points D(safe_maj). Among the minority class sample points contained in the natural neighborhood, after removing noise points and minority class safe points, the remaining minority class sample points are regarded as minority class boundary points D(border_min).

[0088] After the above step S32 divides the abnormal points, noise points, majority class safety points, minority class safety points and minority class boundary points into multiple parts, the abnormal points and noise points are further removed from the discrete bearing vibration signal (S33). At the same time, a hypersphere is constructed for the minority class boundary point part, and the small hyperspheres completely contained in the large hypersphere are merged to form a hypersphere set; according to the radius of each hypersphere, a corresponding sampling ratio is adaptively assigned to each hypersphere, and random oversampling is performed only within the hypersphere (S34), and a specified number of new sample points are generated within each hypersphere by random sampling. After the above process, the data set after the elimination process in step S33 is merged with the data set after the oversampling process in S34 to form a balanced data set with enhanced boundary information, that is, a balanced data set of discrete bearing vibration signals is obtained.

[0089] In one embodiment, in step S34, the sampling ratio corresponding to each hypersphere is calculated based on the total number of samples and the radius of each hypersphere, and the number of samples in each hypersphere is determined. The sampling ratio corresponding to each hypersphere is expressed as: in, Expressed as the number of samples within the i-th hypersphere, , represents the radius of the hypersphere, For the sample point The distance between the farthest natural neighbors, Indicates the total number of adoptions, ,in: represents the imbalance ratio, It is expressed as the number of minority classes after removing outliers and noise points. is the number of majority classes to remove outliers and noise points.

[0090] In addition, in one embodiment, due to the hypersphere oversampling of the natural neighborhood, the unbalanced data set becomes a balanced data set. In order to better detect bearing faults, in one embodiment, a balanced database is further built on this basis to transform the unbalanced problem into a balanced problem. Specifically, the balanced data set is cleaned and preprocessed, and duplicate, erroneous or irrelevant records are removed through data cleaning; the data is preprocessed, such as standardization, normalization or encoding, so that it can be used in subsequent analysis and modeling. Then, the processed balanced data set is divided into a training set, a validation set and a test set. In order to ensure that the categories are balanced during the division process, stratified sampling and other methods can be used to ensure that the proportion of each category in the training set, validation set and test set is the same as the proportion of each category in the balanced data set. In addition, in the process of building a balanced database, the number of samples in each category is regularly monitored to ensure that they remain relatively balanced. Use appropriate performance metrics such as precision, recall, F1 score, AUC-ROC to evaluate the performance of the model and check whether there is any performance degradation due to class imbalance. Finally, based on the evaluation results, adjust the strategies of steps such as data collection, cleaning, preprocessing, and resampling to further optimize the balance of the database. Try to use different algorithms and models to train the data and compare their performance to select the most suitable bearing defect detection model. At the same time, record the process and strategy of building a balanced database in real time, and save the final balanced data set and bearing defect detection model.

[0091] Step S4: input the balanced data set into a bearing defect detection model for training, and the bearing defect detection model outputs the predicted bearing defect.

[0092] In this embodiment, the balanced data set constructed through steps S2-S3 is input into a preset bearing defect detection model, such as a deep learning model, and then the bearing defect detection model is trained to output the predicted bearing defect.

[0093] In summary, the new energy vehicle bearing defect detection method provided by the present invention obtains discrete bearing vibration signals under different fault conditions; then extracts features from the discrete bearing vibration signals to obtain a number of features; and performs hyperspherical oversampling of the natural neighborhood on the several features to obtain a balanced data set. Finally, the balanced data set is input into a bearing defect detection model for training, and the bearing defect detection model outputs predicted bearing defects. This method uses a small sample of unbalanced data to perform new energy bearing detection, ensuring that new energy vehicle bearing detection is carried out efficiently and orderly.

[0094] In addition, refer to Figure 5 , Figure 5 A new energy vehicle bearing defect detection device 300 is shown, which can be implemented by Figure 1The various processes of implementing the new energy vehicle bearing defect detection method are shown.

