Power equipment fault detection method based on neural network

Abstract

The invention discloses an electrical equipment fault detection method based on a neural network, and relates to the technical field of fault detection, and the method comprises the following steps: executing a first operation and a second operation, screening fault components, setting a time window, constructing a multi-channel feature map, training a convolutional neural network, and obtaining a first training model; and obtaining the current fault type according to the first training model. According to the invention, various sensor data and computer vision technologies are combined, and the operation state of the power equipment is captured from multiple dimensions, so that the accuracy of fault detection is improved, the operation state of the power equipment is monitored in real time, abnormal conditions are found in time, the neural network model is utilized to automatically analyze and process data, manual intervention is reduced, and the fault detection efficiency is improved. The automatic and intelligent level of fault detection is improved, the fault mode is identified through learning of historical fault data, deeper analysis is provided for fault diagnosis, utilization of historical data is enhanced, and the value of the data is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a method for detecting faults of power equipment based on a neural network. Background Art

[0002] In recent years, deep learning, especially convolutional neural networks (CNNs), has become the core technology for power equipment fault detection. These algorithms can automatically extract features from images or videos and perform fault diagnosis. They have been widely used in many fields. Power equipment fault detection technology based on neural networks is developing towards higher accuracy, more comprehensive data analysis and greater intelligence.

[0003] Currently, a Chinese invention patent with publication number CN 117541536A discloses a method for detecting power equipment faults based on image feature processing and neural networks. SIFT and HOG are used to extract local image features, enhance the perception of details and textures, and generate candidate regions to reduce the amount of algorithm computation and improve detection speed. However, the related technology does not judge power equipment faults based on multi-dimensional features, lacks comprehensiveness and comprehensiveness in fault judgment, does not conduct a joint analysis of fault types and historical fault data to obtain a fault judgment model, and lacks intelligence and accuracy in fault judgment. Summary of the Invention

[0004] The technical problem solved by the present invention is that the related technology does not judge the faults of power equipment based on multi-dimensional characteristics, lacks the comprehensiveness and comprehensiveness of fault judgment, does not conduct a joint analysis of fault types and historical fault data to obtain a fault judgment model, and lacks the intelligence and accuracy of fault judgment.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, a method for detecting power equipment faults based on a neural network comprises the following steps:

[0006] Step S100, obtaining a power equipment code, setting a first detection condition, and performing a first operation in response to the first detection condition;

[0007] Step S200: obtaining a standard surface image and a standard three-dimensional model of the power equipment based on the serial number of the power equipment, annotating component names and corresponding temperature thresholds in the standard three-dimensional model, setting shooting parameters, acquiring a surface image of the power equipment, preprocessing the surface image of the power equipment, correcting the preprocessed surface image of the power equipment to obtain a first image, calculating similarity between the first image and the standard surface image, setting a similarity threshold, and performing a second operation based on the similarity threshold and the similarity;

[0008] Step S300: setting an acquisition period, monitoring thermal images of the power equipment during the acquisition period, acquiring a three-dimensional thermal image of the power equipment, extracting first temperature values ​​of components, screening faulty components based on the first temperature values, and assigning first numbers to the faulty components;

[0009] Step S400, obtaining any faulty component, matching the historical fault type within the first historical time period with the corresponding historical fault data sequence, looping step S300, obtaining the historical fault first temperature value sequence corresponding to the historical fault three-dimensional thermal image sequence of the faulty component;

[0010] Step S500: Preprocessing a historical fault sound intensity sequence, a historical fault vibration intensity sequence, a historical fault operating current sequence, and a historical fault first temperature value sequence corresponding to any historical fault type to obtain first fault data, setting a time window, splitting the first fault data sequence into subsequences, extracting feature quantities of each subsequence, constructing a multi-channel feature map based on the feature quantities of the subsequences, training a convolutional neural network to obtain a first training model, and assigning a second number to the first training model;

[0011] Step S600, setting the current monitoring time, obtaining the current first temperature value sequence, the current sound intensity sequence, the current vibration intensity sequence and the current operating current sequence within the current monitoring period, and obtaining the current fault type according to the first training model.

[0012] As a preferred solution of the neural network-based power equipment fault detection method of the present invention, step S100 includes the following sub-steps:

[0013] Step S101: Set the code of the device to be detected as the task code, sense the power device code through radio frequency identification, compare the power device code with the task code, and obtain a comparison result, wherein the comparison result includes the power device code and the task code being the same, and the power device code and the task code being different;

[0014] Step S102: When the power equipment code and the task code are the same, the label of the comparison result is set as the first label; when the power equipment code and the task code are different, the label of the comparison result is set as the second label;

[0015] Step S103, setting a first detection condition according to the comparison result, wherein the first detection condition is that the label of the comparison result is a second label;

[0016] Step S104, in response to the first detection condition, performing a first operation, wherein the first operation includes jumping to the next power device and collecting a surface image of the power device;

[0017] When the comparison result meets the first detection condition, the first operation is set to collect the surface image of the power device; when the comparison result does not meet the first detection condition, the first operation is set to jump to the next power device.

