A multi-parameter monitoring system and method based on GIL anti-vibration support

Abstract

The present invention discloses a multi-parameter monitoring system and method based on a GIL anti-vibration support. The system includes a functional box, a multi-parameter detection module, an SF6 leakage detection module, and a host computer. The functional box is a box structure, the outer shell is made of high molecular polymer material, and is made of one-piece injection molding. It is provided with 9 slots for placing detection modules, which can charge and transmit data to the 9 detection modules at the same time, and communicate with the host computer through a USB bus; furthermore, the above-mentioned 9 detection modules can include multiple multi-parameter detection modules and multiple SF6 leakage detection modules. The multi-parameter detection module is arranged on the GIL anti-vibration support, installed radially, and measures the current acceleration signal, temperature signal, tensile stress signal, and torsional stress signal. The present invention adopts a wireless installation method, which is different from the wired towing sensor method on the market. It is convenient and flexible to install and can work continuously for a long time.

Description

Technical Field

[0001] The present application belongs to the technical field of electric power transmission support technology, and in particular relates to a multi-parameter monitoring system and method based on a GIL anti-vibration support. Background Art

[0002] GIL (Gas Insulated Transmission Line) is a high-voltage, high-current, long-distance power transmission device that uses SF6, SF6 / N2 or other insulating gases, with the shell and conductor arranged coaxially.

[0003] Although GILs are relatively reliable devices, various potential problems can still arise throughout their lifecycle. The first is defects caused by internal factors. During processing, transportation, installation, and maintenance, GILs may experience loose components, poor contact, weld loss, metal particles in the cavity, metallic burrs, cracks in insulation, and air gaps between insulators and conductors. These can lead to partial discharge failures, unbalanced electromagnetic forces, and resulting vibration during operation. Second, potential failures are caused by external factors. For example, tunnel settlement, vehicle traffic, and rail transit pressure can affect the transmission line pipeline structure within underground tunnels, potentially causing uneven stress, distortion, and vibration of line components, impacting line safety and even causing equipment damage or accidents.

[0004] To solve the above problems, compensation is currently mainly achieved by installing units such as brackets and expansion joints. However, internal and external factors may still affect the deformation and stress of components such as GIL housings, guide rods, and insulators, thereby causing changes in equipment sealing and insulation characteristics, posing a serious threat to the long-term operational reliability of GIL equipment.

[0005] Common causes of mechanical failure in GILs include external casing vibration, vibration caused by electromagnetic forces and magnetostriction, mechanical damage caused by temperature-induced stress and strain, and potential exposure of the GIL's flange joints and internal conductor contact points. Developing an equipment status monitoring system that integrates temperature, vibration acceleration, stress, and leakage signals is crucial for improving the GIL circuit's vibration and torsion resistance, enabling GIL health assessments and defect warnings, and ensuring the system's long-term safe and reliable operation. However, currently, few existing technologies offer systems and methods for real-time monitoring and diagnosis of common mechanical failures in GILs. Summary of the Invention

[0006] Purpose of the invention: In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a multi-parameter monitoring system and monitoring method based on a GIL anti-vibration bracket.

[0007] Technical solution: According to a first aspect of the present invention, a multi-parameter monitoring system based on a GIL anti-vibration support is provided, comprising a multi-parameter detection module, an SF6 leakage detection module and a host computer;

[0008] The multi-parameter detection module is provided on the GIL anti-vibration support and radially installed between two GIL anti-vibration supports, and is used to measure the real-time acceleration signal, temperature signal, tensile stress signal and torsional stress signal of the GIL pipeline, and upload the relevant signals to the host computer for processing in real time;

[0009] The SF6 leakage detection module is provided on the flange joint of the GIL anti-vibration bracket and is used to detect the current SF6 leakage. The SF6 leakage detection module includes an SF6 closed chamber, a closed chamber clamp, an SF6 detection unit and a first control circuit board module. The SF6 closed chamber is assembled from four detachable arc shells and is used to be clamped on the outer shell of the GIL pipeline flange joint and fastened by the closed chamber clamp. One end of the SF6 detection unit is inserted into the SF6 closed chamber to measure whether there is an SF6 signal in the SF6 closed chamber, and the SF6 detection unit is electrically connected to the input end of the first control circuit board module;

[0010] The host computer performs multiple data analyses on vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals, and SF6 signals in real time through an improved network recognition model, thereby characterizing the health status of key structural components of the GIL and diagnosing and preventing machine failures of mechanical components.

[0011] Further, including:

[0012] The multi-parameter detection module includes a housing, and a temperature sensor module, an acceleration sensor module, a stress sensor module, a second control circuit board module, and a large-capacity battery module installed in the housing. The temperature sensor module is used to measure the temperature signal of the measured point, includes a temperature sensor model sht35, and is connected to the second control circuit board module via an I2C serial bus mode;

[0013] The acceleration sensor module is used to measure the vibration acceleration signal of the measured point, includes a single-axis acceleration sensor of model ADXL103, provides a voltage output after signal conditioning, and is electrically connected to the input end of the second control circuit board module;

[0014] There are two stress sensor modules, which are radially and axially installed inside the multi-parameter detection module respectively, and are electrically connected to the input end of the second control circuit board module;

[0015] The second control circuit board module includes a power conversion circuit, a storage circuit, a processor circuit and a data interface circuit;

[0016] The power conversion circuit converts the output voltage of the large-capacity battery module into 5V and 3.3V power through the power conversion circuit to provide working voltage for the storage circuit, processor circuit, data interface circuit, temperature sensor module, acceleration sensor module and stress sensor module;

[0017] The storage circuit includes a NAND memory model W29N08GVSIAA for storing temperature signals, acceleration signals, and stress signals, and is electrically connected to the processor circuit via a multiplexed 8-bit bus standard;

[0018] The processor circuit includes a main control chip of model STM32F103VET6, which controls the operation of the entire detection module, collects signals from the temperature sensor, acceleration sensor, and stress signal, and stores them through the storage circuit;

[0019] The data interface circuit adopts a magnetic contact interface to charge and read data for the multi-parameter detection module.

