A refilling pump maintenance auxiliary system based on point cloud data superposition

Abstract

This invention relates to the field of top-charge pump technology, specifically disclosing a top-charge pump maintenance assistance system based on point cloud data overlay. This system addresses the problems of inaccurate point cloud data, low defect identification rate, and inconsistent maintenance quality in traditional top-charge pump maintenance. The system includes an auxiliary maintenance module and a maintenance data recording and analysis module. The auxiliary maintenance module comprises an intelligent inspection guidance module, an auxiliary measurement module, and a digital results storage module. The auxiliary maintenance module is connected to a data analysis module, which includes a point cloud data overlay comparison module, a measurement data comparison module, and a model data comparison module. The data analysis module is also connected to a data processing module, which includes an AI defect identification module and an AI fault prediction module. The point cloud data overlay comparison module is configured with an improved dynamic registration algorithm. This invention, through point cloud data overlay, dynamic registration algorithm, and deep learning model, accurately identifies impeller defects and improves fault prediction capabilities, reducing misidentification and omissions.

Description

Technical Field

[0001] This invention relates to the field of top-charge pump technology, and more specifically, to a top-charge pump maintenance auxiliary system based on point cloud data overlay. Background Technology

[0002] The charging pump in nuclear power plants is used to supply water to the main shaft pump seal, make-up water for the volume control of the chemical enrichment system, and supply water in case of high-pressure safety injection system failure. Its disassembly and maintenance mainly involves steps such as disassembly, measurement, repair or replacement, remeasurement, reassembly, testing, and functional reverification. It is characterized by high disassembly difficulty, numerous measurement parameters, and meticulous reassembly work. Once a failure occurs, it can lead to abnormalities in the reactor cooling system, causing serious safety accidents and huge economic losses. Traditional charging pump maintenance mainly relies on manual experience and simple tools for inspection and repair. Manual inspection is difficult to detect some subtle wear, corrosion, and other defects. Furthermore, for complex internal structures, inspection methods are limited. The maintenance process lacks standardized and digital guidance, making operational errors prone to occur, resulting in inconsistent maintenance quality. Outdated data recording and management methods hinder effective data analysis and fault prediction. In particular, the complex structure of the top-charge pump, with easily obstructed areas such as the interface between the blades and the shaft, makes it difficult for traditional testing methods to comprehensively and accurately obtain information from these areas, resulting in incomplete test results and overlooking potential faults. The maintenance process for the top-charge pump is cumbersome, involving the disassembly and assembly of multiple components. Without intuitive and detailed operating instructions, operational errors are prone to occur, affecting maintenance quality and progress. Furthermore, previous maintenance data records were scattered and lacked effective classification and indexing, making it difficult to comprehensively analyze historical data. This hinders the provision of strong support for fault prediction and maintenance decisions, making it impossible to predict potential faults in the top-charge pump in a timely and accurate manner. Maintenance can only be carried out after a fault occurs, increasing maintenance costs and safety risks. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, this invention provides a point cloud data overlay-based auxiliary system for the maintenance of charging pumps. By overlaying point cloud data, using dynamic registration algorithms and deep learning models, it accurately identifies impeller defects and improves fault prediction capabilities, effectively reducing misidentification and omissions, significantly improving detection efficiency and reliability, and providing reliable technical support for the maintenance of charging pumps.

[0004] During routine maintenance and overhauls, nuclear power plants transport charging pumps to a dedicated testing area for disassembly and repair. The repair team typically consists of 4 to 8 people. To prevent human-caused damage and misoperation, there are procedures required for the repair process. Operators must follow the procedures step by step. However, paper work orders cannot constrain the on-site operating procedures. Workers may operate based on their own understanding, resulting in substandard repair quality and secondary damage to equipment. The critical dimensions of the impeller diameter and blade thickness of the charging pump are within the preset tolerance range. However, fluctuations in the data, as shown by the measurement results, indicate a slight decrease in impeller diameter and a certain degree of thinning in blade thickness compared to the last maintenance. While the variation range remains within the allowable tolerance range, without combining the trend of long-term maintenance data and comparing it with previous measurement data, it is impossible to clearly identify areas of localized material loss on the impeller blade surface. Consequently, the cause of this phenomenon cannot be fully analyzed. Without comparison with the original design model, the impact of this blade change on the impeller cannot be determined. Traditionally, relying on manual and experience-based analysis of data in this scenario presents significant challenges. Incomplete data and a lack of experience reduce the accuracy of the analysis results, affecting the detection of potential problems during maintenance and overhaul of the charging pump. This increases the risk of sudden failures after maintenance and fails to provide strong support for the maintenance of the charging pump.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A maintenance assistance system for a charging pump based on point cloud data overlay includes an auxiliary maintenance module and a maintenance data recording and analysis module. The auxiliary maintenance module is connected to the maintenance data recording and analysis module. The auxiliary maintenance module includes an intelligent maintenance guidance module, an auxiliary measurement module, and a digital result storage module. The auxiliary measurement module is sequentially connected to the intelligent maintenance guidance module and the digital result storage module. The auxiliary maintenance module is connected to a data analysis module, which includes a point cloud data overlay comparison module, a measurement data comparison module, and a model data comparison module. The data analysis module is connected to a data processing module, which includes an AI defect identification module and an AI fault prediction module. The auxiliary measurement module is connected to the point cloud data overlay comparison module, which is configured with an improved dynamic registration algorithm and introduces a dynamic weight factor. The system adaptively adjusts to impeller curvature differences, identifies dimensional thinning, crack, and wear areas at set threshold levels, and uses an AI defect identification module to locate and label defect areas based on a deep learning model that fuses point cloud data features at three voxel scales: 0.1mm, 0.5mm, and 1.0mm. The AI ​​fault prediction module constructs a dynamic trend model based on historical point cloud data and current measurement results, monitors blade thickness and impeller diameter changes in real time, and predicts future faults by combining gradient features and current cycle features, issuing early warnings. In the point cloud overlay comparison module, a dynamic registration algorithm allows alignment of point cloud data from different scanning perspectives, introducing a dynamic weighting factor adapted to the impeller morphology of the charging pump to adaptively adjust the registration strategy according to impeller curvature differences. The formula for the dynamic registration algorithm is:

