A pressure-free device detection method, system and device

Abstract

The present invention provides a method, system and device for detecting a pressure-free device, which relates to the field of intelligent detection. The method comprises: regularly collecting images of the pressure-free device at the cable to obtain a device image sequence; performing key frame extraction to obtain a device key image sequence; using a convolutional neural network to extract fault-related features to obtain a fault-related feature distribution sequence; predicting the occurrence of faults based on the fault-related feature distribution sequence to output a historical predicted fault probability sequence; performing fault analysis in future time zones based on the historical predicted fault probability sequence to output a predicted fault probability interval sequence as a fault detection result. The method aims to solve the technical problems that traditional pressure-free device detection methods rely heavily on manual inspections, have low detection efficiency, poor accuracy, and are unable to detect potential defects in a timely manner. The method can effectively improve the automation and intelligence of pressure-free device detection, and significantly enhance the accuracy, real-time performance, and efficiency of pressure-free device fault detection.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection, and in particular to a pressure-free device detection method, system and device. Background Art

[0002] Traditional methods for detecting pressure-free devices mainly rely on manual inspections or regular checks to detect potential equipment failures. Although manual inspections are effective in some cases, they have obvious limitations. First, manual inspections usually rely on the experience and judgment of operators and are easily affected by fatigue, negligence, and subjective factors, resulting in inaccurate and missed detection results. Secondly, due to the long inspection cycle, many potential faults and defects may quietly occur between inspection cycles and cannot be discovered and handled in time, thereby increasing the risk of equipment failure. Thirdly, manual inspection efficiency is low, especially when the pressure-free device is in a high-risk working environment or large-scale application. The frequency and coverage of manual inspections often cannot meet the needs of real-time monitoring and rapid response.

[0003] Therefore, there is an urgent need for a new detection method that can monitor the operating status of the pressure-free device in real time and automatically, and detect potential faults in a timely manner to ensure the stable operation of the equipment, improve the efficiency and accuracy of detection, and reduce dependence on manual intervention. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, system, and device for detecting pressure-free devices, which are designed to address the technical problems of traditional pressure-free device detection methods that rely heavily on manual inspections, resulting in low detection efficiency, poor accuracy, and inability to detect potential defects in a timely manner. The methods include:

[0005] In the first aspect, the present invention provides a method for detecting a voltage-free device, comprising: regularly capturing images of the voltage-free device at a preset time interval to obtain a device image sequence; performing key frame extraction on the device image sequence to obtain a device key image sequence; using a convolutional neural network to extract fault-related features of the device key image sequence to obtain a fault-related feature distribution sequence; predicting the occurrence of a voltage-free device fault based on the fault-related feature distribution sequence, and outputting a historical predicted fault probability sequence; performing fault analysis in a future time zone based on the historical predicted fault probability sequence, and outputting a predicted fault probability interval sequence as a fault detection result of the voltage-free device.

[0006] Preferably, the pressure-free device detection method also includes: configuring key parts of the pressure-free device, wherein the key parts include but are not limited to terminal connection parts, locking components, positioning cylinders and drive components; regularly collecting images of multiple key parts of the pressure-free device at the cable at preset time intervals, and fusing them into a complete image according to the position coordinates of the parts as a device image to obtain a device image sequence.

[0007] Preferably, the pressure-free device detection method also includes: selecting the device image at the earliest monitoring time point in the device image sequence as the first image, setting the first image as the first key image, and setting the adjacent image of the first image as the second image; performing a similarity comparison on the first image and the second image according to a preset comparison strategy, and outputting the first image similarity; if the first image similarity is greater than or equal to a preset similarity threshold, discarding the second image, and performing iterative similarity comparison of subsequent images based on the first image; if the first image similarity is less than the preset similarity threshold, setting the second image as the second key image, and performing iterative similarity comparison of subsequent images based on the second image, until all images in the device image sequence are compared, and multiple key images are obtained, which are combined to construct a device key image sequence.

[0008] Preferably, the pressure-free device detection method also includes: selecting a first similarity comparison algorithm according to pixel similarity measurement; selecting a second similarity comparison algorithm according to feature similarity measurement; selecting a third similarity comparison algorithm according to structural similarity measurement; and obtaining a preset comparison strategy based on the combination of the first similarity comparison algorithm, the second similarity comparison algorithm and the third similarity comparison algorithm.

[0009] Preferably, the pressure-free device detection method also includes: configuring fault-related features, wherein the fault-related features include at least displacement features, deformation features and surface corrosion; using the fault-related features as constraints, querying historical detection records of similar pressure-free devices, collecting sample device image sets and sample fault-related feature distribution sets, wherein the fault-related feature distribution includes fault-related features of multiple key parts, and the key parts are marked with position coordinates; using the sample device image set and the sample fault-related feature distribution set, supervised training and testing of the convolutional neural network until the network converges to obtain a feature extractor; using the feature extractor, sequentially extracting fault-related features from multiple device key images in the device key image sequence, and outputting a fault-related feature distribution sequence.

[0010] Preferably, the pressure-free device detection method also includes: collecting a sample fault-related feature distribution set based on historical detection records of similar pressure-free devices, and counting the failure frequency of the pressure-free device under different sample fault-related feature distributions as the sample failure probability to obtain a sample failure probability set; using the sample fault-related feature distribution set and the sample failure probability set to train the generator and discriminator of the adversarial neural network until the generation loss function and the discriminant loss function converge to obtain a fault prediction model; using the fault prediction model to predict the occurrence of pressure-free device failures in sequence according to the fault-related feature distribution sequence, and output a historical predicted fault probability sequence.

