Equipment fault detection method and device, computer program product and electronic equipment

Abstract

The invention discloses an equipment fault detection method and device, a computer program product and electronic equipment. The method relates to the field of artificial intelligence or other related fields, and comprises the following steps: obtaining an initial inspection image, processing the initial inspection image to obtain an inspection image, the initial inspection image being an image collected when an inspection device inspects a target device; obtaining a feature recognition model, inputting the inspection image into the feature recognition model, and outputting an equipment recognition result, the feature recognition model being obtained by training a convolutional neural network model, and the equipment recognition result being used for representing probabilities of different fault types when the target equipment has a fault; and determining a target fault type of the target equipment according to the equipment identification result. Through the method and the device, the problems of low detection efficiency and low detection accuracy when the target equipment is detected in the prior art are solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence or other related fields, and specifically, to a method, apparatus, computer program product and electronic device for detecting equipment failure. Background Art

[0002] With the continuous expansion of the scale of power systems and the continuous advancement of technology, the safe and stable operation of power systems has become an important guarantee for social development. In modern power systems, main grid equipment is the core component of power transmission, and its health directly affects the reliability and power supply quality of the entire system. However, due to the influence of various factors such as the complex and changeable natural environment and equipment aging, main grid equipment faces many challenges, such as wildfires, lightning damage, foreign body intrusion, etc. These events may not only lead to power supply interruptions, but also trigger more serious chain reactions, endangering public safety.

[0003] Traditional equipment monitoring and fault detection methods mainly rely on manual inspections and regular maintenance. Manual inspections are time-consuming and labor-intensive, and it is difficult to achieve 24-hour uninterrupted monitoring, resulting in low efficiency. In addition, the above methods are subject to personnel working hours and weather conditions. Bad weather and personnel working conditions may affect the inspection results. In emergency situations, manual inspections may not be able to respond in time, and there may also be problems with slow response speeds.

[0004] In order to improve the monitoring efficiency and response speed of the power system, in recent years, with the rapid development of artificial intelligence technology, the power industry has been widely explored, and intelligent inspection equipment has emerged. They can automatically perform inspection tasks and collect status information of main network equipment through high-definition cameras, infrared thermal imagers, acoustic wave detectors and other sensors. However, intelligent inspection equipment lacks the ability to effectively capture and analyze fault characteristics in complex environments, especially when dealing with changeable and complex fault types such as wildfires, lightning damage and foreign object intrusion. The recognition accuracy is often reduced due to the influence of environmental factors, and the above-mentioned equipment fails to fully consider the differences in the urgency and response strategies of different fault types, which limits the actual application effect of the early warning system.

[0005] With regard to the problems of low detection efficiency and low detection accuracy when detecting target devices in related technologies, no effective solution has been proposed so far. Summary of the invention

[0006] The main purpose of the present application is to provide a method, apparatus, computer program product and electronic device for detecting equipment failure, so as to solve the problems of low detection efficiency and low detection accuracy when detecting target equipment in the related art.

[0007] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for detecting equipment failure is provided. The method comprises: obtaining an initial inspection image, processing the initial inspection image, and obtaining an inspection image, wherein the initial inspection image is an image collected when the inspection device inspects the target equipment; obtaining a feature recognition model, inputting the inspection image into the feature recognition model, and outputting an equipment recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the equipment recognition result is used to characterize the probability of different fault types when the target equipment fails; and determining the target fault type of the target equipment according to the equipment recognition result.

[0008] Furthermore, the feature recognition model includes a convolutional pooling layer, an attention module and a classification layer. The inspection image is input into the feature recognition model, and the output device recognition result includes: using the convolutional pooling layer to extract features from the inspection image to obtain M feature images, wherein the features include at least: color features, texture features, shape features and position features, and M is a positive integer; using the attention module to calculate the attention weights of the M feature images to obtain M image attention weights, and performing weighted calculation on the corresponding feature images based on each image attention weight to obtain M weighted feature maps; using the classification layer and the M weighted feature maps to calculate the recognition probability of different fault types of the target device to obtain the device recognition result.