[0020] Further, including:

[0021] The first control circuit board module includes: a power conversion circuit, a storage circuit, a processor circuit and a data interface circuit;

[0022] The power conversion circuit converts the output voltage of the large-capacity battery module into 5V and 3.3V power through the power conversion circuit to provide working voltage for the storage circuit, processor circuit, data interface circuit, temperature sensor module, acceleration sensor module and stress sensor module;

[0023] The storage circuit mainly includes a NAND memory of model W29N08GVSIAA for storing leakage signals, and is electrically connected to the processor circuit via a multiplexed 8-bit bus standard;

[0024] The processor circuit mainly includes a main control chip of model STM32F103VET6, which controls the operation of the entire detection module, collects leakage signals, and stores them through the storage circuit;

[0025] The data interface circuit adopts a magnetic contact interface to charge and read data for the SF6 leakage detection module.

[0026] Further, including:

[0027] The host computer performs multiple data analysis on the vibration acceleration signal, temperature signal, tensile stress signal, torsional stress signal and SF6 signal in real time through the improved network recognition model, specifically including the following steps:

[0028] S1 forms a corresponding data set by collecting historical vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals and SF6 signals during normal operation of the GIL bracket at multiple time points;

[0029] S2 converts the signal dataset in the dataset into a format suitable for training the improved SE-ResNet twin network model. Specifically, the signal data is visualized as a one-dimensional / two-dimensional image, or frequency domain features are extracted to obtain a corresponding preprocessed dataset, and the preprocessed dataset is divided into corresponding training sets, validation sets, and test sets.

[0030] S3 builds an improved SE-ResNet twin network model, specifically including:

[0031] S31 chooses ResNet50 as the baseline model because it achieves a good balance between model performance and model accuracy. Its overall structure consists of convolutional layers, pooling layers, and fully connected layers. To improve sensitivity to key features, a compression and excitation (SE) module is added after the convolutional layers in ResNet50. This can improve the hierarchical representation of features while maintaining a low computational overhead. The SE module compresses the features of each channel, generates channel importance weights, and adjusts the response of each channel by multiplying them with the channel feature elements.

[0032] The SE module compresses the features of each channel into real numbers to represent the feature distribution on the channel, and then extracts the attention factor on each channel through excitation, regards it as an important parameter of the feature channel, and multiplies it with the feature element on the corresponding channel to obtain the result;

[0033] S32 calculates the Euclidean distance between the feature vectors extracted by the improved SE-ResNet to obtain the corresponding set D, the data value D in the set D i Used to measure the similarity between two corresponding inputs;

[0034] S33 Then, a contrast function is defined, which aims to minimize the distance between samples of the same type and maximize the distance between samples of different types.

[0035] Further, including:

[0036] The contrast function is expressed as:

[0037]

[0038] Among them, y is the true label value. When y=1, it corresponds to a positive sample pair. The contrast function converts the distance D iSquare it and minimize it to reduce the distance between similar samples. The value of this item is

[0039] When y=0, that is, when it corresponds to a negative sample pair, the contrast function will be minimized (mD i ) 2 , where m is a boundary distance threshold, ensuring that the distance between heterogeneous samples is greater than m. i When it exceeds m, the loss of this item is 0; therefore, during training, m is initialized to a certain value and allowed to automatically adjust to a suitable value through backpropagation during the training process.

[0040] Further, including:

[0041] The host computer performs multiple data analysis on vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals and SF6 signals in real time through an improved network recognition model, and also includes:

[0042] S4 uses the training set to train the improved SE-ResNet twin network model, adjusts the parameters through the data of the training set, performs forward propagation and backward propagation through a batch of data in the training set each time, and calculates the value of the loss function until the m value set by the comparison function is reached;

[0043] S5 uses the Adam optimization algorithm to help the model converge effectively during training, thereby continuously optimizing the model parameters.

[0044] Further, including:

[0045] In step S2, the signal data is visualized as a two-dimensional image to obtain a corresponding pre-processed data set, including:

[0046] S21 uses the Grammar cross-angle field difference method to encode the corresponding data sequences of vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals, and SF6 signals into two-dimensional images, so that the health of the samples can be detected later using the image feature processing method. Specifically:

[0047] Normalization and GADF encoding are as follows:

[0048] The collected one-dimensional signal data X={x1,x2,......x n Normalize to [-1,1] according to the time series;

[0049] S22 converts the normalized value into an angle The polar coordinate system of the time radius code r is expressed as:

[0050]

[0051] Among them, t i represents time, and N represents the constant factor in the transformation process;

[0052] After S23 transforms the time series into a polar coordinate system, the temporal correlation of different time intervals is identified by considering the angle sum and angle difference between each point;

[0053] S24 can transform a given time series into a two-dimensional feature image that is symmetrical along the diagonal through GADF image encoding, and retain the time-related features in the time series data.

[0054] Further, including:

[0055] The visualizing the signal data into a two-dimensional image to obtain a corresponding preprocessed data set further includes:

[0056] S25 labels the collected data according to the operating status to form classification labels for supervised learning in the health monitoring model. Specifically:

[0057] Each GADF image corresponds to a state label, the normal state is marked as 1, and the sample with abnormal signal is marked as -1, and the data is semi-randomly combined into two tuples;

[0058] For the corresponding SF6 signal, since the sample size of the SF6 signal before and after leakage is significantly lower than that of the normal state, a secondary labeling method is adopted, that is, one sample in the binary group is the normal label, and the other one is marked as 1 if it is normal, and marked as 0 if it is a sample before or after leakage.