[0011] Preferably, the pressure-free device detection method also includes: in a two-dimensional coordinate system, fitting the historical predicted fault probability sequence according to the chronological order of monitoring time to generate a historical predicted fault probability curve; configuring a future time zone, wherein the future time zone includes several consecutive monitoring time nodes; in a two-dimensional coordinate system, extending according to the historical predicted fault probability curve, predicting and obtaining several predicted fault probabilities of several monitoring time nodes, and constructing a predicted fault probability sequence; performing fluctuation error analysis on the historical predicted fault probability sequence, and outputting a fluctuation error ratio; expanding the predicted fault probability sequence according to the fluctuation error ratio, and outputting a predicted fault probability interval sequence.

[0012] Preferably, the pressure-free device detection method also includes: performing mean calculation on multiple historical predicted failure probabilities in the historical predicted failure probability sequence to generate a mean of historical predicted failure probabilities; taking the mean of historical predicted failure probability as a benchmark, performing deviation amplitude analysis on the multiple historical predicted failure probabilities respectively, and performing mean calculation on multiple deviation amplitudes to obtain the fluctuation error ratio.

[0013] In the second aspect, the present invention also provides a pressure-free device detection system for executing a pressure-free device detection method as described in the first aspect, including: an image acquisition module for regularly acquiring images of the pressure-free device at the cable at preset time intervals to obtain a device image sequence; a key frame extraction module for performing key frame extraction on the device image sequence to obtain a device key image sequence; a fault feature extraction module for performing fault-related feature extraction on the device key image sequence using a convolutional neural network to obtain a fault-related feature distribution sequence; a fault occurrence prediction module for predicting the occurrence of a pressure-free device fault based on the fault-related feature distribution sequence and outputting a historical predicted fault probability sequence; a detection result acquisition module for performing fault analysis in a future time zone based on the historical predicted fault probability sequence and outputting a predicted fault probability interval sequence as a fault detection result of the pressure-free device.

[0014] In a third aspect, the present invention further provides a pressure-free device detection device, which can be executed by a pressure-free device detection method described in any one of the first aspects above.

[0015] The embodiments of the present invention include the following advantages:

[0016] By regularly capturing images of the pressure-free device at the cable at preset time intervals, a device image sequence is obtained; then, key frames are extracted from the device image sequence to obtain a device key image sequence; then, a convolutional neural network is used to extract fault-related features from the device key image sequence to obtain a fault-related feature distribution sequence; further, the occurrence of a pressure-free device fault is predicted based on the fault-related feature distribution sequence, and a historical predicted fault probability sequence is output; finally, based on the historical predicted fault probability sequence, a fault analysis is performed in the future time zone, and a predicted fault probability interval sequence is output as the fault detection result of the pressure-free device. In other words, by extracting key features through image recognition and analyzing fault-related features, the probability of failure of the device can be predicted in real time, and the fault probability interval can be given through historical data analysis, which can effectively improve the automation and intelligence of the pressure-free device detection, thereby significantly improving the accuracy, real-time performance and efficiency of the pressure-free device fault detection, and providing reliable protection for the stable operation of electrical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the steps of a pressure-free device detection method of the present invention;

[0018] Figure 2 This is a structural schematic diagram of a pressure-free device detection system of the present invention.

[0019] Description of reference numerals:

[0020] Image acquisition module 11, key frame extraction module 12, fault feature extraction module 13, fault occurrence prediction module 14, detection result acquisition module 15. DETAILED DESCRIPTION

[0021] The present invention solves the technical problems of traditional pressure-free device detection methods, which rely heavily on manual inspections and suffer from low detection efficiency, poor accuracy, and inability to detect potential defects in a timely manner, by providing a method, system, and device for detecting pressure-free devices. By extracting key features through image recognition and analyzing fault-related features, the probability of device failure can be predicted in real time, and the fault probability interval can be given through historical data analysis. This can effectively improve the automation and intelligence of pressure-free device detection, thereby significantly improving the accuracy, real-time performance, and efficiency of pressure-free device fault detection, providing reliable protection for the stable operation of electrical equipment.

[0022] Below, the technical solutions of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should also be noted that, for the convenience of description, only the parts related to the present invention, rather than all, are shown in the accompanying drawings.

[0023] For example 1, please refer to the attached Figure 1 The present invention provides a method for detecting a pressure-free device, which specifically includes the following steps:

[0024] S1: regularly collecting images of the pressure-free device at the cable according to a preset time interval to obtain a device image sequence.

[0025] Furthermore, step S1 of the present invention further includes:

[0026] S11: configuring key parts of the pressure-free device, wherein the key parts include but are not limited to the terminal connection part, the locking assembly, the positioning cylinder and the driving assembly; S12: regularly collecting images of multiple key parts of the pressure-free device at the cable at preset time intervals, and fusing them into a complete image according to the position coordinates of the parts as the device image, thereby obtaining a device image sequence.

[0027] Specifically, in this application, the pressure-free device refers to a pressure-free terminal block, which is a terminal with a simple design that can complete wire connection without the use of traditional crimping tools. It is widely used in the electrical field and is suitable for occasions where electrical connections need to be quickly installed, disassembled, maintained or replaced.

