A Fault Data Augmentation Method

Abstract

The present application discloses a method, apparatus, computer device and storage medium for fault data augmentation. The method includes: obtaining a first fault sample set and a second fault sample set; calculating the maximum Euclidean distance between each first sample in each fault class in the first fault sample set and the centroid of the fault class; calculating the Euclidean distances between any second sample in the second fault sample set and the centroids of some or all fault classes in the first fault sample set, to obtain a plurality of Euclidean distances corresponding to the second samples one-to-one with respect to the fault classes; and then taking the second samples whose Euclidean distances are less than or equal to the maximum Euclidean distance of the corresponding fault classes as augmented samples of the fault classes corresponding to the Euclidean distances of the second samples, to obtain a fault augmented data set of the first fault sample set. By using this method, the originally less fault data can be augmented into more abundant fault data, so as to provide a more complete data basis for fault identification.

Description

Technical Field

[0001] The present application belongs to the technical field of fault diagnosis and health management, and in particular relates to a fault data expansion method. Background Art

[0002] With the development of modern production and advancements in science and technology, industrial equipment is becoming increasingly complex, functional, and automated. Due to various unavoidable factors, equipment can sometimes experience various failures, reducing or even losing its intended functionality. Therefore, utilizing the various data generated within industrial systems to diagnose and identify faults within equipment and even the entire system is an effective means of maintaining the safety and reliability of mechanical equipment and the entire industrial system.

[0003] Traditional technologies typically collect operational data directly from equipment or systems and use this data for fault diagnosis. However, this directly collected data often exhibits a long-tail distribution, with more normal data and less abnormal data. This significantly increases the difficulty of modeling abnormal data, which can lead to reduced fault identification performance. Summary of the Invention

[0004] Therefore, in response to the above-mentioned defects of the prior art, the present application provides a fault data expansion method, apparatus, computer equipment, and readable storage medium, the purpose of which is to solve the problem of missing fault data samples during the fault identification process.

[0005] To achieve the above objectives, according to a first aspect of the present application, a method for expanding fault data is provided, the method comprising:

[0006] Obtaining a first fault sample set and a second fault sample set;

[0007] For any fault class in the first fault sample set, calculate the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class, and obtain multiple maximum Euclidean distances corresponding to each fault class;

[0008] For any second sample in the second fault sample set, calculating the Euclidean distance between the second sample and some or all of the centroids of the fault classes in the first fault sample set, and obtaining a plurality of Euclidean distances corresponding to the second samples in one-to-one correspondence with the fault classes;

[0009] The second sample whose Euclidean distance of the second sample is less than or equal to the maximum Euclidean distance is used as the augmented sample of the fault class corresponding to the Euclidean distance of the second sample, to obtain a fault augmented data set of the first fault sample set.

[0010] Preferably, the obtaining methods of the first fault sample set and the second fault sample include:

[0011] Obtain a first data set and a second data set;

[0012] Perform PCA dimensionality reduction on the first data set to obtain the first fault sample set, and perform PCA dimensionality reduction on the second data set to obtain the second fault sample set.

[0013] Preferably, the obtaining method of the centroids of each fault class in the first fault sample set includes:

[0014] Use the K-Means algorithm to cluster the first fault sample set to obtain the fault classes in the first fault sample set and their corresponding centroids.

[0015] Preferably, the above fault data augmentation method further includes:

[0016] For any one of the minority classes in the fault augmented data set, for any one of the third samples in the minority class, calculate the Euclidean distance between the third sample and all other samples in the minority class except the third sample;

[0017] According to the Euclidean distances between the third sample and all other samples, select several nearest neighbors of the third sample;

[0018] Sample from the nearest neighbors at a preset magnification to obtain several neighboring samples of the third sample;

[0019] Transform all the neighboring samples of the third sample to obtain the fault augmented sample set of the minority class.

[0020] Preferably, the step of transforming all the neighboring samples of the third sample to obtain the fault augmented sample set of the minority class specifically includes:

[0021] For each neighboring sample of the third sample, randomly select a natural number between 0 and 1, multiply it by the distance between the third sample and the neighboring sample, and then add the third sample to obtain an augmented sample group of the third sample;

[0022] All the augmented sample groups of the third sample constitute the fault augmented sample set of the minority class.

