Real-time detection system for three-dimensional mold surface defects based on deep learning

Abstract

The application relates to the technical field of defect detection, and specifically discloses a three-dimensional mold surface defect real-time detection system based on deep learning, which comprises the following modules: a data collection module: a mold area is divided through a data recording unit, point cloud and image data at each moment are recorded, and a historical data set is generated; a preprocessing unit generates a geometric defect positioning curve by using historical point cloud data and generates an RGB defect positioning curve based on image data; a data analysis module: a safety interval is determined according to the geometric defect positioning curve, defect frequencies of each area are counted, and a high-frequency defect area is identified; an RGB related area is positioned through the RGB defect positioning curve; a point cloud density correction module: an RGB sensitive area is determined in combination with the RGB related area, and an initial point cloud density is set; during detection, the RGB value of the RGB sensitive area is monitored in real time, and other areas are scanned at the initial density; after the safety interval ends, the point cloud density of each area is dynamically adjusted according to the defect frequency, and the detection efficiency is optimized.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and more specifically to a real-time detection system for surface defects of three-dimensional molds based on deep learning. Background Technology

[0002] In the existing technological system, surface inspection of 3D molds typically involves acquiring point cloud data of the 3D mold using specific measurement equipment or techniques. After acquiring the point cloud data, a 3D mold is modeled, and the original model is compared with the real-time acquired model. If these deviations exceed the pre-set allowable range, it can be determined that the 3D mold has defects, such as surface wear, deformation, scratches, etc.

[0003] This modeling process is not simply a matter of piling up data; rather, it requires the use of advanced mathematical models and algorithms to process and analyze large amounts of point cloud data. Through preprocessing operations such as filtering, noise reduction, and stitching of the point cloud data, noise points and outliers are removed. Then, the discrete point cloud data is fitted into a continuous surface model to accurately reproduce the true shape and surface features of the 3D mold.

[0004] However, in real-world production environments, surface inspection of 3D molds often needs to be completed within a very short timeframe to promptly identify and address problems, ensuring production continuity and product quality stability. High-precision 3D reconstruction algorithms suffer from excessively high computational complexity, requiring significant computational resources and time to run. Even with high-performance computing equipment, real-time performance requirements are difficult to meet. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time detection system for surface defects of three-dimensional molds based on deep learning, and to solve the following technical problems.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A deep learning-based real-time detection system for 3D mold surface defects includes:

[0008] Data collection module: a data recording unit and a preprocessing unit. The data recording unit divides the mold into several areas to record point cloud data and image data of the mold at various times during its service life. Through the data recording unit, historical point cloud data and historical image data of the mold are obtained.

[0009] The preprocessing unit generates a geometric defect location curve based on the historical point cloud data, and generates an RGB defect location curve based on the historical image data.

[0010] Data analysis module: Based on the geometric defect location curve, determine the safety range and obtain the defect frequency and high-frequency defect area of ​​each region of the mold; based on the RGB defect location curve, determine the RGB related area;

[0011] Point cloud density correction module: Based on the RGB related regions, obtain the RGB sensitive regions and set the initial density of the point cloud; when monitoring the surface defects of the mold, monitor the RGB values ​​of the RGB sensitive regions in real time, and monitor the point cloud data of the remaining regions in real time with the initial density within the safe range; after the safe range ends, adjust the point cloud density of each region of the mold according to the defect frequency of each region of the mold.

[0012] As a further aspect of the present invention: the usage period is the time interval between the time when the mold is put into use and the time when it is put out of use, wherein the time when the mold is put into use is the time when the mold is put out of use; and according to a preset time interval, a number of times are selected equally within the usage period; the image data includes images of each side of the mold.

[0013] As a further aspect of the present invention: the process of obtaining the point cloud model and image data includes:

[0014] A three-dimensional coordinate system is established, the mold is placed in the three-dimensional coordinate system, the center points of each region of the mold are obtained, and each center point is numbered; and the three-dimensional coordinates corresponding to each center point in the three-dimensional coordinate system are obtained to obtain the point cloud data of the mold; the image data includes the RGB values ​​of each region of the mold.