[0095] A new energy vehicle bearing defect detection device 300, adopting Figure 1 The new energy vehicle bearing defect detection method shown in , the device at least comprises:

[0096] The acquisition module 301 is used to obtain discrete bearing vibration signals under different fault conditions.

[0097] The feature extraction module 302 is used to extract features from the discrete bearing vibration signal to obtain a number of features.

[0098] The data set construction module 303 is used to perform hyperspherical oversampling of the natural neighborhood of the features to obtain a balanced data set.

[0099] The model training module 304 is used to input the balance data set into a bearing defect detection model for training, and the bearing defect detection model outputs the predicted bearing defect.

[0100] In addition, those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the module described can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0101] In addition, the embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by the processor, the steps of the above-mentioned new energy vehicle bearing defect detection method are implemented, and the same technical effect can be achieved. The readable storage medium can be a tangible medium, which can contain or store a program for use by an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. The readable storage medium can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0102] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be applied, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0103] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method.

[0104] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A new energy vehicle bearing defect detection method, characterized in that: Include: Obtain discrete bearing vibration signals under different fault conditions; The empirical wavelet transform is used to extract the features of the discrete bearing vibration signal and obtain several features, including: Performing signal preprocessing on the discrete bearing vibration signal; The preprocessed discrete bearing vibration signal is decomposed into several IMF components through empirical mode decomposition, including: Find all the maximum and minimum points of the preprocessed discrete bearing vibration signal, and fit the maximum envelope and minimum envelope of the maximum and minimum points respectively through the cubic spline function; Calculate the average value of the maximum envelope and the minimum envelope to obtain a mean value signal; Based on the difference between the preprocessed discrete bearing vibration signal and the mean signal, an n-order IMF component is obtained; The difference between the discrete bearing vibration signal and each order IMF component is used to obtain a disturbance signal corresponding to the order IMF component; Decomposing the discrete bearing vibration signal into a weighted sum of an n-th order IMF component and a disturbance signal corresponding to the n-th order IMF component; When the IMF mean is zero, the screening is stopped, where the IMF mean is zero as follows: Among them, when the value of standard deviation SD is between 0.2 and 0.3, is the first-order IMF component of the bearing vibration signal after K iterations, is the first-order IMF component of the bearing vibration signal after K-1 iterations; For each IMF component, wavelet transform is performed to obtain the frequency domain representation of each IMF component, where the wavelet basis function uses Mexican Hat wavelet function and Daubechies wavelet; Extract wavelet coefficients of different frequency bands to obtain the features; For these features, hyperspherical oversampling of natural neighborhoods is performed to obtain a balanced data set, including: For the discrete bearing vibration signal after feature extraction, search for natural neighbors to form a natural neighborhood; According to the natural neighbor features, the discrete bearing vibration signals with different characteristics are divided into multiple parts including abnormal points, noise points, majority class safe points, minority class safe points and minority class boundary points; Eliminate abnormal points and noise points from discrete bearing vibration signals; A hypersphere is constructed for the minority class boundary points, and the small hyperspheres completely contained in the large hypersphere are merged to form a hypersphere set; According to the radius of each hypersphere, a corresponding sampling ratio is adaptively allocated to each hypersphere, and random oversampling is performed only within the hypersphere to obtain a balanced data set of discrete bearing vibration signals; the balanced data set is input into a bearing defect detection model for training, and the bearing defect detection model outputs predicted bearing defects.

2. The method according to claim 1, characterized in that Select sample points from several features; Calculate the Euclidean distance between each sample point and other sample points to build a distance matrix; According to the distance matrix, search for the natural neighbors of each sample point to form the natural neighborhood; If a first sample point does not belong to the natural neighborhood, the first sample point is regarded as an outlier; If a first sample point belongs to the natural neighborhood, and a second sample point belongs to a neighbor of the first sample point, and the second sample point is different from the first sample point after adding a label, then the first sample point is regarded as a noise point; If a first sample point belongs to a neighbor of the natural neighborhood, and a second sample point belongs to the neighbor, and the second sample point is different from the first sample point after adding a label, then the first sample point is regarded as a minority class safe point; The set of majority class sample points after removing outliers and noise points is recorded as majority class safe points; Among the minority class sample points contained in the natural neighborhood, noise points and minority class safe points are removed, and the remaining minority class sample points are regarded as minority class boundary points.

