Surface defect visual inspection control method based on vacuum pressure difference micro-deformation characteristics

Abstract

The invention discloses a surface defect visual detection control method based on vacuum pressure difference micro-deformation characteristics, and belongs to the field of intelligent detection control. Comprising a detection setting off-line process of visual imaging detection system coordinate calibration, workpiece defect detection image region division, workpiece defect detection scanning motion planning and workpiece defect detection regulation and control parameter setting. And a detection control online process of workpiece positioning clamping and cabin entering control, workpiece defect detection task initialization, workpiece surface image generation process control, sealed cabin air pressure detection and regulation, deformation feature image comparison and defect identification, workpiece cabin exiting control and tool unloading, and workpiece defect detection task completion judgment. According to the invention, for the micro-deformation of the defect surface of the workpiece in the preset vacuum degree environment, methods of workpiece surface image acquisition, block splicing, similarity calculation and the like are provided, the problem of visual detection of the fine defects on the surface of the workpiece is solved, and the accuracy, efficiency and robustness of the visual detection method are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection and control, and specifically to a visual detection and control method for surface defects based on the micro-deformation characteristics of vacuum pressure difference. Background Technology

[0002] In the field of product quality inspection in manufacturing, surface defect detection of workpieces is a crucial link in ensuring product performance and safety. Especially for high-precision workpieces, the missed or false detection of defects such as micro-cracks and dents can lead to risks in subsequent use. Among current mainstream surface defect detection technologies, traditional manual inspection relies on human experience, resulting in low efficiency, strong subjectivity, and a tendency to miss micro-defects, making it difficult to meet the demands of large-scale, high-precision inspection. While conventional machine vision inspection equipment has achieved automation, it largely relies on the inherent shape of the workpiece surface or the imaging characteristics of a single light source to identify defects. For workpieces with smooth surfaces and indistinct defect features, it is difficult to effectively capture defect information, limiting detection accuracy. Furthermore, existing visual inspection methods have significant shortcomings in terms of accuracy and adaptability: the lack of precise control over the inspection environment during the inspection process makes it impossible to stimulate the manifestation of workpiece defects through external conditions, resulting in small differences between the imaging of defects and normal surfaces, failing to guarantee the accuracy of defect detection. In addition, some integrated inspection equipment lacks a collaborative control mechanism between the detection and control modules, leading to mismatched operating rhythms. This not only affects inspection efficiency but may also cause interruptions in the inspection process due to conflicts between modules, making it difficult to balance accuracy, efficiency, and robustness. Moreover, with the increasing demands for product quality in the manufacturing industry and the widespread adoption of multi-variety, small-batch production models, the industry urgently needs a surface defect visual inspection and control method that can effectively stimulate defect features, flexibly adapt to different workpieces, comprehensively inspect all surfaces, and coordinate the inspection and control processes. Summary of the Invention

[0003] To address the technical shortcomings of existing detection and control methods, this invention proposes a surface defect visual detection and control method based on the micro-deformation characteristics of vacuum pressure difference. This method uses vacuum environment regulation to induce micro-deformation in workpiece defects, active light illumination to highlight surface visual features, multi-degree-of-freedom orientation adjustment to ensure no blind spots in surface detection, and coordinated control of environmental control, orientation adjustment, and detection devices to improve detection efficiency. This method solves the problems of poor workpiece adaptability, low defect detection accuracy, low equipment operating efficiency, and susceptibility to environmental interference in existing methods.

[0004] This invention is achieved through the following technical solution:

[0005] A surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics includes an offline process for visual imaging inspection system coordinate calibration, workpiece defect detection image region division, workpiece defect detection scanning motion planning, and workpiece defect detection control parameter setting; and an online process for workpiece positioning, clamping and chamber entry control, workpiece defect detection task initialization, workpiece surface image generation process control, sealed chamber air pressure detection and control, deformation feature image comparison and defect recognition, workpiece exit control and tooling unloading, and workpiece defect detection task completion determination.

[0006] The offline detection setup process includes the following steps:

[0007] Step 1: Coordinate calibration of the visual imaging detection system: First, the intrinsic parameters of the visual camera in the visual imaging detection system are calibrated using the Zhang Zhengyou calibration method, including focal length, principal point coordinates, and radial distortion parameters. Second, the extrinsic parameters of the position and attitude of the visual camera mounted on the two-degree-of-freedom linear slide are calibrated using the Eye-in-Hand calibration method, including the rotation matrix and translation vector between the camera coordinate system and the slide coordinate system. The position and attitude error of the visual camera mounted on the two-degree-of-freedom linear slide is corrected using the extrinsic parameters. Then, the reference zero point of the two-degree-of-freedom linear slide and the vacuum chamber track is found according to the positioning reference hole of the cabin structure component, and the scanning motion coordinate system of the two-degree-of-freedom linear slide and the displacement and attitude adjustment mechanism on the workpiece transport vehicle running on the vacuum chamber track is established.

[0008] Step 2: Workpiece defect detection image region segmentation: Calculate the region at distance h based on the intrinsic parameters of the vision camera. w Image region A on the imaging plane w The length and width dimensions; for diameter d w Length l w For a cylindrical workpiece, considering the length a of the imaging plane region. c Width b c The vision camera, using the region overlap design method, first divides the cylindrical workpiece into ρ regions along the circumference. y The overlap rate is uniformly divided into n y There are several long strip-shaped image regions, each with a width of b. y Then, each elongated image region is aligned with the axis of the cylindrical workpiece by ρ. z The overlap rate is uniformly divided into n z There are 3 rectangular image blocks, each with a length of a. z Guarantee the above n y ×n z The surface formed by assembling the image blocks can completely cover the outer surface of the cylindrical workpiece.

[0009] Step 3: Setting workpiece defect detection control parameters: For cylindrical workpieces, set the internal air pressure value p1 of the high vacuum sealed chamber and the internal air pressure value p2 of the low vacuum sealed chamber, where p2>p1, and the vacuum pressure difference Δp=p2-p1; set the image similarity threshold s of the workpiece surface images under high vacuum and low vacuum conditions. y ;

[0010] Step 4: Workpiece Defect Detection Scanning Motion Planning: The workpiece defect detection scanning motion is the relative motion between the vision camera and the cylindrical workpiece formed by the rotation of the cylindrical workpiece in the circumferential direction controlled by the displacement and attitude adjustment mechanism and the translation of the vision camera carried by the two-degree-of-freedom linear slide in the horizontal plane. It includes circumferential rotation scanning motion and two-dimensional translational scanning motion. The circumferential rotation scanning motion planning generates the equivalent circumferential attitude adjustment angular displacement, and the two-dimensional translational scanning motion planning generates the equivalent lateral centerline displacement and longitudinal step line displacement. The scanning motion planning command is composed of the equivalent lateral centerline displacement, longitudinal step line displacement, and circumferential attitude adjustment angular displacement.

[0011] The online detection and control process includes the following steps:

[0012] Step 5: Workpiece Positioning, Clamping, and Cabin Entry Control: At the start of the online detection and control process, the workpiece transport vehicle is positioned at the beginning of the tooling base track on the tooling base platform. The cylindrical workpiece is placed between the active roller conveyor and the passive roller conveyor, with one end of the cylindrical workpiece aligned with the reference zero point of the displacement and attitude adjustment mechanism. A workpiece overlength detection sensor is used to detect whether the other end of the cylindrical workpiece is overlength. If the other end is overlength, the online detection and control process terminates; otherwise, the vacuum chamber door is opened, and the workpiece transport vehicle is started to move into the cabin on the tooling base track. The tooling base track is aligned with the vacuum chamber track. After the workpiece transport vehicle passes through the vacuum chamber door, it enters the vacuum chamber track until it is detected by the guide rail limiter at the end of the vacuum chamber track. At this point, the workpiece transport vehicle stops, the vacuum chamber door is closed, and the cabin structure and the vacuum chamber door form a sealed cabin. Proceed to Step 6.

[0013] Step 6: Initialize the workpiece defect detection task: Initialize the roller drive motor of the active roller conveyor to the reference zero point, and initialize the sequence number k of the circular attitude adjustment angular displacement sequence. y =0; Initialize the motion slide of the two-degree-of-freedom linear slide to the reference zero point, execute the lateral centering displacement command through the lateral drive motor, control the vision camera carried by the motion slide to align with the axis of the cylindrical workpiece laterally, reach the centering reference zero point, and initialize the sequence number k of the longitudinal stepping linear displacement sequence. z =0; Open the active light source arrays on both sides of the top of the cabin structure component, and form specific optical imaging features on the upper half of the outer circular surface of the cylindrical workpiece by actively projecting specific light; Proceed to step 7;

[0014] Step 7: Workpiece surface image generation process control: Initialize process control variable n c =0;

[0015] Step 7.1: High Vacuum Control of the Sealed Chamber: Set the target gas pressure inside the sealed chamber under high vacuum conditions to p1, and the process control variable n c Add 1, proceed to step 8;

[0016] Step 7.2: Low Vacuum Control of the Sealed Chamber: Set the target gas pressure inside the sealed chamber under low vacuum conditions to p2, and the process control variable n c Add 1, proceed to step 8;

[0017] Step 7.3: Camera longitudinal scan image generation: First, the vision camera carried by the motion slide is controlled by the longitudinal drive motor to complete a longitudinal stepping line displacement equivalent l. z Then, a vision camera is used to acquire a portion of the upper half of the outer surface of the cylindrical workpiece, forming a rectangular image block; the longitudinal image scanning process of longitudinal stepping line displacement equivalent drive and image acquisition of the outer surface of the cylindrical workpiece is repeated until n images are acquired. z n rectangular image blocks are used to correct the image block stitching method. z A rectangular image block forms a long strip image region; then, the longitudinal drive motor controls the motion slide to carry the vision camera back to the centering reference zero point, and proceeds to step 7.4;

[0018] Step 7.4: Circular Image Acquisition Completion Determination: Determine the circumferential attitude adjustment angular displacement sequence number k. y =n y If the condition is not met, the circumferential image acquisition is incomplete, and proceed to step 7.5; otherwise, proceed to step 7.6.

