Data augmentation method, fault detection model training method, device and medium

Abstract

The application relates to the technical field of data processing, and particularly provides a data enhancement method, a training method and device of a fault detection model and a medium, and aims to solve the problem of low accuracy of training samples in application. To this end, the data enhancement method comprises the following steps: obtaining actual operation data of a target bearing in a preset time period, obtaining target fault data from a preset fault information library based on the actual operation data, obtaining a proportion of the target fault data in the preset fault information library based on a fault type of the target fault data, if the proportion is less than or equal to a preset proportion threshold, performing dynamic mask enhancement processing on the actual operation data based on the target fault data to obtain sample enhancement features, generating a preset number of pieces of enhanced data based on the sample enhancement features and preset fault data generation parameters, and adding the enhanced data to the preset fault information library, so that the quality of the data can be greatly improved, and the use efficiency and accuracy of the data can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a data augmentation method, a training method for a fault detection model, an apparatus, and a medium. Background Technology

[0002] In recent years, fault detection models trained based on deep learning have achieved great success in fault type detection tasks. The performance of fault detection models largely depends on massive amounts of training samples; however, in practical applications, it is difficult to obtain a large number of effective training samples. In deep learning-based fault detection, the following situation exists: when the training dataset is small, the trained fault detection model achieves high accuracy on the training set, but relatively low accuracy on the test set or in real-world applications, indicating poor generalization ability.

[0003] Existing technologies primarily improve model accuracy on test sets or in real-world applications by augmenting training samples. For example, data augmentation based on a single parameter in the training samples, or using fixed augmentation rules, can lead to significant discrepancies between the augmented training samples and the real samples, thus affecting the accuracy of fault detection models. Furthermore, fixed augmentation rules cannot adapt to dynamically changing fault characteristics, causing feature distortion and further impacting the detection performance of fault detection models. Summary of the Invention

[0004] This application addresses the shortcomings of the prior art by proposing a data augmentation method, a fault detection model training method, an apparatus, and a medium.

[0005] In a first aspect, embodiments of this application provide a data augmentation method, comprising: acquiring actual operating data of a target bearing over a preset time period; acquiring target fault data from a preset fault information database based on the actual operating data, the target fault data including fault types; acquiring the proportion of the target fault data in the preset fault information database based on the fault types; if the proportion is less than or equal to a preset proportion threshold, performing dynamic masking augmentation processing on the actual operating data based on the target fault data to obtain sample augmentation features; generating a preset number of augmented data based on the sample augmentation features and preset fault data generation parameters; and adding the augmented data to the preset fault information database.

[0006] Secondly, embodiments of this application provide a training method for a fault detection model, comprising: acquiring target data, wherein the target data is acquired from a preset fault information database in the data augmentation method described in the first aspect; dividing the target data into a training set, a validation set, and a test set; training and validating the fault detection model using the training set and the validation set; testing the validated fault detection model using the test set, and obtaining the trained fault detection model based on the test results.

[0007] Thirdly, embodiments of this application provide a data enhancement device, comprising: a first acquisition module, configured to acquire actual operating data of a target bearing over a preset time period; a second acquisition module, configured to acquire target fault data from a preset fault information database based on the actual operating data, the target fault data including fault types; a third acquisition module, configured to acquire the proportion of the target fault data in the preset fault information database based on the fault types; a dynamic mask enhancement processing module, configured to perform dynamic mask enhancement processing on the actual operating data based on the target fault data if the proportion is less than or equal to a preset proportion threshold, to obtain sample enhancement features; a generation module, configured to generate a preset number of enhanced data entries based on the sample enhancement features and preset fault data generation parameters; and a data addition module, configured to add the enhanced data to the preset fault information database.

[0008] Fourthly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method described in the first aspect above.

[0009] Fifthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0010] The technical solutions provided in this application embodiment have at least the following technical effects or advantages:

[0011] This application embodiment obtains actual operating data of the target bearing over a preset time period. Based on the actual operating data, it retrieves target fault data from a preset fault information database. The target fault data includes fault types. Then, based on the fault types, it obtains the proportion of the target fault data in the preset fault information database. If the proportion is less than or equal to a preset proportion threshold, it performs dynamic masking enhancement processing on the actual operating data based on the target fault data to obtain sample enhancement features. Based on the sample enhancement features and preset fault data generation parameters, it generates a preset number of enhanced data entries and adds them to the preset fault information database. By enhancing the minority classes with a small proportion, it avoids the data imbalance caused by the excessive expansion of majority class samples. The system dynamically determines whether enhancement is needed based on the real-time proportion of each type of fault in the fault information database, enabling the system to adapt to changes in data distribution. Furthermore, this application can also selectively enhance and generate a smaller number of minority class samples, ensuring that the model is not biased towards majority class faults when using data from the fault information database for training, thereby improving the recognition rate of rare faults and greatly enhancing the efficiency of data utilization.

[0012] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0013] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0014] Figure 1 A flowchart of a data augmentation method provided in an embodiment of this application is shown;

[0015] Figure 2 A flowchart illustrating a training method for a fault detection model provided in an embodiment of this application is shown;

[0016] Figure 3 This paper shows a schematic diagram of the structure of a data enhancement device provided in an embodiment of this application;

[0017] Figure 4 A schematic diagram of the structure of a training device for a fault detection model provided in an embodiment of this application is shown;

[0018] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0019] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of this application are shown in the drawings, it should be understood that this application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0020] In recent years, fault detection models trained based on deep learning have achieved great success in fault type detection tasks. The performance of fault detection models largely depends on massive amounts of training samples; however, in practical applications, it is difficult to obtain a large number of effective training samples. In deep learning-based fault detection, the following situation exists: when the training dataset is small, the trained fault detection model achieves high accuracy on the training set, but relatively low accuracy on the test set or in real-world applications, indicating poor generalization ability.

[0021] Existing technologies primarily improve model accuracy on test sets or in real-world applications by augmenting training samples. For example, data augmentation based on a single parameter in the training samples, or using fixed augmentation rules, can lead to significant discrepancies between the augmented training samples and the real samples, thus affecting the accuracy of fault detection models. Furthermore, fixed augmentation rules cannot adapt to dynamically changing fault characteristics, causing feature distortion and further impacting the detection performance of fault detection models.

