Automated visual inspection method and system for appearance defects of automotive interior parts

Abstract

The application discloses a kind of automotive interior part appearance defect automated visual inspection method and system, belong to automotive appearance detection technical field.The application debugs light source combination, exposure parameter and brightness ratio and is saved as template, defines left and right side region of interest.Control conveying line sends interior part to detection area and adopts fixed and follow-up combined transmission mode, calls light source template and shoots image, detects defect after pre-processing using algorithm, and the results are summarized to judge eligibility.Unqualified piece is moved to NG product unloading area, and the angle of gripper is adjusted to reinspection line;read the identification associated automotive parts, extract defect characteristics, and use the model to determine whether it is a regular defect based on vehicle information.For irregular defects, use the defect as a node, build a directed edge based on material and defect characteristics, select the earliest defect as the initial node, traverse to form the impact path chain, distinguish between root cause and derivative defects, and generate a solution.

Description

Technical Field

[0001] This invention relates to the field of automotive exterior inspection technology, specifically to an automated visual inspection method and system for automotive interior parts exterior defects. Background Technology

[0002] The appearance quality of automotive interior parts directly impacts a vehicle's perceived quality and market competitiveness. Defects such as black spots, scratches, and burrs on their surfaces reduce product value and can even affect the user experience. With the rapid development of the automotive industry, the demands for efficiency, accuracy, and automation in the inspection of interior parts appearance defects are increasing. Traditional inspection methods are no longer sufficient to meet the needs of rapid defect identification, classification, and traceability in large-scale production, necessitating a highly efficient and intelligent automated vision inspection solution.

[0003] Existing technologies mostly employ either purely fixed or purely motion-following conveyors. Purely fixed conveyors require frequent start-stop cycles, resulting in low inspection efficiency and difficulty adapting to continuous production scenarios. Purely motion-following conveyors, lacking a stable positioning reference, are prone to inspection deviations due to slight workpiece movement. The sorting and defect analysis of non-conforming parts are fragmented, only allowing for simple pass / fail judgments, failing to trace the root cause and derivative relationships of defects, making it difficult to develop targeted improvement solutions, leading to the recurrence of similar defects. Light source parameters are mostly generic settings, without customized templates for different interior material characteristics, easily causing problems such as reflections and shadows, affecting defect imaging quality and increasing the misjudgment rate of subsequent inspection algorithms. Summary of the Invention

[0004] The purpose of this invention is to provide an automated visual inspection method and system for automotive interior parts to address the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, this application provides an automated visual inspection method for exterior defects of automotive interior parts, comprising the following steps: Adjust the light source combination, set the exposure parameters and brightness ratio, and save it as a light source template; define the left and right sides of the interior parts' regions of interest; The control conveyor line drives the interior trim parts to the defect detection area, using a combination of fixed and moving conveyor methods; the light source template is called to capture images of the interior trim parts in the regions of interest on the left and right sides; the interior trim part images are preprocessed, and the defect detection algorithm is called to detect different types of defects; the defect information detected on the left and right sides is summarized to obtain the defect detection results; Based on the defect detection results, determine whether the interior parts are qualified; when the interior parts are unqualified, transfer them to the NG product unloading area, adjust the gripper angle according to the defect location, and clamp them to the NG re-inspection line; read the markings on the surface of the interior parts and establish a unique association with the corresponding automotive parts; extract defect features from the defect detection results, call the pre-trained defect routine judgment model, and output whether the defect is a routine defect; For interior parts with unconventional defects, each defect is used as a node, and directed edges are established between nodes based on the material properties of the interior parts and the physical properties of the defects. The earliest defect node is selected in the graph structure to form an initial set of defect locations. Starting from the initial defect node, subsequent nodes can be reached by traversing along the directed edges of the graph structure to form an influence path chain, distinguishing between root defects and derived defects and generating solutions.

[0006] In conjunction with the first aspect, in the first embodiment of the first aspect of this application, the adjustment of the light source combination, setting exposure parameters and brightness ratio, and saving it as a light source template, includes: Automotive interior parts are classified according to their material properties. For a specific type of interior part, typical samples containing different defects, sizes, and shapes are selected and fixed in the calibration fixture at the inspection station to ensure that the sample position is consistent with the positioning reference during actual inspection. Suitable light source types and combinations are selected, and the light source angle, brightness output ratio, and camera exposure parameters of each light source are adjusted. Iterative optimization is carried out according to the defect imaging quality assessment standard. When the image of a certain type of defect is blurred, the light source intensity in the corresponding direction is increased. When there is reflection interference, the light source angle is adjusted or the exposure time is reduced. When all target defects of the typical sample can be identified in the image without interference, a candidate light source template for the material is formed; when different typical samples of the same material are replaced, the candidate light source template is called for imaging verification. When the stability of the defects meets the detection requirements, it is confirmed to be valid and saved as a light source template.

[0007] In conjunction with the first aspect, in a second embodiment of the first aspect of this application, the definition of the left and right sides of the interior trim component's region of interest includes: Based on the structural characteristics of automotive interior parts and the locations where defects are prone to occur, and in conjunction with product design drawings, the areas that need to be focused on for inspection are determined, while background or non-critical areas that do not need to be inspected are excluded. At the inspection station, the placement reference of the interior parts is determined by positioning fixtures. Based on the placement reference, the boundary of the area to be inspected and the corresponding specific physical location range are marked, forming a closed area outline in the image. By capturing images of interior parts, it is determined whether the area contained by the camera coordinate boundary completely covers the target detection area. When there is a deviation, the boundary marker is adjusted, and the process is repeated and verified until the region contour matches the detection requirements. The left and right regions of interest are saved respectively.

