Vacuum pump impeller defect visual intelligent identification method and system

Abstract

The present application relates to the technical field of data processing, in particular to a kind of vacuum pump impeller defect visual intelligent identification method and system.The method comprises: obtaining the two-dimensional image data of vacuum pump impeller, three-dimensional point cloud data;Extract visual feature vector and three-dimensional geometric parameter vector;Determine the physical type of each defect and confidence;When confidence is greater than pre-set confidence threshold, select the mechanical model corresponding to the physical type, calculate risk index;According to risk index, the defect of vacuum pump impeller is classified by severity, and the grade result is obtained;Correct grade result, obtain final grade result;Based on final grade result, generate the defect monitoring report of vacuum pump impeller.The present application can improve the classification accuracy of the physical type of defect.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a visual intelligent recognition method and system for defects in vacuum pump impellers. Background Technology

[0002] Traditional vacuum pump impeller defect detection has long relied on manual visual inspection or conventional ultrasonic testing. Manual visual inspection is affected by human eye resolution, subjective fatigue, and lighting conditions, resulting in a high rate of missed detections of minute pores, cold shuts, or hairline cracks. Conventional ultrasonic testing is ill-suited for full-coverage scanning of the complex three-dimensional curved surface of the impeller. With the development of machine vision technology, methods based on grayscale threshold segmentation and template matching have been gradually applied to the surface inspection of metal castings, but their adaptability to scenes with varied defect morphologies and strong background texture interference is limited. In recent years, breakthroughs in deep learning, especially the YOLO series of object detection frameworks, have significantly improved the robustness of identifying low-contrast, minute defects.

[0003] However, existing deep learning methods only output classification results of "defective" or "no defect", without combining material properties and operating parameters to quantify the risk assessment of defects. They cannot distinguish the risk differences under different locations and operating conditions, and are also difficult to support predictive maintenance decisions for in-service impellers. Summary of the Invention

[0004] This invention provides a visual intelligent recognition method and system for vacuum pump impeller defects to solve existing problems.

[0005] The present invention provides a visual intelligent identification method for defects in vacuum pump impellers, which adopts the following technical solution: One embodiment of the present invention provides a visual intelligent recognition method for defects in a vacuum pump impeller, the method comprising the following steps: Acquire two-dimensional image data and three-dimensional point cloud data of the vacuum pump impeller; The visual feature vector of each defect is extracted from the two-dimensional image data, and the three-dimensional geometric parameter vector of each defect is extracted from the three-dimensional point cloud data. The visual feature vector and the three-dimensional geometric parameter vector are weighted and matched with a preset defect physical type database to determine the physical type and confidence level of each defect. The preset defect physical type database includes at least the feature threshold ranges corresponding to cracks, pores, wear and corrosion. When the confidence level is less than or equal to the preset confidence threshold, the defect is manually reviewed. When the confidence level is greater than the preset confidence threshold, the material parameters and operating parameters of the vacuum pump impeller are obtained according to the physical type. The mechanical model corresponding to the physical type is selected, and the material parameters and operating parameters are substituted into the mechanical model to calculate the risk index used to characterize the degree of defect danger. The mechanical model includes a fracture mechanics model, a dynamic balance model or a finite element model, and the risk index includes at least one of the following: remaining fatigue life, safety factor and unbalance amount. The defects of the vacuum pump impeller are classified into severity levels based on risk indicators, and the level results are obtained. The historical defect data of the vacuum pump impeller is obtained, the growth rate of the defect size over time in the historical defect data is calculated, and the grade result is corrected according to the growth rate to obtain the final grade result. Based on the final grade results, a defect monitoring report for the vacuum pump impeller is generated.

[0006] Furthermore, the visual feature vector of each defect is extracted from the two-dimensional image data, specifically including: The Otsu threshold segmentation method was used to extract the defect region, and the average gray level of the defect region was calculated. Construct a gray-level co-occurrence matrix and calculate texture contrast and texture entropy; The Canny edge detection algorithm is used to extract defect edges, and curves are fitted to the defect edge points to calculate the edge curvature. The average gray level, texture contrast, texture entropy, and edge curvature are combined to form the visual feature vector of the defect.

[0007] Furthermore, the three-dimensional geometric parameter vector for each defect is extracted from the three-dimensional point cloud data, specifically including: The ICP algorithm is used to register the 3D point cloud data with the standard CAD model to obtain the registered point cloud data. The point cloud data after registration is compared with the standard CAD model to obtain the point cloud of the defect area. The average distance between each defect point in the defect area point cloud and the corresponding point in the standard CAD model is determined as the defect depth. Calculate the projected area of ​​each defect on the vacuum pump impeller; Calculate the longest axis length for each defect; Calculate the coordinates of the center of each defect relative to the vacuum pump impeller shaft to obtain the relative center coordinates of each defect; Calculate the Euclidean distance between the location coordinates of each defect and the center coordinates of the stress concentration region; The defect depth, projected area, longest axis length, relative center coordinates, and Euclidean distance are combined to form the three-dimensional geometric parameter vector of the defect.

[0008] Furthermore, the visual feature vector and three-dimensional geometric parameter vector are weighted and matched against a pre-defined defect physical type database to determine the physical type and confidence level of each defect. Specifically, this includes: Assign a first weight to the visual feature vector and a second weight to the three-dimensional geometric parameter vector; Calculate the sum of squared differences between the visual feature vector of each defect and the visual feature reference values ​​of various types of defects in the preset defect physical type database, as well as the sum of squared differences between the three-dimensional geometric parameter vector and the three-dimensional geometric parameter reference values ​​of various types of defects in the preset defect physical type database; The sum of squared differences in the visual feature reference values ​​of various defects and the product of the first weight are determined as the visual contribution components of various defects. The product of the sum of squared differences in the reference values ​​of the three-dimensional geometric parameters of various defects and the second weight is determined as the three-dimensional geometric components of various defects. The sum of the visual contribution components and the three-dimensional geometric components of various defects is determined as the weighted difference between the defect and all other defects. The defect type with the smallest weighted difference is determined as the physical type of the defect; The confidence level is calculated based on the ratio of the minimum weighted difference of the defect to the preset maximum matching degree, where the minimum weighted difference is negatively correlated with the confidence level.

