Rotary machine equipment maintenance method and system based on digital twinning technology

Abstract

The invention relates to the technical field of testing and overhauling, in particular to a rotating machine equipment overhauling method and system based on the digital twinning technology, and the method comprises the steps: constructing a transformer substation three-dimensional virtual model which comprises a physical space, a virtual space and a digital service space; acquiring sensor data of a to-be-modeled object through a sensor assembly, processing the data, constructing a three-dimensional model, and filling the three-dimensional model into the virtual space; the physical space and the virtual space are mapped to the digital service space, the equipment operation state and fault positioning information are displayed in real time through the digital service space, overhaul suggestions are provided, the overhaul device is controlled by generating an overhaul instruction to overhaul the rotating machine equipment, and efficient monitoring and maintenance of the transformer substation equipment are achieved.

Description

Technical Field

[0001] The present invention relates to the field of testing and maintenance technology, and in particular to a transfer equipment maintenance method and system based on digital twin technology. Background Art

[0002] With the increasing complexity of power systems and the improvement of automation levels, substations, as key nodes for power transmission and distribution, have gradually become an important part of power grid operations. In order to ensure the safe and stable operation of the power system, it is necessary to monitor various types of equipment in the substation, especially the health status and operating efficiency of the transfer equipment. However, traditional substation equipment maintenance methods mainly rely on regular maintenance and passive troubleshooting, which is not only inefficient, but also has problems such as insufficient detection accuracy, inaccurate fault location, and high maintenance costs.

[0003] Digital twin technology is introduced into existing technologies to monitor power systems. Digital twin technology builds digital virtual models of physical equipment to achieve real-time interaction and mapping between physical space and virtual space, providing technical support for equipment operation monitoring, fault diagnosis, maintenance decision-making, etc. Digital twin technology can not only monitor the operating status of equipment in real time, but also predict faults and optimize maintenance through simulation, thereby significantly improving the efficiency and reliability of equipment management.

[0004] For example, the patent with publication number CN107168238A discloses an integrated system for substation equipment maintenance and information management, including a controller, which is connected to a touch panel and a data storage device, and also includes a QR code reader for scanning and identifying a QR code; the data storage device is used to store relevant information of the device, documents recording the device operation method and operation steps, and operation videos; the QR code is used to represent the relevant information of the device; the touch panel is used to display the documents and operation videos of the device operation method and operation steps; the user scans the QR code through the QR code reader to obtain the QR code information, and obtains the relevant information of the represented device through the QR code information; when the controller obtains the QR code information through the user, the relevant information of the represented device is matched with the relevant information of the device in the data storage device; when the match is successful, the touch panel is triggered to display the information. This invention solves the problem that the existing equipment data information is scattered and unconcentrated, and the maintenance efficiency is low.

[0005] The above patents have the problem raised by this background technology: the transfer equipment in the substation involves different sensor data. How to effectively integrate and process these heterogeneous data to generate a high-precision three-dimensional model is still a problem to be solved. In order to solve the above problems, this application designs a transfer equipment maintenance method and system based on digital twin technology. Summary of the invention

[0006] The technical problem to be solved by the present invention is to address the deficiencies in the prior art and provide a transfer equipment maintenance method and system based on digital twin technology. The method includes constructing a three-dimensional virtual model of a substation including a physical space, a virtual space and a digital service space; acquiring sensor data of the object to be modeled through a sensor component, processing the data, constructing a three-dimensional model and filling it into the virtual space; mapping the physical space and the virtual space to the digital service space, displaying the equipment operating status, fault location information and providing maintenance suggestions in real time through the digital service space, and controlling the maintenance device to perform maintenance on the transfer equipment by generating maintenance instructions.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A turnaround equipment maintenance system based on digital twin technology, the system comprising an equipment virtual module, an equipment operation status monitoring module and a test and maintenance control module;

[0009] The equipment virtualization module is used to perform three-dimensional visualization of the transfer equipment in the substation, and the equipment virtualization module is configured with a data processing strategy and a three-dimensional modeling strategy;

[0010] The equipment operation status monitoring module is used to monitor the transfer equipment through a visual interface, and to perform real-time analysis on the monitoring data, identify abnormal data patterns, and calculate analysis results;

[0011] The test and maintenance control module is used to generate a test and maintenance plan according to the analysis result, convert the test and maintenance plan into a machine control instruction, and output the machine control instruction;

[0012] The test and maintenance control module is configured with a maintenance control strategy, which analyzes the analysis results according to an intelligent decision-making algorithm, generates a test and maintenance plan, and generates maintenance instructions through a control algorithm, and controls the maintenance device to perform maintenance on the transfer equipment through the maintenance instructions.

[0013] The equipment operation status monitoring module comprises:

[0014] Equipment monitoring unit: used to monitor the transfer equipment through sensors and obtain the monitoring data of the equipment;

[0015] Real-time data analysis unit: performs real-time analysis on all monitoring data, identifies abnormal data patterns and calculates analysis results;

[0016] The equipment operation status monitoring module is configured with a data analysis strategy, which includes temperature analysis logic, vibration analysis logic, pressure analysis logic, current analysis logic and flow analysis logic;

[0017] The temperature analysis logic is used to analyze the temperature data collected by the temperature sensor to determine whether there is an overheating fault;

[0018] The pressure analysis logic is used to analyze the pressure data collected by the pressure sensor, identify the trend and fluctuation of pressure over time, and determine whether there is overpressure and pressure relief;

[0019] The flow analysis logic is used to analyze the flow data collected by the turbine flow meter to determine whether there is a flow exceeding the limit;

[0020] The temperature analysis logic, vibration analysis logic, pressure analysis logic, current analysis logic and flow analysis logic are configured in the real-time data analysis unit.

[0021] The vibration analysis logic is used to analyze the vibration signal collected by the vibration sensor to determine whether there is abnormal vibration of the equipment bearing, specifically including:

[0022] Reconstruct the vibration signal;

[0023] Performing Fourier transform on the reconstructed vibration signal, dividing the Fourier spectrum of the vibration signal into frequency bands with different center frequencies;

[0024] Iterate each frequency band and calculate the optimal solution that satisfies the minimum sum of bandwidths of each modal component.

[0025] The power spectrum is calculated according to the optimal solution of each modal component, the characteristic frequency of the bearing signal of the transfer equipment is obtained, the characteristic frequency of the bearing signal of the transfer equipment is compared with the theoretical frequency of the bearing signal of the transfer equipment, and the vibration signal analysis result is output.

[0026] The current analysis logic is used to analyze the current signal collected by the current sensor to determine whether there is load startup and current overload, specifically including:

[0027] The current signal is rectified and smoothed, and the frequency domain analysis of the current signal is performed through short-time Fourier transform to extract high-frequency and low-frequency components;

[0028] Compare the high frequency component of the current signal with the current detection threshold. If it is greater than or equal to the current detection threshold, it indicates that there is a current overload.

[0029] The low-frequency component of the current signal is compared with the current initial threshold value. If it is greater than or equal to the current initial threshold value, it indicates that a load is started.

[0030] The test and maintenance control module comprises:

[0031] A solution generation unit is used to process the analysis results and generate a minimum cost test and maintenance solution;

[0032] A scheme execution unit, used to convert the test and maintenance scheme into a machine instruction executable by the system, and control the maintenance device to perform maintenance on the transfer equipment through the machine instruction;

[0033] The execution feedback unit is used to monitor the operating status of the transfer equipment in real time and compare it with the expected maintenance results. According to the real-time feedback results, the maintenance control strategy and machine control instructions are automatically adjusted;

[0034] The maintenance control strategy includes an optimal solution generation logic and a control logic. The optimal solution generation logic is configured in the solution generation unit, and the control logic is configured in the solution execution unit.

[0035] The optimal solution generation logic includes:

[0036] Construct a genetic algorithm and define the objective function and constraints;

[0037] The analysis results and the equipment operation status are used as the genetic algorithm input parameters, the input parameters are iteratively optimized by the genetic algorithm, and the optimal solution of the objective function is output;

[0038] The optimal solution of the objective function is used as the test and maintenance plan.

[0039] The device virtual module includes:

[0040] A three-dimensional virtual model building unit, used to build a three-dimensional virtual model of the substation and perform data mapping and interaction between the physical space, virtual space and digital service space;

[0041] A sensor data acquisition unit, used to obtain an external image of the object to be modeled and a laser scanning point cloud of the object to be modeled through a variety of sensor components;

[0042] The data processing unit is used to process the external data and internal data of the object to be modeled and to perform three-dimensional visualization of the transfer equipment in the substation.

[0043] The three-dimensional virtual model building unit comprises:

[0044] Physical space unit: includes physical equipment and its layout, and collects physical data of equipment through sensors;

[0045] Virtual space unit: digital representation of the equipment and environment of the physical space, including the 3D model of the equipment, the virtual environment and simulation data;

[0046] Digital service space unit: used to provide digital services for real-time monitoring, data analysis and maintenance guidance;

[0047] The data processing strategy includes image processing logic and point cloud processing logic, wherein the image processing logic is used to extract surface features of the transfer equipment through image processing technology, and the point cloud processing logic is used to extract internal features of the transfer equipment;

[0048] The data processing unit comprises:

[0049] Image processing subunit: processes external data and extracts external structural features;

[0050] Point cloud processing subunit: processes internal data and extracts internal point cloud features;

[0051] Feature fusion subunit: fuses external structural features and internal point cloud features to generate a 3D model of the object to be modeled;

[0052] Model merging subunit: merge multiple 3D models to build a complete 3D model of the substation;

[0053] The three-dimensional modeling strategy includes feature fusion logic, and the feature fusion logic is used to fuse external structural features and internal point cloud features.

