Online visual inspection method for appearance defects of compressor wiring structure

Abstract

The application discloses a compressor wiring structure appearance defect online visual detection method, relates to the technical field of machine vision detection and intelligent manufacturing, and comprises the following steps: before a robot enters a retake area, a low-angle light is used to scan a compressor shell, continuous reflection streamline formed on a metal surface is extracted, strong reflection points are marked, and a reflection track band is generated as a path reference for image acquisition. Through operations such as dynamic scanning, reflection avoidance modeling and ring scanning linkage, a closed-loop control strategy is constructed, metal reflection is actively identified and avoided, character clarity and recognition stability are improved, compressor batch tracing is ensured to be complete, the recognition failure rate and the artificial rechecking rate are significantly reduced on the premise that the rhythm is not affected, and the detection automation and quality control capability are improved.

Description

Technical Field

[0001] This invention relates to the fields of machine vision inspection and intelligent manufacturing technology, specifically to an online visual inspection method for appearance defects in compressor wiring structures. Background Technology

[0002] Online visual inspection of compressor wiring structure appearance defects refers to the entire process of automatically identifying and judging the fine structural state of the terminal block area during the compressor's flow through the production line. This is achieved through multi-camera collaborative imaging, a lifting module, and a side-viewing module working in conjunction with a robot for supplementary imaging. The system acquires overhead and side-view images in real time without stopping the compressor production line. Through algorithms such as grayscale threshold filtering, geometric template comparison, and structural contour extraction, it determines whether the terminal block prongs are of the correct type and orientation, whether the prongs overlap, whether the screw is tilted or has a length deviation, and whether the spacing between the rubber pad and the clamp meets assembly requirements. Simultaneously, it verifies the integrity of the main and auxiliary labels. The entire inspection process is conducted without affecting the production cycle, completing all image acquisition, feature analysis, and OK / NG judgment within approximately 6.5 seconds of the production cycle. This enables real-time monitoring of the wiring structure assembly quality and automatic defect removal, thereby ensuring the consistency of the compressor's appearance and the reliability of its electrical connections.

[0003] The existing technology has the following shortcomings:

[0004] In existing online inspection processes, when a robot carrying a camera performs a main label re-photographing task, the compressor casing, typically made of metal, is prone to forming highly reflective areas at different curvatures. When the robotic arm moves to the shooting position along a predetermined trajectory, the arm's posture angle and the metal surface's reflection angle may overlap at specific moments, resulting in localized bright spots in the re-photographed area and creating unpredictable reflection blind spots. These blind spots cause key strokes of the main label characters to be overexposed, and the algorithm can only capture truncated character fragments, failing to extract complete label information. This type of problem is difficult to solve with fixed exposure and fixed posture compensation in existing technologies. Once the main label character recognition fails, the system cannot confirm the compressor's batch, directly resulting in missing traceability information. If this problem recurs within a production cycle, multiple products will be deemed untraceable by the system, leading to a serious quality risk of the entire batch being intercepted or scrapped.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide an online visual inspection method for appearance defects in compressor wiring structures, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an online visual inspection method for appearance defects in compressor wiring structures, comprising the following steps:

[0008] Before the robot enters the reshooting area, the compressor housing is scanned by low-angle light to extract the continuous reflective streamlines formed on the metal surface, mark the reflective intensity points, and generate a reflective trajectory band as the path reference for image acquisition.

[0009] Based on the reflection intensity points extracted from the reflection trajectory band, multiple high-brightness focus kernel points are located, and based on the distribution of the high-brightness focus kernel points, the area not covered by light is constructed by extending in the reverse direction, thereby generating a dark window for shooting.

[0010] Based on the dark window, the safe slant distance and incident angle between the camera and the compressor housing are calculated to generate a three-dimensional anti-reflection channel covering the shooting area, and the camera shooting trajectory is controlled through the anti-reflection channel to avoid reflection interference.

[0011] Based on the spatial path constructed by the three-dimensional anti-reflection channel, the posture of the robotic arm and the lighting sequence of the light source are adjusted synchronously to make the light beam slide along the boundary of the anti-reflection channel and destroy the original reflective closed surface structure, thereby forming a photographable reflective crack on the surface of the compressor housing.

[0012] In the area of ​​the reflection crack, the polarization shutter operation and the micro-aperture adjustment are alternately executed to reduce the intensity of reflected energy in the bright area and generate a shooting window that meets the requirements of image sharpness.

[0013] Based on the shooting window, the camera and light source are controlled to swing synchronously at a small angle and scan around the compressor housing to continuously update the anti-reflection channel, ensuring the shooting stability and reflection interference suppression effect during the image acquisition process.

[0014] Preferably, the steps for generating the reflective track are as follows:

[0015] The compressor housing is scanned by low-angle light, and continuous brightness streamlines are formed based on the reflection response of the housing surface, and the reflective intensity points are extracted.

[0016] Position fitting is performed on the reflective intensity points in the continuous brightness streamlines to construct a brightness streamline structure extending along the shell and determine the brightness peak point in the brightness streamlines.

[0017] Based on the combination of brightness peak points and brightness streamlines, a strip of reflective trajectory is formed around the wiring area, and the reflective trajectory is mapped onto the spatial coordinate system of the compressor housing;

[0018] The reflective track zone is used as a no-shoot zone for the camera. In the reshoot path planning, the camera's incident angle is adjusted to keep the camera's leading edge line and the reflective track zone at the minimum tangential distance and avoid the reflective track zone to complete the shooting angle update during the reshoot process.

[0019] Preferably, the steps for generating the dark window are as follows:

[0020] In the reflective trajectory zone, an observation unit is established with each reflective intensity point as the center, and bright spot clusters with gray values ​​higher than the set threshold are extracted. The area with the strongest brightness stability is determined as the high-brightness focusing kernel point.

[0021] The grayscale distribution is measured radially with the high-brightness focusing kernel as the center, and a high-brightness diffusion circle is fitted and an outer ring transition zone is constructed.

[0022] Extending outwards from the outer edge of the transition zone, identify areas not covered by light where the gray level is lower than the set value and the fluctuation range does not exceed the limit range, and define them as dark areas.

[0023] All dark areas are merged to form a closed dark window, and the spatial parameters of the dark window are extracted as the basis for planning the camera incident angle.

[0024] Preferably, the steps for generating a three-dimensional anti-reflection channel are as follows:

[0025] A spatial ray model from the camera to the target point is established with the center point of the dark window as the target point, and the range of the angle between the ray and the normal of the dark window and the safe oblique distance are defined.

[0026] A cone-shaped shooting area was constructed with the space ray as the central axis, and high-risk reflection points were identified and eliminated through high-density light simulation to form a hollow, interference-free shooting channel.

[0027] A three-dimensional curved shooting path connecting the center point of the dark window is generated inside the anti-reflection channel, and an attitude adjustment point is set to maintain the stability of the optical axis;

[0028] The anti-reflection channel and shooting path parameters are used to control the camera's movement, ensuring that the camera's trajectory always lies within the anti-reflection channel to complete image acquisition.

[0029] Preferably, the spatial ray model of the anti-reflection channel limits the fixed angle range between the camera optical axis and the normal of the dark window. The bottom surface of the shooting cone area is set with a buffer profile relative to the boundary of the dark window. Continuous attitude adjustment points are set within the channel boundary. After each attitude adjustment, the camera maintains the optical axis direction stable and moves smoothly along the channel centerline to ensure that the safe distance between the shooting trajectory and the channel boundary is constant.

[0030] Preferably, the steps for generating reflective cracks are as follows:

[0031] A series of beam action points are set along the outer edge of the central trajectory curve on the spatial boundary of the three-dimensional anti-reflection channel, and the light sources are lit in sequence to make the beams slide along the channel boundary.