[0023] Preferably, the above fault data augmentation method further includes:

[0024] Use a generative adversarial network to perform data enhancement on the fault augmented sample set to obtain an augmented fault data set.

[0025] According to the second aspect of the present application, there is also provided a fault data augmentation device, which includes:

[0026] An acquisition module, configured to acquire a first fault sample set and a second fault sample set;

[0027] A first calculation module, configured to calculate, for any fault class in the first fault sample set, the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class, and obtain a plurality of maximum Euclidean distances corresponding to each fault class one by one;

[0028] A second calculation module, configured to calculate, for any second sample in the second fault sample set, the Euclidean distance between the second sample and the centroid of some or all of the fault classes in the first fault sample set, and obtain a plurality of Euclidean distances corresponding to the second sample corresponding to each fault class one by one;

[0029] An augmented sample acquisition module, using the second sample whose Euclidean distance is less than or equal to the maximum Euclidean distance as the augmented sample of the fault class corresponding to the Euclidean distance of the second sample, and obtaining a fault augmented data set of the first fault sample set.

[0030] Preferably, the acquisition module is further configured to acquire a first data set and a second data set, perform PCA dimensionality reduction on the first data set to obtain the first fault sample set, and perform PCA dimensionality reduction on the second data set to obtain the second fault sample set.

[0031] According to the third aspect of the present application, there is also provided a computer device, including at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit is caused to execute the steps of the method described in any one of the above.

[0032] According to the fourth aspect of the present application, there is also provided a computer-readable medium, which stores a computer program executable by a computer device, and when the computer program runs on the computer device, the computer device is caused to execute the steps of the method described in any one of the above.

[0033] Generally speaking, compared with the prior art through the above technical solutions of the inventive concept, the above fault data augmentation method, device, computer device and storage medium obtain a first fault sample set and a second fault sample set; for any fault class in the first fault sample set, calculate the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class, and obtain a plurality of maximum Euclidean distances corresponding one by one to each fault class; for any second sample in the second fault sample set, calculate the Euclidean distance between the second sample and the centroids of some or all fault classes in the first fault sample set, and obtain a plurality of Euclidean distances corresponding to the second sample corresponding to each fault class; and then use the second sample whose Euclidean distance is less than or equal to the maximum Euclidean distance as the augmented sample of the fault class corresponding to the Euclidean distance of the second sample, and obtain the fault augmented data set of the first fault sample set. In this way, by using the association between the first fault sample set and the second fault sample set to reconstruct the first fault sample set, and then obtaining the augmented data set of the first fault sample set, the originally less fault data can be augmented into more abundant fault data, so as to provide a more complete data basis for fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 FIG. 1 is a schematic structural diagram of an application scenario of the fault data augmentation method provided by an embodiment of the present application;

[0035] Figure 2 FIG. 2 is a schematic flowchart of the fault data augmentation method provided by an embodiment of the present application;

[0036] Figure 3 FIG. 3 is a structural block diagram of a processing operation guidance device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0038] For the sake of easy understanding, this application first introduces the application scenario applicable to the fault data augmentation method provided herein. Refer to Figure 1, the computer device 102 is electrically connected to the sensor group 104. The data storage system can store the data that the computer device 102 needs to process. The data storage system can be integrated on the computer device 102, or placed in the cloud or other network servers. After the sensor group 104 collects the first fault sample set and the second fault sample set, it transmits the collected first fault sample set and the second fault sample set to the computer device 102. The computer device 102 obtains the first fault sample set and the second fault sample set. For any fault class in the first fault sample set, the computer device 102 calculates the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class, and obtains a plurality of maximum Euclidean distances corresponding to each fault class one by one; for any second sample in the second fault sample set, the computer device 102 calculates the Euclidean distance between the second sample and the centroid of some or all fault classes in the first fault sample set, and obtains a plurality of Euclidean distances corresponding to the second sample corresponding to the fault class one by one; furthermore, the computer device 102 uses the second sample whose Euclidean distance is less than or equal to the maximum Euclidean distance as the augmented sample of the fault class corresponding to the Euclidean distance of the second sample, and obtains the fault augmented data set of the first fault sample set. Among them, the computer device 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers and portable wearable devices.

