Fault identification method, fault identification system, electronic device, and storage medium

Abstract

The embodiment of the application provides a fault identification method, a fault identification system, an electronic device and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: generating a predicted fault image according to a device fault label and a fault image; obtaining predicted discrimination data by discriminating the device fault image and the predicted fault image through a fault discrimination submodel, determining a fault data enhancement model according to the device fault image, the predicted fault image, the device fault label and the predicted discrimination data; obtaining a fault enhanced image by performing data enhancement according to the fault data enhancement model, and obtaining a predicted fault classification label by performing fault identification on the enhanced image according to an initial classification model; determining a fault identification model according to the predicted fault classification label and the device fault label; and obtaining a target fault label by performing fault identification on a target fault image according to the fault identification model. The embodiment of the application can improve the accuracy of fault identification under a small amount of fault samples.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a fault identification method, a fault identification system, an electronic device and a storage medium. BACKGROUND

[0002] Currently, since device fault information cannot be obtained in time when device management is performed, fault identification processing is usually performed in a planned maintenance or after-the-fact maintenance manner, and such a processing manner usually leads to over-maintenance, resource waste and the like. In order to solve these problems, the method for fault identification in the related art usually uses a model to automatically diagnose the device fault type. However, since the number of times of device failure in actual production is limited, only a small amount of samples with fault labels are available when the model is trained, and a large amount of manually labeled fault data is required for training of the model, thereby affecting the identification accuracy of fault information. Therefore, how to provide a method capable of improving the fault identification accuracy with a small amount of fault samples has become a technical problem to be solved. SUMMARY

[0003] The main purpose of the embodiments of the present application is to provide a fault identification method, a fault identification system, an electronic device and a storage medium, which can improve the accuracy of fault identification with a small amount of fault samples.

[0004] To achieve the above-mentioned purpose, a first aspect of the embodiments of the present application provides a fault identification method, which comprises:

[0005] obtaining a device fault image and a device fault label of the device fault image;

[0006] inputting the device fault image and the device fault label into a preset initial network model, wherein the initial network model comprises a fault image generation sub-model and a fault discrimination sub-model;

[0007] inputting the device fault label into the fault image generation sub-model for image generation processing to obtain a predicted fault image, wherein the fault label of the predicted fault image is the same as the device fault label;

[0008] taking the device fault image and the predicted fault image as discrimination images, and inputting the discrimination images into the fault discrimination sub-model for image discrimination processing to obtain predicted discrimination data;

[0009] adjusting parameters of the initial network model according to the device fault image, the predicted fault image, the device fault label and the predicted discrimination data to obtain a fault data enhancement model;

[0010] input the device fault image and the device fault label into the fault data augmentation model for data augmentation processing to obtain a fault augmented image, wherein the fault label of the fault augmented image is the same as the device fault label;

[0011] input the device fault image and the fault augmented image as a fault classification image into a preset initial classification model for fault identification processing to obtain a predicted fault classification label;

[0012] adjust parameters of the initial classification model according to the predicted fault classification label and the device fault label to obtain a fault identification model;

[0013] obtain a target fault image of a device to be tested, and input the target fault image into the fault identification model for fault identification processing to obtain a target fault label.

[0014] In some embodiments, the obtaining of the device fault image and the device fault label of the device fault image comprises:

[0015] obtain device operation data of a sample device;

[0016] perform data segmentation processing on the device operation data according to a preset data length to obtain device operation sub-data;

[0017] perform short-time Fourier transform on the device operation sub-data to obtain device transformed data;

[0018] perform normalization processing on the device transformed data to obtain device normalized data;

[0019] perform image conversion processing on the device normalized data to obtain the device fault image;

[0020] perform fault type marking on the device fault image according to the device operation data to obtain a device fault label of the device fault image.

[0021] In some embodiments, the inputting of the device fault label into the fault image generation sub-model for image generation processing to obtain a predicted fault image comprises:

[0022] perform fault operation data generation according to a preset data length and a preset distribution function to obtain initial fault operation data;

[0023] perform data encoding processing on the initial fault operation data to obtain an initial fault operation vector;

[0024] perform label encoding processing on the device fault label to obtain a fault label vector;

[0025] The initial fault operation vector and the fault label vector are input to the fault image generation sub-model for image generation processing to obtain the predicted fault image.

[0026] In some embodiments, the fault image generation sub-model includes a deconvolution layer, a normalization layer, and a generation activation layer, and the inputting of the initial fault operation vector and the fault label vector to the fault image generation sub-model for image generation processing to obtain the predicted fault image includes:

[0027] The initial fault operation vector and the fault label vector are input to the fault image generation sub-model for image generation processing to obtain the predicted fault image.

[0028] According to the deconvolution layer, the predicted fault input vector is subjected to fault vector deconvolution processing to obtain a deconvolution fault generation vector.

[0029] According to the normalization layer, the deconvolution fault generation vector is subjected to normalization processing to obtain a normalized fault generation vector.

[0030] According to the generation activation layer, the normalized fault generation vector is subjected to de-linearization processing to obtain a target fault generation vector.

[0031] According to the target fault generation vector, image conversion processing is performed to obtain the predicted fault image.

[0032] In some embodiments, the fault discrimination sub-model includes a convolution layer, a standardization layer, and a discrimination activation layer.

[0033] The inputting of the discrimination image to the fault discrimination sub-model for image discrimination processing to obtain predicted discrimination data includes:

[0034] The discrimination image is subjected to feature extraction to obtain a fault discrimination vector.

[0035] According to the convolution layer, the fault discrimination vector is subjected to fault vector convolution processing to obtain a fault convolution discrimination vector.

[0036] According to the standardization layer, the fault convolution discrimination vector is subjected to standardization processing to obtain a standard fault discrimination vector.

[0037] According to the discrimination activation layer, the standard fault discrimination vector is subjected to de-linearization processing to obtain the predicted discrimination data.

[0038] In some embodiments, the predicted discrimination data includes first discrimination data and second discrimination data.

[0039] The parameter adjustment is performed on the initial network model according to the device fault image, the predicted fault image, the device fault label and the predicted discrimination data, to obtain a fault data enhancement model, and the parameter adjustment comprises:

[0040] An initial device fault loss calculation is performed according to the device fault image, the first discrimination data and the device fault label, to obtain initial device fault loss data;

[0041] A generated device fault loss calculation is performed according to the predicted fault image, the second discrimination data and the device fault label, to obtain generated device fault loss data;

[0042] The parameter adjustment is performed on the initial network model according to the initial device fault loss data and the generated device fault loss data, to obtain the fault data enhancement model.