[0018] As a preferred solution of the neural network-based power equipment fault detection method of the present invention, step S200 includes the following sub-steps:

[0019] Step S201: When the first operation is to capture a surface image of the power equipment, setting shooting parameters, the shooting parameters including a shooting angle and a shooting distance;

[0020] Step S202: Acquire a surface image of the power equipment, automatically set surface labels on the surface image of the power equipment, the surface labels including top, bottom, left, right, front, and back, and preprocess the surface image of the power equipment, the preprocessing including denoising and filtering.

[0021] Step S203: Retrieve a device database, input the serial number of the power device into the device database, and obtain a standard surface image and a standard three-dimensional model of the power device. The standard three-dimensional model includes surface labels, component names, and temperature thresholds of the component names. The component names include bushing, winding, core, and cooler.

[0022] Step S204, correcting the pre-processed surface image of the power equipment according to the standard surface image of the power equipment, wherein the correction includes magnification, reduction, and rotation to obtain a first image, and traversing the first image to automatically generate a simulated three-dimensional model;

[0023] Step S205, calculating the similarity between the first image and the standard surface image using a cosine similarity formula, setting a similarity threshold, and performing a second operation based on the similarity threshold and the similarity, wherein the second operation includes changing shooting parameters and collecting thermal images of the power equipment;

[0024] The correction logic of the pre-processed surface image of the power equipment includes:

[0025] Obtain any surface image and its corresponding standard image, identify the outer contour of the surface image through machine vision, record it as the first outer contour, identify the outer contour of the corresponding standard image, record it as the second outer contour, set the second outer contour on the bottom layer, set the first outer contour on the top layer, and set the first outer contour to a semi-transparent state. By reducing, enlarging or rotating, the first outer contour and the second outer contour are overlapped to obtain a first image, the overlap is represented by an angle not exceeding a first angle threshold, and the first image is output.

[0026] As a preferred solution of the neural network-based power equipment fault detection method of the present invention, the setting logic of the second operation includes:

[0027] Obtaining a similarity, setting a second value as a similarity threshold, comparing the similarity with the second similarity threshold, and when the similarity is less than the second similarity threshold, setting the second operation to change shooting parameters; and when the similarity is greater than or equal to the second similarity threshold, setting the second operation to capture a thermal image of the power equipment;

[0028] The logic for changing the shooting parameters includes:

[0029] The third value and the fourth value are set as the change in the shooting angle and the change in the shooting distance respectively, the shooting angle and the shooting distance are continuously increased or continuously decreased, and the similarity is continuously calculated until the similarity is greater than or equal to the second similarity threshold, then the shooting parameters are stopped from being changed, and the new shooting angle and the new shooting distance are set and stored as the shooting parameters of the corresponding surface.

[0030] As a preferred solution of the neural network-based power equipment fault detection method of the present invention, step S300 includes the following sub-steps:

[0031] Step S301: When the second operation is to collect thermal images of the power equipment, a first time period is set as a collection period, and a three-dimensional thermal image of a simulated three-dimensional model of the power equipment is monitored during the first time period to obtain component names and corresponding temperature thresholds in the standard three-dimensional model;

[0032] Step S302: Select the location of the component in the three-dimensional thermal image of the simulated three-dimensional model and label it with the component name in the standard three-dimensional model to obtain a thermal map of each component, extract the temperature value of the component thermal map area, and calculate the average of the temperature values, which is recorded as a first temperature value;

[0033] Step S303: compare the first temperature value with the corresponding heat threshold, select the faulty component according to the comparison result, and assign a first number to the faulty component. The comparison result includes the first temperature value being greater than the corresponding heat threshold, and the first temperature value being less than or equal to the corresponding heat threshold. The first number is represented by X. i , where i is a natural number, and the screening logic for the faulty components includes:

[0034] When the comparison result shows that the first temperature value is greater than the corresponding heat threshold, the component is set as a faulty component; when the comparison result shows that the first temperature value is less than or equal to the corresponding heat threshold, the component is deleted from the simulated three-dimensional model.