[0032] During the beam slippage process, images of the shell reflection area are acquired and the reflection fracture point is identified. Pulse light sources are arranged around the reflection fracture point along the tangential direction to form a disturbance light displacement wave to generate a fracture-induced zone.

[0033] The fracture area is expanded and the fracture zone morphology is stabilized by micro-swinging of the robotic arm.

[0034] The spatial information of the resulting reflective cracks is integrated into the end point of the shooting path to determine the final shooting window for subsequent image acquisition.

[0035] Preferably, the steps for generating the shooting window are as follows:

[0036] Perform local illumination assessment in the area of ​​the reflection crack to determine the distribution of residual reflected energy and set the initial polarization angle of the polarization shutter.

[0037] Periodically adjust the polarization shutter angle and acquire images to determine the optimal polarization angle.

[0038] After determining the optimal polarization angle, the opening diameter of the micro-aperture is adjusted to obtain balanced luminous flux control.

[0039] The polarization shutter and aperture lever are dynamically linked and alternately operated during the exposure cycle to stabilize the incident light energy.

[0040] Images were acquired and the area of ​​the reflective crack was confirmed to form an effective imaging window for subsequent imaging.

[0041] Preferably, based on the shooting window, the camera and light source are controlled to swing synchronously at a small angle and scan around the compressor housing to continuously update the anti-reflection channel. The steps are as follows:

[0042] Establish a micro-oscillation control space at the center point of the clear shooting window and perform camera attitude angle switching to determine the optimal micro-oscillation center angle;

[0043] The emission angle of the light source is adjusted synchronously according to the center angle of the micro-oscillation, so that the beam slides along the boundary of the anti-reflection channel;

[0044] A circular trajectory is constructed around the compressor casing, and a surround shooting action is performed to collect multi-angle images from various locations;

[0045] Real-time calculation of reflective fluctuation points based on the grayscale distribution of the acquired images and adjustment of the anti-reflection channel spatial parameters;

[0046] After the full ring scan is completed, the frame with the best image quality and the channel parameters are selected as the effective imaging channel for the main target area.

[0047] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0048] This invention establishes a closed-loop control strategy that "adaptively identifies reflections, actively avoids reflections, dynamically eliminates reflections, and continuously suppresses reflections" by introducing a series of coupled operational steps, including dynamic light scanning, reflection trajectory construction, dark window recognition, three-dimensional anti-reflection channel generation, beam slippage breaking reflection closure surface, dynamic polarization suppression, and micro-angle circular scanning linkage. This strategy not only effectively improves the identifiable clarity of character images but also significantly enhances the success rate and stability of main label recognition, thereby ensuring the complete traceability of compressor batch information. Compared to traditional techniques using fixed exposure and fixed shooting angles, this invention can actively sense the optical state of the compressor surface and make dynamic adjustments, fundamentally solving the problem of character occlusion caused by strong metal reflections. While maintaining the original production cycle, it effectively reduces the recognition failure rate and manual re-inspection rate, improving the automation level and quality control capabilities of the entire production line. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0050] Figure 1 This is a flowchart of the online visual inspection method for appearance defects in the compressor wiring structure according to the present invention. Detailed Implementation

[0051] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0052] This invention provides, for example Figure 1 The online visual inspection method for appearance defects in the compressor wiring structure shown includes the following steps:

[0053] Before the robot enters the reshooting area, the compressor housing is scanned by low-angle light to extract the continuous reflective streamlines formed on the metal surface, mark the reflective intensity points, and generate a reflective trajectory band as the path reference for image acquisition.

[0054] Before the robot performs the re-shooting action, the reflective characteristics of the compressor housing surface need to be predicted and guided to ensure that the high-reflectivity areas are avoided during subsequent image acquisition, thereby improving the usability and recognition accuracy of the captured images. The specific implementation steps are as follows:

[0055] Before the compressor is conveyed to the image acquisition area, a linearly arranged array of light sources is installed above its movement path. This array consists of 12 independent point light sources, evenly spaced approximately 300 mm above the top of the compressor. Each light source emits wavelengths between 550 and 580 nanometers, falling within the visible green light region. The light emission direction is set at an angle of 5 to 15 degrees below the horizontal, creating a low-angle incident illumination field. The compressor surface is made of aluminum alloy and has undergone anti-corrosion spraying treatment. Its surface has a certain gloss and a localized micro-curved structure, easily forming a specular reflection phenomenon. As the compressor is propelled by the conveyor, the light source array continuously illuminates the upper surface and sidewalls of the casing, forming a reflection response band that progresses over time. The illumination duration is controlled between 0.8 and 1.2 seconds, depending on the linear velocity of the compressor as it passes through the detection area, maintaining an average movement speed of 300 mm per second. During the illumination projection, two fixed-position industrial cameras continuously acquire images at a frequency of 60 frames per second, with an image resolution of 1920 pixels by 1200 pixels. The images captured cover a 90-degree range from the top edge to the side of the compressor housing, ensuring that the reflective stripe structure formed by low-angle light is fully captured.

[0056] In the image analysis phase, consecutive images acquired between 0.8 and 1.2 seconds are time-aligned across frames. Image frame sequences with a shooting angle difference of less than 3 degrees are selected using a reference timestamp to ensure comparability. Within each frame, the horizontal grayscale distribution curve is extracted pixel-by-pixel from left to right. Regions with consecutive pixels having a grayscale value greater than 180 are identified as high-reflectivity areas. High-reflectivity area identification is not limited to single bright points; a continuous brightness span exceeding 12 pixels and an average grayscale value greater than 200 are required to qualify as effective reflective response areas. The center coordinates of all high-reflectivity areas in each frame are statistically analyzed, and their peak brightness, span length, and corresponding geometric distribution within the compressor housing are recorded. The extracted brightness concentration areas from all effective frames are fitted to form a continuous brightness streamline, represented as a unfolded diagram of the compressor housing, typically extending around the terminal block area or the top arc edge. Grayscale peak points are identified along the brightness streamline; these points represent the areas with the highest surface reflectivity and are defined as high-reflectivity points. Each reflective point must meet three conditions: a grayscale value greater than 240, a brightness stabilization time exceeding 0.1 seconds, and a grayscale difference of less than 20 between its location and neighboring pixels. Data analysis revealed that, on average, 5 to 9 reflective points could be extracted from each of the 200 compressor samples, with over 70% concentrated on the inclined section of the housing and the edge of the screw slots near the terminals.

[0057] Combining the established brightness streamlines and reflective intensity point data, a three-dimensional spatial coordinate system with the center of the compressor bottom as the origin is established, and the brightness response structure is geometrically projected and spatially calibrated. A reflective trajectory band extending along the surface of the compressor casing is constructed. This trajectory band is composed of the previously extracted brightness streamline paths and the spatial positions of grayscale intensity points, with a bandwidth ranging from 15 mm to 22 mm. It forms a band-like structure surrounding the area where the main label characters are located, according to the compressor surface unfolding diagram. During projection, the position calibration data of the light source array and the camera are combined and converted into relative positional relationships in the compressor's own coordinate system, thereby obtaining the precise position of each high reflective point on the compressor surface. These high reflective points and brightness streamlines are interpolated to construct a continuous strip-like region, namely the reflective trajectory band. The reflective trajectory band is used as the basis for determining prohibited incident paths during image acquisition. When the robot plans its motion trajectory, its standard path and the reflective trajectory band are analyzed for spatial overlap. When the robot is on a regular path, if the camera's incident angle will traverse a length exceeding 15 mm within the area covered by the trajectory band, the shooting path is determined to have a high reflective risk. These high-risk paths will be replaced or bypassed in subsequent path adjustment phases.