[0022] Based on this, embodiments of this application provide a data augmentation method. The specific solutions of these embodiments are described below with reference to the accompanying drawings.

[0023] See Figure 1 The flowchart shown represents a data augmentation method, which specifically includes the following steps:

[0024] Step 101: Obtain the actual operating data of the target bearing over a preset time period.

[0025] In this embodiment, the actual operating data refers to the data collected when the target bearing is running in a real-world application environment. The actual operating data may include the actual bearing speed, actual vibration signal, actual temperature signal, and actual acoustic signal of the target bearing, etc. Specifically, this actual operating data is time-series data; that is, the actual vibration signal, actual temperature signal, and actual acoustic signal of the target bearing are all time-series data. Specifically, the target vibration signal includes target vibration signals corresponding to multiple time points, the target temperature signal includes target temperature signals corresponding to multiple time points, and the target acoustic signal includes target acoustic signals corresponding to multiple time points.

[0026] The actual vibration signal refers to the signal of mechanical vibration generated by the internal structure of the target bearing due to force during operation.

[0027] The actual temperature signal refers to the real-time temperature value of the target bearing during operation.

[0028] Actual acoustic signal refers to the sound signal generated during the operation of the target bearing.

[0029] In one embodiment, sensors can be used to acquire actual operating data of the target bearing over a preset time period. Specifically, a vibration acceleration sensor can be used to acquire the actual vibration signal of the target bearing over a preset time period, an acoustic sensor or acoustic emission sensor can be used to acquire the actual acoustic signal of the target bearing over a preset time period, a temperature sensor can be used to acquire the actual temperature signal of the target bearing over a preset time period, and a speed sensor can be used to acquire the actual bearing speed of the target bearing over a preset time period.

[0030] Step 102: Based on actual operating data, obtain target fault data from a preset fault information database. The target fault data includes fault types.

[0031] In one embodiment, the preset fault information database can be constructed before obtaining target fault data from the preset fault information database based on actual operating data. Specifically, constructing the preset fault information database includes: pre-collecting test operating data such as preset vibration signals, preset temperature signals, and preset acoustic signals corresponding to different preset bearing speeds, different test bearing components, different fault degrees, and different fault stages of the test bearing; then, based on the test operating data, calculating the preset fault frequencies of each bearing component corresponding to the test bearing and the preset fault evaluation parameters corresponding to the test bearing; wherein the preset fault evaluation parameters include the preset peak-to-peak value, preset fractal dimension, preset impact characteristics, and preset spectral density corresponding to the preset vibration signals; finally, establishing the preset fault information database based on the test operating data, the preset bearing speeds and preset fault frequencies of each bearing component, and the preset fault evaluation parameters corresponding to the test bearing.

[0032] It should be noted that the test operation data is time series data. That is, the preset vibration signals, preset temperature signals and preset acoustic signals corresponding to different preset bearing speeds, different test bearing components, different fault degrees and different fault stages are all time series data. In other words, the preset vibration signals include preset vibration signals corresponding to multiple time points, the preset temperature signals include preset temperature signals corresponding to multiple time points, and the preset acoustic signals include preset acoustic signals corresponding to multiple time points.

[0033] Based on the above implementation, in some modified implementations, the test bearing assembly may include the inner ring, outer ring, and rolling elements of the test bearing. The preset failure frequency of each bearing assembly corresponding to the test bearing is calculated respectively. Specifically, it may be: based on the preset bearing speed and geometric parameters of the test bearing, the preset failure frequency of the inner ring of the test bearing is calculated by the following formula (1), the preset failure frequency of the outer ring of the test bearing is calculated by the following formula (2), and the preset failure frequency of the rolling elements of the test bearing is calculated by the following formula (3). The preset bearing geometric parameters include the bearing pitch diameter, contact angle, number of rolling elements, and rolling element diameter of the preset bearing.

[0034]

[0035] in, This refers to the preset failure frequency for testing the inner ring of the bearing. This refers to the preset failure frequency for testing the outer ring of the bearing. This refers to the preset fault frequency of the rolling elements of the test bearing, r refers to the preset bearing speed of the test bearing, n refers to the number of rolling elements of the test bearing, d refers to the diameter of the rolling elements of the test bearing, D refers to the bearing pitch diameter of the test bearing, and α refers to the contact angle of the test bearing.

[0036] Based on the above implementation method, in some modified implementation methods, the preset fault assessment parameters corresponding to the test bearing are calculated respectively. Specifically, it can be: based on the preset vibration signals corresponding to different preset bearing speeds, different test bearing components, different fault degrees, and different fault stages of the test bearing, the preset peak-to-peak value of the test bearing is calculated by the following formula (4), the preset fractal dimension of the test bearing is calculated by the following formula (5), the preset impact characteristics of the test bearing are calculated by the following formula (6), and the preset spectral density of the test bearing is calculated by the following formula (7).

[0037]

[0038] Where PP refers to the preset peak-to-peak value of the test bearing, D refers to the preset fractal dimension of the test bearing, I refers to the preset impact characteristic of the test bearing, and S(k) refers to the preset spectral density at the k-th frequency point of the test bearing. This refers to the preset vibration signal corresponding to time point t. This refers to the preset length of the vibration signal to cover the preset vibration signal, and N refers to the number of times the preset vibration signal is covered. This number is the ratio of the total length of the vibration signal to the preset length of the vibration signal to cover the preset vibration signal. This refers to the average value of the preset vibration signal. This refers to the standard deviation of the preset vibration signal. This refers to the preset sampling frequency, M refers to the preset total number of frequency points, W refers to the window function normalization factor, X(k) refers to the amplitude of the Fourier transform, w(t) refers to the Hanning window function, j refers to the imaginary unit, and k refers to the time point index of the fault frequency, where k = 0, 1, 2, … , M-1.