[0035] As a preferred solution of the neural network-based power equipment fault detection method of the present invention, step S400 includes the following sub-steps:

[0036] Step S401: obtaining any faulty component, accessing a fault database, inputting the name of the faulty component into the fault database, and matching the historical fault type and the corresponding historical fault data sequence within a first historical time period corresponding to the name of the faulty component, wherein the historical fault data sequence includes a historical fault sound intensity sequence, a historical fault vibration intensity sequence, a historical fault operating current sequence, and a historical fault three-dimensional thermal image sequence, and the historical fault data sequence represents each historical fault data collected within a continuous monitoring time;

[0037] The historical fault types include core heating, core noise, winding short circuit, winding open circuit, winding deformation, cooler failure, poor cooler heat dissipation, insulation breakdown, insulation degradation, bushing flashover or breakage, bushing grounding short circuit and oil leakage, among which the historical fault types of bushing include bushing flashover or breakage, bushing grounding short circuit, insulation breakdown and insulation degradation, the historical fault types of winding include winding short circuit, winding open circuit and winding deformation, the historical fault types of core include core heating and core noise, and the historical fault types of cooler include cooler failure, poor cooler heat dissipation and oil leakage;

[0038] Step S402 , looping step S300 , obtaining a historical fault first temperature value sequence corresponding to a historical fault three-dimensional thermal image sequence of the faulty component.

[0039] As a preferred solution of the neural network-based power equipment fault detection method of the present invention, step S500 includes the following sub-steps:

[0040] Step S501: Preprocessing the historical fault sound intensity sequence, historical fault vibration intensity sequence, historical fault operating current sequence, and historical fault first temperature value sequence corresponding to any historical fault type, wherein the preprocessing includes outlier removal and normalization. The normalization process causes each data to be distributed between 0 and 1, thereby obtaining first fault data.

[0041] Step S502: Set a time window for the second time period, divide the first historical time period according to the time window, and split the first fault data sequence into subsequences, wherein the subsequences include a historical fault sound intensity subsequence, a historical fault vibration intensity subsequence, a historical fault operating current subsequence, and a historical fault first temperature value subsequence;

[0042] Step S503: extracting feature quantities of each subsequence, wherein the feature quantities include sound feature quantities, vibration feature quantities, current feature quantities, and temperature feature quantities. The sound feature quantities are spectral features, the vibration feature quantities are variance features, the current feature quantities are mean features, and the temperature feature quantities are rate of change features. The spectral features are obtained by short-time Fourier transform.

[0043] Step S504: Set channels, including sound channel, vibration channel, current channel, and temperature channel, construct a multi-channel feature map, and integrate the multi-channel feature map into a multi-dimensional array. The expression of the multi-dimensional array is:

[0044] SZ=(N,T,W)

[0045] Among them, SZ is a multidimensional array, N is the number of samples, that is, the total number of data included in the subsequence, T is the number of channels, and W is the dimension of the feature quantity, which is recorded as the total number of categories of the feature quantity.

[0046] Step S505: Using the multi-channel feature map as input and the historical fault type as output, the convolutional neural network is trained to obtain a first training model. The first training model is assigned a second number. The convolutional neural network is selected from CNN or RNN. The second number is represented by X. ij , where j is a natural number, the second number is represented by the first number X i The first training model of the j-th fault type of the fault component.

[0047] As a preferred solution of the neural network-based power equipment fault detection method of the present invention, step S600 includes the following sub-steps:

[0048] Step S601: Set the third time period as the current monitoring time, and obtain the current first temperature value sequence, the current sound intensity sequence, the current vibration intensity sequence, and the current operating current sequence within the third time period;

[0049] Step S603: automatically generate a current multi-channel feature map, retrieve a first training model, input the current multi-channel feature map into the first training model, and obtain the current fault type.

[0050] In a second aspect, the present invention provides an electronic device comprising a memory, a processor and a memory storing computer-readable instructions. When the computer-readable instructions are executed by the processor, the steps in any one of the methods described above are executed.

[0051] In a third aspect, the present invention provides a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps in any one of the methods described above are executed.

[0052] The beneficial effects of the present invention are as follows: by combining multiple sensor data and computer vision technology, the operating status of power equipment can be captured from multiple dimensions, thereby improving the accuracy of fault detection, monitoring the operating status of power equipment in real time, promptly discovering abnormal situations, reducing potential equipment damage and accident risks, and automatically analyzing and processing data using a neural network model, reducing manual intervention, and improving the automation and intelligence level of fault detection. By learning from historical fault data, fault modes can be identified, providing more in-depth analysis for fault diagnosis, enhancing the utilization of historical data, and improving the value of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of the basic flow of a neural network-based power equipment fault detection method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0055] Example, see Figure 1 , as an embodiment of the present invention, provides a method for detecting power equipment faults based on a neural network, comprising the following steps:

[0056] Step S100, obtaining a power equipment code, setting a first detection condition, and performing a first operation in response to the first detection condition;

[0057] Step S200: obtaining a standard surface image and a standard three-dimensional model of the power equipment based on the serial number of the power equipment, annotating component names and corresponding temperature thresholds in the standard three-dimensional model, setting shooting parameters, acquiring a surface image of the power equipment, preprocessing the surface image of the power equipment, correcting the preprocessed surface image of the power equipment to obtain a first image, calculating similarity between the first image and the standard surface image, setting a similarity threshold, and performing a second operation based on the similarity threshold and the similarity;

