Non-contact scanning detection system for industrial parts based on external camera

Abstract

The application belongs to the technical field of quality inspection and analysis of industrial part production, and specifically discloses an industrial part non-contact scanning detection system based on an external camera, which comprises: an initial inspection information acquisition module, which acquires the defect type, position, size and material type of a suspected defect area; an imaging interference judgment module, which extracts feature values from 2D images and 3D point cloud data based on the material type, compares and judges whether there is imaging interference and determines the interference strength; a re-inspection demand analysis module, which determines the reference defect risk weight based on the position, defect type and size, and calculates the re-inspection demand index by comprehensively considering the imaging interference strength; and a re-inspection determination feedback terminal, which determines the areas with an index higher than a threshold value as re-inspection targets and arranges them into a re-inspection queue from high to low according to the index. The application solves the problems of high initial inspection false alarm rate and unreasonable allocation of re-inspection resources of industrial parts coexisting on multi-material surfaces, and improves the response efficiency of high-risk defects.

Description

Technical Field

[0001] This invention belongs to the field of industrial parts inspection technology, and more specifically, relates to a non-contact scanning inspection system for industrial parts based on an external camera. Background Technology

[0002] During the production of industrial parts, the surface of these parts often contains multiple areas with different reflective properties, such as machined surfaces, cast or forged blank surfaces, and welded or coated reinforced surfaces. This can lead to the misjudgment of reflective spots in machined areas, oxide scale color differences in blank areas, and texture features in reinforced coating areas as actual material defects during surface inspection. Therefore, suspected defects need to be re-inspected to ensure the accuracy of quality inspection.

[0003] Currently, when light spot interference on the machined surface and actual cracks on the blank surface output the same confidence level during initial inspection and simultaneously reach the re-inspection trigger standard, the system will uniformly transmit both to the re-inspection station. However, the re-inspection process cannot distinguish whether the former is caused by imaging interference or the latter is a real defect. In industrial parts inspection, the urgency of re-inspection varies significantly depending on the location, type, and inspection status of different defects. The current re-inspection judgment based on a uniform re-inspection trigger standard is prone to causing low-risk surface defects to repeatedly occupy the re-inspection station, resulting in re-inspection congestion, while high-risk defects cannot be identified in a timely manner. This leads to delays in adjusting part process parameters and affects the overall quality control of industrial parts. Summary of the Invention

[0004] In view of this, in order to solve the above problems, a non-contact scanning inspection system for industrial parts based on an external camera is proposed.

[0005] The objective of this invention can be achieved through the following technical solution: This invention provides a non-contact scanning and inspection system for industrial parts based on an external camera. The system includes: a preliminary inspection information acquisition module, which acquires information on suspected defect areas recorded after the preliminary inspection of industrial parts at the quality inspection station. The information includes at least the defect type, location, size, and material type. The material type includes at least one of milled surfaces, cast blank surfaces, and wear-resistant weld overlay surfaces.

[0006] The imaging interference judgment module extracts a first feature value representing texture anomalies from the initial 2D image based on the material type, and extracts a second feature value representing geometric anomalies from the 3D point cloud data. The two are compared to determine whether imaging interference exists. If it exists, the intensity of the imaging interference is determined.

[0007] The re-inspection demand analysis module determines the baseline defect risk weight based on the location, defect type, and size of the suspected defect area, and calculates the re-inspection demand index for each suspected defect area by combining the imaging interference intensity and the baseline defect risk weight.

[0008] The re-inspection confirmation feedback terminal identifies suspected defect areas with a re-inspection demand index higher than the re-inspection trigger threshold as targets to be re-inspected, and arranges each target to be re-inspected into the re-inspection queue according to the re-inspection demand index from high to low.

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Based on the imaging characteristics of multiple materials that may coexist on the same surface of industrial parts, the present invention extracts texture features from 2D images and geometric features from 3D point clouds for the material type of suspected defect areas and compares them. This solves the problem that the production quality inspection cannot distinguish between real defects and imaging interference due to the difference in reflection of different materials. It realizes the accurate identification of imaging interference and the quantification of interference intensity. Thus, it can scientifically reduce the re-inspection priority of suspected defects according to the interference intensity, avoid the ineffective investment of re-inspection resources on false positive defects caused by strong interference, and indirectly improve the reliability and efficiency of the overall quality inspection process.