[0007]

[0008] In the formula: E is the total registration error, i is the index of a point in the point cloud data, n is the total number of point cloud data, and ω i For point p i The weights are dynamically adjusted based on the curvature of the region where the point is located. Let A and B be the coordinates of the i-th point in two point cloud datasets A and B, respectively. Let C be the Euclidean distance between the i-th point in two point cloud datasets A and B, j be the point index of a local region in the impeller model, m be the total number of points in the local region of the impeller model, and C be the Euclidean distance between the i-th point in the two point cloud datasets A and B. j Let ||C| be the curvature feature of the j-th point in a local region of the impeller primitive point cloud data, α be the regularization parameter, and ||C| be the curvature feature. j || 2 Let be the curvature deviation at the j-th point in a local region of the impeller model.

[0009] As a further embodiment of the present invention, the auxiliary measurement module includes a rotating platform, on which a measuring head is provided. The measuring head includes a binocular lens, a projection head, a projection light source device, a grating generator, an ambient light filtering module, a calibration plate, and a vibration adaptive module. The measuring head is connected to a point cloud synthesis computer via a data transmission line. The point cloud synthesis computer is connected to a data segmentation module. The bottom of the rotating platform is provided with a cross-wheel type stand, which includes a main shaft and a horizontal crossbar. The main shaft is longer than 2 meters, and the horizontal crossbar is longer than 0.8 meters.

[0010] As a further aspect of the present invention, the measuring head operates in continuous scanning mode, with a maximum single scanning range of 1000m and a single scanning time of less than or equal to 0.2 seconds. It is located directly above the rotating platform. The binocular lens, including a left camera and a right camera, is located in front of the measuring head and faces the object being scanned at a set angle. The projection head is positioned in front of the binocular lens and 0.3cm higher than the base camera lens. The projection light source is located 4.2cm below the centerline between the projection head and the base dual camera lens. The projection light source uses a wavelength range of 440–480nm. Blue light is used as the projection light source. The grating generator works with the projection head to generate a digital cross-shaped grating pattern, which is then projected onto the target surface and overlaps with the field of view of the binocular lens. The distorted grating pattern is captured by the binocular lens and embedded in the projection head. The ambient light filter module is used to ensure that the red laser irradiation on the surface of the measured object does not have an impact. The interface of the calibration board provides graphic guidance for measuring head calibration and lens settings. The vibration adaptive module is used to detect vibration during the measurement process. If the vibration exceeds the preset range, it issues a command to rescan 3 to 5 times. After the vibration stops, it automatically and continuously scans.

[0011] As a further embodiment of the present invention, the measuring head is wirelessly connected to an optical probe, which is a ruby ​​probe. The probe sphere center position is displayed in real time, and real-time dot detection is performed. The dot data and the scanning model are in the same coordinate system, which is used to detect the geometric dimensions of cylindrical, circular, oblong, conical, spherical curved surface points and planes in the blind spot area.

[0012] As a further aspect of the present invention, the scanning angle of the measuring head includes: a binocular lens angle, a left scanning angle, and a right scanning angle. The binocular lens angle is the scanning angle of the binocular lens, the left scanning angle is the scanning angle composed of the left camera and the projection head, and the right scanning angle is the scanning angle composed of the right camera and the projection head.

[0013] As a further aspect of the present invention, the data analysis module also includes a back projection module connected to the point cloud data overlay comparison module, the measurement data comparison module, and the model data comparison module, which is used to project the three-dimensional coordinates of the measurement data and the point cloud comparison detection data onto the workpiece image of the upper filling pump mold.

[0014] As a further aspect of the present invention, the process of the intelligent maintenance guidance module providing visualized intelligent maintenance guidance is as follows:

[0015] Step 1, Visual Guidance: Based on the 3D model of the top charge pump, a 3D animation demonstrating the assembly and disassembly of the top charge pump is provided. The animation includes the assembly and disassembly sequence of the top charge pump components, operating tools, operating steps, and operating precautions. The demonstration process is accompanied by various forms of text, images, videos, and audio to explain the details and key points of each step.

[0016] Step 2, Assisted Maintenance: Provides a digital maintenance method based on on-site maintenance processes and steps. Using the component disassembly and assembly sequence as nodes, it guides maintenance personnel to scan the charging pump components one by one, save the digital results of measurement data, point cloud data, and model data. The scanned model data is compared with the component model. If it does not match the node model, it indicates an abnormal step and guides the maintenance personnel to scan the charging pump components according to the correct steps. Each guided step includes a demonstration of the operation, a voice introduction to guide the scanning, and saving data. The maintenance personnel are allowed to watch the operation demonstration first and become familiar with the steps before being guided to scan and save data, and the components are monitored during the maintenance process.

[0017] Step 3, record maintenance data: Record and save maintenance start time, maintenance personnel, disassembly and assembly sequence, and charging pump component scan data. Support tagging based on unit, charging pump number, and maintenance time, save maintenance data, form a multi-time point maintenance database based on the charging pump, and form a two-dimensional data index of charging pump number and maintenance time, and maintenance data management based on number and equipment.

[0018] Step 4, Generate Maintenance Report: Automatically generate maintenance summary reports based on the last day of each maintenance project.