[0058] Step S300: setting an acquisition period, monitoring thermal images of the power equipment during the acquisition period, acquiring a three-dimensional thermal image of the power equipment, extracting first temperature values ​​of components, screening faulty components based on the first temperature values, and assigning first numbers to the faulty components;

[0059] Step S400, obtaining any faulty component, matching the historical fault type within the first historical time period with the corresponding historical fault data sequence, looping step S300, obtaining the historical fault first temperature value sequence corresponding to the historical fault three-dimensional thermal image sequence of the faulty component;

[0060] Step S500: Preprocessing a historical fault sound intensity sequence, a historical fault vibration intensity sequence, a historical fault operating current sequence, and a historical fault first temperature value sequence corresponding to any historical fault type to obtain first fault data, setting a time window, splitting the first fault data sequence into subsequences, extracting feature quantities of each subsequence, constructing a multi-channel feature map based on the feature quantities of the subsequences, training a convolutional neural network to obtain a first training model, and assigning a second number to the first training model;

[0061] Step S600, setting the current monitoring time, obtaining the current first temperature value sequence, the current sound intensity sequence, the current vibration intensity sequence and the current operating current sequence within the current monitoring period, and obtaining the current fault type according to the first training model.

[0062] The present invention combines multiple sensor data and computer vision technology to capture the operating status of power equipment from multiple dimensions, thereby improving the accuracy of fault detection, monitoring the operating status of power equipment in real time, promptly discovering abnormal situations, and reducing potential equipment damage and accident risks. It uses neural network models to automatically analyze and process data, reducing manual intervention and improving the automation and intelligence level of fault detection. By learning from historical fault data, it identifies fault modes, provides more in-depth analysis for fault diagnosis, enhances the utilization of historical data, and improves the value of data.

[0063] The step S100 includes the following sub-steps:

[0064] Step S101: Set the code of the device to be detected as the task code, sense the power device code through radio frequency identification, compare the power device code with the task code, and obtain a comparison result, wherein the comparison result includes the power device code and the task code being the same, and the power device code and the task code being different;

[0065] Step S102: When the power equipment code and the task code are the same, the label of the comparison result is set as the first label; when the power equipment code and the task code are different, the label of the comparison result is set as the second label;

[0066] Step S103, setting a first detection condition according to the comparison result, wherein the first detection condition is that the label of the comparison result is a second label;

[0067] Step S104, in response to the first detection condition, performing a first operation, wherein the first operation includes jumping to the next power device and collecting a surface image of the power device;

[0068] When the comparison result meets the first detection condition, the first operation is set to collect the surface image of the power device; when the comparison result does not meet the first detection condition, the first operation is set to jump to the next power device.

[0069] In specific implementation, the coding of power equipment is sensed by radio frequency identification (RFID) technology and compared with the task coding, which can ensure that the detection operation is only targeted at equipment with specific tasks, avoiding invalid detection of irrelevant equipment. A comparison mechanism for task coding and power equipment coding is set to quickly distinguish target equipment from non-target equipment, thereby optimizing the detection process and improving detection efficiency. Through the setting of coding and labels, it is easier to track the equipment's fault history and maintenance records, which is conducive to the long-term maintenance and management of the equipment.

[0070] The step S200 includes the following sub-steps:

[0071] Step S201: When the first operation is to capture a surface image of the power equipment, setting shooting parameters, the shooting parameters including a shooting angle and a shooting distance;

[0072] Step S202: Acquire a surface image of the power equipment, automatically set surface labels on the surface image of the power equipment, the surface labels including top, bottom, left, right, front, and back, and preprocess the surface image of the power equipment, the preprocessing including denoising and filtering.

[0073] Step S203: Retrieve a device database, input the serial number of the power device into the device database, and obtain a standard surface image and a standard three-dimensional model of the power device. The standard three-dimensional model includes surface labels, component names, and temperature thresholds of the component names. The component names include bushing, winding, core, and cooler.

[0074] Step S204, correcting the pre-processed surface image of the power equipment according to the standard surface image of the power equipment, wherein the correction includes magnification, reduction, and rotation to obtain a first image, and traversing the first image to automatically generate a simulated three-dimensional model;

[0075] Step S205, calculating the similarity between the first image and the standard surface image using a cosine similarity formula, setting a similarity threshold, and performing a second operation based on the similarity threshold and the similarity, wherein the second operation includes changing shooting parameters and collecting thermal images of the power equipment;

[0076] The correction logic of the pre-processed surface image of the power equipment includes:

[0077] Obtain any surface image and its corresponding standard image, identify the outer contour of the surface image through machine vision, record it as the first outer contour, identify the outer contour of the corresponding standard image, record it as the second outer contour, set the second outer contour on the bottom layer, set the first outer contour on the top layer, and set the first outer contour to a semi-transparent state. By reducing, enlarging or rotating, the first outer contour and the second outer contour are overlapped to obtain a first image, the overlap is represented by an angle not exceeding a first angle threshold, and the first image is output.