[0009] Furthermore, the attention weights of M feature images are calculated using the attention module to obtain M image attention weights, including: calculating the key-value vector of each feature image to obtain M key-value vectors; obtaining a mapping function, and extracting environmental data from each feature image, processing each environmental data according to the mapping function to obtain M query vectors; for each feature image, determining the image weight matrix of each feature image, and determining the preset parameters of each feature image; calculating the preset parameters of each feature image, the key-value vector associated with each feature image, the query vector of each feature image, and the image weight matrix of each feature image to obtain M feature parameters; normalizing each feature parameter to obtain the image attention weight of each feature image.

[0010] Furthermore, determining the image weight matrix of each feature image includes: for a feature image, obtaining an initial weight matrix, generating an initial value of each initial weight matrix using a Gaussian distribution, and obtaining a first adjustment weight matrix; calculating the gradient of the first adjustment weight matrix using a back propagation algorithm to obtain a matrix gradient, and adjusting the first adjustment weight matrix using the matrix gradient to obtain a second adjustment weight matrix; adjusting the second adjustment weight matrix using environmental data of the feature image to obtain a third adjustment weight matrix, and determining the third adjustment weight matrix as the image weight matrix.

[0011] Furthermore, the attention module includes a convolutional layer and a fully connected layer, and calculating the key-value vector of each feature image includes: for a feature image, using the convolutional layer to capture the local pattern of the feature image to obtain a local pattern feature map; obtaining a conversion function, using the conversion function to process the local pattern feature map to obtain a first processed feature map; using the fully connected layer to perform pooling on the first processed feature map to obtain a key-value vector associated with the feature image.

[0012] Furthermore, a mapping function is obtained, and environmental data is extracted from each feature image. Each environmental data is processed according to the mapping function to obtain M query vectors, including: for a feature image, a segmentation algorithm is used to perform segmentation processing on the environmental data in the feature image to obtain N segmented feature images, where N is a positive integer; an edge detection algorithm is used to extract features from the N segmented feature images to obtain N environmental features; the N environmental features are encoded to obtain an environmental vector, and the environmental vector is processed using a mapping function to obtain a query vector of the feature image.

[0013] Furthermore, after determining the target fault type of the target device based on the device identification result, the method also includes: obtaining a warning strategy list, determining the fault level based on the target fault type, wherein the warning strategy list includes a plurality of associations between fault levels and warning strategies, and the fault level is used to characterize the severity of the fault of the target device; extracting the target warning strategy from the warning strategy list using the fault level, and processing the target device based on the target warning strategy.

[0014] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a device for detecting equipment failure is provided. The device comprises: a first acquisition unit, used to acquire an initial inspection image, process the initial inspection image, and obtain an inspection image, wherein the initial inspection image is an image collected when the inspection device inspects the target equipment; a second acquisition unit, used to acquire a feature recognition model, input the inspection image into the feature recognition model, and output the equipment recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the equipment recognition result is used to characterize the probability of different fault types when the target equipment fails; a determination unit, used to determine the target fault type of the target equipment according to the equipment recognition result.

[0015] According to another aspect of an embodiment of the present invention, a computer storage medium is provided. The computer storage medium is used to store a program. When the program is executed, the device where the computer storage medium is located is controlled to execute a method for detecting device failure.

[0016] According to another aspect of an embodiment of the present invention, an electronic device is provided, comprising one or more processors and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein a method for detecting device failure is executed when the computer-readable instructions are executed.

[0017] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, a method for detecting a device failure is performed.

[0018] Through the present application, the following steps are adopted: obtaining an initial inspection image, processing the initial inspection image, and obtaining an inspection image, wherein the initial inspection image is an image collected when the inspection device inspects the target device; obtaining a feature recognition model, inputting the inspection image into the feature recognition model, and outputting a device recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the device recognition result is used to characterize the probability of different fault types when a fault occurs in the target device; determining a target fault type of the target device according to the device recognition result, thereby solving the problem of low detection efficiency and low detection accuracy when detecting the target device in the related technology, obtaining an inspection image by processing the collected initial inspection image, inputting the inspection image into the feature recognition model, and outputting a device recognition result, and then determining the target fault type of the target device based on the device recognition result, thereby achieving the effect of improving the accuracy and efficiency of identifying device faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 is a flow chart of a method for detecting device failure according to an embodiment of the present application;

[0021] Figure 2 is a schematic diagram of a method for determining a device identification result provided in an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of a device for detecting equipment failure provided according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set between this system and the relevant user or organization. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0028] It should be noted that the collected information used in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse to use it.