[0028] First of all, it is necessary to determine the key parts of the pressure-free device. These parts are the core parts that determine the normal operation and failure of the device. The key parts include but are not limited to the terminal connection part, locking assembly, positioning cylinder and drive assembly. Among them, the terminal connection part is the part where the pressure-free device is connected to the cable, and it is also the key interface of the entire electrical connection. The state of the terminal connection part directly affects the transmission stability of current and signal. Any looseness, poor contact or corrosion may cause failure; the locking assembly is used to fix the various parts of the pressure-free device to ensure that the components will not loosen or shift during operation. The stability of the locking assembly is the key to whether the pressure-free device can operate stably for a long time. If the locking assembly fails, it may cause the entire pressure-free device to fail; the positioning cylinder is used to position the various parts in the device to the specified position to ensure that the device components are precisely aligned and effectively matched. The accuracy of the positioning cylinder is crucial to the structural stability and working performance of the pressure-free device; the drive assembly is responsible for providing power so that the relevant components in the pressure-free device can correctly perform actions such as crimping or unlocking. Failure of the drive assembly may cause the device to fail to operate correctly, so its monitoring is crucial.

[0029] Next, based on the equipment's operating conditions and monitoring requirements, a fixed time interval (e.g., 24 hours, i.e., daily image acquisition) is set to capture images of multiple key locations on the pressure-free device. This ensures the monitoring system can continuously and stably monitor the device's status and promptly detect potential faults. Images of multiple key locations on the cable's pressure-free device are then captured regularly at preset time intervals (e.g., 24 hours). During image acquisition, each location's image has specific location coordinates, meaning each location's image is associated with the device's actual physical location. To obtain a complete image of the device, each location's image is fused based on its actual location coordinates. The fusion process combines image data from different locations and locations into a complete image, creating a comprehensive image of the device's status. This fused image retains information about all key locations, enabling subsequent analysis systems to comprehensively evaluate the device. These device images are then organized into a continuous image sequence, chronologically, to create a device image sequence. Each image represents the device's status at a different point in time. The continuity of the image sequence provides rich data support for subsequent fault detection and prediction.

[0030] S2: extracting key frames from the device image sequence to obtain a device key image sequence.

[0031] Furthermore, step S2 of the present invention further includes:

[0032] S21: Selecting the device image at the earliest monitoring time point in the device image sequence as the first image, setting the first image as the first key image, and setting the image adjacent to the first image as the second image.

[0033] Specifically, in the device image sequence, the device image at the earliest monitoring time point is selected as the first image. This image will show the initial state of the pressure-free device and is the benchmark for subsequent analysis; then, the first image is set as the first key image, which represents the state of the device at the initial stage and is the starting point for all subsequent image comparisons and fault detection; then the adjacent image of the first image is used as the second image.

[0034] S22: performing a similarity comparison on the first image and the second image according to a preset comparison strategy, and outputting a first image similarity.

[0035] Furthermore, step S22 of the present invention further includes:

[0036] S221: Select a first similarity comparison algorithm according to pixel similarity measurement; S222: Select a second similarity comparison algorithm according to feature similarity measurement; S223: Select a third similarity comparison algorithm according to structural similarity measurement; S224: Obtain a preset comparison strategy based on the combination of the first similarity comparison algorithm, the second similarity comparison algorithm and the third similarity comparison algorithm.

[0037] Specifically, first, the first similarity comparison algorithm is selected according to the pixel similarity metric. The pixel similarity metric refers to directly comparing the differences between two images at the pixel level. The pixel similarity metric is usually used to detect intuitive differences in images and compare the color, brightness and other features of each pixel. For example, the mean square error calculates the difference in pixel values ​​between two images. The smaller the value, the higher the image similarity. The pixel comparison algorithm is suitable for detecting simple and direct changes between images, such as obvious damage, dirt or morphological changes on the surface of the device. Next, the second similarity comparison algorithm is selected according to the feature similarity measure. The feature similarity measure refers to extracting representative features (such as edges, corners, textures, etc.) in the image and comparing the similarity between these features. Feature comparison is usually used to compare global features in the image to avoid interference caused by external factors such as lighting and perspective changes. For example, algorithms such as SIFT (Scale-Invariant Feature Transform) and SURF (Speeded Robust Features) extract stable feature points from the image for matching, and are suitable for robust detection of image rotation, scaling and lighting changes. Feature comparison is suitable for identifying high-level features such as the shape, texture, and structure of the device. For example, when the structure of the pressure-free device is deformed or the positioning components become loose, the feature comparison algorithm can capture these changes more sensitively.

[0038] On the other hand, the third similarity comparison algorithm is selected according to the structural similarity metric. The structural similarity metric focuses on the comparison of the overall structure or geometric shape of the image. It not only considers pixels and features, but also considers factors such as the spatial layout and structural relationship of the image. The structural similarity metric method is particularly suitable for comparing the shape and structural information of objects when judging changes in image content, such as Fourier transform and Hough transform of the image. Hough transform is used to detect changes in geometric shapes such as straight lines and circles in the image, and is suitable for the detection of structural features. Structural comparison is suitable for detecting changes in the overall layout and structure of pressure-free devices, such as displacement of positioning cylinders and drive components, damage to locking components, etc. These problems are often early signals of equipment failure.

[0039] The first, second, and third similarity comparison algorithms are then combined using a weighted approach to create a pre-defined comparison strategy. This assigns different weights to each algorithm based on actual needs. For example, for scenarios that are very sensitive to structural changes, structural comparison may be given a higher weight; while detecting minor changes on the device surface may rely more heavily on pixel or feature comparison. By combining the comparison results of the three algorithms, the similarity between images can be comprehensively determined, leading to a more comprehensive and accurate assessment of the status of the pressure-free device.