[0078] In specific implementation, the surface images of power equipment are preprocessed through denoising and filtering to improve image quality, provide clearer data for subsequent image analysis and comparison, automatically set surface labels and shooting parameters to ensure the consistency and standardization of image acquisition, and improve the efficiency and accuracy of image acquisition. By obtaining standard surface images and standard three-dimensional models, an accurate reference standard is provided for the surface images of power equipment, facilitating accurate comparison and analysis. Through machine vision recognition and correction logic, the preprocessed image is automatically corrected to align it with the standard image, thereby improving the accuracy of image comparison.

[0079] The setting logic of the second operation includes:

[0080] Obtaining a similarity, setting a second value as a similarity threshold, comparing the similarity with the second similarity threshold, and when the similarity is less than the second similarity threshold, setting the second operation to change shooting parameters; and when the similarity is greater than or equal to the second similarity threshold, setting the second operation to capture a thermal image of the power equipment;

[0081] The logic for changing the shooting parameters includes:

[0082] The third value and the fourth value are set as the change in the shooting angle and the change in the shooting distance respectively, the shooting angle and the shooting distance are continuously increased or continuously decreased, and the similarity is continuously calculated until the similarity is greater than or equal to the second similarity threshold, then the shooting parameters are stopped from being changed, and the new shooting angle and the new shooting distance are set and stored as the shooting parameters of the corresponding surface.

[0083] In specific implementation, by dynamically adjusting the shooting angle and distance, image acquisition is optimized for different equipment and environmental conditions, the quality and availability of images are improved, and the decision on whether to acquire thermal images is dynamically made based on the similarity results, making the image acquisition process more flexible and able to make the best decision based on actual conditions. By continuously adjusting the shooting parameters and recalculating the similarity until a satisfactory similarity level is reached, the accuracy of subsequent analysis is ensured. The similarity is monitored in real time and the shooting parameters are adjusted as needed, providing a real-time feedback mechanism to ensure the continuity and dynamic adjustment capability of the detection process.

[0084] The step S300 includes the following sub-steps:

[0085] Step S301: When the second operation is to collect thermal images of the power equipment, a first time period is set as a collection period, and a three-dimensional thermal image of a simulated three-dimensional model of the power equipment is monitored during the first time period to obtain component names and corresponding temperature thresholds in the standard three-dimensional model;

[0086] Step S302: Select the location of the component in the three-dimensional thermal image of the simulated three-dimensional model and label it with the component name in the standard three-dimensional model to obtain a thermal map of each component, extract the temperature value of the component thermal map area, and calculate the average of the temperature values, which is recorded as a first temperature value;

[0087] Step S303: Compare the first temperature value with the corresponding heat threshold, screen the faulty components based on the comparison result, and assign a first number to the faulty components. The comparison result includes the first temperature value being greater than the corresponding heat threshold, and the first temperature value being less than or equal to the corresponding heat threshold. The first number is represented by Xi, where i is a natural number. The faulty component screening logic includes:

[0088] When the comparison result shows that the first temperature value is greater than the corresponding heat threshold, the component is set as a faulty component; when the comparison result shows that the first temperature value is less than or equal to the corresponding heat threshold, the component is deleted from the simulated three-dimensional model.

[0089] In specific implementation, by monitoring the three-dimensional thermal images of power equipment within the set acquisition period, the temperature distribution of each component of the equipment can be accurately obtained, providing intuitive temperature data for fault detection. The positions of the components are automatically selected and marked in the three-dimensional thermal image, which simplifies manual operations and improves the efficiency and accuracy of component identification. By calculating the average temperature of the thermal map area of ​​each component and comparing it with the set threshold, components that may have faults can be quickly screened out, thereby improving the speed of fault detection.

[0090] The step S400 includes the following sub-steps:

[0091] Step S401: obtaining any faulty component, accessing a fault database, inputting the name of the faulty component into the fault database, and matching the historical fault type and the corresponding historical fault data sequence within a first historical time period corresponding to the name of the faulty component, wherein the historical fault data sequence includes a historical fault sound intensity sequence, a historical fault vibration intensity sequence, a historical fault operating current sequence, and a historical fault three-dimensional thermal image sequence, and the historical fault data sequence represents each historical fault data collected within a continuous monitoring time;

[0092] The historical fault types include core heating, core noise, winding short circuit, winding open circuit, winding deformation, cooler failure, poor cooler heat dissipation, insulation breakdown, insulation degradation, bushing flashover or breakage, bushing grounding short circuit and oil leakage, among which the historical fault types of bushing include bushing flashover or breakage, bushing grounding short circuit, insulation breakdown and insulation degradation, the historical fault types of winding include winding short circuit, winding open circuit and winding deformation, the historical fault types of core include core heating and core noise, and the historical fault types of cooler include cooler failure, poor cooler heat dissipation and oil leakage;

[0093] Step S402 , looping step S300 , obtaining a historical fault first temperature value sequence corresponding to a historical fault three-dimensional thermal image sequence of the faulty component.