[0029] The present invention is described below in conjunction with preferred implementation steps. Figure 1 is a flow chart of a method for detecting a device failure according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0030] Step S101, acquiring an initial inspection image, and processing the initial inspection image to obtain an inspection image, wherein the initial inspection image is an image captured when an inspection device inspects a target device.

[0031] Specifically, the target device can be a main network device. Due to the unpredictability of the natural environment and the aging of the equipment, it is easy to cause the main network equipment to have faults such as wildfires, lightning damage, and foreign objects intruding into the transmission channel. These faults may not only cause power outages, but also cause more serious safety accidents. Therefore, it is necessary to perform real-time detection of the main network equipment and respond in time based on the detection results. When detecting the target device, the initial inspection image collected by the inspection equipment when inspecting the target device can be collected first, and the image can be pre-processed and other operations can be performed on the image to obtain the inspection image.

[0032] Step S102, obtaining a feature recognition model, inputting the inspection image into the feature recognition model, and outputting a device recognition result, wherein the feature recognition model is trained by a convolutional neural network model, and the device recognition result is used to characterize the probability of different fault types when a target device fails.

[0033] Specifically, the feature recognition model can be trained by a convolutional neural network model. After obtaining the inspection image, the inspection image can be input into the feature recognition model. The convolutional pooling layer in the feature recognition model learns the features of the inspection image to obtain several feature images, which are then transmitted to the attention module. The attention module sets attention weights for the feature images that can reflect the importance of the features in identifying faults. After strengthening the important features, each feature image is weighted with the corresponding attention weight to obtain a weighted feature map, which is then transmitted to the classification layer. Finally, the fault type is identified according to the importance of the features. At this time, the feature recognition model can input the device recognition results used to characterize the probability of different fault types when the main network equipment fails.

[0034] Step S103: determining the target fault type of the target device according to the device identification result.

[0035] Specifically, after the feature recognition model outputs the device identification result, since the device identification result can characterize the probabilities of different fault types when the target device fails, the target fault type of the target device can be determined based on the device identification result, that is, the fault type with the highest probability is selected as the identified target fault type.

[0036] The device fault detection method provided in the embodiment of the present application obtains an initial inspection image, processes the initial inspection image, and obtains an inspection image, wherein the initial inspection image is an image collected when the inspection device inspects the target device; obtains a feature recognition model, inputs the inspection image into the feature recognition model, and outputs a device recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the device recognition result is used to characterize the probabilities of different fault types when the target device fails; determines the target fault type of the target device according to the device recognition result, thereby solving the problems of low detection efficiency and low detection accuracy when detecting the target device in the related art, obtains an inspection image by processing the collected initial inspection image, inputs the inspection image into the feature recognition model, outputs the device recognition result, and then determines the target fault type of the target device based on the device recognition result, thereby achieving the effect of improving the accuracy and efficiency of identifying device faults.

[0037] Figure 2 is a schematic diagram of a method for determining a device identification result according to an embodiment of the present application, such as Figure 2 As shown, optionally, in the device fault detection method provided in the embodiment of the present application, the feature recognition model includes a convolutional pooling layer, an attention module and a classification layer, and the inspection image is input into the feature recognition model, and the output device recognition result includes:

[0038] Step S201, using a convolutional pooling layer to extract features from the inspection image to obtain M feature images, wherein the features include at least: color features, texture features, shape features and position features, and M is a positive integer.

[0039] Step S202, using the attention module to calculate the attention weights of the M feature images, obtain M image attention weights, and perform weighted calculation on the corresponding feature images based on each image attention weight to obtain M weighted feature maps.

[0040] Step S203, using the classification layer and M weighted feature graphs to calculate the recognition probability of different fault types existing in the target device, and obtain the device recognition result.

[0041] Specifically, when using the feature recognition model to identify the inspection image, since the feature recognition model is trained by the convolutional neural network model, the feature recognition model includes a convolutional pooling layer, an attention module and a classification layer. At this time, the convolutional pooling layer can extract color features, texture features, shape features and position features from the inspection image and generate multiple feature images. For example, a color feature map, a texture feature map, a shape feature map and a position feature map can be generated.