3. The method according to claim 1, characterized in that According to the total number of samples and the radius of each hypersphere, the corresponding sampling ratio in each hypersphere is calculated, and the number of samples in each hypersphere is determined; The balanced data set is formed by generating a specified number of new sample points in each hypersphere through random sampling.

4. The method according to claim 3, characterized in that The corresponding sampling ratio in each hypersphere is expressed as: in, Indicates the adoption ratio, ri represents the radius of the hypersphere, and Nc represents the total number of samples.

5. The method according to claim 1, characterized in that: Also includes: Perform data cleaning and preprocessing on the balanced data set; The balanced data set is divided into a training set, a validation set, and a test set, and the proportion of each category in the training set, the validation set, and the test set is the same as the proportion of each category in the balanced data set.

6. The method according to claim 1, characterized in that A vibration sensor is installed on the outer ring of the bearing, and the vibration signal collected by the vibration sensor is amplified and processed by analog-to-digital conversion to obtain a discrete bearing vibration signal.

7. A new energy vehicle bearing defect detection device, characterized in that: The new energy vehicle bearing defect detection method according to any one of claims 1 to 6 is adopted, and the device comprises: Acquisition module, used to obtain discrete bearing vibration signals under different fault conditions; The feature extraction module is used to extract features of the discrete bearing vibration signal using empirical wavelet transform to obtain several features, including: Performing signal preprocessing on the discrete bearing vibration signal; The preprocessed discrete bearing vibration signal is decomposed into several IMF components through empirical mode decomposition, including: Find all the maximum and minimum points of the preprocessed discrete bearing vibration signal, and fit the maximum envelope and minimum envelope of the maximum and minimum points respectively through the cubic spline function; Calculate the average value of the maximum envelope and the minimum envelope to obtain a mean value signal; Based on the difference between the preprocessed discrete bearing vibration signal and the mean signal, an n-order IMF component is obtained; The difference between the discrete bearing vibration signal and each order IMF component is used to obtain a disturbance signal corresponding to the order IMF component; Decomposing the discrete bearing vibration signal into a weighted sum of an n-th order IMF component and a disturbance signal corresponding to the n-th order IMF component; When the IMF mean is zero, the screening is stopped, where the IMF mean is zero as follows: Among them, when the value of standard deviation SD is between 0.2 and 0.3, is the first-order IMF component of the bearing vibration signal after K iterations, is the first-order IMF component of the bearing vibration signal after K-1 iterations; For each IMF component, wavelet transform is performed to obtain the frequency domain representation of each IMF component, where the wavelet basis function uses Mexican Hat wavelet function and Daubechies wavelet; Extract wavelet coefficients of different frequency bands to obtain the features; The data set construction module is used to perform hyperspherical oversampling of the natural neighborhood of the features to obtain a balanced data set, including: For the discrete bearing vibration signal after feature extraction, search for natural neighbors to form a natural neighborhood; According to the natural neighbor features, the discrete bearing vibration signals with different characteristics are divided into multiple parts including abnormal points, noise points, majority class safe points, minority class safe points and minority class boundary points; Eliminate abnormal points and noise points from discrete bearing vibration signals; A hypersphere is constructed for the minority class boundary points, and the small hyperspheres completely contained in the large hypersphere are merged to form a hypersphere set; According to the radius of each hypersphere, a corresponding sampling ratio is adaptively allocated to each hypersphere, and random oversampling is performed only within the hypersphere to obtain a balanced data set of discrete bearing vibration signals; a model training module is used to input the balanced data set into a bearing defect detection model for training, and the bearing defect detection model outputs predicted bearing defects.

Images