[0043] In some embodiments, the fault classification image is input into a preset initial classification model for fault identification processing, to obtain a predicted fault classification label, and the fault identification processing comprises:

[0044] An image superposition processing is performed on the fault classification image, to obtain a fault superposition image;

[0045] The fault superposition image is input into the initial classification model for fault identification processing, to obtain fault classification data;

[0046] The predicted fault classification label of the fault classification image is determined according to the fault classification data.

[0047] To achieve the above object, a second aspect of the embodiments of the present application proposes a fault identification system, which comprises:

[0048] A sample acquisition module is configured to acquire a device fault image and a device fault label of the device fault image;

[0049] A model input module is configured to input the device fault image and the device fault label into a preset initial network model, and the initial network model comprises a fault image generation sub-model and a fault discrimination sub-model;

[0050] An image generation module is configured to input the device fault label into the fault image generation sub-model for image generation processing, to obtain a predicted fault image, and the fault label of the predicted fault image is the same as the device fault label;

[0051] A fault discrimination module is configured to take the device fault image and the predicted fault image as discrimination images, and input the discrimination images into the fault discrimination sub-model for image discrimination processing, to obtain predicted discrimination data;

[0052] The first parameter adjustment module is configured to perform parameter adjustment on the initial network model according to the device fault image, the predicted fault image, the device fault label and the prediction discrimination data, to obtain a fault data enhancement model.

[0053] The data enhancement module is configured to input the device fault image and the device fault label into the fault data enhancement model for data enhancement processing, to obtain a fault enhancement image, wherein the fault label of the fault enhancement image is the same as the device fault label.

[0054] The image classification module is configured to take the device fault image and the fault enhancement image as fault classification images, and input the fault classification images into a preset initial classification model for fault identification processing, to obtain a predicted fault classification label.

[0055] The second parameter adjustment module is configured to perform parameter adjustment on the initial classification model according to the predicted fault classification label and the device fault label, to obtain a fault identification model.

[0056] The fault identification module is configured to obtain a target fault image of a device to be tested, and input the target fault image into the fault identification model for fault identification processing, to obtain a target fault label.

[0057] To achieve the above object, a third aspect of embodiments of the present application provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of the first aspect of embodiments of the present application when executing the computer program.

[0058] To achieve the above object, a fourth aspect of embodiments of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method according to any one of the first aspect of embodiments of the present application.

[0059] The fault identification method, the fault identification system, the electronic device and the storage medium provided in the embodiments of the present application first acquire a device fault image and a device fault label of the device fault image, input the device fault image and the device fault label into a preset initial network model, and the initial network model includes a fault image generation submodel and a fault discrimination submodel. In order to make the generated device enhanced image more similar to the real device fault image, the device fault label is input into the fault image generation submodel for image generation processing, and a predicted fault image is obtained, wherein the fault label of the predicted fault image is the same as the device fault label. Then, the device fault image and the predicted fault image are taken as discrimination images, and the discrimination images are input into the fault discrimination submodel for image discrimination processing, and predicted discrimination data is obtained. The initial network model is adjusted in parameters according to the device fault image, the predicted fault image, the device fault label and the predicted discrimination data, and a fault data enhancement model is obtained. Then, the device fault image and the device fault label are input into the fault data enhancement model for data enhancement processing, and a fault enhanced image is obtained, wherein the fault label of the fault enhanced image is the same as the device fault label. The device fault image and the fault enhanced image are taken as fault classification images, and the fault classification images are input into a preset initial classification model for fault identification processing, and a predicted fault classification label is obtained. The initial classification model is adjusted in parameters according to the predicted fault classification label and the device fault label, and a fault identification model is obtained. A target fault image of a to-be-tested device is acquired, and the target fault image is input into the fault identification model for fault identification processing, and a target fault label is obtained. The fault data enhancement model including the fault image generation submodel and the fault discrimination submodel is constructed, the input device fault image can be data enhanced to expand the number of labeled fault data, and the fault identification model with higher accuracy is obtained according to the fault enhanced image after data enhancement and the device fault image. Therefore, when the target fault image of the to-be-tested device is processed for fault identification according to the fault identification model provided in the embodiments of the present application, the accuracy of fault identification can be improved under a small amount of fault samples. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is a flowchart of the fault identification method provided in the embodiments of the present application;

[0061] Figure 2 is Figure 1 is a flowchart of the specific method of step S110 in the method;

[0062] Figure 3 is Figure 1 is a flowchart of the specific method of step S130 in the method;

[0063] Figure 4 is Figure 3A flow chart of the specific method in step S340;

[0064] Figure 5 is Figure 1 A flow chart of the specific method in step S140;

[0065] Figure 6 is Figure 1 A flow chart of the specific method in step S150;

[0066] Figure 7 is Figure 1 A flow chart of the specific method in step S170;

[0067] Figure 8 is a module structure block diagram of the fault identification system provided by the embodiments of the present application;

[0068] Figure 9 is a hardware structure schematic diagram of the electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0069] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0070] It should be noted that although the functional modules are divided in the device schematic diagram, and the logical order is shown in the flow chart, in some cases, the steps shown or described can be executed in a manner different from the module division in the device or the order in the flow chart. The terms "first", "second", etc. in the specification and claims and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0071] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0072] First, the several terms involved in the present application are analyzed:

[0073] Artificial Intelligence (AI): is a new technical science that studies, develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence; Artificial intelligence is a branch of computer science, artificial intelligence attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence, including robots, language recognition, image recognition, natural language processing and expert systems. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence is also the theory, method, technology and application system of using digital computer or digital computer controlled machine to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results.

[0074] Visual Geometry Group Network (VGG): is an image classification model, also known as VGG-16, which includes 16 hidden layers, including 13 convolutional layers and 3 fully connected layers. This model can show that increasing the depth of the network can affect the final performance of the network to a certain extent.

[0075] Deconvolution: also known as trans rank convolution, the processing process of deconvolution is opposite to that of convolution. Deconvolution can perform dimensionality operation, that is, deconvolution outputs the target image according to the characteristics of the input image, thereby playing the role of image restoration.

[0076] Batch Normalization (BN): is to make the result of convolution meet the normal distribution again, and then input the linear rectifier function, which will not cause gradient disappearance.