[0015] As a further aspect of the present invention: the process of generating the geometric defect location curve includes:

[0016] When the surface of the mold is free of defects, the three-dimensional coordinates of the center points of each region are obtained and recorded as standard coordinates. For the mold at any time, the distance between the three-dimensional coordinates of each center point and the standard coordinates is obtained and recorded as the distance difference. A rectangular coordinate system is established with the center point number as the abscissa and the distance difference as the ordinate. In the rectangular coordinate system, the center point of each number and its corresponding distance difference are fitted to obtain the geometric defect location curve.

[0017] As a further aspect of the present invention: the process of determining the safe zone includes:

[0018] Within the usage period, obtain the slope of each point on the geometric defect location curve at each moment, obtain the sum of the slopes at each point, and obtain the fluctuation value of the geometric defect location curve; record the moment corresponding to the geometric defect location curve when the fluctuation value is first not 0, and denote it as the defect moment.

[0019] Obtain the defect times of all grinding tools during their service life to obtain a set of defect times, and select the minimum value in the set of defect times as the safe cut-off time; obtain the safe interval from the time of use and the safe cut-off time.

[0020] As a further aspect of the present invention: the process of obtaining the defect frequency of each region of the mold includes:

[0021] Obtain the geometric defect location curve of the mold at the end of its use, and denot it as the end time curve; based on the historical point cloud data, obtain the end time curves of all molds; for any numbered center point, obtain the number of end time curves in all end time curves where the vertical coordinate value of the center point is not 0, and obtain the defect frequency f=n / N of the region corresponding to the center point, where n is the number of end time curves where the vertical coordinate value is not 0, and N is the total number of all end time curves.

[0022] As a further aspect of the present invention: the process of determining the RGB related region includes:

[0023] The time corresponding to the first non-zero ordinate value of the geometric defect positioning curve in all geometric defect positioning curves of the high-frequency defect region during its usage cycle is obtained and denoted as the first defect time. The RGB defect positioning curve at the first defect time is obtained and denoted as the first curve. The center point number of the high-frequency defect region is obtained, and the RGB value corresponding to the number is obtained on the first curve and denoted as the first RGB value. The RGB defect positioning curve at the time preceding the first defect time is obtained and denoted as the second curve. The RGB value corresponding to the number is obtained on the second curve and denoted as the second RGB value. If the first RGB value and the second RGB value are not equal, then the high-frequency defect region is an RGB-related region.

[0024] As a further aspect of the present invention: if there are several adjacent RGB-related regions, the area occupied by these several RGB-related regions is recorded as a region set; if the area occupied by the region set exceeds a preset area threshold, the region set is recorded as an RGB sensitive region.

[0025] The beneficial effects of this invention are:

[0026] This invention dynamically adjusts the point cloud acquisition density of different regions based on historical data statistics of defect frequency and RGB sensitive areas. High-density scanning is used in high-frequency defect areas or RGB sensitive areas, while density is reduced in other areas to reduce computational load and achieve a balance between real-time detection and resource consumption. In the early stages of mold use (within a safe range), only basic density monitoring is required, and targeted adjustments are made later to avoid high-load computation throughout the process. High-precision scanning is performed only on key areas (RGB sensitive areas and high-frequency defect areas) to reduce reliance on high-performance hardware. In addition, for areas where RGB values ​​change significantly when defects occur (such as scratches causing changes in reflectivity), this invention directly uses image monitoring to avoid unnecessary high-density point cloud scanning. High-density point cloud detection is retained only in geometrically deformable sensitive areas (such as high-frequency defect areas) to achieve precise allocation of computational load. Attached Figure Description

[0027] The invention will now be further described with reference to the accompanying drawings.

[0028] Figure 1 This is a schematic diagram of the method of the real-time detection system for three-dimensional mold surface defects based on deep learning according to the present invention. Detailed Implementation

[0029] 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.

[0030] Please see Figure 1 As shown, this invention is a real-time detection system for surface defects in three-dimensional molds based on deep learning, comprising:

[0031] Data collection module: a data recording unit and a preprocessing unit. The data recording unit divides the mold into several areas to record point cloud data and image data of the mold at various times during its service life. Through the data recording unit, historical point cloud data and historical image data of the mold are obtained.