[0010] (2) This invention calculates the re-inspection demand index by fusing imaging interference intensity and benchmark defect risk weight to quantify the urgency of re-inspection for each suspected area, breaking the limitation that the unified re-inspection triggering standard cannot distinguish the urgency of defects, and facilitating subsequent targeted re-inspection sorting.

[0011] (3) Based on the re-inspection demand index determined by the imaging interference intensity and the baseline defect risk weight, this invention identifies areas above the trigger threshold as re-inspection targets and dynamically arranges them into the re-inspection queue from high to low according to the re-inspection demand index. This solves the problem of low-risk surface defects repeatedly occupying the re-inspection station, causing congestion, while high-risk defects cannot be confirmed in time, leading to delays in process adjustment. It achieves optimal allocation of re-inspection resources under the coupled interference of multiple materials and multiple defects, greatly improves the response timeliness of high-risk defects and the efficiency of production line adjustment, and also ensures the overall quality control of industrial parts. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the system module connections of the present invention;

[0013] Figure 2 This is a schematic diagram of the overall implementation process of the present invention;

[0014] Figure 3 This is a schematic diagram of the first feature value extraction process of the present invention;

[0015] Figure 4 This is a schematic diagram of the second feature value extraction process of the present invention. Detailed Implementation

[0016] 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.

[0017] Currently, during the initial inspection of industrial parts at the quality control station, the presence of multiple materials on the surface often causes imaging interference such as reflection, color difference, and texture, leading to a large number of non-real defects being falsely reported as suspected areas. Furthermore, existing systems use a uniform re-inspection trigger standard for all suspected areas, ignoring the actual differences in defect location, type, and degree of interference. This results in low-risk defects repeatedly occupying re-inspection stations while high-risk defects are difficult to identify in a timely manner.

[0018] This invention discloses a non-contact scanning and inspection system for industrial parts based on an external camera. By extracting texture and geometric anomaly features according to material type to determine imaging interference, and dynamically sorting the re-inspection queue by integrating the defect risk itself and the interference intensity, it can effectively reduce the false detection rate and optimize the allocation of re-inspection resources, while also significantly improving the response efficiency of high-risk defects.

[0019] Specifically, please refer to Figure 1 and Figure 2 As shown, the non-contact scanning and inspection system for industrial parts based on an external camera provided by the present invention includes: a preliminary inspection information acquisition module, an imaging interference judgment module, a re-inspection requirement analysis module, and a re-inspection determination feedback terminal.

[0020] In the above, the imaging interference judgment module is connected to the initial inspection information acquisition module and the re-inspection demand analysis module, respectively. The re-inspection demand analysis module is also connected to the re-inspection determination feedback terminal.

[0021] The initial inspection information acquisition module acquires information on suspected defect areas recorded after the initial inspection of industrial parts at the quality inspection station. The information includes at least the defect type, location, size, and material type. The material type includes at least one of the following: milled surface, cast blank surface, and wear-resistant weld overlay surface.

[0022] In this embodiment, on the production line, industrial parts that have completed previous processes (such as cleaning and deburring) are transported to the quality inspection station via an automated conveyor belt or AGV. Once the position sensor detects that the industrial part is in place, an industrial camera and laser scanner mounted on the quality inspection station simultaneously acquire 2D images and 3D point cloud data of the part. These images and data are stored in a time-stamped manner, and the pixel coordinates are precisely mapped to the spatial three-dimensional coordinates using pre-calibrated camera-laser extrinsic parameters. Subsequently, a target detection algorithm (such as YOLOv5 or Faster R-CNN) is used to identify suspected defect areas in the 2D images. For each identified suspected defect area, its defect type, location, size, and material type are recorded.

[0023] Understandably, defect types include cracks, pores, pinholes, scratches, pits, and uneven weld cladding, and the physical dimensions of the defects are converted from pixel dimensions to actual lengths using camera calibration parameters.

[0024] It should be noted that industrial parts design drawings typically divide and label different machined surfaces, such as milled surfaces, cast blank surfaces, and wear-resistant weld overlay surfaces. The material type of a suspected defect area can be obtained as follows: Based on the design drawings, a CAD model of the industrial part is constructed. According to the position of the fixed fixture in the workstation, the coordinate system of the CAD model of the industrial part is pre-registered with the physical coordinate system of the actual inspection station. For each suspected defect area, the material type corresponding to the area where that point is located is queried in the CAD model based on its physical coordinates (derived from pixel coordinates).

[0025] It is important to note that for suspected defect areas located on the boundary between two materials, the material type corresponding to the geometric center point of the defect should be taken.