[0077] Rectified Linear Unit (ReLU): ReLU function is a piecewise linear function, if the input is positive, the function will directly output, otherwise, the function will output zero. ReLU can make the model easier to train and can obtain better performance.

[0078] Fault diagnosis of equipment is a core part of equipment management. Since the equipment management cannot obtain the equipment fault information in time, the prior art usually adopts planned maintenance or after-service maintenance to identify and process the fault, which usually leads to over maintenance and resource waste. At present, the common method for fault diagnosis is to perform spectrum analysis on the collected signal data according to a pure physical model of equipment mechanism, and to judge the equipment fault according to the frequency or amplitude size combined with a threshold. However, the traditional mechanism model needs to accurately understand the detailed mechanism of the equipment, and to a great extent, depends on the experience of experts, which is difficult. The spectrum analysis method can roughly judge the general common fault type, but the recognition accuracy is low for complex signal data. In order to solve these problems, the related technology also includes using deep learning, machine learning and other models to automatically diagnose and identify the equipment fault type. However, due to the limited number of equipment failures in actual production, only a small amount of samples with fault labels are available for model training, and a large amount of manually labeled fault data is required for model training, thereby affecting the recognition accuracy of the fault information. Therefore, how to provide a method capable of improving the fault recognition accuracy under a small amount of fault samples has become a technical problem to be solved.

[0079] Therefore, the fault recognition method, the fault recognition system, the electronic device and the storage medium provided by the embodiments of the present application can improve the accuracy of fault recognition under a small amount of fault samples.

[0080] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Artificial intelligence (AI) is the use of digital computers or computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0081] The basic technology of artificial intelligence generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. The software technology of artificial intelligence mainly includes computer vision technology, robot technology, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0082] The fault identification method provided in the embodiments of the present application relates to the technical field of artificial intelligence. The fault identification method provided in the embodiments of the present application can be applied to a terminal, can be applied to a server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart watch or the like; the server can be a stand-alone server or a cloud server providing cloud services, a cloud database, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, a content delivery network (CDN), and basic cloud computing services such as a big data and artificial intelligence platform; and the software can be an application program implementing the fault identification method, but is not limited to the above forms.

[0083] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0084] Please refer to Figure 1 , Figure 1 is an optional flowchart of the fault identification method provided in the embodiments of the present application. In some embodiments of the present application, the fault identification method of the embodiments of the present application includes but is not limited to steps S110 to S190, which are described below in combination with Figure 1 The nine steps are described in detail.

[0085] Step S110, obtaining a device fault image and a device fault label of the device fault image;

[0086] Step S120, inputting the device fault image and the device fault label into a preset initial network model, the initial network model including a fault image generation sub-model and a fault discrimination sub-model;

[0087] Step S130, input the device fault label to the fault image generation sub-model for image generation processing to obtain a predicted fault image, and the fault label of the predicted fault image is the same as the device fault label.

[0088] Step S140, take the device fault image and the predicted fault image as a discrimination image, and input the discrimination image to the fault discrimination sub-model for image discrimination processing to obtain predicted discrimination data.

[0089] Step S150, adjust parameters of the initial network model according to the device fault image, the predicted fault image, the device fault label and the predicted discrimination data to obtain a fault data enhancement model.

[0090] Step S160, input the device fault image and the device fault label to the fault data enhancement model for data enhancement processing to obtain a fault enhanced image, and the fault label of the fault enhanced image is the same as the device fault label.

[0091] Step S170, take the device fault image and the fault enhanced image as a fault classification image, and input the fault classification image to a preset initial classification model for fault recognition processing to obtain a predicted fault classification label.

[0092] Step S180, adjust parameters of the initial classification model according to the predicted fault classification label and the device fault label to obtain a fault recognition model.

[0093] Step S190, obtain a target fault image of a to-be-tested device, and input the target fault image to the fault recognition model for fault recognition processing to obtain a target fault label.

[0094] It should be noted that in an actual application environment, the fault recognition method provided by the embodiments of the present application can be executed by a terminal or a server respectively, or cooperatively executed by the terminal and the server. Wherein, the terminal communicates with the server through a network, for example, taking the fault recognition method executed on the corresponding terminal as an example, specifically: the terminal or the server obtains the device fault image and the device fault label of the device fault image from the local, obtains the fault data enhancement model based on the device fault image and the device fault label, and performs data enhancement based on the fault data enhancement model to obtain a plurality of fault enhanced images, so as to expand the data quantity of the device fault image. Then, the fault recognition model is trained according to the expanded fault image, and the fault recognition model can accurately diagnose and identify the target fault label of the target fault image. The fault recognition method can also be deployed on the server, so that the server can also implement the steps of the above fault recognition method.

[0095] In step S110 of some embodiments, due to the limited number of equipment failures in actual production, only a small number of samples with failure labels are available during model training. In order to expand the data volume of equipment failure images, first, an original data set is obtained, which includes a plurality of equipment failure images and a corresponding equipment failure label for each equipment failure image. The equipment failure image is used to represent the collected operation data of the equipment under various operating states, including normal operation data and failure operation data. Therefore, the equipment failure label includes a normal operation label and a failure operation label.

[0096] Please refer to Figure 2 , Figure 2 is a flowchart of the specific method of step S110 provided by the embodiments of the present application. In some embodiments of the present application, step S110 can specifically include but is not limited to steps S210 to S260, which will be described below in combination with Figure 2 The six steps will be described in detail.

[0097] Step S210, obtaining equipment operation data of a sample equipment;

[0098] Step S220, performing data segmentation processing on the equipment operation data according to a preset data length to obtain equipment operation sub-data;

[0099] Step S230, performing short-time Fourier transform on the equipment operation sub-data to obtain equipment transformed data;

[0100] Step S240, performing normalization processing on the equipment transformed data to obtain equipment normalized data;

[0101] Step S250, performing image conversion processing on the equipment normalized data to obtain equipment failure images;

[0102] Step S260, marking the equipment failure images according to the equipment operation data to obtain equipment failure labels of the equipment failure images.

[0103] In step S210 of some embodiments, the equipment operation data of the equipment is obtained through the sensors installed on the sample equipment. The sample equipment can be a bearing, a reduction gearbox, a motor, a drive machine, etc. The equipment operation data is the equipment operation data of the sample equipment under various operating states, which can be normal operation data or abnormal operation data when a failure occurs. The equipment operation data is vibration signal data, which is equivalent to a time series data. The time length of the obtained equipment operation data is not specifically limited.