[0019] Step 7.5: Workpiece Circumferential Rotation Image Scanning: Based on the equivalent of the circumferential attitude adjustment angular displacement, control the active roller conveyor to drive the cylindrical workpiece to complete one circumferential step angular displacement in the circumferential direction. The circumferential attitude adjustment angular displacement sequence number is k. y Add 1, return to step 7.3;

[0020] Step 7.6: Generation of surface image of cylindrical workpiece: for n y A long strip-shaped image region is used to generate a high-vacuum workpiece surface image I under high-vacuum conditions using an image block stitching and correction method. h Generating low-vacuum workpiece surface images I under low-vacuum conditions d Proceed to step 7.7;

[0021] Step 7.7: Determine if image acquisition is complete: Determine the process control variable n cIf the condition is not met, the image acquisition is not completed and proceed to step 7.2; otherwise, proceed to step 9.

[0022] Step 8: Sealed Chamber Pressure Detection and Control: Target pressure value p inside the sealed chamber. t The actual air pressure value was adjusted to [p] using a sealed chamber air pressure detection and control method. t -Δp t , p t +Δp t Within the range, the Δp t For the target air pressure value p t The error; return to step 7.3;

[0023] Step 9: Deformation Feature Image Comparison and Defect Identification: For high-vacuum workpiece surface images I h and low vacuum workpiece surface image I d First, adaptive image filtering is used to remove image noise; then, n... e ×n e The pixel region is evenly divided into cells in the two images above. The image similarity s(i,j) of each pair of cell images is calculated using the image similarity calculation method. If the image similarity s(i,j) of each pair of cell images is ≥ s y If the surface of the workpiece is free of defects, then there are no defects on the workpiece surface; otherwise, there are defects on the workpiece surface, and the defects are located at s(i, j). y The cell image area; proceed to step 10;

[0024] Step 10: Workpiece exit control and tooling unloading: Open the vacuum chamber door, start the workpiece transport vehicle to move out of the chamber on the vacuum chamber track. After passing through the vacuum chamber door, the workpiece transport vehicle enters the tooling base track. Close the vacuum chamber door until the workpiece transport vehicle stops when detected by the guide rail limiter at the end of the tooling base track. The workpiece transport vehicle returns to the starting position of the tooling base track; remove the cylindrical workpiece between the active roller conveyor and the passive roller conveyor; proceed to step 11.

[0025] Step 11: Workpiece defect detection task completion determination: If there is no new workpiece defect detection task, the entire online process of workpiece defect detection control ends; otherwise, return to step 5.

[0026] Furthermore, the Eye-in-Hand calibration method specifically includes the following steps:

[0027] Step 2.1: Image Acquisition: Prepare a black and white checkerboard template with known physical dimensions and fix it at the calibrated target position; move the two-degree-of-freedom linear slide and acquire a series of images of the black and white checkerboard template from different angles and positions using a vision camera;

[0028] ​Step 2.2: Equation Construction: For each image of the black and white checkerboard template, the following relationship holds:

[0029] (1)

[0030] in, This represents the transformation matrix from the calibration board coordinate system to the camera coordinate system. This represents the transformation matrix from the slide table coordinate system to the slide table base coordinate system;

[0031] Due to the transformation matrix from the slide base coordinate system to the calibration plate coordinate system Keeping the coordinates unchanged, construct the calibration equations from the slide coordinate system to the camera coordinate system:

[0032] (2)

[0033] Step 2.3: Solving for the transformation matrix: The Tais method is used to solve for the transformation matrix from the slide coordinate system to the camera coordinate system. This includes rotation matrices and translation vectors.

[0034] Furthermore, the region overlap design method specifically includes the following steps:

[0035] Step 3.1: Circumferential Region Division: The outer circumferential surface of the cylindrical workpiece is considered as being divided into two parts. It consists of elongated image regions arranged along the axis of the cylindrical workpiece; the length of each elongated image region is... Width is The central angle of the cylindrical workpiece corresponding to each elongated image region is... The corresponding field of view angle of the visual camera is The overlap rate between two adjacent elongated image regions is [missing information]. The central angular displacement of the cylindrical workpiece in each rotation driven by the active roller conveyor is: Maximize the central angular displacement The objective function and constraints are as follows:

[0036] (3)

[0037] (4)

[0038] In the formula, and These are the minimum and maximum overlap rates between two adjacent elongated image regions, and the surface approximation error threshold. This is used to control the surface approximation error of the outer circular surface region of a cylindrical workpiece observed by a vision camera, which is approximately an elongated strip-shaped image region unfolded into a planar diagram. It is the horizontal field of view of the visual camera. It is the angle between two lines that pass through the optical center of the lens and are tangent to the circular cross-section of the cylindrical workpiece. It is the diameter of the circular cross-section of the cylindrical workpiece. It is the diameter of the roller conveyor. It is the axial distance between the active roller conveyor and the passive roller conveyor. It is the focal length of the visual camera. It is the width of the imaging plane of the vision camera. It is the distance from the optical center of the vision camera to the lower generatrix of the cylindrical workpiece. It is the distance from the optical center of the vision camera to the generatrix of the roller conveyor. Z is the distance from the optical center of the vision camera to the lower generatrix of the roller conveyor. + Represents the set of positive integers;

[0039] Step 3.2: Axial Direction Region Division: Divide each elongated graphic region evenly into... A length of The rectangular image patch has an overlap rate of 0.5% between two adjacent rectangular image patches. The parameters , , and The following relationship exists:

[0040] (5)

[0041] Minimize the number of rectangular image blocks The objective function and constraints are as follows:

[0042] (6)

[0043] (7)

[0044] In the formula, and These are the minimum and maximum overlap rates between two adjacent rectangular image patches, respectively. It is the length of the imaging plane of the visual camera;

[0045] Step 3.3: Combining the objective functions and constraints of Steps 3.1 and 3.2, a unified multi-objective nonlinear optimization model is constructed. A multi-objective particle swarm optimization algorithm is then employed to simultaneously maximize the desired outcome. and minimize parameters , , , , , .

[0046] Furthermore, the circumferential rotation scanning motion is realized by a displacement and attitude adjustment mechanism, which includes a roller drive motor, an active roller conveyor, and a passive roller conveyor. The active roller conveyor is driven by the roller drive motor, and the cylindrical workpiece is supported by both the active and passive roller conveyors. The static friction between the active roller conveyor and the cylindrical workpiece drives the cylindrical workpiece to rotate in the circumferential direction. The circumferential rotation scanning motion is planned based on the width b of the elongated image region. y With the diameter d of the cylindrical workpiece w Calculate the equivalent angular displacement θ of circular attitude adjustment y :

[0047] (8)

[0048] Furthermore, the two-dimensional translational scanning motion is realized by a two-degree-of-freedom linear slide, which includes a longitudinal drive motor, a longitudinal compound screw guide rail, a transverse drive motor, and a transverse compound screw guide rail. The longitudinal compound screw guide rail is driven by the longitudinal drive motor, and the transverse compound screw guide rail is driven by the transverse drive motor. The two-dimensional translational scanning motion planning uses the axis of the active roller conveyor as the transverse reference zero point, and calculates the transverse centering displacement l of the transverse drive motor in the direction of the transverse compound screw guide rail when aligning with the axis of the cylindrical workpiece. y :

[0049] (9)

[0050] Among them, l r This is the distance between the passive roller conveyor axis and the active roller conveyor axis;

[0051] The equivalent longitudinal stepping linear displacement l of the longitudinal drive motor in the longitudinal composite screw guide direction z :

[0052] (10)

[0053] Furthermore, the image block stitching correction method specifically includes the following steps:

[0054] Step 4.1: Determining the stitching method: When stitching multiple rectangular image blocks, the image coordinate system of the first rectangular image block is used as the stitching coordinate system, and the axis direction of the cylindrical workpiece is used as the main stitching direction. Calculate the main displacement reference value f of the rectangular image block in the main stitching direction. z :

[0055] f z = a z (1 - ρ z (11)

[0056] When stitching together multiple elongated image regions, the image coordinate system of the first elongated image region is used as the stitching coordinate system, and the radial direction of the cylindrical workpiece is used as the main stitching direction. The main displacement reference value f of the elongated image region in the main stitching direction is calculated. y :

[0057] f y = b y (1 – ρ y (12)

[0058] Step 4.2: Image block registration: In the stitching coordinate system of the rectangular image blocks, using the principal displacement reference value f z As the initial estimate, the image feature registration method is used to calculate the pose transformation matrix T of the next rectangular image block in the stitching coordinate system. b In the stitched coordinate system of the elongated image region, the principal displacement reference value f y As the initial estimate, the image feature registration method is used to calculate the pose transformation matrix T of the next elongated image region in the stitching coordinate system. a ;

[0059] Step 4.3: Image stitching initialization: When stitching multiple rectangular image blocks, assume that the pixel points of the merged rectangular image blocks in the stitching coordinate system are... Initialize the new pixel P in the next rectangular image block that will be merged. b The coordinates (p, q) are (0, 0); when stitching together multiple elongated image regions, assume that the pixel points of the merged elongated image regions in the stitching coordinate system are... Initialize the new pixel P in the next elongated image region that is about to be merged. a The coordinates (p, q) are (0, 0);

[0060] Step 4.4: Stitching Coordinate Transformation: When stitching multiple rectangular image blocks, in the stitching coordinate system, assume the length of the stitched rectangular image block region is... Then set splicing area A. b The boundary is [(0,0), ( Using the pose transformation matrix T b Transform the coordinates (p, q) of the new pixel in the rectangular image block to the stitching coordinate system to obtain the corresponding stitching point coordinates (p, q). t , q t If coordinates (p) t , q t (Exceeding splicing area A) b If so, proceed to step 4.8; otherwise, obtain the new pixel P. b The brightness value I of (p, q)b (p, q), proceed to step 4.5;

[0061] When stitching together multiple elongated image regions, in the stitching coordinate system, assume the width of the stitched elongated image regions is... Then set splicing area A. a The boundary is [(0,0), (l w , Using the pose transformation matrix T a Transform the coordinates (p, q) of the new pixel in the elongated image region to the stitching coordinate system to obtain the corresponding stitching point coordinates (p, q). t , q t If coordinates (p) t , q t (Exceeding splicing area A) a If so, proceed to step 4.8; otherwise, obtain the new pixel P. a The brightness value I of (p, q) a (p, q), proceed to step 4.5;

