A method and system for detecting loss of power metal

By combining neural network with virtual physical models, a destructive detection model that can detect tension, tension and halo voltage of power metal tools is constructed, which solves the problem of single and complex detection methods of power metal tools in the existing technology, and achieves more efficient , more comprehensive detection results.

CN118706196BActive Publication Date: 2024-11-15WENSHANG POWER SUPPLY CO OF STATE GRID SHANDONG ELECTRIC POWER CO
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411194579.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-15
Estimated Expiration
2044-08-29

AI Technical Summary

Technical Problem

The existing technology cannot fully detect the tension, tension and halo voltage problems caused by long-term wind erosion of power tools. The deep learning model is complex, the training speed is slow, and it is easy to overfit, and the detection method is single, and there are loopholes.

Method used

The improved neural network model is combined with the virtual physical model, and the tension-wind speed, tension-wind speed and halo voltage-wind speed functions are constructed by collecting image data and simulated operation data, the threshold is calculated, and the feature extraction and Similarity calculation, build a destructive detection model, and realize multi-faceted detection of power tools.

Benefits of technology

实现了对电力金具的全面检测,避免了检测漏洞,提高了检测准确率,减少了模型大小和训练时间,避免了过拟合问题。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118706196B_ABST
    Figure CN118706196B_ABST
Patent Text Reader

Abstract

The present invention relates to a method and system for detecting lossy power hardware, belonging to the field of lossy detection technology, including: collecting image data, manufacturing data and simulated operation data of the power hardware to be tested, respectively constructing functions between tension, tension resistance and corona inception voltage and wind speed, and calculating the corresponding tension threshold, tension resistance threshold and corona inception voltage threshold; constructing an improved neural network model and a virtual physical model, and combining the improved neural network model and the virtual physical model to obtain a lossy detection model, obtaining the actual operation data and actual image data of the power hardware and inputting them into the lossy detection model, and outputting lossy detection results. The present invention can detect whether the power hardware is lossy and the lossy type from multiple aspects, enriching the detection means of the power hardware, making it more comprehensive, and avoiding the existence of detection loopholes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of lossy detection, and in particular to a lossy detection method and system based on power metal. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Transmission lines are an important way of power transmission in power grids and are the key to maintaining the long-term safe and stable operation of power systems. Transmission lines are generally composed of poles, wires and various power fittings. Among them, power fittings are essential metal accessories that fix, connect and support wires and poles, wires and insulators. Power fittings play an important role in transmitting mechanical loads, electrical loads or providing some kind of protection.

[0004] Power fittings are often connected at high altitudes. Due to the high wind speed at high altitudes, especially in areas that are often affected by wind and sand, the environment is relatively harsh, and power fittings will be damaged to varying degrees. Studies have found that power fittings need to withstand great tension and tensile strength, especially in areas with high wind speeds and strong wind and sand, which will cause the transmission lines and fittings to shake continuously. During the process, the tension and tensile strength of the power fittings will be further increased, which can easily lead to damage to the power fittings. In addition, since the power fittings are eroded by wind for a long time, the surface is damaged and seriously contaminated, which will affect the surface electric field distribution and corona characteristics of the power fittings. It can be seen that wind speed has a huge impact on power fittings. In order to prevent power fittings from failing due to damage, power fittings need to be tested.

[0005] At present, the hardware detection methods are generally divided into ultrasonic detection and deep learning detection. The ultrasonic detection method, such as the patent application number CN202410775997.1, discloses a hardware defect detection method, which uses the ultrasonic propagation principle to detect the hardware to be tested, without damage and with high sensitivity. However, the ultrasonic wave in this patent can only detect whether there are defects on the surface and inside of the hardware, and cannot detect the problems caused by the hardware being subjected to large tensile force, tension resistance and corona initiation voltage for a long time. With the introduction of deep learning technology, the existing technology has gradually adopted a combination of deep learning and image processing technology to detect power hardware. However, only the surface defects can be detected based on the image alone, and the detection means are relatively single. The existing technology cannot perform comprehensive lossy detection of power hardware, and there are detection loopholes. In addition, the existing technology has the problem of complex models and slow training speed in the process of deep learning, and often requires a large number of samples for learning, otherwise there will be overfitting problems. Summary of the invention