[0059] Further, including:

[0060] The two inputs of the improved SE-ResNet twin network model are set as:

[0061] A known normal state sample, or the average features of multiple normal samples, is selected as a comparison benchmark. The features representing the normal state are used as the input of one branch, and the input of the other branch is the state to be judged. In the twin network, by calculating the similarity of the two input samples after being output from the improved SE-ResNet twin network model, the model can determine whether an abnormality has occurred, that is, the model determines the state of the key components of the GIL, thereby realizing real-time monitoring of system faults.

[0062] On the other hand, the present invention also provides a multi-parameter monitoring method based on a GIL anti-vibration support, the method comprising the following steps:

[0063] The multi-parameter detection module is set on the GIL anti-vibration bracket and radially installed between the two GIL anti-vibration brackets to measure the real-time acceleration signal, temperature signal, tensile stress signal and torsional stress signal of the GIL pipeline, and upload the relevant signals to the host computer for processing in real time;

[0064] An SF6 leakage detection module is arranged on the flange joint of the GIL anti-vibration bracket to detect the current SF6 leakage. The SF6 leakage detection module includes an SF6 closed chamber, a closed chamber clamp, an SF6 detection unit and a first control circuit board module. The SF6 closed chamber is assembled from four detachable arc shells and is used to be clamped on the outer shell of the GIL pipeline flange joint and fastened by the closed chamber clamp. One end of the SF6 detection unit is inserted into the SF6 closed chamber to measure whether there is an SF6 signal in the SF6 closed chamber, and the SF6 detection unit is electrically connected to the input end of the first control circuit board module.

[0065] The upper computer uses an improved network identification model to perform real-time analysis of multiple data including vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals, and SF6 signals, thereby characterizing the health status of key structural components of the GIL and diagnosing and preventing machine failures of mechanical components.

[0066] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0067] This invention, by mounting multiple detection modules, simultaneously measures vibration acceleration, temperature, stress, and leakage, significantly impacting the operational status, defect detection, and fault diagnosis of GIL mechanical systems. Specifically, through multi-dimensional data analysis of vibration acceleration, temperature, stress, and leakage, the health of key GIL components can be characterized, and mechanical failures can be diagnosed and prevented.

[0068] Moreover, this system adopts wireless installation mode, which is different from the wired towing sensor mode on the market. It is convenient and flexible to install and can work continuously for a long time. The device adopts a many-to-one mode of multiple detection modules and function boxes, which is suitable for the requirements of simultaneous testing of multiple test points. The device is miniaturized, portable and automated, which is convenient for on-site detection work.

[0069] Finally, the present invention adopts an improved SE-ResNet twin network model for the health status of key structural components, diagnosis and prevention of machine failures of mechanical components, and sets a comparison function that conforms to the collected signals of this application. While constraining model training, it can also distinguish between positive and negative samples, thereby improving the accuracy of model training and recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the conventional technology, the following briefly introduces the drawings required for use in the embodiments or the conventional technology descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0071] Figure 1 A connection diagram of a monitoring system according to an embodiment of the present invention;

[0072] Figure 2 This is a structural diagram of a monitoring system according to an embodiment of the present invention;

[0073] Figure 3 This is a schematic diagram of an SF6 leakage detection module according to an embodiment of the present invention;

[0074] Figure 4 A schematic diagram of a GADF encoding process according to an embodiment of the present invention;

[0075] Figure 5 This is a schematic diagram of the overall network architecture of the improved ResNet50 according to one embodiment of the present invention;

[0076] Figure 6 This is a schematic diagram of the structure of the improved ResNet50 according to one embodiment of the present invention;

[0077] Figure 7 This is a schematic diagram of the SE network structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0078] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention and not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0079] First, to facilitate understanding of the present application, a more comprehensive description of the present application will be provided below with reference to the accompanying drawings. The drawings illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the present application more thorough and comprehensive.

[0080] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein in the specification of this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The term "and / or" as used herein includes any and all couplings of one or more of the associated listed items.

[0081] It will be understood that the terms "first," "second," etc. used herein may be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish a first element from another element.

[0082] The present invention is based on a multi-parameter monitoring system of a GIL anti-vibration support, such as Figure 1 As shown, the structure includes a function box, a multi-parameter detection module, an SF6 leakage detection module and a host computer.

[0083] The functional box is a box structure, the shell is made of polymer material, and is made of one-piece injection molding. It is provided with 9 slots for placing detection modules. It can charge and transmit data to 9 detection modules at the same time and communicate with the host computer through the USB bus.

[0084] Furthermore, the above-mentioned nine detection modules may include multiple multi-parameter detection modules and multiple SF6 leakage detection modules.

[0085] The host computer mainly includes a computer and host computer software installed on the computer. The computer is connected to the function box through a provided USB cable to synchronize the working status of each detection module and read the data. The read data is displayed and published through the host computer software; further, the host computer software mainly realizes the reading and saving of temperature, vibration acceleration, stress, leakage data, data readback and call, data waveform display, channel selection, data processing and other functions.

[0086] like Figure 2 As shown, the multi-parameter detection module is arranged on the GIL anti-vibration bracket, installed radially, and measures the current acceleration signal, temperature signal, tensile stress signal and torsional stress signal;

[0087] like Figure 3 As shown, the SF6 leakage detection module is arranged on the flange joint provided with the GIL, and is used to detect the current SF6 leakage situation.

[0088] The multi-parameter detection module is installed on the GIL anti-vibration bracket by bolts on both sides, and is installed radially. It mainly includes a shell, and a temperature sensor module, an acceleration sensor module, a stress sensor module, a control circuit board module, and a large-capacity battery module installed in the shell.