[0042] The attention module can be further used to calculate the attention weight for the above-mentioned feature image to obtain the corresponding image attention weight, and then each feature image is weighted with the corresponding image attention weight to obtain a weighted feature map, and then the weighted feature map is transmitted to the classification layer, which can include wildfire fault neurons, lightning fault neurons, and foreign body intrusion fault neurons. Then, the activation function softmax function can be used to calculate the recognition probability of the weighted feature map belonging to the above three fault types, and the fault type with the highest recognition probability is selected as the identified fault type, that is, the equipment recognition result is obtained. This embodiment processes the inspection image by using the feature recognition model, and then outputs the equipment recognition result that can determine the equipment fault type, providing data support for the staff to perform equipment maintenance.

[0043] In order to strengthen important features and suppress unimportant features, an attention module can be used to calculate the attention weight of each feature image. Optionally, in the device fault detection method provided in an embodiment of the present application, the attention weights of M feature images are calculated using an attention module to obtain M image attention weights, including: calculating the key value vector of each feature image to obtain M key value vectors; obtaining a mapping function, and extracting environmental data from each feature image, processing each environmental data according to the mapping function to obtain M query vectors; for each feature image, determining the image weight matrix of each feature image, and determining the preset parameters of each feature image; calculating the preset parameters of each feature image, the key value vector associated with each feature image, the query vector of each feature image, and the image weight matrix of each feature image to obtain M feature parameters; normalizing each feature parameter to obtain the image attention weight of each feature image.

[0044] Specifically, it is first necessary to use the convolutional layer in the attention module to calculate the key-value vector of the feature image, and then obtain the query vector based on the environmental data from the feature image, such as light intensity, occluded area, background complexity, etc.

[0045] After determining the image weight matrix and preset parameters of the importance of each feature image in the entire inspection image, the attention weight of the feature image can be calculated according to the query vector, the key value vector, the image weight matrix and the preset parameters. The image attention weight α of each feature image can be calculated by the following formula: i :

[0046]

[0047] Among them, α i represents the image attention weight of the i-th feature image, g(·) represents the activation function, and q env represents the query vector of the i-th feature image, F irepresents the i-th feature image, j represents the feature in the feature image, c represents the color feature, t represents the texture feature, s represents the shape feature, and p represents the position feature. Represents the key value vector corresponding to the i-th feature image, A j Represents a preset parameter, and k represents the total number of feature images. This embodiment calculates the image attention weights of different feature images by using the query vector, key value vector, image weight matrix and preset parameters, which can provide data support for the subsequent output device recognition results, thereby improving the precision and accuracy of recognition.

[0048] The image attention weight needs to be determined by the image weight matrix. Optionally, in the device fault detection method provided in the embodiment of the present application, determining the image weight matrix of each feature image includes: for a feature image, obtaining an initial weight matrix, generating an initial value of each initial weight matrix using a Gaussian distribution, and obtaining a first adjustment weight matrix; calculating the gradient of the first adjustment weight matrix using a back propagation algorithm to obtain a matrix gradient, and adjusting the first adjustment weight matrix using the matrix gradient to obtain a second adjustment weight matrix; adjusting the second adjustment weight matrix using the environmental data of the feature image to obtain a third adjustment weight matrix, and determining the third adjustment weight matrix as the image weight matrix.

[0049] Specifically, before calculating the image attention weight, it is also necessary to determine the image weight matrix of each feature image. First, define the initialized weight matrix, that is, obtain the initial weight matrix, and then generate random numbers through Gaussian distribution to assign initial values ​​to the elements in the initial weight matrix, and then obtain the first adjusted weight matrix.

[0050] The back propagation algorithm is further used to calculate the gradient of the loss function relative to each adjustment weight matrix, and these gradients are used to update the adjustment weight matrix, that is, first the input feature image is sent to the neural network, and the output of each layer is calculated until the final prediction result is obtained, and then an appropriate loss function is used to measure the difference between the prediction and the actual label, and the gradient of the loss function relative to each adjustment weight matrix is ​​calculated to obtain the matrix gradient, and finally the optimizer is used to adjust the first adjustment weight matrix based on the matrix gradient to obtain the second adjustment weight matrix. Finally, the elements in the second adjustment weight matrix are dynamically updated using environmental factors to adapt to the recognition requirements under different environmental conditions, and the image weight matrix of each feature image is obtained.