[0009] Furthermore, when the physical type is crack, the corresponding mechanical model is determined to be a fracture mechanics model. The step of substituting material parameters and operating condition parameters into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard specifically includes: The critical crack length is determined based on the fracture toughness and local maximum stress in the material parameters. The remaining fatigue life is estimated as a risk indicator based on the current crack length, critical crack length, and local maximum stress.

[0010] Furthermore, when the physical type is wear or corrosion, the corresponding mechanical model is determined to be a dynamic equilibrium model. The calculation of risk indicators characterizing the degree of defect hazard by substituting material parameters and operating condition parameters into this mechanical model specifically includes: Calculate the amount of missing mass based on the volume of the defect and the material density in the material parameters; The unbalance is obtained by multiplying the mass loss amount by the distance from the defect center to the impeller shaft. The imbalance is used as a risk indicator.

[0011] Furthermore, when the physical type is porosity, the corresponding mechanical model is determined to be a finite element model. The step of substituting material parameters and operating condition parameters into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard specifically includes: Based on the depth, projected area, longest axis length of the defect, yield strength in the material parameters, and design rotation speed in the working condition parameters, a finite element model is established to calculate the local maximum stress at the defect. The safety factor is determined as a risk indicator based on the ratio of yield strength to local maximum stress.

[0012] Furthermore, based on risk indicators, the defects of the vacuum pump impeller are classified into severity levels, resulting in a classification result, specifically including: The calculated risk indicators are compared with the threshold ranges corresponding to each level; The level corresponding to the threshold range to which the risk indicator belongs is determined as the level result of the defect.

[0013] Furthermore, historical defect data of the vacuum pump impeller is obtained, the growth rate of defect size over time in the historical defect data is calculated, and the grading result is corrected according to the growth rate to obtain the final grading result, specifically including: Extract the defect size of the same defect at different points in time from historical defect data; The rate of increase of defect size over time is calculated based on the ratio of the change in defect size to the time interval. The ranking results are adjusted based on the growth rate to obtain the final ranking results.

[0014] This invention proposes a visual intelligent recognition system for vacuum pump impeller defects, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the visual intelligent recognition method for vacuum pump impeller defects.

[0015] The beneficial effects of the technical solution of the present invention are: In this embodiment of the invention, by fusing two-dimensional images with three-dimensional point clouds, weighted matching is used to accurately identify the physical type (cracks, porosity, wear, corrosion) and confidence level of defects. Based on this, fracture mechanics, dynamic balancing, or finite element models are adaptively selected according to the defect type. Combined with material properties and operating parameters, risk indicators such as remaining fatigue life, safety factor, or unbalance are quantitatively calculated, thus achieving a scientific risk level classification. At the same time, the growth rate of historical defect data is introduced to dynamically correct the level results, and finally, an inspection report oriented towards operation and maintenance decision-making is generated. This effectively bridges the gap between traditional visual recognition and engineering applications, and solves the problem that existing methods cannot distinguish the differences in location / operating condition risks and cannot support the predictive maintenance needs of in-service impellers. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a visual intelligent recognition method for vacuum pump impeller defects provided in one embodiment of the present invention; Figure 2 This is a structural diagram of a visual intelligent recognition system for vacuum pump impeller defects, provided in one embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a visual intelligent recognition method for vacuum pump impeller defects proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the visual intelligent recognition method for vacuum pump impeller defects provided by the present invention.

[0021] This invention provides a method and system for visual intelligent recognition of defects in vacuum pump impellers. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of a visual intelligent recognition method for vacuum pump impeller defects according to an embodiment of the present invention. The method includes the following steps: S101. Acquire two-dimensional image data and three-dimensional point cloud data of the vacuum pump impeller.

[0022] In this embodiment, the vacuum pump impeller is fixed on a rotating platform, and the binocular vision or structured light equipment is adjusted to be directly in front of the impeller to ensure that the equipment resolution is not less than 1920×1080 and the point cloud accuracy is not higher than 0.01mm.

[0023] For example, the device is started and the rotating platform is controlled to rotate at a constant speed in 15° increments. A set of two-dimensional RGB images and three-dimensional point cloud data are collected for each rotation step, for a total of 24 sets, covering the entire circumferential surface of the impeller.

[0024] The collected 3D point cloud was denoised using a statistical filtering method: the mean distance between each point and its neighboring points was calculated, and points whose distance exceeded three times the standard deviation of the mean were identified as outliers and removed, resulting in high-precision 3D point cloud data with a point cloud accuracy of no more than 0.01 mm.

[0025] The acquired two-dimensional RGB image is subjected to illumination compensation using histogram equalization: the gray-level histogram of the image is stretched from a concentrated distribution to a uniform distribution to eliminate the influence of uneven illumination on image quality, resulting in an illumination-compensated two-dimensional RGB image, which is stored in .jpg format and has an image contrast of not less than 0.3.

[0026] This completes the acquisition of two-dimensional image data and three-dimensional point cloud data of the vacuum pump impeller.

[0027] S102. Extract the visual feature vector of each defect from the two-dimensional image data, and extract the three-dimensional geometric parameter vector of each defect from the three-dimensional point cloud data.

[0028] In this embodiment, the visual feature vector of each defect is extracted from the two-dimensional image data, specifically including: The Otsu threshold segmentation method was used to extract the defect region, and the average gray level of the defect region was calculated. Construct a gray-level co-occurrence matrix and calculate texture contrast and texture entropy; The Canny edge detection algorithm is used to extract defect edges, and curves are fitted to the defect edge points to calculate the edge curvature. The average gray level, texture contrast, texture entropy, and edge curvature are combined to form the visual feature vector of the defect.