[0089] The temperature sensor module is used to measure the temperature signal of the measured point, mainly including a temperature sensor of model sht35, which is connected to the control circuit board module through the I2C serial bus mode;

[0090] The acceleration sensor module is used to measure the vibration acceleration signal of the measured point, and mainly includes a single-axis acceleration sensor of model ADXL103, which provides a voltage output after signal conditioning and is electrically connected to the input end of the control circuit board module;

[0091] The stress sensor modules are provided with two, which are radially and axially mounted inside the multi-parameter detection module respectively, and are electrically connected to the input end of the control circuit board module;

[0092] The control circuit board module includes a power conversion circuit, a storage circuit, a processor circuit and a data interface circuit;

[0093] The power conversion circuit converts the output voltage of the large-capacity battery module into 5V and 3.3V power through the power conversion circuit to provide operating voltage for the storage circuit, processor circuit, data interface circuit, temperature sensor module, acceleration sensor module and stress sensor module;

[0094] The storage circuit mainly includes a NAND memory model W29N08GVSIAA for storing temperature signals, acceleration signals, and stress signals, and is electrically connected to the processor circuit via a multiplexed 8-bit bus standard;

[0095] The processor circuit mainly includes a main control chip of model STM32F103VET6, which controls the operation of the entire detection module, collects signals from the temperature sensor, acceleration sensor, and stress signal, and stores them through the storage circuit;

[0096] The data interface circuit adopts a magnetic contact interface for connecting to the function box to charge and read data for the multi-parameter detection module.

[0097] The SF6 leakage detection module is installed on the flange joint of the GIL. Its structure mainly includes SF6 sealed chamber, gasket, sealed chamber clamp, SF6 detection unit, control circuit board module and large-capacity battery module.

[0098] The SF6 sealed chamber is assembled from four detachable arc shells made of stainless steel, which are used to be clamped on the shell of the GIL flange joint and fastened by the gasket and the sealed chamber clamp. The SF6 detection unit mainly includes an SF6 sensor with model BA-2200-004. One end of the SF6 sensor is inserted into the SF6 sealed chamber to measure whether there is an SF6 signal in the SF6 sealed chamber, and is electrically connected to the input end of the control circuit board module.

[0099] The control circuit board module includes a power conversion circuit, a storage circuit, a processor circuit and a data interface circuit;

[0100] The power conversion circuit converts the output voltage of the large-capacity battery module into 5V and 3.3V power through the power conversion circuit to provide operating voltage for the storage circuit, processor circuit, data interface circuit, temperature sensor module, acceleration sensor module and stress sensor module;

[0101] The storage circuit mainly includes a NAND memory model W29N08GVSIAA for storing leakage signals, and is electrically connected to the processor circuit via a multiplexed 8-bit bus standard;

[0102] The processor circuit mainly includes a main control chip of model STM32F103VET6, which controls the operation of the entire detection module, collects leakage signals, and stores them through the storage circuit;

[0103] The data interface circuit adopts a magnetic contact interface for connecting to the functional box to charge and read data for the SF6 leakage detection module.

[0104] The multi-parameter monitoring system based on the GIL anti-vibration support structurally includes a function box, a multi-parameter detection module, an SF6 leakage detection module and a host computer. The function box is connected to the host computer via a USB interface and a USB connection cable.

[0105] The above-mentioned function box is also provided with a circuit board module and a mounting bracket module. The circuit board module is placed at the bottom of the internal box of the function box. The electrical connection is composed of two sets of bus modules, namely a power bus module and a data bus module. The power bus module converts the 220V voltage into the input power supply voltage of the large-capacity battery module provided in the detection module through a switching power supply, which is used to charge each detection module; the mounting bracket module is an aluminum alloy flat bracket, which is fixed to the upper middle part of the function box by screws and is provided with a mounting hole for the external USB interface of the circuit board module, a mounting hole for the power button, 9 detection module placement slots and a mounting hole for the charging status indicator light;

[0106] Specifically, the function box is connected to an external 220V power supply. By pressing the power button of the function box, the detection module inserted in the card slot can be charged. When charging is completed, the charging status indicator light 7 changes from red to green, indicating that charging is complete; the data bus module is used for the work synchronization and data reading of each detection module. It completes the communication with the host computer by connecting to the USB interface of the function box, and waits for the work instructions of the host computer to receive the corresponding working status.

[0107] The host computer uses the twin network model to analyze multiple data such as vibration acceleration, temperature data, stress, and leakage in real time to characterize the health status of key GIL structural components and diagnose and prevent mechanical component failures. The model construction and training process is as follows:

[0108] Step (1) Arrange sensor acquisition and data set preparation:

[0109] The multi-parameter monitoring module is placed on the GIL anti-vibration bracket to collect real-time data, and the temperature, vibration, stress and other data under normal operating conditions and before and after pipeline leakage are read respectively.

[0110] Data preprocessing includes operations such as data cleaning and normalization. Data cleaning uses methods such as interpolation and mean filling to fill in data with missing values ​​and errors, which can improve model training efficiency and reliability. Normalizing the data samples can improve model accuracy and generalization. The Gramian Angular Difference Field (GADF) method is then used to encode the one-dimensional time series into a two-dimensional image, allowing for subsequent use of image feature processing to perform sample health checks.

[0111] Normalization and GADF encoding are as follows:

[0112] The collected one-dimensional signal X={x1,x2,......x n Normalize the time series to between [-1,1], the formula is as follows:

[0113]

[0114] Normalized values ​​converted to angles Polar coordinate system of time radius encoding r:

[0115]

[0116] Among them, t i Represents time, and N represents the constant factor in the conversion process.