[0104] In step S220 of some embodiments, in order to improve the efficiency of fault identification, the equipment operation data is subjected to data segmentation processing according to a preset data length, that is, the original equipment operation data is segmented into at least one equipment operation sub-data containing the same preset data length, that is, the data length of the equipment operation sub-data is less than that of the equipment operation data, and each equipment operation sub-data contains the same amount of operation information.

[0105] It should be noted that when the equipment operation data is subjected to data segmentation processing, when the length of the last equipment operation sub-data after data segmentation is less than the preset data length, the last equipment operation sub-data can be subjected to data padding to make the data length of the equipment operation sub-data meet the preset data length, wherein the padded data is normal operation data and does not affect the fault type; or the last equipment operation sub-data can be subjected to data rejection to only keep the equipment operation sub-data meeting the preset data length.

[0106] In step S230 of some embodiments, since the collected equipment operation data is mostly non-stationary sequence data, in order to be able to analyze the deeper data information contained in the equipment operation data in each frequency band, each equipment operation sub-data after segmentation is subjected to short-time Fourier transform (STFT) to obtain operation sample sub-data. The multiple operation sample sub-data are subjected to superposition processing of short-time stationary signals to obtain equipment transform data.

[0107] It should be noted that the equipment transform data can contain N operation sample sub-data, N being a positive integer greater than or equal to 1, and the equipment transform data obtained by the embodiments of the present application is equivalent to signal data in complex form obtained by superposition of N operation sample sub-data. After obtaining the equipment transform data, the amplitude data of each operation sample sub-data can be obtained according to the amplitude calculation method of the complex form of Fourier transform.

[0108] In step S240 of some embodiments, in order to limit the obtained equipment transform data within a certain range, thereby eliminating the adverse effects caused by singular sample data, the equipment transform data is subjected to normalization processing, that is, the amplitude data of each operation sample sub-data is normalized to the range of [0, 255] to obtain multiple normalized amplitude data of the equipment transform data, and the multiple normalized amplitude data are combined to obtain equipment normalized data.

[0109] In step S250 of some embodiments, the device normalized data is image conversion processed according to the preset image size, so as to convert the equal-length device running sub-data into the device fault image with the same size. Specifically, the normalized amplitude data is equivalent to the pixel value of the device fault image. For example, the device running sub-data includes N normalized amplitude data, and the preset image size is MxM, where M is a positive integer. The normalized amplitude data is image conversion processed according to the preset image size, that is, for the i-th normalized amplitude data, the horizontal coordinate value j is obtained by performing a modulus operation on i divided by M, and the vertical coordinate value k is obtained by performing a floor operation on i divided by M and adding 1 to the obtained result, where i, j, and k are integers. The image conversion processing is performed on all normalized amplitude data of the device normalized data according to the preset image size, and the device fault image with the image size of MxM is obtained, and the device fault image is a gray image.

[0110] In step S260 of some embodiments, the device fault image is labeled with a fault type according to the running condition of the collected device running data, and a device fault label of the device fault image is obtained. The device fault label includes a normal running label and a fault running label, where the fault running label is used to represent a plurality of preset fault types. For example, when the sample device is a bearing, the corresponding fault running label can include an inner ring fault label, an outer ring fault label, a rolling element fault label, a bearing shedding label, etc. When the sample device is a reduction gearbox, the corresponding fault running label can include a shaft imbalance fault label, a gear fault label, etc. The embodiment of the present application converts the original one-dimensional time sequence signal into a two-dimensional gray image to obtain input data suitable for the two-dimensional initial network model.

[0111] In step S120 of some embodiments, in order to solve the problem of limited sample quantity of the existing fault label, and in order to avoid the problem of unstable training of the generative adversarial network, the initial network model constructed based on the conditional generative adversarial network structure can be used to increase the richness of the original data set and improve the balance of the original data set. Specifically, the device fault image and the device fault label are input into a preset initial network model, and the initial network model includes a fault image generation sub-model and a fault judgment sub-model.

[0112] In step S130 of some embodiments, the fault image generation sub-model is used to generate a predicted fault image with the same device fault label, that is, the device fault label is input into the fault image generation sub-model for image generation processing, and a predicted fault image is obtained, and the fault label of the predicted fault image is the same as the device fault label.

[0113] Please refer to Figure 3 ,Figure 3 is a flow chart of a specific method of step S130 provided by the embodiments of the present application. In some embodiments of the present application, step S130 can specifically include but is not limited to steps S310 to S340, which are described below in combination with Figure 3 The four steps are described in detail.

[0114] Step S310, generating fault operation data according to a preset data length and a preset distribution function to obtain initial fault operation data;

[0115] Step S320, performing data encoding processing on the initial fault operation data to obtain an initial fault operation vector;

[0116] Step S330, performing label encoding processing on the equipment fault label to obtain a fault label vector;

[0117] Step S340, inputting the initial fault operation vector and the fault label vector into a fault image generation sub-model to perform image generation processing to obtain a predicted fault image.

[0118] In steps S310 and S320 of some embodiments, the preset data length is denoted as a, a is any positive integer, such as 10, 20, etc., and the preset distribution function is a generation function flexibly selected according to needs, such as a Gaussian distribution function. For example, when the preset distribution function is set as a Gaussian distribution function, denoted as P d , the Gaussian distribution P d with a mean of 0 and a standard deviation of 1 is used to generate the initial fault operation data. That is, a plurality of initial fault operation data with a data length of a can be randomly generated. Then, data encoding processing is performed on each of the generated initial fault operation data to obtain an initial fault operation vector, and the initial fault operation vector can be expressed in the form of a 1 x a vector.

[0119] In step S330 and step S340 of some embodiments, in order to enable the fault image generation sub-model to generate samples of the specified feature, that is, to generate predicted fault images with the same label as the device fault label, the device fault label is used as prior information of the image generation sub-model, and on the basis of this prior information, the generated image content is closer to the required expanded image content. Specifically, the device fault label is encoded by one-hot encoding to obtain a fault label vector. For example, when the label type of the device fault label is k, the length of the encoded fault label vector is k. Then, the initial fault operation vector and the fault label vector are input into the fault image generation sub-model for image generation processing to obtain a predicted fault image. The embodiment of the present application can effectively limit the strong free generation ability of the original generative adversarial network to generate fault images with the required label by inputting the encoded fault label vector into the fault image generation sub-model.