[0040] Next, a similarity analysis is performed on the first and second images according to the pre-configured comparison strategy (including pixel similarity, feature similarity, and structural similarity). The similarity between the images is calculated, reflecting the degree of visual and structural similarity between the images. The comparison algorithm outputs the first image similarity, which is a similarity score between the first and second images. The similarity value is usually expressed as a percentage, numeric value, or distance metric. A high similarity indicates that the changes between the two images are small; a low similarity indicates that the two images are significantly different, possibly indicating a change in device status.

[0041] S23: If the similarity of the first image is greater than or equal to a preset similarity threshold, the second image is discarded, and the first image is used as a benchmark to perform iterative similarity comparison of subsequent images; S24: If the similarity of the first image is less than the preset similarity threshold, the second image is set as a second key image, and the second image is used as a benchmark to perform iterative similarity comparison of subsequent images until all images in the device image sequence are compared, and multiple key images are obtained, which are combined to construct a device key image sequence.

[0042] Specifically, a preset similarity threshold is configured, which can be set according to the actual scenario, such as 80%. If the similarity of the first image is greater than or equal to the preset similarity threshold, it means that the second image has little visual and structural changes with the first image, and the device status is stable or there is no obvious abnormality. In this case, the second image is discarded and the first image is used as the benchmark for subsequent image comparisons.

[0043] If the similarity between the first image and the second image is less than the preset similarity threshold, it means that there is a large difference between the two images, which may be due to a change in the device status. In this case, the second image is regarded as a new key image and used as a benchmark for subsequent image comparisons. If images with lower similarity continue to be found, the key images are updated and compared until all images in the entire device image sequence are processed. Finally, after a series of iterations and comparisons, a set of key images will be obtained and combined into a complete device key image sequence, which can reflect the state changes that occur in the pressure-free device during operation. By extracting key images, the state changes of the device at different time points can be accurately captured, thereby providing reliable image data support for subsequent fault detection and effectively improving the accuracy and efficiency of detection.

[0044] S3: Using a convolutional neural network, extract fault-related features from the key image sequence of the device to obtain a fault-related feature distribution sequence.

[0045] Furthermore, step S3 of the present invention further includes:

[0046] S31: Configure fault association features, wherein the fault association features include at least displacement features, deformation features and surface corrosion; S32: Using the fault association features as constraints, query historical inspection records of similar pressure-free devices, collect sample device image sets and sample fault association feature distribution sets, wherein the fault association feature distribution includes fault association features of multiple key parts, and the key parts are marked with position coordinates; S33: Use the sample device image set and the sample fault association feature distribution set to supervise training and testing of the convolutional neural network until the network converges to obtain a feature extractor; S34: Use the feature extractor to extract fault association features of multiple device key images in the device key image sequence in sequence, and output a fault association feature distribution sequence.

[0047] Specifically, fault-related features are configured. Fault-related features refer to physical quantities or state parameters that are closely related to the failure or performance changes of the pressure-free device. These features can reflect the changes caused by different factors (such as mechanical pressure, environmental influence, usage time, etc.) during the use of the device. Among them, the fault-related features include at least displacement features, deformation features and surface corrosion. The displacement feature refers to the movement of key components in the pressure-free device (such as positioning cylinder, crimping sleeve, etc.) relative to the original position. The size of the displacement is related to whether the device is subjected to excessive pressure, collision or vibration. A larger displacement usually indicates that the device may be loose, fall off or damaged. The deformation feature focuses on the shape change of the components of the pressure-free device, such as the deformation of the crimping plate, the bending of the clamping ring, etc. The equipment may be deformed during operation due to external factors such as pressure and temperature, which is usually a precursor to failure. The surface corrosion feature focuses on the degree of corrosion of the surface material of the device, especially metal parts. Corrosion is usually caused by environmental factors (such as moisture, chemicals, etc.), which will directly affect the electrical connection performance of the device and even cause short circuit or equipment failure. The purpose of configuring these fault-related features is to monitor and analyze the status changes of the pressure-free device from multiple dimensions. By extracting these features, a more comprehensive assessment of the health status of the device can be made, and possible faults can be predicted.

[0048] Next, using the fault-related features as constraints, historical inspection records of similar pressure-free devices are queried. These records contain past inspection data, including image data and fault feature data, which can be used as a historical reference for equipment failure analysis. By querying historical records, the failure modes and patterns of similar devices in the past can be discovered and learned, providing a basis for predicting failures of new equipment. Then, based on the historical inspection records, a sample device image set and a sample fault-related feature distribution set are collected. The sample device image set refers to a collected image data set containing different device states (normal, faulty, etc.). By comparing these images, the image features of the device in different fault states can be identified. For example, there may be structural changes, corrosion marks, or looseness between a normal device and a faulty device. The sample fault-related feature distribution set refers to the feature data associated with the device failure collected by analyzing historical records. These data include information such as the displacement, deformation, and corrosion degree of different components. The feature distribution set can help the system understand the different feature distributions exhibited by different parts during the fault process. Among them, the fault correlation feature distribution includes the fault correlation features of multiple key parts. The key parts are marked with position coordinates, that is, the spatial position of the component in the device. This helps to associate the fault features with specific parts, thereby accurately locating the location where the fault occurs during the detection process.

[0049] Then, the sample device image set and the sample fault-related feature distribution set are used as sample data and divided into a sample training set and a sample test set according to a certain ratio, wherein the training set usually accounts for 75% to 90% and the test set usually accounts for 10% to 25%. Convolutional neural networks (CNNs) are a type of deep learning algorithm widely used in fields such as image processing and computer vision. Their core advantage lies in their ability to automatically extract features from raw data and are more efficient than traditional machine learning methods when processing image data. A feature extractor is then constructed based on the convolutional neural network. The feature extractor includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer is used to pass image data to subsequent convolutional layers for feature extraction. The convolutional layer is the core part of the feature extractor, responsible for extracting local features from the input image. Each convolutional layer includes multiple convolution kernels, which slide across the image and extract different features (such as edges, textures, colors, etc.) through convolution operations (i.e., matrix multiplication and addition). The pooling layer is used to downsample the feature map, reducing its size, thereby reducing the amount of computation, avoiding overfitting, and retaining important information in the image. The fully connected layer is used to integrate the previously extracted features. The neurons in this layer are connected to all neurons in the previous layer and are used to integrate high-level information of the image. The output layer provides the final prediction result.