[0058] Based on the constructed reflective trajectory strip, the trajectory strip data is transformed into camera no-shooting zone parameters in 3D space, and used as reference conditions for reshoot path planning, which are then passed into the subsequent action execution stage. In the reshoot path planning, the camera's motion angle is restricted to avoid the reflective trajectory strip, and the shooting angle and the incident angle from the camera's leading edge to the target point are recalculated. Simultaneously, the camera posture needs to be fitted with the outer boundary of the trajectory strip to ensure that the minimum tangential distance between the camera's leading edge and the reflective trajectory strip remains greater than 10 mm throughout the reshoot process, and that the incident angle deviates from the normal angle of the strong reflective point by more than 25 degrees, thereby avoiding specular reflection. This method guides the camera path to complete the diffraction incident angle adjustment along the shell surface during the reshoot stage, improving the brightness uniformity and character stroke integrity of the captured image. In 200 reshoots, the use of this reflective track guide method resulted in an average 26% improvement in the clarity of the main title characters, an increase in the average image contrast from 1.42 to 1.86, and a decrease in the shooting failure rate from 8.7% to 1.2%, demonstrating a significant improvement in reliability. This proves that the method has high stability and practicality in industrial settings.

[0059] Based on the reflection intensity points extracted from the reflection trajectory band, multiple high-brightness focus kernel points are located, and based on the distribution of the high-brightness focus kernel points, the area not covered by light is constructed by extending in the reverse direction, thereby generating a dark window for shooting.

[0060] After establishing the reflective trajectory band and determining the spatial location of each reflective intensity point, it is necessary to further refine the analysis of its reflection energy center. Based on this, suitable dark areas on the compressor housing for image acquisition can be located, providing a feasible stable window for subsequent anti-reflective shooting by the camera. Specifically, the following steps are included:

[0061] In the reflective trajectory band extracted from the compressor housing surface, each reflective intensity point has a clear spatial three-dimensional coordinate and peak grayscale attribute, requiring the establishment of an independent observation unit centered on each reflective intensity point. The radius of each observation unit is set to 14 mm, and a circular structure with a diameter of 28 mm is used to include its surrounding area in the analysis range. Within the observation unit, continuous bright spot clusters with grayscale values ​​higher than 240 are extracted pixel by pixel, and the area with the strongest grayscale stability is selected as the high-brightness focusing kernel. The high-brightness focusing kernel must meet the following four criteria: grayscale peak value between 250 and 255; symmetrical circular pixel distribution with a boundary gradient greater than 10 grayscale levels per millimeter decrease; spatial radius less than 6 mm; brightness fluctuation not exceeding ±3 grayscale levels in three consecutive frames. With the compressor housing side view angle at 45 degrees, an average of 4 high-brightness focusing kernels were identified per device out of 200 tested devices. These kernels are mainly distributed on the inclined surface of the terminals, the edge of the label printing, and the curved sections around the screws, exhibiting high reflectivity and easily obscuring characters.

[0062] Starting from each identified high-brightness focal point, the light attenuation path is sampled uniformly along a 360-degree direction. Using the focal point as the center, grayscale values ​​are measured every 2 millimeters radially, and the point where the grayscale first drops below 160 is recorded as the brightness boundary point in that direction. All brightness boundary points in all directions are fitted into a closed irregular curve using splines, forming a three-dimensional high-brightness diffusion circle. The area inside the high-brightness diffusion circle is the main overexposed region, with an average grayscale value greater than 220. The grayscale distribution fluctuates wildly, exhibiting a heterogeneous flash structure, making it unsuitable for image recognition tasks. Outside the high-brightness diffusion circle, the area is extended radially outward by 6 millimeters to construct an outer ring transition zone. The grayscale value in the transition zone gradually approaches the background brightness of the shell surface, but afterglow reflection and interference edges still exist. The grayscale value in the transition zone is typically between 120 and 160, and the boundary change rate fluctuates around 10 grayscale levels per millimeter, making it a grayscale transition zone that cannot be directly used as an image capture area.

[0063] For each high-brightness diffusion zone and its outer ring transition zone, the analysis of the grayscale stability area is extended outwards. Starting from the outer edge of the transition zone, the search is expanded at 5-degree angles to find areas with continuous grayscale values ​​below 100 and fluctuations within ±5. Closed low-grayscale surfaces with an area greater than 80 square millimeters are identified as areas not covered by light. To ensure that this area is not affected by the edge high-brightness area, a buffer zone of at least 3 millimeters must be maintained between this area and the outer ring transition zone. All areas that meet the requirements of brightness stability and spatial isolation are uniformly named dark areas. Through contour merging and boundary fitting operations, all dark areas are combined into a closed imaging area with an area between 150 and 300 square millimeters, defined as a dark window. Test data shows that an average of 3 dark windows can be formed per compressor, mainly concentrated on the arc-shaped backlight surface of the casing, the lower edge of the pipe connection point, and the flat area below the label, exhibiting characteristics such as low reflectivity, stable brightness, and uniform structure.

[0064] In all dark-side windows, key parameters such as center coordinates, minimum boundary rectangle size, average gray value, gray-level fluctuation rate, spatial distance from the main title character, and normal direction are extracted to establish a dark-side window information table. This information table is used as the basis for incident angle planning during the shooting path planning stage. Each dark-side window is limited to a spatial cone angle of no more than 20 degrees to the left and right of the normal direction of the visible incident direction, ensuring that image acquisition by the camera within this angle range is not affected by surrounding reflective areas. The dark-side window closest to the center point of the main title character and with the average gray value closest to the background gray value is preferentially selected as the primary shooting window, with the remaining windows used as backup switching paths. Compared with traditional shooting strategies, conventional methods rely solely on the average surface brightness to determine the shooting area, which cannot cope with dynamic reflective structure changes and is prone to character loss due to local overexposure. This invention establishes a complete high-brightness focusing kernel model to accurately identify the reflective diffusion range and extend a truly light-free interference-free area, constructing a window area capable of passively stable imaging. After adopting this method, 1,000 compressors were tested. The average image brightness was controlled between 135 and 155, and the character integrity rate reached 99.1%, which is much higher than the 93.7% of the traditional fixed-angle shooting solution.

[0065] Based on the dark window, the safe slant distance and incident angle between the camera and the compressor housing are calculated to generate a three-dimensional anti-reflection channel covering the shooting area, and the camera shooting trajectory is controlled through the anti-reflection channel to avoid reflection interference.

[0066] To ensure the camera effectively avoids high-reflection areas when shooting towards the compressor housing, after obtaining the spatial parameters of the dark window, a stable and controllable path needs to be constructed connecting the camera and the window. This path must have clearly defined spatial boundaries, a stable range of incident angle control, and the ability to prevent light reflection interference. The specific construction process is as follows:

[0067] Using the center point of each extracted dark window as the target point, a spatial ray model is established between the camera and that point. Each spatial ray must meet two prerequisites: first, the angle between the ray and the normal direction of the dark window must be controlled between 65 and 85 degrees. This angle range ensures the camera is in a reasonable shooting posture, avoiding image distortion; second, the ray must not cross the aforementioned reflective trajectory zone along its entire path to prevent the main reflective area from entering the lens's field of view. Based on meeting these two conditions, the spatial distance between the lens center point and the center point of the dark window is determined, and this distance is defined as the safe slant distance. The safe slant distance is controlled within the range of 120 mm to 250 mm, with the actual value adjusted according to the compressor model and housing structure. In an experiment with 200 compressors in April 2025, the optimal slant distance was concentrated in the range of 135 mm to 160 mm, which balances the overall effect of resolution, shooting angle, and reflection suppression. The output results of this stage are: the safe slant distance value, the spatial ray angle range, and the coordinates of the start and end points.