[0019] As a further aspect of the present invention, the process of identifying and locating defective areas of the charging pump impeller by the AI ​​defect identification module includes:

[0020] Step 1, Data Acquisition and Preprocessing: Obtain point cloud data of the charging pump impeller after registration and alignment using an improved dynamic registration algorithm, perform data cleaning, process the point cloud data according to a predetermined voxelization scale, and convert it into a structure suitable for the input of a deep learning model.

[0021] Step 2, Feature Extraction: Based on different voxelization scales, extract texture, shape and depth feature information at different levels from the point cloud data. Input the voxelized point cloud data into the deep learning model and extract local features through convolutional layers. The point cloud data at each scale goes through three convolutional layers to extract feature data of the impeller surface.

[0022] Step 3, Feature Fusion: The data from different scales are fused, and the importance of defect identification is weighted according to the features of each scale. The features from different scales are then spliced ​​together along the feature dimension.

[0023] Step 4, Defect Classification and Prediction: Defect classification is performed using the fully connected layer of a deep neural network. The feature vector of each point cloud is matched with a predefined defect category, and the probability value of whether each point is a defect is output. When the probability is greater than a set threshold, it is marked as a defect area. Based on the probability map output by the model, the location of the defect is determined and the defect area is marked. The coordinate range of the defect is determined by regression method.

[0024] Step 5, Defect Region Annotation and Post-processing: After detecting the point cloud region of the defect, further region extraction is performed to determine the shape and size of the defect. The defect is then smoothed using a Gaussian smoothing filter. Misidentified areas in the annotated defect region are corrected based on point cloud data overlap technology and an improved dynamic registration algorithm. The defect region is highlighted in the 3D point cloud map using a 3D visualization tool. Based on the defect identification results, the wear and damage degree of the impeller are analyzed.

[0025] As a further aspect of the present invention, in step five, using the point cloud data from the first scan and the current scan, the diameter and blade thickness features of the impeller are used to correct misidentified areas in the marked defect regions based on point cloud data overlap technology and an improved dynamic registration algorithm. The formula for correcting misidentified areas is as follows:

[0026] P cor (p i ) = P smo (p i )·[1-I(p i )•E(p i )•F(p i )]

[0027] In the formula: P cor (p i P represents the position coordinates of the i-th point after correction of the misidentified region. smo (p i Let I(p) be the coordinates of the i-th point after smoothing. i ) is the misidentification correction index function, if point p i If it belongs to the misidentified region, then I(p) i E(p) = 1, otherwise 0, i ) is the error threshold function, used to measure the error at point p. i Whether the error exceeds the set threshold is determined based on the point cloud registration error and the change in geometric features, F(p) i (p) is the judgment point based on impeller diameter and blade thickness. i Is the function within the normal region, if point p i The impeller blade diameter and thickness are within the permissible range, F(p) i=1, otherwise 0.

[0028] As a further aspect of the present invention, the process of the AI ​​fault prediction module in predicting impeller faults includes:

[0029] Step 1: Merge historical and current data: Merge historical point cloud data with current impeller blade thickness and impeller diameter data, and construct a dynamic trend model through time series analysis to show the changing trends of impeller blade thickness and impeller diameter over time;

[0030] Step II, Obtain gradient and periodic features: Analyze the gradient of change of impeller blade thickness and impeller diameter by calculating the gradient of change, and perform periodic analysis of the data;

[0031] Step III, Real-time Monitoring and Fault Prediction: Real-time monitoring of impeller parameter changes recorded during previous maintenance and overhauls, combined with dynamic trend models to perform machine learning predictions on the current data, and comparison of the prediction results output by the model with preset thresholds. If the prediction results exceed the preset thresholds, potential component faults are identified, and an early warning is issued through the intelligent maintenance guidance module in the auxiliary maintenance module to remind maintenance personnel to take preventive maintenance measures.

[0032] The technical advantages of the charging pump maintenance auxiliary system based on point cloud data overlay proposed in this invention are as follows:

[0033] This invention utilizes point cloud data overlay technology, an improved dynamic registration algorithm, and a deep learning-based defect identification and trend prediction model to accurately register and analyze point cloud data. This not only improves the accuracy and reliability of maintenance operations but also reduces misidentification and omissions caused by traditional reliance on experience-based judgment. It enhances detection efficiency and fault prediction capabilities, enabling early detection of potential problems in impeller components and proactively reducing the occurrence of sudden failures. This provides reliable technical support for the maintenance and operation of charging pumps. Attached Figure Description

[0034] Figure 1 These are disassembly interface diagrams and model disassembly interface diagrams of the charging pump device in this invention;

[0035] Figure 2 This is a schematic diagram of the intelligent maintenance of the charging pump of the present invention;

[0036] Figure 3 This is the data analysis interface for the charging pump maintenance in this invention;

[0037] Figure 4 This is the interface for viewing the invention report;

[0038] Figure 5 This is a disassembly interface diagram of the upper charging pump coupling of the present invention;

[0039] Figure 6 For maintenance process data analysis diagrams;

[0040] Figure 7 This is a screenshot of the maintenance management interface.

[0041] Figure 8 A comparison chart of measurement data and model data for the charging pump assembly;

[0042] Figure 9 Image of the actual measuring head;

[0043] Figure 10 This is a diagram of a tripod model;

[0044] Figure 11 This is a picture of the actual rotating platform.

[0045] Figure 12 A schematic diagram showing the division of the rotating platform into different areas;

[0046] Figure 13 The diagram shows the effect of the measurement head's detection overlapping with the actual image, the back projection function, and the dynamic tracking function.