[0008] In conjunction with the first aspect, in the third embodiment of the first aspect of this application, the control conveyor line drives the interior trim parts to the defect detection area, employing a conveying method combining fixed and follow-up mechanisms, including: Start the conveyor line and place the interior trim pieces into the positioning fixtures on the conveyor line, ensuring that the interior trim pieces are aligned with the positioning reference of the fixtures to achieve initial fixation. The conveyor line drives the fixtures and interior trim pieces to move along a preset path, using a combination of fixed and moving conveying methods: when the movement reaches the preset fixed detection section, the conveyor line stops operating, and the interior trim pieces are in a fixed state. At this time, the fixtures maintain a stable position through a mechanical locking structure. When the movement is in a non-fixed detection section, the interior trim pieces move synchronously with the conveyor line and are in a moving state. The position detection sensor located at the entrance of the inspection area monitors the position of the interior parts in real time. When the interior parts are detected to have reached the preset position at the boundary of the inspection area, the corresponding positioning mechanism is triggered depending on whether the interior parts are fixed or moving: when they are fixed, it is confirmed that the interior parts have stopped at the fixed reference position in the inspection area; when they are moving, the position data of the interior parts are collected in real time by the vision sensor, and the conveyor line adjusts its running speed to bring the interior parts into the effective inspection range of the inspection area.

[0009] In conjunction with the first aspect, in the fourth embodiment of the first aspect of this application, the step of calling the light source template to capture images of the interior trim parts in the left and right sides of the region of interest includes: The system identifies the material type of the current interior trim component, retrieves the corresponding light source template based on the material type, sends a command to the light source controller, and activates the light source according to the light source template for preheating. After the light source stabilizes, a positioning command is sent to the inspection robot. When the interior trim component is in a fixed state, the first robot, carrying a camera, moves to the preset fixed inspection position on the right, and the second robot, carrying a camera, moves to the preset fixed inspection position on the left. When the interior trim component is in a moving state, the robot monitors the position of the interior trim component in real time through a vision sensor, dynamically adjusting the X-axis and Y-axis displacement and rotation angle to ensure that the inspection posture matches the position of the interior trim component. The robot arms are placed with their working ranges staggered. The system ensures that the camera lens axis is aligned with the center of the corresponding region of interest on the left and right sides, and captures images of the interior trim component in the region of interest on both sides.

[0010] In conjunction with the first aspect, in the fifth embodiment of the first aspect of this application, the preprocessing of the interior part image and the invocation of a defect detection algorithm to detect different types of defects include: The interior component images are preprocessed to unify their size and color space. The images are then cropped to retain the effective portion containing the region of interest and to remove irrelevant background. The cropped images are scaled according to the detection accuracy requirements. An adaptive filtering algorithm is used to eliminate high-frequency noise interference in the images to address random noise. The brightness differences in the images caused by uneven lighting are corrected by adjusting the brightness in different areas. Color features are extracted from interior parts images. The color parameters of pixels are compared with the color reference range of qualified areas to filter out abnormal pixels whose colors exceed the reference range. Connectivity analysis is performed on abnormal pixels to determine whether they form independent point-like objects. Size features are combined to determine whether they are black dots or discolored dots. Morphological operations are used to enhance the contours of small foreign objects in the image, highlighting the linear features of fuzz and the blocky features of particles. Edge extraction is performed, and feature analysis is used to distinguish between fuzz and particles. Length, width, and distribution density are combined to filter out non-defect interference. Based on the edge detection algorithm, the edge contour of the interior parts surface is extracted with reference to the standard edge data of NG materials. The deviation between the edge contour and the standard contour is calculated to identify protrusions that exceed the preset tolerance range, and burrs and flash are identified. Texture features are extracted from the image, and the difference operation is performed with the texture model of qualified areas to filter out texture abnormal areas. When there is a continuous linear gray-scale change band, it is judged as a scratch. When there is a uniform texture abnormality, it is judged as sandpaper mark.

[0011] In conjunction with the first aspect, in the sixth embodiment of the first aspect of this application, the step of extracting defect features from the defect detection results, calling a pre-trained defect routine judgment model, and outputting whether the defect is a routine defect includes: Defect features are extracted from defect detection results, including defect type, specific location within the vehicle, quantity, size, and morphological features. These extracted features are then structured. Corresponding component information is retrieved, including component model, production batch, production standard, usage environment requirements, past quality inspection records, and common defect distribution patterns for components of the same model. This information is then organized into structured data. The defect features are combined with the component information to form a comprehensive dataset for model input. The routine defect judgment model is trained based on defect data from components of the same model. During training, the input data is scaled to a uniform size, and the component surface is divided into four faces: A, B, C, and D. Different defect thresholds are set for different defect types, with the thresholds decreasing sequentially from surface A to surface D, reflecting increasingly stringent defect control. A convolutional neural network algorithm is used for training to learn the common defect feature patterns corresponding to different parts. The fused part information and defect feature data are input into the model in the format required by the model. The model calculates the matching degree by comparing the current defect feature with the common defect feature library of parts of the same model. Among them, defects on surfaces A and B are common defects, while defects on surfaces C and D are non-common defects. Defects on surface A do not require processing, while defects on surface B require rework. The model outputs the result based on the matching degree threshold to determine whether the current defect is a common defect.

[0012] In conjunction with the first aspect, in the seventh embodiment of the first aspect of this application, the step of establishing directed edges between unconventional interior parts, using each defect as a node, based on the material properties of the interior parts and the physical properties of the defects, includes: Identify the type, location, and corresponding interior trim material of unconventional defects. Treat each defect as a node in the graph, and associate its location with the material information and physical properties of the defect itself in the node attributes. Analyze the material properties of the interior trim and the physical properties of the defects at the locations corresponding to each node to determine the interaction between different materials and defect properties. Perform pairwise comparisons on all defect nodes and determine the influence relationship based on the material properties and physical properties of the defects. When the characteristics of defect P, combined with the material properties of its location, directly or indirectly cause the generation or deterioration of defect Q, it is determined that P has an influence on Q. For defect nodes that have an impact relationship, a directed edge is established from P to Q, and the edge attributes are labeled according to the impact, forming a graph structure that includes all non-standard defect nodes and the impact relationships between nodes.