[0117] After converting the time series into a polar coordinate system, the temporal correlation of different time intervals is identified by considering the angle sum and angle difference between each point. The GADF is defined as follows:

[0118]

[0119] like Figure 4 As shown in Figure 3, GADF image encoding can transform a given time series into a two-dimensional feature image that is symmetrical along the diagonal line, and can retain the time-related features in the time series data.

[0120] The collected data is labeled according to operating status (normal, before and after a leak) to form classification labels for supervised learning in the health monitoring model. Each GADF image corresponds to a state label, with the normal state labeled 1 and samples before and after a leak labeled -1. The data is semi-randomly grouped into two-tuples. Because the number of samples before and after a leak is significantly lower than that of samples in the normal state, one sample in the two-tuple is labeled normal, and the other is labeled 1 if normal and 0 if it is a sample before and after a leak.

[0121] The dataset is partitioned into training, validation, and test sets in a 7:2:1 ratio. Stratified sampling is used to ensure that each class is represented equally in each subset. Each GADF image and its corresponding label are saved in a standard data format: the image is saved as a png file, and the corresponding data label is stored in the corresponding labels.csv file.

[0122] Step (2) Improved SE-ResNet twin network model training

[0123] Based on the prepared dataset, we consider using a twin network and using the data labels as samples to train the input network. The basic idea of ​​the twin network is to input the input data into two identical neural networks at the same time, and the two networks share the same weights and parameters. By learning the representation of the input data in the two networks, the twin network can calculate the similarity between the two input samples. Both inputs of the network are image data generated by GADF encoding, and feature extraction is performed by the improved ResNet50. The overall network architecture of the model is as follows Figure 5 shown.

[0124] like Figure 6As shown in the figure, ResNet50 was chosen as the baseline model because it strikes a good balance between model performance and accuracy. Its overall structure consists of convolutional layers, pooling layers, and fully connected layers. To improve sensitivity to key features, a compression and excitation (SE) module is added to ResNet50, which improves the hierarchical representation of features while maintaining low computational overhead. The SE module compresses the features of each channel, generates channel importance weights, and adjusts the response of each channel by multiplying them with the channel feature elements.

[0125] The structure of the SE module is as follows Figure 7 As shown in the figure, it compresses the features of each channel into real numbers to characterize the feature distribution on the channel, and then extracts the attention factor on each channel through excitation, regards it as an important parameter of the feature channel, and multiplies it with the feature elements on the corresponding channel to obtain the result.

[0126] The Euclidean distance is calculated for the feature vectors extracted by the Twin SE-ResNet, and the resulting value D is used to measure the similarity between the two inputs. Subsequently, a contrast loss is defined. The contrast function aims to minimize the distance between samples of the same class while maximizing the distance between samples of different classes. The contrast function used here is:

[0127]

[0128] Among them, y is the true label value. When y=1, it corresponds to a positive sample pair. The contrast function converts the distance D i Square it and minimize it to reduce the distance between similar samples. The value of this item is

[0129] When y=0, that is, when it corresponds to a negative sample pair, the contrast function will be minimized (mD i ) 2 , where m is a boundary distance threshold, ensuring that the distance between heterogeneous samples is greater than m. i When the value exceeds m, the loss is zero; therefore, during training, m is initialized to a certain value and automatically adjusted to an appropriate value through backpropagation. This contrast function constrains the model while minimizing the distance between samples of the same type and maximizing the distance between samples of different types, thereby distinguishing between positive and negative samples and improving the accuracy of model training and recognition.

[0130] Step (3) Model training:

[0131] The monitoring module is used to collect various data collected during the normal operation of the GIL bracket and various data collected before and after the failure of the GIL bracket, and the GADF method is used to convert them into two-dimensional feature images. During the training phase of the model, the model will adjust the parameters through the data of the training set. Each time, a batch of data from the training set is used for forward propagation and backpropagation, the value of the loss function is calculated, and the Adam optimization algorithm is used to help the model converge effectively during the training process, thereby continuously optimizing the model parameters. After feature extraction and selection of data such as acceleration, temperature, stress, and leakage, a database is constructed for offline training of the model. The training set is used to train the twin network, and the validation set is used to evaluate the model and perform hyperparameter adjustment to prevent overfitting until the network model reaches a high level of accuracy. Finally, the network model is saved for online diagnosis. After training, the model is evaluated using the test set to ensure the generalization performance of the model on unseen data.

[0132] Step (4) System integration and testing:

[0133] The constructed model is deployed on the host computer, which monitors the fault status of the GIL bracket online, collects monitoring module data, and converts it into a GADF feature image. Since the network is a twin structure, it requires two inputs and determines whether an abnormality has occurred by judging the distance in the sample space. Therefore, during the inference phase of the model, a known normal state sample or the average feature of multiple normal samples will be selected as a comparison benchmark. The feature representing the normal state is used as the input of one branch, and the input of the other branch is the state to be judged. In the twin network, by calculating the similarity of the two input samples, the model can determine whether an abnormality has occurred. The model determines the status of the key components of the GIL, thereby realizing real-time monitoring of system faults. If a fault occurs, an alarm must be issued before or after the leak so that detailed inspection and processing can be carried out as soon as possible.

[0134] The testing process of this system includes:

[0135] Before the test, the multi-parameter detection module is installed radially on the GIL anti-vibration bracket by bolts; the SF6 leakage detection module is installed on the GIL flange interface, and the detection time is three days.

[0136] Before testing, each detection module needs to be synchronized and set up, and the host computer sends a start-of-work command to each detection module through the data bus connected to the USB interface; after the test is completed, each detection module needs to be synchronized and set up, and the host computer sends an end-of-work command to each detection module through the data bus connected to the USB interface, and reads vibration acceleration, temperature data, stress data, and leakage data. This helps the host computer to compare and analyze the test data of components in the same time period.