[0047] Figure 14 This is a schematic diagram illustrating the use of an optical probe. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] To achieve the above objectives, the present invention provides the following technical solution:

[0050] Example 1

[0051] like Figures 1 to 8As shown, this invention proposes a point cloud data overlay-based auxiliary system for the maintenance of charging pumps. It includes an auxiliary maintenance module and a maintenance data recording and analysis module, both connected to the maintenance data recording and analysis module. The auxiliary maintenance module includes an intelligent maintenance guidance module, an auxiliary measurement module, and a digital results storage module. The auxiliary measurement module is sequentially connected to the intelligent maintenance guidance module and the digital results storage module. The auxiliary maintenance module is also connected to a data analysis module, which includes a point cloud data overlay comparison module, a measurement data comparison module, and a model data comparison module. The data analysis module is further connected to a data processing module, which includes an AI defect identification module and an AI fault prediction module. The auxiliary measurement module is connected to the point cloud data overlay comparison module, which is configured with an improved dynamic registration algorithm. The AI ​​defect identification module uses a dynamic weighting factor to adaptively adjust the impeller curvature difference, identifying dimensional thinning, crack, and wear areas at set threshold levels. It locates and labels defect areas based on a deep learning model that fuses point cloud data features at three voxel scales: 0.1mm, 0.5mm, and 1.0mm. The AI ​​fault prediction module constructs a dynamic trend model based on historical point cloud data and current measurement results, monitoring blade thickness and impeller diameter changes in real time. Combining gradient features and current cycle features, it predicts future faults and issues early warnings. In the point cloud overlay comparison module, the dynamic registration algorithm allows alignment of point cloud data from different scanning perspectives. It introduces a dynamic weighting factor adapted to the impeller morphology of the charging pump to adaptively adjust the registration strategy according to impeller curvature differences. The formula for the dynamic registration algorithm is:

[0052]

[0053] In the formula: E is the total registration error, i is the index of a point in the point cloud data, n is the total number of point cloud data, and ω i For point p i The weights are dynamically adjusted based on the curvature of the region where the point is located. Let A and B be the coordinates of the i-th point in two point cloud datasets A and B, respectively. Let C be the Euclidean distance between the i-th point in two point cloud datasets A and B, j be the point index of a local region in the impeller model, m be the total number of points in the local region of the impeller model, and C be the Euclidean distance between the i-th point in the two point cloud datasets A and B. j Let ||C| be the curvature feature of the j-th point in a local region of the impeller primitive point cloud data, α be the regularization parameter, and ||C| be the curvature feature. j || 2 Let be the curvature deviation at the j-th point in a local region of the impeller model.

[0054] To clearly illustrate the technical effect of the dynamic registration algorithm in the above technical solution, data from test groups 1, 2, and 3 are selected for specific explanation. The test equipment used in each group employs the auxiliary measurement module described in the technical solution, maintaining stable temperature and humidity conditions in the scanning environment. Test group 1, based on the initial overhaul data of the charging pump impeller, focuses on detecting early wear or thinning areas on the impeller surface. During the initial overhaul, the impeller diameter was within the standard tolerance range, and localized roughness began to appear at the blade edges, not yet reaching the severe wear stage. The blade edge thickness decreased by 0.3 mm, and localized surface roughness increased, but the fault threshold was not reached. Test group 2, based on the data from the first major overhaul, shows that some areas of the impeller blades had undergone repair... The damaged blades were replaced. Minor warping at the blade joints and mismatch in surface texture in the repaired area can lead to changes in localized hardness or surface smoothness. A dynamic registration algorithm was used to identify the geometric deformation of the joint area and check the consistency of the surface texture after repair. Test group 3, based on historical data after the second major overhaul, focused on detecting fatigue and potential cracks in the impeller after prolonged use. Microcracks appeared at the blade edges, propagating radially with drastic curvature changes. Point cloud overlay technology and a dynamic registration algorithm were used to accurately identify the crack location and morphology, assessing the impact of the cracks on the overall impeller structure. The control group used traditional manual experience-based inspection methods and simple registration techniques. The total registration error value E and curvature deviation of the four groups were compared. j || 2 The data is shown in Table 1:

[0055] Table 1 Comparison of Key Parameters in the Detection Area

[0056]

[0057] The reduction in curvature deviation indicates that the dynamic registration algorithm significantly improves the alignment accuracy of point cloud data, especially when dealing with complex shapes (such as impeller surfaces), reducing errors caused by curvature differences.

[0058] The changes in parameters such as curvature change rate, curvature deviation, and area of ​​the thinned region indicate that by improving the algorithm, the defect area of ​​the impeller can be identified and calibrated more accurately, avoiding the errors caused by roughness or uneven surface in traditional methods.

[0059] Overall, the reduction in E (registration error) and curvature deviation improves the accuracy of defect areas, avoids the risk of missed detection or misjudgment, and enhances the reliability of fault prediction and maintenance assistance.

[0060] Example 2

[0061] like Figures 9 to 12 As shown, the auxiliary measurement module includes a rotating platform ( Figure 11The rotating platform is displayed as a whole, and is divided into several areas (such as...). Figure 12 As shown), it can place several charging pump assemblies together on a rotating platform and scan all components on the platform at once. After scanning, it automatically divides the components into several scanning parts. The rotating platform is equipped with a measuring head (such as...). Figure 9 As shown), the measuring head includes a binocular lens, a projection head, a projection light source device, a grating generator, an ambient light filter module, a calibration plate, and a vibration adaptive module. The measuring head is connected to a point cloud synthesis computer via a data transmission line. The point cloud synthesis computer is connected to a data segmentation module. The bottom of the rotating platform is equipped with a cross-wheel type tripod (such as...). Figure 10 As shown, the cross-wheel tripod includes a main shaft and a horizontal crossbar. The main shaft is longer than 2 meters, and the horizontal crossbar is longer than 0.8 meters. It can move freely indoors, facilitating measurement.