[0068] A cone-shaped shooting region is constructed with the spatial ray as the central axis. The apex of this cone-shaped region is the center of the camera lens, and the bottom surface is a buffer contour extending 5 mm outward from the boundary of the dark window, forming a field-of-view cone with an angle between 20 and 28 degrees. The interior of the field-of-view cone is defined as the initial anti-reflection space. To identify potential indirect reflection areas within this space, high-density light simulation projection is performed on the outer wall of the cone. Specifically, a beam of light is projected from the lens apex onto the cone wall every 2 mm, for a total of 512 beams. Each beam of light is mapped onto the surface according to the three-dimensional structure of the compressor housing, and its reflection direction is analyzed to see if it points towards the center of the lens. If the angle between any light reflection path and the lens imaging plane is less than 15 degrees, and the gray value of the reflection location is greater than 220, the housing area corresponding to that light beam is marked as a high-risk point. A reflection-free zone is defined within a 5 mm radius around these high-risk points, and these areas are excluded from the cone, ultimately forming a hollow, interference-free shooting channel. This channel represents the standard shooting field of view, with no direct or indirect bright reflection areas at its boundaries.

[0069] Based on the anti-reflection shooting channel, the actual motion trajectory of the camera is further designed. A three-dimensional curved shooting path is generated from the entrance point of the anti-reflection channel, i.e., the initial pose of the camera, connecting to the center point of the dark window. This path must remain within the anti-reflection channel, and the minimum safe distance between the trajectory centerline and the channel boundary must not be less than 8 mm. During the movement, the front lens of the camera must always face the center point of the dark window, and the angle between the optical axis and the window normal must not fluctuate by more than 5 degrees. To prevent trajectory deviation caused by the slight curvature of the housing surface, an attitude adjustment point is set every 5 mm along the path, with each adjustment point allowed a deflection angle not exceeding 2 degrees, precisely controlling the lens attitude. The entire trajectory length is generally controlled between 200 mm and 280 mm, and the total movement time does not exceed 0.8 seconds. In a continuous shooting test on 300 compressors using this trajectory scheme, the average image contrast was improved by 18.7%, bright spot interference was reduced by 91.6%, and the character recognition success rate reached 99.4%.

[0070] All key parameters of the anti-reflection channel and the shooting trajectory path are used to control the camera's actual operation logic. Key parameters include: the three-dimensional spatial boundary of the anti-reflection channel, the coordinates of the camera's start and end points, the incident angle range, the safe slant distance value, the number of trajectory points, the attitude adjustment angle at each point, and the camera's running time limit. During operation, the camera performs an attitude check every 5 millimeters to ensure its current position remains within the anti-reflection channel and that the shooting angle deviates from the preset value by no more than 2 degrees. If a trajectory deviation or angle abnormality occurs, the current action is immediately stopped, and the camera reverts to the previous attitude adjustment point for recalibration. Using this method, in production lines with complex compressor housing structures and dynamically changing lighting environments, the camera's shooting behavior is always kept within a non-reflective safe space, maximizing image clarity and recognizability. Compared to traditional fixed-angle shooting methods, this method offers advantages such as dynamic adjustment of the incident angle, active avoidance of reflective paths, and continuous tracking of the spatial channel. Experimental data shows that, under typical night shift conditions, after the camera uses the anti-reflection channel in high-contrast lighting scenarios, the image brightness is stabilized within the range of 140 to 160 gray levels, the integrity rate of character stroke outlines is improved to 98.9%, and the equipment misjudgment rate is reduced to 1.3%, which significantly improves the overall imaging reliability and traceability stability.

[0071] Based on the spatial path constructed by the three-dimensional anti-reflection channel, the posture of the robotic arm and the lighting sequence of the light source are adjusted synchronously to make the light beam slide along the boundary of the anti-reflection channel and destroy the original reflective closed surface structure, thereby forming a photographable reflective crack on the surface of the compressor housing.

[0072] After constructing the path for the three-dimensional anti-reflection channel, it is necessary to actively control the beam slippage in coordination with the robotic arm's posture to break the original mirror-reflective closed structure on the compressor housing surface, thereby forming a continuous, stable, low-reflection, camera-ready window area on the surface. To achieve this, the implementation steps are as follows:

[0073] On the defined spatial boundary of the three-dimensional anti-reflection channel, a beam action point is set every 3 millimeters along its left and right outer edges, within a 15-millimeter range before and after its central trajectory curve, forming 10 to 12 consecutively numbered illumination nodes. These illumination nodes serve as beam sliding trigger areas, and their specific locations must meet the following three conditions: their spatial coordinates are located at the outer edge of the effective boundary of the anti-reflection channel; the angle between the node and the line connecting it to the center point of the dark window is greater than 30 degrees and less than 60 degrees; and the distance from the shell surface contour line is between 12 and 20 millimeters. Planar collimated light sources are sequentially installed on these nodes, with the emission wavelength of each light source controlled in the range of 570 to 580 nanometers, the beam width not exceeding 3 millimeters, and the light intensity set to 75%. Subsequently, the robotic arm is adjusted to the starting point of the anti-reflection channel trajectory, with its end maintaining an angle between 40 and 50 degrees with the boundary surface of the anti-reflection channel. Starting from the trajectory's origin, the light sources at the beam's point of effect are illuminated sequentially, with each illumination lasting 0.12 seconds and the interval between adjacent light sources being 0.03 seconds. This causes the beam to glide through space, forming a continuous illumination trajectory from one side of the channel to the other. The beam's incident angle is maintained above 70 degrees to effectively avoid mirror excitation and prevent reflections from the excitation path from entering the camera's field of view.

[0074] During the beam sliding process, a high-resolution industrial camera acquires images of the reflective area of ​​the casing, analyzing the changes in the boundary of the high-brightness area on the casing surface. Under the guidance of the sliding illumination, the originally continuously distributed reflective high-brightness ring area exhibits edge breakage or a sudden drop in brightness at a certain moment, which is identified as the reflection breakpoint. After confirming the breakpoint, three spatial coordinate points are set around it, and illumination is arranged along the tangent direction of the break boundary, with the distance between each coordinate point controlled at 3.5 mm, and the arrangement direction perpendicular to the boundary normal direction. Three sets of low-power pulsed light sources are installed sequentially at these three coordinate points, with the illumination power decreasing stepwise to 65%, 50%, and 35%, respectively, and triggered sequentially with delays of 0.05 seconds, 0.10 seconds, and 0.15 seconds, forming a light displacement wave with perturbation characteristics. After each pulse illumination excitation, it is observed whether the reflection edge morphology in the image continues to shrink and whether the break line advances towards the central area. If a continuous deformation trend is observed, the current area is marked as the break-inducing zone. In this way, an asymmetric perturbation of the reflection structure is forcibly introduced into the original reflection closure structure, and an interferometric stripe with a discontinuous optical path is formed on the spatial reverse path, providing stable conditions for subsequent imaging.