[0039] Based on the above implementation method, in some modified implementation methods, a preset fault information database is established according to the test operation data, the preset bearing speed and preset fault frequency of each bearing assembly, and the preset fault evaluation parameters corresponding to the test bearing. Specifically, it can be: based on the preset test bearing assembly corresponding to the test bearing, the fault type of each bearing assembly is determined, and a corresponding mapping relationship is established between the test operation data, the preset bearing speed and preset fault frequency of each bearing assembly, and the preset fault evaluation parameters corresponding to the test bearing, thereby obtaining the preset fault information database shown in Table (1) below:

[0040]

[0041] Table (1)

[0042] This application's embodiments establish a benchmark database that comprehensively covers the fault evolution process by pre-collecting test operation data under multiple operating conditions (different speeds, components, fault degrees and stages) and multiple modes (vibration, temperature, acoustic signals), significantly improving the adaptability and generalization ability of the fault detection model to complex industrial scenarios. Secondly, based on physical mechanisms, preset fault frequencies and multi-dimensional fault assessment parameters (such as peak-to-peak value, fractal dimension, impact characteristics, and spectral density) are extracted and associated, achieving a precise mapping from raw data to interpretable fault characteristics. This not only enhances the credibility of diagnostic results but also provides quantitative basis for fault location, severity assessment, and early warning. Finally, this structured knowledge base can effectively support supervised learning and transfer learning, reduce dependence on the number of on-site fault samples, accelerate model deployment, and form reusable enterprise digital assets. This promotes the intelligent transformation of equipment operation and maintenance from passive response to proactive prediction, demonstrating outstanding technical advantages in improving diagnostic accuracy, reducing false alarms and missed alarms, and extending equipment life.

[0043] Based on the above implementation, in some modified implementations, the actual operating data may include the actual bearing speed and the actual vibration signal. Based on the actual operating data, target fault data is obtained from a preset fault information database. Specifically, this may involve: calculating the actual fault frequency of each target bearing component based on the actual bearing speed and the geometric dimensions of the target bearing; then, calculating the actual fault assessment parameters of the target bearing based on the actual vibration signal; and finally, obtaining the target fault data from the preset fault information database based on the actual fault frequency and actual fault assessment parameters of each target bearing component.

[0044] It should be noted that the geometric parameters of the target bearing may include the bearing pitch diameter, contact angle, number of rolling elements and rolling element diameter. Each target bearing assembly may include the inner ring, outer ring and rolling elements. Actual fault assessment parameters may include the actual peak-to-peak value, actual fractal dimension and actual impact characteristics corresponding to the actual vibration signal.

[0045] Furthermore, based on the actual bearing speed and the geometric dimensions of the target bearing, the actual failure frequency of each target bearing component of the target bearing can be calculated. Formulas (1), (2) and (3) in the above embodiments can be used to calculate the actual failure frequency of the inner ring, outer ring and rolling elements of the target bearing respectively.

[0046] Furthermore, based on the actual vibration signal, the actual fault assessment parameters of the target bearing are calculated, that is, based on the actual vibration signal, the actual peak-to-peak value, actual fractal dimension and actual impact characteristics of the target bearing are calculated. Formulas (4), (5) and (6) in the above embodiments can be used to calculate them respectively.

[0047] Based on the above implementation methods, in some modified implementation methods, target fault data is obtained from a preset fault information database based on the actual fault frequency and actual fault assessment parameters of each target bearing assembly. Specifically, this may include: calculating the similarity between the actual operating data and each preset fault data based on the actual fault frequency and actual fault assessment parameters of each target bearing assembly and the preset fault frequency and preset fault assessment parameters of each preset fault data in the preset fault information database; and then determining the target fault data based on the preset fault data corresponding to the minimum similarity and the actual fault frequency and actual fault assessment parameters.

[0048] It should be noted that, based on the above implementation method for constructing a preset fault information database, the preset fault information database includes multiple preset fault data. Each preset fault data includes fault type, fault degree, fault stage, preset bearing speed, preset fault frequency, and preset fault evaluation parameters. The preset fault evaluation parameters include preset peak-to-peak value, preset fractal dimension, and preset impact characteristics, etc.

[0049] Based on the above implementation methods, in some modified implementation methods, the similarity between the actual operating data and each preset fault data is calculated based on the actual fault frequency and actual fault assessment parameters of each target bearing assembly and the preset fault frequency and preset fault assessment parameters of each preset bearing assembly in the preset fault information database. Specifically, the similarity can be calculated using the following formula (8): the actual fault frequency, actual peak-to-peak value, actual fractal dimension, and actual impact characteristics of each target bearing assembly, and the preset fault frequency, preset peak-to-peak value, preset fractal dimension, and preset impact characteristics of each preset bearing assembly in the preset fault information database.

[0050]

[0051] Here, Sim refers to similarity. This refers to the actual failure frequency. This refers to the preset fault frequency. This refers to the actual peak-to-peak value. This refers to the preset peak-to-peak value. This refers to the actual fractal dimension. This refers to the preset fractal dimension. This refers to actual impact characteristics. This refers to the preset impact characteristics. This refers to the preset first weight. This refers to the preset second weight. This refers to the preset third weight. This refers to the preset fourth weight.

[0052] It should be noted that, , , and All weights can be pre-set by those skilled in the art according to actual needs, or they can be weights obtained by adjusting pre-set weights according to actual needs. This application's embodiments do not impose specific limitations. Furthermore, , , and The total weight is 1.

[0053] Based on the above implementation methods, in some modified implementation methods, target fault data is determined based on the preset fault data corresponding to the minimum similarity, the actual fault frequency, and the actual fault evaluation parameters. Specifically, this can be done by: determining the final bearing assembly based on the fault type in the preset fault data corresponding to the minimum similarity, then determining the final bearing assembly as the target bearing assembly, using the fault degree in the preset fault data corresponding to the minimum similarity as the target fault degree, using the fault stage in the preset fault data corresponding to the minimum similarity as the target fault stage, using the actual fault frequency as the target fault frequency, and using the actual fault evaluation parameters as the target fault evaluation parameters, i.e., using the actual peak-to-peak value as the target peak-to-peak value, the actual fractal dimension as the target fractal dimension, the actual impact characteristics as the target impact characteristics, and the actual spectral density as the target spectral density. The target bearing assembly, the fault type, the target fault degree, the target fault stage, the target fault frequency, and the target fault evaluation parameters in the preset fault data corresponding to the minimum similarity are then merged to obtain the target fault data.