[0006] In order to solve the technical problems in the prior art, the present invention proposes a method and system for detecting damage to power hardware. By introducing a virtual physical model and improving the traditional neural network, on the one hand, it enriches the detection means of power hardware and avoids detection loopholes. On the other hand, it improves the problems existing in traditional deep learning.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a method for detecting damage to power metal, comprising:

[0009] Collect image data and manufacturing data of the electrical fittings to be tested, and pre-process the acquired image data;

[0010] Collect simulated operation data of the electric fittings to be tested, and construct tension-wind speed function, tension resistance-wind speed function and corona inception voltage-wind speed function respectively; and calculate the tension threshold, tension resistance threshold and corona inception voltage threshold of the electric fittings to be tested;

[0011] An improved neural network model is constructed, the image data of the electrical fittings to be tested is input into the improved neural network model, and the corresponding lossy type is output;

[0012] According to the manufacturing data, a virtual physical model is constructed, and the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function as well as the tension threshold, the tension resistance threshold and the corona inception voltage threshold are input into the virtual physical model, and the destructive detection model is obtained in combination with the improved neural network model; the actual operation data and the actual image data of the power fittings are obtained and input into the destructive detection model, and the destructive detection results are output.

[0013] According to a further technical solution, the simulated operation data of the electric fittings to be tested include: simulated tensile force data, simulated tensile resistance data and simulated corona inception voltage data of the electric fittings; and the manufacturing data include: painting data, cutting data, grinding data and polishing data.

[0014] A further technical solution is that the specific steps of constructing the tension-wind speed function and calculating the tension threshold of the electrical fittings to be tested are: obtaining simulated tension data and the wind speed corresponding to the simulated tension data, drawing a tension-wind speed curve, and calculating the functional relationship in the curve to construct the tension-wind speed function; taking the derivative of the tension-wind speed function, collecting all derivative values ​​when the tension is reduced, and calculating the average value of all derivative values, and using it as the tension threshold.

[0015] A further technical solution is that the specific steps of constructing the tension resistance-wind speed function and calculating the tension resistance threshold of the electrical fittings to be tested are: obtaining simulated tension resistance data and the wind speed corresponding to the simulated tension resistance data, drawing a tension resistance-wind speed curve, and calculating the functional relationship in the curve to construct the tension resistance-wind speed function; taking the derivative of the tension resistance-wind speed function, collecting all derivative values ​​when the tension resistance is reduced, and calculating the average value of all derivative values, and using it as the tension resistance threshold.

[0016] A further technical solution is that the specific steps of constructing the corona inception voltage-wind speed function and calculating the corona inception voltage threshold of the electrical fittings to be tested are: obtaining simulated corona inception voltage data and the wind speed corresponding to the simulated corona inception voltage data, drawing a corona inception voltage-wind speed curve, and calculating the functional relationship in the curve to construct the corona inception voltage-wind speed function; taking the derivative of the corona inception voltage-wind speed function, collecting all derivative values ​​when the corona inception voltage is reduced, and calculating the average value of all derivative values, and using it as the corona inception voltage threshold.

[0017] In a further technical solution, the construction of the improved neural network includes fusing features and calculating similarity, and the specific process of fusing features is:

[0018] First, select 150×150×3 image data and input it into the deep residual network for feature extraction. Then select the feature maps of the last three residual structures as the first feature maps, which are , and ,right and After the maximum pooling operation, the three feature maps are superimposed to generate a new feature map as the second feature map. Finally, a convolution operation is performed to obtain the third feature map, which is used to calculate similarity.

[0019] In a further technical solution, the similarity is calculated by:

[0020] Similar ),in, and The result of paving each channel of the third feature map by fusion features of the image data according to the pixel points is: for The transpose of , Q represents the number of channels of the feature map.

[0021] In a second aspect, the present invention provides a power metal damage detection system, comprising:

[0022] The data acquisition and processing module is configured to: acquire image data and manufacturing data of the power fittings to be tested, and pre-process the acquired image data;

[0023] The threshold calculation module is configured to: collect simulated operation data of the power fittings to be tested, respectively construct tension-wind speed function, tension resistance-wind speed function and corona initiation voltage-wind speed function; and calculate the tension threshold, tension resistance threshold and corona initiation voltage threshold of the power fittings to be tested;

[0024] The neural network building module is configured to: build an improved neural network model, input the image data of the power fittings to be tested into the improved neural network model, and output corresponding lossy features;

[0025] The lossy detection module is configured to: construct a virtual physical model according to the manufacturing data, input the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function as well as the tension threshold, the tension resistance threshold and the corona inception voltage threshold into the virtual physical model, and obtain a lossy detection model in combination with the improved neural network model; obtain the actual operation data and actual image data of the power fittings and input them into the lossy detection model, and output the lossy detection results.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. The present invention obtains the corresponding normal threshold value by collecting the tensile force, tensile strength and corona inception voltage data of the power fittings, so that in the subsequent lossy detection process, it is possible to detect whether the power fittings are damaged and the type of damage from multiple aspects.