[0054] The image processing logic is configured in the image processing subunit, and the point cloud processing logic is configured in the point cloud processing subunit;

[0055] The feature fusion logic is configured in the feature fusion subunit.

[0056] The image processing logic comprises:

[0057] Compressing the external image of the object to be modeled to obtain a compressed reconstructed image;

[0058] Perform camera calibration on the image sensor, obtain the extrinsic parameter matrix and intrinsic parameter matrix of the image sensor, perform coordinate transformation on the compressed reconstructed image according to the extrinsic parameter matrix and the intrinsic parameter matrix, and transform the pixels of the compressed reconstructed image from the camera coordinate system to the world coordinate system;

[0059] Calculate the global mean of all pixels of the compressed reconstructed image after coordinate transformation, perform weighted sampling through the upsampling function, and calculate the global feature map;

[0060] Compare the confidence of the pixel points of the global feature map with the confidence threshold, retain the pixel points greater than the confidence threshold, and obtain the surface feature map;

[0061] Performing regional cutting on the surface feature map, extracting the regional temporal features after the surface feature map is cut according to the temporal feature extraction function, and extracting the regional spatial features after the surface feature map is cut through the attention mechanism;

[0062] The regional temporal features and regional spatial features of the surface feature map are fused to obtain the bilinear regional features. The bilinear regional features of all regions are integrated through the pooling function to calculate the external structural features.

[0063] The calculation formula of the external structural characteristics is:

[0064]

[0065] Among them, Gb fd represents the external structural features, a represents the cut area of ​​a single surface feature map, A represents the total number of cut areas of the surface feature map, b{·} represents the pooling function, T f (·) represents the time feature extraction function, hb fi represents the surface feature map, * represents the fusion operation, Ex[·] represents the excitation function in the attention mechanism, W represents the horizontal length of the surface feature map, H represents the vertical length of the surface feature map, n represents the unit horizontal length of the surface feature map, m represents the unit vertical length of the surface feature map, v a Represents the convolution kernel corresponding to the cut area.

[0066] The point cloud processing logic includes:

[0067] Preprocessing the laser scanning point cloud according to the laser scanning point cloud of the internal parts of the object to be modeled;

[0068] Find the most suitable 3D plane through plane fitting, perform coordinate transformation on the pre-processed 3D point cloud data, and transfer the laser scanning point cloud from the radar coordinate system to the world coordinate system;

[0069] The world coordinate system is divided into grids at equal intervals, the laser scanning point cloud is represented by voxels, and the voxel index of the laser scanning point cloud is projected to the corresponding grid, the points in the voxel are connected, the local feature information is enhanced, and the voxel features of the laser scanning point cloud are obtained;

[0070] The laser scanning point cloud voxel features are clustered according to the clustering algorithm, and the grids where the laser scanning point cloud voxel features are located under the same central cluster are merged to extract the internal point cloud features. The calculation formula of the internal point cloud features is:

[0071]

[0072] Among them, T arepresents the internal point cloud features corresponding to the voxel features of the laser scanning point cloud in the same central cluster after clustering, i represents a single voxel feature of the laser scanning point cloud in the central cluster, M represents the total number of voxel features of the laser scanning point cloud in the central cluster, and η i represents the eigenvalue of the voxel feature of the i-th laser scanning point cloud, d represents the clustering threshold corresponding to the central cluster, r2 represents the farthest distance scanned by the laser scanner, r1 represents the closest distance scanned by the laser scanner, and p i_x represents the voxel horizontal coordinate of the i-th laser scanning point cloud voxel feature, p i_y The voxel ordinate represents the feature of the i-th laser scanning point cloud voxel, p i_z Represents the voxel vertical coordinate of the i-th laser scanning point cloud voxel feature.

[0073] The cluster analysis comprises:

[0074] Preliminarily clustering the voxel features of the laser scanning point cloud by using a DBSCAN clustering algorithm to obtain core points;

[0075] The core points after DBSCAN clustering are used as the initial center clusters of the K-means clustering algorithm, and the Euclidean distances from the voxel features of the laser scanning point cloud to the corresponding initial center clusters are calculated;

[0076] Clustering is performed according to the Euclidean distance, and the average clustering error of the initial central cluster is calculated. The central cluster is updated according to the average clustering error, and it is determined whether the central cluster meets the convergence condition after the update. If the convergence condition is not met, the central cluster is continued to be updated. If the convergence condition is met, the clustering result is output.

[0077] The feature fusion logic includes:

[0078] Converting the external structural features and the internal point cloud features into a feature sequence, wherein the feature sequence includes an image point cloud feature sequence and a laser scanning point cloud feature sequence;

[0079] Constructing geometric feature extraction module, surface feature extraction module, internal feature extraction module and spatial feature extraction module;

[0080] Processing the image point cloud feature sequence by the geometric feature extraction module and the surface feature extraction module to extract geometric features and surface features;

[0081] Processing the laser scanning point cloud feature sequence by the internal feature extraction module and the spatial feature extraction module to extract internal features and spatial features;

[0082] The geometric features, the surface features, the internal features and the spatial features are mapped into a tuple order, and fusion features are calculated through pooling operations and residual connections.

[0083] The surface feature extraction module is used to extract and integrate surface shape features and surface structure features from the image point cloud feature sequence, and the surface feature extraction module includes a surface shape feature extraction submodule, a surface structure feature extraction submodule and a feature cascade submodule;

[0084] The geometric feature extraction module is used to identify and extract geometric shape features from the image point cloud feature sequence, and the geometric feature extraction module includes a geometric shape detection submodule, a normal vector calculation submodule and a curvature estimation submodule;

[0085] The spatial feature extraction module is used to extract the spatial layout features of the object from the laser scanning point cloud feature sequence, and the spatial feature extraction module includes a spatial encoding submodule, a spatial transformation submodule and a spatial feature aggregation submodule;

[0086] The internal feature extraction module is used to extract the internal material distribution features of the object from the laser scanning point cloud feature sequence.

[0087] The calculation formula of the fusion feature is:

[0088] F fu =b{Con[En(T1),En(T2),...,En(T r )]},

[0089] Among them, F fu represents fusion features, Con[·] represents residual connection function, En(·) represents local enhancement function, T r Represents the rth feature group in the tuple sequence.

[0090] A method for maintenance of transfer equipment based on digital twin technology, the method comprising:

[0091] Constructing a three-dimensional virtual model of a substation, wherein the three-dimensional virtual model of the substation includes a physical space, a virtual space, and a digital service space;

[0092] Determine an object to be modeled according to the physical space, and obtain sensor data of the object to be modeled through a sensor component;

[0093] Processing the sensor data, constructing a three-dimensional model of a substation, and filling the three-dimensional model of the substation into the virtual space;

[0094] Mapping the physical space and the virtual space to the digital service space, and displaying the operation status, fault location information and test and maintenance plan of the equipment in real time through the digital service space;

[0095] A maintenance instruction is generated according to the test and detection scheme, and a maintenance device is controlled by the maintenance instruction to perform maintenance on the transfer equipment.

[0096] Compared with the prior art, the present invention has the following beneficial effects:

[0097] 1. The present invention achieves a high degree of synchronization between physical space and virtual space by constructing a three-dimensional virtual model including physical space, virtual space and digital service space. The present invention can obtain device data in the physical space in real time through sensor components, and integrate these data into the virtual space to generate an accurate three-dimensional model. This three-dimensional model can not only display the geometric shape of the equipment, but also reflect the operating status of the equipment, providing comprehensive visualization support for equipment maintenance and troubleshooting.

[0098] 2. The present invention acquires external images and point cloud data through image sensors and laser scanners, extracts external structural features and internal point cloud features, generates a three-dimensional model through feature fusion and modeling software, and finally performs fault troubleshooting and early warning to achieve efficient monitoring and maintenance of substation transfer equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0100] Figure 1 This is a flow chart of a method for maintenance of transfer equipment based on digital twin technology in Embodiment 1 of the present invention;

[0101] Figure 2 This is a schematic diagram of the structure of a three-dimensional virtual model of a substation in Embodiment 1 of the present invention;

[0102] Figure 3 The flowchart of the external structure feature extraction disclosed in the present invention;

[0103] Figure 4 The present invention discloses a laser scanning point cloud voxel feature clustering flow chart;

[0104] Figure 5 It is a grid clustering result diagram disclosed in the present invention;

[0105] Figure 6 It is a feature fusion network structure diagram disclosed in the present invention;

[0106] Figure 7 It is a structural diagram of the face feature extraction module disclosed in the present invention;

[0107] Figure 8 It is a structural diagram of the geometric feature extraction module disclosed in the present invention;

[0108] Fig. 9 This is a structural diagram of the spatial feature extraction module disclosed in the present invention;

[0109] Fig.10 This is a structural schematic diagram of a transfer equipment maintenance system based on digital twin technology in the present invention. DETAILED DESCRIPTION

[0110] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0111] Example 1

[0112] See also Figure 1 , an embodiment provided by the present invention: a method for maintenance of transfer equipment based on digital twin technology, the specific steps of the method are as follows:

[0113] S1: construct a three-dimensional virtual model of the substation and determine the object to be modeled, wherein the three-dimensional virtual model of the substation includes a physical space, a virtual space and a digital service space;

[0114] S2: acquiring an external image of the object to be modeled through an image sensor, processing the external image, and extracting external structural features of the object to be modeled;

[0115] S3: acquiring a laser scanning point cloud of the object to be modeled by a laser scanner, processing the laser scanning point cloud, and extracting internal point cloud features of the object to be modeled;

[0116] S4: constructing a feature fusion network, taking the external structural features and the internal point cloud features as input parameters of the feature fusion network, training the input parameters through the feature fusion network, and calculating fusion features;

[0117] S5: importing the fusion features into the modeling software for processing, outputting the three-dimensional model of the object to be modeled, merging all the three-dimensional models, constructing a three-dimensional model of the substation, and filling the three-dimensional model of the substation into the virtual space;

[0118] S6: Mapping the physical space and the virtual space to the digital service space, and monitoring the operating status of the transfer equipment of the substation through the digital service space;

[0119] S7: According to the monitoring results, check the suspected faulty equipment, locate the number and position of the faulty equipment, display it in the digital service space, and formulate a maintenance plan;

[0120] S8: Convert the maintenance plan into maintenance instructions, and control the maintenance device to perform maintenance on the transfer equipment through the maintenance instructions.