[0050] The sample training set and sample test set are then used to supervise training and test the feature extractor, respectively. During the training process, images in the training set are input into the feature extractor, and each input image is processed through convolution layers, activation functions, pooling layers, etc. to extract features from the image. Then, through multiple layers of convolution and pooling operations, the network will automatically learn low-level and high-level features in the image. For example, in an image of electrical equipment, the network may learn corrosion on the surface of the equipment, loose wiring parts, or other fault features. The output of the feature extractor is then compared with the actual label (fault type, fault location, etc.), and the error between the predicted result and the true label is calculated. The difference is usually measured using a cross-entropy loss function or a mean squared error loss function. Further, based on the calculated loss value, the weights in the network are adjusted using a backpropagation algorithm to reduce the error. Through multiple iterative training, the weights of the network will be gradually optimized, so that the feature extractor can better extract features related to the fault. The training process usually sets a stopping condition, such as loss value convergence or reaching a preset training round. Once the training is completed, the feature extractor will have the ability to extract relevant features from the image. During the testing process, the images in the test set are input into the trained feature extractor and the same feature extraction process is performed. Next, the network output is compared with the actual labels in the test set to check whether the extracted features can accurately reflect the fault information. Then, based on the difference between the network output results and the actual labels, the performance of the feature extractor is evaluated using evaluation indicators (such as accuracy). If the performance of the feature extractor on the test set is not ideal, it can be tuned until the expected output accuracy is achieved, resulting in a trained feature extractor.

[0051] Finally, the feature extractor is used to extract fault-related features from multiple key images of the device in the key image sequence in sequence. The feature extractor performs multi-level feature extraction tasks on each key image, learns features at different levels (such as edges, textures, shapes, colors, etc.), and finally obtains feature data related to equipment failure. The output fault-related features include the above-mentioned numerical features such as displacement, deformation, corrosion degree, or other more complex fault indication signals; the extracted fault-related features will form a fault-related feature distribution sequence in chronological order or image order, providing data support for subsequent fault prediction and analysis.

[0052] S4: Predicting the occurrence of a pressure-free device failure based on the fault-related feature distribution sequence, and outputting a historical predicted failure probability sequence.

[0053] Furthermore, step S4 of the present invention further includes:

[0054] S41: Based on the historical detection records of similar pressure-free devices, a sample fault-related feature distribution set is collected, and the failure frequency of the pressure-free device under different sample fault-related feature distributions is counted as the sample failure probability to obtain a sample failure probability set; S42: Using the sample fault-related feature distribution set and the sample failure probability set, the generator and discriminator of the adversarial neural network are trained until the generation loss function and the discriminant loss function converge to obtain a fault prediction model; S43: Using the fault prediction model, the pressure-free device failure is predicted in sequence according to the fault-related feature distribution sequence, and a historical predicted fault probability sequence is output.

[0055] Specifically, key feature data related to equipment failures are collected from historical inspection records of similar pressure-free devices. These data can come from historical maintenance records, sensor data, regular inspections or image detection results, etc., to obtain a sample fault-related feature distribution set; then, the frequency of failures of the pressure-free device under different sample fault-related feature distributions is counted. For example, if the equipment frequently fails under certain specific fault-related features (such as displacement, corrosion, deformation, etc.), then there may be a strong correlation between the appearance of these features and the occurrence of failures. The statistical results form a sample failure probability set, that is, the failure probability of each sample. These failure probabilities are obtained by analyzing the failure data in historical records and reflect the possibility of equipment failure under specific conditions.

[0056] Next, the sample fault association feature distribution set and the sample fault probability set are used as training data to train the generator and discriminator of the adversarial neural network. The goal of the generator is to generate realistic sample fault probabilities from the input fault association feature distribution, close to the actual fault distribution; the goal of the discriminator is to determine whether a sample comes from a real fault dataset (historical records) or pseudo-data generated by the generator. The training process is carried out through adversarial training. The generator and discriminator compete with each other, and the generator is continuously optimized to generate more realistic fault probability distributions, while the discriminator is continuously optimized to improve its ability to distinguish between real and fake data. During the training process, the generator and discriminator are optimized using a loss function. The generation loss function is used to measure the difference between the fault probability samples generated by the generator and the real fault data; the discriminant loss function is used to measure the accuracy of the discriminator in distinguishing between real samples and generated samples. Training will continue until both the generation loss function and the discriminant loss function converge, that is, the generator can produce samples close to real data and the discriminator can correctly distinguish between real and generated samples. At this time, training is terminated to obtain a trained fault prediction model, which can predict the failure probability of the device based on the fault association feature distribution.

[0057] Then, a feature distribution sequence formed by fault-related features (such as displacement, deformation, corrosion, etc.) extracted from the key image sequence of the device is used as input data, and these feature distribution sequences are input into the trained fault prediction model. The model will predict the failure probability of the pressure-free device based on these input features. Each time point or each key image corresponds to a failure probability, indicating the possibility of the device failing in the current state. The model outputs a historical predicted failure probability sequence, that is, the failure probability corresponding to each time point or each key image. This sequence can provide the probability trend of device failure in the future.