[0028] 2. The present invention improves the traditional neural network, including fusing features and calculating similarity. The similarity of two sets of image data is judged by comparing the inner product size of the feature graphs of the two sets of image data. Whether the two sets of image data belong to the same category is measured based on the similarity, and the binary classification task of defective or not is completed. While ensuring a high accuracy rate, the training speed is faster, the problem of overfitting is avoided, and the overall model is smaller.

[0029] 3. The present invention introduces a virtual physical model and combines it with an improved neural network to further achieve more comprehensive detection of electrical fittings. At the same time, this application constructs a virtual physical model based on electrical fittings, enriches the detection methods of electrical fittings, makes them more comprehensive, and avoids detection loopholes. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0031] Figure 1 This is a flow chart of the destructive testing of the present invention; DETAILED DESCRIPTION

[0032] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0034] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0035] Embodiment 1

[0036] In this embodiment, Figure 1 As shown, this embodiment provides a method for detecting damage to power metal, the steps comprising:

[0037] S1: Collect image data and manufacturing data of the power fittings to be tested, and pre-process the acquired image data;

[0038] S2: Collect the simulated operation data of the power fittings to be tested, and construct the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function respectively; and calculate the tension threshold, tension resistance threshold and corona inception voltage threshold of the power fittings to be tested;

[0039] S3: construct an improved neural network model, input the image data of the power fittings to be tested into the improved neural network model, and output the corresponding lossy type;

[0040] S4: According to the manufacturing data, a virtual physical model is constructed, and the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function as well as the tension threshold, the tension resistance threshold and the corona inception voltage threshold are input into the virtual physical model, and the destructive detection model is obtained in combination with the improved neural network model; the actual operation data and the actual image data of the power fittings are obtained and input into the destructive detection model, and the destructive detection results are output.

[0041] In step S1 , the collected manufacturing data includes: shape data, painting data, cutting data, grinding data, polishing data, and the like.

[0042] In step S1, the method for preprocessing the acquired image data is: using the improved MixUp algorithm to enhance the image data. Specifically, the essence of the improved MixUp algorithm is to select any two groups of images in each batch and fuse them into a new image according to a set ratio. The specific formula is as follows:

[0043] = +( ;

[0044] in, and Represents two groups of images selected from the same batch, represents an arbitrary number sampled from a beta distribution, and The value of is between 0 and 1. All acquired image data are preprocessed by improving the MixUp algorithm. Since the loss on the electrical fittings is often small or not very lossy, the traditional image enhancement method cannot learn the features on the image data very well. In this embodiment, the image data is enhanced by using the improved MixUp algorithm, which improves the generalization ability of the image processing model for processing small lossy images or images with less loss.

[0045] In step S2, the collected simulated operation data of the power fitting to be tested includes: simulated tensile force data, simulated tensile resistance data and simulated corona inception voltage data of the power fitting;

[0046] In step S2, the specific method of constructing the tension-wind speed function and calculating the tension threshold of the electrical fitting to be tested is:

[0047] Method for constructing tension-wind speed function: obtain several simulated tension data of the power fittings to be tested, recorded as { , , … }, , , … Indicates the first, second, third, ... nth simulated tension data obtained, where n is a positive integer; the wind speed corresponding to each simulated tension data is obtained, recorded as { , , … }, , , … Indicates the wind speed corresponding to the 1st, 2nd, 3rd, ... nth simulated tension data, where n is a positive integer;

[0048] A tension-wind speed curve is drawn using the several simulated tension data of the electrical fittings to be tested obtained above and the wind speed corresponding to each simulated tension data. The corresponding functional relationship in the tension-wind speed curve is calculated using the piecewise function method as L=f(S), where L represents the simulated tension data, S represents the wind speed, and f(S) is the functional relationship between the two.