[0121] See also Figure 2 , a schematic diagram of the structure of a three-dimensional virtual model of a substation in an embodiment of the present invention, wherein the physical space includes actual physical equipment in the substation, and the virtual space is a digital representation of the physical space, including virtual objects and a virtual environment. The virtual space is based on computer technology and is a digital space constructed from different granularities according to application needs, i.e., a digital mapping of the physical equipment of the substation in a computer environment, and describes the geometric structural features and external behavioral performance of the physical equipment objects through computers. Therefore, the virtual entity not only includes three-dimensional geometric models of physical equipment objects, such as three-dimensional models of transformers, three-dimensional models of high-voltage switchgear, three-dimensional models of busbar equipment, three-dimensional models of capacitors, three-dimensional models of distribution devices, and three-dimensional models of cable equipment, but also includes geometric constraint models, dynamic functional behavior models, and fault behavior models. The digital service space is used to provide digital services such as real-time monitoring, data analysis, fault diagnosis, and maintenance guidance;

[0122] The physical space can be regarded as a perceptible object, which is the target object of maintenance digitization. Mastering the physical space is the premise of realizing three-dimensional digitization. The physical space includes physical objects and sensor components. The physical objects include transformers, high-voltage switchgear, busbar equipment, capacitors, distribution devices and cable equipment. The sensor components include image sensors, laser scanners and monitoring sensors. The transformer includes a main transformer and an auxiliary transformer. The main transformer includes a casing, a cooling system, a radiator and an insulating oil tank. The auxiliary transformer is used as an auxiliary power transformer for the power supply system. The high-voltage switchgear includes a circuit breaker, a disconnector and a load switch. The circuit breaker is used to cut off or close the circuit, and its casing, operating mechanism, and internal structure need to be inspected. The disconnector is used to isolate the circuit, and the modeling should include its knife switch, transmission mechanism and other components. The load switch is used to carry and cut off the load current, and the modeling should include its housing and operating mechanism. The busbar equipment is used to transmit high-voltage current conductors, and the modeling should include busbar support insulators, connection clamps and fixtures. The capacitor includes capacitor banks and reactors. The capacitor bank is used for reactive power compensation, and the modeling should include capacitors, connecting wires and insulation supports. The reactor is used for current limiting or filtering, and the modeling should include reactor coils and iron cores. The distribution device is used to distribute electricity, and the modeling should include the distribution device housing, internal switches, protection devices and terminals. The cable equipment includes high-voltage and low-voltage cables;

[0123] In the process of building digital twins, the physical space is used as the object of data collection. Through devices such as image sensors and laser scanners, the external and internal features of the equipment in the physical space are obtained. These feature data are the basis for subsequent modeling and analysis. The precise description of the physical space helps to ensure the accuracy of the data and provide reliable data support for the construction of the virtual space.

[0124] See also Figure 3 , the external structure feature extraction flow chart of the embodiment of the present invention, the specific steps of S2 are as follows:

[0125] S2.1: Preprocessing the external image of the object to be modeled, wherein the preprocessing includes image smoothing and edge enhancement, and compressing the preprocessed image by an image compression algorithm to generate a compressed reconstructed image;

[0126] By smoothing and edge enhancement, the noise is reduced and the contrast of the image is enhanced. The compression algorithm should ensure that the key structural information in the image is preserved as much as possible while reducing the image file size.

[0127] S2.2: Perform camera calibration on the image sensor to obtain the extrinsic parameter matrix and intrinsic parameter matrix of the image sensor. Camera calibration is a technology that obtains the intrinsic and extrinsic parameters of a camera through multiple images at different angles. The intrinsic parameter matrix describes the intrinsic properties of the camera, such as focal length, optical center position, etc. The extrinsic parameter matrix describes the position and direction of the camera in the world coordinate system. The compressed and reconstructed image is subjected to coordinate transformation according to the extrinsic and intrinsic parameter matrices, and the pixels of the compressed and reconstructed image are transferred from the camera coordinate system to the world coordinate system. Through the coordinate transformation, the two-dimensional pixel position in the image is associated with the three-dimensional coordinate in the real physical space, providing more accurate spatial positioning;

[0128] S2.3: Calculate the global mean of all pixels of the compressed and reconstructed image after coordinate transformation, extract the local feature map, perform weighted sampling on the local feature map through the upsampling function, superimpose the weighted sampled local feature map with the compressed and reconstructed image, calculate the global feature map, and compare the confidence of the pixels of the global feature map with the confidence threshold, retain the pixels greater than the confidence threshold, obtain the surface feature map, and enhance the representation ability of image features. The global mean calculation captures the overall features of the image, while the local feature extraction focuses on the details in the image. By superimposing the compressed and reconstructed image and the local feature map, the importance of local features can be enhanced globally, thereby providing a more comprehensive feature representation for subsequent feature fusion;

[0129] S2.4: Perform regional cutting on the surface feature map, divide the feature map into multiple small regions, each small region represents a different structural part in the image, and then use the time feature extraction function to extract the time dimension features of each region to obtain the change information at different time points. Then, use the attention mechanism to extract the spatial features of each region. The attention mechanism can focus on the key areas in the image and extract representative spatial information. The regional cutting operation makes the feature extraction more refined and can capture the features of different parts of the image. The time feature extraction function provides dynamic information of each region, and the attention mechanism ensures the accurate extraction of spatial features. Through this multi-dimensional feature extraction, the model can better understand the complex structure and dynamic changes of the image;

[0130] S2.5: After obtaining the temporal and spatial features of each region, use the bilinear pooling operation to fuse these features. Bilinear pooling achieves a full combination of temporal and spatial features by calculating the outer product of the two features, generating bilinear regional features. Then, the pooling function is used to comprehensively process the bilinear regional features of all regions to generate external structural features. The feature fusion process fully integrates the feature information of different dimensions, so that the external structural features can fully reflect the structural details and overall morphology in the image. The pooling operation further refines these features, removes redundant information, and enhances the model's discrimination ability. The calculation formula of the external structural features is:

[0131]

[0132] Among them, Gb fd represents the external structural features, a represents the cut area of ​​a single surface feature map, A represents the total number of cut areas of the surface feature map, b{·} represents the pooling function, T f (·) represents the time feature extraction function, hb fi represents the surface feature map, * represents the fusion operation, Ex[·] represents the excitation function in the attention mechanism, W represents the horizontal length of the surface feature map, H represents the vertical length of the surface feature map, n represents the unit horizontal length of the surface feature map, m represents the unit vertical length of the surface feature map, v a Represents the convolution kernel corresponding to the cut area.

[0133] The main purpose of compression processing is to reduce redundant information of the image, reduce the overhead of data storage and processing, and maintain the integrity of key feature information. This step effectively improves the efficiency of subsequent image processing and provides efficient input data for feature extraction and 3D modeling. The specific steps of S2.1 are as follows:

[0134] S2.1.1: De-noising the surface image, wherein the noise includes on-site light noise, operation noise of each unit itself, circuit switching noise, and image sensor photoelectric tube noise;

[0135] S2.1.2: Perform sparse representation on the denoised surface image and perform mask sampling through Fourier transform to obtain a sparse surface image;

[0136] S2.1.3: Cut the sparse surface image, cut the original image of size h*w into blocks of size p*q, obtain the block image matrix, construct the observation matrix sequence, match different sampling rates according to the composition complexity of different regions of the block image, and perform sparse sampling compression in the Fourier domain in a distributed multi-point collaborative manner;

[0137] S2.1.4: Divide the preprocessed image into segmented images with row length = column length = γ according to matrix binarization, obtain the centroid position of the segmented image, calculate the segmented image box dimension according to the centroid position, and compare the segmented image box dimension with the background threshold;

[0138] S2.1.5: If it is greater than or equal to the background threshold, keep the segmented image; if it is less than the background threshold, delete the segmented image, traverse all segmented images, fuse the remaining segmented images, and obtain the compressed reconstructed image, where γ represents the length of rows and columns.