[0094] In specific implementation, by retrieving historical fault data from the fault database, a comprehensive analysis of the faulty components is conducted, including multi-dimensional data such as sound, vibration, current, and thermal images, to match all historical fault types related to the faulty components, ensuring the comprehensiveness of fault diagnosis and avoiding missing any possible causes of the fault.

[0095] The step S500 includes the following sub-steps:

[0096] Step S501: Preprocessing the historical fault sound intensity sequence, historical fault vibration intensity sequence, historical fault operating current sequence, and historical fault first temperature value sequence corresponding to any historical fault type, wherein the preprocessing includes outlier removal and normalization. The normalization process causes each data to be distributed between 0 and 1, thereby obtaining first fault data.

[0097] Step S502: Set a time window for the second time period, divide the first historical time period according to the time window, and split the first fault data sequence into subsequences, wherein the subsequences include a historical fault sound intensity subsequence, a historical fault vibration intensity subsequence, a historical fault operating current subsequence, and a historical fault first temperature value subsequence;

[0098] Step S503: extracting feature quantities of each subsequence, wherein the feature quantities include sound feature quantities, vibration feature quantities, current feature quantities, and temperature feature quantities. The sound feature quantities are spectral features, the vibration feature quantities are variance features, the current feature quantities are mean features, and the temperature feature quantities are rate of change features. The spectral features are obtained by short-time Fourier transform.

[0099] Step S504: Set channels, including sound channel, vibration channel, current channel, and temperature channel, construct a multi-channel feature map, and integrate the multi-channel feature map into a multi-dimensional array. The expression of the multi-dimensional array is:

[0100] SZ=(N,T,W)

[0101] Among them, SZ is a multidimensional array, N is the number of samples, that is, the total number of data included in the subsequence, T is the number of channels, and W is the dimension of the feature quantity, which is recorded as the total number of categories of the feature quantity.

[0102] Step S505: Using the multi-channel feature map as input and the historical fault type as output, the convolutional neural network is trained to obtain a first training model. The first training model is assigned a second number. The convolutional neural network is selected from CNN or RNN. The second number is represented by X. ij , where j is a natural number, and the second number represents the first training model of the j-th fault type of the fault component with the first number Xi.

[0103] In the specific implementation, a time window is set and the data sequence is segmented according to the time window to flexibly capture the fault characteristics in different time periods, improve the model's adaptability to time changes, extract multi-dimensional feature quantities such as sound, vibration, current and temperature from each subsequence, comprehensively describe the fault characteristics, and improve the accuracy of fault detection. Convolutional neural networks (CNN) or recurrent neural networks (RNN) are used to train multi-channel feature maps to automatically learn complex patterns and relationships in the data and improve the performance of fault classification.

[0104] The step S600 includes the following sub-steps:

[0105] Step S601: Set the third time period as the current monitoring time, and obtain the current first temperature value sequence, the current sound intensity sequence, the current vibration intensity sequence, and the current operating current sequence within the third time period;

[0106] Step S603: automatically generate a current multi-channel feature map, retrieve a first training model, input the current multi-channel feature map into the first training model, and obtain the current fault type.

[0107] The present invention combines multiple sensor data and computer vision technology to capture the operating status of power equipment from multiple dimensions, thereby improving the accuracy of fault detection, monitoring the operating status of power equipment in real time, promptly discovering abnormal situations, and reducing potential equipment damage and accident risks. It uses neural network models to automatically analyze and process data, reducing manual intervention and improving the automation and intelligence level of fault detection. By learning from historical fault data, it identifies fault modes, provides more in-depth analysis for fault diagnosis, enhances the utilization of historical data, and improves the value of data.

[0108] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium may be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work 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 Functions specified in one or more boxes