[0026] After completing the above recording, the defect type, location, size, and material type of each suspected defect area are associated and stored as the information basis for subsequent imaging interference judgment and re-inspection requirement analysis. The imaging interference judgment module is then activated.

[0027] The imaging interference judgment module extracts a first feature value representing texture anomalies from the initial 2D image based on the material type, and extracts a second feature value representing geometric anomalies from the 3D point cloud data. It compares the two to determine whether imaging interference exists. If it exists, it determines the intensity of the imaging interference.

[0028] It should be noted that when imaging interference (such as reflective spots, oxide scale color difference, or cladding texture) occurs, local areas on the surface of industrial parts will exhibit obvious texture anomalies in 2D images, while these areas will not show corresponding geometric deformations in 3D point cloud data. Therefore, embodiments of the present invention can identify imaging interference by comparing the relationship between their first and second feature values ​​with their respective thresholds.

[0029] For example, please refer to Figure 3 As shown, extracting the first feature value representing texture anomalies from the initial 2D image includes the following steps:

[0030] A1. Based on the material type of the suspected defect area (such as cast blank surface, milled surface or wear-resistant weld overlay surface), obtain the GLCM parameters corresponding to the material type. The GLCM parameters include at least the window height, window width and gray-level quantization level of the gray-level co-occurrence matrix.

[0031] It should be noted that the GLCM parameters can be obtained by statistically analyzing 2D images of qualified samples of the corresponding material type. The specific statistical method is as follows: collect no less than 200 defect-free qualified samples, construct a qualified sample set, calculate the coefficient of variation of contrast and energy for each sample, design multiple sets of candidate parameters, wherein the window height and width are in the range of 5 pixels to 25 pixels, the gray quantization level is in the range of 8 levels to 64 levels, calculate the average value of the coefficient of variation of contrast and energy for each set of candidate parameters, and take the candidate parameter with the smallest average value as the GLCM parameter for the corresponding material type.

[0032] A2. Using the center pixel of the suspected defect area as the center, determine a rectangular window based on the window height and window width, and extract the original gray values ​​of all pixels within the rectangular window from the 2D image.

[0033] Understandably, the rectangular window is determined as follows: taking the center pixel coordinates as a reference, the height of the window is rounded up and down by half, and the width of the window is rounded up and down by half, thus obtaining the rectangular window.

[0034] A3. Perform grayscale compression quantization on the original grayscale values ​​according to the grayscale quantization level to generate a quantized grayscale value matrix. For example, if the original grayscale level is 256 levels and the quantization level is set to 32 levels, then every 8 consecutive grayscale levels will be compressed into 1 quantization level.

[0035] A4. Perform gray-level co-occurrence matrix calculation on the quantized gray-level value matrix, and extract contrast and energy from the gray-level co-occurrence matrix as texture statistical features. The gray-level co-occurrence matrix calculation is a common existing calculation method, which will not be described in detail here.

[0036] A5. Compare the extracted contrast and energy with the standard contrast and standard energy corresponding to the same material type, and calculate the relative deviation value respectively.

[0037] Understandably, the standard contrast and standard energy are obtained as follows: for each material type, the contrast and energy of each sample in the qualified sample set are calculated according to steps A1 to A4 respectively, a contrast sequence and an energy sequence are constructed, the arithmetic mean of the contrast sequence and the arithmetic mean of the energy sequence are calculated, and these are used as the standard contrast and standard energy corresponding to that material type.

[0038] A6. The relative deviation values ​​of contrast and energy are linearly weighted and summed to obtain the weighted comprehensive deviation, which is used as the first feature value. The sum of the contrast weight and the energy weight is 1, and the specific weight values ​​are preset according to the material type.

[0039] Understandably, for milled surfaces, imaging interference (such as reflective spots) primarily alters the clarity and depth contrast of the machined texture, thus being more sensitive to contrast; therefore, contrast has a higher weight than energy. For cast blank surfaces, imaging interference (such as oxide scale color difference) mainly disrupts the uniformity of grayscale distribution, making energy more sensitive to this; therefore, energy is given a higher weight than contrast. For wear-resistant weld overlay surfaces, the cladding texture exhibits both directionality and randomness; contrast and energy have comparable descriptive abilities for interference, thus both are given equal weight.

[0040] It should be noted that the first feature value mentioned above is used to quantify the overall deviation between the texture features of the suspected defect area and the qualified sample of the same material. The larger the first feature value, the greater the deviation between the texture features of the suspected defect area and the texture features of the qualified sample of the same material, that is, the more serious the imaging interference.