[0062] The measuring head operates in continuous scanning mode, with a maximum single scan range of 1000m and a single scan time of less than or equal to 0.2 seconds. It is located directly above the rotating platform. The binocular lenses, including a left and right camera, are located at the front of the measuring head and are angled toward the object being scanned. The projection head is positioned in front of the binocular lenses and is 0.3cm higher than the base camera lenses. The projection light source is located 4.2cm below the centerline between the projection head and the base dual camera lenses. The projection light source uses LEN blue light with a wavelength range of 440-480nm as the projection light source. There is no need for a powder sprayer to affix labels for scanning, ensuring the safety and reliability of the test data. The technology effectively reduces or controls noise during the scanning process, avoiding the orange peel effect of the scanning components. The light projected by the measuring head has consistent brightness throughout the entire range, and the four corners cannot be darker than the center. Verification is performed by comparison. The grating generator works in conjunction with the projection head to generate a digital crosshair grating pattern, which is projected onto the target surface and overlaps with the field of view of the binocular lens. The binocular lens captures the distorted grating pattern and embeds it within the projection head. An ambient light filter module ensures that red laser light irradiating the measured object surface does not have an impact. The calibration board interface provides graphical guidance for measuring head calibration and lens settings. A vibration adaptive module detects vibration during the measurement process. If the vibration exceeds a preset range, it issues a command to rescan 3 to 5 times, and automatically resumes scanning after the vibration stops. The measuring head operates in a temperature range of +5℃ to +35℃ and is non-condensable. Within different temperature ranges, a thermometer is used to identify the temperature of the calibration board for input, and automatic temperature compensation is performed. A 10-GigE data link is established via fiber optic cable to complete the transmission of image and control signals.

[0063] It should be noted that the measuring head is wirelessly connected to an optical probe (such as...). Figure 14As shown, the optical probe is a ruby ​​probe, and its probe bar is a pre-calibrated group of label points (the relationship between the label points and the center of the sphere). Measuring the label point group of the probe bar automatically establishes the sphere center measurement point, possessing dynamic reference characteristics, displaying the probe sphere center position in real time, and performing real-time dot detection. The dot data and the scanning model are in the same coordinate system, used to detect the geometric dimensions of cylindrical, circular, oblong, conical, and spherical curved surface points and planes in blind spot areas. The scanning angles of the measuring head include: binocular lens angle, left scanning angle, and right scanning angle. The binocular lens angle is the scanning angle of the binocular lens; the left scanning angle is the scanning angle composed of the left camera and the projection head; and the right scanning angle is the scanning angle composed of the right camera and the projection head. This triple-view scanning technology ensures that more features of the complex structure of the charging pump assembly are measured, and better scans the detailed structure of the parts.

[0064] It should be noted that the data analysis module also includes a back projection module connected to the point cloud data overlay comparison module, the measurement data comparison module, and the model data comparison module. This module is used to project the three-dimensional coordinates of the measurement data and point cloud comparison detection data onto the workpiece image of the upper filling pump mold, enhancing the component's recognizability (e.g., Figure 13 (As shown).

[0065] The auxiliary measurement module combines full-area contact scanning and contact 3D point measurement through optical probes. It transmits data to the real world in a virtual alignment manner through component alignment and positioning to achieve dynamic tracking. The measurement data is then back-projected onto the component to achieve back-projection measurement. The scanning process displays the changes in the component's image in real time, and the measurement is adjusted at different angles and positions. Position and volume parameters are assisted in real-time scanning, and laser focusing is supported in conjunction with scanning.

[0066] The auxiliary detection module's interface for detecting the top-charge pump blades includes an airfoil detection workspace, a toolbar with detection process guidance, and an I-Inspect button for profile section airfoil detection functions such as accumulation points, centerlines, inlet and outlet edge geometry, chord lines, and supports profile shape and position detection. It features an offset centerline and expert parameter settings for profile shape and position detection. It also includes profile edge thickness detection, blade disk scanning and detection, bow and tilt detection, and analysis of accumulation axis, accumulation point arc, profile centroid, profile shape and position, profile boundary points, profile boundary circles, profile boundary ellipses, profile chord lines (axial force chord, double tangent chord, maximum chord, aerodynamic chord), maximum profile thickness, and profile boundary thickness.

[0067] It should be noted that, as Figure 2 As shown, the process of the intelligent maintenance guidance module providing visualized intelligent maintenance guidance is as follows:

[0068] Step 1, Visual Guidance: Based on the 3D model of the top charge pump, a 3D animation demonstrating the assembly and disassembly of the top charge pump is provided. The animation includes the assembly and disassembly sequence of the top charge pump components, operating tools, operating steps, and operating precautions. The demonstration process is accompanied by various forms of text, images, videos, and audio to explain the details and key points of each step.

[0069] Step 2, Assisted Maintenance: Provides a digital maintenance method based on on-site maintenance processes and steps. Using the component disassembly and assembly sequence as nodes, it guides maintenance personnel to scan the charging pump components one by one, save the digital results of measurement data, point cloud data, and model data. The scanned model data is compared with the component model. If it does not match the node model, it indicates an abnormal step and guides the maintenance personnel to scan the charging pump components according to the correct steps. Each guided step includes a demonstration of the operation, a voice introduction to guide the scanning, and saving data. The maintenance personnel are allowed to watch the operation demonstration first and become familiar with the steps before being guided to scan and save data, and the components are monitored during the maintenance process.

[0070] Step 3, record maintenance data: Record and save maintenance start time, maintenance personnel, disassembly and assembly sequence, and charging pump component scan data. Support tagging based on unit, charging pump number, and maintenance time, save maintenance data, form a multi-time point maintenance database based on the charging pump, and form a two-dimensional data index of charging pump number and maintenance time, and maintenance data management based on number and equipment.