[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting power equipment faults based on a neural network, characterized in that: The following steps are involved: Step S100, obtaining a power equipment code, setting a first detection condition, and performing a first operation in response to the first detection condition; Step S200: obtaining a standard surface image and a standard three-dimensional model of the power equipment based on the serial number of the power equipment, annotating component names and corresponding temperature thresholds in the standard three-dimensional model, setting shooting parameters, acquiring a surface image of the power equipment, preprocessing the surface image of the power equipment, correcting the preprocessed surface image of the power equipment to obtain a first image, calculating similarity between the first image and the standard surface image, setting a similarity threshold, and performing a second operation based on the similarity threshold and the similarity; Step S300: setting an acquisition period, monitoring thermal images of the power equipment during the acquisition period, acquiring a three-dimensional thermal image of the power equipment, extracting first temperature values ​​of components, screening faulty components based on the first temperature values, and assigning first numbers to the faulty components; Step S400, obtaining any faulty component, matching the historical fault type within the first historical time period with the corresponding historical fault data sequence, looping step S300, obtaining the historical fault first temperature value sequence corresponding to the historical fault three-dimensional thermal image sequence of the faulty component; Step S500: Preprocessing a historical fault sound intensity sequence, a historical fault vibration intensity sequence, a historical fault operating current sequence, and a historical fault first temperature value sequence corresponding to any historical fault type to obtain first fault data, setting a time window, splitting the first fault data sequence into subsequences, extracting feature quantities of each subsequence, constructing a multi-channel feature map based on the feature quantities of the subsequences, training a convolutional neural network to obtain a first training model, and assigning a second number to the first training model; Step S600, setting the current monitoring time, obtaining the current first temperature value sequence, the current sound intensity sequence, the current vibration intensity sequence and the current operating current sequence within the current monitoring period, and obtaining the current fault type according to the first training model.

2. The method for detecting power equipment faults based on a neural network according to claim 1, wherein: The step S100 includes the following sub-steps: Step S101: Set the code of the device to be detected as the task code, sense the power device code through radio frequency identification, compare the power device code with the task code, and obtain a comparison result, wherein the comparison result includes the power device code and the task code being the same, and the power device code and the task code being different; Step S102: When the power equipment code and the task code are the same, the label of the comparison result is set as the first label; when the power equipment code and the task code are different, the label of the comparison result is set as the second label; Step S103, setting a first detection condition according to the comparison result, wherein the first detection condition is that the label of the comparison result is a second label; Step S104, in response to the first detection condition, performing a first operation, wherein the first operation includes jumping to the next power device and collecting a surface image of the power device; When the comparison result meets the first detection condition, the first operation is set to collect the surface image of the power device; when the comparison result does not meet the first detection condition, the first operation is set to jump to the next power device.

3. The method for detecting power equipment faults based on a neural network according to claim 1, wherein: The step S200 includes the following sub-steps: Step S201: When the first operation is to capture a surface image of the power equipment, setting shooting parameters, the shooting parameters including a shooting angle and a shooting distance; Step S202: Acquire a surface image of the power equipment, automatically set surface labels on the surface image of the power equipment, the surface labels including top, bottom, left, right, front, and back, and preprocess the surface image of the power equipment, the preprocessing including denoising and filtering. Step S203: Retrieve a device database, input the serial number of the power device into the device database, and obtain a standard surface image and a standard three-dimensional model of the power device. The standard three-dimensional model includes surface labels, component names, and temperature thresholds of the component names. The component names include bushing, winding, core, and cooler. Step S204, correcting the pre-processed surface image of the power equipment according to the standard surface image of the power equipment, wherein the correction includes magnification, reduction, and rotation to obtain a first image, and traversing the first image to automatically generate a simulated three-dimensional model; Step S205, calculating the similarity between the first image and the standard surface image using a cosine similarity formula, setting a similarity threshold, and performing a second operation based on the similarity threshold and the similarity, wherein the second operation includes changing shooting parameters and collecting thermal images of the power equipment; The correction logic of the pre-processed surface image of the power equipment includes: Obtain any surface image and its corresponding standard image, identify the outer contour of the surface image through machine vision, record it as the first outer contour, identify the outer contour of the corresponding standard image, record it as the second outer contour, set the second outer contour on the bottom layer, set the first outer contour on the top layer, and set the first outer contour to a semi-transparent state. By reducing, enlarging or rotating, the first outer contour and the second outer contour are overlapped to obtain a first image, the overlap is represented by an angle not exceeding a first angle threshold, and the first image is output.

4. The method for detecting power equipment faults based on a neural network according to claim 3, wherein: The setting logic of the second operation includes: Obtaining a similarity, setting a second value as a similarity threshold, comparing the similarity with the second similarity threshold, and when the similarity is less than the second similarity threshold, setting the second operation to change shooting parameters; and when the similarity is greater than or equal to the second similarity threshold, setting the second operation to capture a thermal image of the power equipment; The logic for changing the shooting parameters includes: The third value and the fourth value are set as the change in the shooting angle and the change in the shooting distance respectively, the shooting angle and the shooting distance are continuously increased or continuously decreased, and the similarity is continuously calculated until the similarity is greater than or equal to the second similarity threshold, then the shooting parameters are stopped from being changed, and the new shooting angle and the new shooting distance are set and stored as the shooting parameters of the corresponding surface.