[0041] Considering that the texture deviation reflected by the first feature value may originate from the actual geometric defects on the surface of industrial parts (such as cracks, pores, scratches, etc.), or from external interference during the imaging process (such as uneven lighting, lens stains, surface reflections, etc.), while real defects are usually accompanied by synchronous anomalies in the three-dimensional geometric shape, and imaging interference will not produce equivalent geometric fluctuations in the 3D point cloud.

[0042] Based on this, please refer to Figure 4 As shown, after completing the first feature value extraction, the second feature value representing geometric anomalies is extracted from the 3D point cloud data. The specific extraction process includes:

[0043] B1. Obtain the surface fitting order and neighborhood radius corresponding to the material type of the suspected defect area.

[0044] Understandably, the surface fitting order is determined as follows: For each material type, 3D point cloud data of each sample under that material type is obtained from the qualified sample set. For each candidate radius (from 1 mm to 10 mm), the local surface fitting residual of each point is calculated, the standard deviation of the residuals of all points within each sample is obtained, and then the arithmetic mean of the standard deviations of all samples is calculated as the average residual standard deviation under that radius. A standard deviation sequence is constructed, and the first difference of this standard deviation sequence is calculated to obtain the first difference sequence. The candidate radius corresponding to the first local minimum point in the first difference sequence is determined as the neighborhood radius. If no local minimum point exists, the smallest candidate radius is taken.

[0045] The fitting order is determined as follows: First-order and second-order polynomial surfaces are used for fitting, and the standard deviations of their respective fitting residuals are calculated. The order with the smaller standard deviation of the fitting residuals is selected as the fitting order. If the two are equal, first-order fitting is selected.

[0046] B2. Using the neighborhood radius as the search radius and the center point of the suspected defect area as the sphere center, extract the spatial three-dimensional coordinates of all points whose distance from the sphere center is less than or equal to the search radius from the 3D point cloud data, and construct a three-dimensional coordinate sequence.

[0047] B3. Take all points within the search radius as analysis points. Based on the surface fitting order determined in step B1, use the least squares method to perform polynomial surface fitting on the three-dimensional coordinate sequence constructed in step B2 to generate a fitting reference surface. Calculate the height residual value of each analysis point relative to the fitting reference surface and construct a height residual sequence.

[0048] Understandably, when the surface fitting order is first, the surface fitting function is: , where x, y, and z represent the x-axis coordinates, y-axis coordinates, and z-axis coordinates of the analysis point, respectively.

[0049] When the surface fitting order is second, the surface fitting function is: ,in, to These are all surface coefficients to be fitted, which can be solved using the least squares method.

[0050] B4. When the material type is a milled surface, calculate the standard deviation of the height residual sequence and use the standard deviation as the second characteristic value. When the material type is a cast blank surface or a wear-resistant weld overlay surface, calculate the average value of the Gaussian curvature change and use the average value as the second characteristic value.

[0051] The average value of the Gaussian curvature change is calculated using the following sub-steps:

[0052] B41. Obtain the spatial three-dimensional coordinates of each analysis point.

[0053] B42. For each analysis point, with the analysis point as the center and the neighborhood radius as the search radius, extract all analysis points whose distance from the analysis point is less than or equal to the search radius, and use them as the neighborhood point set of the analysis point.

[0054] B43. Using the spatial three-dimensional coordinates of each point in the neighborhood point set as input, perform local quadratic surface fitting (the fitting model is the same as the second-order surface fitting), and calculate the Gaussian curvature value of the analysis point based on the fitting results.

[0055] The formula for calculating the Gaussian curvature value K is: , , , Let be the second-order partial derivatives of the surface fitting function along the x and y directions at the analysis point, respectively. and These are the first-order partial derivatives of the surface fitting function along the x and y directions at the analysis point, respectively, and the values ​​of each partial derivative are obtained by differentiating the second-order surface fitting function at the analysis point.

[0056] B44. Take each analysis point as the current analysis point in turn, extract the neighborhood point set of the current analysis point, calculate the average value of the Gaussian curvature values ​​of all points in the neighborhood point set, and use it as the Gaussian curvature benchmark value of the current analysis point.

[0057] B45. Calculate the absolute value of the difference between the Gaussian curvature value of the current analysis point and the Gaussian curvature reference value, and use it as the change in Gaussian curvature of the current analysis point.