[0137] The charging process is as follows:

[0138] Place the detection module in the function box;

[0139] Start the power switch button of the function box;

[0140] The detection module starts charging and the charging indicator light turns red;

[0141] After the detection module is fully charged, the charging indicator turns green.

[0142] The synchronization process is as follows:

[0143] Place the detection module in the function box;

[0144] Send start / end work synchronization instructions to each detection module through the host computer;

[0145] Each synchronized detection module sends a synchronization response instruction back to the host computer;

[0146] The host computer displays the device synchronization success. If the synchronization fails, the failed device number is displayed.

[0147] The data transmission process is as follows:

[0148] Place the detection module in the function box;

[0149] Read and set via the host computer software;

[0150] After setting, click the Read button to read the data;

[0151] After reading, the read data is saved, displayed, processed and other related operations are performed.

[0152] Specifically:

[0153] Device initialization:

[0154] Place the detection module in the function box and press the power-on button on the function box. The host computer will send a "start work" synchronization command to the function box, which will then distribute the synchronization command to each detection module. After receiving the synchronization command, each detection module will send a synchronization response command back to the host computer. At this time, the software on the host computer will display that the synchronization of each detection module is successful. If the synchronization of a detection module fails, the device number of the detection module that failed synchronization will be displayed in the software on the host computer. At this time, you can try to synchronize again. If the synchronization still fails, you need to check whether there is poor contact between the connection contacts between the detection module and the function box. After all detection modules have been synchronized successfully, install all detection modules in the location to be tested. After 72 hours of continuous operation, recycle each detection module and place it back in the function box. Repeat the above instruction synchronization steps to carry out the "end work" synchronization process. After completion, wait for the next step of data transmission.

[0155] Data transmission:

[0156] Place the detection modules in the function box and have the host computer identify each detection module. If all detection modules can be recognized by the host computer, data reading can begin. If a detection module cannot be recognized, the device number of the detection module will be displayed on the host computer. Check whether there is any poor contact between the detection module and the function box. After all detection modules are successfully connected, set up data reading through the host computer software. After the software settings are completed, read the data through the USB interface and read the test data from the 9 detection modules into the host computer in turn for storage. After all detection module data has been read and saved, the host computer software performs data processing, data graphing, and display.

[0157] Charging process:

[0158] After each test is completed, each detection module needs to be charged for use in the next test. The charging process is as follows:

[0159] Place each detection module in the function box;

[0160] Start the power switch button of the function box;

[0161] If the connection between each detection module and the charging contact of the function box is normal, the charging indicator light will turn red and charging will begin. If the connection between the detection module and the contact of the function box is abnormal, the indicator light will not light up, indicating that the device connection is abnormal. In this case, you need to manually check the device contact connection. After the connection problem is resolved, each detection module will start charging.

[0162] If the detection module is fully charged, the charging indicator light turns green, indicating that charging is complete.

[0163] Turn off the power button on the function box to end charging.

[0164] On the other hand, the present invention also provides a multi-parameter monitoring method based on a GIL anti-vibration support, the method comprising the following steps:

[0165] The multi-parameter detection module is set on the GIL anti-vibration bracket and radially installed between the two GIL anti-vibration brackets to measure the real-time acceleration signal, temperature signal, tensile stress signal and torsional stress signal of the GIL pipeline, and upload the relevant signals to the host computer for processing in real time;

[0166] An SF6 leakage detection module is arranged on the flange joint of the GIL anti-vibration bracket to detect the current SF6 leakage. The SF6 leakage detection module includes an SF6 closed chamber, a closed chamber clamp, an SF6 detection unit and a first control circuit board module. The SF6 closed chamber is assembled from four detachable arc shells and is used to be clamped on the outer shell of the GIL pipeline flange joint and fastened by the closed chamber clamp. One end of the SF6 detection unit is inserted into the SF6 closed chamber to measure whether there is an SF6 signal in the SF6 closed chamber, and the SF6 detection unit is electrically connected to the input end of the first control circuit board module.

[0167] The upper computer uses an improved network identification model to perform real-time analysis of multiple data including vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals, and SF6 signals, thereby characterizing the health status of key structural components of the GIL and diagnosing and preventing machine failures of mechanical components.

[0168] Other technical features of the multi-parameter monitoring method based on the GIL anti-vibration support described in the present invention correspond to those of the multi-parameter monitoring system based on the GIL anti-vibration support, and will not be repeated here.

[0169] The scope of protection of the present disclosure is not limited to the above-described embodiments. Obviously, those skilled in the art may make various modifications and variations to the present disclosure without departing from the scope and spirit of the present disclosure. If such modifications and variations fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is intended to include such modifications and variations.