[0054] This application's embodiments transform actual bearing speeds and geometric parameters from practical applications into actual fault frequencies with clear physical meaning. These frequencies are then combined with actual fault assessment parameters extracted from actual vibration signals, achieving a multi-dimensional and refined description of equipment status and providing a comprehensive information foundation for fault diagnosis. Furthermore, by introducing a similarity calculation model, real-time operational data is intelligently compared with a vast number of benchmark cases in a pre-set fault information database, ensuring the scientific rigor and objectivity of the initial fault identification process and effectively reducing reliance on operator experience. Finally, an innovative decision-making mechanism based on minimum similarity is adopted. By systematically integrating matching results, original observation parameters, and historical fault knowledge, a complete diagnostic conclusion is formed, encompassing specific fault components, types, degrees, and stages. This not only achieves precise fault location and severity assessment, significantly improving the accuracy and interpretability of diagnostic results, but also provides direct and reliable quantitative evidence for subsequent predictive maintenance decisions, thereby enhancing the overall intelligence level of the fault detection system.

[0055] Step 103: Based on the fault type, obtain the proportion of the target fault data in the preset fault information database.

[0056] In one implementation, the number of data of a certain fault type in a preset fault information database can be obtained based on the fault type, and the ratio of this number to the total number of data in the preset fault information database can be used as the proportion of the target fault data in the preset fault information database.

[0057] Step 104: If the proportion is less than or equal to the preset proportion threshold, then based on the target fault data, perform dynamic masking enhancement processing on the actual operating data to obtain sample enhancement features.

[0058] In this embodiment, the preset percentage threshold is a boundary value used to determine whether to perform solid-state mask enhancement processing on the actual running data. The preset percentage threshold can be a percentage threshold preset by someone skilled in the art according to actual needs, or it can be a percentage threshold obtained by adjusting a preset percentage threshold according to actual needs. This embodiment does not impose specific limitations. For example, the preset percentage threshold can be preset to 0.15, meaning that if the percentage is less than or equal to 0.15, then dynamic mask enhancement processing is performed on the actual running data.

[0059] In one embodiment, the target fault data may include the target bearing assembly, fault type, target fault degree, target fault stage, target fault frequency, and target fault assessment parameters. Based on the target fault degree, a dynamic mask ratio can be determined. Then, based on the target fault assessment parameters, the enhancement weight of the actual operating data can be determined. Finally, based on the dynamic mask ratio and the enhancement weight of the actual operating data, dynamic mask enhancement processing is performed on the actual operating data to obtain sample enhancement features.

[0060] Based on the above implementation method, in some modified implementation methods, the dynamic mask ratio is determined based on the target fault degree. Specifically, it can be: pre-establish a mapping relationship between each preset fault degree setting and the dynamic mask ratio, and then match the target fault degree with the preset fault degree. If the match is successful, the corresponding dynamic mask ratio is obtained.

[0061] It should be noted that the preset fault severity can include minor, moderate, and severe. For example, the mapping relationship between each preset fault severity setting and the dynamic mask ratio can be shown in the following table (2):

[0062]

[0063] Table (2)

[0064] Based on the above implementation method, in some modified implementation methods, the dynamic mask ratio is determined based on the target fault degree. Specifically, the dynamic mask ratio can be determined using the following formula (9) based on the target fault degree:

[0065]

[0066] Where M represents the dynamic mask ratio.

[0067] Based on the above implementation methods, in some modified implementation methods, the enhancement weights of the actual operating data are determined based on the target fault assessment parameters. Specifically, the enhancement weights corresponding to the actual vibration signals are calculated using the following formula (10) based on the target impact characteristics and the actual temperature signal, the second enhancement weights corresponding to the actual temperature signals are calculated using the following formula (11), and the third enhancement weights corresponding to the actual acoustic signals are calculated using the following formula (12).

[0068]

[0069] in, This refers to the first enhanced weight. This refers to the second increase in weight. This refers to the third step of increasing body weight. This refers to the target's impact characteristics. This refers to the preset impact characteristic threshold. This refers to the actual temperature signal. This refers to the preset temperature threshold. This refers to the preset first adjustment coefficient. This refers to the preset second adjustment coefficient.

[0070] Based on the above implementation methods, in some modified implementation methods, dynamic mask enhancement processing is performed on the actual operating data based on the dynamic mask ratio and the enhancement weights of the actual operating data to obtain sample enhancement features. Specifically, this can be done by: converting the one-dimensional actual vibration signal in the actual operating data into a two-dimensional actual vibration time-spectrum image; normalizing the actual vibration time-spectrum image, the actual temperature signal, and the actual sound signal in the actual operating data to obtain a normalized actual vibration time-spectrum image, a normalized actual temperature signal, and a normalized actual sound signal; aligning the normalized actual temperature signal and the normalized actual sound signal with the normalized actual vibration time-spectrum image based on the timestamp of the normalized actual vibration time-spectrum image; then fusing the aligned normalized actual vibration time-spectrum image, the normalized actual temperature signal, and the normalized actual sound signal based on the first enhancement weight corresponding to the actual vibration signal, the second enhancement weight corresponding to the actual temperature signal, and the third enhancement weight corresponding to the actual acoustic signal to obtain a fusion matrix; and then performing dynamic mask enhancement processing on the actual operating data based on the fusion matrix and the dynamic mask ratio to obtain sample enhancement features.

[0071] Based on the above implementation, in some modified implementations, the one-dimensional actual vibration signal in the actual operating data is converted into a two-dimensional actual vibration time-frequency image. Specifically, the complex Morlet wavelet transform algorithm can be used, and the actual vibration signal is used to calculate the actual vibration time-frequency coefficients using the following formula (13). Based on the actual vibration time-frequency coefficients, the actual vibration time-frequency image is calculated:

[0072]

[0073] in, This refers to the spectral coefficients during actual vibration, and 'a' refers to the preset scale parameter. This refers to the preset displacement parameters. This refers to the complex Morlet wavelet conjugate function. This refers to the actual vibration signal at time point t, and d refers to the operation symbol, i.e., taking the integral over t.