[0062] Step 4.5: Overlapping Area Judgment: In the stitching coordinate system of the rectangular image blocks, check the coordinate position (p t , q t ) corresponding pixels Does it already have a brightness value? In the stitching coordinate system of the elongated image region, check the coordinate position (p... t , q t ) corresponding pixels Does the above pixel already have a brightness value? or Already have brightness values or Then the new pixel P b (p, q) or P a Transform (p, q) to the overlapping region of the stitching coordinate system, then proceed to step 4.6; if the above pixels... or If there is no brightness value, then the new pixel P b (p, q) or P a Transform (p, q) to the extended region of the spliced ​​coordinate system, and proceed to step 4.7;

[0063] Step 4.6: Overlapping area processing: Set a new pixel P b (p, q) and P a The brightness fusion weights for (p, q) are w b and w a Then, in the stitching coordinate system of the rectangular image blocks, the overlapping pixels are calculated. New brightness value for

[0064] (13)

[0065] Using overlapping pixels New brightness value Update original brightness value :

[0066] (14)

[0067] In the stitching coordinate system of the elongated image region, the overlapping pixels are calculated. New brightness value for

[0068] (15)

[0069] Using overlapping pixels New brightness value Update original brightness value :

[0070] (16)

[0071] Proceed to step 4.8;

[0072] Step 4.7: Expanding the region processing: In the stitching coordinate system of the rectangular image patch, the new pixel point P... b Brightness values ​​of (p, q) Assign the value to the corresponding pixel. :

[0073] (17)

[0074] In the stitching coordinate system of the elongated image region, the new pixel P is... a Brightness values ​​of (p, q) Assign the value to the corresponding pixel. :

[0075] (18)

[0076] Proceed to step 4.8;

[0077] Step 4.8: Stitching Completion Detection: In the stitching coordinate system of the rectangular image block, determine the new pixel point P. b Does the coordinate (p, q) traverse all regions of the next rectangular image patch? In the stitching coordinate system of the elongated image region, determine whether the new pixel P... aCheck if the coordinates (p, q) have traversed all regions of the next long strip image area; if the coordinates (p, q) have traversed all regions, then the image block stitching is complete, and proceed to step 4.9; otherwise, update the coordinates (p, q) of the new pixel point, and proceed to step 4.4.

[0078] Step 4.9: Post-processing of stitched areas: In the stitched image area, starting from coordinates (0, 0), check each pixel in turn to see if it has a brightness value; if the pixel does not have a brightness value, take the average brightness of all pixels with brightness values ​​in the eight neighboring areas of the pixel as the brightness value of the pixel.

[0079] Furthermore, the method for detecting and controlling the air pressure in the sealed chamber specifically includes the following steps:

[0080] Step 5.1: In-cabin air pressure detection: Multiple air pressure sensors installed around the vacuum-adjustable cabin are used to detect the air pressure values ​​at multiple points inside the cabin in real time;

[0081] Step 5.2: Construction of the in-cabin flow field: Based on the turbulent kinetic energy and dissipation rate transport equations of the k-ε turbulence model, the Reynolds stress is correlated with the time-averaged velocity gradient through the eddy viscosity assumption. The pressure-velocity coupled field is calculated by operator splitting and multiple pressure corrections of the PISO algorithm. A real-time pressure field model is constructed for the real-time air pressure values ​​at multiple points inside the cabin.

[0082] Step 5.3: In-cabin pressure control: Based on a real-time pressure field model, a model predictive control method is used to control the real-time vacuum level inside the cabin: When it is necessary to increase the vacuum level inside the cabin, the intake solenoid valve is closed and the exhaust solenoid valve is opened, and air inside the cabin is drawn out by the vacuum pump and discharged to the outside of the sealed cabin through the exhaust solenoid valve; when it is necessary to decrease the vacuum level inside the cabin, the vacuum pump and the exhaust solenoid valve are closed and the intake solenoid valve is opened, and outside air enters the cabin through the intake solenoid valve; when the actual air pressure is close to the given air pressure, the opened exhaust solenoid valve or intake solenoid valve is gradually closed; when the actual air pressure reaches the given air pressure, all solenoid valves are closed.

[0083] Furthermore, the image feature similarity calculation method specifically includes the following steps:

[0084] Step 6.1: Image block registration: Low vacuum workpiece surface image I d The image coordinate system is used as the reference coordinate system, with (0, 0, 0) as the reference coordinate system. T As the initial estimate, the image feature registration method is used to calculate the surface image I of the high-vacuum workpiece. h Pose transformation matrix T in the reference coordinate system h ;

[0085] Step 6.2: Image coordinate transformation: Based on the pose transformation matrix T h Image I of the surface of a high-vacuum workpiece h Transform to the reference coordinate system to obtain the reference workpiece surface image I. c ;

[0086] Step 6.3: Cell-related calculations: For image I c Image I of cell i in row i and column j c (i, j) and image I d Image I of cell i in row i and column j d (i, j), the pixel correlation calculation method is used to obtain the image I of each pair of cells. c (i, j) and I d Image similarity s(i, j) to (i, j):

[0087] (19)

[0088] Wherein, for the cell image I in the i-th row and j-th column c (i, j), Let be the brightness of the pixel in the p-th row and q-th column. For cell image I c The average brightness of (i, j); for the cell image I in the i-th row and j-th column. d (i, j), Let be the brightness of the pixel in the p-th row and q-th column. For cell image I d The average brightness of (i, j).

[0089] Compared with the prior art, the present invention has at least the following beneficial effects or advantages:

[0090] This invention proposes a surface defect visual inspection and control method based on the micro-deformation characteristics of vacuum pressure difference. By adjusting the vacuum environment to induce micro-deformation of workpiece defects, active light illumination to highlight surface visual features, multi-degree-of-freedom orientation adjustment to ensure no blind spots in the inspection of all surfaces, and coordinated control of the environmental control, orientation adjustment, and inspection device to improve inspection efficiency, this method solves the problems of poor workpiece adaptability, low defect detection accuracy, low equipment operating efficiency, and susceptibility to environmental interference in existing methods. It significantly improves the accuracy, efficiency, and robustness of visual inspection of workpiece surface defects. Attached Figure Description

[0091] The present invention will now be described in further detail with reference to the accompanying drawings:

[0092] Figure 1 This is a schematic diagram of the composition of the surface defect visual inspection equipment based on the micro-deformation characteristics of vacuum pressure difference in this invention;

[0093] Figure 2 This is a schematic diagram of the adjustable vacuum chamber in this invention;

[0094] Figure 3 This is a schematic diagram of the composition of the visual imaging detection system in this invention;

[0095] Figure 4 This is a schematic diagram of the multi-degree-of-freedom detection and positioning fixture in this invention;

[0096] Figure 5 This is the overall flowchart of the visual inspection and control method for workpiece defects in this invention;

[0097] Figure 6 This is a flowchart of the workpiece surface image generation and control method in this invention;

[0098] Figure 7 This is a schematic diagram of the Eye-in-Hand calibration method in this invention;

[0099] Figure 8 This is a schematic diagram of the circumferential region division of the cylindrical workpiece in this invention;

[0100] Figure 9 This is a schematic diagram illustrating the division of the axial region of the cylindrical workpiece in this invention;

[0101] Figure 10 This is a schematic diagram of the camera imaging model along the axial direction of the cylindrical workpiece in this invention;

[0102] Figure 11 This is a schematic diagram of the displacement and attitude adjustment mechanism in this invention;

[0103] Figure 12 This is a schematic diagram of the structure of the two-degree-of-freedom linear slide in this invention;

[0104] Figure 13 This is a flowchart of the image block stitching correction method in this invention;

[0105] In the diagram, A is the vacuum-adjustable chamber, B is the visual imaging inspection system, C is the multi-degree-of-freedom inspection and positioning fixture, and D is the defect detection and control system; A1 is the chamber structural component, A2 is the vacuum chamber door, A3 is the vacuum degree control unit, A4 is the vacuum chamber guide rail, and A5 is the guide rail limiter; B1 is the two-degree-of-freedom linear slide, B2 is the visual camera, and B3 is the active light source array; C1 is the fixture base platform, C2 is the workpiece transport vehicle, and C3 is the displacement and attitude adjustment mechanism; B1-1 is the motion slide, B1-2 is the longitudinal drive motor, and B1-3 is... B1-4 is the horizontal drive shaft of the slide table; B1-5 is the longitudinal composite screw guide rail; B1-6 is the longitudinal composite nut slider; B1-7 is the transverse drive motor; B1-8 is the transverse composite screw guide rail; C3-1 is the roller conveyor drive motor; C3-2 is the active roller conveyor; C3-3 is the fixed passive roller conveyor; C3-4 is the displacement passive roller conveyor; C3-5 is the passive roller conveyor base; C3-6 is the roller conveyor displacement guide rail; C3-7 is the roller conveyor displacement slider; C3-8 is the roller conveyor displacement screw; C3-9 is the roller conveyor displacement nut. Detailed Implementation

[0106] 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0107] The present invention will now be described in further detail with reference to the accompanying drawings.