[0120] It should be noted that the device fault label can also be encoded by a dummy variable, that is, the encoding method of the label is not limited.

[0121] Please refer to Figure 4 , Figure 4 is a flowchart of the specific method of step S340 provided by the embodiment of the present application. In some embodiments of the present application, the fault image generation sub-model includes a deconvolution layer, a normalization layer and a generation activation layer. Step S340 can include but is not limited to steps S410 to S450. The following will be described in combination with Figure 4 The five steps will be described in detail.

[0122] Step S410, merging the initial fault operation vector and the fault label vector to obtain a predicted fault input vector;

[0123] Step S420, performing fault vector deconvolution processing on the predicted fault input vector according to the deconvolution layer to obtain a deconvolution fault generation vector;

[0124] Step S430, performing normalization processing on the deconvolution fault generation vector according to the normalization layer to obtain a normalized fault generation vector;

[0125] Step S440, performing de-linearization processing on the normalized fault generation vector according to the generation activation layer to obtain a target fault generation vector;

[0126] Step S450, performing image conversion processing on the target fault generation vector to obtain a predicted fault image.

[0127] In steps S410 to S450 of some embodiments, in order to input to the fault image generation sub-model for image generation, first, the initial fault operation vector and the fault label vector are combined into a feature vector to obtain a predicted fault input vector. When the initial fault operation vector is 1 x d and the length of the fault label vector is k, the predicted fault input vector can be expressed in the form of a 1 x (k+d) vector. The predicted fault input vector is used to standardize the image features that need to be learned. In order to map low-dimensional features to high-dimensional features to more accurately restore the detailed features of the input image, the predicted fault input vector is subjected to fault vector deconvolution processing according to the deconvolution layer. In order to more accurately restore the predicted fault image according to the input image features, three deconvolution layers are set, and the three deconvolution layers are sequentially connected. The predicted fault input vector is sequentially subjected to fault vector deconvolution processing by the three deconvolution layers. In order to make the convolution result meet the preset distribution function again, the deconvolution fault generation vector output by each deconvolution layer is subjected to normalization processing according to the normalization layer constructed based on the BN method to obtain a normalized fault generation vector, so that the deep network model is more easily converged and the risk of model overfitting is reduced. Then, the normalized fault generation vector is subjected to de-linearization processing according to the generation activation layer constructed based on ReLU, and the target fault generation vector is output after the generation activation layer corresponding to the last deconvolution layer, so that the model is more easily trained and better performance can be obtained. Finally, the target fault generation vector is subjected to image conversion processing to obtain a predicted fault image with the same size as the device fault image. For example, if the image size of the device fault image is M x M, the image size of the predicted fault image is also M x M.

[0128] In step S140 of some embodiments, in order to judge the image generation ability of the fault image generation sub-model, specifically, the initial discrimination label corresponding to the device fault image is set as a first label, and the initial discrimination label corresponding to the predicted fault image is set as a second label. The device fault image and the predicted fault image are mixed to obtain an image discrimination set, which includes a discrimination image. The discrimination image, the initial discrimination label corresponding to the discrimination image, and the device fault image corresponding to the discrimination image are input to the fault discrimination sub-model for image discrimination processing to obtain predicted discrimination data. The predicted discrimination data is used to represent the probability value of the input discrimination image belonging to the original device fault image and the predicted fault image generated by the fault image generation sub-model.

[0129] Please refer to Figure 5 , Figure 5 is a flowchart of the specific method of step S140 provided by the embodiments of the present application. In some embodiments of the present application, the fault discrimination sub-model includes a convolution layer, a normalization layer, and a discrimination activation layer. Step S140 can specifically include but is not limited to steps S510 to S540, which will be described below in combination with Figure 5The four steps are described in detail.

[0130] In step S510, feature extraction is performed on the discrimination image to obtain a fault discrimination vector.

[0131] In step S520, the fault vector convolution processing is performed on the fault discrimination vector according to the convolution layer to obtain a fault convolution discrimination vector.

[0132] In step S530, the fault convolution discrimination vector is standardized according to the standardization layer to obtain a standard fault discrimination vector.

[0133] In step S540, the standard fault discrimination vector is de-linearized according to the discrimination activation layer to obtain prediction discrimination data.

[0134] In steps S510 to S540 of some embodiments, in order to accurately determine the probability value of the equipment fault image or the prediction fault image to which the discrimination image belongs, first, feature extraction is performed on the discrimination image to obtain a fault discrimination vector. In order to obtain the key feature information of the image, the fault vector convolution processing is performed on the fault discrimination vector according to the convolution layer to obtain a fault convolution discrimination vector. Among them, the fault discrimination sub-model sets three two-dimensional convolution layers, and a standardization layer based on the BN method is set after each convolution layer to standardize the fault convolution discrimination vector and obtain a standard fault discrimination vector. Then, the standard fault discrimination vector is de-linearized according to the discrimination activation layer based on the sigmoid function, that is, the classification result is mapped to the [0, 1] interval, and the prediction discrimination data is output after the discrimination activation layer corresponding to the last convolution layer. Then, the classification probability values of the prediction fault images generated by the original equipment fault image and the fault image generation sub-model can be obtained.

[0135] In step S150 of some embodiments, the loss of the fault data enhancement model is determined by the fault image generation sub-model and the fault discrimination sub-model. The fault image generation sub-model is denoted as G, the fault discrimination sub-model is denoted as D, the model parameters corresponding to G are denoted as θ G , and the model parameters corresponding to D are denoted as θ D . Specifically, when adjusting the parameters of the initial network model according to the equipment fault image, the prediction fault image, the equipment fault label and the prediction discrimination data, first, the model parameters of G are fixed to train the model parameters of D, so that D can accurately distinguish the real samples and the generated samples as much as possible, that is, the model discrimination accuracy of D reaches the preset discrimination accuracy threshold, such as 0.95, 0.96, etc. In the training process of D, the gradient descent method can be used to update the model parameters θ DAfter the training of D is completed, the model parameters of D are fixed to train the model parameters of G, so that the generated predicted failure images and the real device failure images are as close as possible, that is, until D cannot distinguish whether the discrimination image sample comes from the output of G or the real device failure image. In the training process of G, the gradient descent method can also be used to update the model parameters of G. After continuously updating D and G for multiple iterations, the final predicted failure images generated by G are very close to the given device failure images, thereby achieving the purpose of expanding the number of failure samples.