[0058] S5: Based on the historical predicted fault probability sequence, perform fault analysis in the future time zone and output a predicted fault probability interval sequence as a fault detection result of the pressure-free device.

[0059] Furthermore, step S5 of the present invention further includes:

[0060] S51: In a two-dimensional coordinate system, the historical predicted fault probability sequence is fitted according to the chronological order of monitoring time to generate a historical predicted fault probability curve; S52: A future time zone is configured, wherein the future time zone includes a number of consecutive monitoring time nodes; S53: In a two-dimensional coordinate system, the historical predicted fault probability curve is extended to predict and obtain a number of predicted fault probabilities of a number of monitoring time nodes, and a predicted fault probability sequence is constructed.

[0061] Specifically, in a two-dimensional coordinate system, with time as the horizontal axis and failure probability as the vertical axis, the data points in the historical predicted failure probability series are connected to form a curve. This means that these data points are fitted (for example, using polynomial fitting, spline interpolation, or other curve fitting methods) to obtain a smooth historical predicted failure probability curve. The fitted curve can reflect the trend of equipment failure probability over time and help observe the fluctuation pattern of failure probability and changes over long time periods. Next, a future time zone is configured, where the future time zone includes several consecutive monitoring time nodes. These monitoring time nodes are the time range used for prediction and early warning after the historical data, for example, the next few days or weeks. Further, in the two-dimensional coordinate system, the historical predicted failure probability curve is extended, that is, the curve is extended to the monitoring time nodes in the future time zone in the two-dimensional coordinate system. This step is essentially to use the trend of historical data for extrapolation, that is, to predict the failure probability of future time nodes based on historical data, and to predict the failure probability of several time nodes in the future through the extended curve or fitting model. For example, if the historical data shows an upward trend, the extended curve may predict an increase in the failure probability in the future period of time. The predicted failure probability of each monitoring time node represents the possibility of the equipment failing at that time point; finally, these predicted failure probabilities are arranged in chronological order to construct a predicted failure probability sequence. This sequence can help predict changes in the failure probability of the equipment in future time zones and reflect the changing trend of the equipment's failure risk over time.

[0062] S54: performing a fluctuation error analysis on the historical predicted fault probability sequence, and outputting a fluctuation error ratio.

[0063] Furthermore, step S54 of the present invention further includes:

[0064] S541: Calculate the mean of multiple historical predicted failure probabilities in the historical predicted failure probability sequence to generate the mean of historical predicted failure probabilities; S542: Based on the mean of historical predicted failure probability, perform deviation amplitude analysis on the multiple historical predicted failure probabilities respectively, and calculate the mean of multiple deviation amplitudes to obtain the fluctuation error ratio.

[0065] Specifically, the method first calculates the mean of multiple historical predicted failure probability values ​​in the sequence to obtain a historical predicted failure probability mean, which represents the average level of failure probability of the equipment during the past monitoring period. Next, using the historical predicted failure probability mean as a benchmark, the method performs a deviation analysis on each of the multiple historical predicted failure probabilities. The deviation amplitude represents the degree of difference between the failure probability at a certain point in time and the mean. That is, the deviation amplitude is the ratio of the absolute value of the difference between the historical predicted failure probability and the historical predicted failure probability mean to the historical predicted failure probability mean, resulting in multiple deviation amplitudes. Finally, the method calculates the mean of the multiple deviation amplitudes to obtain the fluctuation error ratio, which reflects the degree of fluctuation in the historical predicted failure probability. If the fluctuation error ratio is small, it indicates that the failure probability is relatively stable and the prediction accuracy is high. If the fluctuation error ratio is large, it indicates that the failure probability fluctuates greatly, and there may be high uncertainty or prediction error.

[0066] S55: Expanding the predicted fault probability sequence according to the fluctuation error ratio, and outputting a predicted fault probability interval sequence.

[0067] Specifically, the multiple predicted failure probabilities in the predicted failure probability sequence are expanded based on the fluctuation error ratio. Each predicted failure probability value is expanded into an interval, with the lower limit being the product of 1 minus the fluctuation error ratio and the predicted failure probability, and the upper limit being the product of 1 plus the fluctuation error ratio and the predicted failure probability. Each predicted failure probability value is expanded upward and downward to generate a failure probability interval (i.e., upper and lower limits) based on the fluctuation error ratio, reflecting the predicted range of the failure probability. Finally, these intervals are arranged in chronological order to form a predicted failure probability interval sequence, providing a reference for subsequent fault warning and equipment maintenance decisions.

[0068] In summary, the pressure-free device detection method provided by the present invention has the following technical effects:

[0069] By regularly capturing images of the pressure-free device at the cable at preset time intervals, a device image sequence is obtained; then, key frames are extracted from the device image sequence to obtain a device key image sequence; then, a convolutional neural network is used to extract fault-related features from the device key image sequence to obtain a fault-related feature distribution sequence; further, the occurrence of a pressure-free device fault is predicted based on the fault-related feature distribution sequence, and a historical predicted fault probability sequence is output; finally, based on the historical predicted fault probability sequence, a fault analysis is performed in the future time zone, and a predicted fault probability interval sequence is output as the fault detection result of the pressure-free device. In other words, by extracting key features through image recognition and analyzing fault-related features, the probability of failure of the device can be predicted in real time, and the fault probability interval can be given through historical data analysis, which can effectively improve the automation and intelligence of the pressure-free device detection, thereby significantly improving the accuracy, real-time performance and efficiency of the pressure-free device fault detection, and providing reliable protection for the stable operation of electrical equipment.