[0108] like Figures 1 to 4As shown, a surface defect visual inspection equipment based on vacuum pressure difference micro-deformation characteristics includes a vacuum-adjustable chamber A, a multi-degree-of-freedom detection and positioning fixture C, a visual imaging inspection system B, and a defect detection and control system D. The vacuum-adjustable chamber includes a chamber structure A1, a vacuum door A2, a vacuum degree control unit A3, a vacuum chamber track A4, and a guide rail limiter A5. The chamber structure A1 and the vacuum door A2 form a sealed chamber, and the internal air pressure of the sealed chamber is adjusted by the vacuum degree control unit A3. The multi-degree-of-freedom detection and positioning fixture C includes a fixture base platform C1, a workpiece transport vehicle C2, and a displacement and attitude adjustment mechanism C3. The workpiece transport vehicle C... 2. The workpiece is transported into and out of the sealed chamber. When the vacuum level inside the sealed chamber changes, the defective parts on the workpiece surface undergo micro-deformation. The displacement and attitude adjustment mechanism C3 changes the workpiece's posture, so that all surfaces of the workpiece sequentially enter the field of view of the visual imaging detection system. The visual imaging detection system B includes a two-degree-of-freedom linear slide B1, a vision camera B2, and an active light source array B3. The active light source array B3 projects colored light onto the workpiece surface, and the micro-deformed parts on the workpiece surface form different visual features. The two-degree-of-freedom linear slide B1 carries the vision camera B2 to collect visual feature images of the workpiece surface. The defect detection control system D includes... The system includes a control computer, a power supply, and a human-machine interface terminal. The input terminal of the control computer is electrically connected to the pressure sensor and vision camera B2 of the vacuum degree control unit A3. The output terminal of the control computer is electrically connected to the vacuum chamber door A2, the vacuum pump A3-1 of the vacuum degree control unit A3, the workpiece transport cart C2, the displacement and attitude adjustment mechanism C3, and the two-degree-of-freedom linear slide B1. The input and output terminals of the control computer are electrically connected to the human-machine interface terminal, receiving control commands from the terminal to open the vacuum chamber door A2, control the workpiece transport cart C2 to transport workpieces into or out of the chamber, and control the workpiece transport cart C2 to enter the reference position for chamber detection via the guide rail limiter A5. The system controls the vacuum chamber door A2 to close and form a sealed chamber with the chamber structure component A1. It also coordinates the displacement and attitude adjustment mechanism C3 and the two-degree-of-freedom linear slide B1 to sequentially bring all surfaces of the workpiece into the field of view of the vision camera B2. The vacuum degree control unit A3 adjusts the internal air pressure of the sealed chamber from atmospheric pressure to a predetermined vacuum state. The vision camera B2 collects atmospheric pressure characteristic images and vacuum characteristic images of the workpiece surface under the predetermined vacuum state. By comparing these images, it identifies abnormal areas in the atmospheric pressure and vacuum characteristic images, thereby identifying defects with micro-deformation on the workpiece surface under the predetermined vacuum state. Therefore, a surface defect visual inspection equipment based on vacuum pressure difference micro-deformation characteristics utilizes an adjustable vacuum chamber A to provide an air pressure regulation environment, a multi-degree-of-freedom detection and positioning fixture C to realize workpiece transportation and attitude adjustment, a visual imaging detection system B to collect defect characteristic images, and a defect detection control system D to coordinate the actions of each system and complete defect identification.Through the full-process automation of "environmental control - posture adjustment - visual acquisition - intelligent analysis", it is especially suitable for hidden surface defects that are difficult to identify by traditional visual inspection, and achieves accurate detection of hidden defects on the surface of workpieces.

[0109] like Figure 5 As shown, white rounded rectangles represent visual inspection control methods; gridded right-angled rectangles represent primary components of surface defect visual inspection equipment, such as multi-degree-of-freedom inspection and positioning fixture C, visual imaging inspection system B, and defect detection and control system D; vertical right-angled rectangles represent secondary components of surface defect visual inspection equipment, such as two-degree-of-freedom linear slide B1. The dashed lines represent the boundary between the offline and online stages of the surface defect visual inspection control methods; dashed arrows represent parameter or attribute settings made by various visual inspection control sub-methods to the surface defect visual inspection equipment components in the offline stage; dashed lines represent the association between various visual inspection control sub-methods and corresponding surface defect visual inspection equipment components in the online stage; and solid arrows represent the logical order relationship between various visual inspection control sub-methods.

[0110] In this implementation case, for the aforementioned visual inspection equipment for workpiece surface defects, a surface defect visual inspection control method based on the micro-deformation characteristics of vacuum pressure difference includes an offline process for visual imaging inspection system coordinate calibration, workpiece defect detection image region division, workpiece defect detection scanning motion planning, and workpiece defect detection control parameter setting, as well as an online process for workpiece positioning, clamping and chamber entry control, workpiece defect detection task initialization, workpiece surface image generation process control, sealed chamber air pressure detection and control, deformation feature image comparison and defect recognition, workpiece exit control and tooling unloading, and workpiece defect detection task completion determination.

[0111] like Figure 5 As shown above the dashed line, the offline detection setup process includes the following steps:

[0112] Step 1: Coordinate Calibration of the Visual Imaging Inspection System: First, the intrinsic parameters of the visual camera B2 in the visual imaging inspection system B are calibrated using the Zhang Zhengyou calibration method, including focal length, principal point coordinates, and radial distortion parameters. Second, the extrinsic parameters of the position and attitude of the visual camera B2 mounted on the two-degree-of-freedom linear slide B1 are calibrated using the Eye-in-Hand calibration method, including the rotation matrix and translation vector between the camera coordinate system and the slide coordinate system. The extrinsic parameters are used to correct the position and attitude errors of the visual camera mounted on the two-degree-of-freedom linear slide B1. Then, based on the positioning reference holes of the cabin structural component A1, the reference zero points of the two-degree-of-freedom linear slide B1 and the vacuum chamber track A4 are aligned, establishing the scanning motion coordinate system between the two-degree-of-freedom linear slide B1 and the displacement and attitude adjustment mechanism C3 on the workpiece transport vehicle C2 running on the vacuum chamber track A4. The coordinate calibration sub-method of the visual imaging inspection system calibrates the intrinsic and extrinsic parameters of the visual imaging inspection system B.

[0113] Step 2: Workpiece defect detection image region segmentation: Calculate the region at distance h based on the intrinsic parameters of the vision camera. w Image region A on the imaging plane w The length and width dimensions; for diameter d w Length l w For a cylindrical workpiece, considering the length a of the imaging plane region. c Width b c The vision camera, using the region overlap design method, first divides the cylindrical workpiece into ρ regions along the circumference. y The overlap rate is uniformly divided into n y There are several long strip-shaped image regions, each with a width of b. y Then, each elongated image region is aligned with the axis of the cylindrical workpiece by ρ. z The overlap rate is uniformly divided into n z There are 3 rectangular image blocks, each with a length of a. z Guarantee the above n y ×n z The surface formed by assembling the image blocks can completely cover the outer surface of the cylindrical workpiece. The sub-method for dividing the workpiece defect detection image region is to divide the defect detection control system D into a long strip image region and a set of rectangular image blocks for the cylindrical workpiece.

[0114] Step 3: Setting workpiece defect detection control parameters: For cylindrical workpieces, set the internal air pressure value p1 of the high vacuum sealed chamber and the internal air pressure value p2 of the low vacuum sealed chamber, where p2>p1, and the vacuum pressure difference Δp=p2-p1; set the image similarity threshold s of the workpiece surface images under high vacuum and low vacuum conditions. yThe sub-method for setting control parameters for workpiece defect detection sets the internal air pressure values ​​of the sealed chamber under high and low vacuum conditions, as well as the image similarity threshold of the workpiece surface image under the above two conditions, for the defect detection control system D.

[0115] Step 4: Workpiece Defect Detection Scanning Motion Planning: The workpiece defect detection scanning motion is formed by the relative motion between the vision camera and the cylindrical workpiece, created by the rotation of the cylindrical workpiece in the circumferential direction controlled by the displacement and attitude adjustment mechanism C3 and the translation of the vision camera carried by the two-degree-of-freedom linear slide B1 in the horizontal plane. This includes circumferential rotation scanning motion and two-dimensional translational scanning motion. The circumferential rotation scanning motion planning generates circumferential attitude adjustment angular displacement equivalents, and the two-dimensional translational scanning motion planning generates lateral centerline displacement and longitudinal step line displacement equivalents. These lateral centerline displacement, longitudinal step line displacement equivalents, and circumferential attitude adjustment angular displacement equivalents constitute the scanning motion planning command. The workpiece defect detection scanning motion planning sub-method sets the circumferential attitude adjustment angular displacement equivalent for the multi-degree-of-freedom detection and positioning fixture C, and sets the lateral centerline displacement and longitudinal step line displacement equivalents for the vision imaging detection system B.

[0116] like Figure 5 As shown below the dashed line, the online detection and control process includes the following steps:

[0117] Step 5: Workpiece Positioning, Clamping, and Cabin Entry Control: At the start of the online detection and control process, the workpiece transport vehicle C2 is positioned at the starting position of the tooling base track on the tooling base platform C1. The cylindrical workpiece is placed between the active roller conveyor C3-2 and the passive roller conveyor, with one end of the cylindrical workpiece aligned with the reference zero point of the displacement and attitude adjustment mechanism. A workpiece overlength detection sensor is used to detect whether the other end of the cylindrical workpiece is overlength. If the other end is overlength, the online detection and control process terminates; otherwise, the vacuum chamber door A2 is opened, and the workpiece transport vehicle C2 is started to move into the cabin on the tooling base track. The tooling base track is aligned with the vacuum chamber track A4. After passing through the vacuum chamber door A2, the workpiece transport vehicle C2 enters the vacuum chamber track A4 until it is detected by the guide rail limiter A5 at the end of the vacuum chamber track A4. At this point, the workpiece transport vehicle C2 stops, the vacuum chamber door A2 is closed, and the cabin structure A1 and the vacuum chamber door A2 form a sealed cabin. Proceed to Step 6.

[0118] Step 6: Initialize the workpiece defect detection task: Initialize the roller drive motor C3-1 of the active roller conveyor C3-2 to the reference zero point, and initialize the sequence number k of the circular attitude adjustment angular displacement sequence. y =0; Initialize the motion slide B1-1 of the two-degree-of-freedom linear slide B1 to the reference zero point, execute the lateral centering displacement command through the lateral drive motor B1-6, control the vision camera B2 carried by the motion slide B1-1 to align with the axis of the cylindrical workpiece in the lateral direction, reach the centering reference zero point, and initialize the sequence number k of the longitudinal stepping linear displacement sequence.z =0; Open the active light source array B3 on both sides of the top of the cabin structure A1, and form specific optical imaging features on the upper half of the outer circular surface of the cylindrical workpiece by actively projecting specific light; Proceed to step 7.

[0119] like Figure 6 As shown, the workpiece surface image generation process control sub-method includes multiple steps within the dashed box, and these internal steps are related to the sealed chamber air pressure detection and control sub-method. The area outside the dashed box represents the workpiece surface image generation process control, sealed chamber air pressure detection and control, and deformation feature image comparison and defect identification sub-methods. The following section combines... Figure 6 The specific explanation of step 7, the control process for generating the workpiece surface image, is as follows.