[0071] Step 4, Generate Maintenance Report: Automatically generate maintenance summary reports based on the last day of each maintenance project.

[0072] The auxiliary maintenance module for the charging pump also includes a charging pump fault tree database, spare parts management functions, and maintenance management functions, such as... Figure 8 As shown, the comparison chart is formed by measurement data and model data, which can intuitively show the comparison results of measurement data and model data in each step of the standard maintenance process for each component.

[0073] It should be noted that the process of identifying and locating defective areas in the charging pump impeller by the AI ​​defect recognition module includes:

[0074] Step 1, Data Acquisition and Preprocessing: Obtain point cloud data of the charging pump impeller after registration and alignment using an improved dynamic registration algorithm, perform data cleaning, process the point cloud data according to a predetermined voxelization scale, and convert it into a structure suitable for the input of a deep learning model.

[0075] Step 2, Feature Extraction: Based on different voxelization scales, extract texture, shape and depth feature information at different levels from the point cloud data. Input the voxelized point cloud data into the deep learning model and extract local features through convolutional layers. The point cloud data at each scale goes through three convolutional layers to extract feature data of the impeller surface.

[0076] Step 3, Feature Fusion: The data from different scales are fused, and the importance of defect identification is weighted according to the features of each scale. The features from different scales are then spliced ​​together along the feature dimension.

[0077] Step 4, Defect Classification and Prediction: Defect classification is performed using the fully connected layer of a deep neural network. The feature vector of each point cloud is matched with a predefined defect category, and the probability value of whether each point is a defect is output. When the probability is greater than a set threshold, it is marked as a defect area. Based on the probability map output by the model, the location of the defect is determined and the defect area is marked. The coordinate range of the defect is determined by regression method.

[0078] Step 5, Defect Region Annotation and Post-processing: After detecting the point cloud region of the defect, further region extraction is performed to determine the shape and size of the defect. The defect is then smoothed using a Gaussian smoothing filter. Misidentified areas in the annotated defect region are corrected based on point cloud data overlap technology and an improved dynamic registration algorithm. The defect region is highlighted in the 3D point cloud map using a 3D visualization tool. Based on the defect identification results, the wear and damage degree of the impeller are analyzed.

[0079] By combining dynamic registration algorithms and deep learning technology, the defect areas of the charging pump impeller can be identified and located with high precision. The multi-scale method of feature extraction and feature fusion ensures that details at different levels are fully preserved, and accurately classifies and predicts defect areas such as micro-cracks and thinning. Combined with Gaussian filtering smoothing and three-dimensional visualization annotation, not only is the data post-processing optimized, but also the misidentified areas are reduced, providing intuitive and reliable data support for maintenance decisions and improving the overall accuracy, efficiency and maintenance safety of the inspection.

[0080] It should be noted that in step five, using the point cloud data from the first scan and the current scan, the diameter and thickness features of the impeller are used to correct misidentified areas in the marked defect regions based on point cloud data overlap technology and an improved dynamic registration algorithm. The formula for correcting misidentified areas is as follows:

[0081] P cor (p i ) = P smo (p i )·[1-I(p i )·E(p i )·F(p i )]

[0082] In the formula: P cor (p i P represents the position coordinates of the i-th point after correction of the misidentified region. smo (p iLet I(p) be the coordinates of the i-th point after smoothing. i ) is the misidentification correction index function, if point p i If it belongs to the misidentified region, then I(p) i E(p) = 1, otherwise 0, i ) is the error threshold function, used to measure the error at point p. i Whether the error exceeds the set threshold is determined based on the point cloud registration error and the change in geometric features, F(p) i (p) is the judgment point based on impeller diameter and blade thickness. i Is the function within the normal region, if point p i The impeller blade diameter and thickness are within the permissible range, F(p) i =1, otherwise 0.

[0083] Due to the complex structure and high maintenance difficulty of the impeller in top-charge pumps, traditional methods cannot accurately identify local defect areas, especially when there are errors in the point cloud data or significant differences in impeller curvature, which can easily lead to mislabeling. However, by combining the point cloud data from the initial and current scans with the impeller diameter and blade thickness characteristics for misidentification correction, an improved dynamic registration algorithm can adaptively adjust the registration weights to accurately eliminate mislabeled points, ensuring more precise labeling of defect areas on the impeller surface. This avoids subsequent analysis and maintenance quality issues caused by missing data or mislabeling. Using the point cloud data from the initial and current scans, and combining the impeller diameter and blade thickness characteristics with an improved dynamic registration algorithm for misidentification correction, significantly reduces mislabeled points caused by curvature differences and registration errors, reduces omissions or false detections in defect area labeling, improves the alignment accuracy of point cloud data, optimizes the accurate labeling of defect areas, effectively improves the reliability of data analysis and maintenance quality, ensures that subtle defects on the complex surface of the impeller can be fully detected, provides more reliable and efficient data support for top-charge pump maintenance decisions, reduces operational risks, and enhances the safety and efficiency of maintenance work.

[0084] It should be noted that the process of the AI ​​fault prediction module in predicting impeller faults includes:

[0085] Step 1: Merge historical and current data: Merge historical point cloud data with current impeller blade thickness and impeller diameter data, and construct a dynamic trend model through time series analysis to show the changing trends of impeller blade thickness and impeller diameter over time;

[0086] Step II, Obtain gradient and periodic features: Analyze the gradient of change of impeller blade thickness and impeller diameter by calculating the gradient of change, and perform periodic analysis of the data;

[0087] Step III, Real-time Monitoring and Fault Prediction: Real-time monitoring of impeller parameter changes recorded during previous maintenance and overhauls, combined with dynamic trend models to perform machine learning predictions on the current data, and comparison of the prediction results output by the model with preset thresholds. If the prediction results exceed the preset thresholds, potential component faults are identified, and an early warning is issued through the intelligent maintenance guidance module in the auxiliary maintenance module to remind maintenance personnel to take preventive maintenance measures.