[0032] The preprocessing unit generates a geometric defect location curve based on the historical point cloud data, and generates an RGB defect location curve based on the historical image data.

[0033] In a preferred embodiment of the present invention, the usage period is the time period between the time when the mold is put into use and the time when it is put out of use, wherein the time when the mold is put into use is the time when the mold is put out of use is the time when it is put out of use; and according to a preset time interval, a number of times are selected equally within the usage period; the image data includes images of each side of the mold.

[0034] In a preferred embodiment of the present invention, the process of obtaining the point cloud model and image data includes:

[0035] A three-dimensional coordinate system is established, the mold is placed in the three-dimensional coordinate system, the center points of each region of the mold are obtained, and each center point is numbered; and the three-dimensional coordinates corresponding to each center point in the three-dimensional coordinate system are obtained to obtain the point cloud data of the mold; the image data includes the RGB values ​​of each region of the mold;

[0036] The mold surface is divided into several regions, and a center point number is assigned to each region to establish a spatial mapping relationship in a three-dimensional coordinate system. The three-dimensional coordinates (x, y, z) of the center point of each region are obtained by a 3D scanner (such as LiDAR or structured light), and the geometric deformation is recorded to obtain point cloud data. RGB images of the mold surface are acquired by an industrial camera to record color or texture changes and obtain image data.

[0037] In a preferred embodiment of the present invention, the process of generating the geometric defect location curve includes:

[0038] When the surface of the mold is free of defects, the three-dimensional coordinates of the center points of each region are obtained and recorded as standard coordinates. For the mold at any time, the distance between the three-dimensional coordinates of each center point and the standard coordinates is obtained and recorded as the distance difference. A rectangular coordinate system is established with the center point number as the abscissa and the distance difference as the ordinate. In the rectangular coordinate system, the center point of each number and its corresponding distance difference are fitted to obtain the geometric defect location curve.

[0039] Based on the standard coordinates when there are no defects, the deviation (distance difference) between the real-time point cloud coordinates and the standard values ​​is calculated, and a curve is fitted to quantify the degree of deformation.

[0040] In a preferred embodiment of the present invention, the process of generating the RGB defect location curve includes:

[0041] Based on the historical image data, the RGB values ​​of each region on the surface of the mold are obtained and recorded as the RGB values ​​of the center point of the region. The center point of each number and its corresponding RGB value are fitted to obtain the RGB defect location curve.

[0042] Extract the RGB value change curves of each region over time to capture color anomalies (such as color differences caused by oxidation or scratches).

[0043] Data analysis module: Based on the geometric defect location curve, determine the safety range and obtain the defect frequency and high-frequency defect area of ​​each region of the mold; based on the RGB defect location curve, determine the RGB related area;

[0044] In a preferred embodiment of the present invention, the process of determining the safe zone includes:

[0045] Within the usage period, obtain the slope of each point on the geometric defect location curve at each moment, obtain the sum of the slopes at each point, and obtain the fluctuation value of the geometric defect location curve; record the moment corresponding to the geometric defect location curve when the fluctuation value is first not 0, and denote it as the defect moment.

[0046] Obtain the defect times of all grinding tools during their service life to obtain a set of defect times, and select the minimum value in the set of defect times as the safe cut-off time; obtain the safe interval from the time of commissioning and the safe cut-off time;

[0047] By measuring the slope changes at each point on the geometric defect location curve, the cumulative fluctuation value of mold deformation is quantified; the moment when the first non-zero fluctuation value appears is marked as the defect initiation time, and the earliest defect time of all molds is determined by combining historical data.

[0048] In a preferred embodiment of the present invention, the process of obtaining the defect frequency of each region of the mold includes:

[0049] Obtain the geometric defect location curve of the mold at the end of the use, and denot it as the end time curve; based on the historical point cloud data, obtain the end time curves of all molds; for any numbered center point, obtain the number of end time curves with a non-zero ordinate value of the center point in all end time curves, and obtain the defect frequency f=n / N of the region corresponding to the center point, where n is the number of end time curves with a non-zero ordinate value and N is the total number of all end time curves;

[0050] Based on historical data, the number of defects occurring in each area at the end of use is counted, and the defect frequency is calculated. Areas with a frequency exceeding a threshold for high-frequency defects need to be monitored closely.