[0139] The specific steps of S3 are as follows:

[0140] S3.1: preprocessing the laser scanning point cloud according to the laser scanning point cloud of the internal parts of the object to be modeled;

[0141] S3.2: During the point cloud data acquisition process, the posture and position of the object to be modeled are different in different scanning angles. According to the least squares method, the direction and position of the plane are adjusted to minimize the sum of the squares of the distances from all points to the plane. The most suitable three-dimensional plane is found through plane fitting. The pre-processed three-dimensional point cloud data is converted to convert the laser scanning point cloud from the radar coordinate system to the world coordinate system. Under different scanning angles, the surface point cloud data is uniformly represented in the same coordinate system, which provides spatial consistency for subsequent data processing and feature extraction.

[0142] S3.3: Divide the world coordinate system into grids at equal intervals, voxelize the laser scanning point cloud, project the voxel index of the laser scanning point cloud to the corresponding grid, connect the points in the voxel, strengthen the local feature information, and obtain the voxel features of the laser scanning point cloud. The voxelization process divides the space into uniform small cube units, and gathers the points in the same voxel into a voxel point. This representation method not only reduces the amount of data, but also strengthens the local feature information, so that the detailed features of the object to be modeled can be better preserved and represented.

[0143] S3.4: Perform preliminary clustering of the voxel features of the laser scanning point cloud according to the DBSCAN clustering algorithm to obtain core points, and determine the number of cluster center clusters by the number of core points. DBSCAN classifies points in high-density areas into the same category by analyzing the density of points, and can effectively identify and ignore noise points. Determine the number of preliminary cluster center clusters by the number of core points. The clustering results of this step help identify the main areas of the internal structure of the object to be modeled, laying the foundation for subsequent accurate clustering;

[0144] S3.5: Use the core points after DBSCAN clustering as the initial center cluster of the K-means clustering algorithm, calculate the Euclidean distance from the voxel features of the laser scanning point cloud to each of the initial center clusters, use the core points obtained by DBSCAN clustering as the initial center cluster of the K-means algorithm, and use the K-means algorithm to further optimize the clustering results. K-means performs clustering by calculating the Euclidean distance from the voxel features to the initial center cluster. The selection of the initial center cluster has a great influence on the final convergence effect of the K-means algorithm. By using the DBSCAN clustering results as the initial point, the convergence speed and clustering quality of the K-means algorithm can be improved;

[0145] S3.6: Perform clustering according to the Euclidean distance, and calculate the average clustering error of the initial central cluster. Update the central cluster according to the average clustering error, and determine whether the central cluster meets the convergence condition after the update. If it does not meet the convergence condition, continue to update the central cluster. If it meets the convergence condition, output the clustering result. In each iteration, the K-means algorithm reallocates each point to the nearest central cluster and updates the position of the central cluster. This process continues to iterate until the average clustering error meets the convergence condition and the clustering result is stable. Through this process, the algorithm can more accurately extract the internal features of the object to be modeled and finally output the clustering result.

[0146] S3.7: Extract the voxel features of the laser scanning point cloud after clustering, and merge the grids where the voxel features of the laser scanning point cloud are located under the same central cluster, ensuring that the feature points belonging to the same category are aggregated together, making the classification of voxel features more accurate. The point cloud features finally extracted can reflect the internal morphological features and potential defects, and extract the internal point cloud features. The calculation formula of the internal point cloud features is:

[0147]

[0148] Among them, T a represents the internal point cloud features corresponding to the voxel features of the laser scanning point cloud in the same central cluster after clustering, i represents a single voxel feature of the laser scanning point cloud in the central cluster, M represents the total number of voxel features of the laser scanning point cloud in the central cluster, and η i represents the eigenvalue of the voxel feature of the i-th laser scanning point cloud, d represents the clustering threshold corresponding to the central cluster, r2 represents the farthest distance scanned by the laser scanner, r1 represents the closest distance scanned by the laser scanner, and p i_x represents the voxel horizontal coordinate of the i-th laser scanning point cloud voxel feature, p i_y The voxel ordinate represents the feature of the i-th laser scanning point cloud voxel, p i_z Represents the voxel vertical coordinate of the i-th laser scanning point cloud voxel feature.

[0149] See also Figure 4 , the flow chart of clustering voxel features of laser scanning point cloud in an embodiment of the present invention, specifically, by combining the two clustering algorithms of DBSCAN and K-means, clustering analysis is performed on the voxel features of laser scanning point cloud, and more accurate clustering results are obtained by iteratively optimizing the central cluster, so as to better understand the working status of the transfer equipment, which is helpful to improve the operating efficiency of the power system and reduce energy loss. DBSCAN is a density clustering algorithm used to identify closely connected data points in space and divide them into clusters. DBSCAN can be used for preliminary clustering to obtain core points. Core points refer to the core area containing at least a specified number of data points within a given radius. Since each core point may represent a cluster, the number of core points can be used to preliminarily determine the number of central clusters of clustering.

[0150] The kernel points obtained by preliminary clustering of DBSCAN are used as the initial center cluster of the K-means clustering algorithm, providing an initial cluster center position for the K-means algorithm; the Euclidean distance is used for clustering, and each work operation information is assigned to the nearest center cluster by calculating the Euclidean distance from the voxel features of the laser scanning point cloud to each initial center cluster. The clustering average error refers to the average value of the Euclidean distance from all the data points to the center. The clustering average error can be used to update the initial center cluster to improve the clustering effect. The position of the center cluster can be adjusted according to the calculated clustering average error, so as to update the center cluster, and the convergence condition is defined to determine whether the updated center cluster meets the condition to determine whether to end the iteration. If the convergence condition is met, the clustering result is output.

[0151] See also Figure 5 , the grid clustering result diagram of the embodiment of the present invention, first traverses the grid map, finds the first unmarked grid containing the laser scanning point cloud voxel feature, marks it as 1, then searches for other grids containing the laser scanning point cloud voxel feature in the neighborhood of the grid, and determines whether the central cluster of the laser scanning point cloud voxel feature is the same as the central cluster of the previous laser scanning point cloud voxel feature. If they are the same, they are also marked as 1. If they are different, continue to search for other grids, repeatedly mark the neighborhood of the grid until all grids in the connected domain are successfully marked, record the positions of all grids in the connected domain, and then repeat the above operation to find the remaining connected domains in the grid map, and mark them as 2 and 3 according to the numerical sequence, and finally obtain a number of connected areas, that is, a number of grids with the same central cluster in the grid map. At this time, the laser scanning point cloud voxel features stored in each grid in the same connected domain are merged to obtain internal point cloud features;

[0152] Laser scanning point cloud data usually contains a large amount of raw point data, which may be affected by noise, external environment and equipment errors, so as to improve the overall quality of point cloud data. This process can effectively reduce the errors caused by external factors, so that the point cloud data can more truly reflect the actual situation of the tire surface. To ensure the accuracy of subsequent analysis, the preprocessing step first removes noise points and abnormal points. The specific steps for preprocessing laser scanning point cloud are as follows:

[0153] S3.1.1: Calculate the y-axis coordinates of all laser scanning point clouds. If the y-axis coordinate of a point is less than 0, delete the point from the laser scanning point cloud.

[0154] S3.1.2: traverse and query the laser scanning point cloud, and delete the laser scanning point cloud data with the same radar rotation angle and echo distance;

[0155] S3.1.3: Set the fitting plane equation, calculate the coefficients of the fitting plane equation according to the laser scanning point cloud, and calculate the distance from all laser scanning point clouds to the fitting plane, compare the distance from the laser scanning point cloud to the fitting plane with the plane distance threshold, if it is less than or equal to the plane distance threshold, save the laser scanning point cloud, if it is greater than the plane distance threshold, delete the laser scanning point cloud;

[0156] S3.1.4: Gather the laser scanning point cloud in the normal direction and adjust the position and coordinates of the sampling points to denoise the laser scanning point cloud.

[0157] See also Figure 6 , a structural diagram of a feature fusion network according to an embodiment of the present invention, wherein the feature fusion network comprises:

[0158] Input layer, establishes feature sequences according to input parameters, the feature sequences include image point cloud feature sequences and laser scanning point cloud feature sequences, and receives external structural features and internal point cloud features as input parameters. These features represent the surface structure and internal structure of the object to be modeled, respectively, and are key data extracted and processed from the previous steps. External structural features reflect the external geometric shape and texture information of the object to be modeled, and internal point cloud features reflect the internal structure and spatial layout of the object to be modeled;

[0159] A feature extraction layer, used to extract feature vectors of feature sequences, the feature extraction layer includes a geometric feature extraction module, a surface feature extraction module, an internal feature extraction module and a spatial feature extraction module, and the feature vectors include surface features, spatial features, internal features and geometric features;

[0160] The feature fusion layer is used to map the geometric features, the surface features, the internal features and the spatial features into a tuple order, and calculate the fusion features through pooling operations and residual connections.

[0161] The training of the input parameters through the feature fusion network to calculate the fusion features specifically includes:

[0162] The input parameters are used to establish a grey differential equation through the input layer. The grey differential equation is a mathematical model used to describe the dynamic characteristics of the system, which is particularly suitable for dealing with uncertainty and insufficient information. The least squares parameters of the grey differential equation are calculated, and the least squares parameter solution is discretized. The continuous solution obtained by the least squares method is discretized to calculate the fitting value of the feature sequence. These fitting values ​​represent the image point cloud feature sequence and laser scanning point cloud feature sequence generated based on the input parameters. The image point cloud feature sequence reflects the changes in the external features of the object to be modeled, and the laser scanning point cloud feature sequence reflects the changes in the internal features.