[0029] Extracting the 3D geometric parameter vector for each defect from the 3D point cloud data, specifically including: The ICP algorithm is used to register the 3D point cloud data with the standard CAD model to obtain the registered point cloud data. The point cloud data after registration is compared with the standard CAD model to obtain the point cloud of the defect area. The average distance between each defect point in the defect area point cloud and the corresponding point in the standard CAD model is determined as the defect depth. Calculate the projected area of ​​each defect on the vacuum pump impeller; Calculate the longest axis length for each defect; Calculate the coordinates of the center of each defect relative to the vacuum pump impeller shaft to obtain the relative center coordinates of each defect; Calculate the Euclidean distance between the location coordinates of each defect and the center coordinates of the stress concentration region; The defect depth, projected area, longest axis length, relative center coordinates, and Euclidean distance are combined to form the three-dimensional geometric parameter vector of the defect.

[0030] For example, defect identification of vacuum pump impellers is a necessary measure to ensure industrial safety and maintain the core performance of equipment. Impellers endure enormous centrifugal forces during high-speed rotation. If there are hidden pores, inclusions, or micro-cracks at the root, they are highly susceptible to rapid crack propagation or even fracture under alternating stress. The ejected fragments directly threaten personnel and equipment safety; therefore, accurate identification of hidden defects is the first line of defense against catastrophic accidents. Simultaneously, minute defects on the impeller surface caused by cavitation or wear can disrupt the precision clearances within the pump chamber, leading to a decrease in ultimate vacuum and reduced pumping efficiency. Only high-precision inspection can ensure that the pump body meets stringent process specifications over the long term. Traditional manual visual inspection has a high rate of missed inspections and is inefficient, while intelligent vision systems can complete 100% inspection within seconds, significantly reducing quality inspection costs and preventing defective products from flowing into subsequent processes. Furthermore, once the defect images and dimensional data accumulated by automated inspection are integrated into the manufacturing system, a full-process quality archive from raw material to service can be established. This allows for rapid tracing of the root cause of abnormalities and reverse optimization of casting process parameters, thereby driving the continuous evolution of vacuum pump manufacturing towards lean and intelligent closed-loop control.

[0031] Different types of defects exhibit significant differences in their visual features (grayscale, texture, edges), and defect patterns can be initially distinguished through quantified features. The steps for extracting the visual feature vector for each defect from two-dimensional image data specifically include: The Otsu thresholding method was used to extract the defect region, and the average gray level of the defect region was calculated. Average gray level. The calculation formula is:

[0032] in, For the first in the defect area grayscale value of each pixel. This represents the number of pixels in the defect area. This formula is used to quantify the brightness difference in the defect area.

[0033] Construct a gray-level co-occurrence matrix and calculate texture contrast and texture entropy. Texture contrast. The calculation formula is:

[0034] in, and This indicates that the gray values ​​in the gray-level co-occurrence matrix are respectively and , Represents the gray levels in the gray-level co-occurrence matrix. and The probability of both occurring simultaneously. This formula is used to quantify the sharpness of textures.

[0035] Texture Entropy The calculation formula is:

[0036] This formula is used to quantify the complexity of the texture. The distance of the gray-level co-occurrence matrix is ​​set to 1, and the angles are set to 0°, 45°, 90°, and 135° respectively. The average of the four angles is taken as the final texture contrast and texture entropy.

[0037] The Canny edge detection algorithm is used to extract defect edges, curves are fitted to the defect edge points, and the edge curvature is calculated. Edge curvature. The calculation formula is:

[0038] in, The first derivative of the edge curve is represented. This represents the second derivative of the edge curve. This formula is used to quantify the curvature of the edge.

[0039] The visual feature vector of this defect is constructed by combining average gray level, texture contrast, texture entropy, and edge curvature. .

[0040] In this embodiment, the step of extracting the three-dimensional geometric parameter vector of each defect from the three-dimensional point cloud data specifically includes: The ICP algorithm is used to register the 3D point cloud data with the standard CAD model, and the registration error is controlled within 0.02mm to obtain the registered point cloud data.

[0041] Differential calculations are performed on the registered point cloud data and the standard CAD model to obtain the point cloud of the defect area.

[0042] The defect depth is determined by the average distance between each defect point cloud and its corresponding point in the standard CAD model. The calculation formula is:

[0043] in, Indicates the defect area. The distance between the point cloud and the CAD model. This represents the number of point clouds in the defect area. This formula is used to quantify the depth of the defect.

[0044] Calculate the projected area of ​​each defect on the vacuum pump impeller. .

[0045] Calculate the longest axis length for each defect. .

[0046] Calculate the coordinates of the center of each defect relative to the vacuum pump impeller shaft to obtain the relative center coordinates of each defect. .

[0047] Calculate the Euclidean distance between the location coordinates of each defect and the center coordinates of the stress concentration region. The calculation formula is:

[0048] in, These are the coordinates of the center of the stress concentration region. This formula is used to quantify the distance between defects and high-risk areas.

[0049] The defect depth, projected area, longest axis length, relative center coordinates, and Euclidean distance are combined to form the three-dimensional geometric parameter vector of the defect. .

[0050] S103. Perform weighted matching between the visual feature vector and the three-dimensional geometric parameter vector and the preset defect physical type database to determine the physical type and confidence level of each defect. The preset defect physical type database includes at least the feature threshold ranges corresponding to cracks, pores, wear and corrosion.

[0051] In this embodiment, the visual feature vector and the three-dimensional geometric parameter vector are weighted and matched with a preset defect physical type database to determine the physical type and confidence level of each defect. Specifically, this includes: Assign a first weight to the visual feature vector and a second weight to the three-dimensional geometric parameter vector; Calculate the sum of squared differences between the visual feature vector of each defect and the visual feature reference values ​​of various types of defects in the preset defect physical type database, as well as the sum of squared differences between the three-dimensional geometric parameter vector and the three-dimensional geometric parameter reference values ​​of various types of defects in the preset defect physical type database; The sum of squared differences in the visual feature reference values ​​of various defects and the product of the first weight are determined as the visual contribution components of various defects. The product of the sum of squared differences in the reference values ​​of the three-dimensional geometric parameters of various defects and the second weight is determined as the three-dimensional geometric components of various defects. The sum of the visual contribution components and the three-dimensional geometric components of various defects is determined as the weighted difference between the defect and all other defects. The defect type with the smallest weighted difference is determined as the physical type of the defect; The confidence level is calculated based on the ratio of the minimum weighted difference of the defect to the preset maximum matching degree, where the minimum weighted difference is negatively correlated with the confidence level.