[0074] It should be noted that the scale parameter 'a' can be a scale parameter preset by those skilled in the art according to actual needs, or it can be a scale parameter obtained by those skilled in the art after adjusting the preset scale parameter according to actual needs. This application embodiment does not impose specific limitations. The value range of the scale parameter 'a' in this application embodiment can be set to any value from 1 to 64.

[0075] Displacement parameters The displacement parameters can be preset by those skilled in the art according to actual needs, or they can be obtained by those skilled in the art by adjusting preset displacement parameters according to actual needs. This application does not specifically limit the specific displacement parameters. The displacement parameters in this application embodiment... The value can be set to any number between 1 and 2048.

[0076] Furthermore, based on the spectral coefficients during actual vibration, the spectral image during actual vibration is calculated. Specifically, based on the spectral coefficients during actual vibration, the spectral image during actual vibration is calculated using the following formula (14):

[0077]

[0078] in, This refers to the spectral image during actual vibration. This refers to the real part of the spectral coefficients during actual vibration. This refers to the imaginary part of the spectral coefficients during actual vibration.

[0079] Based on the above implementation, in some modified implementations, the actual vibration spectrum image, the actual temperature signal and the actual sound signal in the actual operating data are normalized to obtain the normalized actual vibration spectrum image, the normalized actual temperature signal and the normalized actual sound signal. Specifically, the following formula (15) can be used to normalize the actual vibration spectrum image, the actual temperature signal and the actual sound signal:

[0080]

[0081] Where x represents the spectrum image during actual vibration, This represents the spectrum image of the normalized actual vibration. This represents the smallest image in the normalized actual vibration frequency spectrum image corresponding to each time point. This represents the maximum value in the normalized actual vibration frequency spectrum at each time point; when x represents the actual temperature signal... This represents the normalized actual temperature signal. This represents the minimum temperature in the actual temperature signal corresponding to each time point. This represents the maximum temperature in the actual temperature signal corresponding to each time point; when x represents the actual sound signal... This represents the normalized actual sound signal. This represents the minimum sound in the actual sound signal corresponding to each time point. This represents the maximum volume in the actual sound signal at each point in time.

[0082] Based on the above implementation, in some modified implementations, the normalized actual temperature signal and the normalized actual sound signal are aligned with the normalized actual vibration spectrum image based on the timestamp of the normalized actual vibration spectrum image. Specifically, the actual temperature signal and the actual sound signal are time-aligned using a linear interpolation method based on the timestamp of the normalized actual vibration spectrum image to ensure that the same time window contains three types of modal data.

[0083] Based on the above implementation, in some modified implementations, the normalized actual vibration time-frequency image, normalized actual temperature signal, and normalized actual sound signal are fused based on the first enhancement weight corresponding to the actual vibration signal, the second enhancement weight corresponding to the actual temperature signal, and the third enhancement weight corresponding to the actual acoustic signal, to obtain a fusion matrix. Specifically, based on the first enhancement weight corresponding to the actual vibration signal, the second enhancement weight corresponding to the actual temperature signal, and the third enhancement weight corresponding to the actual acoustic signal, the normalized actual vibration time-frequency image, normalized actual temperature signal, and normalized actual sound signal are fused using the following formula (16) to obtain a fusion matrix:

[0084]

[0085] in, This refers to the fusion matrix. This refers to the first enhanced weight. This refers to the second increase in weight. This refers to the third step of increasing body weight. This refers to the normalized spectrum image matrix of the actual vibration. This refers to the normalized actual temperature signal. This refers to the normalized actual sound signal, and E refers to the pre-set dimension expansion matrix.

[0086] Based on the above implementation method, in some modified implementation methods, dynamic masking enhancement processing is performed on the actual operating data based on the fusion matrix and the dynamic masking ratio to obtain sample enhancement features. Specifically, it can be as follows: a masking matrix is ​​constructed based on the fusion matrix, the dynamic masking ratio and the target fault frequency, and then dynamic masking enhancement processing is performed on the fusion matrix based on the masking matrix to obtain sample enhancement features.

[0087] Furthermore, based on the fusion matrix, the dynamic mask ratio, and the target fault frequency, a mask matrix is ​​constructed. Specifically, this can be done by: determining the number of elements in the mask matrix based on the length and width of the fusion matrix and the dynamic mask ratio; determining the row and column indices of the target fault frequency in the normalized actual vibration spectrum image matrix; and selecting elements of the mask matrix that correspond to the number of elements in the normalized actual vibration spectrum image matrix based on the row and column indices to obtain the mask matrix.

[0088] Furthermore, based on the length and width of the fusion matrix and the dynamic mask ratio, the number of elements in the mask matrix is ​​determined. Specifically, the product of the length, width, and dynamic mask ratio of the fusion matrix can be used as the number of elements in the mask matrix.

[0089] Furthermore, based on the mask matrix, dynamic mask enhancement processing is performed on the fusion matrix to obtain sample enhancement features. Specifically, the mask matrix and the fusion matrix are multiplied element-wise to obtain the sample enhancement features.

[0090] This application embodiment utilizes data augmentation processing of three modalities simultaneously: actual vibration frequency spectrum image, actual temperature signal, and actual sound signal. This comprehensively covers the multifaceted manifestations of bearing faults, avoids the diagnostic blind spots of single-modal data, ensures the consistency of data from different modalities in physical time through timestamp alignment, and eliminates the influence of different dimensions and magnitudes through normalization processing, making subsequent feature fusion and model training more stable and effective.

[0091] Furthermore, by dynamically determining the first enhancement weight corresponding to the actual vibration signal, the second enhancement weight corresponding to the actual temperature signal, and the third enhancement weight corresponding to the actual acoustic signal, the enhanced data can provide high-quality training data for subsequent deep learning models. This greatly reduces the complexity of the model aligning, associating, and extracting features from multiple original data streams, and improves the model's learning efficiency for key features.

[0092] Step 105: Generate a preset number of augmented data based on the sample augmentation features and preset fault data generation parameters.

[0093] In this embodiment, the preset fault data generation parameters refer to the data tags required when generating enhanced data. The preset fault data generation parameters can be generation parameters pre-set by those skilled in the art, or generation parameters obtained by those skilled in the art after adjusting pre-set generation parameters according to actual needs. This embodiment does not impose specific limitations. For example, the preset fault data generation parameters can be data tags generated from data in the preset fault information database shown in Table (1) of the above-described embodiments of this application. For example, the fault type tag can be 1, the fault severity tag can be 2, ..., and so on.