[0070] Example 2: Based on the same inventive concept as the pressure-free device detection method in the above embodiment, the present invention also provides a pressure-free device detection system, please refer to the attached Figure 2 , including: an image acquisition module 11, used to regularly capture images of the pressure-free device at the cable at preset time intervals to obtain a device image sequence; a key frame extraction module 12, used to extract key frames from the device image sequence to obtain a device key image sequence; a fault feature extraction module 13, used to use a convolutional neural network to extract fault-related features from the device key image sequence to obtain a fault-related feature distribution sequence; a fault occurrence prediction module 14, used to predict the occurrence of a fault of the pressure-free device based on the fault-related feature distribution sequence, and output a historical predicted fault probability sequence; a detection result acquisition module 15, used to perform fault analysis in a future time zone based on the historical predicted fault probability sequence, and output a predicted fault probability interval sequence as a fault detection result of the pressure-free device.

[0071] Furthermore, the pressure-free device detection system is also used to: configure the key parts of the pressure-free device, wherein the key parts include but are not limited to the terminal connection part, the locking assembly, the positioning cylinder and the driving assembly; regularly collect images of multiple key parts of the pressure-free device at the cable at preset time intervals, and fuse them into a complete image according to the part position coordinates as the device image, to obtain a device image sequence.

[0072] Furthermore, the pressure-free device detection system is also used to: select the device image at the earliest monitoring time point in the device image sequence as the first image, set the first image as the first key image, and use the adjacent image of the first image as the second image; perform a similarity comparison on the first image and the second image according to a preset comparison strategy, and output the first image similarity; if the first image similarity is greater than or equal to a preset similarity threshold, discard the second image, and use the first image as a benchmark to perform iterative similarity comparison of subsequent images; if the first image similarity is less than the preset similarity threshold, set the second image as the second key image, and use the second image as a benchmark to perform iterative similarity comparison of subsequent images, until all images in the device image sequence are compared, and multiple key images are obtained, which are combined to construct a device key image sequence.

[0073] Furthermore, the pressure-free device detection system is also used to: select a first similarity comparison algorithm according to pixel similarity measurement; select a second similarity comparison algorithm according to feature similarity measurement; select a third similarity comparison algorithm according to structural similarity measurement; and obtain a preset comparison strategy based on the combination of the first similarity comparison algorithm, the second similarity comparison algorithm and the third similarity comparison algorithm.

[0074] Furthermore, the pressure-free device detection system is also used to: configure fault-related features, wherein the fault-related features include at least displacement features, deformation features and surface corrosion; using the fault-related features as constraints, query historical detection records of similar pressure-free devices, collect sample device image sets and sample fault-related feature distribution sets, wherein the fault-related feature distribution includes fault-related features of multiple key parts, and the key parts are marked with position coordinates; use the sample device image set and the sample fault-related feature distribution set to supervise training and testing of the convolutional neural network until the network converges to obtain a feature extractor; use the feature extractor to extract fault-related features of multiple device key images in the device key image sequence in sequence, and output a fault-related feature distribution sequence.

[0075] Furthermore, the pressure-free device detection system is also used to: collect sample fault-related feature distribution sets based on historical detection records of similar pressure-free devices, and count the failure frequencies of the pressure-free devices under different sample fault-related feature distributions as sample failure probabilities to obtain sample failure probability sets; use the sample fault-related feature distribution sets and sample failure probability sets to train the generator and discriminator of the adversarial neural network until both the generation loss function and the discriminant loss function converge to obtain a fault prediction model; use the fault prediction model to predict the occurrence of pressure-free device failures in sequence according to the fault-related feature distribution sequence, and output a historical predicted fault probability sequence.

[0076] Furthermore, the pressure-free device detection system is also used to: fit the historical predicted fault probability sequence in the two-dimensional coordinate system according to the chronological order of monitoring time to generate a historical predicted fault probability curve; configure a future time zone, wherein the future time zone includes several consecutive monitoring time nodes; within the two-dimensional coordinate system, extend according to the historical predicted fault probability curve, predict and obtain several predicted fault probabilities of several monitoring time nodes, and construct a predicted fault probability sequence; perform fluctuation error analysis on the historical predicted fault probability sequence, and output a fluctuation error ratio; expand the predicted fault probability sequence according to the fluctuation error ratio, and output a predicted fault probability interval sequence.

[0077] Furthermore, the pressure-free device detection system is also used to: perform mean calculation on multiple historical predicted failure probabilities in the historical predicted failure probability sequence to generate a mean of historical predicted failure probabilities; based on the mean of historical predicted failure probability, perform deviation amplitude analysis on the multiple historical predicted failure probabilities respectively, and perform mean calculation on multiple deviation amplitudes to obtain the fluctuation error ratio.

[0078] In the third embodiment, based on the same inventive concept as the pressure-free device detection method in the aforementioned embodiment, the present invention further provides a pressure-free device detection device, which can be executed by any one of the pressure-free device detection methods described in the first embodiment.

[0079] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0080] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is intended to include these modifications and variations.