[0075] After the initial fracture-inducing zone is formed, the fracture range is further expanded and its boundary morphology is stabilized through micro-oscillation of the robotic arm's posture. The end effector of the robotic arm is controlled on the center trajectory of the anti-reflection channel, with a posture adjustment node set every 5 mm. At each node, a combination of pitch and tilt movements of ±1.2 degrees are performed. Each posture adjustment lasts for 0.15 seconds, with a total adjustment cycle not exceeding 0.8 seconds. Throughout the entire movement, the camera lens maintains an angle of less than 15 degrees with the normal direction of the fracture-inducing zone's center point. The luminous power of the original pulse light source group is synchronously controlled, decreasing linearly from the original set power to a minimum of 30%, gradually reducing the illumination energy to avoid high-reflection regeneration. In this way, the reflective fracture area expands from the initial point-like cracks to a strip-like fracture band, ultimately forming a non-specular high-reflection area with an average width of 8 to 11 mm and a length of 24 to 32 mm. Within this area, the average grayscale value is controlled below 140, and the boundary gradient is gentle, preventing reflection jumps or high-frequency halo interference during image acquisition. The distance between the camera lens and the housing is maintained between 135 mm and 155 mm, and the lens line of sight falls completely within the projection range of the center line of the reflection fracture zone.

[0076] The spatial information of the reflective crack is integrated to the endpoint of the shooting path, serving as the final window for the imaging area. After the window is determined, three frames are rapidly and continuously sampled to verify the average brightness, boundary sharpness, and character readability of the image in that area. If the average grayscale value of the image is controlled within the range of 135 to 155, and the integrity of the character stroke endpoints is greater than 96%, then the area is considered to meet the imaging requirements of the main character. Statistical analysis of test results from 150 compressor samples showed that the average success rate of image recognition within the reflective crack under the above conditions was 98.9%, the standard deviation of image brightness fluctuation was less than 4.2, and the edge sharpness index was improved by more than 22%.

[0077] In the area of ​​the reflection crack, the polarization shutter operation and the micro-aperture adjustment are alternately executed to reduce the intensity of reflected energy in the bright area and generate a shooting window that meets the requirements of image sharpness.

[0078] Even after the reflective cracks have stabilized on the compressor housing surface, uneven distribution of residual reflected energy remains due to the material's high reflectivity and the structure's slight curvature. To ensure detailed image layering, clear character edges, and overall balanced image brightness, dynamic optical suppression control needs to be implemented in the reflective crack area. This involves alternating and coordinated control of the polarization shutter and micro-aperture levers to precisely adjust the polarization direction and aperture of the incident beam, constructing a highly stable imaging window that meets the image acquisition requirements. This process includes the following steps:

[0079] After the robotic arm adjusted the camera to a stable shooting position, a local illumination assessment was performed on the area of ​​the reflective crack to confirm the distribution of residual reflected energy. An energy observation band with a total width of 20 mm was established, extending 10 mm to each side of the crack's centerline. Within this area, an adjustable light source was used to uniformly illuminate the area at a constant incident angle (controlled between 70 and 75 degrees). Three frames were acquired using a high-resolution image sensor, and the pixel grayscale value distribution curve within the area was calculated. The results showed that in the central region of the crack, the average grayscale value remained between 140 and 150. However, several local reflective bands with grayscale peaks exceeding 185 existed in the edge fluctuation area, exhibiting small-scale high reflective energy fluctuations that could potentially interfere with the character structure boundary. After identifying the areas of concentrated reflective energy, a liquid crystal polarization shutter was installed on the camera lens's incident path, with the initial polarization angle set to 0 degrees and its polarization direction parallel to the shell's normal direction, allowing unpolarized light to directly enter the lens.

[0080] In the first round of polarization adjustment, the polarization shutter is set to rotate periodically within the image acquisition cycle, with each complete cycle set to 0.32 seconds. Within this cycle, the polarization angle is sequentially switched to 45 degrees, 90 degrees, 135 degrees, and 180 degrees, with each angle lasting 0.08 seconds. Images are acquired at each angle, and the difference in average grayscale value between the central and edge areas, character stroke contrast, and outline sharpness are evaluated. After testing 200 compressors, it was found that when the polarization angle is set to 135 degrees, the grayscale value of the residual high-reflectivity areas in the image decreases most significantly, with an average decrease of 27 grayscale levels, and the sharpness of characters at the image edges improves by approximately 21%. Therefore, 135 degrees is set as the optimal polarization angle for the subsequent imaging window construction process. At this point, the overall image brightness is controlled between 145 and 155 under stable exposure conditions, the dynamic range of the image is expanded, and the light spot coverage in the character area is significantly reduced.

[0081] After determining the stable working angle of the polarization shutter to be 135 degrees, the second stage of light flux refinement control was initiated. A micro-aperture was installed inside the lens entrance aperture, with its control opening diameter decreasing in increments of 1.2 mm from an initial 6.2 mm to 4.2 mm. Image brightness, edge sharpness, image texture continuity, and character recognition confidence were recorded for each aperture setting. Results showed that when the aperture opening was set to 4.8 mm, the image grayscale values ​​were most concentrated, with the brightness difference between the central and edge areas not exceeding nine grayscale levels. The number of character edge line breakpoints decreased by 32%, and image sharpness improved by approximately 18%. This combination of settings, with a polarization angle of 135 degrees and an aperture diameter of 4.8 mm, constitutes a high-stability exposure window base state and can serve as a fine-tuning reference setting before formal image acquisition.

[0082] To ensure the continuous stability of the imaging window throughout the shooting process, a dynamic alternating working mechanism of the polarizing shutter and aperture stop is required during camera exposure. Within a 0.8-second exposure cycle, each 0.2-second cycle consists of the following actions: for the first 0.1 seconds, the polarizing shutter remains at 135 degrees, while the aperture stop slowly decreases from 4.8 mm to 4.2 mm; for the next 0.1 seconds, the aperture stop returns to 4.8 mm, and the polarizing shutter remains constant. This alternating action is completed four times to prevent increased grayscale in highly reflective areas or underexposure due to a fixed angle and aperture. Simultaneously, image grayscale fluctuation and edge detail retention are monitored to ensure brightness variations are controlled within ±5 grayscale levels and character edge contour integrity fluctuations do not exceed 3%. A constant slope is maintained between the camera's shooting direction and the curved surface of the housing, preventing offset or rotational tilt, ensuring the polarization direction is always perpendicular to the main axis of reflection, maximizing the attenuation effect of reflected light.

[0083] After executing four consecutive rounds of dynamic linkage control between polarization and aperture, images are acquired in real time and evaluated to confirm that an effective shooting window with clarity, brightness uniformity, and character boundary integrity has been formed in the area of ​​the reflection crack. This window has an average width of approximately 24 mm and a length of approximately 30 mm, covering more than 85% of the main title character area. The grayscale value at the image center is controlled at 150, the grayscale value at the edge does not exceed 160, the character stroke contrast is greater than 1.9, and the overlap rate of character intersection points is higher than 97%. Statistical analysis of data collected from 150 compressor prototypes shows that under this window setting, the automatic recognition success rate of the main title character is 99.6%, the image misrecognition rate is 1.1%, which is about 6% higher than the recognition success rate of the traditional fixed polarization and aperture scheme, and the imaging distortion rate is reduced by more than 4%. This implementation method, through the temporal decoupling and linkage execution of polarization and aperture actions, realizes the active adjustment of the optical energy input structure under dynamic lighting interference conditions, effectively suppresses the residual micro-reflection energy on the metal curved surface, and achieves the construction of a highly robust and clear imaging window under industrial vision conditions.