[0052] For example, the visual feature vector and the three-dimensional geometric parameter vector are weighted and matched against a pre-defined defect physical type database to determine the physical type and confidence level of each defect, specifically including: Assign first weights to visual feature vectors Assign a second weight to the three-dimensional geometric parameter vector. .

[0053] For example, the first weight Set to 0.4, second weight The weight is set to 0.6. The principle behind this weight setting is as follows: three-dimensional geometric parameters (such as defect depth, projected area, length, etc.) directly reflect the degree of physical damage caused by the defect to the impeller structure and have a stronger ability to distinguish defect types, therefore they are given a higher weight; visual features (such as grayscale, texture, edge curvature) are mainly used to initially distinguish the appearance of defects and play an auxiliary role in determining the physical type, therefore they are given a relatively lower weight. It should be noted that the above weight values ​​are illustrative examples, and those skilled in the art can adjust the weights according to the actual detection scenario and accuracy requirements; these adjustments still fall within the protection scope of this invention.

[0054] Calculate the sum of squared differences between the visual feature vector of each defect and the visual feature reference values ​​of various types of defects in the preset defect physical type database, as well as the sum of squared differences between the three-dimensional geometric parameter vector and the three-dimensional geometric parameter reference values ​​of various types of defects in the preset defect physical type database.

[0055] The sum of squared differences in the visual feature reference values ​​of various defects, multiplied by the first weight, is determined as the visual contribution component of each defect.

[0056] The product of the sum of squared differences in the reference values ​​of the three-dimensional geometric parameters of various defects and the second weight is determined as the three-dimensional geometric components of various defects.

[0057] The sum of the visual contribution components and the three-dimensional geometric components of each type of defect is determined as the weighted difference between that defect and all other types of defects. The calculation formula is:

[0058] in, The first visual feature vector representing the defect Item parameters, This represents the first visual feature reference value in the preset defect physical type database corresponding to a certain type of defect. Item parameters; The third element in the three-dimensional geometric parameter vector representing the defect Item parameters, This represents the first reference value of the three-dimensional geometric parameters corresponding to a certain type of defect in the preset defect physical type database. The formula quantifies the degree of matching between the defect and the reference type by calculating a weighted sum of feature differences. The smaller the value, the higher the match.

[0059] The defect type with the smallest weighted difference is determined as the physical type of the defect.

[0060] The confidence level is calculated based on the ratio of the minimum weighted difference to the preset maximum match, where the minimum weighted difference is negatively correlated with the confidence level. (Confidence Level) The calculation formula is:

[0061] in, This represents the minimum weighted difference. This indicates the preset maximum match score. The setting is based on the theoretical maximum difference range of the feature parameters: when each parameter in the visual feature vector and the 3D geometric parameter vector takes the maximum possible difference, the calculated weighted difference is used as the preset maximum matching degree, ensuring that the confidence output is between 0% and 100%. This formula converts the matching degree into a confidence degree, intuitively reflecting the reliability of the judgment result.

[0062] For example, in the preset defect physical type database, the characteristic threshold range for crack-type defects is: linear defects with a length greater than 5 mm, a depth greater than 0.5 mm, and a texture contrast greater than 0.6. The basis for setting this threshold is that cracks usually present a linear shape, are relatively long and deep, and have sharp edges that result in high texture contrast.

[0063] The characteristic threshold range for pore-like defects is: projected area less than 10 mm², defect depth less than 0.3 mm, and texture entropy less than 0.8. This threshold setting is based on the fact that pores typically appear as dots or nearly circular shapes, are small in size and shallow in depth, and have uniform internal grayscale, resulting in low texture entropy.

[0064] The characteristic threshold range for wear defects is as follows: projected area greater than 15 mm², defect depth between 0.1 mm and 0.5 mm, and edge curvature less than 0.1. This threshold is based on the fact that wear typically presents as a planar distribution, with a large area, shallow depth, and gentle edges resulting in low curvature.

[0065] The characteristic threshold range for corrosion defects is as follows: texture entropy greater than 1.2, average gray level more than 20% lower than background gray level, and defect depth greater than 0.3 mm. The basis for setting this threshold is that corrosion areas typically have complex textures, low gray levels, and deep depths.

[0066] It should be noted that the specific values ​​of the above-mentioned feature threshold range are illustrative examples, and those skilled in the art can make adaptive adjustments to the above thresholds based on the actual detection object, detection accuracy requirements, and engineering experience.

[0067] S104. When the confidence level is less than or equal to the preset confidence threshold, the defect is manually reviewed; when the confidence level is greater than the preset confidence threshold, the material parameters and operating parameters of the vacuum pump impeller are obtained according to the physical type, a mechanical model corresponding to the physical type is selected, and the material parameters and operating parameters are substituted into the mechanical model to calculate the risk index used to characterize the degree of defect danger. The mechanical model includes a fracture mechanics model, a dynamic balance model, or a finite element model, and the risk index includes at least one of the following: remaining fatigue life, safety factor, and unbalance.

[0068] In this embodiment, when the physical type is crack, the mechanical model corresponding to this physical type is determined to be a fracture mechanics model. Material parameters and operating condition parameters are substituted into this mechanical model to calculate a risk index used to characterize the degree of defect hazard, specifically including: The critical crack length is determined based on the fracture toughness and local maximum stress in the material parameters. The remaining fatigue life is estimated as a risk indicator based on the current crack length, critical crack length, and local maximum stress.