[0013] In conjunction with the first aspect, in the eighth embodiment of the first aspect of this application, the step of traversing along the directed edges of the graph structure from the initial defect node to reach subsequent nodes to form an influence path chain, distinguishing between root defects and derived defects, and generating a solution includes: Starting from the initial defect node in the graph structure, all subsequent defect nodes that can be reached directly or indirectly are traced according to the direction of the directed edges between nodes; the connection relationship of each node is recorded, and the nodes are sorted out in the order of initial node, directly affected nodes and indirectly affected nodes to form an influence path chain, ensuring that the path chain covers all related defects caused by the initial defect; The starting point of the impact path chain is identified as the root defect. All nodes in the path chain other than the root defect are considered derivative defects. These are further subdivided according to their impact level: nodes directly affected by the root defect are first-level derivative defects, nodes affected by first-level derivative defects are second-level derivative defects, and so on, thus determining the hierarchical relationship between each derivative defect and the root defect. For the root defect, a priority treatment plan is developed based on the material characteristics and defect type of the interior parts at its corresponding location to block the impact path at its source. For derivative defects, a treatment order is determined based on their hierarchical relationship with the root defect, prioritizing first-level derivative defects, followed by second-level and higher derivative defects. When a derivative defect is a direct result of the root defect, its disappearance is checked after treating the root defect. When separate treatment is required, an appropriate repair method is selected based on material characteristics. All treatment steps are integrated to generate a solution.

[0014] Secondly, this application provides an automated visual inspection system for exterior defects of automotive interior parts, comprising: Region of Interest (ROI) Definition Module: Includes a light source template generation unit and an ROI definition unit; wherein, the light source template generation unit adjusts the light source combination, sets the exposure parameters and brightness ratio, and saves it as a light source template; the ROI definition unit defines the left and right ROIs of the interior parts; The defect detection module includes an image acquisition unit, a defect detection unit, and a defect information aggregation unit. The image acquisition unit controls the conveyor line to move interior trim parts to the defect detection area, using a combination of fixed and moving conveyor methods. It calls a light source template to capture images of the interior trim parts in the left and right regions of interest. The defect detection unit preprocesses the interior trim part images and uses a defect detection algorithm to detect different types of defects. The defect information aggregation unit aggregates the defect information detected on the left and right sides to obtain the defect detection results. The routine defect judgment module includes: an interior part acceptance judgment unit, an NG re-inspection clamping unit, an association establishment unit, and a routine defect judgment unit. The interior part acceptance judgment unit determines whether an interior part is acceptable based on defect detection results. When an interior part is unacceptable, the NG re-inspection clamping unit moves it to the NG product unloading area, adjusts the clamping angle according to the defect location, and clamps it to the NG re-inspection line. The association establishment unit reads the markings on the surface of the interior part and establishes a unique association with the corresponding automotive parts. The routine defect judgment unit extracts defect features from the defect detection results, calls a pre-trained routine defect judgment model, and outputs whether the defect is a routine defect. The solution generation module includes a directed edge establishment unit, an initial defect location screening unit, and a solution generation unit. The directed edge establishment unit, for interior parts with unconventional defects, establishes directed edges between nodes based on the material properties of the interior parts and the physical properties of the defects, using each defect as a node. The initial defect location screening unit selects the earliest defect node in the graph structure to form an initial defect location set. The solution generation unit starts from the initial defect node, traverses along the directed edges of the graph structure to reach subsequent nodes, forming an influence path chain, distinguishing between root defects and derived defects, and generating a solution.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention adopts a combination of fixed and follow-up transmission methods. When stable detection is required, it switches to fixed mode to ensure positioning accuracy; when continuous flow is required, it adopts follow-up mode to avoid frequent start and stop, greatly improves overall detection efficiency, and adapts to the needs of mass production.

[0016] 2. This invention uses unique identifiers to associate component information, combines a defect routine judgment model to distinguish between routine and non-routine defects, and constructs a graph structure analysis of non-routine defects to analyze the root causes and derivative relationships, generating targeted solutions to help optimize production processes and reduce the recurrence of similar defects.

[0017] 3. This invention customizes light source templates for different materials and ensures clear defect imaging through parameter iteration, providing high-quality image input for subsequent defect detection algorithms, improving defect recognition accuracy and reducing false positive rate. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the steps of an automated visual inspection method for exterior defects of automotive interior parts according to the present invention; Figure 2 This is a system structure diagram of an automated visual inspection system for exterior defects of automotive interior parts according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Figures 1-2 As shown, the present invention provides a technical solution: like Figure 1 As shown, this application provides an automated visual inspection method for exterior defects of automotive interior parts, including the following steps: Step S100: Debug the light source combination, set the exposure parameters and brightness ratio, and save it as a light source template; define the left and right sides of the interior parts' regions of interest; Specifically, automotive interior parts are classified according to their material properties. For a particular type of interior part, typical samples containing different defects, sizes, and shapes are selected and fixed in the calibration fixture at the inspection station to ensure that the sample position is consistent with the positioning reference during actual inspection. A suitable combination of light sources is selected, and the light source angle, brightness output ratio, and camera exposure parameters of each light source are adjusted. Iterative optimization is carried out based on the defect imaging quality assessment standard. When the image of a certain type of defect is blurred, the light source intensity in the corresponding direction is increased. When there is reflection interference, the light source angle is adjusted or the exposure time is reduced. When all target defects of the typical sample can be identified in the image without interference, a candidate light source template for the material is formed; when different typical samples of the same material are replaced, the candidate light source template is called for imaging verification. When the stability of the defects meets the detection requirements, it is confirmed to be valid and saved as a light source template.