[0058] B46. Take the average value of the Gaussian curvature change of all analysis points in the suspected defect area to obtain the average value of the Gaussian curvature change.

[0059] After obtaining the first feature value and the second feature value, the specific judgment process for comparing the two to determine whether there is imaging interference is as follows: the first feature value obtained in step A6 is compared with the texture anomaly threshold corresponding to the material type of the suspected defect area, and the second feature value obtained in step B4 is compared with the geometric fluctuation anomaly threshold corresponding to the material type of the suspected defect area. If the first feature value is greater than the texture anomaly threshold and the second feature value is less than the geometric fluctuation anomaly threshold, it is determined that there is imaging interference; otherwise, it is determined that there is no imaging interference.

[0060] It should be added that the texture anomaly threshold and the geometric fluctuation anomaly threshold are predetermined in advance by the following methods: take the same qualified sample set as in step A1, calculate the first feature value of each sample according to steps A1 to A6, calculate the second feature value of each sample according to steps B1 to B4, construct the first feature value sequence and the second feature value sequence respectively, and take the 95th percentile of the first feature value sequence and the second feature value sequence as the corresponding threshold.

[0061] Furthermore, when imaging interference is determined to exist, the specific method for determining the intensity of imaging interference is as follows: calculate the first excess ratio of the first feature value exceeding the texture anomaly threshold and the second excess ratio of the geometric fluctuation anomaly threshold exceeding the second feature value, and multiply the first excess ratio and the second excess ratio to obtain the intensity of imaging interference.

[0062] It should be noted that the first excess ratio is calculated as follows: calculate the difference between the first feature value and the texture anomaly threshold, and divide the difference by the first feature value to obtain the first excess ratio.

[0063] The second excess ratio is calculated as follows: calculate the difference between the geometric fluctuation anomaly threshold and the second feature value, and divide the difference by the geometric fluctuation anomaly threshold to obtain the second excess ratio.

[0064] This embodiment is based on the imaging characteristics of industrial parts that may coexist with multiple materials. By extracting texture features from 2D images and geometric features from 3D point clouds for the material type of suspected defect areas and comparing them, it determines whether there is imaging interference. This solves the problem of initial misjudgment of non-real defects due to the difference in reflection of different materials. It realizes accurate identification of imaging interference and quantification of interference intensity, thereby effectively reducing the false detection rate and improving the reliability of the initial inspection results.

[0065] Furthermore, when imaging interference exists, the degree of interference is quantified by calculating the intensity of the imaging interference, providing a data basis for subsequent re-inspection needs analysis, thereby effectively reducing the false detection rate and providing direction for subsequent production optimization.

[0066] After completing the imaging interference assessment, this embodiment initiates the re-inspection requirement analysis module to further perform the re-inspection requirement analysis.

[0067] The re-inspection demand analysis module determines the baseline defect risk weight based on the location, defect type, and size of the suspected defect area, and calculates the re-inspection demand index for each suspected defect area by combining the imaging interference intensity and the baseline defect risk weight.

[0068] It should be noted that in the quality inspection of industrial parts production, the number of suspected defective areas identified during the initial inspection may be large, while resources for re-inspection are limited. Indiscriminately re-inspecting all suspected areas would lead to inefficiency and high costs. Therefore, it is necessary to conduct a risk quantification assessment of each suspected defective area and prioritize the re-inspection of areas with higher risk.

[0069] Based on this, the present invention first determines the baseline defect risk weight based on the location, defect type and size of the suspected defect area. This weight reflects the potential impact of the defect on the safety performance of industrial parts, ignoring imaging interference.

[0070] Specifically, the method for determining the baseline defect risk weight is as follows: D1, obtain the stress field distribution corresponding to the currently inspected industrial part.

[0071] D2. Obtain the equivalent stress value at the center coordinates of the suspected defect area from the stress field distribution. At the same time, extract the maximum equivalent stress value from the stress field distribution. Calculate the ratio of the equivalent stress value corresponding to the center coordinates to the maximum equivalent stress value to obtain the location risk weight.

[0072] It should be noted that, in order to quantify the relative danger level of the suspected defect area, this embodiment of the invention preferably compares the stress level at that location with the maximum stress level on the entire industrial part. This ratio, between 0 and 1, characterizes the relative stress danger level of the suspected defect area. The larger the ratio, the higher the stress at that location, and the greater the risk of the defect propagating or causing failure at that point.

[0073] D3. Determine the safe threshold for defect size in suspected defect areas based on location risk weights.