[0051] It should be noted that the elements in the second adjustment weight matrix can be updated in a variety of ways: an adaptive learning rate method that adjusts the learning rate to adapt to different environmental conditions; a conditional training method that uses a network architecture with additional inputs (i.e., environmental factors) to allow the model to learn directly from the environment; a transfer learning method in which a pre-trained model can be fine-tuned according to a specific environment to adapt to new conditions. This embodiment adjusts the elements of the initial weight matrix to obtain an image weight matrix, laying the foundation for calculating the image attention weight.

[0052] The image attention weight also needs to be determined by the key-value vector. Optionally, in the device fault detection method provided in the embodiment of the present application, the attention module includes a convolutional layer and a fully connected layer. Calculating the key-value vector of each feature image includes: for a feature image, using the convolutional layer to capture the local pattern of the feature image to obtain a local pattern feature map; obtaining a conversion function, and using the conversion function to process the local pattern feature map to obtain a first processed feature map; using the fully connected layer to perform pooling processing on the first processed feature map to obtain a key-value vector associated with the feature image.

[0053] Specifically, the key value vector is extracted from the feature image and can be calculated through the convolution layer and the fully connected layer. It can represent the key information of the feature image. The key value vector can be calculated according to different feature images (color, texture, shape, position). For example, when calculating the key value vector of the color feature map, the convolution kernel of the convolution layer is first used to capture the local pattern of the feature image, such as edges, textures, etc., to obtain the local pattern feature map. After the convolution layer, the conversion function is used The local pattern feature map is processed to obtain a first processed feature map, wherein the conversion function can be a nonlinear activation function such as a ReLU function, or a neural network composed of multiple layers.

[0054] The first processed feature map after the conversion function processing can be further globally pooled to obtain a vector of fixed length, which is used as the key vector associated with the feature image, wherein the global pooling processing method can be global average pooling or global maximum pooling. This embodiment calculates the key vector of each feature image, which can be used for further attention weight calculation to improve the model's ability to understand the inspection image.

[0055] The image attention weight also needs to be determined by the query vector. Optionally, in the device fault detection method provided in the embodiment of the present application, a mapping function is obtained, and environmental data is extracted from each feature image. Each environmental data is processed according to the mapping function to obtain M query vectors, including: for a feature image, a segmentation algorithm is used to segment the environmental data in the feature image to obtain N segmented feature images, where N is a positive integer; an edge detection algorithm is used to extract features from the N segmented feature images to obtain N environmental features; the N environmental features are encoded to obtain an environmental vector, and the environmental vector is processed using a mapping function to obtain a query vector of the feature image.

[0056] Specifically, when calculating the query vector of each feature image, since each feature image includes multiple environmental data, including but not limited to light intensity, occlusion area, background complexity, etc., the light intensity can be obtained by histogram analysis. Then, the environmental data in the feature image is segmented by a segmentation algorithm, wherein the unmarked area is the occlusion area. Then, the edge detection algorithm is used to detect the edge of the image, that is, feature extraction is performed on the segmented feature image. The more edges are detected, the higher the background complexity of the image, and then the environmental features are obtained. Finally, the environmental features are encoded, and the encoded environmental vector is converted using a mapping function to obtain the query vector of the feature image.

[0057] It should be noted that the conversion of the query vector can be achieved through one or more fully connected layers, and if the environmental data is numerical, these values ​​can be used directly for segmentation and other processing; if the environmental data is categorical, it needs to be converted into a dense vector through an embedding layer, and then the dense vector is segmented and other processing is performed. This embodiment converts the query vector obtained by using the environmental data in the feature image, which can be used to calculate the attention weight, thereby improving the model's ability to understand the inspection image.

[0058] Optionally, in the device fault detection method provided in the embodiment of the present application, after determining the target fault type of the target device based on the device identification result, the method also includes: obtaining a warning strategy list, determining the fault level based on the target fault type, wherein the warning strategy list includes a plurality of associations between fault levels and warning strategies, and the fault level is used to characterize the severity of the fault of the target device; extracting the target warning strategy from the warning strategy list using the fault level, and processing the target device based on the target warning strategy.