[0069] When the physical type is wear or corrosion, the corresponding mechanical model is determined to be a dynamic equilibrium model. Material parameters and operating condition parameters are substituted into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard, specifically including: Calculate the amount of missing mass based on the volume of the defect and the material density in the material parameters; The unbalance is obtained by multiplying the mass loss amount by the distance from the defect center to the impeller shaft. The imbalance is used as a risk indicator.

[0070] When the physical type is porosity, the corresponding mechanical model is determined to be a finite element model. Material parameters and operating condition parameters are substituted into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard, specifically including: Based on the depth, projected area, longest axis length of the defect, yield strength in the material parameters, and design rotation speed in the working condition parameters, a finite element model is established to calculate the local maximum stress at the defect. The safety factor is determined as a risk indicator based on the ratio of yield strength to local maximum stress.

[0071] For example, further classifying defects based on accurate detection is a crucial bridge for transforming detection data into actionable maintenance decisions. The impact of defects of varying severity on impeller safety margins differs significantly. Micron-level surface pitting may only require monitoring, while deep cracks reaching the root necessitate immediate shutdown and replacement. Without clear classification, operators are easily caught in a dilemma when faced with alarm signals: either "over-maintenance wastes capacity" or "operating with defects creates hidden dangers." By introducing quantitative classification standards, the system can automatically determine "qualified, downgraded, repaired, or scrapped" based on the correlation between defect size, location, and stress-sensitive areas. This not only makes spare parts scheduling and maintenance planning more data-driven but also allows for the establishment of impeller remaining life prediction models by long-term tracking of the expansion rate of defects at each level. Simultaneously, massive amounts of classification data flow back to the casting process, accurately revealing the processes that generate specific levels of porosity or inclusions, providing fine-grained guidance for process parameter optimization. This truly achieves a leap from "post-event alarms" to "predictive maintenance" and a "lean manufacturing closed loop," ensuring the optimal balance between equipment safety and economic benefits.

[0072] In this embodiment, based on the physical parameters of defects, material properties, and operating conditions, the impact of defects on impeller performance is quantified through an engineering physical model, achieving scientific classification and solving the problems of traditional methods that cannot distinguish the differences in location / operating condition risks and cannot support operation and maintenance decisions.

[0073] When the confidence level is less than or equal to a preset confidence threshold (e.g., 80%), the defect is marked as requiring manual review and will not proceed to the subsequent automatic risk indicator calculation process. Manual review involves inspection personnel conducting a secondary verification of the defect image and point cloud data to determine the authenticity and type of the defect. The review results can be used as system feedback data for subsequent model optimization.

[0074] When the confidence level is greater than the preset confidence threshold, the corresponding risk indicator is calculated based on the physical type.

[0075] When the physical type is crack, the corresponding mechanical model is determined to be a fracture mechanics model. Material parameters and operating condition parameters are substituted into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard, specifically including: Based on the fracture toughness in the material parameters and local maximum stress Determine the critical crack length Critical crack length The calculation formula is:

[0076] in, This is a geometric factor, depending on the crack shape and component geometry. For example, for a surface crack at the impeller blade root, Take 1.12; The maximum local stress at the defect is calculated using centrifugal load based on the design rotation speed in the operating parameters.

[0077] Based on the current crack length Critical crack length and local maximum stress The remaining fatigue life is estimated as a risk indicator. The calculation uses the Paris formula:

[0078] in, Indicates the crack propagation rate. and This represents a material constant, obtained through material fatigue testing. The range of stress intensity factor is represented by the following formula: , Geometric factor This represents the current crack length. The Paris formula is applied from the current crack length. To the critical crack length Integrate to obtain the remaining fatigue life. :

[0079] For example, for 304 stainless steel, under typical operating conditions, Pick , Taking 3.0, the calculated remaining fatigue life can be used as a risk indicator for crack-type defects to determine the remaining safe service life of the impeller.

[0080] When the physical type is wear or corrosion, the corresponding mechanical model is determined to be a dynamic equilibrium model. Material parameters and operating condition parameters are substituted into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard, specifically including: Based on the volume of the defect Material density in material parameters Calculate the amount of missing quality The formula for calculating the amount of missing quality is:

[0081] The defect volume is calculated by three-dimensional reconstruction of the point cloud of the defect region. For example, it is obtained by triangulating the point cloud of the defect region and calculating the volume enclosed by the triangular facets. The material density is obtained from the material mechanical property database based on the material grade.

[0082] Calculate the amount of missing quality and the distance from the defect center to the impeller shaft The product of these two quantities yields the unbalance quantity. :

[0083] in, The distance from the defect center to the impeller shaft is calculated based on the relative center coordinates. The modulus is obtained by calculation.

[0084] Imbalance As a risk indicator. For example, when the imbalance exceeds the impeller's maximum permissible imbalance (e.g., 0.5g). When the speed reaches 0.5 mm, it indicates that the impeller will generate significant vibration under high-speed rotation, and dynamic balancing correction measures need to be taken.

[0085] When the physical type is porosity, the corresponding mechanical model is determined to be a finite element model. Material parameters and operating condition parameters are substituted into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard, specifically including: Based on the depth of the defect Projected area Longest axis length Yield strength in material parameters and the design speed in the operating parameters Establish a finite element model to calculate the local maximum stress at the defect. .

[0086] For example, the three-dimensional geometric parameters of the defect are imported into finite element analysis software (such as ANSYS) to establish a finite element model of the impeller. The mesh of the defect region is refined, with a mesh size no larger than 0.1 mm, to ensure calculation accuracy. A centrifugal load is then applied to the finite element model. , Angular velocity, For impeller mass, Let be the distance from the defect center to the impeller shaft. The local maximum stress at the defect is obtained through finite element analysis. .

[0087] According to yield strength With local maximum stress The ratio of the two factors determines the safety factor. :

[0088] For example, for 304 stainless steel, the yield strength Take 205 MPa. When the safety factor is greater than 1.5, it indicates that the structural safety margin is sufficient; when the safety factor is between 1.0 and 1.5, it indicates that the structure is in a critical state; when the safety factor is less than 1.0, it indicates that the structure has a risk of failure.