[0120] Step 7: Workpiece surface image generation process control: Initialize process control variable n c =0;

[0121] Step 7.1: High Vacuum Control of the Sealed Chamber: Set the target gas pressure inside the sealed chamber under high vacuum conditions to p1, and the process control variable n c Add 1, proceed to step 8. Step 7.1 calls step 8 to perform air pressure detection and control of the sealed chamber. After step 8 is completed, return to step 7.3 to perform the camera longitudinal scan image generation operation.

[0122] Step 7.2: Low Vacuum Control of the Sealed Chamber: Set the target gas pressure inside the sealed chamber under low vacuum conditions to p2, and the process control variable n c Add 1 to proceed to the next step. Step 7.2 calls step 8 to perform air pressure detection and control of the sealed chamber. After step 8 is completed, return to step 7.3 to perform the camera longitudinal scan image generation operation.

[0123] Step 7.3: Camera longitudinal scan image generation: First, the vision camera carried by the motion slide is controlled by the longitudinal drive motor to complete a longitudinal stepping line displacement equivalent l. z Then, a vision camera is used to acquire a portion of the upper half of the outer surface of the cylindrical workpiece, forming a rectangular image block; the longitudinal image scanning process of longitudinal stepping line displacement equivalent drive and image acquisition of the outer surface of the cylindrical workpiece is repeated until n images are acquired. z n rectangular image blocks are used to correct the image block stitching method. z A rectangular image block is formed into a long strip image region; then, the longitudinal drive motor controls the motion slide to carry the vision camera back to the centering reference zero point, and proceeds to step 7.4. Step 7.3 completes the acquisition of rectangular image blocks and the stitching of long strip image regions on the outer surface of a cylindrical workpiece under a circular attitude adjustment angular displacement equivalent.

[0124] Step 7.4: Circular Image Acquisition Completion Determination: Determine the circumferential attitude adjustment angular displacement sequence number k. y =n y If the condition is not met, the circumferential image acquisition is incomplete, and proceed to step 7.5; otherwise, proceed to step 7.6.

[0125] Step 7.5: Workpiece Circumferential Rotation Image Scanning: Based on the equivalent of the circumferential attitude adjustment angular displacement, control the active roller conveyor to drive the cylindrical workpiece to complete one circumferential step angular displacement in the circumferential direction. The circumferential attitude adjustment angular displacement sequence number is k. y Add 1, return to step 7.3. Add a circular attitude adjustment angle displacement, then call step 7.3 again to perform rectangular image block acquisition and long strip image region stitching.

[0126] Step 7.6: Generation of surface image of cylindrical workpiece: for n y A long strip-shaped image region is used to generate a high-vacuum workpiece surface image I under high-vacuum conditions using an image block stitching and correction method. h Generating low-vacuum workpiece surface images I under low-vacuum conditions d Proceed to step 7.7. Step 7.6 stitches together all the elongated image regions to form a complete image of the outer cylindrical surface of the workpiece.

[0127] Step 7.7: Determine if image acquisition is complete: Determine the process control variable n c If the condition is not met, the image acquisition is not completed. Proceed to step 7.2 to acquire and stitch together the image area of ​​the outer cylindrical surface of the workpiece under low vacuum conditions. Otherwise, proceed to step 9.

[0128] Step 8: Sealed Chamber Pressure Detection and Control: Target pressure value p inside the sealed chamber. t The actual air pressure value was adjusted to [p] using a sealed chamber air pressure detection and control method. t -Δp t , p t +Δp t Within the range, the Δp t For the target air pressure value p t The error is checked; return to step 7.3. After step 8 is called, return to step 7.3 to generate the camera's longitudinal scan image.

[0129] Step 9: Deformation Feature Image Comparison and Defect Identification: For high-vacuum workpiece surface images I h and low vacuum workpiece surface image I d First, adaptive image filtering is used to remove image noise; then, n... e ×n eThe pixel region is evenly divided into cells in the two images above. The image similarity s(i,j) of each pair of cell images is calculated using the image similarity calculation method. If the image similarity s(i,j) of each pair of cell images is ≥ s y If the surface of the workpiece is free of defects, then there are no defects on the workpiece surface; otherwise, there are defects on the workpiece surface, and the defects are located at s(i, j). y The cell image area; proceed to step 10. Step 9: Determine whether there are defects on the workpiece surface based on the image similarity between the two images under high vacuum and low vacuum conditions, and identify the location of the defects in the images.

[0130] Step 10: Workpiece exit control and tooling unloading: Open the vacuum chamber door, start the workpiece transport vehicle to run out of the chamber on the vacuum chamber track. After passing through the vacuum chamber door, the workpiece transport vehicle enters the tooling base track. Close the vacuum chamber door until the workpiece transport vehicle stops when it is detected by the guide rail limiter at the end of the tooling base track. The workpiece transport vehicle returns to the starting position of the tooling base track; remove the cylindrical workpiece between the active roller conveyor and the passive roller conveyor; proceed to step 11.

[0131] Step 11: Workpiece defect detection task completion determination: If there is no new workpiece defect detection task, the entire online process of workpiece defect detection control ends; otherwise, return to step 5.

[0132] like Figure 7 As shown in this implementation example, the Eye-in-Hand calibration method specifically includes the following steps:

[0133] Step 2.1: Image Acquisition: Prepare a black and white checkerboard template with known physical dimensions and fix it at the calibrated target position; move the two-degree-of-freedom linear slide and acquire a series of images of the black and white checkerboard template from different angles and positions using a vision camera;

[0134] Step 2.2: Equation Construction: For each image of the black and white checkerboard template, the following relationship holds:

[0135] (1)

[0136] in, This represents the transformation matrix from the calibration board coordinate system to the camera coordinate system. This represents the transformation matrix from the slide table coordinate system to the slide table base coordinate system;

[0137] Due to the transformation matrix from the slide base coordinate system to the calibration plate coordinate system Keeping the coordinates unchanged, construct the calibration equations from the slide coordinate system to the camera coordinate system:

[0138] (2) ​

[0139] Step 2.3: Solving for the transformation matrix: The Tais method is used to solve for the transformation matrix from the slide coordinate system to the camera coordinate system. This includes rotation matrices and translation vectors.

[0140] like Figure 8 , Figure 9 and Figure 10 As shown in this implementation example, the region overlap design method first divides the side of the cylindrical workpiece into multiple adjacent and partially overlapping elongated image regions along the circumferential direction, then divides each elongated image region into multiple adjacent and partially overlapping rectangular image blocks along the axial direction, and finally constructs a multi-objective nonlinear optimization model to solve for the relevant image region division parameters. Specifically, it includes the following steps:

[0141] Step 3.1: Circumferential Region Division: The outer circumferential surface of the cylindrical workpiece is considered as being divided into two parts. It consists of elongated image regions arranged along the axis of the cylindrical workpiece; the length of each elongated image region is... Width is The central angle of the cylindrical workpiece corresponding to each elongated image region is... The corresponding field of view angle of the visual camera is The overlap rate between two adjacent elongated image regions is [missing information]. The central angular displacement of the cylindrical workpiece in each rotation driven by the active roller conveyor is: Maximize the central angular displacement The objective function and constraints are as follows:

[0142] (3)

[0143] (4)

[0144] In the formula, and These are the minimum and maximum overlap rates between two adjacent elongated image regions, and the surface approximation error threshold. This is used to control the surface approximation error of the outer circular surface region of a cylindrical workpiece observed by a vision camera, which is approximately an elongated strip-shaped image region unfolded into a planar diagram. It is the horizontal field of view of the visual camera. It is the angle between two lines that pass through the optical center of the lens and are tangent to the circular cross-section of the cylindrical workpiece. It is the diameter of the circular cross-section of the cylindrical workpiece. It is the diameter of the roller conveyor. It is the axial distance between the active roller conveyor and the passive roller conveyor. It is the focal length of the visual camera. It is the width of the imaging plane of the vision camera. It is the distance from the optical center of the vision camera to the lower generatrix of the cylindrical workpiece. It is the distance from the optical center of the vision camera to the generatrix of the roller conveyor. Z is the distance from the optical center of the vision camera to the lower generatrix of the roller conveyor. + Represents the set of positive integers;

[0145] Step 3.2: Axial Direction Region Division: Divide each elongated graphic region evenly into... A length of The rectangular image patch has an overlap rate of 0.5% between two adjacent rectangular image patches. The parameters , , and The following relationship exists:

[0146] (5)

[0147] Minimize the number of rectangular image blocks The objective function and constraints are as follows:

[0148] (6)

[0149] (7)

[0150] In the formula, and These are the minimum and maximum overlap rates between two adjacent rectangular image patches, respectively. It is the length of the imaging plane of the visual camera;

[0151] Step 3.3: Combining the objective functions and constraints of Steps 3.1 and 3.2, a unified multi-objective nonlinear optimization model is constructed. A multi-objective particle swarm optimization algorithm is then employed to simultaneously maximize the desired outcome. and minimize parameters , , , , , .

[0152] like Figure 11As shown in this embodiment, the circumferential rotation scanning motion is achieved by the displacement and attitude adjustment mechanism C3. The displacement and attitude adjustment mechanism C3 includes a roller drive motor C3-1, an active roller conveyor C3-2, and a passive roller conveyor (divided into a fixed passive roller conveyor C3-3 and a displacement passive roller conveyor C3-4). The active roller conveyor C3-2 is driven by the roller drive motor C3-1. The cylindrical workpiece is supported by both the active roller conveyor C3-2 and the passive roller conveyor. The static friction between the active roller conveyor C3-2 and the cylindrical workpiece drives the cylindrical workpiece to rotate in the circumferential direction. The circumferential rotation scanning motion is planned based on the width b of the elongated image region. y With the diameter d of the cylindrical workpiece w Calculate the equivalent angular displacement θ of circular attitude adjustment y :

[0153] (8)

[0154] like Figure 12 As shown in this embodiment, the two-dimensional translational scanning motion is realized by a two-degree-of-freedom linear slide B1. The two-degree-of-freedom linear slide B1 includes a longitudinal drive motor B1-2, a longitudinal composite screw guide rail B1-4, a transverse drive motor B1-6, and a transverse composite screw guide rail B1-7. The longitudinal composite screw guide rail B1-4 is driven by the longitudinal drive motor B1-2, and the transverse composite screw guide rail B1-7 is driven by the transverse drive motor B1-6. The two-dimensional translational scanning motion is planned with the axis of the active roller conveyor C3-2 as the transverse reference zero point. The transverse centering displacement l of the transverse drive motor in the direction of the transverse composite screw guide rail B1-7 is calculated when aligning with the axis of the cylindrical workpiece. y :

[0155] (9)

[0156] Among them, l r This is the distance between the axis of the passive roller conveyor C3-4 and the axis of the active roller conveyor C3-2.