Claims

1. A method for detecting a pressure-free device, characterized in that: Methods include: regularly collecting images of the pressure-free device at the cable at preset time intervals to obtain a sequence of device images; Extracting key frames from the device image sequence to obtain a device key image sequence; Using a convolutional neural network, fault-related features are extracted from a key image sequence of the device to obtain a fault-related feature distribution sequence; Predicting the occurrence of a pressure-free device failure based on the fault-related feature distribution sequence, and outputting a historical predicted failure probability sequence; Based on the historical predicted fault probability sequence, performing fault analysis in a future time zone, and outputting a predicted fault probability interval sequence as a fault detection result of the pressure-free device; The process of predicting the occurrence of a pressure-free device failure based on the fault correlation feature distribution sequence and outputting a historical prediction failure probability sequence includes: Based on the historical inspection records of similar pressure-free devices, a sample fault correlation feature distribution set is collected, and the failure frequency of the pressure-free devices under different sample fault correlation feature distributions is counted as the sample failure probability to obtain the sample failure probability set; Using the sample fault association feature distribution set and the sample fault probability set, the generator and the discriminator of the adversarial neural network are trained until both the generation loss function and the discriminant loss function converge, thereby obtaining a fault prediction model; Using the fault prediction model, predicting the occurrence of pressure-free device faults in sequence according to the fault correlation feature distribution sequence, and outputting a historical predicted fault probability sequence; Wherein, based on the historical predicted fault probability sequence, performing fault analysis in the future time zone and outputting a predicted fault probability interval sequence includes: In a two-dimensional coordinate system, the historical predicted fault probability sequence is fitted according to the monitoring time sequence to generate a historical predicted fault probability curve; Configuring a future time zone, wherein the future time zone includes a number of consecutive monitoring time nodes; In a two-dimensional coordinate system, the historical predicted failure probability curve is extended to predict and obtain a number of predicted failure probabilities at a number of monitoring time nodes, and a predicted failure probability sequence is constructed; Performing a fluctuation error analysis on the historical predicted fault probability sequence and outputting a fluctuation error ratio; The predicted fault probability sequence is expanded according to the fluctuation error ratio, and a predicted fault probability interval sequence is output.

2. A pressure-free device detection method according to claim 1, characterized in that: Regularly capture images of the pressure-free device at the cable at preset time intervals, including: Configure key parts of the pressure-free device, wherein the key parts include but are not limited to the terminal connection part, the locking component, the positioning cylinder and the driving component; Images of multiple key parts of the pressure-free device at the cable are collected regularly at preset time intervals, and are fused into a complete image according to the position coordinates of the parts as the device image to obtain a device image sequence.

3. A pressure-free device detection method according to claim 1, characterized in that: Extracting key frames from the device image sequence to obtain a device key image sequence includes: Selecting the device image at the earliest monitoring time point in the device image sequence as the first image, setting the first image as the first key image, and using an image adjacent to the first image as the second image; Performing a similarity comparison on the first image and the second image according to a preset comparison strategy, and outputting a first image similarity; If the similarity of the first image is greater than or equal to a preset similarity threshold, the second image is discarded, and subsequent images are iteratively compared based on the first image; If the similarity of the first image is less than a preset similarity threshold, the second image is set as the second key image, and the second image is used as a benchmark to perform iterative similarity comparison of subsequent images until all images in the device image sequence are compared, and multiple key images are obtained, which are combined to construct a device key image sequence.

4. A pressure-free device detection method according to claim 3, characterized in that: The configuration method of the preset comparison strategy includes: Selecting a first similarity comparison algorithm according to pixel similarity measurement; Selecting a second similarity comparison algorithm according to the feature similarity measure; Selecting a third similarity comparison algorithm according to the structural similarity metric; A preset comparison strategy is obtained according to the combination of the first similarity comparison algorithm, the second similarity comparison algorithm and the third similarity comparison algorithm.

5. A pressure-free device detection method according to claim 2, characterized in that: Using a convolutional neural network, fault-related features are extracted from the key image sequence of the device to obtain a fault-related feature distribution sequence, including: Configuring fault-related features, wherein the fault-related features at least include displacement features, deformation features, and surface corrosion degree; Using the fault correlation characteristics as constraints, query historical inspection records of similar pressure-free devices, collect sample device image sets and sample fault correlation characteristic distribution sets, wherein the fault correlation characteristic distribution includes fault correlation characteristics of multiple key parts, and the key parts are marked with location coordinates; Using the sample device image set and the sample fault-related feature distribution set, supervised training and testing are performed on a convolutional neural network until the network converges to obtain a feature extractor; The feature extractor is used to sequentially extract fault-related features from a plurality of device key images in the device key image sequence, and a fault-related feature distribution sequence is output.

6. A pressure-free device detection method according to claim 1, characterized in that: Performing a fluctuation error analysis on the historical predicted fault probability sequence and outputting a fluctuation error ratio includes: Calculating the mean of multiple historical predicted failure probabilities in the historical predicted failure probability sequence to generate a mean of the historical predicted failure probabilities; Based on the mean value of the historical predicted failure probabilities, the deviation amplitudes of the multiple historical predicted failure probabilities are analyzed respectively, and the mean values ​​of the multiple deviation amplitudes are calculated to obtain the fluctuation error ratio.

7. A pressure-free device detection system, characterized in that: The steps for implementing the pressure-free device detection method according to any one of claims 1 to 6 include: An image acquisition module is used to regularly acquire images of the pressure-free device at the cable according to a preset time interval to obtain a sequence of device images; A key frame extraction module, configured to extract key frames from the device image sequence to obtain a device key image sequence; A fault feature extraction module is used to extract fault-related features from a key image sequence of the device using a convolutional neural network to obtain a fault-related feature distribution sequence; A fault occurrence prediction module is used to predict the occurrence of a pressure-free device fault based on the fault correlation feature distribution sequence and output a historical predicted fault probability sequence; The detection result obtaining module is used to perform fault analysis in a future time zone based on the historical predicted fault probability sequence, and output a predicted fault probability interval sequence as a fault detection result of the pressure-free device.

Images