[0021] Furthermore, based on the structural characteristics of automotive interior parts and the locations where defects are prone to occur, and in conjunction with product design drawings, areas that require key inspection are identified, while background or non-critical areas that do not require inspection are excluded. At the inspection station, the placement reference of the interior parts is determined by positioning fixtures. Based on the placement reference, the boundaries of the area to be inspected and the corresponding specific physical location range are marked, forming a closed area outline in the image. By capturing images of interior parts, it is determined whether the area contained by the camera coordinate boundary completely covers the target detection area. When there is a deviation, the boundary marker is adjusted, and the process is repeated and verified until the region contour matches the detection requirements. The left and right regions of interest are saved respectively.

[0022] In one specific embodiment, automotive interior parts are categorized into three types based on their material: leather, plastic, and fabric. Taking a black piano lacquer plastic interior part as an example, a typical sample containing 0.3mm pinholes, 6mm scratches, and 0.5mm discolored spots is selected. The sample is fixed using a positioning fixture, and the alignment error between the sample edge and the fixture's baseline is controlled within ±0.2mm, consistent with the actual testing positioning baseline.

[0023] A combination of 30° ring light and 45° strip light was used, with initial parameters: ring light brightness 500 lux, strip light brightness 300 lux, and camera exposure 18ms. At this time, the pinhole image was blurred (grayscale difference 35%), and the scratch showed a 1.5mm overexposed area due to reflection.

[0024] Iterative optimization: The brightness of the strip light at the pinhole was increased to 400 lux, and the grayscale difference was improved to 70%; the ring light angle was adjusted to 270°, the exposure time was shortened to 12ms, and the reflective area was reduced from 25% to 4%. Once all defects were clearly identifiable, candidate templates were formed. Three samples of the same material were used for verification; 25 out of 26 defects showed stable imaging (clarity ≥90%), confirming their effectiveness and saving the template.

[0025] Taking the interior components of the car's center console as an example, based on the design drawings, the left-side button panel (which is prone to dust accumulation and discoloration) and the right-side display screen bezel (which is prone to scratches) are identified as the key inspection areas.

[0026] The center console is fixed by positioning fixtures. The boundary of the left area is marked with the midpoint of the lower edge of the panel as the reference origin: physical range X-axis 110-210mm, Y-axis 280-360mm; right area X-axis 320-420mm, Y-axis 280-360mm, forming a closed rectangular outline in the image.

[0027] Image verification: The left area completely covers the button panel (without any missing edges), and the right area precisely frames the display screen border. Due to a 0.8mm deviation in the upper limit of the X-axis of the right area, the boundary was adjusted to 320-419.2mm on the X-axis. After re-verification, the area outline perfectly matches the detection area, and the parameters are saved.

[0028] Step S200: Control the conveyor line to drive the interior trim parts to the defect detection area, using a combination of fixed and moving conveyor methods; call the light source template to capture images of the interior trim parts in the left and right regions of interest; preprocess the interior trim part images, call the defect detection algorithm to detect different types of defects; summarize the defect information detected on the left and right sides to obtain the defect detection results; Specifically, the conveyor line is started, and the interior trim is placed in the positioning fixture of the conveyor line to ensure that the interior trim is aligned with the positioning reference of the fixture, thus achieving initial fixation. The conveyor line drives the fixture and interior trim to move along a preset path, using a combination of fixed and follow-up conveying methods: when it moves to the preset fixed detection section, the conveyor line stops, and the interior trim is in a fixed state. At this time, the fixture maintains its position through a mechanical locking structure. When moving through a non-fixed detection section, the interior trim moves synchronously with the conveyor line and is in a follow-up state. The position detection sensor located at the entrance of the inspection area monitors the position of the interior parts in real time. When the interior parts are detected to have reached the preset position at the boundary of the inspection area, the corresponding positioning mechanism is triggered depending on whether the interior parts are fixed or moving: when they are fixed, it is confirmed that the interior parts have stopped at the fixed reference position in the inspection area; when they are moving, the position data of the interior parts are collected in real time by the vision sensor, and the conveyor line adjusts its running speed to bring the interior parts into the effective inspection range of the inspection area.

[0029] Furthermore, the material type of the current interior component is identified, and the corresponding light source template is retrieved based on the material type. An instruction is sent to the light source controller to activate the light source template for preheating. After the light source stabilizes, a positioning instruction is sent to the inspection robot. When the interior component is in a fixed state, the first robot, carrying a camera, moves to a preset fixed inspection position on the right, and the second robot, carrying a camera, moves to a preset fixed inspection position on the left. When the interior component is in a moving state, the robot monitors the position of the interior component in real time using a vision sensor, dynamically adjusting the X-axis and Y-axis displacement and rotation angle to ensure that the inspection posture matches the position of the interior component. The robot arms are placed with their working ranges staggered. The camera lens axis is aligned with the center of the corresponding region of interest on each side to capture images of the interior component in the left and right regions of interest.