[0039] In one embodiment, as Figure 2 shown, a fault data augmentation method is provided. Taking the method applied to the Figure 1 computer device in as an example for illustration, it includes the following steps:

[0040] Step 201, obtain the first fault sample set and the second fault sample set.

[0041] Among them, the first fault sample set is the target threshold fault sample set, the second fault sample set is the source threshold fault sample set, the first fault sample set is composed of multiple first samples, and the second fault sample set is composed of multiple second samples. The first fault sample set and the second fault sample set are two data sets collected by the sensor group under different circumstances. It can be understood that the sensor collects the first fault sample set in the first case and the second fault sample set in the second case. The first and the second are only used to make a naming distinction for the fault sample sets collected by the sensor group under different circumstances, and the fault sample sets collected by the sensor group itself are not restricted.

[0042] Specifically, after the sensor group collects the first fault sample set and the second fault sample set respectively under different circumstances, it transmits the collected first fault sample set and the second fault sample set to the computer device, and the computer device obtains the first fault sample set and the second fault sample set.

[0043] Step 202: For any fault class in the first fault sample set, calculate the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class, and obtain a plurality of maximum Euclidean distances corresponding one-to-one to each fault class.

[0044] Specifically, after the computer device obtains the first fault sample set and the second fault sample set, since the first fault sample set contains multiple fault classes, and each fault class contains multiple first samples and the centroid of the fault class, for any fault class in the first fault sample set, first calculate the Euclidean distance between each first sample in the fault class and the centroid of the fault class, and then determine the maximum Euclidean distance corresponding to the centroid of the fault class from the Euclidean distances between each first sample in the fault class and the centroid of the fault class. In this way, for each fault class in the first fault sample set, a plurality of maximum Euclidean distances corresponding one-to-one to each fault class can be obtained.

[0045] Step 203: For any second sample in the second fault sample set, calculate the Euclidean distance between the second sample and the centroids of some or all fault classes in the first fault sample set, and obtain a plurality of Euclidean distances corresponding to each fault class and corresponding to the second sample.

[0046] Specifically, after the computer device determines a plurality of maximum Euclidean distances corresponding one-to-one to each fault class in the first fault sample set, for any second sample in the second fault sample set, calculate the Euclidean distance between the second sample and the centroids of some or all fault classes in the first fault sample set, and obtain a plurality of Euclidean distances corresponding to each fault class and corresponding to the second sample. In this way, for each second sample in the second fault sample set, a plurality of Euclidean distances corresponding to each fault class and corresponding to the second sample can be obtained.

[0047] Step 204: Use the second samples whose Euclidean distances are less than or equal to the maximum Euclidean distance as augmented samples of the fault class corresponding to the Euclidean distance of the second sample, and obtain the fault augmented data set of the first fault sample set.

[0048] Specifically, after the computer device obtains a plurality of maximum Euclidean distances corresponding one-to-one to each fault class in the first fault sample set and the Euclidean distances corresponding to each second sample in the second fault sample set and each fault class, compare the Euclidean distances corresponding to each second sample and each fault class with the maximum Euclidean distances corresponding to each fault class respectively. If the Euclidean distance corresponding to a second sample and a fault class is less than or equal to the maximum Euclidean distance corresponding to the fault class, then use the second sample as the augmented sample of the fault class. For example: For one of the fault classes F i , its centroid is C i , and the computer device obtains the maximum Euclidean distance d of the fault class through Step 202maxi For one of the second samples \(x\) in the second fault sample set j the computer device obtains the second sample \(x\) through step 203 j and \(C\) i the Euclidean distance therebetween, and then the second sample \(x\) j and \(C\) i the Euclidean distance therebetween and the maximum Euclidean distance \(d\) of \(F\) i are compared. If the Euclidean distance between \(x\) maxi and \(C\) j is less than or equal to \(d\) i then \(x\) maxi is used as the augmented sample of the fault class \(F\) j In this way, by comparing and selecting the Euclidean distances between each second sample in the second fault sample set and each fault class one by one with the maximum Euclidean distances corresponding to each fault class, the augmented sample sets of each fault class in the first fault sample set can be obtained respectively, and the augmented sample sets of each fault class constitute the fault augmented data set of the first fault sample set. i In this embodiment, by obtaining the first fault sample set and the second fault sample set; for any fault class in the first fault sample set, calculating the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class to obtain a plurality of maximum Euclidean distances corresponding to each fault class one by one; for any second sample in the second fault sample set, calculating the Euclidean distance between the second sample and the centroid of some or all fault classes in the first fault sample set to obtain a plurality of Euclidean distances corresponding to the second sample corresponding to each fault class one by one; and then using the second sample whose Euclidean distance is less than or equal to the maximum Euclidean distance as the augmented sample of the fault class corresponding to the Euclidean distance of the second sample, the fault augmented data set of the first fault sample set is obtained. In this way, by using the association between the first fault sample set and the second fault sample set to reconstruct the first fault sample set, and then obtaining the augmented data set of the first fault sample set, the originally less fault data can be expanded into richer fault data, so as to provide a more complete data basis for fault recognition.