[0074] D4. Compare the defect size with the defect size safety threshold, and output the baseline defect risk weight based on the comparison results.

[0075] Understandably, if the defect size does not exceed the defect size safety threshold, the ratio of the defect size to the defect size safety threshold is calculated to obtain the defect size risk weight. The product of the defect size risk weight and the location risk weight is used as the baseline defect risk weight. This product form ensures that the baseline risk weight only increases significantly when the location is dangerous and the size is close to the limit. In low-stress areas or for extremely small sizes, the product result approaches 0.

[0076] If the defect size exceeds the defect size safety threshold, the baseline defect risk weight is assigned a value of 1 to ensure that a mandatory re-inspection can be triggered. The baseline defect risk weight is between 0 and 1, reflecting the risk level of the defect under the dual factors of size and location.

[0077] The following additional explanations are required during the specific execution of steps D1 to D4:

[0078] It should be noted that the severity of defects (cracks, porosity, inclusions, etc.) in industrial parts depends primarily on the actual stress level at that location under working conditions. The higher the stress level, the lower the tolerance for defects. For example, in load-bearing areas subjected to alternating loads, such as shaft shoulder transition fillets, gear tooth roots, or bracket mounting surfaces, once microcracks are present, these microcracks may propagate to a through-crack within hours. Therefore, higher stress areas require stricter defect size safety thresholds (smaller safety thresholds), while lower stress areas can have more lenient thresholds (larger safety thresholds).

[0079] The process for determining the defect size safety threshold is as follows: Based on the material type and defect type of the suspected defect area, obtain the critical risk defect size and the minimum allowable defect size of the currently inspected industrial part at the location of maximum stress, corresponding to the combination of material type and defect type, and denoted as follows: and Among them, the critical risk defect size and the minimum allowable defect size can be determined according to the safety standards for industrial parts production. This indicates the maximum acceptable defect size at that location. This indicates the lower limit of the defect size allowed by the process standard.

[0080] The location risk weight is denoted as Calculate the defect size safety threshold , , This represents a function that minimizes the value; this formula allows the safety threshold to be set within a certain range. to The risk weights adjust linearly with location, when (At the point of maximum stress) The safety threshold is tightened to the minimum value when (When located in a stress-free position) The safety threshold is relaxed to the maximum value, that is, the greater the location risk weight, the smaller the defect size safety threshold.

[0081] It should also be noted that the stress field distribution can be pre-calculated using the finite element method. Taking a certain type of large load-bearing connecting bracket as an example, the material is ZG35CrMo, with an elastic modulus of 206 GPa, Poisson's ratio of 0.3, and a density of 7850 kg / m³. Fixed constraints are applied at the mounting holes at the bottom of the bracket, and a rated working load of 200 kN is applied to the top bearing flange surface, in the direction of the main load. Tetrahedral elements are used to divide the mesh, with an element size of 5 mm, and linear statics are performed. After solving, the equivalent stress and three-dimensional spatial coordinates of each node on the bracket surface are extracted to form the stress field distribution. For other types of industrial parts, the same method can be used to construct the stress field distribution based on the actual design drawings.

[0082] Furthermore, since the aforementioned imaging interference intensity reflects the degree of interference to the 2D image in the initial inspection, when the imaging interference intensity is high, texture anomalies mainly originate from imaging noise rather than actual defects. In this case, the reliability of defect size and type information extracted from the image is low. If a re-inspection is still initiated with a high baseline risk weight, it may lead to false alarms and waste re-inspection resources. Therefore, it is necessary to introduce an imaging reliability coefficient to reduce the baseline risk weight.

[0083] Based on this, the re-inspection demand index is calculated using the following formula: .

[0084] in, This indicates the re-inspection demand index. Indicates the intensity of imaging interference. This represents the benchmark risk weight.

[0085] This represents the imaging reliability coefficient. The greater the imaging interference intensity, the worse the reliability of the baseline risk weight assessment result, and the lower the corresponding need for re-inspection. This imaging reliability coefficient makes... The closer the re-inspection demand index is to 1, the closer it approaches 0. The re-inspection demand index for smaller defects is mainly determined by the baseline defect risk weight.

[0086] It is important to note that when the benchmark risk weight... When the value is 1, the risk has reached its maximum and a re-inspection needs to be forcibly triggered. That is, when the baseline defect risk weight is equal to 1, the re-inspection demand index is assigned a value of 1.

[0087] After completing the above calculations, the re-inspection demand index for each suspected defect area is output to the subsequent re-inspection confirmation feedback terminal for dynamic scheduling of the re-inspection queue.