[0163] The feature extraction layer includes a geometric feature extraction module, a surface feature extraction module, an internal feature extraction module and a spatial feature extraction module.

[0164] See also Figure 7 , a structural diagram of a surface feature extraction module according to an embodiment of the present invention, wherein the surface feature extraction module is responsible for extracting and integrating surface shape features and surface structure features from an image point cloud feature sequence, and realizes efficient extraction of complex three-dimensional surface features through hierarchical processing and high-level cascade operations;

[0165] The face feature extraction module includes a face shape feature extraction submodule, a face structure feature extraction submodule and a feature cascade submodule;

[0166] The surface shape feature extraction submodule includes a multi-layer perceptron network for extracting shape features from the image point cloud feature sequence. The multi-layer perceptron network (MLP) is a feedforward neural network composed of multiple fully connected layers. Each layer of neurons receives the output of the neurons in the previous layer and performs nonlinear transformation. In each layer of MLP, the image point cloud feature sequence is first averaged and pooled. Average pooling is a downsampling method that reduces the size of the feature map and retains important statistical information by calculating the average value of the input features. This operation helps to process features symmetrically and reduce sensitivity to the asymmetry of the input data. After average pooling, a one-dimensional convolution operation is applied to the feature sequence. One-dimensional convolution is used to extract local patterns in the sequence and capture shape changes by sliding windows. The feature sequence after convolution is standardized by normalization to improve the stability and convergence speed of training, and the image point cloud feature sequence is described in shape. Finally, it is activated according to the ReLU activation function to obtain the surface shape feature, which is used to describe the local and global information of the surface shape of the object surface.

[0167] The surface structure feature extraction submodule includes a set of Gaussian kernels for learning. The Gaussian kernel function is a kernel method that can map the input features to a higher-dimensional feature space by projection, thereby capturing more complex structural patterns. Here, the Gaussian kernel function is used to learn the radial distance, azimuth and tilt angle of the image point cloud feature sequence. The Gaussian kernel function operates on the image point cloud feature sequence to calculate the radial distance (the distance from the reference point to the surface point), azimuth (the angle of the surface point relative to the reference axis) and tilt angle (the angle between the surface and the reference plane) of each face. These parameters describe the spatial position and orientation of each face and are important components of the surface structure features. After obtaining the radial distance, azimuth and tilt angle, these features are encoded using convolution operations. The convolution operation can capture the local structural features of each face, including local concavity, surface texture, etc., thereby encoding complex geometric structure information into feature vectors that can be used for subsequent processing, and obtaining surface structure features, which are used to describe the spatial relationship between various parts of the object surface.

[0168] The feature cascade submodule is used to cascade the features obtained from the surface shape feature extraction submodule and the surface structure feature extraction submodule. The cascade operation is to connect two feature vectors along the feature dimension to form a new, higher-dimensional feature vector. This operation retains all the information of the original features and combines them together to provide a more comprehensive feature expression. Finally, the feature cascade submodule outputs the fused surface features, which contain both surface shape features and surface structure features, and can fully describe the shape information and structure information of the object surface.

[0169] See also Figure 8 , a structural diagram of a geometric feature extraction module of an embodiment of the present invention, wherein the geometric feature extraction module is responsible for identifying and extracting basic geometric shape features from an image point cloud feature sequence. These geometric features include planes, curved surfaces, and edges, etc., which constitute the basic geometric structure of the object surface. The function of the geometric feature extraction module is not only to identify these basic shapes, but also to refine them into feature vectors for subsequent processing through precise mathematical and computational methods;

[0170] The geometric feature extraction module includes a geometric shape detection submodule, a normal vector calculation submodule and a curvature estimation submodule;

[0171] The geometric shape detection submodule identifies the basic geometric shapes in the point cloud through RANSAC and least squares fitting to ensure the reliability of geometric feature extraction. In point cloud data, the plane is one of the most common geometric features. This can be achieved through the RANSAC algorithm. RANSAC is an iterative algorithm that finds the plane equation that conforms to the largest point cloud part through continuous sampling and verification. The advantage of RANSAC is that it can still accurately detect plane features in the presence of noise and outliers. For non-planar parts, curves and surfaces are common geometric shapes. For surface detection, least squares fitting is a commonly used technique. This technology finds the best fitting surface by minimizing the sum of the squares of the distance between the point cloud and the fitting surface, and extracts complex surface features in the point cloud.

[0172] The normal vector calculation submodule is used to calculate the normal vector of each point. The consistency of the normal vector field is ensured by smoothing, the influence of noise on geometric analysis is reduced, and the stability of feature extraction is improved. The normal vector is an important tool for describing geometric shapes. In order to calculate the normal vector, it is necessary to first determine the neighborhood of each point, find a preset number of nearest neighbors for each point through the KNN algorithm, and then use these neighborhood points for plane fitting. Then, the covariance matrix of the neighborhood point set is calculated by the PCA algorithm, and the eigenvalue decomposition is performed on it. The PCA method can extract the main direction, that is, the eigenvector. The eigenvector corresponding to the minimum eigenvalue of the eigenvector is the normal vector. The advantage of PCA lies in its sensitivity to the local point cloud structure and its ability to accurately capture subtle changes on complex surfaces.

[0173] The curvature estimation submodule is used to calculate the key parameters that describe the degree of surface curvature. Through discrete methods and normal vector gradient analysis, the curvature estimation submodule can accurately extract surface curvature information and help identify the curvature features and local extreme points of the surface. The combined use of different curvature types enables the curvature estimation submodule to perform a more comprehensive analysis of the surface geometry, thereby improving the shape description ability of the model. By analyzing the change of the normal vector of each point according to the discrete method, the average curvature of the object surface can be calculated. By calculating the directional change rate of the normal vector field, the curvature of the surface in different directions can be determined. Aggregation is performed through a Gaussian function to calculate the Gaussian curvature of the object surface. Gaussian curvature and average curvature are the two most commonly used curvature types. Gaussian curvature reflects the overall curvature of the surface, while average curvature reflects more local features.

[0174] See also Fig. 9, a structural diagram of a spatial feature extraction module in an embodiment of the present invention, wherein the spatial feature extraction module is used to extract the spatial layout features of an object from a laser scanning point cloud feature sequence. These spatial features not only describe the relative position and distance between points in the point cloud data, but also involve higher-level spatial relationships, such as proximity, symmetry, and global spatial structure. By extracting these spatial features, key spatial information can be provided for subsequent 3D modeling, analysis, and application.

[0175] The spatial feature extraction module includes a spatial encoding submodule, a spatial transformation submodule and a spatial feature aggregation submodule.

[0176] The spatial encoding submodule enhances the network's perception of the spatial relationship of the point cloud by introducing clear spatial position information, making feature extraction more accurate and comprehensive. First, the point cloud is normalized. The normalization process includes adjusting the coordinates of the point cloud data to a uniform scale range (usually mapped to a unit cube) to ensure that point clouds from different scans are comparable. After normalization, the point cloud data is further standardized, the center of the point cloud is moved to the origin of the coordinates, and the distribution of the point cloud is adjusted to have a uniform scale. This step eliminates the inconsistency caused by data offset and scale differences, assigns a position code to each point, and maps the position information of the point into a high-dimensional space, so that the model can better perceive and process spatial information. The position encoding method usually uses sine and cosine functions to generate high-dimensional position vectors, which are associated with the spatial layout of the point cloud.

[0177] The spatial transformation submodule performs spatial alignment on the point cloud data. It uses adaptive learning to automatically adjust the point cloud to a standard posture during processing, thereby reducing the impact of changes in perspective and position. A 3x3 affine transformation matrix is ​​learned through a small neural network. This matrix can rotate, translate, scale, and other operations on the input point cloud data, so that the point cloud data is aligned in space. This process reduces spatial deformation while retaining key spatial information in the point cloud data.

[0178] The spatial feature aggregation submodule is used to extract local spatial features from point cloud data, construct a high-dimensional spatial feature representation by analyzing the relative position and distance between points in the neighborhood, and use the KNN algorithm to find the nearest K neighbors for each point to form a local neighborhood. The KNN algorithm calculates the Euclidean distance between points in a high-dimensional space to determine the nearest neighbor point set. The parameter K of KNN can be tuned experimentally to best balance the capture of local details and computational complexity. After finding the neighborhood point set, the feature vector describing the local structure of the point cloud is extracted by analyzing the relative position and distance of these neighborhood points. The specific method includes calculating the centroid, maximum and minimum distance, local density, etc. of the neighborhood point set. These feature vectors encode the microstructure of the point cloud and are the specific manifestation of the local spatial layout. A three-dimensional convolution operation is applied in the three-dimensional space of the point cloud to capture the local spatial pattern by sliding the convolution kernel in space. The 3D convolution network can identify complex features in the local space, such as volume structure, spatial symmetry, etc. Through multi-layer 3D convolutional networks, spatial features at different levels can be extracted, and features of different scales can be organized hierarchically to form a pyramid structure. This structure can effectively fuse spatial features from different scales, allowing the network to focus on local details and global layout at the same time. At the top level of the pyramid, feature fusion technology is used to aggregate features of different scales into a unified spatial feature representation. Common feature fusion methods include weighted averaging, maximum pooling, or using a fully connected layer to integrate features of different scales.