[0049] In one of the embodiments, the obtaining methods of the first fault sample set and the second fault sample include: obtaining a first data set and a second data set; performing PCA dimensionality reduction on the first data set to obtain the first fault sample set, and performing PCA dimensionality reduction on the second data set to obtain the second fault sample set. <l

[0050] Among them, the first data set is the target threshold data set, and the second data set is the source threshold data set.

[0051]

[0052] ​Specifically, the first fault sample set and the second fault sample set can be obtained by performing PCA dimensionality reduction on the data sets collected by the sensor group. After the sensor group collects the first data set and the second data set under different conditions, the first data set and the second data set are transmitted to the computer device through the network, and the computer device performs PCA dimensionality reduction on the first data set and the second data set respectively, thereby obtaining the first fault sample set and the second fault sample set.

[0053] In this embodiment, after performing PCA dimensionality reduction on the data set collected by the sensor, fault data augmentation is performed, and then an augmented sample set with low dimensions is obtained. In this way, the augmented sample set with low dimensions can be used to prevent overfitting from occurring during subsequent fault identification.

[0054] In one of the embodiments, after the computer device obtains the first fault sample set, a similarity measurement method including but not limited to the K-Means algorithm is used to cluster the first fault sample set, thereby obtaining the fault classes in the first fault sample set and their corresponding centroids.

[0055] In one of the embodiments, the above data augmentation method further includes: for any minority class in the fault augmented data set, for any third sample in the minority class, calculating the Euclidean distance between the third sample and all other samples in the minority class except the third sample; according to the Euclidean distance between the third sample and all other samples, selecting several nearest neighbors of the third sample; sampling from the nearest neighbors at a preset magnification to obtain several neighboring samples of the third sample; transforming all the neighboring samples of the third sample to obtain a fault augmented sample set of the minority class.

[0056] Among them, the minority class is a fault class with fewer augmented samples in the augmented sample set, and the third sample is an augmented sample in the minority class.

[0057] Specifically, after the computer device obtains the fault augmented data set of the first fault sample set, since the fault augmented data set of the first fault sample set is composed of the augmented sample sets of each fault class, there are fewer augmented samples in the augmented sample sets of some fault classes. For these fault classes with fewer augmented samples, data augmentation can be continued. First, for any third sample in any one of the minority classes, calculate the Euclidean distance between the third sample and all other samples in the minority class except the third sample. Then, according to the Euclidean distances between the third sample and other samples, select several samples with the smallest distances from other samples in ascending order as the nearest neighbors of the third sample. Then, sample from the nearest neighbors at a preset sampling magnification to obtain several neighboring samples of the third sample. Furthermore, transform each sampled neighboring sample of the third sample to obtain an augmented sample group of the third sample. The augmented sample groups of all third samples in the minority class form the fault augmented sample set of the minority class.