[0030] Furthermore, the interior component images are preprocessed to unify their size and color space, crop the images to retain the effective portion containing the region of interest, remove irrelevant background, and scale the cropped images according to the detection accuracy requirements. For random noise in the images, an adaptive filtering algorithm is used to eliminate high-frequency noise interference. The brightness difference in the images caused by uneven lighting is corrected by adjusting the brightness in different areas. Color features are extracted from interior parts images. The color parameters of pixels are compared with the color reference range of qualified areas to filter out abnormal pixels whose colors exceed the reference range. Connectivity analysis is performed on abnormal pixels to determine whether they form independent point-like objects. Size features are combined to determine whether they are black dots or discolored dots. Morphological operations are used to enhance the contours of small foreign objects in the image, highlighting the linear features of fuzz and the blocky features of particles. Edge extraction is performed, and feature analysis is used to distinguish between fuzz and particles. Length, width, and distribution density are combined to filter out non-defect interference. Based on the edge detection algorithm, the edge contour of the interior parts surface is extracted with reference to the standard edge data of NG materials. The deviation between the edge contour and the standard contour is calculated to identify protrusions that exceed the preset tolerance range, and burrs and flash are identified. Texture features are extracted from the image, and the difference operation is performed with the texture model of qualified areas to filter out texture abnormal areas. When there is a continuous linear gray-scale change band, it is judged as a scratch. When there is a uniform texture abnormality, it is judged as sandpaper mark.

[0031] In one specific embodiment, a black piano-lacquered plastic center console interior trim is placed into a positioning fixture on the conveyor line. Through the engagement of a slot and a positioning pin, the alignment error between the edge of the trim trim and the fixture reference is ensured to be ≤0.3mm. The conveyor line drives the fixture to rotate at a speed of 150mm / s, employing a combination of fixed and moving mechanisms: when rotating to the fixed detection section (1.2m in length), the conveyor line stops within 0.8s, the mechanical locking structure of the fixture is activated, and position fluctuations are controlled within ±0.1mm; when rotating outside the detection section, the trim trim moves synchronously with the conveyor line, and the position deviation in the moving state is ≤0.5mm / s.

[0032] When the laser sensor (accuracy ±0.05mm) at the entrance of the inspection area detects that the interior trim part has reached the preset position at the boundary (500mm from the inspection center), it is confirmed to stop at the reference position (X=1000mm, Y=500mm) in the fixed state; in the follow-up state, the vision sensor (sampling frequency 60Hz) collects position data in real time, and the conveyor line finely adjusts the speed to 148mm / s so that the interior trim part enters the effective range of the inspection area (X=1200-1800mm, Y=450-550mm) after 5s.

[0033] The interior trim is identified as black piano lacquer plastic. The corresponding light source templates are selected: 30° ring light (270° range, 500 lux) and 45° strip light (400 lux). After the light source is turned on, it is preheated for 2.5 seconds and the brightness is stabilized within ±1.5%.

[0034] In the stationary state, the first robotic arm carries the camera to a preset position on the right (X=1600mm, Y=500mm, 350mm from the interior trim), while the second robotic arm moves to a preset position on the left (X=1400mm, Y=500mm). In the servo state, the vision sensor detects an X-axis offset of +0.6mm in the interior trim, and the robotic arm dynamically adjusts the X-axis displacement by +0.6mm and the rotation angle by +0.5° to ensure that the lens axis is aligned with the center of the region of interest (deviation ≤0.2mm). Both cameras capture images simultaneously at a resolution of 2048×1536 pixels, with no motion blur.

[0035] Image preprocessing: cropping to retain the left button panel (X=420-860 pixels, Y=650-920 pixels) and the right display screen border (X=1100-1540 pixels, Y=650-920 pixels), removing 65% of irrelevant background; scaling to 1280×960 pixels according to detection accuracy requirements.

[0036] After applying the adaptive filtering algorithm, the number of random noise points decreased from 22 per frame to 3 per frame, while the details of pinhole edges were fully preserved. Regional adjustments were made to address the brightness differences caused by uneven lighting (regional brightness standard deviation 38), reducing the standard deviation to 9 and improving the contrast between out-of-color points and the background by 25%.

[0037] In color feature extraction, the acceptable color reference range is grayscale value 180-220. 237 abnormal pixels were screened out. Connectivity analysis identified two different color points (size 0.6mm×0.5mm and 0.4mm×0.3mm), both of which exceeded the 0.3mm upper limit.

[0038] After morphological operations (erosion + dilation) to enhance the contour, one hair (4.2 mm in length, aspect ratio 8:1) and one particle (0.35 mm in diameter, aspect ratio 1.1:1) were extracted from the edge, and two non-defect interferences <0.2 mm were filtered out.

[0039] Edge detection, using NG material standard edge data as a reference, calculated a protrusion on the right side border (deviation 0.15mm), exceeding the preset tolerance (≤0.1mm), and was judged as a burr. Texture difference calculation revealed a scratch on the left panel (length 6.8mm, continuous grayscale variation band), with no sandpaper marks. Summary of detection results: 2 discolored spots, 1 lint, 1 particle, 1 burr, and 1 scratch.

[0040] Step S300: Based on the defect detection results, determine whether the interior parts are qualified; when the interior parts are unqualified, transfer them to the NG product unloading area, adjust the gripper angle according to the defect location, and clamp them to the NG re-inspection line; read the markings on the surface of the interior parts and establish a unique association with the corresponding automotive parts; extract defect features from the defect detection results, call the pre-trained defect routine judgment model, and output whether the defect is a routine defect; Specifically, defect features are extracted from defect detection results, including defect type, specific location within the vehicle, quantity, size, and morphological features. These extracted features are then structured. Corresponding component information is retrieved, including component model, production batch, production standard, usage environment requirements, past quality inspection records, and common defect distribution patterns for components of the same model. This information is then organized into structured data. The defect features are combined with the component information to form a comprehensive dataset for model input. The routine defect judgment model is trained based on defect data from components of the same model of automobile. During training, the input data is scaled to a uniform size, and the component surface is divided into four faces: A, B, C, and D. Each surface has a corresponding defect threshold set for different defect types, with the threshold decreasing sequentially from surface A to surface D, reflecting increasingly stringent control over defects. A convolutional neural network algorithm is used for training to learn the common defect feature patterns corresponding to different components. The fused component information and defect feature data are input into the model in the required format. The model calculates the matching degree by comparing the current defect feature with a common defect feature library of components of the same model. Defects on surfaces A and B are considered common defects, while defects on surfaces C and D are considered uncommon defects. Defects on surface A do not require processing, while defects on surface B require rework. The model outputs the result based on the matching degree threshold to determine whether the current defect is a common defect.