[0059] Specifically, since the cause of the target equipment failure may be due to the natural environment or equipment aging, such as wildfires, lightning damage, and foreign objects intruding into the transmission channel, different causes will lead to complex and changeable fault types, so the fault types can be classified and early warning strategies can be formulated according to the classified levels. After identifying the target fault type, the fault level is determined by the fault type, and then a list of early warning strategies including the association between multiple fault levels and early warning strategies is obtained, and then the early warning strategy corresponding to the fault level is found from the list, and finally the target equipment is processed based on the determined early warning strategy.

[0060] It should be noted that the purpose of fault level classification is to organize responses based on the urgency, potential destructiveness and processing priority of the fault. Therefore, the fault level of the fault type can be divided into: emergency level, high risk level, and low risk level. For example, when the fault type is an emergency level, the emergency plan should be immediately activated and a professional team should be dispatched to handle it on site; when the fault type is a high risk level, a patrol team should be dispatched to conduct on-site inspections, assess risks, and take preventive measures if necessary; when the fault type is a low risk level, it can be recorded and continuously monitored, regular inspections can be arranged, and emergency supplies and equipment can be prepared. The early warning strategy can include four links: monitoring, early warning triggering, response actions, and follow-up tracking. For example, robots and drones can be used to monitor wildfires; lightning monitoring networks can be used to monitor lightning activities, and cameras and infrared detectors can be used to monitor foreign object intrusion in power transmission channels; thresholds can be set for fire power distance, lightning density, foreign object size, and foreign object location, and corresponding early warning signals can be set for fault type levels. When the fault data exceeds the threshold, the corresponding early warning signal is automatically triggered and an alarm is issued. This embodiment determines different early warning levels by using the target fault type, and then determines different early warning strategies, which can not only reduce labor costs, but also achieve the purpose of accurate processing.

[0061] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0062] The embodiment of the present application also provides a device for detecting a device failure. It should be noted that the device for detecting a device failure in the embodiment of the present application can be used to execute the method for detecting a device failure provided in the embodiment of the present application. The device for detecting a device failure provided in the embodiment of the present application is introduced below.

[0063] Figure 3 is a schematic diagram of a device failure detection apparatus provided according to an embodiment of the present application, such as Figure 3 As shown, the device includes: a first acquisition unit 30, a second acquisition unit 31, and a determination unit 32.

[0064] A first acquisition unit 30 is used to acquire an initial inspection image and process the initial inspection image to obtain an inspection image, wherein the initial inspection image is an image collected when the inspection device inspects the target device;

[0065] The second acquisition unit 31 is used to acquire a feature recognition model, input the inspection image into the feature recognition model, and output a device recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the device recognition result is used to characterize the probability of different fault types when a fault occurs in the target device;

[0066] The determination unit 32 is used to determine the target fault type of the target device according to the device identification result.

[0067] The device for detecting equipment failure provided in the embodiment of the present application obtains an initial inspection image through a first acquisition unit 30, processes the initial inspection image, and obtains an inspection image, wherein the initial inspection image is an image collected when the inspection device inspects the target equipment; the second acquisition unit 31 obtains a feature recognition model, inputs the inspection image into the feature recognition model, and outputs an equipment recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the equipment recognition result is used to characterize the probabilities of different fault types when a fault occurs in the target equipment; the determination unit 32 determines the target fault type of the target equipment according to the equipment recognition result, thereby solving the problems of low detection efficiency and low detection accuracy when detecting the target equipment in the related art, obtains an inspection image by processing the collected initial inspection image, inputs the inspection image into the feature recognition model, outputs the equipment recognition result, and then determines the target fault type of the target equipment based on the equipment recognition result, thereby achieving the effect of improving the accuracy and efficiency of identifying equipment failures.

[0068] Optionally, in the equipment failure detection device provided in the embodiment of the present application, the second acquisition unit 31 includes: a first extraction module, used to use a convolutional pooling layer to extract features from the inspection image to obtain M feature images, wherein the features include at least: color features, texture features, shape features and position features, and M is a positive integer; a first calculation module, used to use an attention module to calculate the attention weights of the M feature images to obtain M image attention weights, and perform weighted calculation on the corresponding feature images based on each image attention weight to obtain M weighted feature maps; a second calculation module, used to use a classification layer and M weighted feature maps to calculate the recognition probability of different fault types of the target device to obtain a device recognition result.