[0089] Thus, based on different physical types, the remaining fatigue life of crack-type defects, the unbalance of wear / corrosion-type defects, and the safety factor of porosity-type defects are obtained as risk indicators.

[0090] S105. Classify the severity level of defects in the vacuum pump impeller according to the risk indicators, and obtain the level results.

[0091] In this embodiment, the defects of the vacuum pump impeller are classified into severity levels according to risk indicators to obtain the level results, specifically including: The calculated risk indicators are compared with the threshold ranges corresponding to each level; The level corresponding to the threshold range to which the risk indicator belongs is determined as the level result of the defect.

[0092] For example, the calculated risk index is compared with the threshold range corresponding to each level; the level corresponding to the threshold range to which the risk index belongs is determined as the level result of the defect.

[0093] Pre-defined severity level standards are established, and feasible severity level standards are developed based on risk indicators and engineering experience to ensure that the classification results can support operation and maintenance decisions.

[0094] For example, four severity levels are preset: mild, moderate, severe, and extremely severe, with each level corresponding to a different risk indicator threshold range.

[0095] For crack-type defects, the remaining fatigue life is used as the risk indicator for classification: When the remaining fatigue life is greater than During the next cycle, it is classified as a minor level, indicating that the defect has minimal impact on impeller safety and can continue to operate normally. When the remaining fatigue life is between to Between cycles, it is classified as medium level, indicating that the defect has a certain risk of expansion, and it is recommended to strengthen monitoring; When the remaining fatigue life is between to Between cycles, it is classified as severe, indicating a high risk of defect propagation, and it is recommended to schedule maintenance in the short term; When the remaining fatigue life is less than During the next cycle, it is classified as extremely serious, indicating that the defect is about to cause failure and the machine should be stopped and replaced immediately.

[0096] For wear or corrosion defects, the degree of imbalance is used as a risk indicator for classification: When the imbalance is less than 0.1g When the temperature is mm, it is classified as a minor level; When the imbalance is between 0.1g mm to 0.5g When the value is between mm, it is classified as medium grade; When the imbalance is between 0.5g mm to 1.0g When the value is between mm, it is classified as a severe level; When the imbalance is greater than 1.0g When the temperature reaches mm, it is classified as extremely severe.

[0097] For porosity defects, the safety factor is used as a risk indicator for classification: When the safety factor is greater than 2.0, it is classified as a minor level; When the safety factor is between 1.5 and 2.0, it is classified as medium level; When the safety factor is between 1.0 and 1.5, it is classified as a severe level; When the safety factor is less than 1.0, it is classified as extremely serious.

[0098] It should be noted that the specific threshold values ​​for the above-mentioned classification are illustrative examples, and those skilled in the art can make adaptive adjustments to the above thresholds based on actual equipment requirements, safety standards, and engineering experience.

[0099] S106. Obtain historical defect data of the vacuum pump impeller, calculate the growth rate of defect size over time in the historical defect data, and correct the grade result according to the growth rate to obtain the final grade result.

[0100] In this embodiment, historical defect data of the vacuum pump impeller is acquired, the growth rate of defect size over time in the historical defect data is calculated, and the grade result is corrected according to the growth rate to obtain the final grade result. Specifically, this includes: Extract the defect size of the same defect at different points in time from historical defect data; The rate of increase of defect size over time is calculated based on the ratio of the change in defect size to the time interval. The ranking results are adjusted based on the growth rate to obtain the final ranking results.

[0101] For example, for in-service impellers, defects expand over time, and combining historical growth trends can more accurately assess the current risk. In this embodiment, historical defect data of the vacuum pump impeller is obtained, the growth rate of defect size over time in the historical defect data is calculated, and the rating result is corrected according to the growth rate to obtain the final rating result, specifically including: Extract the dimensions of the same defect at different points in time from historical defect data. For example, import the historical defect detection records of the impeller from the operation and maintenance management system, including the dimensions (such as crack length), location, and detection time of the same defect recorded in each detection.

[0102] Calculate the rate of increase in defect size over time based on the ratio of the change in defect size to the time interval. The formula for calculating the growth rate is:

[0103] in, Indicates the size of the defect currently detected. This indicates the size of defects detected in the past. This represents the time interval between two detections. This formula is used to quantify the rate of defect growth, reflecting the development trend of risk.

[0104] The ranking results are adjusted based on the growth rate to obtain the final ranking results.

[0105] For example, a first threshold and a second threshold are preset. The calculated growth rate is compared with the first threshold and the second threshold: When the growth rate exceeds the first threshold, the grade result is upgraded by one level (e.g., from moderate to severe). When the growth rate is between the second threshold and the first threshold, the grade result is promoted by half a grade (e.g., from medium grade to medium+ grade). When the growth rate is less than or equal to the second threshold, the grading results remain unchanged.

[0106] For example, the first threshold is set to 0.1 mm / h, and the second threshold is set to 0.05 mm / h. When the crack propagation rate exceeds 0.1 mm / h, it indicates that the defect is propagating rapidly, the risk is significantly increased, and the level needs to be upgraded by one level; when the propagation rate is between 0.05 mm / h and 0.1 mm / h, it indicates that the defect has a certain propagation trend, and the level is upgraded by half a level; when the propagation rate is below 0.05 mm / h, it indicates that the defect is propagating slowly, and the level remains unchanged.

[0107] It should be noted that the specific values ​​of the first threshold, the second threshold, and the adjustment coefficient mentioned above are illustrative examples. Those skilled in the art can make adaptive adjustments to the above parameters according to actual equipment requirements, safety standards, and engineering experience.

[0108] S107. Based on the final grade results, generate a defect monitoring report for the vacuum pump impeller.

[0109] In this embodiment, a defect monitoring report for the vacuum pump impeller is generated based on the final grade result, specifically including: Integrate defect location, physical type, 3D dimensions, severity level, risk impact analysis, and maintenance recommendations to generate a visual inspection report.