[0136] Please refer to Figure 6 , Figure 6 is a flowchart of a specific method of step S150 provided by the embodiments of the present application. In some embodiments of the present application, the prediction discrimination data includes first discrimination data and second discrimination data, and step S150 can specifically include but is not limited to steps S610 to S630, which will be described below in combination with Figure 6 The three steps will be described in detail.

[0137] Step S610, performing initial device failure loss calculation according to the device failure image, the first discrimination data and the device failure label to obtain initial device failure loss data;

[0138] Step S620, performing generated device failure loss calculation according to the predicted failure image, the second discrimination data and the device failure label to obtain generated device failure loss data;

[0139] Step S630, adjusting the parameters of the initial network model according to the initial device failure loss data and the generated device failure loss data to obtain the failure data enhancement model.

[0140] In steps S610 to S630 of some embodiments, first, the model parameters θ G and θ D of G and D are randomly initialized. When optimizing G, since the failure image generation sub-model is only related to the generated device failure image, initial device failure loss calculation is performed according to the device failure image, the real discrimination data and the device failure label to obtain initial device failure loss data. In order to improve the correct discrimination ability of the failure discrimination sub-model to the image, generated device failure loss calculation is performed according to the predicted failure image, the generated discrimination data and the device failure label to obtain generated device failure loss data. In order to improve the high-quality failure samples under a small number of failure samples, the loss function of the initial network model is taken as the target loss function, denoted as V. The parameters of the initial network model are adjusted according to the initial device failure loss data and the generated device failure loss data to obtain the failure data enhancement model. The target loss function corresponding to the initial network model can be as shown in formula (1).

[0141]

[0142] min G minimizing initial device failure loss data; max D maximizing generated device failure loss data; P data(x) denotes the original data set of real device failure images, x denotes each device failure image, y denotes the device failure label of the device failure image, E(·) denotes the calculation of the expected value, D(·) denotes the predicted discriminant data output by the fault discriminant sub-model after processing the input discriminant data, d denotes the predicted fault image generated by the fault image generation sub-model, P d(d) denotes a set of predicted fault images generated according to a Gaussian distribution function, and G(d|y) denotes a predicted fault image generated under the known device failure label.

[0143] It should be noted that, since the initial discriminant label corresponding to the device failure image in the image discriminant set is the first label, and the initial discriminant label corresponding to the predicted fault image is the second label, the first discriminant data is used to represent the probability data of the input discriminant image being discriminated as the first label, and the second discriminant data is used to represent the probability data of the input discriminant image being discriminated as the second label.

[0144] It should be noted that, in order to improve the generation image quality of the fault image generation sub-model to obtain the fault sample image meeting the requirements, the present embodiment minimizes the probability that the predicted fault image generated by the fault image generation sub-model is correctly discriminated in the fault discriminant sub-model. The present embodiment maximizes the probability that the input discriminant data is correctly discriminated, for example, when the input is a device failure image, the probability value of identifying the first label corresponding to the device failure image is maximized, and when the input is a predicted fault image, the second label value corresponding to the predicted fault image is maximized.

[0145] In step S160 of some embodiments, after obtaining the fault data enhancement model with good data enhancement effect, the device failure image and each device failure label are input into the fault data enhancement model for data enhancement processing, to obtain a plurality of fault enhanced images corresponding to each device failure label. The fault enhanced image has the same fault label as the device failure label from which it is derived. The device failure label includes a normal operation label and different types of fault operation labels.

[0146] In step S170 of some embodiments, in order to improve the accuracy of fault identification, the device fault image and the fault enhanced image are image mixed to obtain an image classification set, the image classification set including a fault classification image and an initial classification label corresponding to the fault classification image, the initial classification label being a device fault label of the device fault image or the fault enhanced image from which it originates. The fault classification image is input to a preset initial classification model for fault identification processing to obtain a predicted fault classification label.

[0147] Please refer to Figure 7 , Figure 7 is a flowchart of a specific method of step S170 provided by the embodiments of the present application. In some embodiments of the present application, step S170 can specifically include but is not limited to steps S710 to S730, which will be described below in combination with Figure 7 The three steps will be described in detail.

[0148] Step S710, image superposition processing is performed on the fault classification image to obtain a fault superposition image;

[0149] Step S720, the fault superposition image is input to the initial classification model for fault identification processing to obtain fault classification data;

[0150] Step S730, a predicted fault classification label of the fault classification image is determined according to the fault classification data.

[0151] In steps S710 to S730 of some embodiments, an initial classification model can be constructed based on VGG-16, and the input of the model is set to be three-channel color image data of 224x224x3, wherein 224x224 is used for the size of the image. Since the obtained fault classification image is a grayscale image, i.e., single-channel data, image superposition processing is performed on the fault classification image to obtain an initial superposition image, which is three-channel color image data. Then, linear image interpolation is used to perform image transformation on the image of each channel of the initial superposition image to obtain a fault superposition image, the size of the image of each channel of the fault superposition image being 224x224. The fault superposition image is input to the initial classification model for fault identification processing to obtain fault classification data, the fault classification data being used to represent classification probability values of the fault superposition image belonging to each preset fault type. Numerical comparison is performed on the multiple classification probability values to determine the predicted fault classification label of the fault classification image.

[0152] It should be noted that the initial classification model can include five classification convolution modules, each of which includes a classification convolution layer and a classification pooling layer. After sequentially passing through each classification convolution module, the input sequentially connects three fully connected layers to map the feature space calculated by the previous convolution, pooling and other layers to the sample label space. Then, the probability that the input image belongs to which class of fault is obtained by normalizing processing through the softmax function.

[0153] In step S180 of some embodiments, in order to ensure the accuracy of fault classification, the initial classification model is adjusted in parameters according to the predicted fault classification label and the device fault label, to obtain a fault recognition model.

[0154] In step S190 of some embodiments, in actual application, the system corresponding to the fault recognition method of the present application can be installed on the terminal. Therefore, when performing fault recognition operation, the terminal generates a fault recognition request and sends the fault recognition request and the target fault image of the to-be-tested device to the fault recognition system. The fault recognition system responds to the fault recognition request and performs fault recognition processing by the fault recognition model to obtain a target fault label, which is used to represent the predicted fault type to which the target fault image belongs.