[0157] The equivalent longitudinal stepping linear displacement l of the longitudinal drive motor B1-2 in the longitudinal composite lead screw guide B1-4 direction z :

[0158] (10)

[0159] like Figure 13 As shown in this embodiment, the image block stitching correction method specifically includes the following steps:

[0160] Step 4.1: Determining the stitching method: When stitching multiple rectangular image blocks, the image coordinate system of the first rectangular image block is used as the stitching coordinate system, and the axis direction of the cylindrical workpiece is used as the main stitching direction. Calculate the main displacement reference value f of the rectangular image block in the main stitching direction.z :

[0161] f z = a z (1 - ρ z (11)

[0162] When stitching together multiple elongated image regions, the image coordinate system of the first elongated image region is used as the stitching coordinate system, and the radial direction of the cylindrical workpiece is used as the main stitching direction. The main displacement reference value f of the elongated image region in the main stitching direction is calculated. y :

[0163] f y = b y (1 – ρ y (12)

[0164] Step 4.2: Image block registration: In the stitching coordinate system of the rectangular image blocks, using the principal displacement reference value f z As the initial estimate, the image feature registration method is used to calculate the pose transformation matrix T of the next rectangular image block in the stitching coordinate system. b In the stitched coordinate system of the elongated image region, the principal displacement reference value f y As the initial estimate, the image feature registration method is used to calculate the pose transformation matrix T of the next elongated image region in the stitching coordinate system. a ;

[0165] Step 4.3: Image stitching initialization: When stitching multiple rectangular image blocks, assume that the pixel points of the merged rectangular image blocks in the stitching coordinate system are... Initialize the new pixel P in the next rectangular image block that will be merged. b The coordinates (p, q) are (0, 0); when stitching together multiple elongated image regions, assume that the pixel points of the merged elongated image regions in the stitching coordinate system are... Initialize the new pixel P in the next elongated image region that is about to be merged. a The coordinates (p, q) are (0, 0);

[0166] Step 4.4: Stitching Coordinate Transformation: When stitching multiple rectangular image blocks, in the stitching coordinate system, assume the length of the stitched rectangular image block region is... Then set splicing area A. b The boundary is [(0,0), ( Using the pose transformation matrix T b Transform the coordinates (p, q) of the new pixel in the rectangular image block to the stitching coordinate system to obtain the corresponding stitching point coordinates (p, q). t , qt If coordinates (p) t , q t (Exceeding splicing area A) b If so, proceed to step 4.8; otherwise, obtain the new pixel P. b The brightness value I of (p, q) b (p, q), proceed to step 4.5;

[0167] When stitching together multiple elongated image regions, in the stitching coordinate system, assume the width of the stitched elongated image regions is... Then set splicing area A. a The boundary is [(0,0), (l w , Using the pose transformation matrix T a Transform the coordinates (p, q) of the new pixel in the elongated image region to the stitching coordinate system to obtain the corresponding stitching point coordinates (p, q). t , q t If coordinates (p) t , q t (Exceeding splicing area A) a If so, proceed to step 4.8; otherwise, obtain the new pixel P. a The brightness value I of (p, q) a (p, q), proceed to step 4.5;

[0168] Step 4.5: Overlapping Area Judgment: In the stitching coordinate system of the rectangular image blocks, check the coordinate position (p t , q t ) corresponding pixels Does it already have a brightness value? In the stitching coordinate system of the elongated image region, check the coordinate position (p... t , q t ) corresponding pixels Does the above pixel already have a brightness value? or Already have brightness values or Then the new pixel P b (p, q) or P a Transform (p, q) to the overlapping region of the stitching coordinate system, then proceed to step 4.6; if the above pixels... or If there is no brightness value, then the new pixel P b (p, q) or P a Transform (p, q) to the extended region of the spliced ​​coordinate system, and proceed to step 4.7;

[0169] Step 4.6: Overlapping area processing: Set a new pixel Pb (p, q) and P a The brightness fusion weights for (p, q) are w b and w a Then, in the stitching coordinate system of the rectangular image blocks, the overlapping pixels are calculated. New brightness value for

[0170] (13)

[0171] Using overlapping pixels New brightness value Update original brightness value :

[0172] (14)

[0173] In the stitching coordinate system of the elongated image region, the overlapping pixels are calculated. New brightness value for

[0174] (15)

[0175] Using overlapping pixels New brightness value Update original brightness value :

[0176] (16)

[0177] Proceed to step 4.8;

[0178] Step 4.7: Expanding the region processing: In the stitching coordinate system of the rectangular image patch, the new pixel point P... b Brightness values ​​of (p, q) Assign the value to the corresponding pixel. :

[0179] (17)

[0180] In the stitching coordinate system of the elongated image region, the new pixel P is... a Brightness values ​​of (p, q) Assign the value to the corresponding pixel. :

[0181] (18)

[0182] Proceed to step 4.8;

[0183] Step 4.8: Stitching Completion Detection: In the stitching coordinate system of the rectangular image block, determine the new pixel point P.b Does the coordinate (p, q) traverse all regions of the next rectangular image patch? In the stitching coordinate system of the elongated image region, determine whether the new pixel P... a Check if the coordinates (p, q) have traversed all regions of the next long strip image area; if the coordinates (p, q) have traversed all regions, then the image block stitching is complete, and proceed to step 4.9; otherwise, update the coordinates (p, q) of the new pixel point, and proceed to step 4.4.

[0184] Step 4.9: Post-processing of stitched areas: In the stitched image area, starting from coordinates (0, 0), check each pixel in turn to see if it has a brightness value; if the pixel does not have a brightness value, take the average brightness of all pixels with brightness values ​​in the eight neighboring areas of the pixel as the brightness value of the pixel.

[0185] In this embodiment, the method for detecting and controlling the air pressure of the sealed chamber specifically includes the following steps:

[0186] Step 5.1: In-cabin air pressure detection: Multiple air pressure sensors installed around the vacuum-adjustable cabin are used to detect the air pressure values ​​at multiple points inside the cabin in real time;

[0187] Step 5.2: Construction of the in-cabin flow field: Based on the turbulent kinetic energy and dissipation rate transport equations of the k-ε turbulence model, the Reynolds stress is correlated with the time-averaged velocity gradient through the eddy viscosity assumption. The pressure-velocity coupled field is calculated by operator splitting and multiple pressure corrections of the PISO algorithm. A real-time pressure field model is constructed for the real-time air pressure values ​​at multiple points inside the cabin.

[0188] Step 5.3: In-cabin pressure control: Based on a real-time pressure field model, a model predictive control method is used to control the real-time vacuum level inside the cabin: When it is necessary to increase the vacuum level inside the cabin, the intake solenoid valve is closed and the exhaust solenoid valve is opened, and air inside the cabin is drawn out by the vacuum pump and discharged to the outside of the sealed cabin through the exhaust solenoid valve; when it is necessary to decrease the vacuum level inside the cabin, the vacuum pump and the exhaust solenoid valve are closed and the intake solenoid valve is opened, and outside air enters the cabin through the intake solenoid valve; when the actual air pressure is close to the given air pressure, the opened exhaust solenoid valve or intake solenoid valve is gradually closed; when the actual air pressure reaches the given air pressure, all solenoid valves are closed.

[0189] In this implementation example, the image feature similarity calculation method specifically includes the following steps:

[0190] Step 6.1: Image block registration: Low vacuum workpiece surface image I d The image coordinate system is used as the reference coordinate system, with (0, 0, 0) as the reference coordinate system. T As the initial estimate, the image feature registration method is used to calculate the surface image I of the high-vacuum workpiece. hPose transformation matrix T in the reference coordinate system h ;

[0191] Step 6.2: Image coordinate transformation: Based on the pose transformation matrix T h Image I of the surface of a high-vacuum workpiece h Transform to the reference coordinate system to obtain the reference workpiece surface image I. c ;

[0192] Step 6.3: Cell-related calculations: For image I c Image I of cell i in row i and column j c (i, j) and image I d Image I of cell i in row i and column j d (i, j), the pixel correlation calculation method is used to obtain the image I of each pair of cells. c (i, j) and I d Image similarity s(i, j) to (i, j):

[0193] (19)

[0194] Wherein, for the cell image I in the i-th row and j-th column c (i, j), Let be the brightness of the pixel in the p-th row and q-th column. For cell image I c The average brightness of (i, j); for the cell image I in the i-th row and j-th column. d (i, j), Let be the brightness of the pixel in the p-th row and q-th column. For cell image I d The average brightness of (i, j).