[0058] In one embodiment, transforming all neighboring samples of the third samples to obtain the fault augmented sample set of the minority class specifically includes: for each neighboring sample of the third sample, randomly select a natural number from 0 to 1, multiply it by the distance between the third sample and the neighboring sample, and then add the third sample to obtain the augmented sample group of the third sample; the augmented sample groups of all third samples constitute the fault augmented sample set of the minority class.

[0059] Specifically, after the computer device samples from the nearest neighbors of the third sample at a preset sampling magnification to obtain several neighboring samples of the third sample, for any sampled neighboring sample of the third sample, randomly select a natural number from 0 to 1, multiply it by the distance between the neighboring sample and the corresponding third sample, and then add the corresponding third sample itself to obtain the augmented sample of the corresponding third sample. In this way, by transforming all the sampled neighboring samples of the third sample, the augmented sample group of the third sample can be obtained, and thus the augmented sample groups of all third samples constitute the fault augmented sample set of the minority class.

[0060] In the above embodiment, by transforming the samples in the minority classes in the fault augmented data set of the first fault sample set, these minority classes can be further augmented with data, making the originally less fault data more abundant.

[0061] In another embodiment, after the computer device obtains the fault augmented sample set of the minority class, it can also use a generative adversarial network to perform data enhancement on the fault augmented sample set to obtain an augmented fault data set. In this way, the amount of data in the fault augmented sample set can be further increased, thereby improving the generalization ability of the subsequent fault recognition model.

[0062] Next, a specific embodiment is used to illustrate the fault data augmentation method of the present application in detail:

[0063] Step 1, obtain the source domain data set X = {x s} containing m samplings of n sensors (1 ≤ i ≤ m i ), and each x s is an n-dimensional original feature vector composed of the sampling data of n sensors. Let X i = {x fs} ( i x is an n-dimensional vector) be the fault sample set in the source threshold data set X, and there are N i classes of faults in X fs . s

[0064] Step 2, perform PCA dimensionality reduction on the source domain data set. Specifically, calculate the normalized covariance matrix of X T :

[0065]

[0066] where R is an n×n matrix;

[0067] Calculate the eigenvalues and eigenvectors of R, and arrange the eigenvalues in descending order:

[0068] λ1 ≥ λ2 ≥ λ3 … ≥ λ n ............................................(2)

[0069] After rearranging the eigenvectors according to the eigenvalues, we get:

[0070] P nn = [p1, p2, …, p n .........................................(3)

[0071] Select the first k eigenvectors corresponding to the eigenvalues whose cumulative sum is greater than 85% for PCA dimensionality reduction, that is, k satisfies:

[0072]

[0073] Calculate the eigenvector matrix P nk corresponding to the k eigenvalues. After dimensionality reduction of the source domain data set:

[0074]

[0075] where, contains k eigenvalues, This is the dataset after dimensionality reduction.

[0076] Step 3: Construct a target domain dataset Y = {y d}(t represents time, 1 ≤ j ≤ m j ), where each sample is an n-dimensional original feature vector composed of sampling data from n sensors. Let Y d = {y fd}( j}( y j is an n-dimensional vector) be the set of faulty samples in dataset Y, and Y fd contains N d types of faults.

[0077] Step 4: According to the method in Step 2, perform PCA dimensionality reduction on the target domain dataset. Here, directly select the eigenvector matrix corresponding to the first k eigenvalues of its feature matrix as the basis to obtain the dimensionality-reduced target domain dataset

[0078] Step 5: For each fault class F in i (1 ≤ i ≤ n d ), use a similarity measurement method including but not limited to the K-means clustering algorithm to determine the centroid of this fault class. For example, use the method of mean calculation to calculate the mean of F i on k dimensions as the centroid C i of this fault class, and find the maximum distance d maxi from the centroid in the Euclidean distance within this class.

[0079] Step 6: For each faulty sample x in j , if its Euclidean distance from the above-mentioned fault class centroid C i is less than or equal to dmax i , then take x j as the augmented sample of F i . In this way, the fault augmented dataset of the dimensionality-reduced target domain dataset is constructed.

[0080] Step 7: For the minority classes in the fault augmented dataset, use the SMOTE algorithm for sample augmentation. Specifically:

[0081] a) For the sample y j in the minority class, calculate its distances to all other samples in the minority class sample set using the Euclidean distance as the standard to obtain the k-nearest neighbors.