[0094] The preset number can be a number pre-set by those skilled in the art according to actual needs, or it can be a number obtained by adjusting the preset number according to actual needs. This application embodiment does not make a specific limitation. For example, the preset number can be set such that after the number of generated enhanced data is added to the preset fault information database, the deviation of the total number of its corresponding fault types from the total number of other fault types is less than or equal to 5%.

[0095] In one implementation, sample enhancement features and preset fault data generation parameters can be input into a pre-trained enhanced sample generation model, and the enhanced sample generation model can be used to generate a preset number of enhanced data.

[0096] In this embodiment of the application, by selectively supplementing minority fault data, the generation efficiency of minority fault data can be greatly improved. This can effectively solve the problem that when using data in the preset fault information database as training data, the training data is insufficient and the fault types are unbalanced, resulting in low detection accuracy of the trained fault detection model. As a result, the generalization ability of the fault detection model under complex working conditions such as variable speed-load, high temperature and strong noise is significantly improved.

[0097] Step 106: Add the enhanced data to the preset fault information database.

[0098] Based on the above implementation, in some modified implementations, before adding the enhanced data to the preset fault information database, the method further includes: obtaining the frequency matching degree between the target fault frequency of the target fault data and the enhanced fault frequency of the enhanced data, and / or obtaining the impact characteristic similarity between the target impact characteristic of the target fault data and the enhanced impact characteristic of the enhanced data, and / or obtaining the fractal dimension consistency between the target fractal dimension of the target fault data and the enhanced fractal dimension of the enhanced data. Then, based on the frequency matching degree and / or impact characteristic similarity and / or fractal dimension consistency, the method determines whether the enhanced data is qualified.

[0099] Furthermore, the frequency matching degree between the target fault frequency of the target fault data and the enhanced fault frequency of the enhanced data can be obtained by using the following formula (17):

[0100]

[0101] Here, FFM refers to frequency matching degree. This refers to the enhanced spectral density corresponding to the enhanced fault frequency at the k-th time point in the enhanced data. This refers to the index of time points that increase the frequency of failures. This refers to the target spectral density corresponding to the target fault frequency at the k-th time point in the target fault data. This refers to the index of the time point of the target failure frequency. This refers to the preset fault frequency range, and d refers to the operation symbol, i.e., the operation on... , Calculate the points.

[0102] Furthermore, the similarity of the impact characteristics of the target impact characteristics of the target fault data and the enhanced impact characteristics of the enhanced data can be obtained by using the following formula (18) to calculate the similarity of the impact characteristics of the target impact characteristics of the target fault data and the enhanced impact characteristics of the enhanced data:

[0103]

[0104] Here, ICS refers to Impact Similarity. This refers to enhancing impact characteristics. This refers to the target impact characteristics, where N represents the total number of time points representing the target impact characteristics. This refers to the enhanced impact characteristics corresponding to the i-th time point. This refers to the target impact characteristics at the i-th time point.

[0105] Furthermore, to obtain the fractal dimension consistency between the target fractal dimension of the target fault data and the enhanced fractal dimension of the augmented data, specifically, the fractal dimension consistency between the target fractal dimension of the fault data and the enhanced fractal dimension of the augmented data can be calculated using the following formula (19):

[0106]

[0107] Here, FDC refers to fractal dimension uniformity. This refers to enhancing the fractal dimension. This refers to the target fractal dimension.

[0108] Based on the above implementation methods, in some modified implementation methods, the qualification of the enhancement data is determined based on the frequency matching degree and / or the impact characteristic similarity and / or the fractal dimension consistency. Specifically, this can be as follows: determine whether the frequency matching degree is greater than or equal to a preset frequency matching degree threshold; if the frequency matching degree is greater than or equal to the frequency matching degree threshold, the enhancement data is deemed qualified; otherwise, the enhancement data is deemed unqualified. And / or, determine whether the impact characteristic similarity is greater than or equal to a preset impact characteristic similarity threshold; if the impact characteristic similarity is greater than or equal to the impact characteristic similarity threshold, the enhancement data is deemed qualified; otherwise, the enhancement data is deemed unqualified. And / or, determine whether the fractal dimension consistency is greater than or equal to a preset fractal dimension consistency threshold; if the fractal dimension consistency is greater than or equal to the fractal dimension consistency threshold, the enhancement data is deemed qualified; otherwise, the enhancement data is deemed unqualified.

[0109] Furthermore, if the enhanced data is qualified, it is added to the preset fault information database; if the enhanced data is unqualified, a preset number of enhanced data entries are regenerated based on the sample enhancement features and the preset fault data generation parameters.

[0110] Based on the above implementation, in some modified implementations, before regenerating a preset number of augmented data based on sample augmentation features and preset fault data generation parameters, a calibration mechanism for the augmented sample generation model can be triggered to calibrate the noise scheduling parameters of the augmented sample generation model, thereby obtaining a calibrated augmented sample generation model. Then, using the calibrated augmented sample generation model, a preset number of augmented data are regenerated based on sample augmentation features and preset fault data generation parameters.

[0111] Furthermore, the noise scheduling parameters of the enhanced sample generation model are calibrated. Specifically, this can be done by: adjusting the noise scaling factor based on the target fault severity in the target fault data and a preset corrected noise coefficient β. That is, setting a noise scaling factor and dynamically adjusting it according to the target fault severity. For a minor fault, the noise scaling factor is set to 0.5, halving the noise coefficient to ensure weak features. For a moderate fault, the noise scaling factor is set to 0.75, slightly reducing the noise coefficient to balance features and diversity. For a severe fault, the noise scaling factor is set to 1.5, increasing the noise level and enhancing sample diversity.