[0049] The method for calculating the tension threshold of the power fitting to be tested is as follows: the corresponding function in the calculated tension-wind speed curve is derivated to obtain the derivative function of the tension-wind speed function: = ,in Indicates the rate of change of simulated tension data, It represents the function between the rate of change of simulated tension data and wind speed. It can be seen that when the derivative is less than 0, it means that the independent variable is decreasing. Then, based on the derivative function of the tension-wind speed function, when When <0, it means that the simulated tension data decreases, and the derivative values ​​of all simulated tension decreases are collected and recorded as { , , ……、 },in , , ……、 represents the derivative value when the 1st, 2nd, 3rd, ... kth simulated tension data collected are reduced, k is a positive integer and less than n;

[0050] Calculate the average value of the collected derivative values ​​when the simulated tension decreases. The specific formula is as follows:

[0051] = ;in, It represents the average value of the derivative value when the tension decreases in all simulations, Represents the derivative value when the simulated tension data is reduced, i∈[1, k], and the average value of all derivative values ​​obtained As the tensile force threshold of the electrical fittings to be tested.

[0052] The specific method for constructing the tension resistance-wind speed function and calculating the tension resistance threshold of the power fittings to be tested is:

[0053] Method for constructing the tension resistance-wind speed function: obtain several simulated tension resistance data of the power fittings to be tested, recorded as { , , … }, , , … Indicates the first, second, third, ... nth simulated tension resistance data obtained, where n is a positive integer; the wind speed corresponding to each simulated tension resistance data is obtained, recorded as { , , … }, , , … Indicates the wind speed corresponding to the 1st, 2nd, 3rd, ... nth simulated tension resistance data, where n is a positive integer;

[0054] A tension resistance-wind speed curve is drawn using the several simulated tension resistance data of the electrical fittings to be tested obtained above and the wind speed corresponding to each simulated tension resistance data. The corresponding functional relationship in the tension resistance-wind speed curve is calculated using the piecewise function method as H=g(S), where H represents the simulated tension resistance data, S represents the wind speed, and g(S) is the functional relationship between the two.

[0055] The method for calculating the tension resistance threshold value of the power fittings to be tested is as follows: the corresponding function in the calculated tension resistance-wind speed curve is derivated to obtain the derivative function of the tension resistance-wind speed function: = ,in Indicates the rate of change of simulated tensile strength data, Represents the function between the rate of change of simulated tension resistance data and wind speed, based on the derivative function of the tension resistance-wind speed function, when When <0, it means that the simulated tensile strength data is reduced, and the derivative values ​​of all simulated tensile strength reductions are collected and recorded as { , , ……、 },in , , ……、 It represents the derivative value when the 1st, 2nd, 3rd, ... kth simulated tensile resistance data collected are reduced, k is a positive integer and less than n;

[0056] Calculate the average value of the collected simulated tensile resistance reduction derivative values, the specific formula is as follows:

[0057] = ;in, represents the average of all derivative values, Represents the derivative value when the simulated tension resistance data is reduced, i∈[1, k], and the average value of all derivative values ​​obtained As the tension resistance threshold of the electrical fittings to be tested.

[0058] The specific method for constructing the corona inception voltage-wind speed function and calculating the corona inception voltage threshold of the power fitting to be tested is:

[0059] Method for constructing the corona inception voltage-wind speed function: obtain several simulated corona inception voltage data of the power fittings to be tested, denoted as { , , … }, , , … Indicates the 1st, 2nd, 3rd, ... nth simulated corona inception voltage data obtained, where n is a positive integer; the wind speed corresponding to each simulated corona inception voltage data is obtained, and is recorded as { , , … }, , , … Indicates the wind speed corresponding to the 1st, 2nd, 3rd, ... nth simulated corona inception voltage data, where n is a positive integer;

[0060] A corona inception voltage-wind speed curve is drawn using the several simulated corona inception voltage data of the electrical fittings to be tested obtained as mentioned above and the wind speed corresponding to each simulated corona inception voltage data. The corresponding functional relationship in the corona inception voltage-wind speed curve is calculated using the piecewise function method as D=x(S), where D represents the simulated corona inception voltage data, S represents the wind speed, and x(S) is the functional relationship between the two.