[0195] The above describes in detail specific implementation examples of the present invention. The present invention has many specific applications. The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for visual inspection and control of surface defects based on the micro-deformation characteristics of vacuum pressure difference, characterized in that... The offline process of detection settings includes coordinate calibration of the visual imaging detection system, image region division of workpiece defect detection, scanning motion planning of workpiece defect detection, and setting of control parameters for workpiece defect detection; and online process of detection control including workpiece positioning and clamping and chamber entry control, initialization of workpiece defect detection task, process control of workpiece surface image generation, air pressure detection and control of sealed chamber, comparison of deformation feature images and defect identification, control of workpiece exit from chamber and tooling unloading, and determination of completion of workpiece defect detection task. The offline detection setup process includes the following steps: Step 1: Coordinate calibration of the visual imaging detection system: First, the intrinsic parameters of the visual camera in the visual imaging detection system are calibrated using the Zhang Zhengyou calibration method, including focal length, principal point coordinates, and radial distortion parameters. Second, the extrinsic parameters of the position and attitude of the visual camera mounted on the two-degree-of-freedom linear slide are calibrated using the Eye-in-Hand calibration method, including the rotation matrix and translation vector between the camera coordinate system and the slide coordinate system. The position and attitude error of the visual camera mounted on the two-degree-of-freedom linear slide is corrected using the extrinsic parameters. Then, the reference zero point of the two-degree-of-freedom linear slide and the vacuum chamber track is found according to the positioning reference hole of the cabin structure component, and the scanning motion coordinate system of the two-degree-of-freedom linear slide and the displacement and attitude adjustment mechanism on the workpiece transport vehicle running on the vacuum chamber track is established. Step 2: Workpiece defect detection image region segmentation: Calculate the region at distance h based on the intrinsic parameters of the vision camera. w Image region A on the imaging plane w The length and width dimensions; for diameter d w Length l w For a cylindrical workpiece, considering the length a of the imaging plane region. c Width b c The vision camera, using the region overlap design method, first divides the cylindrical workpiece into ρ regions along the circumference. y The overlap rate is uniformly divided into n y There are several long strip-shaped image regions, each with a width of b. y Then, each elongated image region is aligned with the axis of the cylindrical workpiece by ρ. z The overlap rate is uniformly divided into n z There are 3 rectangular image blocks, each with a length of a. z Guarantee the above n y ×n z The surface formed by assembling the image blocks can completely cover the outer surface of the cylindrical workpiece. Step 3: Setting workpiece defect detection control parameters: For cylindrical workpieces, set the internal air pressure value p1 of the high vacuum sealed chamber and the internal air pressure value p2 of the low vacuum sealed chamber, where p2>p1, and the vacuum pressure difference Δp=p2-p1; set the image similarity threshold s of the workpiece surface images under high vacuum and low vacuum conditions. y ; Step 4: Workpiece Defect Detection Scanning Motion Planning: The workpiece defect detection scanning motion is the relative motion between the vision camera and the cylindrical workpiece, formed by the rotation of the cylindrical workpiece in the circumferential direction controlled by the displacement and attitude adjustment mechanism and the translation of the vision camera carried by the two-degree-of-freedom linear slide in the horizontal plane. It includes circumferential rotation scanning motion and two-dimensional translational scanning motion. The circumferential rotation scanning motion planning generates the equivalent circumferential attitude adjustment angular displacement, and the two-dimensional translational scanning motion planning generates the equivalent lateral centerline displacement and longitudinal step line displacement. The scanning motion planning command is composed of the equivalent lateral centerline displacement, longitudinal step line displacement, and circumferential attitude adjustment angular displacement. The online detection and control process includes the following steps: Step 5: Workpiece Positioning, Clamping, and Cabin Entry Control: At the start of the online detection and control process, the workpiece transport vehicle is positioned at the beginning of the tooling base track on the tooling base platform. The cylindrical workpiece is placed between the active roller conveyor and the passive roller conveyor, with one end of the cylindrical workpiece aligned with the reference zero point of the displacement and attitude adjustment mechanism. A workpiece overlength detection sensor is used to detect whether the other end of the cylindrical workpiece is overlength. If the other end is overlength, the online detection and control process terminates; otherwise, the vacuum chamber door is opened, and the workpiece transport vehicle is started to move into the cabin on the tooling base track. The tooling base track is aligned with the vacuum chamber track. After the workpiece transport vehicle passes through the vacuum chamber door, it enters the vacuum chamber track until it is detected by the guide rail limiter at the end of the vacuum chamber track. At this point, the workpiece transport vehicle stops, the vacuum chamber door is closed, and the cabin structure and the vacuum chamber door form a sealed cabin. Proceed to Step 6. Step 6: Initialize the workpiece defect detection task: Initialize the roller drive motor of the active roller conveyor to the reference zero point, and initialize the sequence number k of the circular attitude adjustment angular displacement sequence. y =0; Initialize the motion slide of the two-degree-of-freedom linear slide to the reference zero point, execute the lateral centering displacement command through the lateral drive motor, control the vision camera carried by the motion slide to align with the axis of the cylindrical workpiece laterally, reach the centering reference zero point, and initialize the sequence number k of the longitudinal stepping linear displacement sequence. z =0; Open the active light source arrays on both sides of the top of the cabin structure component, and form specific optical imaging features on the upper half of the outer circular surface of the cylindrical workpiece by actively projecting specific light; Proceed to step 7; Step 7: Workpiece surface image generation process control: Initialize process control variable n c =0; Step 7.1: High Vacuum Control of the Sealed Chamber: Set the target gas pressure inside the sealed chamber under high vacuum conditions to p1, and the process control variable n c Add 1, proceed to step 8; Step 7.2: Low Vacuum Control of the Sealed Chamber: Set the target gas pressure inside the sealed chamber under low vacuum conditions to p2, and the process control variable n c Add 1, proceed to step 8; Step 7.3: Camera longitudinal scan image generation: First, the vision camera carried by the motion slide is controlled by the longitudinal drive motor to complete a longitudinal stepping line displacement equivalent l. z Then, a vision camera is used to acquire a portion of the upper half of the outer surface of the cylindrical workpiece, forming a rectangular image block; the longitudinal image scanning process of longitudinal stepping line displacement equivalent drive and image acquisition of the outer surface of the cylindrical workpiece is repeated until n images are acquired. z n rectangular image blocks are used to correct the image block stitching method. z A rectangular image block forms a long strip image region; then, the longitudinal drive motor controls the motion slide to carry the vision camera back to the centering reference zero point, and proceeds to step 7.4; Step 7.4: Circular Image Acquisition Completion Determination: Determine the circumferential attitude adjustment angular displacement sequence number k. y =n y If the condition is not met, the circumferential image acquisition is incomplete, and proceed to step 7.5; otherwise, proceed to step 7.

6. Step 7.5: Workpiece Circumferential Rotation Image Scanning: Based on the equivalent of the circumferential attitude adjustment angular displacement, control the active roller conveyor to drive the cylindrical workpiece to complete one circumferential step angular displacement in the circumferential direction. The circumferential attitude adjustment angular displacement sequence number is k. y Add 1, then return to step 7.3; Step 7.6: Generation of surface image of cylindrical workpiece: for n y A long strip-shaped image region is used to generate a high-vacuum workpiece surface image I under high-vacuum conditions using an image block stitching and correction method. h Generating low-vacuum workpiece surface images I under low-vacuum conditions d Proceed to step 7.7; Step 7.7: Determine if image acquisition is complete: Determine the process control variable n c If the condition is not met, the image acquisition is not completed and proceed to step 7.2; otherwise, proceed to step 9. Step 8: Sealed Chamber Pressure Detection and Control: Target pressure value p inside the sealed chamber. t The actual air pressure value was adjusted to [p] using a sealed chamber air pressure detection and control method. t -Δp t , p t +Δp t Within the range, the Δp t For the target air pressure value p t The error; return to step 7.3; Step 9: Deformation Feature Image Comparison and Defect Identification: For high-vacuum workpiece surface images I h and low vacuum workpiece surface image I d First, adaptive image filtering is used to remove image noise; then, n... e ×n e The pixel region is uniformly divided into cells in the two images above. The image similarity s(i, j) of each pair of cell images is calculated using the image similarity calculation method. If the image similarity s(i, j) of each pair of cell images is ≥ s y If the surface of the workpiece is free of defects, then there are no defects on the workpiece surface; otherwise, there are defects on the workpiece surface, and the defects are located at s(i, j). y The cell image area; proceed to step 10;​ Step 10: Workpiece exit control and tooling unloading: Open the vacuum chamber door, start the workpiece transport vehicle to move out of the chamber on the vacuum chamber track. After passing through the vacuum chamber door, the workpiece transport vehicle enters the tooling base track. Close the vacuum chamber door until the workpiece transport vehicle stops when detected by the guide rail limiter at the end of the tooling base track. The workpiece transport vehicle returns to the starting position of the tooling base track; remove the cylindrical workpiece between the active roller conveyor and the passive roller conveyor; proceed to step 11. Step 11: Workpiece defect detection task completion determination: If there is no new workpiece defect detection task, the entire online process of workpiece defect detection control ends; otherwise, return to step 5.

2. The surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics according to claim 1, characterized in that: The Eye-in-Hand hand-eye calibration method specifically includes the following steps: Step 2.1: Image Acquisition: Prepare a black and white checkerboard template with known physical dimensions and fix it at the calibrated target position; move the two-degree-of-freedom linear slide and acquire a series of images of the black and white checkerboard template from different angles and positions using a vision camera; Step 2.2: Equation Construction: For each image of the black and white checkerboard template, the following relationship holds: (1) in, This represents the transformation matrix from the calibration board coordinate system to the camera coordinate system. This represents the transformation matrix from the slide table coordinate system to the slide table base coordinate system; Due to the transformation matrix from the slide base coordinate system to the calibration plate coordinate system Keeping the coordinates unchanged, construct the calibration equations from the slide coordinate system to the camera coordinate system: (2) Step 2.3: Solving for the transformation matrix: The Tais method is used to solve for the transformation matrix from the slide coordinate system to the camera coordinate system. This includes rotation matrices and translation vectors.

3. The surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics according to claim 1, characterized in that: The region overlap design method specifically includes the following steps: Step 3.1: Circumferential Region Division: The outer circumferential surface of the cylindrical workpiece is considered as being divided into two parts. It consists of elongated image regions arranged along the axis of the cylindrical workpiece; the length of each elongated image region is... Width is The central angle of the cylindrical workpiece corresponding to each elongated image region is... The corresponding field of view angle of the visual camera is The overlap rate between two adjacent elongated image regions is [missing information]. The central angular displacement of the cylindrical workpiece in each rotation driven by the active roller conveyor is: Maximize the central angular displacement The objective function and constraints are as follows: (3) (4) In the formula, and These are the minimum and maximum overlap rates between two adjacent elongated image regions, and the surface approximation error threshold. This is used to control the surface approximation error of the outer circular surface region of a cylindrical workpiece observed by a vision camera, which is approximately an elongated strip-shaped image region unfolded into a planar diagram. It is the horizontal field of view of the visual camera. It is the angle between two lines that pass through the optical center of the lens and are tangent to the circular cross-section of the cylindrical workpiece. It is the diameter of the circular cross-section of the cylindrical workpiece. It is the diameter of the roller conveyor. It is the axial distance between the active roller conveyor and the passive roller conveyor. It is the focal length of the visual camera. It is the width of the imaging plane of the vision camera. It is the distance from the optical center of the vision camera to the lower generatrix of the cylindrical workpiece. It is the distance from the optical center of the vision camera to the generatrix of the roller conveyor. Z is the distance from the optical center of the vision camera to the lower generatrix of the roller conveyor. + Represents the set of positive integers; Step 3.2: Axial Direction Region Division: Divide each elongated graphic region evenly into... A length of The rectangular image patch has an overlap rate of 0.5% between two adjacent rectangular image patches. The parameters , , and The following relationship exists: (5) Minimize the number of rectangular image blocks The objective function and constraints are as follows: (6) (7) In the formula, and These are the minimum and maximum overlap rates between two adjacent rectangular image patches, respectively. It is the length of the imaging plane of the visual camera; Step 3.3: Combining the objective functions and constraints of Steps 3.1 and 3.2, a unified multi-objective nonlinear optimization model is constructed. A multi-objective particle swarm optimization algorithm is then employed to simultaneously maximize the desired outcome. and minimize parameters , , , , , .