Claims

1. A multi-parameter monitoring system based on GIL anti-vibration support, characterized in that: Including multi-parameter detection module, SF6 leakage detection module and host computer; The multi-parameter detection module is provided on the GIL anti-vibration support and radially installed between two GIL anti-vibration supports, and is used to measure the real-time acceleration signal, temperature signal, tensile stress signal and torsional stress signal of the GIL pipeline, and upload the relevant signals to the host computer for processing in real time; The SF6 leakage detection module is provided on the flange joint of the GIL anti-vibration bracket and is used to detect the current SF6 leakage. The SF6 leakage detection module includes an SF6 closed chamber, a closed chamber clamp, an SF6 detection unit and a first control circuit board module. The SF6 closed chamber is assembled from four detachable arc shells and is used to be clamped on the outer shell of the GIL pipeline flange joint and fastened by the closed chamber clamp. One end of the SF6 detection unit is inserted into the SF6 closed chamber to measure whether there is an SF6 signal in the SF6 closed chamber, and the SF6 detection unit is electrically connected to the input end of the first control circuit board module; The host computer performs multiple data analyses on vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals, and SF6 signals in real time through an improved network recognition model, thereby characterizing the health status of key structural components of the GIL and diagnosing and preventing machine failures of mechanical components. The host computer performs multiple data analysis on the vibration acceleration signal, temperature signal, tensile stress signal, torsional stress signal and SF6 signal in real time through the improved network recognition model, specifically including the following steps: S1 forms a corresponding data set by collecting historical vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals and SF6 signals during normal operation of the GIL bracket at multiple time points; S2 converts the signal dataset in the dataset into a format suitable for training the improved SE-ResNet twin network model. Specifically, the signal data is visualized as a one-dimensional / two-dimensional image, or frequency domain features are extracted to obtain a corresponding preprocessed dataset, and the preprocessed dataset is divided into corresponding training sets, validation sets, and test sets. S3 builds an improved SE-ResNet twin network model, specifically including: S31 chooses ResNet50 as the baseline model because it achieves a good balance between model performance and model accuracy. Its overall structure consists of convolutional layers, pooling layers, and fully connected layers. To improve sensitivity to key features, a compression and excitation (SE) module is added after the convolutional layers in ResNet50. This can improve the hierarchical representation of features while maintaining a low computational overhead. The SE module compresses the features of each channel, generates channel importance weights, and adjusts the response of each channel by multiplying them with the channel feature elements. The SE module compresses the features of each channel into real numbers to represent the feature distribution on the channel, and then extracts the attention factor on each channel through excitation, regards it as an important parameter of the feature channel, and multiplies it with the feature element on the corresponding channel to obtain the result; S32 calculates the Euclidean distance between the feature vectors extracted by the improved SE-ResNet to obtain the corresponding set D, the data value D in the set D i Used to measure the similarity between two corresponding inputs; S33 then defines a contrast function, wherein the contrast function aims to minimize the distance between samples of the same class and maximize the distance between samples of different classes; The contrast function is expressed as: Among them, y is the true label value. When y=1, it corresponds to a positive sample pair. The contrast function converts the distance D i Square it and minimize it to reduce the distance between similar samples. The value of this item is When y=0, that is, when it corresponds to a negative sample pair, the contrast function will be minimized (mD i ) 2 , where m is a boundary distance threshold, ensuring that the distance between heterogeneous samples is greater than m. i When it exceeds m, the loss of this item is 0; therefore, during training, m is initialized to a certain value and allowed to automatically adjust to a suitable value through backpropagation during the training process.

2. The multi-parameter monitoring system based on the GIL anti-vibration support according to claim 1, characterized in that: The multi-parameter detection module includes a housing, and a temperature sensor module, an acceleration sensor module, a stress sensor module, a second control circuit board module, and a large-capacity battery module installed in the housing. The temperature sensor module is used to measure the temperature signal of the measured point, includes a temperature sensor model sht35, and is connected to the second control circuit board module via an I2C serial bus mode; The acceleration sensor module is used to measure the vibration acceleration signal of the measured point, includes a single-axis acceleration sensor of model ADXL103, provides a voltage output after signal conditioning, and is electrically connected to the input end of the second control circuit board module; There are two stress sensor modules, which are radially and axially installed inside the multi-parameter detection module respectively, and are electrically connected to the input end of the second control circuit board module; The second control circuit board module includes a power conversion circuit, a storage circuit, a processor circuit and a data interface circuit; The power conversion circuit converts the output voltage of the large-capacity battery module into 5V and 3.3V power through the power conversion circuit to provide working voltage for the storage circuit, processor circuit, data interface circuit, temperature sensor module, acceleration sensor module and stress sensor module; The storage circuit includes a NAND memory model W29N08GVSIAA for storing temperature signals, acceleration signals, and stress signals, and is electrically connected to the processor circuit via a multiplexed 8-bit bus standard; The processor circuit includes a main control chip of model STM32F103VET6, which controls the operation of the entire detection module, collects signals from the temperature sensor, acceleration sensor, and stress signal, and stores them through the storage circuit; The data interface circuit adopts a magnetic contact interface to charge and read data for the multi-parameter detection module.

3. The multi-parameter monitoring system based on the GIL anti-vibration support according to claim 1, characterized in that: The first control circuit board module includes: a power conversion circuit, a storage circuit, a processor circuit and a data interface circuit; The power conversion circuit converts the output voltage of the large-capacity battery module into 5V and 3.3V power through the power conversion circuit to provide working voltage for the storage circuit, processor circuit, data interface circuit, temperature sensor module, acceleration sensor module and stress sensor module; The storage circuit mainly includes a NAND memory of model W29N08GVSIAA for storing leakage signals, and is electrically connected to the processor circuit via a multiplexed 8-bit bus standard; The processor circuit mainly includes a main control chip of model STM32F103VET6, which controls the operation of the entire detection module, collects leakage signals, and stores them through the storage circuit; The data interface circuit adopts a magnetic contact interface to charge and read data for the SF6 leakage detection module.

4. The multi-parameter monitoring system based on the GIL anti-vibration support according to claim 1, characterized in that: The host computer performs multiple data analysis on vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals and SF6 signals in real time through an improved network recognition model, and also includes: S4 uses the training set to train the improved SE-ResNet twin network model, adjusts the parameters through the data of the training set, performs forward propagation and backward propagation through a batch of data in the training set each time, and calculates the value of the loss function until the m value set by the comparison function is reached; S5 uses the Adam optimization algorithm to help the model converge effectively during training, thereby continuously optimizing the model parameters.