[0082] b) Set a sampling ratio according to the sample imbalance ratio to determine the sampling magnification N. For each minority-class sample a, randomly select several neighboring samples from its k nearest neighbors.

[0083] c) For each randomly selected neighboring sample b, construct a new sample c with the original sample a according to the following formula:

[0084] c = a + rand(0,1) × |a - b|...............................(6)

[0085] The final target-domain fault augmented dataset can be obtained according to the above method. The method of transfer learning can be used for fault recognition of the obtained fault augmented dataset.

[0086] This embodiment uses the information in the existing source-domain fault samples to make up for the small-sample fault data in the target domain. Through the existing clustering information in the target domain, similar samples in the source domain are quickly found for expansion, reducing the interference of irrelevant information. At the same time, new minority classes are synthesized for the small-sample classes for expansion, as much as possible ensuring the balance of the target-domain samples. It provides good feature classification and data for the subsequent transfer learning fault recognition, can greatly reduce the complexity of the classification model, and improve the prediction accuracy and robustness.

[0087] The embodiments of the present application can be widely applied to the health management system of large power plants to help achieve rapid fault recognition of the devices and ensure stable operation.

[0088] Based on the same inventive concept, as Figure 3 shown, the embodiments of the present application also provide a fault data augmentation device 300, including: an acquisition module 301, a first calculation module 302, a second calculation module 303, and an augmented sample acquisition module 304, where:

[0089] The acquisition module is used to acquire a first fault sample set and a second fault sample set.

[0090] The first calculation module is used to calculate the maximum Euclidean distance between each first sample in a fault class and the centroid of the fault class in the first fault sample set for any fault class in the first fault sample set, and obtain a plurality of maximum Euclidean distances corresponding to each fault class one by one.

[0091] The second calculation module is used to calculate the Euclidean distance between a second sample in the second fault sample set and the centroids of some or all fault classes in the first fault sample set for any second sample in the second fault sample set, and obtain a plurality of corresponding second sample Euclidean distances corresponding to the fault classes one by one.

[0092] The augmented sample acquisition module takes the second samples whose Euclidean distance is less than or equal to the maximum Euclidean distance as the augmented samples of the fault class corresponding to the Euclidean distance of the second samples, and obtains the fault augmented data set of the first fault sample set.

[0093] In one embodiment, the acquisition module is further configured to acquire a first data set and a second data set, perform PCA dimensionality reduction on the first data set to obtain a first fault sample set, and perform PCA dimensionality reduction on the second data set to obtain the second fault sample set.

[0094] In one embodiment, the above-mentioned fault data augmentation device further includes a clustering module, which is used to cluster the first fault sample set by using the K-Means algorithm to obtain the fault classes in the first fault sample set and their corresponding centroids.

[0095] In one embodiment, the above-mentioned fault data augmentation device further includes an augmented sample acquisition module, which is used to calculate the Euclidean distance between any third sample in a minority class in the fault augmented data set and all other samples in the minority class except the third sample; and select several nearest samples of the third sample according to the Euclidean distance between the third sample and all other samples; then sample from the nearest samples at a preset magnification to obtain several neighboring samples of the third sample; and transform all the neighboring samples of the third sample to obtain the fault augmented sample set of the minority class.

[0096] In one embodiment, the above-mentioned augmented sample acquisition module is further configured to, for each neighboring sample of the third sample, randomly select a natural number from 0 to 1, multiply it by the distance between the third sample and the neighboring sample, and then add the third sample to obtain the augmented sample group of the third sample; all the augmented sample groups of the third sample constitute the fault augmented sample set of the minority class.

[0097] In one embodiment, the above-mentioned fault data augmentation device further includes a data enhancement module, which is used to perform data enhancement on the fault augmented sample set by using a generative adversarial network to obtain the augmented fault data set.