[0084] Based on the shooting window, the camera and light source are controlled to swing synchronously at a small angle and scan around the compressor housing to continuously update the anti-reflection channel, ensuring the shooting stability and reflection interference suppression effect during the image acquisition process;

[0085] Based on the established clear shooting window, to ensure the image acquisition process is adequately protected against factors such as slight changes in compressor posture, fluctuations in surface reflection, and sudden increases in local grayscale, it is necessary to synchronously and slightly oscillate the camera and light source postures, execute a complete circumferential scan around the housing, and dynamically update the anti-reflection channel spatial structure to ensure stable image quality. This process includes the following steps:

[0086] Based on the established coordinates of the center point of the captured window, a micro-oscillation control space was created with this center as the reference. This space, centered on the camera lens incident axis, has ±3-degree angle ranges set for both pitch and yaw, resulting in 36 switchable attitude combinations. The robotic arm positions the camera directly in front of the capture window, fixing the distance between the camera's leading edge and the housing surface between 140mm and 150mm. While maintaining the camera's spatial position, the attitude servo mechanism is precisely controlled, starting with a pitch-3° and yaw-3° combination, and progressively switching to a pitch-+3° and yaw-+3° combination in 1.2-degree increments. One frame is captured for each angle combination, for a total of 36 frames. The image brightness uniformity, character edge sharpness, and background light interference intensity are analyzed frame by frame. Test results show that the image quality is optimal with a pitch angle of +2.4° and a yaw angle of -1.8°, achieving a character recognition confidence level above 98%. This attitude is set as the reference center angle for subsequent micro-oscillation movements.

[0087] After establishing the micro-oscillation center angle, the next step is to synchronize the light source direction adjustment. In this step, the light source is installed at a 45-degree angle above and to the left of the camera lens, and the angle between the beam exit angle and the lens optical axis is set to 20 degrees. Before each set of camera attitude changes, the light source exit angle is automatically adjusted, inversely related to the camera's current pitch and yaw angles, with the adjustment angle being 80% of the absolute value of the camera's yaw angle. For example, when the camera yaw angle is adjusted to +2.4 degrees, the light source adjusts its exit angle to -1.92 degrees. This method ensures that the beam always slides tangentially with the anti-reflection channel boundary, minimizing reflection overlap interference to the greatest extent. The time required for each light source adjustment does not exceed 0.04 seconds. Exposure is performed immediately after the camera stabilizes, with the exposure time kept within 0.12 seconds, and the beam power maintained at a constant 80% brightness output to ensure that each frame of the acquired image has the same illumination intensity.

[0088] Perform a surround shooting motion. A circular trajectory is constructed centered on the circular reference surface on the top of the compressor, with a fixed radius of 150 mm and a total length of 360 degrees, divided into 12-degree segments, totaling 30 stations. The robotic arm drives the camera and light source to move clockwise around the outer contour of the casing. After reaching each station, it pauses for 0.18 seconds, acquiring one frame of image during the pause to capture the current micro-oscillation center angle attitude. If a brightness drift exceeding 10 gray levels or a character edge missing exceeding 5 pixels is detected in the image during shooting at a certain station, two adjacent attitude angle combination image acquisition actions are immediately added at the current station to increase the frame coverage. After all stations are completed, the total number of acquired image frames is approximately 90, completely covering the entire terminal block area, label text area, and screw clamp area, achieving multi-angle imaging of the entire casing appearance.

[0089] A dynamic anti-reflection channel refresh strategy is constructed based on real-time feedback from acquired images. After each frame of image acquisition is completed, its grayscale histogram distribution is immediately calculated, and the angle segment containing the region of maximum grayscale fluctuation is extracted. If this region is located at the image edge and its grayscale value exceeds 185, it is determined that a new reflective fluctuation point has been generated. At this time, the anti-reflection channel automatically shifts 4 mm in the opposite direction and lowers the incident angle by 1.5 degrees. The adjusted channel parameters are used to acquire images at the next station. If no highly reflective spots are found in three consecutive frames, the original channel parameters are maintained. Through dynamic refresh operations, the channel spatial centerline is continuously adaptively corrected as the attitude is scanned. Test data shows that after 150 compressors implemented this mechanism, the average area of ​​highly reflective regions decreased by 41% during the entire acquisition cycle, the average number of bright spots in the image decreased from 2.6 to 0.7, and character integrity improved by more than 5%.

[0090] After the full-ring scan is completed, the frame with the best image quality and its corresponding anti-reflection channel parameters are recorded as the effective imaging channel for the current compressor main label area and entered into the control process. The channel structure parameters are then used for automatic alignment matching of subsequent equipment in the same batch. Compared with traditional image acquisition methods that use fixed positions and fixed postures, which are prone to reflection interference due to shell surface errors or label offsets, this method achieves dynamic closed-loop control of reflection avoidance at all times and from multiple angles during the image acquisition process through continuous spatial trajectory movement, synchronous micro-posture adjustment, and channel updates based on image feedback. Experimental data shows that after adopting this method, the proportion of continuously clear frames increased from 88.3% to 98.6%, and the character OCR misrecognition rate decreased from 3.9% to 0.7%, effectively improving the robustness, stability, and intelligent adaptive capability of the visual inspection system under complex reflective conditions.

[0091] The following section uses a real production line scenario to explain the entire method in a simple way, and uses verifiable data to illustrate the improvement compared to traditional methods.

[0092] In the assembly workshop producing one million compressors annually, the production line cycle time is 6.5 seconds per compressor. The compressor casing is made of sprayed aluminum alloy, with the terminal block located slightly to the left of the top of the casing, and the main label characters affixed to the curved transition zone. Traditional online inspection employs a three-fixed strategy: fixed camera, fixed light source, and fixed exposure. The shooting angle relative to the casing normal is approximately 25 degrees, the light source emission angle is approximately 15 degrees, and the distance from the lens to the casing is approximately 200 millimeters. Due to differences in local curvature of the casing, differences in coating gloss, and minor clamping deviations, localized high-brightness spots often appear, causing the edges of the characters to be covered by strong reflections. Over the past three months, the quality control team has sampled a total of 20,000 products from the three production lines, with an average first-time recognition success rate of 92.7%. Among these, 0.9% of the samples were deemed untraceable, mainly due to localized overexposure causing missing strokes in the characters. To avoid batch traceability risks, process engineers set up a manual re-inspection and re-shooting station. The manual re-inspection takes an average of 23 seconds, requires two employees, and the average number of re-inspections per shift is between 680 and 820 units, which directly affects the stability of the cycle time.

[0093] After introducing the method of this invention, the entire production line does not change its pace or transport rhythm; only the shooting behavior is adjusted. For ease of understanding, the following uses a batch of 1000 units as an example to illustrate how the entire process, from predicting reflections, constructing windows, and stabilizing shooting, operates, and how key data changes.

[0094] First, 0.8 seconds before each compressor arrives at the shooting station, a low-angle light source sweeps across the upper surface of the casing and the area around the terminals at an angle of 5 to 15 degrees. Two industrial cameras capture reflective stripes at 60 frames per second and a resolution of 1920 x 1200. After grayscale thresholding, the reflective streamlines form a banded distribution on the casing's unfolded diagram, commonly concentrated at the lower edge of the terminals, the top arc segment, and below the label. Statistics from 1000 units show that an average of 5 to 9 strong reflective points are identified per unit, with grayscale peak values ​​often exceeding 240. Based on these strong points, a dark window is constructed at the outer edge of the strong points, requiring an average grayscale value less than 100, an area of ​​at least 150 square millimeters, and a stable fluctuation of less than ±5 grayscale levels over three frames. This window guides the camera to shoot from a location that is not glaring, avoiding blindly approaching reflective areas. In simpler terms, it first finds areas so bright they are barely visible, then reverses this process to find a stable, dark, and uniform safety window.