[0088] The re-inspection confirmation feedback terminal identifies suspected defect areas with a re-inspection demand index higher than the re-inspection trigger threshold as targets to be re-inspected, and arranges each target to be re-inspected into the re-inspection queue according to the re-inspection demand index from high to low.

[0089] In this embodiment of the invention, the default value for the re-inspection trigger threshold is 0.5. The specific value can be dynamically adjusted based on the actual re-inspection capacity of the production line. As a concrete example, assuming the production line can re-inspect a total of M industrial parts per day, and there are currently N targets awaiting re-inspection in the re-inspection queue, the remaining acceptable capacity is MN. All targets awaiting re-inspection are sorted from highest to lowest according to their re-inspection demand index. If the total number of targets awaiting re-inspection is greater than the remaining acceptable capacity, the index value at the (M-N+1)th position in the sorted sequence is taken as the new re-inspection trigger threshold. If the total number of targets awaiting re-inspection is less than or equal to the remaining acceptable capacity, the re-inspection trigger threshold is not adjusted.

[0090] Specifically, the method for arranging each target to be re-inspected into the re-inspection queue can be a dynamic priority arranging method. That is, whenever a new target to be re-inspected is added, it is automatically inserted into the corresponding position in the queue according to its re-inspection demand index, ensuring that high-index targets to be re-inspected are at the front of the queue and receive priority for re-inspection.

[0091] Understandably, once the re-inspection queue is generated, the system automatically sends data such as the location information of the target to be re-inspected, defect type, and re-inspection demand index to the re-inspection station or operator terminal, and provides feedback on the current re-inspection progress and results. After the re-inspection is completed, the terminal can update the inspection record.

[0092] It should be noted that the targets in the re-inspection queue will be sent to the re-inspection station in sequence for manual or higher-precision instrument re-inspection. Since the queue has been strictly sorted according to the re-inspection demand index, suspected defect areas with high risk and low imaging interference can be identified first, while false positive areas caused by low risk or strong interference are delayed or directly filtered out, thereby effectively avoiding congestion of re-inspection resources and ensuring the timely handling of real defects.

[0093] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A non-contact scanning and inspection system for industrial parts based on an external camera, characterized in that, The system includes: The initial inspection information acquisition module acquires information on suspected defect areas recorded after the initial inspection of industrial parts at the quality inspection station. The information includes at least the defect type, location, size, and material type. The material type includes at least one of the following: milled surface, cast blank surface, and wear-resistant weld overlay surface. The imaging interference judgment module extracts the first feature value representing texture anomalies from the initial 2D image based on the material type, and extracts the second feature value representing geometric anomalies from the 3D point cloud data. It compares the two to determine whether imaging interference exists. If it exists, it determines the intensity of the imaging interference. The re-inspection demand analysis module determines the baseline defect risk weight based on the location, defect type and size of the suspected defect area, and calculates the re-inspection demand index of each suspected defect area by combining the imaging interference intensity and the baseline defect risk weight. The re-inspection confirmation feedback terminal identifies suspected defect areas with a re-inspection demand index higher than the re-inspection trigger threshold as targets to be re-inspected, and arranges each target to be re-inspected into the re-inspection queue according to the re-inspection demand index from high to low.

2. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 1, characterized in that: The extraction of the first feature value representing texture anomalies specifically includes: Based on the material type of the suspected defect area, obtain the window height, window width, and gray level of the gray-level co-occurrence matrix corresponding to that material type; Using the center pixel of the suspected defect area as the center, a rectangular window is determined according to the window height and window width, and the original gray values ​​of all pixels within the rectangular window range are extracted from the 2D image. The original grayscale values ​​are subjected to grayscale compression quantization processing based on the grayscale quantization level to generate a quantized grayscale value matrix. The gray-level co-occurrence matrix is ​​calculated on the quantized gray-level value matrix, and contrast and energy are extracted from the gray-level co-occurrence matrix as texture statistical features; The extracted contrast and energy values ​​are compared with the standard contrast and standard energy values ​​corresponding to the same material type, and the relative deviation values ​​are calculated respectively. The relative deviation values ​​of contrast and energy are linearly weighted and summed to obtain the weighted overall deviation, which is then used as the first characteristic value.

3. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 1, characterized in that: The extraction of the second feature value representing the geometric anomaly specifically includes: Obtain the surface fitting order and neighborhood radius corresponding to the material type of the suspected defect region; Using the neighborhood radius as the search radius and the center point of the suspected defect area as the sphere center, the spatial three-dimensional coordinates of all points whose distance from the sphere center is less than or equal to the search radius are extracted from the 3D point cloud data to construct a three-dimensional coordinate sequence. All points within the search radius are used as analysis points. Polynomial surface fitting is performed on the three-dimensional coordinate sequence according to the surface fitting order to generate a fitting reference surface. The height residual value of each analysis point relative to the fitting reference surface is calculated to construct a height residual sequence. When the material type is a milled surface, calculate the standard deviation of the height residual sequence and use it as the second characteristic value. When the material type is a cast blank surface or a wear-resistant weld overlay surface, calculate the average value of the Gaussian curvature change and use this average value as the second characteristic value.

4. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 3, characterized in that: The calculation of the average value of the Gaussian curvature change specifically includes: Obtain the three-dimensional spatial coordinates of each analysis point; For each analysis point, with the analysis point as the center and the neighborhood radius as the search radius, extract all analysis points whose distance from the analysis point is less than or equal to the search radius, and use them as the neighborhood point set of the analysis point; Using the spatial three-dimensional coordinates of each point in the neighborhood point set as input, local surface fitting is performed, and the Gaussian curvature value of the analysis point is calculated based on the fitting results. Each analysis point is taken as the current analysis point in turn. The neighborhood point set of the current analysis point is extracted, and the average value of the Gaussian curvature values ​​of all points in the neighborhood point set is calculated as the Gaussian curvature benchmark value of the current analysis point. Calculate the absolute value of the difference between the Gaussian curvature value at the current analysis point and the Gaussian curvature reference value, and use it as the change in Gaussian curvature at the current analysis point; The average value of the Gaussian curvature change is obtained by averaging the changes in Gaussian curvature at all analysis points within the suspected defect area.

5. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 1, characterized in that: The specific method for determining whether imaging interference exists is as follows: The first feature value is compared with the texture anomaly threshold corresponding to the material type of the suspected defect area, and the second feature value is compared with the geometric fluctuation anomaly threshold corresponding to the material type of the suspected defect area. If the first feature value is greater than the texture anomaly threshold and the second feature value is less than the geometric fluctuation anomaly threshold, it is determined that there is imaging interference; otherwise, it is determined that there is no imaging interference.

6. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 5, characterized in that: The specific method for determining the intensity of the imaging interference is as follows: Calculate the first excess ratio of the first feature value to the texture anomaly threshold, and the second excess ratio of the geometric fluctuation anomaly threshold to the second feature value; Multiply the first excess ratio by the second excess ratio to obtain the imaging interference intensity.

7. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 1, characterized in that: The method for determining the baseline defect risk weight is as follows: Obtain the stress field distribution corresponding to the currently inspected industrial part, and calculate the location risk weight based on the stress field distribution and the center coordinates of the suspected defect area; Determine the safe threshold for defect size in suspected defect areas based on location risk weights; If the defect size does not exceed the defect size safety threshold, the ratio of the defect size to the defect size safety threshold is calculated to obtain the defect size risk weight. The defect size risk weight is then multiplied by the location risk weight to obtain the baseline defect risk weight. If the defect size exceeds the defect size safety threshold, the baseline defect risk weight is assigned a value of 1.

8. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 7, characterized in that: The process for determining the location risk weight is as follows: The equivalent stress value at the center coordinates of the suspected defect area is obtained from the stress field distribution, and the maximum equivalent stress value is extracted from the stress field distribution. The location risk weight is obtained by calculating the ratio of the equivalent stress value corresponding to the center coordinate to the maximum equivalent stress value.

9. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 7, characterized in that: The process for determining the defect size safety threshold is as follows: Based on the material type and defect type of the suspected defect area, obtain the critical risk defect size and the minimum allowable defect size of the currently inspected industrial part at the location of maximum stress, corresponding to the combination of material type and defect type, and denoted as follows: and ; The location risk weight is denoted as Calculate the defect size safety threshold , .

10. The non-contact scanning and inspection system for industrial parts based on an external camera as described in claim 1, characterized in that: The specific calculation process for the re-inspection demand index is as follows: If the baseline defect risk weight is equal to 1, then the re-inspection demand index is assigned a value of 1; Otherwise, subtract the imaging interference intensity from 1 to obtain the imaging reliability coefficient, and use the product of the imaging reliability coefficient and the baseline defect risk weight as the re-inspection demand index.

Images