[0179] After multi-view, multi-scale analysis and deep learning model processing, the internal feature extraction module generates a comprehensive feature vector, which represents the internal structural features of the object. This feature vector not only includes local and global geometric information, but also reflects the material distribution and spatial relationship inside the object.

[0180] The feature fusion layer maps spatial features, internal features, surface features and geometric features into a tuple sequence [T(Q1, K1, J1, P1), T(Q2, K2, J2, P2), ..., T(Q r ,K r ,J r ,P r )], where T(·) represents the tuple order, Q r represents the corresponding input of the rth feature in the geometric feature, K r represents the corresponding input of the rth feature in the spatial feature, J r represents the corresponding input of the rth feature in the face feature, P rRepresents the corresponding input of the rth feature in the internal feature, performs dot product on the corresponding input features in each candidate sequence in the tuple sequence, calculates the correlation score after the dot product of the candidate sequence by the Pearson correlation coefficient, compresses the correlation score according to the sigmoid function, and compares it with the correlation threshold. If it is greater than the correlation threshold, the candidate sequence is retained. If it is less than or equal to the correlation threshold, the candidate sequence is filtered, and the local features of the candidate sequence retained in the tuple sequence are collected and enhanced. The enhanced candidate sequences are spliced ​​along the head sequence of the tuple sequence, and the local features are fused through residual connection. The feature vector after residual connection fusion is pooled to obtain the fused feature. The calculation formula of the fused feature is:

[0181] F fu =b{Con[En(T1),En(T2),...,En(T r )]},

[0182] Among them, F fu represents fusion features, Con[·] represents residual connection function, En(·) represents local enhancement function, T r Represents the rth feature group in the tuple sequence;

[0183] The S5 imports the fusion features into the modeling software for processing, finally outputs the three-dimensional model of the object to be modeled, and merges all the three-dimensional models to construct a complete three-dimensional model of the substation;

[0184] Before importing the fused features into the modeling software, it is first necessary to convert these features into a data format that the modeling software can recognize. The converted fused feature data is imported into the modeling software. During the import process, ensure that the scale, unit and coordinate system of the model are consistent with the settings of the modeling software to avoid data distortion or dislocation. After completing the merging of the various models, integrate these models into a complete substation scene. Scene integration includes the layout, positioning and scaling of the model so that each component is in the correct position. The scene management tool of the modeling software can be used to organize and control the relationship and interaction between different models. During the scene construction process, physical properties such as material, light reflectivity, collision properties, etc. are assigned to different model components. These physical properties are crucial for simulation and interaction in virtual space. By assigning the correct physical properties, the model not only has visual realism, but also can correctly respond to environmental changes and user operations in the virtual space. After building a complete substation 3D model, an integrity check and verification is performed. This includes checking the model for defects, overlaps or inconsistencies, and adjusting and correcting them through repair tools. The integrity check ensures the quality and accuracy of the final model so that it can be successfully applied in the virtual space.

[0185] Before filling the substation 3D model into the virtual space, it is necessary to set the environmental parameters of the virtual space. This includes setting the background, light source, terrain, etc. of the virtual space so that the 3D model matches the virtual environment. By adjusting the environmental parameters, a more realistic virtual scene can be created, and the generated substation 3D model can be imported into the virtual space. During the import process, adjust the position, angle and scale of the model to ensure its correct placement in the virtual space. Virtual reality tools can be used to preview and adjust the position of the model for the best display effect, and add interactive functions to the substation 3D model in the virtual space. This may include functions such as clicking to view detailed information, dynamically simulating equipment operation, and remote operation. By integrating interactive functions, users can interact with the 3D model in the virtual space for simulation testing, operation training or display.

[0186] The digital service space visualizes sensor data from the physical space and the three-dimensional model of the virtual space through the Internet of Things platform. The digital service space includes equipment operation status monitoring, troubleshooting, fault warning and maintenance guidance. The equipment operation status monitoring includes temperature monitoring, vibration monitoring, pressure monitoring, current monitoring and flow monitoring. The equipment operation status monitoring also includes real-time analysis of all monitoring data, identification of abnormal data patterns, and calculation of analysis results. The troubleshooting is used to predict faults based on historical data and analysis results through artificial intelligence algorithms. The fault warning is used to automatically issue a warning signal when signs of failure appear in the transfer equipment, reminding the staff to perform testing, inspection and maintenance work. The maintenance guidance is used to generate maintenance operation guidance, including the cause of the failure, recommended repair steps and spare parts requirements.

[0187] The temperature monitoring is used to monitor the temperature data of the transfer equipment according to the temperature sensor installed in the physical space, clean and preprocess the collected temperature data, draw a frequency distribution histogram of the temperature data, and output the temperature status analysis result of the transfer equipment according to the frequency distribution histogram.

[0188] The pressure monitoring is used to monitor the pressure data of the transfer equipment based on the pressure sensor installed in the physical space, and pre-process the pressure data, including filtering and denoising, data smoothing and outlier detection to ensure the accuracy of subsequent analysis. Through the time series analysis method, it is detected whether there is an abnormal pressure increase or decrease trend, and the pressure signal is decomposed into different frequency components through wavelet analysis, so as to detect high-frequency abnormal fluctuations.

[0189] The current monitoring is used to collect the current signal in the power system of the transfer equipment in real time based on the current sensor installed in the physical space, such as the Hall effect sensor or the shunt, rectify and smooth the current signal to obtain a stable DC signal, and perform frequency domain analysis on the current signal through short-time Fourier transform to identify different working states in the system operation. The low-frequency component in the current may correspond to the normal working state, while the high-frequency component may indicate abnormal load fluctuations in the system. By setting the detection threshold of the instantaneous current, the system can quickly identify the current anomaly that may damage the equipment and take corresponding protective measures. During the operation of the equipment, it is monitored in real time and it is determined whether there is current anomaly, thereby avoiding potential risks such as equipment overload or short circuit.

[0190] The flow monitoring is used to collect flow data in the fluid system of the transfer equipment based on flow sensors installed in the physical space, such as turbine flowmeters, electromagnetic flowmeters and ultrasonic flowmeters. In order to ensure the accuracy of the data, the output of the flow sensor is first calibrated, including temperature compensation and zero drift correction. By calculating the input-output balance of the flow, it is possible to detect whether there is leakage or blockage in the system. For example, by comparing the readings of different flow meters, abnormal flow changes in a certain part of the pipeline or equipment can be identified, thereby inferring potential problems, and predicting future flow trends through time series prediction models, and identifying possible abnormal situations in advance. For example, if the predicted flow value exceeds or falls below the preset safety range, the system can issue a warning in advance and initiate corresponding protective measures.

[0191] The vibration monitoring is used to collect vibration signals during the operation of the transfer equipment according to a vibration sensor installed in the physical space, perform adaptive signal decomposition on the vibration signal, calculate the kurtosis value of each component, sort the components in descending order according to the kurtosis value, select the top ten components for signal reconstruction, eliminate the vibration signal noise, perform Fourier transform on the reconstructed vibration signal, divide the frequency domain boundary of the Fourier spectrum of the vibration signal by scale space transformation, divide the Fourier spectrum of the vibration signal into frequency bands with different center frequencies, each frequency band corresponds to an intrinsic mode function, iterate each intrinsic mode function, and calculate the intrinsic mode function that satisfies each mode. The optimal solution when the bandwidth sum of the modal components is minimized is obtained, the power spectrum is calculated according to the optimal solution of each modal component, the characteristic frequency of the rolling bearing signal is obtained, the component with the largest energy in the power spectrum is selected as the rolling bearing inner ring signal, the component with the second largest energy in the power spectrum is selected as the rolling bearing outer ring signal, the component with the third largest energy in the power spectrum is selected as the rolling bearing rolling element signal, and the component with the second largest energy in the power spectrum is selected as the rolling bearing cage signal. The characteristic frequency of the rolling bearing signal is compared with the theoretical frequency of the rolling bearing, and the vibration signal analysis result is output, wherein the vibration signal analysis result includes the normal state of the bearing, the fault state of the bearing and the abnormal state of the bearing.

[0192] Example 2

[0193] See also Fig.10 , the present invention provides an embodiment: a transfer equipment maintenance system based on digital twin technology, the system comprising an equipment virtual module, an equipment operation status monitoring module and a test maintenance control module;

[0194] The equipment virtualization module is used to perform three-dimensional visualization of the transfer equipment in the substation, and the equipment virtualization module is configured with a data processing strategy and a three-dimensional modeling strategy;

[0195] The equipment operation status monitoring module is used to monitor the transfer equipment through a visual interface, and to perform real-time analysis on the monitoring data, identify abnormal data patterns, and calculate analysis results;

[0196] The test and maintenance control module is used to generate a test and maintenance plan according to the analysis result, convert the test and maintenance plan into a machine control instruction, and output the machine control instruction;

[0197] The test and maintenance control module is configured with a maintenance control strategy, which analyzes the analysis results according to an intelligent decision-making algorithm, generates a test and maintenance plan, and generates maintenance instructions through a control algorithm, and controls the maintenance device to perform maintenance on the transfer equipment through the maintenance instructions.