[0095] Secondly, after determining the coordinates of the dark window, the method calculates the safe slant distance and incident angle between the lens and the housing to generate a three-dimensional anti-reflection channel. A spatial ray is established with the window center as the endpoint and the lens center as the starting point, and this ray is expanded into a conical channel: the apex is at the lens, the bottom surrounds the window boundary and extends outward by 5 mm, and the cone angle is 20 to 28 degrees. Light projection simulation is performed every 2 mm along the outer wall of the channel. Any position where the angle of reflection back to the lens is less than 15 degrees and the grayscale is higher than 220 is designated as a danger zone and eliminated. After the channel is eliminated, a hollow safety corridor is formed. The camera approaches the target along the centerline of the channel, maintaining a distance of 135 to 160 mm between the lens and the housing, with an angle fluctuation not exceeding ±5 degrees. In actual measurements of 1000 units, the effective field of view of the channel averaged 180 square centimeters, which was sufficient to cover the target characters and wiring components.

[0096] The third step is to break up the glaring closed loop. Based on the spatial curve of the channel boundary, a beam action point is set every 3 millimeters along the outer edge of the channel, for a total of 10 to 12 points. The light source is lit sequentially from action point 1 to action point N, each time for 0.12 seconds, with an interval of 0.03 seconds, allowing the beam to slide along the boundary. During the sliding process, a break or collapse will appear at the boundary of the high-brightness ring, forming a reflective crack. To open the crack, three low-power pulsed lights are arranged near the break point along the tangential direction of the boundary, with powers of 65%, 50%, and 35% respectively, triggered with delays of 0.05 seconds, 0.10 seconds, and 0.15 seconds, respectively. These are superimposed with small pitch and tilt movements of ±1.2 degrees by the robotic arm, once every 0.15 seconds, for a maximum of 5 times. Finally, a cracked band with an average width of 8 to 11 millimeters and a length of 24 to 32 millimeters is formed on the surface of the casing, the average grayscale value drops below 140, and the bright spot no longer closes back to the lens. A statistical analysis of 1,000 units showed that the success rate of fracture zone formation reached over 99%, with unsuccessful samples concentrated in a very small number of parts with coating defects.

[0097] The fourth step is to clean the window. Within a 10mm radius on either side of the center line of the crack, alternately use the polarizing shutter and the micro-aperture lever. Cycle the polarization angles to 45°, 90°, 135°, and 180°, pausing for 0.08 seconds at each setting with a 0.02-second interval; advance the aperture in three settings: 6.2mm, 5.0mm, and 4.2mm. A comparative experiment with 100 units showed that with 135° polarization and a 4.8mm aperture, the center brightness decreased by 28 gray levels, while the character edge sharpness improved by approximately 19%. Within a 0.8-second exposure cycle, perform a fine-tuning cycle of polarization followed by aperture adjustment every 0.2 seconds, maintaining the window brightness between 145 and 155, and the edge brightness no more than 160, avoiding overexposure or underexposure. For 1000 samples, the window stability rate was 99.3%, and the character readability rate was 99.7%.

[0098] The fifth step is to stabilize the window and follow its movement. Once the window is fixed, the camera and light source synchronously micro-oscillate within a range of ±3 degrees of pitch and ±3 degrees of yaw, performing a circular scan around the casing with a radius of 150 mm. The circumference is divided into 30 stations, each pausing for approximately 0.18 seconds. At each station, a frame is first taken at a reference center angle. If a bright area exceeding 185 grayscale is detected at a corner, the channel centerline is immediately shifted 4 mm in the opposite direction, and the incident angle is lowered by 1.5 degrees. The next station continues with the new parameters. The entire circle is completed within 2.4 seconds, producing approximately 90 frames covering the terminals, labels, clamps, and screws in a multi-view format. Statistics show that the proportion of continuously clear frames increased from 88.3% before importation to 98.6%, the average area of ​​high-reflectivity regions decreased by 41%, and the number of bright spots decreased from 2.6 per frame to 0.7. More importantly, the circular scan marks the optimal angle segment for subsequent alignment with the same model, ensuring the next batch starts in the correct position.

[0099] To make the results more intuitive, A / B comparison data for two weeks are presented here. The comparison was conducted under the same production line, shifts, work groups, and batches of raw materials. The first week used the same three fixed methods, while the second week used the method of this invention, with 1000 units sampled in each week.

[0100] The success rate of primary character recognition increased from 92.7% to 99.4%, an increase of 6.7 percentage points.

[0101] The percentage of cases that cannot be traced due to overexposure and masking decreased from 0.9% to 0.08%.

[0102] Manual re-inspection volume: reduced from about 780 units per shift to less than 60 units per shift, and the number of re-inspection positions reduced from two people to one person doing both, saving about 1.5 person-shifts of manpower.

[0103] Number of retries per shot: decreased from an average of 0.31 to 0.04.

[0104] The average time for single-piece shooting and judgment has been reduced from 1.9 seconds to 1.2 seconds, leaving a 0.7-second buffer for the overall production line rhythm, and the assembly rhythm fluctuation rate has been reduced from ±9% to ±3%.

[0105] Character edge sharpness index (unit normalization): improved from 1.58 to 1.93.

[0106] Average image contrast: improved from 1.52 to 1.86.

[0107] The proportion of bright spots in the image decreased from 5.8% to 1.1%.

[0108] Batch interception incidents triggered by visual non-traceability: decreased from 3 to 0 within a two-week observation period.

[0109] Rework and scrap caused by lack of traceability: rework decreased from 145 pieces per week to 19 pieces per week, and scrap was reduced to zero.

[0110] These changes are not the result of a single parameter fine-tuning, but rather a series of effects: first, finding the dark window; then, building a 3D channel; breaking reflection closure; then, using polarization and aperture to clean the window; and finally, using micro-oscillation and circular scanning to make the window follow the target. Traditional three-fixed methods lock exposure, angle, and light source, relying on luck to avoid reflections; the method of this invention first actively identifies the bright area, then searches for a stable dark area in reverse, and builds a safe channel in space. A moving beam is used to create a slit on the surface of the shell, and residual energy is suppressed by polarization and aperture. Finally, the window is continuously refreshed as its position changes. Common issues on the production line, such as batch differences in spraying, tooling indentations, and label misalignment, no longer pose fatal interferences under this combination of finding, avoiding, breaking, suppressing, and following.

[0111] For ease of reading, the key metrics are summarized below:

[0112] Indicator Name Traditional three fixed values Numerical values ​​of the method of the present invention Increase illustrate Main character recognition success rate 92.7% 99.4% +6.7 percentage points One shot, clear and readable Untraceable proportion 0.90% 0.08% -0.82 percentage points Almost eliminate batch loss Manual re-inspection volume (per shift) Approximately 780 units Less than 60 units −92% Release manpower and rhythm Number of retries per item 0.31 times 0.04 times −87% Basically eliminate the need for reshoots Time consumed for single-item shooting and judgment 1.9 seconds 1.2 seconds -0.7 seconds Allowing for a 6.5-second tempo. Character edge sharpness (normalized) 1.58 1.93 +22% Clearer outline Image contrast 1.52 1.86 +22% Higher level of detail High brightness spot area ratio 5.8% 1.1% −81% Overexposure significantly reduced Continuous clear frame ratio 88.3% 98.6% +10.3 percentage points Stable output of ring scan Batch interception events (two weeks) 3 cases 0 Clear Tracing stability Rework volume (weekly) 145 items 19 items −87% Quality loss decreased Scrapping volume (weekly) 12 items 0 items Clear Avoid the risk of scrapping

[0113] This example clearly demonstrates that the method of this invention transforms the unfavorable situation of reflection as a random disturbance into a controllable process that is observable, avoidable, modifiable, suppressable, and adaptable. Furthermore, it significantly improves the identification success rate and traceability stability without altering the production cycle. The above data comes from a comparison of the same batch on the same production line; the changes are limited to shooting behavior and optical control strategies. Therefore, the conclusions are reproducible in the field and have promotional value.