[0051] In a preferred embodiment of the present invention, the process of determining the high-frequency defect region includes:

[0052] A frequency threshold is set, and regions where the defect frequency exceeds the frequency threshold are recorded as high-frequency defect regions; otherwise, they are recorded as low-frequency defect regions.

[0053] In a preferred embodiment of the present invention, the process of determining the RGB related region includes:

[0054] The process involves obtaining the geometric defect location curve corresponding to the first non-zero ordinate value among all geometric defect location curves of the high-frequency defect region during its usage cycle, denoted as the first defect time; obtaining the RGB defect location curve at the first defect time, denoted as the first curve; obtaining the number of the center point of the high-frequency defect region; obtaining the RGB value corresponding to the number on the first curve, denoted as the first RGB value; and obtaining the RGB defect location curve at the time preceding the first defect time, denoted as the second curve; obtaining the RGB value corresponding to the number on the second curve, denoted as the second RGB value; if the first RGB value and the second RGB value are not equal, then the high-frequency defect region is an RGB-related region.

[0055] Compare the RGB values ​​of the first defect moment in the high-frequency defect area with those of the previous moment. If the RGB values ​​change abruptly, the area is identified as an RGB-related region.

[0056] Point cloud density correction module: Based on the RGB related regions, obtain the RGB sensitive regions and set the initial density of the point cloud; when monitoring the surface defects of the mold, monitor the RGB values ​​of the RGB sensitive regions in real time, and monitor the point cloud data of the remaining regions in real time with the initial density within the safe range; after the safe range ends, adjust the point cloud density of each region of the mold according to the defect frequency of each region of the mold.

[0057] It should be noted that image monitoring is used for RGB sensitive areas, while point cloud monitoring is retained for geometrically sensitive areas. The hybrid detection architecture achieves data synchronization through the PCIe bus. The point cloud density is dynamically adjusted, with an initial density for the entire area within the safe zone and increased point cloud density for high-frequency areas outside the safe zone. A layered architecture is also adopted, with the front end based on FPGA to achieve fast comparison of image RGB values ​​and the back end based on GPU to accelerate point cloud feature extraction.

[0058] In a preferred embodiment of the present invention, if there are several adjacent RGB related regions, the area occupied by these several RGB related regions is recorded as a region set. If the area occupied by the region set exceeds a preset area threshold, the region set is recorded as an RGB sensitive region.

[0059] In a preferred embodiment of the present invention, the initial density of the point cloud is a preset number of point clouds per unit area of ​​the mold surface.

[0060] In a preferred embodiment of the present invention, the process of real-time monitoring of the RGB values ​​of the RGB sensitive area includes:

[0061] If the RGB value of a region within the RGB sensitive area changes compared to the previous moment, then a defect appears on the surface of the mold in that region.

[0062] In a preferred embodiment of the present invention, the process of adjusting the point cloud density of each region of the mold includes:

[0063] If the initial density is denoted as Id, then the point cloud density of the region is obtained. , where k is a preset correction coefficient and k > 0;

[0064] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A real-time detection system for surface defects of three-dimensional molds based on deep learning, characterized in that, include: Data collection module: a data recording unit and a preprocessing unit. The data recording unit divides the mold into several areas to record point cloud data and image data of the mold at various times during its service life. Through the data recording unit, historical point cloud data and historical image data of the mold are obtained. The preprocessing unit generates a geometric defect location curve based on the historical point cloud data, and generates an RGB defect location curve based on the historical image data. Data analysis module: Based on the geometric defect location curve, determine the safety range and obtain the defect frequency and high-frequency defect area of ​​each region of the mold; based on the RGB defect location curve, determine the RGB related area; Point cloud density correction module: Based on the RGB related regions, obtain the RGB sensitive regions and set the initial density of the point cloud; when monitoring the surface defects of the mold, monitor the RGB values ​​of the RGB sensitive regions in real time, and monitor the point cloud data of the remaining regions in real time with the initial density within the safe range; after the safe range ends, adjust the point cloud density of each region of the mold according to the defect frequency of each region of the mold.