[0198] The equipment operation status monitoring module comprises:

[0199] Equipment monitoring unit: used to monitor the transfer equipment through sensors and obtain the monitoring data of the equipment;

[0200] Real-time data analysis unit: performs real-time analysis on all monitoring data, identifies abnormal data patterns and calculates analysis results;

[0201] The equipment operation status monitoring module is configured with a data analysis strategy, which includes temperature analysis logic, vibration analysis logic, pressure analysis logic, current analysis logic and flow analysis logic;

[0202] The temperature analysis logic is used to analyze the temperature data collected by the temperature sensor to determine whether there is an overheating fault;

[0203] The pressure analysis logic is used to analyze the pressure data collected by the pressure sensor, identify the trend and fluctuation of pressure over time, and determine whether there is overpressure and pressure relief;

[0204] The flow analysis logic is used to analyze the flow data collected by the turbine flow meter to determine whether there is a flow exceeding the limit;

[0205] The temperature analysis logic, vibration analysis logic, pressure analysis logic, current analysis logic and flow analysis logic are configured in the real-time data analysis unit.

[0206] The vibration analysis logic is used to analyze the vibration signal collected by the vibration sensor to determine whether there is abnormal vibration of the equipment bearing, specifically including:

[0207] Reconstruct the vibration signal;

[0208] Performing Fourier transform on the reconstructed vibration signal, dividing the Fourier spectrum of the vibration signal into frequency bands with different center frequencies;

[0209] Iterate each frequency band and calculate the optimal solution that satisfies the minimum sum of bandwidths of each modal component.

[0210] The power spectrum is calculated according to the optimal solution of each modal component, the characteristic frequency of the bearing signal of the transfer equipment is obtained, the characteristic frequency of the bearing signal of the transfer equipment is compared with the theoretical frequency of the bearing signal of the transfer equipment, and the vibration signal analysis result is output.

[0211] The current analysis logic is used to analyze the current signal collected by the current sensor to determine whether there is load startup and current overload, specifically including:

[0212] The current signal is rectified and smoothed, and the frequency domain analysis of the current signal is performed through short-time Fourier transform to extract high-frequency and low-frequency components;

[0213] Compare the high frequency component of the current signal with the current detection threshold. If it is greater than or equal to the current detection threshold, it indicates that there is a current overload.

[0214] The low-frequency component of the current signal is compared with the current initial threshold value. If it is greater than or equal to the current initial threshold value, it indicates that a load is started.

[0215] The test and maintenance control module comprises:

[0216] A solution generation unit is used to process the analysis results and generate a minimum cost test and maintenance solution;

[0217] A scheme execution unit, used to convert the test and maintenance scheme into a machine instruction executable by the system, and control the maintenance device to perform maintenance on the transfer equipment through the machine instruction;

[0218] The execution feedback unit is used to monitor the operating status of the transfer equipment in real time and compare it with the expected maintenance results. According to the real-time feedback results, the maintenance control strategy and machine control instructions are automatically adjusted;

[0219] The maintenance control strategy includes an optimal solution generation logic and a control logic. The optimal solution generation logic is configured in the solution generation unit, and the control logic is configured in the solution execution unit.

[0220] The optimal solution generation logic includes:

[0221] Construct a genetic algorithm and define the objective function and constraints;

[0222] The analysis results and the equipment operation status are used as the genetic algorithm input parameters, the input parameters are iteratively optimized by the genetic algorithm, and the optimal solution of the objective function is output;

[0223] The optimal solution of the objective function is used as the test and maintenance plan.

[0224] The device virtual module includes:

[0225] A three-dimensional virtual model building unit, used to build a three-dimensional virtual model of the substation and perform data mapping and interaction between the physical space, virtual space and digital service space;

[0226] A sensor data acquisition unit, used to obtain an external image of the object to be modeled and a laser scanning point cloud of the object to be modeled through a variety of sensor components;

[0227] The data processing unit is used to process the external data and internal data of the object to be modeled and to perform three-dimensional visualization of the transfer equipment in the substation.

[0228] The three-dimensional virtual model building unit comprises:

[0229] Physical space subunit: includes physical equipment and its layout, and collects physical data of equipment through sensors;

[0230] Virtual space subunit: digital representation of the equipment and environment of the physical space, including the 3D model of the equipment, the virtual environment and simulation data;

[0231] Digital service space subunit: used to provide digital services for real-time monitoring, data analysis and maintenance guidance;

[0232] The data processing strategy includes image processing logic and point cloud processing logic, wherein the image processing logic is used to extract surface features of the transfer equipment through image processing technology, and the point cloud processing logic is used to extract internal features of the transfer equipment;

[0233] The data processing unit comprises:

[0234] Image processing subunit: processes external data and extracts external structural features;

[0235] Point cloud processing subunit: processes internal data and extracts internal point cloud features;

[0236] Feature fusion subunit: fuses external structural features and internal point cloud features to generate a 3D model of the object to be modeled;

[0237] Model merging subunit: merge multiple 3D models to build a complete 3D model of the substation;

[0238] The three-dimensional modeling strategy includes feature fusion logic, and the feature fusion logic is used to fuse external structural features and internal point cloud features.

[0239] The image processing logic is configured in the image processing subunit, and the point cloud processing logic is configured in the point cloud processing subunit;

[0240] The feature fusion logic is configured in the feature fusion subunit.

[0241] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A transfer equipment maintenance system based on digital twin technology, characterized in that: The system includes an equipment virtual module, an equipment operation status monitoring module and a test and maintenance control module; The equipment virtualization module is used to perform three-dimensional visualization of the transfer equipment in the substation, and the equipment virtualization module is configured with a data processing strategy and a three-dimensional modeling strategy; The equipment operation status monitoring module monitors the transfer equipment through a visual interface, and performs real-time analysis on the monitoring data, identifies abnormal data patterns, and calculates the analysis results; The test and maintenance control module generates a test and maintenance plan according to the analysis result, converts the test and maintenance plan into a machine control instruction, and outputs the machine control instruction; The test and maintenance control module is configured with a maintenance control strategy, which analyzes the analysis results according to an intelligent decision-making algorithm, generates a test and maintenance plan, and generates maintenance instructions through a control algorithm, and controls the maintenance device to perform maintenance on the transfer equipment through the maintenance instructions.

2. According to claim 1, a transfer equipment maintenance system based on digital twin technology is characterized in that: The equipment operation status monitoring module comprises: Equipment monitoring unit: monitors the transfer equipment through sensors and obtains equipment monitoring data; Real-time data analysis unit: performs real-time analysis on all monitoring data, identifies abnormal data patterns and calculates analysis results; The equipment operation status monitoring module is configured with a data analysis strategy, which includes temperature analysis logic, vibration analysis logic, pressure analysis logic, current analysis logic and flow analysis logic; The temperature analysis logic is used to analyze the temperature data collected by the temperature sensor to determine whether there is an overheating fault; The pressure analysis logic is used to analyze the pressure data collected by the pressure sensor, identify the trend and fluctuation of pressure over time, and determine whether there is overpressure and pressure relief; The flow analysis logic is used to analyze the flow data collected by the turbine flow meter to determine whether there is a flow exceeding the limit; The temperature analysis logic, vibration analysis logic, pressure analysis logic, current analysis logic and flow analysis logic are configured in the real-time data analysis unit.

3. According to claim 2, a transfer equipment maintenance system based on digital twin technology is characterized in that: The vibration analysis logic is used to analyze the vibration signal collected by the vibration sensor to determine whether there is abnormal vibration of the equipment bearing, specifically including: Reconstruct the vibration signal; Performing Fourier transform on the reconstructed vibration signal, dividing the Fourier spectrum of the vibration signal into frequency bands with different center frequencies; Iterate each frequency band and calculate the optimal solution that satisfies the minimum sum of bandwidths of each modal component. The power spectrum is calculated according to the optimal solution of each modal component, the characteristic frequency of the bearing signal of the transfer equipment is obtained, the characteristic frequency of the bearing signal of the transfer equipment is compared with the theoretical frequency of the bearing signal of the transfer equipment, and the vibration signal analysis result is output.

4. According to claim 2, a transfer equipment maintenance system based on digital twin technology is characterized in that: The current analysis logic analyzes the current signal collected by the current sensor to determine whether there is load startup and current overload, specifically including: The current signal is rectified and smoothed, and the frequency domain analysis of the current signal is performed through short-time Fourier transform to extract high-frequency and low-frequency components; Compare the high frequency component of the current signal with the current detection threshold. If it is greater than or equal to the current detection threshold, it indicates that there is a current overload. The low-frequency component of the current signal is compared with the current initial threshold value. If it is greater than or equal to the current initial threshold value, it indicates that a load is started.

5. According to claim 1, a transfer equipment maintenance system based on digital twin technology is characterized in that: The test and maintenance control module comprises: A solution generation unit is used to process the analysis results and generate a minimum cost test and maintenance solution; A scheme execution unit, used to convert the test and maintenance scheme into a machine instruction executable by the system, and control the maintenance device to perform maintenance on the transfer equipment through the machine instruction; The execution feedback unit is used to monitor the operating status of the transfer equipment in real time and compare it with the expected maintenance results. According to the real-time feedback results, the maintenance control strategy and machine control instructions are automatically adjusted; The maintenance control strategy includes an optimal solution generation logic and a control logic. The optimal solution generation logic is configured in the solution generation unit, and the control logic is configured in the solution execution unit.