5. The multi-parameter monitoring system based on the GIL anti-vibration support according to claim 4 is characterized in that: In step S2, the signal data is visualized as a two-dimensional image to obtain a corresponding pre-processed data set, including: S21 uses the Grammar cross-angle field difference method to encode the corresponding data sequences of vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals, and SF6 signals into two-dimensional images, so that the health of the samples can be detected later using the image feature processing method. Specifically: Normalization and GADF encoding are as follows: The collected one-dimensional signal data X={x1,x2,......x n Normalize to [-1,1] according to the time series; S22 converts the normalized value into an angle The polar coordinate system of the time radius code r is expressed as: Among them, t i represents time, and N represents the constant factor in the transformation process; After S23 transforms the time series into a polar coordinate system, the temporal correlation of different time intervals is identified by considering the angle sum and angle difference between each point; S24 can transform a given time series into a two-dimensional feature image that is symmetrical along the diagonal through GADF image encoding, and retain the time-related features in the time series data.

6. The multi-parameter monitoring system based on the GIL anti-vibration support according to claim 5, characterized in that: The visualizing the signal data into a two-dimensional image to obtain a corresponding preprocessed data set further includes: S25 labels the collected data according to the operating status to form classification labels for supervised learning in the health monitoring model. Specifically: Each GADF image corresponds to a state label, the normal state is marked as 1, and the sample with abnormal signal is marked as -1, and the data is semi-randomly combined into two tuples; For the corresponding SF6 signal, since the sample size of the SF6 signal before and after leakage is significantly lower than that of the normal state, a secondary labeling method is adopted, that is, one sample in the binary group is the normal label, and the other one is marked as 1 if it is normal, and marked as 0 if it is a sample before or after leakage.

7. The multi-parameter monitoring system based on the GIL anti-vibration support according to claim 5, characterized in that: The two inputs of the improved SE-ResNet twin network model are set as: A known normal state sample, or the average features of multiple normal samples, is selected as a comparison benchmark. The features representing the normal state are used as the input of one branch, and the input of the other branch is the state to be judged. In the twin network, by calculating the similarity of the two input samples after being output from the improved SE-ResNet twin network model, the model can determine whether an abnormality has occurred, that is, the model determines the state of the key components of the GIL, thereby realizing real-time monitoring of system faults.

8. A multi-parameter monitoring method based on a GIL anti-vibration support, characterized by: The method comprises the following steps: The multi-parameter detection module is set on the GIL anti-vibration bracket and radially installed between the two GIL anti-vibration brackets to measure the real-time acceleration signal, temperature signal, tensile stress signal and torsional stress signal of the GIL pipeline, and upload the relevant signals to the host computer for processing in real time; An SF6 leakage detection module is arranged on the flange joint of the GIL anti-vibration bracket to detect the current SF6 leakage. The SF6 leakage detection module includes an SF6 closed chamber, a closed chamber clamp, an SF6 detection unit and a first control circuit board module. The SF6 closed chamber is assembled from four detachable arc shells and is used to be clamped on the outer shell of the GIL pipeline flange joint and fastened by the closed chamber clamp. One end of the SF6 detection unit is inserted into the SF6 closed chamber to measure whether there is an SF6 signal in the SF6 closed chamber, and the SF6 detection unit is electrically connected to the input end of the first control circuit board module. The host computer uses an improved network recognition model to perform real-time analysis of multiple data including vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals, and SF6 signals, thereby characterizing the health status of key GIL structural components and diagnosing and preventing mechanical failures. The host computer performs multiple data analysis on the vibration acceleration signal, temperature signal, tensile stress signal, torsional stress signal and SF6 signal in real time through the improved network recognition model, specifically including the following steps: S1 forms a corresponding data set by collecting historical vibration acceleration signals, temperature signals, tensile stress signals, torsional stress signals and SF6 signals during normal operation of the GIL bracket at multiple time points; S2 converts the signal dataset in the dataset into a format suitable for training the improved SE-ResNet twin network model. Specifically, the signal data is visualized as a one-dimensional / two-dimensional image, or frequency domain features are extracted to obtain a corresponding preprocessed dataset, and the preprocessed dataset is divided into corresponding training sets, validation sets, and test sets. S3 builds an improved SE-ResNet twin network model, specifically including: S31 chooses ResNet50 as the baseline model because it achieves a good balance between model performance and model accuracy. Its overall structure consists of convolutional layers, pooling layers, and fully connected layers. To improve sensitivity to key features, a compression and excitation (SE) module is added after the convolutional layers in ResNet50. This can improve the hierarchical representation of features while maintaining a low computational overhead. The SE module compresses the features of each channel, generates channel importance weights, and adjusts the response of each channel by multiplying them with the channel feature elements. The SE module compresses the features of each channel into real numbers to represent the feature distribution on the channel, and then extracts the attention factor on each channel through excitation, regards it as an important parameter of the feature channel, and multiplies it with the feature element on the corresponding channel to obtain the result; S32 calculates the Euclidean distance between the feature vectors extracted by the improved SE-ResNet to obtain the corresponding set D, the data value D in the set D i Used to measure the similarity between two corresponding inputs; S33 then defines a contrast function, wherein the contrast function aims to minimize the distance between samples of the same class and maximize the distance between samples of different classes; The contrast function is expressed as: Among them, y is the true label value. When y=1, it corresponds to a positive sample pair. The contrast function converts the distance D i Square it and minimize it to reduce the distance between similar samples. The value of this item is When y=0, that is, when it corresponds to a negative sample pair, the contrast function will be minimized (mD i ) 2 , where m is a boundary distance threshold, ensuring that the distance between heterogeneous samples is greater than m. i When it exceeds m, the loss of this item is 0; therefore, during training, m is initialized to a certain value and allowed to automatically adjust to a suitable value through backpropagation during the training process.

Images