[0110] For example, the location of defects is marked on the three-dimensional model of the impeller to visually demonstrate the spatial distribution of the defects. The report lists the physical type of each defect (such as crack, porosity, wear or corrosion), three-dimensional dimensions (such as defect depth, projected area, longest axis length, etc.), and the final rating result (such as minor, moderate, severe, and extremely severe).

[0111] Risk impact analysis is tailored to the defect type and severity level. For example, for a crack-like defect located at the blade root, if the severity level is "severe," the risk impact analysis would be: "The crack is located in the stress concentration area at the blade root, with a remaining fatigue life of approximately..." In subsequent cycles, there is a risk of expansion leading to fracture. For wear or corrosion defects, if the imbalance is large, the risk impact analysis is: "The defect causes the impeller dynamic imbalance to exceed the allowable range, which may cause severe vibration during high-speed operation." For porosity defects, if the safety factor is low, the risk impact analysis is: "Porosity defects lead to localized stress concentration, resulting in insufficient structural safety margin." The maintenance recommendations are generated based on the final severity level. For example, for minor defects, the recommendation is: "Continue use, monitor according to the regular cycle." For medium defects, the recommendation is: "Downgrade use, shorten the detection cycle, and focus on key areas." For severe defects, the recommendation is: "Repair within a specified time, and schedule a shutdown for maintenance within 3 months." For extremely severe defects, the recommendation is: "Immediate shutdown and replacement, prohibit further operation."

[0112] The inspection report can be generated in PDF format, and includes two-dimensional images of the impeller, defect annotation diagrams of the three-dimensional model, and risk indicator charts (such as bar charts of remaining fatigue life, trend charts of unbalance, and comparison charts of safety factors), which makes it easy for maintenance personnel to intuitively understand the inspection results.

[0113] In addition, all data from this inspection (including visual feature vectors, three-dimensional geometric parameter vectors, defect physical types, severity levels, risk indicators, and maintenance recommendations) will be stored in a database to establish a full lifecycle file of defects and associate it with a unique impeller identifier (such as the impeller serial number or equipment number) to facilitate subsequent querying, comparative analysis, and dynamic risk assessment.

[0114] In summary, in this embodiment of the invention, by constructing a progressive framework of "multi-source information fusion + physical model-driven," firstly, two-dimensional images and three-dimensional point clouds are acquired using binocular vision or structured light. The visual feature vectors and three-dimensional geometric parameter vectors of defects are extracted and weighted matching is performed with a preset defect physical type database to accurately identify the physical type (cracks, porosity, wear, corrosion) and confidence level of defects. For defects with a confidence level higher than the threshold, a fracture mechanics model, dynamic balancing model, or finite element model is adaptively selected based on their physical type. Combined with material parameters and operating condition parameters, risk indicators such as remaining fatigue life, unbalance, or safety factor are quantitatively calculated to achieve a scientific risk level classification. Simultaneously, for in-service impellers, historical defect data is introduced to calculate the growth rate of defect size over time, and the level results are dynamically corrected accordingly. Finally, a visualized inspection report containing defect location, physical type, three-dimensional size, risk indicators, and operation and maintenance suggestions is generated. This method effectively bridges the gap between traditional visual recognition and engineering applications, solving the problems that existing technologies can only determine "whether there is a defect" and cannot distinguish the risk differences under different locations and operating conditions, and cannot support the predictive maintenance needs of in-service impellers. It significantly improves the intelligence level of vacuum pump impeller defect detection and operation and maintenance decision support capabilities.

[0115] This invention also proposes a visual intelligent recognition system for vacuum pump impeller defects; please refer to [link / reference]. Figure 2 The diagram shows a structural diagram of a vacuum pump impeller defect visual intelligent recognition system provided in an embodiment of the present invention. The system includes: a data acquisition module 101, a data analysis module 102, and a defect recognition module 103.

[0116] The data acquisition module 101 is used to acquire two-dimensional image data and three-dimensional point cloud data of the vacuum pump impeller; The data analysis module 102 is used to extract the visual feature vector of each defect from the two-dimensional image data and the three-dimensional geometric parameter vector of each defect from the three-dimensional point cloud data. The visual feature vector and the three-dimensional geometric parameter vector are weighted and matched with a preset defect physical type database to determine the physical type and confidence level of each defect. The preset defect physical type database includes at least the feature threshold ranges corresponding to cracks, pores, wear and corrosion. When the confidence level is less than or equal to the preset confidence threshold, the defect is manually reviewed. When the confidence level is greater than the preset confidence threshold, the material parameters and operating parameters of the vacuum pump impeller are obtained according to the physical type. The mechanical model corresponding to the physical type is selected, and the material parameters and operating parameters are substituted into the mechanical model to calculate the risk index used to characterize the degree of defect danger. The mechanical model includes a fracture mechanics model, a dynamic balance model or a finite element model, and the risk index includes at least one of the following: remaining fatigue life, safety factor and unbalance amount. The defects of the vacuum pump impeller are classified into severity levels based on risk indicators, and the level results are obtained. The historical defect data of the vacuum pump impeller is obtained, the growth rate of the defect size over time in the historical defect data is calculated, and the grade result is corrected according to the growth rate to obtain the final grade result. The defect identification module 103 is used to generate a defect monitoring report for the vacuum pump impeller based on the final grade results.

[0117] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the vacuum pump impeller defect visual intelligent recognition system and the vacuum pump impeller defect visual intelligent recognition method embodiment provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiment, which will not be repeated here.