[0112] This application embodiment obtains actual operating data of the target bearing over a preset time period. Based on the actual operating data, it retrieves target fault data from a preset fault information database. The target fault data includes fault types. Then, based on the fault types, it obtains the proportion of the target fault data in the preset fault information database. If the proportion is less than or equal to a preset proportion threshold, it performs dynamic masking enhancement processing on the actual operating data based on the target fault data to obtain sample enhancement features. Based on the sample enhancement features and preset fault data generation parameters, it generates a preset number of enhanced data entries and adds them to the preset fault information database. By enhancing the minority classes with a small proportion, it avoids the data imbalance caused by the excessive expansion of majority class samples. The system dynamically determines whether enhancement is needed based on the real-time proportion of each type of fault in the fault information database, enabling the system to adapt to changes in data distribution. Furthermore, this application can also selectively enhance and generate a smaller number of minority class samples, ensuring that the model is not biased towards majority class faults when using data from the fault information database for training, thereby improving the recognition rate of rare faults and greatly enhancing the efficiency of data utilization.

[0113] See Figure 2 This application also provides a method for training a fault detection model, which specifically includes the following steps:

[0114] Step 201: Obtain the target data.

[0115] In this embodiment, the target data is obtained from the preset fault information database in the data augmentation method described in the above embodiments.

[0116] Step 202: Divide the target data into training set, validation set and test set.

[0117] Step 203: Use the training set and validation set to train and validate the fault detection model.

[0118] Step 204: Use the test set to test the validated fault detection model, and obtain the trained fault detection model based on the test results.

[0119] The training method for the fault detection model provided in this application embodiment is based on the same inventive concept as the data augmentation method provided in the above embodiment, and has the same beneficial effects as the method used, operated or implemented.

[0120] See Figure 3 This application also provides a data enhancement device, which includes:

[0121] The first acquisition module 301 is used to acquire the actual operating data of the target bearing over a preset time period.

[0122] The second acquisition module 302 is used to acquire target fault data from a preset fault information database based on the actual operating data, wherein the target fault data includes fault types;

[0123] The third acquisition module 303 is used to acquire the proportion of the target fault data in the preset fault information database based on the fault type.

[0124] The dynamic mask enhancement processing module 304 is used to perform dynamic mask enhancement processing on the actual operating data based on the target fault data if the proportion is less than or equal to a preset proportion threshold, so as to obtain sample enhancement features.

[0125] The generation module 305 is used to generate a preset number of enhanced data based on the sample enhancement features and preset fault data generation parameters;

[0126] The data adding module 306 is used to add the enhanced data to the preset fault information database.

[0127] The data enhancement device provided in this application embodiment and the data enhancement method provided in the above embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.

[0128] See Figure 4 This application also provides a training device for a fault detection model, the device comprising:

[0129] The fourth acquisition module 401 is used to acquire target data, which is obtained from the preset fault information database in the data augmentation method described in the above embodiments.

[0130] The data partitioning module 402 is used to partition the target data into a training set, a validation set, and a test set.

[0131] The model training and validation module 403 is used to train and validate the fault detection model using the training set and the validation set;

[0132] The testing module 404 is used to test the verified fault detection model using the test set, and to obtain the trained fault detection model based on the test results.

[0133] The training device for the fault detection model provided in this application embodiment and the training method for the fault detection model provided in the above embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, run or implement.

[0134] This application also provides an electronic device corresponding to the training and data augmentation method for the fault detection model provided in the foregoing embodiments. Please refer to... Figure 5 This illustrates a schematic diagram of an electronic device provided by some embodiments of this application. For example... Figure 5 As shown, the electronic device 50 may include: a processor 500, a memory 501, a bus 502, and a communication interface 503. The processor 500, the communication interface 503, and the memory 501 are connected via the bus 502. The memory 501 stores a computer program that can run on the processor 500. When the processor 500 runs the computer program, it executes the training and data augmentation method of the fault detection model provided in any of the foregoing embodiments of this application.

[0135] The memory 501 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 503 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0136] Bus 502 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. Memory 501 is used to store programs. After receiving an execution instruction, processor 500 executes the program. The training and data augmentation method for the fault detection model disclosed in any of the foregoing embodiments of this application can be applied to processor 500, or implemented by processor 500.

[0137] The processor 500 may be an integrated circuit with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 500 or by instructions in software form. The processor 500 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 501. The processor 500 reads the information in memory 501 and, in conjunction with its hardware, completes the steps of the above method.

[0138] The electronic device provided in this application embodiment and the training and data augmentation method for the fault detection model provided in this application embodiment are based on the same inventive concept and have the same beneficial effects as the methods they adopt, operate or implement.

[0139] This application also provides a computer-readable storage medium corresponding to the training and data augmentation method for the fault detection model provided in the foregoing embodiments, wherein a computer program (i.e., a program product) is stored thereon, and the computer program, when run by a processor, executes the training and data augmentation method for the fault detection model provided in any of the foregoing embodiments.

[0140] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0141] This application also provides a computer program product corresponding to the training and data augmentation method for the fault detection model provided in the foregoing embodiments, including a computer program that is executed by a processor to implement the training and data augmentation method for the fault detection model provided in the above embodiments.

[0142] The computer-readable storage medium and computer program product provided in the above embodiments of this application are based on the same inventive concept as the training and data augmentation method of the fault detection model provided in the embodiments of this application, and have the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0143] It should be noted that:

[0144] The algorithms and displays provided herein are not inherently related to any particular computer, virtual device, or other equipment. Various general-purpose devices can also be used in conjunction with the teachings herein. The required structure for constructing such devices is apparent from the above description. Furthermore, this application is not directed to any particular programming language. It should be understood that the content of this application described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of this application.

[0145] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0146] Similarly, it should be understood that, in order to simplify this application and aid in understanding one or more of the various aspects of the invention, in the description of exemplary embodiments of this application above, various features of this application are sometimes grouped together into a single embodiment, figure, or description thereof.

[0147] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed herein and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed herein may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0148] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments.