4. The surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics according to claim 1, characterized in that: The circumferential rotation scanning motion is achieved by a displacement and attitude adjustment mechanism, which includes a roller drive motor, an active roller conveyor, and a passive roller conveyor. The active roller conveyor is driven by the roller drive motor, and the cylindrical workpiece is supported by both the active and passive roller conveyors. The static friction between the active roller conveyor and the cylindrical workpiece drives the cylindrical workpiece to rotate in the circumferential direction. The circumferential rotation scanning motion is planned based on the width b of the elongated image region. y With the diameter d of the cylindrical workpiece w Calculate the equivalent angular displacement θ of circular attitude adjustment y : (8)。 5. The surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics according to claim 1, characterized in that: The two-dimensional translational scanning motion is achieved by a two-degree-of-freedom linear slide, which includes a longitudinal drive motor, a longitudinal compound screw guide rail, a transverse drive motor, and a transverse compound screw guide rail. The longitudinal compound screw guide rail is driven by the longitudinal drive motor, and the transverse compound screw guide rail is driven by the transverse drive motor. The two-dimensional translational scanning motion planning uses the axis of the active roller conveyor as the transverse reference zero point, and calculates the transverse centering displacement l of the transverse drive motor in the direction of the transverse compound screw guide rail when aligning with the axis of the cylindrical workpiece. y : (9) Among them, l r This is the distance between the passive roller conveyor axis and the active roller conveyor axis; The equivalent longitudinal stepping linear displacement l of the longitudinal drive motor in the longitudinal composite screw guide direction z : (10)。 6. The surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics according to claim 1, characterized in that: The image block stitching correction method specifically includes the following steps: Step 4.1: Determining the stitching method: When stitching multiple rectangular image blocks, the image coordinate system of the first rectangular image block is used as the stitching coordinate system, and the axis direction of the cylindrical workpiece is used as the main stitching direction. Calculate the main displacement reference value f of the rectangular image block in the main stitching direction. z : f z = a z (1 - p z ) (11) When stitching together multiple elongated image regions, the image coordinate system of the first elongated image region is used as the stitching coordinate system, and the radial direction of the cylindrical workpiece is used as the main stitching direction. The main displacement reference value f of the elongated image region in the main stitching direction is calculated. y : f y = b y (1 – p y ) (12) Step 4.2: Image block registration: In the stitching coordinate system of the rectangular image blocks, using the principal displacement reference value f z As the initial estimate, the image feature registration method is used to calculate the pose transformation matrix T of the next rectangular image block in the stitching coordinate system. b In the stitched coordinate system of the elongated image region, the principal displacement reference value f y As the initial estimate, the image feature registration method is used to calculate the pose transformation matrix T of the next elongated image region in the stitching coordinate system. a ; Step 4.3: Image stitching initialization: When stitching multiple rectangular image blocks, assume that the pixel points of the merged rectangular image blocks in the stitching coordinate system are... Initialize the new pixel P in the next rectangular image block that will be merged. b The coordinates (p, q) are (0, 0); when stitching together multiple elongated image regions, assume that the pixel points of the merged elongated image regions in the stitching coordinate system are... Initialize the new pixel P in the next elongated image region that is about to be merged. a The coordinates (p, q) are (0, 0); Step 4.4: Stitching Coordinate Transformation: When stitching multiple rectangular image blocks, in the stitching coordinate system, assume the length of the stitched rectangular image block region is... Then set splicing area A. b The boundary is [(0,0), ( Using the pose transformation matrix T b Transform the coordinates (p, q) of the new pixel in the rectangular image block to the stitching coordinate system to obtain the corresponding stitching point coordinates (p, q). t , q t If coordinates (p) t , q t (Exceeding splicing area A) b If so, proceed to step 4.8; otherwise, obtain the new pixel P. b The brightness value I of (p,q) b (p, q), proceed to step 4.5; When stitching together multiple elongated image regions, in the stitching coordinate system, assume the width of the stitched elongated image regions is... Then set splicing area A. a The boundary is [(0,0), (l w , Using the pose transformation matrix T a Transform the coordinates (p, q) of the new pixel in the elongated image region to the stitching coordinate system to obtain the corresponding stitching point coordinates (p, q). t ,q t If coordinates (p) t , q t (Exceeding splicing area A) a If so, proceed to step 4.8; otherwise, obtain the new pixel P. a The brightness value I of (p, q) a (p, q), proceed to step 4.5; Step 4.5: Overlapping Area Judgment: In the stitching coordinate system of the rectangular image blocks, check the coordinate position (p t , q t ) corresponding pixels Does it already have a brightness value? In the stitching coordinate system of the elongated image region, check the coordinate position (p... t ,q t ) corresponding pixels Does the above pixel already have a brightness value? or Already have brightness values or Then the new pixel P b (p, q) or P a Transform (p, q) to the overlapping region of the stitching coordinate system, then proceed to step 4.6; if the above pixels... or If there is no brightness value, then the new pixel P b (p, q) or P a Transform (p, q) to the extended region of the spliced ​​coordinate system, and proceed to step 4.7; Step 4.6: Overlapping area processing: Set a new pixel P b (p, q) and P a The brightness fusion weights for (p, q) are w b and w a Then, in the stitching coordinate system of the rectangular image blocks, the overlapping pixels are calculated. New brightness value for (13) Using overlapping pixels New brightness value Update original brightness value : (14) In the stitching coordinate system of the elongated image region, the overlapping pixels are calculated. New brightness value for (15) Using overlapping pixels New brightness value Update original brightness value : (16) Proceed to step 4.8; Step 4.7: Expanding the region processing: In the stitching coordinate system of the rectangular image patch, the new pixel point P is... b Brightness values ​​of (p, q) Assign the value to the corresponding pixel. : (17) In the stitching coordinate system of the elongated image region, the new pixel P is... a Brightness values ​​of (p, q) Assign the value to the corresponding pixel. : (18) Proceed to step 4.8; Step 4.8: Stitching Completion Detection: In the stitching coordinate system of the rectangular image block, determine the new pixel point P. b Does the coordinate (p, q) traverse all regions of the next rectangular image patch? In the stitching coordinate system of the elongated image region, determine whether the new pixel P... a Check if the coordinates (p, q) have traversed all regions of the next long strip image area; if the coordinates (p, q) have traversed all regions, then the image block stitching is complete, and proceed to step 4.9; otherwise, update the coordinates (p, q) of the new pixel point, and proceed to step 4.

4. Step 4.9: Post-processing of stitched areas: In the stitched image area, starting from coordinates (0, 0), check each pixel in turn to see if it has a brightness value; if the pixel does not have a brightness value, take the average brightness of all pixels with brightness values ​​in the eight neighboring areas of the pixel as the brightness value of the pixel.

7. The surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics according to claim 1, characterized in that: The method for detecting and regulating the air pressure of the sealed chamber specifically includes the following steps: Step 5.1: In-cabin air pressure detection: Multiple air pressure sensors installed around the vacuum-adjustable cabin are used to detect the air pressure values ​​at multiple points inside the cabin in real time; Step 5.2: Construction of the in-cabin flow field: Based on the turbulent kinetic energy and dissipation rate transport equations of the k-ε turbulence model, the Reynolds stress is correlated with the time-averaged velocity gradient through the eddy viscosity assumption. The pressure-velocity coupled field is calculated by operator splitting and multiple pressure corrections of the PISO algorithm. A real-time pressure field model is constructed for the real-time air pressure values ​​at multiple points inside the cabin. Step 5.3: In-cabin pressure control: Based on a real-time pressure field model, a model predictive control method is used to control the real-time vacuum level inside the cabin: When it is necessary to increase the vacuum level inside the cabin, the intake solenoid valve is closed and the exhaust solenoid valve is opened, and air inside the cabin is drawn out by the vacuum pump and discharged to the outside of the sealed cabin through the exhaust solenoid valve; when it is necessary to decrease the vacuum level inside the cabin, the vacuum pump and the exhaust solenoid valve are closed and the intake solenoid valve is opened, and outside air enters the cabin through the intake solenoid valve; when the actual air pressure is close to the given air pressure, the opened exhaust solenoid valve or intake solenoid valve is gradually closed; when the actual air pressure reaches the given air pressure, all solenoid valves are closed.

8. The surface defect visual inspection and control method based on vacuum pressure difference micro-deformation characteristics according to claim 1, characterized in that: The image feature similarity calculation method specifically includes the following steps: Step 6.1: Image block registration: Low vacuum workpiece surface image I d The image coordinate system is used as the reference coordinate system, with (0,0,0) as the reference coordinate system. T As the initial estimate, the image feature registration method is used to calculate the surface image I of the high-vacuum workpiece. h Pose transformation matrix T in the reference coordinate system h ; Step 6.2: Image coordinate transformation: Based on the pose transformation matrix T h Image I of the surface of a high-vacuum workpiece h Transform to the reference coordinate system to obtain the reference workpiece surface image I. c ; Step 6.3: Cell-related calculations: For image I c Image I of cell i in row i and column j c (i, j) and image I d Image I of cell i in row i and column j d (i, j), the pixel correlation calculation method is used to obtain the image I of each pair of cells. c (i, j) and I d Image similarity s(i, j) to (i, j): (19) Wherein, for the cell image I in the i-th row and j-th column c (i, j), Let be the brightness of the pixel in the p-th row and q-th column. For cell image I c The average brightness of (i, j); for the cell image I in the i-th row and j-th column. d (i, j), Let be the brightness of the pixel in the p-th row and q-th column. For cell image I d The average brightness of (i, j).

Images