[0061] The method for calculating the corona inception voltage threshold of the power fitting to be tested is as follows: the corresponding function in the calculated corona inception voltage-wind speed curve is derivated to obtain the derivative function of the corona inception voltage-wind speed function: = ,in Indicates the rate of change of the simulated corona inception voltage data, It represents the function between the changing rate of the simulated corona inception voltage data and the wind speed. Based on the derivative function of the corona inception voltage-wind speed function, when When it is less than 0, it means that the simulated corona inception voltage data is reduced. The derivative values ​​of all simulated corona inception voltages when they are reduced are collected and recorded as { , , ……、 },in , , ……、 represents the derivative value when the 1st, 2nd, 3rd, ... kth simulated corona inception voltage data collected are reduced, k is a positive integer and less than n;

[0062] Calculate the average value of the collected derivative values ​​when the simulated corona inception voltage decreases. The specific formula is as follows:

[0063] = ;in, represents the average of all derivative values, represents the derivative value when the simulated corona inception voltage data is reduced, i∈[1, k], and the average value of all derivative values ​​obtained is As the corona inception voltage threshold of the electrical fittings to be tested.

[0064] In step S3, the loss on the power fittings is often small or not much. Traditional neural networks often lose their generalization ability during the processing and lack adaptability to such loss, resulting in low resolution of feature maps, weak expression ability, and difficulty in expressing detailed features, which ultimately leads to poor accuracy. This embodiment improves the traditional neural network and constructs an improved neural network. The improved neural network includes an input layer, a convolution layer, an activation function, a pooling layer, and an output layer. The improvements are: fusion features and calculation similarity, specifically:

[0065] First, select 150×150×3 image data and input it into the deep residual neural network (ResNet) for feature extraction. Select the feature maps of the last three residual structures as the first feature maps, which are , and ,in The size is 20×20×128, The size is 10×10×256, The size of is 3×3×512, and After the maximum pooling operation, and The size is converted to 3×3×128 and 3×3×256, and then the maximum pooling operation is performed and and Superposition is performed to generate a new feature map as the second feature map, the size of the second feature map is 3×3×896, and finally a convolution operation is performed, and the number of channels is adjusted to 128 to obtain the third feature map, the size of the third feature map is 3×3×128, and it is used to calculate the similarity.

[0066] In the process of calculating the similarity, this embodiment conceives that after obtaining the third feature map, each channel is flattened according to the pixels to obtain a feature matrix, and the feature matrix obtained by the third feature map of one image data is multiplied by the transpose of the feature matrix obtained by the third feature map of another image data. The specific calculation method of the similarity is:

[0067] Similar ),in, and The result of paving each channel of the third feature map by fusion features of the image data according to the pixel points is: for The transpose of Q represents the number of channels of the feature map. The image data of the power fittings to be tested is input into the improved neural network model, and the corresponding lossy type is output.

[0068] In step S4, a virtual physical model is constructed according to the manufacturing data. In this embodiment, the virtual physical model is selected as a digital twin model. The specific process includes: collecting manufacturing data including but not limited to shape data, painting data, galvanizing data, cutting data, grinding data, and polishing data, etc.; modeling the power fittings according to the above data collected. A virtual physical model based on the physical power fittings is constructed. After the virtual physical model based on the physical power fittings is constructed, the tension-wind speed function, the tension resistance-wind speed function, the corona inception voltage-wind speed function, the tension threshold, the tension resistance threshold, and the corona inception voltage threshold are input into the virtual physical model.

[0069] In order to be able to perform more comprehensive lossy detection on power fittings and enrich lossy detection methods, in this embodiment, the improved neural network constructed in step S3 is combined with a virtual physical model based on physical power fittings to obtain a lossy detection model. The lossy detection model specifically includes: a basic layer, a data interaction layer, a model construction layer, a specific practice layer and a deep learning layer, wherein the basic layer includes physical data such as the specific shape and size of the power fittings, the painting time, the total amount of galvanizing, the cutting position, the grinding time and the polishing time. The data of the basic layer will be transmitted to the data interaction layer, which is responsible for saving and integrating the physical data of the power fittings. It is a combination of a real physical environment and a virtual environment. The model construction layer includes physical device twins of the power fittings tension, tension resistance and corona initiation voltage data sensors, and each twin obtains the simulated tension, tension resistance and corona initiation voltage data of the power fittings through sensors. The specific practice layer can realize the visualization of the virtual physical model driven by the simulated tension, tension resistance and corona initiation voltage data of the corresponding power fittings. The deep learning layer is relatively independent and includes the improved neural network constructed above, which is used to output lossy types and enrich the means of lossy detection of power equipment.