[0149] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the virtual machine creation apparatus according to embodiments of this application. This application can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such an implementation of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0150] It should be noted that the above embodiments are illustrative of this application and not limiting of it, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0151] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data augmentation method, characterized in that, The method comprises the following steps: acquiring actual operation data of a target bearing in a preset time period; based on the actual operation data, acquiring target fault data from a preset fault information library, the target fault data including fault type, target fault degree and target fault evaluation parameter; based on the fault type, acquiring the proportion of the target fault data in the preset fault information library; if the proportion is less than or equal to a preset proportion threshold, performing dynamic mask enhancement processing on the actual operation data based on the target fault data to obtain sample enhancement features; based on the sample enhancement features and preset fault data generation parameters, generating a preset number of pieces of enhanced data; adding the enhanced data to the preset fault information library; the dynamic mask enhancement processing on the actual operation data based on the target fault data to obtain sample enhancement features comprises the following steps: based on the target fault degree, determining a dynamic mask proportion; based on the target fault evaluation parameter, determining an enhancement weight of the actual operation data, the actual operation data including actual vibration signal, actual temperature signal and actual sound signal; converting the one-dimensional actual vibration signal into a two-dimensional actual vibration time-frequency spectrum image, and performing normalization processing on the actual vibration time-frequency spectrum image, the actual temperature signal and the actual sound signal to obtain normalized actual vibration time-frequency spectrum image, normalized actual temperature signal and normalized actual sound signal, and aligning the normalized actual temperature signal and the normalized actual sound signal with the normalized actual vibration time-frequency spectrum image based on the timestamp of the normalized actual vibration time-frequency spectrum image as a reference; based on a first enhancement weight corresponding to the actual vibration signal, a second enhancement weight corresponding to the actual temperature signal and a third enhancement weight corresponding to the actual sound signal, performing fusion processing on the aligned normalized actual vibration time-frequency spectrum image, the normalized actual temperature signal and the normalized actual sound signal to obtain a fusion matrix; based on the fusion matrix and the dynamic mask proportion, performing dynamic mask enhancement processing on the actual operation data to obtain sample enhancement features.

2. The data augmentation method of claim 1, wherein, The actual operation data includes actual vibration signal and actual bearing speed, and the target fault data is acquired from the preset fault information library based on the actual operation data, which comprises the following steps: based on the actual bearing speed and the geometric size parameters of the target bearing, calculating the actual fault frequency of each target bearing component of the target bearing, the geometric size parameters including bearing pitch diameter, contact angle, number of rolling elements and rolling element diameter, and the each target bearing component including inner ring, outer ring and rolling element; based on the actual vibration signal, calculating the actual fault evaluation parameter of the target bearing, the actual fault evaluation parameter including actual peak-to-peak value, actual fractal dimension and actual impact characteristic corresponding to the actual vibration signal; based on the actual fault frequency of each target bearing component and the actual fault evaluation parameter, acquiring target fault data from the preset fault information library.

3. The data augmentation method of claim 2, wherein, The target fault data is obtained from a preset fault information library based on the actual failure frequency and the actual failure evaluation parameter of each target bearing assembly, and the target fault data includes: The similarity between the actual running data and each preset fault data is calculated based on the actual failure frequency and the actual failure evaluation parameter of each target bearing assembly and the preset failure frequency and the preset failure evaluation parameter of each preset bearing assembly of each preset fault data in the preset fault information library, and the preset failure evaluation parameter includes a preset peak-peak value, a preset fractal dimension, and a preset impact characteristic; The target fault data is determined based on the preset fault data corresponding to the minimum similarity and the actual failure frequency and the actual failure evaluation parameter.

4. The data augmentation method of any one of claims 1-3, wherein, Before the enhanced data is added to the preset fault information library, the method further includes: a frequency matching degree between a target failure frequency of the target fault data and an enhanced failure frequency of the enhanced data is obtained; and / or, an impact characteristic similarity between a target impact characteristic of the target fault data and an enhanced impact characteristic of the enhanced data is obtained; and / or, a fractal dimension consistency between a target fractal dimension of the target fault data and an enhanced fractal dimension of the enhanced data is obtained; whether the enhanced data is qualified is determined based on the frequency matching degree and / or the impact characteristic similarity and / or the fractal dimension consistency.

5. The data augmentation method of claim 4, wherein, The method further includes: if the enhanced data is qualified data, the enhanced data is added to the preset fault information library; if the enhanced data is unqualified data, a preset number of pieces of enhanced data are regenerated based on the sample enhanced feature and a preset fault data generation parameter. 6.A training method of a fault detection model, comprising: obtaining target data from a preset fault information library in the data enhancement method of any one of claims 1-5; dividing the target data into a training set, a validation set, and a test set; training and validating the fault detection model using the training set and the validation set; testing the validated fault detection model using the test set, and obtaining the trained fault detection model based on the test result.

7. A data enhancement apparatus, characterized by, The method includes: a first obtaining module configured to obtain actual running data of a target bearing in a preset time period; a second obtaining module configured to obtain target fault data from a preset fault information library based on the actual running data, the target fault data including a fault type; a third obtaining module configured to obtain a proportion of the target fault data in the preset fault information library based on the fault type; a dynamic mask enhancement processing module configured to perform dynamic mask enhancement processing on the actual running data based on the target fault data to obtain a sample enhanced feature if the proportion is less than or equal to a preset proportion threshold; a generating module configured to generate a preset number of pieces of enhanced data based on the sample enhanced feature and a preset fault data generation parameter; a data adding module configured to add the enhanced data to the preset fault information library. The dynamic mask enhancement processing module is configured to determine a dynamic mask ratio based on the target fault degree. The actual operation data includes actual vibration signals, actual temperature signals, and actual sound signals. The actual vibration signals are converted into two-dimensional actual vibration time-frequency spectrum images, and the actual vibration time-frequency spectrum images, the actual temperature signals, and the actual sound signals are normalized to obtain normalized actual vibration time-frequency spectrum images, normalized actual temperature signals, and normalized actual sound signals. The normalized actual temperature signals and the normalized actual sound signals are aligned with the normalized actual vibration time-frequency spectrum images based on timestamps of the normalized actual vibration time-frequency spectrum images. The aligned normalized actual vibration time-frequency spectrum images, the normalized actual temperature signals, and the normalized actual sound signals are fused based on first enhancement weights corresponding to the actual vibration signals, second enhancement weights corresponding to the actual temperature signals, and third enhancement weights corresponding to the actual sound signals to obtain a fusion matrix.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The actual operation data is dynamically masked and enhanced based on the fusion matrix and the dynamic mask ratio to obtain sample enhancement features.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The processor executes the computer program to implement the method of any one of claims 1-6. The computer program is executed by the processor to implement the method of any one of claims 1-6.

Images