6. According to claim 5, a transfer equipment maintenance system based on digital twin technology is characterized in that: The optimal solution generation logic includes: Construct a genetic algorithm and define the objective function and constraints; The analysis results and the equipment operation status are used as the genetic algorithm input parameters, the input parameters are iteratively optimized by the genetic algorithm, and the optimal solution of the objective function is output; The optimal solution of the objective function is used as the test and maintenance plan.

7. According to claim 1, a transfer equipment maintenance system based on digital twin technology is characterized in that: The device virtual module includes: A three-dimensional virtual model building unit, used to build a three-dimensional virtual model of the substation and perform data mapping and interaction between the physical space, virtual space and digital service space; A sensor data acquisition unit, which acquires an external image of the object to be modeled and a laser scanning point cloud of the object to be modeled through a variety of sensor components; The data processing unit is used to process the external data and internal data of the object to be modeled and to perform three-dimensional visualization of the transfer equipment in the substation.

8. According to claim 7, a transfer equipment maintenance system based on digital twin technology is characterized in that: The three-dimensional virtual model building unit comprises: Physical space unit: includes physical equipment and its layout, and collects physical data of equipment through sensors; Virtual space unit: digital representation of the equipment and environment of the physical space, including the 3D model of the equipment, the virtual environment and simulation data; Digital service space unit: used to provide digital services such as real-time monitoring, data analysis and maintenance guidance.

9. According to claim 8, a transfer equipment maintenance system based on digital twin technology is characterized in that: The data processing unit comprises: Image processing subunit: processes external data and extracts external structural features; Point cloud processing subunit: processes internal data and extracts internal point cloud features; Feature fusion subunit: fuses external structural features and internal point cloud features to generate a 3D model of the object to be modeled; Model merging subunit: merge multiple 3D models to build a complete 3D model of the substation; The data processing strategy includes image processing logic and point cloud processing logic, wherein the image processing logic extracts surface features of the transfer equipment through image processing technology, and the point cloud processing logic is used to obtain internal features of the transfer equipment; The three-dimensional modeling strategy includes feature fusion logic, and the feature fusion logic is used to fuse external structural features and internal point cloud features; The image processing logic is configured in the image processing subunit, and the point cloud processing logic is configured in the point cloud processing subunit; The feature fusion logic is configured in the feature fusion subunit.

10. The transfer equipment maintenance system based on digital twin technology according to claim 9 is characterized in that: The image processing logic comprises: Compressing the external image of the object to be modeled to obtain a compressed reconstructed image; Perform camera calibration on the image sensor, obtain the extrinsic parameter matrix and intrinsic parameter matrix of the image sensor, perform coordinate transformation on the compressed reconstructed image according to the extrinsic parameter matrix and the intrinsic parameter matrix, and transform the pixels of the compressed reconstructed image from the camera coordinate system to the world coordinate system; Calculate the global mean of all pixels of the compressed reconstructed image after coordinate transformation, perform weighted sampling through the upsampling function, and calculate the global feature map; Compare the confidence of the pixel points of the global feature map with the confidence threshold, retain the pixel points greater than the confidence threshold, and obtain the surface feature map; Performing regional cutting on the surface feature map, extracting the regional temporal features after the surface feature map is cut according to the temporal feature extraction function, and extracting the regional spatial features after the surface feature map is cut through the attention mechanism; The regional temporal features and regional spatial features of the surface feature map are fused to obtain the bilinear regional features. The bilinear regional features of all regions are integrated through the pooling function to calculate the external structural features.

11. The transfer equipment maintenance system based on digital twin technology according to claim 10, characterized in that: The calculation formula of the external structural characteristics is: Among them, Gb fd represents the external structural features, a represents the cut area of ​​a single surface feature map, A represents the total number of cut areas of the surface feature map, b{·} represents the pooling function, T f (·) represents the time feature extraction function, hb fi represents the surface feature map, * represents the fusion operation, Ex[·] represents the excitation function in the attention mechanism, W represents the horizontal length of the surface feature map, H represents the vertical length of the surface feature map, n represents the unit horizontal length of the surface feature map, m represents the unit vertical length of the surface feature map, v a Represents the convolution kernel corresponding to the cut area.

12. According to claim 9, a transfer equipment maintenance system based on digital twin technology is characterized in that: The point cloud processing logic includes: Preprocessing the laser scanning point cloud according to the laser scanning point cloud of the internal parts of the object to be modeled; Find the most suitable 3D plane through plane fitting, perform coordinate transformation on the pre-processed 3D point cloud data, and transfer the laser scanning point cloud from the radar coordinate system to the world coordinate system; The world coordinate system is divided into grids at equal intervals, the laser scanning point cloud is represented by voxels, and the voxel index of the laser scanning point cloud is projected to the corresponding grid, the points in the voxel are connected, the local feature information is enhanced, and the voxel features of the laser scanning point cloud are obtained; The laser scanning point cloud voxel features are clustered according to the clustering algorithm, and the grids where the laser scanning point cloud voxel features are located under the same central cluster are merged to extract the internal point cloud features. The calculation formula of the internal point cloud features is: Among them, T a represents the internal point cloud features corresponding to the voxel features of the laser scanning point cloud in the same central cluster after clustering, i represents a single voxel feature of the laser scanning point cloud in the central cluster, M represents the total number of voxel features of the laser scanning point cloud in the central cluster, and η i represents the eigenvalue of the voxel feature of the i-th laser scanning point cloud, d represents the clustering threshold corresponding to the central cluster, r2 represents the farthest distance scanned by the laser scanner, r1 represents the closest distance scanned by the laser scanner, and p i_x represents the voxel horizontal coordinate of the i-th laser scanning point cloud voxel feature, p i_y The voxel ordinate represents the feature of the i-th laser scanning point cloud voxel, p i_z Represents the voxel vertical coordinate of the i-th laser scanning point cloud voxel feature.

13. The transfer equipment maintenance system based on digital twin technology according to claim 12 is characterized in that: The cluster analysis comprises: Preliminarily clustering the voxel features of the laser scanning point cloud by using a DBSCAN clustering algorithm to obtain core points; The core points after DBSCAN clustering are used as the initial center clusters of the K-means clustering algorithm, and the Euclidean distances from the voxel features of the laser scanning point cloud to the corresponding initial center clusters are calculated; Clustering is performed according to the Euclidean distance, and the average clustering error of the initial central cluster is calculated. The central cluster is updated according to the average clustering error, and it is determined whether the central cluster meets the convergence condition after the update. If the convergence condition is not met, the central cluster is continued to be updated. If the convergence condition is met, the clustering result is output.

14. The transfer equipment maintenance system based on digital twin technology according to claim 9 is characterized in that: The feature fusion logic includes: Converting the external structural features and the internal point cloud features into a feature sequence, wherein the feature sequence includes an image point cloud feature sequence and a laser scanning point cloud feature sequence; Constructing geometric feature extraction module, surface feature extraction module, internal feature extraction module and spatial feature extraction module; Processing the image point cloud feature sequence by the geometric feature extraction module and the surface feature extraction module to extract geometric features and surface features; Processing the laser scanning point cloud feature sequence by the internal feature extraction module and the spatial feature extraction module to extract internal features and spatial features; The geometric features, the surface features, the internal features and the spatial features are mapped into a tuple order, and fusion features are calculated through pooling operations and residual connections.

15. The transfer equipment maintenance system based on digital twin technology according to claim 14, characterized in that: The surface feature extraction module is used to extract and integrate surface shape features and surface structure features from the image point cloud feature sequence, and the surface feature extraction module includes a surface shape feature extraction submodule, a surface structure feature extraction submodule and a feature cascade submodule; The geometric feature extraction module is used to identify and extract geometric shape features from the image point cloud feature sequence, and the geometric feature extraction module includes a geometric shape detection submodule, a normal vector calculation submodule and a curvature estimation submodule; The spatial feature extraction module is used to extract the spatial layout features of the object from the laser scanning point cloud feature sequence, and the spatial feature extraction module includes a spatial encoding submodule, a spatial transformation submodule and a spatial feature aggregation submodule; The internal feature extraction module is used to extract the internal material distribution features of the object from the laser scanning point cloud feature sequence.

16. The transfer equipment maintenance system based on digital twin technology according to claim 15, characterized in that: The calculation formula of the fusion feature is: F fu =b{With[In(T1),In(T2),...,In(T r )]}, Among them, F fu represents fusion features, Con[·] represents residual connection function, En(·) represents local enhancement function, T r represents the rth feature group in the tuple sequence, and b{·} represents the pooling function.

17. A method for maintenance of transfer equipment based on digital twin technology, the method being implemented based on a transfer equipment maintenance system based on digital twin technology as claimed in any one of claims 1 to 16, characterized in that: The method comprises: Constructing a three-dimensional virtual model of a substation, wherein the three-dimensional virtual model of the substation includes a physical space, a virtual space, and a digital service space; Determine an object to be modeled according to the physical space, and obtain sensor data of the object to be modeled through a sensor component; Processing the sensor data, constructing a three-dimensional model of a substation, and filling the three-dimensional model of the substation into the virtual space; Mapping the physical space and the virtual space to the digital service space, and displaying the operation status, fault location information and test and maintenance plan of the equipment in real time through the digital service space; A maintenance instruction is generated according to the test and detection scheme, and a maintenance device is controlled by the maintenance instruction to perform maintenance on the transfer equipment.

Images