[0041] In one specific embodiment, based on the defect detection results of the black piano lacquer plastic center console (2 discoloration spots, 1 fuzzy thread, 1 particle, 1 burr, and 1 scratch), the product qualification standard for this model was applied: maximum size of discoloration spot ≤ 0.3mm, fuzzy thread length ≤ 3mm, particle diameter ≤ 0.3mm, burr deviation ≤ 0.1mm, and scratch length ≤ 5mm. Since all defects exceeded the standard limits, the product was deemed unqualified.

[0042] The control system commands the gripper device to activate, adjusting the angle according to the location of each defect: for burrs on the right display screen bezel (coordinates X=1550mm, Y=520mm), the gripper rotates 10° to avoid the protrusion; for scratches on the left panel (X=1380mm, Y=490mm), the gripper is raised 5mm to avoid contact damage. The gripping force is set to 10N, and the transfer process takes 8 seconds. The interior trim is then placed smoothly on the NG re-inspection line with a positioning deviation ≤0.8mm.

[0043] The unique laser-engraved mark on the interior parts surface is "KT20240518-369". After being captured by an industrial camera and recognized by OCR (recognition time 0.5s, accuracy 100%), it was matched with the parts database and associated with the following information: model "KT-ZK-012", production batch "202405", production standard "Q / KT018-2024", and operating environment requirement "-30℃~70℃". Quality records of the same batch of products over the past three months show that common defects are "≤0.2mm discoloration spots" and "≤2mm burrs", with no records of burrs or scratches.

[0044] Structured defect features were extracted from the test results: discolored dots (2 locations, left button panel, 0.6mm×0.5mm / 0.4mm×0.3mm, oval), fuzz (1 piece, right display bezel, 4.2mm, linear), particle (1 piece, left button panel, 0.35mm, round), burr (1 location, right display bezel, deviation 0.15mm, raised), and scratch (1 scratch, left button panel, 6.8mm, linear).

[0045] Information on parts of the same model was retrieved to form a comprehensive dataset, which was then input into a routine defect judgment model (based on CNN training, with input data scaled to 512×512 pixels). This model divides the center console surface into surface A (top decorative area), surface B (central functional area), surface C (button / display area), and surface D (edge ​​seam area), with different color point thresholds of 0.5mm, 0.4mm, 0.3mm, and 0.2mm respectively; and scratch thresholds of 8mm, 6mm, 5mm, and 3mm respectively.

[0046] All current defects are located on the C-side (button / display area). The sizes of discolored dots, fuzz, particles, burrs, and scratches all exceed the C-side threshold. Model comparison yielded a matching rate of 32% (below the 60% threshold for conventional defects), resulting in the conclusion that all defects are non-standard. Burrs and scratches, in particular, have a matching rate of only 15% because there are no similar records in this batch; these should be prioritized for investigation into production anomalies.

[0047] Step S400: For interior parts with non-standard defects, each defect is used as a node, and directed edges are established between nodes based on the material properties of the interior parts and the physical properties of the defects; the earliest defect node is selected in the graph structure to form an initial set of defect locations; starting from the initial defect node, subsequent nodes are reached by traversing along the directed edges of the graph structure to form an influence path chain, distinguishing between root defects and derived defects and generating solutions.

[0048] Specifically, the types, locations, and corresponding interior trim materials of unconventional defects are determined. Each defect is treated as a node in the graph, and its attributes are associated with the material information of its location and the physical characteristics of the defect itself. The material characteristics of the interior trim and the physical characteristics of the defects at the corresponding locations of each node are analyzed to determine the interaction between different materials and defect characteristics. All defect nodes are compared pairwise, and the influence relationship is determined based on the material characteristics and defect physical characteristics. When the characteristics of defect P, combined with the material characteristics of its location, directly or indirectly cause the generation or deterioration of defect Q, it is determined that P has an influence on Q. For defect nodes that have an impact relationship, a directed edge is established from P to Q, and the edge attributes are labeled according to the impact, forming a graph structure that includes all non-standard defect nodes and the impact relationships between nodes.

[0049] Furthermore, starting from the initial defect node in the graph structure, all subsequent defect nodes that can be reached directly or indirectly are traced in sequence according to the direction of the directed edges between nodes; the connection relationship of each node is recorded, and the nodes are sorted out in the order of initial node, directly affected nodes and indirectly affected nodes to form an influence path chain, ensuring that the path chain covers all related defects caused by the initial defect; The starting point of the impact path chain is identified as the root defect. All nodes in the path chain other than the root defect are considered derivative defects. These are further subdivided according to their impact level: nodes directly affected by the root defect are first-level derivative defects, nodes affected by first-level derivative defects are second-level derivative defects, and so on, thus determining the hierarchical relationship between each derivative defect and the root defect. For the root defect, a priority treatment plan is developed based on the material characteristics and defect type of the interior parts at its corresponding location to block the impact path at its source. For derivative defects, a treatment order is determined based on their hierarchical relationship with the root defect, prioritizing first-level derivative defects, followed by second-level and higher derivative defects. When a derivative defect is a direct result of the root defect, its disappearance is checked after treating the root defect. When separate treatment is required, an appropriate repair method is selected based on material characteristics. All treatment steps are integrated to generate a solution.