[0069] Optionally, in the equipment failure detection device provided in the embodiment of the present application, the second acquisition unit 31 includes: a third calculation module, used to calculate the key value vector of each feature image to obtain M key value vectors; a first acquisition module, used to obtain a mapping function, and extract environmental data from each feature image, and process each environmental data according to the mapping function to obtain M query vectors; a determination module, used to determine the image weight matrix of each feature image and determine the preset parameters of each feature image for each feature image; a fourth calculation module, used to calculate the preset parameters of each feature image, the key value vector associated with each feature image, the query vector of each feature image and the image weight matrix of each feature image to obtain M feature parameters; a first processing module, used to normalize each feature parameter to obtain the image attention weight of each feature image.

[0070] Optionally, in the equipment failure detection device provided in the embodiment of the present application, the second acquisition unit 31 includes: a second acquisition module, used to obtain an initial weight matrix for a feature image, generate an initial value of each initial weight matrix using a Gaussian distribution, and obtain a first adjustment weight matrix; a fifth calculation module, used to calculate the gradient of the first adjustment weight matrix using a back propagation algorithm to obtain a matrix gradient, and adjust the first adjustment weight matrix using the matrix gradient to obtain a second adjustment weight matrix; an adjustment module, used to adjust the second adjustment weight matrix using the environmental data of the feature image to obtain a third adjustment weight matrix, and determine the third adjustment weight matrix as the image weight matrix.

[0071] Optionally, in the equipment failure detection device provided in the embodiment of the present application, the second acquisition unit 31 includes: a capture module, which is used to capture the local pattern of a feature image using a convolution layer to obtain a local pattern feature map; a third acquisition module, which is used to obtain a conversion function, and use the conversion function to process the local pattern feature map to obtain a first processed feature map; a second processing module, which is used to perform pooling processing on the first processed feature map using a fully connected layer to obtain a key-value vector associated with the feature image.

[0072] Optionally, in the equipment failure detection device provided in the embodiment of the present application, the second acquisition unit 31 includes: a third processing module, used to perform segmentation processing on the environmental data in a feature image using a segmentation algorithm to obtain N segmented feature images, where N is a positive integer; a second extraction module, used to perform feature extraction on the N segmented feature images using an edge detection algorithm to obtain N environmental features; an encoding module, used to encode the N environmental features to obtain an environmental vector, and process the environmental vector using a mapping function to obtain a query vector of the feature image.

[0073] Optionally, in the device for detecting equipment failure provided in the embodiment of the present application, the device further includes: a third acquisition unit, for obtaining a warning strategy list after determining the target failure type of the target device according to the device identification result, and determining the fault level according to the target failure type, wherein the warning strategy list includes a plurality of associations between fault levels and warning strategies, and the fault level is used to characterize the severity of the failure of the target device; an extraction unit, for extracting the target warning strategy from the warning strategy list using the fault level, and processing the target device based on the target warning strategy.

[0074] The above-mentioned device failure detection apparatus includes a processor and a memory. The above-mentioned first acquisition unit 30, second acquisition unit 31, determination unit 32, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0075] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the problems of low detection efficiency and low detection accuracy in detecting the target device in the related technology can be solved by adjusting the kernel parameters.

[0076] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0077] An embodiment of the present invention provides a computer storage medium, which is used to store a program. When the program is running, the device where the computer storage medium is located is controlled to execute a device failure detection method.

[0078] Figure 4 is a schematic diagram of an electronic device provided according to an embodiment of the present application, such as Figure 4 As shown, an embodiment of the present invention provides an electronic device, the electronic device 40 includes a processor, a memory, and a program stored in the memory and executable on the processor, the processor is used to execute a computer-readable instruction, wherein the computer-readable instruction executes a method for detecting a device failure when it is executed. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0079] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of a device failure detection method in each embodiment of the present application.