[0070] In step S4, the tension-wind speed function, the tension resistance-wind speed function and the corona initiation voltage-wind speed function as well as the tension threshold, the tension resistance threshold and the corona initiation voltage threshold are input into the virtual physical model, and the method for detecting the power fittings is as follows:

[0071] Collect real-time wind speed According to the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function, the corresponding tension, tension resistance and corona inception voltage are calculated in real time as shown below:

[0072] =f( );

[0073] =g( );

[0074] =x( );

[0075] in, , and They respectively represent the real-time corresponding tension, tension resistance and corona inception voltage data.

[0076] when ≥ When the tension on the electrical fitting is abnormal, the destructive test result is output and the tension is abnormal. < When , the destructive test result is output and the tension is judged to be normal;

[0077] when ≥ When the tension resistance of the power fittings is abnormal, the destructive test results are output and the tension resistance is abnormal. < When the destructive test result is output, it is determined that the tensile strength is normal;

[0078] when < When the corona inception voltage of the power fitting is abnormal, the destructive test result is output and the corona inception voltage is determined to be abnormal. ≥ When the damage detection result is output, it is determined that the corona inception voltage is normal. When it is determined that there is abnormal damage, this embodiment can optimize and adjust the manufacturing data, improve the power fittings from the root, and obtain better power fittings.

[0079] Through the lossy detection model, on the one hand, after inputting the image data of the power fittings, the improved neural network in the deep learning layer can output the specific lossy type, and can clearly obtain the defects on the surface of the power fittings, including loose nuts, missing nuts, surface corrosion, etc. On the other hand, according to virtual physics, it can be considered whether the power fittings are lossy under the conditions of tension, tensile strength and corona inception voltage. In this embodiment, digital twin technology and deep learning technology are combined to enrich the lossy detection means and avoid the existence of detection loopholes. According to the designed virtual physical model, when the tension, tensile strength and corona inception voltage are abnormal, the manufacturing data can be optimized and adjusted, so as to obtain new and better power fittings in all aspects.

[0080] Embodiment 2

[0081] This embodiment provides a power metal damage detection system, which specifically includes the following modules:

[0082] The data acquisition and processing module is configured to: acquire image data and manufacturing data of the power fittings to be tested, and pre-process the acquired image data;

[0083] The threshold calculation module is configured to: collect simulated operation data of the power fittings to be tested, respectively construct tension-wind speed function, tension resistance-wind speed function and corona initiation voltage-wind speed function; and calculate the tension threshold, tension resistance threshold and corona initiation voltage threshold of the power fittings to be tested;

[0084] The neural network building module is configured to: build an improved neural network model, input the image data of the power fittings to be tested into the improved neural network model, and output corresponding lossy features;

[0085] The lossy detection module is configured to: construct a virtual physical model according to the manufacturing data, input the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function as well as the tension threshold, the tension resistance threshold and the corona inception voltage threshold into the virtual physical model, and obtain a lossy detection model in combination with the improved neural network model; obtain the actual operation data and actual image data of the power fittings and input them into the lossy detection model, and output the lossy detection results.

[0086] It should be noted here that each module in this embodiment corresponds one-to-one to the method in Embodiment 1, and the specific implementation process is the same, which will not be repeated here.

[0087] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0088] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present invention.

Claims

1. A method for detecting loss of power metal, characterized in that: include: Collect image data and manufacturing data of the electrical fittings to be tested, and pre-process the acquired image data; Collect simulated operation data of the electric fittings to be tested, and construct tension-wind speed function, tension resistance-wind speed function and corona inception voltage-wind speed function respectively; and calculate the tension threshold, tension resistance threshold and corona inception voltage threshold of the electric fittings to be tested; An improved neural network model is constructed, the image data of the electrical fittings to be tested is input into the improved neural network model, and the corresponding lossy type is output; According to the manufacturing data, a virtual physical model is constructed, and the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function as well as the tension threshold, the tension resistance threshold and the corona inception voltage threshold are input into the virtual physical model, and the destructive detection model is obtained in combination with the improved neural network model; the actual operation data and the actual image data of the power fittings are obtained and input into the destructive detection model, and the destructive detection results are output.