5. The method for detecting power equipment faults based on a neural network according to claim 1, wherein: The step S300 includes the following sub-steps: Step S301: When the second operation is to collect thermal images of the power equipment, a first time period is set as a collection period, and a three-dimensional thermal image of a simulated three-dimensional model of the power equipment is monitored during the first time period to obtain component names and corresponding temperature thresholds in the standard three-dimensional model; Step S302: Select the location of the component in the three-dimensional thermal image of the simulated three-dimensional model and label it with the component name in the standard three-dimensional model to obtain a thermal map of each component, extract the temperature value of the component thermal map area, and calculate the average of the temperature values, which is recorded as a first temperature value; Step S303: compare the first temperature value with the corresponding heat threshold, select the faulty component according to the comparison result, and assign a first number to the faulty component. The comparison result includes the first temperature value being greater than the corresponding heat threshold, and the first temperature value being less than or equal to the corresponding heat threshold. The first number is represented by X. i , where i is a natural number, and the screening logic for the faulty components includes: When the comparison result shows that the first temperature value is greater than the corresponding heat threshold, the component is set as a faulty component; when the comparison result shows that the first temperature value is less than or equal to the corresponding heat threshold, the component is deleted from the simulated three-dimensional model.

6. The method for detecting power equipment faults based on a neural network according to claim 1, wherein: The step S400 includes the following sub-steps: Step S401: obtaining any faulty component, accessing a fault database, inputting the name of the faulty component into the fault database, and matching the historical fault type and the corresponding historical fault data sequence within a first historical time period corresponding to the name of the faulty component, wherein the historical fault data sequence includes a historical fault sound intensity sequence, a historical fault vibration intensity sequence, a historical fault operating current sequence, and a historical fault three-dimensional thermal image sequence, and the historical fault data sequence represents each historical fault data collected within a continuous monitoring time; The historical fault types include core heating, core noise, winding short circuit, winding open circuit, winding deformation, cooler failure, poor cooler heat dissipation, insulation breakdown, insulation degradation, bushing flashover or breakage, bushing grounding short circuit and oil leakage, among which the historical fault types of bushing include bushing flashover or breakage, bushing grounding short circuit, insulation breakdown and insulation degradation, the historical fault types of winding include winding short circuit, winding open circuit and winding deformation, the historical fault types of core include core heating and core noise, and the historical fault types of cooler include cooler failure, poor cooler heat dissipation and oil leakage; Step S402 , looping step S300 , obtaining a historical fault first temperature value sequence corresponding to a historical fault three-dimensional thermal image sequence of the faulty component.

7. The method for detecting power equipment faults based on a neural network according to claim 1, wherein: The step S500 includes the following sub-steps: Step S501: Preprocessing the historical fault sound intensity sequence, historical fault vibration intensity sequence, historical fault operating current sequence, and historical fault first temperature value sequence corresponding to any historical fault type, wherein the preprocessing includes outlier removal and normalization. The normalization process causes each data to be distributed between 0 and 1, thereby obtaining first fault data. Step S502: Set a time window for the second time period, divide the first historical time period according to the time window, and split the first fault data sequence into subsequences, wherein the subsequences include a historical fault sound intensity subsequence, a historical fault vibration intensity subsequence, a historical fault operating current subsequence, and a historical fault first temperature value subsequence; Step S503: extracting feature quantities of each subsequence, wherein the feature quantities include sound feature quantities, vibration feature quantities, current feature quantities, and temperature feature quantities. The sound feature quantities are spectral features, the vibration feature quantities are variance features, the current feature quantities are mean features, and the temperature feature quantities are rate of change features. The spectral features are obtained by short-time Fourier transform. Step S504: Set channels, including sound channel, vibration channel, current channel, and temperature channel, construct a multi-channel feature map, and integrate the multi-channel feature map into a multi-dimensional array. The expression of the multi-dimensional array is: SZ=(N,T,W) Among them, SZ is a multidimensional array, N is the number of samples, that is, the total number of data included in the subsequence, T is the number of channels, and W is the dimension of the feature quantity, which is recorded as the total number of categories of the feature quantity. Step S505: Using the multi-channel feature map as input and the historical fault type as output, the convolutional neural network is trained to obtain a first training model. The first training model is assigned a second number. The convolutional neural network is selected from CNN or RNN. The second number is represented by X. ij , where j is a natural number, the second number is represented by the first number X i The first training model of the j-th fault type of the fault component.

8. The method for detecting power equipment faults based on a neural network according to claim 1, wherein: The step S600 includes the following sub-steps: Step S601: Set the third time period as the current monitoring time, and obtain the current first temperature value sequence, the current sound intensity sequence, the current vibration intensity sequence, and the current operating current sequence within the third time period; Step S603: automatically generate a current multi-channel feature map, retrieve a first training model, input the current multi-channel feature map into the first training model, and obtain the current fault type.

9. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores a computer program that can be called by the processor, and the processor executes the neural network-based power equipment fault detection method as described in claims 1-8 by calling the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer is caused to execute the power equipment fault detection method based on a neural network as claimed in claims 1 to 8.

Images