[0155] It should be noted that after determining the fault type to which the target fault image belongs, the recognized information related to the fault type can be sent to the terminal for display, or the matched type and the processing process information of the fault type can be stored in a set constructed by the historical processing mode, to obtain the processing process information of the fault type corresponding to the device, and the matched type and the processing process information are sent to the corresponding processing object, so that the processing object can refer to the processing process information to repair the to-be-tested device, and the efficiency of fault solving can be improved.

[0156] Please refer to Figure 8 , Figure 8 is a module structure schematic diagram of the fault recognition system provided by the embodiments of the present application. In some embodiments of the present application, the fault recognition system includes:

[0157] The sample acquisition module 810 is configured to acquire a device fault image and a device fault label of the device fault image.

[0158] The model input module 820 is configured to input the device fault image and the device fault label to a preset initial network model. The initial network model includes a fault image generation sub-model and a fault discrimination sub-model.

[0159] The image generation module 830 is configured to input the device fault label to the fault image generation sub-model to perform image generation processing, to obtain a predicted fault image. The fault label of the predicted fault image is the same as the device fault label.

[0160] The fault discrimination module 840 is configured to take the device fault image and the predicted fault image as discrimination images, input the discrimination images to the fault discrimination sub-model for image discrimination processing, and obtain predicted discrimination data.

[0161] The first parameter adjustment module 850 is configured to perform parameter adjustment on the initial network model according to the device fault image, the predicted fault image, the device fault label, and the predicted discrimination data, and obtain a fault data enhancement model.

[0162] The data enhancement module 860 is configured to input the device fault image and the device fault label to the fault data enhancement model for data enhancement processing, and obtain a fault enhancement image, wherein the fault label of the fault enhancement image is the same as the device fault label.

[0163] The image classification module 870 is configured to take the device fault image and the fault enhancement image as fault classification images, input the fault classification images to a preset initial classification model for fault recognition processing, and obtain a predicted fault classification label.

[0164] The second parameter adjustment module 880 is configured to perform parameter adjustment on the initial classification model according to the predicted fault classification label and the device fault label, and obtain a fault recognition model.

[0165] The fault recognition module 890 is configured to obtain a target fault image of a to-be-tested device, input the target fault image to the fault recognition model for fault recognition processing, and obtain a target fault label.

[0166] It should be noted that the fault recognition system of the embodiments of the present application is used to execute the fault recognition method described above, and the fault recognition system of the embodiments of the present application corresponds to the fault recognition method described above. For details of the training process, please refer to the fault recognition method described above, which will not be described here.

[0167] The embodiments of the present application also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor implements the fault recognition method of the embodiments of the present application when executing the computer program.

[0168] The electronic device can be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a vehicle-mounted computer, and the like.

[0169] The embodiments of the present application will be described in detail below. Figure 9 The electronic device of the embodiments of the present application will be described in detail.

[0170] The embodiments of the present application will be described in detail below. Figure 9 , Figure 9 The hardware structure of the electronic device of another embodiment is shown. The electronic device includes:

[0171] The processor 910 can be implemented in a manner of a general central processing unit (CPU), a microprocessor, an application specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute related programs to implement the technical solutions provided by the embodiments of the present application.

[0172] The memory 920 can be implemented in a manner of a read only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 920 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are stored in the memory 920 and are called and executed by the processor 910 to implement the fault identification method of the embodiments of the present application.

[0173] The input / output interface 930 is configured to implement information input and output.

[0174] The communication interface 940 is configured to implement the communication interaction between the device and other devices. The communication can be implemented in a wired manner (for example, a USB, a network cable, etc.) or in a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0175] The bus 950 is configured to transmit information between various components (for example, the processor 910, the memory 920, the input / output interface 930, and the communication interface 940) of the device.

[0176] The processor 910, the memory 920, the input / output interface 930, and the communication interface 940 are connected to each other in the device through the bus 950.

[0177] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the fault identification method of the embodiments of the present application.

[0178] The embodiment of the application provides a fault identification method, a fault identification system, an electronic device and a storage medium. The device fault image and the device fault label of the device fault image are obtained, the device fault image and the device fault label are input into a preset initial network model, and the initial network model comprises a fault image generation submodel and a fault discrimination submodel. In order to make the generated device enhanced image more similar to the real device fault image, the device fault label is input into the fault image generation submodel for image generation processing, and a predicted fault image is obtained, wherein the fault label of the predicted fault image is the same as the device fault label. Then, the device fault image and the predicted fault image are used as discrimination images, and the discrimination images are input into the fault discrimination submodel for image discrimination processing, and prediction discrimination data is obtained. The initial network model is adjusted according to the device fault image, the predicted fault image, the device fault label and the prediction discrimination data, and a fault data enhancement model is obtained. Then, the device fault image and the device fault label are input into the fault data enhancement model for data enhancement processing, and a fault enhanced image is obtained, wherein the fault label of the fault enhanced image is the same as the device fault label. The device fault image and the fault enhanced image are used as fault classification images, and the fault classification images are input into a preset initial classification model for fault identification processing, and a predicted fault classification label is obtained. The initial classification model is adjusted according to the predicted fault classification label and the device fault label, and a fault identification model is obtained. The target fault image of the to-be-tested device is obtained, and the target fault image is input into the fault identification model for fault identification processing, and a target fault label is obtained. According to the fault data enhancement model comprising the fault image generation submodel and the fault discrimination submodel, the data of the input device fault image can be enhanced, the number of labeled fault data is expanded, and the fault identification model with higher accuracy is obtained according to the fault enhanced image and the device fault image. Therefore, when the target fault image of the to-be-tested device is processed according to the fault identification model provided by the embodiment of the application, the cost of large-scale manual labeling can be effectively saved, and efficient fault diagnosis can be realized, that is, the accuracy of fault identification can be improved under a small number of fault samples.

[0179] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory can include a high-speed random access memory and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory that is remotely arranged relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0180] The embodiments described in the specification are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0181] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and can include more or fewer steps than shown in the figures, or combine certain steps, or different steps.

[0182] The device embodiments described above are merely illustrative, and the units described as separate components can or can not be physically separated, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0183] Those skilled in the art can understand that all or some of the steps in the above disclosed method, the functional modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0184] The terms "first", "second", "third", "fourth" and the like (if any) in the specification of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0185] It should be understood that, in the application, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B and A and B existing at the same time, wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b and c can be single or multiple.