[0114] This invention establishes a closed-loop control strategy that "adaptively identifies reflections, actively avoids reflections, dynamically eliminates reflections, and continuously suppresses reflections" by introducing a series of coupled operational steps, including dynamic light scanning, reflection trajectory construction, dark window recognition, three-dimensional anti-reflection channel generation, beam slippage breaking reflection closure surface, dynamic polarization suppression, and micro-angle circular scanning linkage. This strategy not only effectively improves the identifiable clarity of character images but also significantly enhances the success rate and stability of main label recognition, thereby ensuring the complete traceability of compressor batch information. Compared to traditional techniques using fixed exposure and fixed shooting angles, this invention can actively sense the optical state of the compressor surface and make dynamic adjustments, fundamentally solving the problem of character occlusion caused by strong metal reflections. While maintaining the original production cycle, it effectively reduces the recognition failure rate and manual re-inspection rate, improving the automation level and quality control capabilities of the entire production line.

[0115] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for on-line visual inspection of appearance defects of a compressor wiring structure, characterized in that, The method comprises the following steps: Before the robot enters the retake area, the compressor shell is scanned by low-angle light, the continuous reflection streamline formed by the metal surface is extracted, the strong reflection points are marked, and the reflection trajectory band is generated as the path reference for image acquisition; According to the reflection strong points extracted in the reflection trajectory band, a plurality of highlight focus kernel points are located, and based on the distribution of the highlight focus kernel points, the non-covered area of the light is constructed in reverse to generate a dark window for shooting; Based on the dark window, the safe slope distance and the incident angle between the camera and the compressor shell are calculated, a three-dimensional anti-reflection channel covering the shooting area is generated, and the camera shooting track is controlled through the anti-reflection channel to avoid reflection interference; According to the space path constructed by the three-dimensional anti-reflection channel, the mechanical arm posture and the light source lighting sequence are adjusted synchronously, the light beam is made to slide along the boundary of the anti-reflection channel, and the original reflection closed surface structure is destroyed, so that a reflection breaking seam suitable for shooting is formed on the surface of the compressor shell; In the reflection breaking seam area, the polarization shutter operation and the micro-shutter disc action are alternately performed to reduce the reflection energy intensity of the highlight area, and a shooting window meeting the image definition requirement is generated; Based on the shooting window, the camera and the light source are controlled to swing synchronously at a small angle, and the ring scanning around the compressor shell is performed to continuously update the anti-reflection channel, so that the shooting stability and the reflection interference suppression effect in the image acquisition process are guaranteed.

2. The compressor wiring structure appearance defect on-line visual inspection method according to claim 1, characterized in that, The reflection trajectory band generation step is as follows: The compressor shell is scanned by low-angle light, the continuous brightness streamline is formed according to the reflection response of the shell surface, and the reflection strong points are extracted; The reflection strong points in the continuous brightness streamline are position-fitted to construct the brightness streamline structure extending along the shell and determine the brightness peak points in the brightness streamline; Based on the combination of the brightness peak points and the brightness streamline, a belt-shaped reflection trajectory band surrounding the wiring area is formed, and the reflection trajectory band is mapped into the spatial coordinate system of the compressor shell; The reflection trajectory band is taken as the camera forbidden shooting interval to guide the camera incident angle adjustment in the retake path planning, so that the camera leading edge line maintains the minimum tangential distance with the reflection trajectory band and avoids the reflection trajectory band to complete the shooting angle update in the retake process.

3. The compressor wiring structure appearance defect on-line visual inspection method according to claim 2, characterized in that, The dark window generation step is as follows: An observation unit is established in the reflection trajectory band with each reflection strong point as the center, and the bright point cluster with the gray value higher than the set threshold is extracted, and the area with the strongest brightness stability is determined as the highlight focus kernel point; The gray scale distribution is measured radially with the highlight focus kernel point as the center, the highlight diffusion ring is fitted, and the outer ring transition zone is constructed; From the outer edge of the transition zone, the non-covered area of the light whose gray scale is lower than the set value and whose fluctuation range does not exceed the limited range is identified, which is defined as the dark area; All dark areas are combined to form a closed dark window, and the spatial parameters of the dark window are extracted as the basis for camera incident angle planning.

4. The compressor wiring structure appearance defect on-line visual inspection method according to claim 3, characterized by, The three-dimensional anti-reflection channel generation step is as follows: A space ray model from the camera to the target point is established with the center point of the dark window as the target point, and the included angle range between the space ray and the normal line of the dark window and the safe slope distance are limited; A shooting conical area is constructed with the space ray as the central axis, reflection high-risk points are identified and removed through high-density light simulation to form a hollow non-interference shooting channel. A three-dimensional curve shooting path connecting the center points of the dark surface window is generated inside the anti-reflection channel, and a posture adjustment point is set to keep the optical axis stable; The anti-reflection channel and the shooting path parameters are used to control the camera operation, ensuring that the camera motion trajectory is always located inside the anti-reflection channel to complete image acquisition.

5. The compressor wiring structure appearance defect on-line visual inspection method according to claim 4, characterized by, The space ray model of the anti-reflection channel limits the fixed angle range between the camera optical axis and the normal of the dark surface window, sets a buffer contour for the bottom surface of the shooting conical region relative to the dark surface window boundary, and sets continuous posture adjustment points inside the channel boundary. The camera keeps the optical axis direction stable and moves smoothly along the channel center line after each posture adjustment, ensuring that the safety distance between the shooting trajectory and the channel boundary is constant.

6. The compressor wiring structure appearance defect on-line visual inspection method according to claim 5, characterized by, The reflection breaking crack generation steps are as follows: Set continuous beam action points along the outer edge of the center trajectory curve on the three-dimensional anti-reflection channel space boundary, and light up the light source in turn to make the light beam slide along the channel boundary; Collect images of the shell reflection area during the light beam sliding process and identify the reflection breaking points, arrange pulsed light sources around the reflection breaking points in the tangent direction to form disturbance light displacement waves to generate a breaking induction area; Expand the breaking range and stabilize the breaking band shape through micro swing operation of the mechanical arm posture; Integrate the formed reflection breaking crack space information to the shooting path endpoint to determine the final shootable window for subsequent image acquisition.

7. The compressor wiring structure appearance defect on-line visual inspection method according to claim 6, characterized by, The shooting window generation steps are as follows: Perform local lighting evaluation in the reflection breaking crack area, determine the residual reflection energy distribution, and set the initial polarization shutter polarization angle; Periodically adjust the polarization shutter angle and collect images to determine the best polarization angle state; After determining the best polarization angle, adjust the opening diameter of the micro diaphragm dial to obtain light flux control with balanced brightness; Alternately perform dynamic linkage operation of the polarization shutter and the diaphragm dial during the exposure period to stabilize the light incident energy; Collect images and confirm that the reflection breaking crack area forms an effective shooting window for subsequent imaging.

8. The compressor wiring structure appearance defect on-line visual inspection method according to claim 7, characterized by, Based on the shooting window, control the camera and light source to swing synchronously at a small angle, and perform ring scanning around the compressor shell. The continuous update anti-reflection channel steps are as follows: Establish a micro swing control space at the center point of the clear shooting window and perform camera posture angle switching to determine the best micro swing center angle; Synchronously adjust the light source exit angle according to the micro swing center angle, so that the light beam slides along the anti-reflection channel boundary; Build a circumferential trajectory around the compressor shell and perform ring shooting action, collecting multi-angle images at each station; Based on the gray scale distribution of the collected images, calculate the reflection fluctuation points in real time and adjust the anti-reflection channel space parameters; After completing the full ring scan, select the image frame with the best image quality and the channel parameters as the main target area effective imaging channel.

Images