[0050] In one specific embodiment, for five non-standard defects of the black piano lacquer plastic center console, the information of each node is determined: scratch H (6.8mm, left button panel, piano lacquer layer thickness 80μm, hardness 3H), discolored spot Y1 (0.6mm×0.5mm), Y2 (0.4mm×0.3mm), fuzz M (4.2mm, right display screen bezel), particle K (0.35mm), and burr C (deviation 0.15mm).

[0051] Analysis of material and defect characteristics: Scratches on the piano lacquer surface lead to lacquer layer damage, and the exposed substrate easily attracts dust, forming particles; static electricity generated by friction at the burr attachment points accelerates the formation of discolored spots; friction between burr protrusions and tooling may cause scratches. Pairwise comparisons determined the influence relationships: the damaged area of ​​scratch H leads to particle K adhesion; the static electricity of burr M causes discolored spots Y1 and Y2; friction between burr C and tooling produces scratch H. A directed edge H→K, M→Y1, M→Y2, C→H was established, with edge attributes labeled "lacquer layer damage attracts dust," "static electricity attracts discolored particles," and "mechanical friction causes scratches."

[0052] By filtering through the timestamps of the detection records, spur C (detection time 09:12:35) is the earliest appearing node and is used as the initial node. A path chain is formed by traversing along the directed edges: C→H→K; C→H (independent branch); M→Y1, M→Y2 (parallel branches). Here, C directly points to H, H directly points to K, and M directly points to Y1 and Y2. All nodes are covered by the path chain.

[0053] Burr C is determined to be the root defect; scratch H is a first-level derivative defect; particle K and fuzz M are second-level derivative defects; and discolored spots Y1 and Y2 are third-level derivative defects.

[0054] Treatment plan: Root defect C was treated with laser polishing (5W power, 0.2mm spot diameter) to remove the raised portion, with the deviation controlled within 0.05mm after polishing; primary defect H was treated with piano lacquer repair agent (viscosity 200cP), curing temperature 60℃, duration 15 minutes; secondary defect K was removed with an anti-static adhesive pen (viscosity grade 2N / cm), and M was wiped with a lint-free cloth soaked in 95% alcohol; tertiary defects Y1 and Y2 were treated after M, and then treated with the same color paint (thickness 30μm). A follow-up inspection was conducted 24 hours after treatment; all defects had disappeared, and the surface gloss was restored to 90GU (original standard 85-95GU).

[0055] like Figure 2 As shown, this application provides an automated visual inspection system for exterior defects of automotive interior parts, including: Region of Interest (ROI) Definition Module: Includes a light source template generation unit and an ROI definition unit; wherein, the light source template generation unit adjusts the light source combination, sets the exposure parameters and brightness ratio, and saves it as a light source template; the ROI definition unit defines the left and right ROIs of the interior parts; The defect detection module includes an image acquisition unit, a defect detection unit, and a defect information aggregation unit. The image acquisition unit controls the conveyor line to move interior trim parts to the defect detection area, using a combination of fixed and moving conveyor methods. It calls a light source template to capture images of the interior trim parts in the left and right regions of interest. The defect detection unit preprocesses the interior trim part images and uses a defect detection algorithm to detect different types of defects. The defect information aggregation unit aggregates the defect information detected on the left and right sides to obtain the defect detection results. The routine defect judgment module includes: an interior part acceptance judgment unit, an NG re-inspection clamping unit, an association establishment unit, and a routine defect judgment unit. The interior part acceptance judgment unit determines whether an interior part is acceptable based on defect detection results. When an interior part is unacceptable, the NG re-inspection clamping unit moves it to the NG product unloading area, adjusts the clamping angle according to the defect location, and clamps it to the NG re-inspection line. The association establishment unit reads the markings on the surface of the interior part and establishes a unique association with the corresponding automotive parts. The routine defect judgment unit extracts defect features from the defect detection results, calls a pre-trained routine defect judgment model, and outputs whether the defect is a routine defect. The solution generation module includes a directed edge establishment unit, an initial defect location screening unit, and a solution generation unit. The directed edge establishment unit, for interior parts with unconventional defects, establishes directed edges between nodes based on the material properties of the interior parts and the physical properties of the defects, using each defect as a node. The initial defect location screening unit selects the earliest defect node in the graph structure to form an initial defect location set. The solution generation unit starts from the initial defect node, traverses along the directed edges of the graph structure to reach subsequent nodes, forming an influence path chain, distinguishing between root defects and derived defects, and generating a solution.