[0098] For the specific limitations of the fault data augmentation device, reference can be made to the limitations of the fault data augmentation method in the above text, which will not be elaborated here. Each module in the above-mentioned fault data augmentation device can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in the processor of the mixed reality device in hardware form or independent of it, or stored in the memory of the mixed reality device in software form, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0099] The content described in this specification is only an illustration of the present invention. Those skilled in the art to which the present invention pertains can make various modifications or supplements to the specific embodiments described, or use similar methods for substitution, as long as they do not deviate from the content of this specification of the present invention or exceed the scope defined by this claims, and shall fall within the protection scope of the present invention.

Claims

1. A method for fault data augmentation, characterized in that, Including: Obtain a first fault sample set and a second fault sample set; For any fault class in the first fault sample set, calculate the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class, and obtain a plurality of maximum Euclidean distances corresponding one-to-one to each fault class; For any second sample in the second fault sample set, calculate the Euclidean distance between the second sample and the centroids of some or all of the fault classes in the first fault sample set, and obtain a plurality of Euclidean distances corresponding to the second sample corresponding one-to-one to the fault classes; Use the second samples whose Euclidean distances are less than or equal to the maximum Euclidean distance as augmented samples of the fault class corresponding to the Euclidean distance of the second sample, and obtain a fault augmented data set of the first fault sample set.

2. The fault data expansion method according to claim 1, wherein The obtaining methods of the first fault sample set and the second fault sample set include: Obtain a first data set and a second data set; Perform PCA dimensionality reduction on the first data set to obtain the first fault sample set, and perform PCA dimensionality reduction on the second data set to obtain the second fault sample set.

3. The fault data expansion method according to claim 1, wherein The obtaining method of the centroids of each fault class in the first fault sample set includes: Use the K-Means algorithm to cluster the first fault sample set to obtain the fault classes in the first fault sample set and their corresponding centroids.

4. The fault data expansion method according to claim 1, characterized in that, Also included: For any minority class in the fault augmented data set, for any third sample in the minority class, calculate the Euclidean distance between the third sample and all other samples in the minority class except the third sample; Select several nearest neighbors of the third sample according to the Euclidean distance between the third sample and all other samples; Sample from the nearest neighbors at a preset magnification to obtain several neighboring samples of the third sample; Transform all the neighboring samples of the third sample to obtain a fault expansion sample set of the minority class.

5. The fault data expansion method according to claim 4, characterized in that The step of transforming all the neighboring samples of the third sample to obtain a fault expansion sample set of the minority class specifically includes: For each neighboring sample of the third sample, randomly select a natural number from 0 to 1, multiply it by the distance between the third sample and the neighboring sample, and then add the third sample to obtain an augmented sample group of the third sample; All the augmented sample groups of the third sample constitute the fault expansion sample set of the minority class.

6. The fault data expansion method according to claim 4 or 5, characterized in that Also included: Use a generative adversarial network to perform data augmentation on the fault expansion sample set to obtain an augmented fault data set.

7. A fault data expansion device, characterized in that, Including: An acquisition module, configured to obtain a first fault sample set and a second fault sample set; A first calculation module, configured to calculate, for any fault class in the first fault sample set, the maximum Euclidean distance between each first sample in the fault class and the centroid of the fault class, and obtain a plurality of maximum Euclidean distances corresponding one-to-one to each fault class; A second calculation module, configured to calculate, for any second sample in the second fault sample set, the Euclidean distance between the second sample and the centroids of some or all of the fault classes in the first fault sample set, and obtain a plurality of Euclidean distances corresponding to the second sample corresponding one-to-one to the fault classes; The augmented sample acquisition module takes the second samples whose Euclidean distance is less than or equal to the maximum Euclidean distance as the augmented samples of the fault class corresponding to the Euclidean distance of the second samples, and obtains the fault augmented data set of the first fault sample set.

8. The fault data expansion device according to claim 7, wherein, The acquisition module is further configured to acquire a first data set and a second data set, perform PCA dimensionality reduction on the first data set to obtain the first fault sample set, and perform PCA dimensionality reduction on the second data set to obtain the second fault sample set.

9. A computer device, characterized in that, It includes at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit is caused to execute the steps of the method according to any one of claims 1 to 6.

10. A computer-readable medium, characterized in that, It stores a computer program executable by a computer device, and when the computer program runs on the computer device, the computer device is caused to execute the steps of the method according to any one of claims 1 to 6.

Images