2. A method for detecting loss of power metal as claimed in claim 1, characterized in that: The simulated operation data of the electric fittings to be tested include: simulated tensile force data, simulated tensile resistance data and simulated corona inception voltage data of the electric fittings; the manufacturing data include: painting data, cutting data, grinding data and polishing data.

3. A method for detecting loss of power metal as claimed in claim 1, characterized in that: The specific steps of constructing the tension-wind speed function and calculating the tension threshold of the electrical fittings to be tested are: obtaining simulated tension data and the wind speed corresponding to the simulated tension data, drawing a tension-wind speed curve, and calculating the functional relationship in the curve to construct the tension-wind speed function; taking the derivative of the tension-wind speed function, collecting all derivative values ​​when the tension is reduced, and calculating the average value of all derivative values ​​as the tension threshold.

4. A method for detecting loss of power metal as claimed in claim 1, characterized in that: The specific steps of constructing the tension resistance-wind speed function and calculating the tension resistance threshold of the electrical hardware to be tested are: obtaining simulated tension resistance data and the wind speed corresponding to the simulated tension resistance data, drawing a tension resistance-wind speed curve, and calculating the functional relationship in the curve to construct the tension resistance-wind speed function; taking the derivative of the tension resistance-wind speed function, collecting all derivative values ​​when the tension resistance is reduced, and calculating the average value of all derivative values ​​as the tension resistance threshold.

5. The method for detecting loss of electric power metal according to claim 1, characterized in that: The specific steps of constructing the corona inception voltage-wind speed function and calculating the corona inception voltage threshold of the electrical fitting to be tested are: obtaining simulated corona inception voltage data and the wind speed corresponding to the simulated corona inception voltage data, drawing a corona inception voltage-wind speed curve, and calculating the functional relationship in the curve to construct the corona inception voltage-wind speed function; taking the derivative of the corona inception voltage-wind speed function, collecting all derivative values ​​when the corona inception voltage is reduced, and calculating the average value of all derivative values ​​as the corona inception voltage threshold.

6. A method for detecting loss of power metal as claimed in claim 1, characterized in that: The construction of the improved neural network includes fusing features and calculating similarity. The specific process of fusing features is as follows: First, select 150×150×3 image data and input it into the deep residual network for feature extraction. Then select the feature maps of the last three residual structures as the first feature maps, which are , and ,right and After the maximum pooling operation, the three feature maps are superimposed to generate a new feature map as the second feature map. Finally, a convolution operation is performed to obtain the third feature map, which is used to calculate the similarity.

7. A method for detecting loss of power metal as claimed in claim 6, characterized in that: The similarity is calculated as follows: Similar ),in, and The result of paving each channel of the third feature map by fusion features of the image data according to the pixel points is: for The transpose of , Q represents the number of channels of the feature map.

8. An electric power metal damage detection system, characterized in that: include: The data acquisition and processing module is configured to: acquire image data and manufacturing data of the power fittings to be tested, and pre-process the acquired image data; The threshold calculation module is configured to: collect simulated operation data of the power fittings to be tested, respectively construct tension-wind speed function, tension resistance-wind speed function and corona initiation voltage-wind speed function; and calculate the tension threshold, tension resistance threshold and corona initiation voltage threshold of the power fittings to be tested; The neural network building module is configured to: build an improved neural network model, input the image data of the power fittings to be tested into the improved neural network model, and output corresponding lossy features; The lossy detection module is configured to: construct a virtual physical model according to the manufacturing data, input the tension-wind speed function, the tension resistance-wind speed function and the corona inception voltage-wind speed function as well as the tension threshold, the tension resistance threshold and the corona inception voltage threshold into the virtual physical model, and obtain a lossy detection model in combination with the improved neural network model; obtain the actual operation data and actual image data of the power fittings and input them into the lossy detection model, and output the lossy detection results.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for detecting damage to electric metal parts as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for detecting damage to electric power metal as described in any one of claims 1 to 7 are implemented.

Citation Information

Patent Citations

  • Hardware defect detection method

    CN118348126A

  • Image recognition method and system based on transfer learning and ResNet50 neural network

    CN116824239A

  • Electric power fitting surface intelligent detection method and detection system

    CN117607019A

  • Sample expansion method and system for electric power operation distance measurement task

    CN117975188A

  • Neural network-based image recognition method and apparatus, electronic device, and medium

    WO2022213465A1