[0088] In summary, this invention utilizes point cloud data overlay technology, an improved dynamic registration algorithm, and a deep learning-based defect identification and trend prediction model to accurately register and analyze point cloud data. This not only improves the accuracy and reliability of maintenance operations but also reduces misidentification and omissions caused by traditional reliance on experience-based judgment. It enhances detection efficiency and fault prediction capabilities, enabling early detection of potential problems in impeller components and proactively reducing the occurrence of sudden failures. This provides a reliable technical guarantee for the maintenance and operation of the charging pump.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0090] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A point cloud data overlay-based auxiliary system for the maintenance of a charging pump, comprising an auxiliary maintenance module and a maintenance data recording and analysis module, characterized in that, The auxiliary maintenance module includes an intelligent maintenance guidance module, an auxiliary measurement module, and a digital results storage module. The auxiliary measurement module is sequentially connected to the intelligent maintenance guidance module and the digital results storage module. The auxiliary maintenance module is also connected to a data analysis module, which includes a point cloud data overlay and comparison module, a measurement data comparison module, and a model data comparison module. This data analysis module is connected to a data processing module, which includes an AI defect identification module and an AI fault prediction module. The auxiliary measurement module is connected to the point cloud data overlay and comparison module, which is equipped with an improved dynamic registration algorithm. This module introduces a dynamic weighting factor to adaptively adjust for impeller curvature differences and identify dimensional thinning at set threshold levels. In the crack and wear areas, the AI ​​defect identification module locates and labels defect areas based on a deep learning model that fuses point cloud data features at three voxel scales: 0.1mm, 0.5mm, and 1.0mm. The AI ​​fault prediction module constructs a dynamic trend model based on historical point cloud data and current measurement results, monitoring changes in blade thickness and impeller diameter in real time. Combining gradient features and current cycle characteristics, it predicts future faults and issues early warnings. In the point cloud overlay and comparison module, the dynamic registration algorithm allows alignment of point cloud data from different scanning perspectives. It introduces a dynamic weighting factor adapted to the impeller morphology of the charging pump to adaptively adjust the registration strategy according to impeller curvature differences. The formula for the dynamic registration algorithm is: In the formula: E is the total registration error, i is the index of a point in the point cloud data, n is the total number of point cloud data, and ω i For point p i The weights are dynamically adjusted based on the curvature of the region where the point is located. Let A and B be the coordinates of the i-th point in two point cloud datasets A and B, respectively. Let C be the Euclidean distance between the i-th point in two point cloud datasets A and B, j be the point index of a local region in the impeller model, m be the total number of points in the local region of the impeller model, and C be the Euclidean distance between the i-th point in the two point cloud datasets A and B. j Let be the curvature feature of the j-th point in a local region of the impeller primitive point cloud data, α be the regularization parameter, and ||C|| be the curvature feature. j || 2 The curvature deviation at the j-th point in a local region of the impeller model; The process by which the AI ​​defect recognition module identifies and locates defective areas on the impeller of the charging pump includes: Step 1, Data Acquisition and Preprocessing: Obtain point cloud data of the charging pump impeller after registration and alignment using an improved dynamic registration algorithm, perform data cleaning, process the point cloud data according to a predetermined voxelization scale, and convert it into a structure suitable for the input of a deep learning model. Step 2, Feature Extraction: Based on different voxelization scales, extract texture, shape and depth feature information at different levels from the point cloud data. Input the voxelized point cloud data into the deep learning model and extract local features through convolutional layers. The point cloud data at each scale goes through three convolutional layers to extract feature data of the impeller surface. Step 3, Feature Fusion: The data from different scales are fused, and the importance of defect identification is weighted according to the features of each scale. The features from different scales are then spliced ​​together along the feature dimension. Step 4, Defect Classification and Prediction: Defect classification is performed using the fully connected layer of a deep neural network. The feature vector of each point cloud is matched with a predefined defect category, and the probability value of whether each point is a defect is output. When the probability is greater than a set threshold, it is marked as a defect area. Based on the probability map output by the model, the location of the defect is determined and the defect area is marked. The coordinate range of the defect is determined by regression method. Step 5, Defect Region Annotation and Post-processing: After detecting the point cloud region of the defect, further region extraction is performed to determine the shape and size of the defect. A Gaussian smoothing filter is used for smoothing. Misidentified areas in the annotated defect region are corrected using point cloud data overlap technology and an improved dynamic registration algorithm. The defect region is highlighted in the 3D point cloud map using a 3D visualization tool. The wear and damage degree of the impeller are analyzed based on the defect identification results. Using the point cloud data from the first and current scans, the diameter and blade thickness features of the impeller are used to correct misidentified areas in the annotated defect region using point cloud data overlap technology and an improved dynamic registration algorithm. The formula for correcting misidentified areas is: P cor (p i )=P smo (p i )·[1-I(p i )·E(p i )·F(p i )] In the formula: P cor (p i P represents the position coordinates of the i-th point after correction of the misidentified region. smo (p i Let I(p) be the coordinates of the i-th point after smoothing. i ) is the misidentification correction index function, if point p i If it belongs to the misidentified region, then I(p) i E(p) = 1, otherwise 0, i ) is the error threshold function, used to measure the error at point p. i Whether the error exceeds the set threshold is determined based on the point cloud registration error and the change in geometric features, F(p) i (p) is the judgment point based on impeller diameter and blade thickness. i Whether a function belongs to the normal region, if point p i The impeller blade diameter and thickness are within the permissible range, F(p) i =1, otherwise 0.