[0118] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0119] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0120] 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 principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for visual intelligent recognition of defects in a vacuum pump impeller, characterized in that, include: Acquire two-dimensional image data and three-dimensional point cloud data of the vacuum pump impeller; The visual feature vector of each defect is extracted from the two-dimensional image data, and the three-dimensional geometric parameter vector of each defect is extracted from the three-dimensional point cloud data. The visual feature vector and the three-dimensional geometric parameter vector are weighted and matched with a preset defect physical type database to determine the physical type and confidence level of each defect. The preset defect physical type database includes at least the feature threshold ranges corresponding to cracks, pores, wear and corrosion. When the confidence level is less than or equal to the preset confidence threshold, the defect is manually reviewed. When the confidence level is greater than the preset confidence threshold, the material parameters and operating parameters of the vacuum pump impeller are obtained according to the physical type. The mechanical model corresponding to the physical type is selected, and the material parameters and operating parameters are substituted into the mechanical model to calculate the risk index used to characterize the degree of defect danger. The mechanical model includes a fracture mechanics model, a dynamic balance model or a finite element model, and the risk index includes at least one of the following: remaining fatigue life, safety factor and unbalance amount. The defects of the vacuum pump impeller are classified into severity levels based on risk indicators, and the level results are obtained. The historical defect data of the vacuum pump impeller is obtained, the growth rate of the defect size over time in the historical defect data is calculated, and the grade result is corrected according to the growth rate to obtain the final grade result. Based on the final grade results, a defect monitoring report for the vacuum pump impeller is generated.

2. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, The extraction of the visual feature vector for each defect from the two-dimensional image data specifically includes: The Otsu threshold segmentation method was used to extract the defect region, and the average gray level of the defect region was calculated. Construct a gray-level co-occurrence matrix and calculate texture contrast and texture entropy; The Canny edge detection algorithm is used to extract defect edges, and curves are fitted to the defect edge points to calculate the edge curvature. The average gray level, texture contrast, texture entropy, and edge curvature are combined to form the visual feature vector of the defect.

3. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, The extraction of the three-dimensional geometric parameter vector for each defect from the three-dimensional point cloud data specifically includes: The ICP algorithm is used to register the 3D point cloud data with the standard CAD model to obtain the registered point cloud data. The point cloud data after registration is compared with the standard CAD model to obtain the point cloud of the defect area. The average distance between each defect point in the defect area point cloud and the corresponding point in the standard CAD model is determined as the defect depth. Calculate the projected area of ​​each defect on the vacuum pump impeller; Calculate the longest axis length for each defect; Calculate the coordinates of the center of each defect relative to the vacuum pump impeller shaft to obtain the relative center coordinates of each defect; Calculate the Euclidean distance between the location coordinates of each defect and the center coordinates of the stress concentration region; The defect depth, projected area, longest axis length, relative center coordinates, and Euclidean distance are combined to form the three-dimensional geometric parameter vector of the defect.

4. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, The step of performing a weighted matching between visual feature vectors and three-dimensional geometric parameter vectors and a preset defect physical type database to determine the physical type and confidence level of each defect specifically includes: Assign a first weight to the visual feature vector and a second weight to the three-dimensional geometric parameter vector; Calculate the sum of squared differences between the visual feature vector of each defect and the visual feature reference values ​​of various types of defects in the preset defect physical type database, as well as the sum of squared differences between the three-dimensional geometric parameter vector and the three-dimensional geometric parameter reference values ​​of various types of defects in the preset defect physical type database; The sum of squared differences in the visual feature reference values ​​of various defects and the product of the first weight are determined as the visual contribution components of various defects. The product of the sum of squared differences in the reference values ​​of the three-dimensional geometric parameters of various defects and the second weight is determined as the three-dimensional geometric components of various defects. The sum of the visual contribution components and the three-dimensional geometric components of various defects is determined as the weighted difference between the defect and all other defects. The defect type with the smallest weighted difference is determined as the physical type of the defect; The confidence level is calculated based on the ratio of the minimum weighted difference of the defect to the preset maximum matching degree, where the minimum weighted difference is negatively correlated with the confidence level.

5. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, When the physical type is crack, the corresponding mechanical model is determined to be a fracture mechanics model. The step of substituting material parameters and operating condition parameters into this mechanical model to calculate risk indicators used to characterize the degree of defect hazard specifically includes: The critical crack length is determined based on the fracture toughness and local maximum stress in the material parameters. The remaining fatigue life is estimated as a risk indicator based on the current crack length, critical crack length, and local maximum stress.

6. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, When the physical type is wear or corrosion, the corresponding mechanical model is determined to be a dynamic equilibrium model. The calculation of risk indicators characterizing the degree of defect hazard by substituting material parameters and operating condition parameters into this mechanical model specifically includes: Calculate the amount of missing mass based on the volume of the defect and the material density in the material parameters; The unbalance is obtained by multiplying the mass loss amount by the distance from the defect center to the impeller shaft. The imbalance is used as a risk indicator.

7. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, When the physical type is porosity, the corresponding mechanical model is determined to be a finite element model. The step of substituting material parameters and operating condition parameters into this mechanical model to calculate risk indicators characterizing the degree of defect hazard specifically includes: Based on the depth, projected area, longest axis length of the defect, yield strength in the material parameters, and design rotation speed in the working condition parameters, a finite element model is established to calculate the local maximum stress at the defect. The safety factor is determined as a risk indicator based on the ratio of yield strength to local maximum stress.

8. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, The method of classifying the severity of defects in the vacuum pump impeller based on risk indicators to obtain a classification result specifically includes: The calculated risk indicators are compared with the threshold ranges corresponding to each level; The level corresponding to the threshold range to which the risk indicator belongs is determined as the level result of the defect.

9. The method for visual intelligent recognition of defects in a vacuum pump impeller according to claim 1, characterized in that, The process of acquiring historical defect data of the vacuum pump impeller, calculating the growth rate of defect size over time in the historical defect data, and correcting the grade result based on the growth rate to obtain the final grade result specifically includes: Extract the defect size of the same defect at different points in time from historical defect data; The rate of increase of defect size over time is calculated based on the ratio of the change in defect size to the time interval. The ranking results are adjusted based on the growth rate to obtain the final ranking results.

10. A visual intelligent recognition system for defects in a vacuum pump impeller, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the visual intelligent recognition method for vacuum pump impeller defects as described in any one of claims 1-9.

Images