[0186] In several embodiments provided in the application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0187] The units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on multiple network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0188] In addition, the functional units in each embodiment of the application can be integrated into a processing unit, or each unit can exist physically, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0189] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes multiple instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0190] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, and are not limited to the scope of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A fault identification method, characterized in that, The method includes: Acquire equipment fault images and equipment fault tags for the equipment fault images; The equipment fault image and the equipment fault label are input into a preset initial network model, which includes a fault image generation sub-model and a fault discrimination sub-model. The equipment fault label is input into the fault image generation sub-model for image generation processing to obtain a predicted fault image, wherein the fault label of the predicted fault image is the same as the equipment fault label. The equipment fault image and the predicted fault image are used as discrimination images, and the discrimination images are input into the fault discrimination sub-model for image discrimination processing to obtain the predicted discrimination data; The parameters of the initial network model are adjusted based on the equipment fault image, the predicted fault image, the equipment fault label, and the predicted discrimination data to obtain a fault data augmentation model. The equipment fault image and the equipment fault label are input into the fault data augmentation model for data augmentation processing to obtain a fault augmented image, wherein the fault label of the fault augmented image is the same as the equipment fault label; The equipment fault image and the fault enhancement image are used as fault classification images, and the fault classification images are input into a preset initial classification model for fault identification processing to obtain predicted fault classification labels; The parameters of the initial classification model are adjusted based on the predicted fault classification label and the equipment fault label to obtain the fault identification model. A target fault image of the device under test is acquired and input into the fault recognition model for fault recognition processing to obtain a target fault label.

2. The method according to claim 1, characterized in that, The acquisition of the device fault image and the device fault tag of the device fault image includes: Obtain the equipment operation data of the sample equipment; The device operation data is segmented according to a preset data length to obtain device operation sub-data. Perform a short-time Fourier transform on the device operation sub-data to obtain the device transformed data; The device transformation data is normalized to obtain device normalized data; The image is converted based on the normalized data of the equipment to obtain the fault image of the equipment; The equipment fault image is labeled with the fault type based on the equipment operation data to obtain the equipment fault label of the equipment fault image.

3. The method according to claim 1, characterized in that, The step of inputting the equipment fault label into the fault image generation sub-model for image generation processing to obtain a predicted fault image includes: Initial fault operation data is obtained by generating fault operation data based on preset data length and preset distribution function. The initial fault operation data is encoded to obtain the initial fault operation vector; The equipment fault tags are processed by tag encoding to obtain fault tag vectors; The initial fault operation vector and the fault label vector are input into the fault image generation sub-model for image generation processing to obtain the predicted fault image.

4. The method according to claim 3, characterized in that, The fault image generation sub-model includes a deconvolution layer, a normalization layer, and a generative activation layer. The step of inputting the initial fault running vector and the fault label vector into the fault image generation sub-model for image generation processing to obtain the predicted fault image includes: The initial fault operation vector and the fault label vector are merged to obtain the predicted fault input vector; The predicted fault input vector is deconvolved by the deconvolution layer to obtain the deconvolution fault generation vector. The normalization layer is used to normalize the deconvolution fault generation vector to obtain a normalized fault generation vector. The normalized fault generation vector is delinearized according to the generation activation layer to obtain the target fault generation vector. The predicted fault image is obtained by performing image transformation processing based on the target fault generation vector.

5. The method according to claim 1, characterized in that, The fault discrimination sub-model includes a convolutional layer, a normalization layer, and a discriminative activation layer; The step of inputting the discrimination image into the fault discrimination sub-model for image discrimination processing to obtain predicted discrimination data includes: Feature extraction is performed on the discrimination image to obtain the fault discrimination vector; The fault discrimination vector is processed by the convolutional layer to obtain the fault convolutional discrimination vector. The fault convolutional discrimination vector is standardized according to the standardization layer to obtain a standard fault discrimination vector; The standard fault discrimination vector is delinearized according to the discrimination activation layer to obtain the predicted discrimination data.

6. The method according to claim 1, characterized in that, The prediction and discrimination data includes first discrimination data and second discrimination data; The step of adjusting the parameters of the initial network model based on the equipment fault image, the predicted fault image, the equipment fault label, and the predicted discrimination data to obtain a fault data augmentation model includes: Based on the equipment fault image, the first discrimination data and the equipment fault label, the initial equipment fault loss is calculated to obtain the initial equipment fault loss data. Based on the predicted fault image, the second discrimination data, and the equipment fault label, the generated equipment fault loss calculation is performed to obtain the generated equipment fault loss data; The parameters of the initial network model are adjusted based on the initial equipment failure loss data and the generated equipment failure loss data to obtain the failure data enhancement model.

7. The method according to any one of claims 1 to 6, characterized in that, The step of inputting the fault classification image into a preset initial classification model for fault identification processing to obtain a predicted fault classification label includes: The fault classification images are overlaid to obtain fault overlay images; The fault overlay image is input into the initial classification model for fault identification processing to obtain fault classification data; The predicted fault classification label of the fault classification image is determined based on the fault classification data.

8. A fault identification system, characterized in that, The system includes: The sample acquisition module is used to acquire equipment fault images and equipment fault labels for the equipment fault images; The model input module is used to input the equipment fault image and the equipment fault label into a preset initial network model, the initial network model including a fault image generation sub-model and a fault discrimination sub-model; An image generation module is used to input the equipment fault label into the fault image generation sub-model for image generation processing to obtain a predicted fault image, wherein the fault label of the predicted fault image is the same as the equipment fault label. The fault discrimination module is used to take the equipment fault image and the predicted fault image as discrimination images, and input the discrimination images into the fault discrimination sub-model for image discrimination processing to obtain predicted discrimination data; The first parameter adjustment module is used to adjust the parameters of the initial network model based on the equipment fault image, the predicted fault image, the equipment fault label and the predicted discrimination data to obtain a fault data enhancement model. The data augmentation module is used to input the equipment fault image and the equipment fault label into the fault data augmentation model for data augmentation processing to obtain a fault augmented image, wherein the fault label of the fault augmented image is the same as the equipment fault label; The image classification module is used to take the equipment fault image and the fault enhancement image as fault classification images, and input the fault classification images into a preset initial classification model for fault identification processing to obtain predicted fault classification labels; The second parameter adjustment module is used to adjust the parameters of the initial classification model according to the predicted fault classification label and the equipment fault label to obtain a fault identification model. The fault identification module is used to acquire the target fault image of the device under test, and input the target fault image into the fault identification model for fault identification processing to obtain the target fault label.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images