2. The charging pump maintenance auxiliary system based on point cloud data overlay according to claim 1, characterized in that, The auxiliary measurement module includes a rotating platform with a measuring head. The measuring head includes a binocular lens, a projection head, a projection light source, a grating generator, an ambient light filter module, a calibration plate, and a vibration adaptive module. The measuring head is connected to a point cloud synthesis computer via a data transmission line. The point cloud synthesis computer is connected to a data segmentation module. The bottom of the rotating platform is equipped with a cross-wheel type support, which includes a main shaft and a horizontal crossbar. The main shaft is longer than 2 meters, and the horizontal crossbar is longer than 0.8 meters.

3. The charging pump maintenance auxiliary system based on point cloud data overlay according to claim 2, characterized in that, The measuring head operates in continuous scanning mode, with a maximum single scan range of 1000m and a single scan time of less than or equal to 0.2 seconds. It is located directly above the rotating platform. The binocular lens, including a left and right camera, is located at the front of the measuring head and faces the object being scanned at a set angle. The projection head is positioned in front of the binocular lens and is 0.3cm higher than the base camera lens. The projection light source is located 4.2cm below the centerline between the projection head and the base dual camera lens. The projection light source uses LEN blue light with a wavelength range of 440-480nm as the projection light source. The grating generator works with the projection head to generate a digital cross-shaped grating pattern and projects it onto the target surface, coinciding with the field of view of the binocular lens. The binocular lens captures the distorted grating pattern and embeds it within the projection head. An ambient light filter module ensures that red laser light irradiating the surface of the measured object does not have an impact. The calibration board interface provides graphic guidance for measuring head calibration and lens settings. The vibration adaptive module detects vibration during the measurement process. If the vibration exceeds a preset range, it issues a command to rescan 3 to 5 times. After the vibration stops, it automatically and continuously scans.

4. The charging pump maintenance auxiliary system based on point cloud data overlay according to claim 3, characterized in that, The measuring head is wirelessly connected to an optical probe, which is a ruby ​​probe. It displays the position of the probe sphere center in real time and performs real-time point detection. The point data and the scanning model are in the same coordinate system. It is used to detect the geometric dimensions of cylindrical, circular, oblong, conical, and spherical curved surface points and planes in the blind spot area.

5. The charging pump maintenance auxiliary system based on point cloud data overlay according to claim 4, characterized in that, The scanning angles of the measuring head include: binocular lens angle, left scanning angle, and right scanning angle. The binocular lens angle is the scanning angle of the binocular lens, the left scanning angle is the scanning angle composed of the left camera and the projection head, and the right scanning angle is the scanning angle composed of the right camera and the projection head.

6. The charging pump maintenance auxiliary system based on point cloud data overlay according to claim 4, characterized in that, The data analysis module also includes a back projection module that is connected to the point cloud data overlay comparison module, the measurement data comparison module, and the model data comparison module. This module is used to project the three-dimensional coordinates of the measurement data and point cloud comparison detection data onto the workpiece image of the upper filling pump mold.

7. The charging pump maintenance auxiliary system based on point cloud data overlay according to claim 1, characterized in that, The process of providing visualized intelligent maintenance guidance by the intelligent maintenance guidance module is as follows: Step 1, Visual Guidance: Based on the 3D model of the top charge pump, a 3D animation demonstrating the assembly and disassembly of the top charge pump is provided. The animation includes the assembly and disassembly sequence of the top charge pump components, operating tools, operating steps, and operating precautions. The demonstration process is accompanied by various forms of text, images, videos, and audio to explain the details and key points of each step. Step 2, Assisted Maintenance: Provides a digital maintenance method based on on-site maintenance processes and steps. Using the component disassembly and assembly sequence as nodes, it guides maintenance personnel to scan the charging pump components one by one, save the digital results of measurement data, point cloud data, and model data. The scanned model data is compared with the component model. If it does not match the node model, it indicates an abnormal step and guides the maintenance personnel to scan the charging pump components according to the correct steps. Each guided step includes a demonstration of the operation, voice introduction to guide the scanning and data saving. The maintenance personnel can watch the operation demonstration first, familiarize themselves with the steps, and then be guided to scan and save data, monitoring the components during the maintenance process. Step 3, record maintenance data: Record and save maintenance start time, maintenance personnel, disassembly and assembly sequence, and charging pump component scan data. Support tagging based on unit, charging pump number, and maintenance time, save maintenance data, form a multi-time point maintenance database based on the charging pump, and form a two-dimensional data index of charging pump number and maintenance time, and maintenance data management based on number and equipment. Step 4, Generate Maintenance Report: Automatically generate maintenance summary reports based on the last day of each maintenance project.

8. The charging pump maintenance auxiliary system based on point cloud data overlay according to claim 1, characterized in that, The process of impeller fault prediction by the AI ​​fault prediction module includes: Step 1: Merge historical and current data: Merge historical point cloud data with current impeller blade thickness and impeller diameter data, and construct a dynamic trend model through time series analysis to show the changing trends of impeller blade thickness and impeller diameter over time; Step II, Obtain gradient and periodic features: Analyze the gradient of change of impeller blade thickness and impeller diameter by calculating the gradient of change, and perform periodic analysis of the data; Step III, Real-time Monitoring and Fault Prediction: Real-time monitoring of impeller parameter changes recorded during previous maintenance and overhauls, combined with dynamic trend models to perform machine learning predictions on the current data, and comparison of the prediction results output by the model with preset thresholds. If the prediction results exceed the preset thresholds, potential component faults are identified, and an early warning is issued through the intelligent maintenance guidance module in the auxiliary maintenance module to remind maintenance personnel to take preventive maintenance measures.

Images