[0080] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0084] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0085] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0086] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0087] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0088] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for detecting equipment failure, characterized in that: include: Acquire an initial inspection image, and process the initial inspection image to obtain an inspection image, wherein the initial inspection image is an image captured when the inspection device inspects the target device; Acquire a feature recognition model, input the inspection image into the feature recognition model, and output a device recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the device recognition result is used to characterize the probability of different fault types when the target device fails; A target fault type of the target device is determined according to the device identification result.

2. The method according to claim 1, characterized in that The feature recognition model includes a convolutional pooling layer, an attention module, and a classification layer. The inspection image is input into the feature recognition model, and the output device recognition result includes: Extract features from the inspection image using the convolutional pooling layer to obtain M feature images, wherein the features include at least: color features, texture features, shape features, and position features, and M is a positive integer; Calculating the attention weights of the M feature images using the attention module to obtain M image attention weights, and performing weighted calculation on the corresponding feature images based on each image attention weight to obtain M weighted feature maps; The classification layer and the M weighted feature maps are used to calculate the recognition probability of different fault types existing in the target device to obtain the device recognition result.

3. The method according to claim 2, characterized in that The attention weights of the M feature images are calculated using the attention module to obtain the M image attention weights, including: Calculate the key value vector of each feature image to obtain M key value vectors; Obtain a mapping function, extract environmental data from each feature image, and process each environmental data according to the mapping function to obtain M query vectors; For each feature image, determining an image weight matrix for each feature image, and determining preset parameters for each feature image; Calculate the preset parameters of each feature image, the key value vector associated with each feature image, the query vector of each feature image, and the image weight matrix of each feature image to obtain M feature parameters; Each feature parameter is normalized to obtain the image attention weight of each feature image.

4. The method according to claim 3, characterized in that The image weight matrix for each feature image is determined by: For a feature image, an initial weight matrix is ​​obtained, and an initial value of each initial weight matrix is ​​generated using a Gaussian distribution to obtain a first adjustment weight matrix; Calculating the gradient of the first adjustment weight matrix using a back propagation algorithm to obtain a matrix gradient, and adjusting the first adjustment weight matrix using the matrix gradient to obtain a second adjustment weight matrix; The second adjustment weight matrix is ​​adjusted using the environmental data of the feature image to obtain a third adjustment weight matrix, and the third adjustment weight matrix is ​​determined as the image weight matrix.

5. The method according to claim 3, characterized in that: The attention module includes a convolutional layer and a fully connected layer, and the key value vector of each feature image is calculated including: For a feature image, using the convolution layer to capture a local pattern of the feature image to obtain a local pattern feature map; Acquire a conversion function, and use the conversion function to process the local pattern feature map to obtain a first processed feature map; The first processed feature map is pooled using the fully connected layer to obtain a key-value vector associated with the feature image.

6. The method according to claim 3, characterized in that A mapping function is obtained, and environmental data is extracted from each feature image. Each environmental data is processed according to the mapping function to obtain M query vectors including: For a feature image, a segmentation algorithm is used to segment the environment data in the feature image to obtain N segmented feature images, where N is a positive integer; Using an edge detection algorithm to extract features from the N segmented feature images to obtain N environmental features; The N environmental features are encoded to obtain an environmental vector, and the environmental vector is processed using the mapping function to obtain a query vector of the feature image.

7. The method according to claim 1, characterized in that After determining the target fault type of the target device according to the device identification result, the method further includes: Obtaining a warning strategy list, and determining a fault level according to the target fault type, wherein the warning strategy list includes a correlation relationship between multiple fault levels and warning strategies, and the fault level is used to characterize the severity of the fault of the target device; A target early warning strategy is extracted from the early warning strategy list using the fault level, and the target device is processed based on the target early warning strategy.

8. A device for detecting equipment failure, characterized in that: include: A first acquisition unit is used to acquire an initial inspection image and process the initial inspection image to obtain an inspection image, wherein the initial inspection image is an image collected when the inspection device inspects the target device; A second acquisition unit is used to acquire a feature recognition model, input the inspection image into the feature recognition model, and output a device recognition result, wherein the feature recognition model is obtained by training a convolutional neural network model, and the device recognition result is used to characterize the probability of different fault types when the target device fails; A determination unit is used to determine a target fault type of the target device according to the device identification result.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the device failure detection method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the device failure detection method described in any one of claims 1 to 7.

Images