2. The real-time detection system for surface defects of three-dimensional molds based on deep learning according to claim 1, characterized in that, The usage period is the time interval between the time when the mold is put into use and the time when it is put out of use. The time when the mold is put into use is the time when the mold is put out of use. The time when the mold is put out of use is the time when it is put out of use. A number of times are selected equally within the usage period according to a preset time interval. The image data includes images of each side of the mold.

3. The real-time detection system for three-dimensional mold surface defects based on deep learning according to claim 1, characterized in that, The process of obtaining the point cloud data and image data includes: A three-dimensional coordinate system is established, the mold is placed in the three-dimensional coordinate system, the center points of each region of the mold are obtained, and each center point is numbered; and the three-dimensional coordinates corresponding to each center point in the three-dimensional coordinate system are obtained to obtain the point cloud data of the mold; the image data includes the RGB values ​​of each region of the mold.

4. The real-time detection system for three-dimensional mold surface defects based on deep learning according to claim 1, characterized in that, The process of generating the geometric defect location curve includes: When the surface of the mold is free of defects, the three-dimensional coordinates of the center points of each region are obtained and recorded as standard coordinates. For the mold at any time, the distance between the three-dimensional coordinates of each center point and the standard coordinates is obtained and recorded as the distance difference. A rectangular coordinate system is established with the center point number as the abscissa and the distance difference as the ordinate. In the rectangular coordinate system, the center point of each number and its corresponding distance difference are fitted to obtain the geometric defect location curve.

5. The real-time detection system for three-dimensional mold surface defects based on deep learning according to claim 2, characterized in that, The process of determining the safe zone includes: Within the usage period, obtain the slope of each point on the geometric defect location curve at each moment, obtain the sum of the slopes at each point, and obtain the fluctuation value of the geometric defect location curve; record the moment corresponding to the geometric defect location curve when the fluctuation value is first not 0, and denote it as the defect moment. Obtain the defect times of all grinding tools during their service life to obtain a set of defect times, and select the minimum value in the set of defect times as the safe cut-off time; obtain the safe interval from the time of use and the safe cut-off time.

6. The real-time detection system for three-dimensional mold surface defects based on deep learning according to claim 2, characterized in that, The process of obtaining the defect frequency of each region of the mold includes: Obtain the geometric defect location curve of the mold at the end of its use, and denot it as the end time curve; based on the historical point cloud data, obtain the end time curves of all molds; for any numbered center point, obtain the number of end time curves in all end time curves where the vertical coordinate value of the center point is not 0, and obtain the defect frequency f=n / N of the region corresponding to the center point, where n is the number of end time curves where the vertical coordinate value is not 0, and N is the total number of all end time curves.

7. The real-time detection system for surface defects of three-dimensional molds based on deep learning according to claim 1, characterized in that, The process of determining the RGB-related region includes: The time corresponding to the first non-zero ordinate value of the geometric defect positioning curve in all geometric defect positioning curves of the high-frequency defect region during its usage cycle is obtained and denoted as the first defect time. The RGB defect positioning curve at the first defect time is obtained and denoted as the first curve. The center point number of the high-frequency defect region is obtained, and the RGB value corresponding to the number is obtained on the first curve and denoted as the first RGB value. The RGB defect positioning curve at the time preceding the first defect time is obtained and denoted as the second curve. The RGB value corresponding to the number is obtained on the second curve and denoted as the second RGB value. If the first RGB value and the second RGB value are not equal, then the high-frequency defect region is an RGB-related region.

8. The real-time detection system for surface defects of three-dimensional molds based on deep learning according to claim 1, characterized in that, If there are several adjacent RGB-related regions, the area occupied by these several RGB-related regions is recorded as a region set. If the area occupied by the region set exceeds a preset area threshold, the region set is recorded as an RGB sensitive region.

Images