[0056] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An automated visual inspection method for exterior defects of automotive interior parts, characterized in that, Includes the following steps: Adjust the light source combination, set the exposure parameters and brightness ratio, and save it as a light source template; define the left and right sides of the interior parts' regions of interest; The control conveyor line drives the interior trim parts to the defect detection area, using a combination of fixed and moving conveyor methods; the light source template is called to capture images of the interior trim parts in the regions of interest on the left and right sides; the interior trim part images are preprocessed, and the defect detection algorithm is called to detect different types of defects; the defect information detected on the left and right sides is summarized to obtain the defect detection results; The preprocessing of interior component images and the invocation of defect detection algorithms to detect different types of defects include: The interior component images are preprocessed to unify their size and color space. The images are then cropped to retain the effective portion containing the region of interest and to remove irrelevant background. The cropped images are scaled according to the detection accuracy requirements. An adaptive filtering algorithm is used to eliminate high-frequency noise interference in the images to address random noise. The brightness differences in the images caused by uneven lighting are corrected by adjusting the brightness in different areas. Color features are extracted from interior parts images. The color parameters of pixels are compared with the color reference range of qualified areas to filter out abnormal pixels whose colors exceed the reference range. Connectivity analysis is performed on abnormal pixels to determine whether they form independent point-like objects. Size features are combined to determine whether they are black dots or discolored dots. Morphological operations are used to enhance the contours of small foreign objects in the image, highlighting the linear features of fuzz and the blocky features of particles. Edge extraction is performed, and feature analysis is used to distinguish between fuzz and particles. Length, width, and distribution density are combined to filter out non-defect interference. Based on the edge detection algorithm, the edge contour of the interior parts surface is extracted with reference to the standard edge data of NG materials. The deviation between the edge contour and the standard contour is calculated to identify protrusions that exceed the preset tolerance range, and burrs and flash are identified. Texture features are extracted from the image, and difference operations are performed with the texture model of qualified areas to filter out texture abnormal areas. When there is a continuous linear gray-scale change band, it is identified as a scratch. When there is a uniform texture abnormality, it is identified as sandpaper marks. The control conveyor line drives the interior parts to the defect detection area, employing a combination of fixed and follow-up conveying methods, including: Start the conveyor line and place the interior trim pieces into the positioning fixtures on the conveyor line, ensuring that the interior trim pieces are aligned with the positioning reference of the fixtures to achieve initial fixation. The conveyor line drives the fixtures and interior trim pieces to move along a preset path, using a combination of fixed and moving conveying methods: when the movement reaches the preset fixed detection section, the conveyor line stops operating, and the interior trim pieces are in a fixed state. At this time, the fixtures maintain a stable position through a mechanical locking structure. When the movement is in a non-fixed detection section, the interior trim pieces move synchronously with the conveyor line and are in a moving state. The position detection sensor located at the entrance of the inspection area monitors the position of the interior parts in real time. When the interior parts are detected to have reached the preset position at the boundary of the inspection area, the corresponding positioning mechanism is triggered depending on whether the interior parts are fixed or moving: when they are fixed, it is confirmed that the interior parts have stopped at the fixed reference position in the inspection area; when they are moving, the position data of the interior parts are collected in real time by the vision sensor, and the conveyor line adjusts its running speed to bring the interior parts into the effective inspection range of the inspection area. Based on the defect detection results, determine whether the interior parts are qualified; when the interior parts are unqualified, transfer them to the NG product unloading area, adjust the gripper angle according to the defect location, and clamp them to the NG re-inspection line; read the markings on the surface of the interior parts and establish a unique association with the corresponding automotive parts; extract defect features from the defect detection results, call the pre-trained defect routine judgment model, and output whether the defect is a routine defect; The step of extracting defect features from defect detection results, calling a pre-trained defect routine judgment model, and outputting whether the defect is a routine defect includes: Defect features are extracted from defect detection results, including defect type, specific location within the vehicle, quantity, size, and morphological features. These extracted features are then structured. Corresponding component information is retrieved, including component model, production batch, production standard, usage environment requirements, past quality inspection records, and common defect distribution patterns for components of the same model. This information is then organized into structured data. The defect features are combined with the component information to form a comprehensive dataset for model input. The routine defect judgment model is trained based on defect data from components of the same model. During training, the input data is scaled to a uniform size, and the component surface is divided into four faces: A, B, C, and D. Different defect thresholds are set for different defect types, with the thresholds decreasing sequentially from surface A to surface D, reflecting increasingly stringent defect control. A convolutional neural network algorithm is used for training to learn common defect feature patterns corresponding to different components. The fused component information and defect feature data are input into the model in the required format. The model calculates the matching degree by comparing the current defect feature with a common defect feature library of components of the same model. Defects on surfaces A and B are considered common defects, while defects on surfaces C and D are considered non-common defects. Defects on surface A do not require processing, while defects on surface B require rework. The model outputs the result based on the matching degree threshold to determine whether the current defect is a common defect. For interior parts with unconventional defects, each defect is used as a node, and directed edges are established between nodes based on the material properties of the interior parts and the physical properties of the defects. The earliest defect node is selected in the graph structure to form an initial set of defect locations. Starting from the initial defect node, subsequent nodes can be reached by traversing along the directed edges of the graph structure to form an influence path chain, distinguishing between root defects and derived defects and generating solutions. The process of starting from the initial defect node, traversing along the directed edges of the graph structure to reach subsequent nodes to form an influence path chain, distinguishing between root defects and derived defects, and generating solutions includes: Starting from the initial defect node in the graph structure, all subsequent defect nodes that can be reached directly or indirectly are traced according to the direction of the directed edges between nodes; the connection relationship of each node is recorded, and the nodes are sorted out in the order of initial node, directly affected nodes and indirectly affected nodes to form an influence path chain, ensuring that the path chain covers all related defects caused by the initial defect; The starting point of the impact path chain is identified as the root defect. All nodes in the path chain other than the root defect are considered derivative defects. These are further subdivided according to their impact level: nodes directly affected by the root defect are first-level derivative defects, nodes affected by first-level derivative defects are second-level derivative defects, and so on, thus determining the hierarchical relationship between each derivative defect and the root defect. For the root defect, a priority treatment plan is developed based on the material characteristics and defect type of the interior parts at its corresponding location to block the impact path at its source. For derivative defects, a treatment order is determined based on their hierarchical relationship with the root defect, prioritizing first-level derivative defects, followed by second-level and higher derivative defects. When a derivative defect is a direct result of the root defect, its disappearance is checked after treating the root defect. When separate treatment is required, an appropriate repair method is selected based on material characteristics. All treatment steps are integrated to generate a solution.

2. The automated visual inspection method for exterior defects of automotive interior parts according to claim 1, characterized in that, The debugging light source combination, setting exposure parameters and brightness ratio, and saving it as a light source template includes: Automotive interior parts are classified according to their material properties. For a specific type of interior part, typical samples containing different defects, sizes, and shapes are selected and fixed in the calibration fixture at the inspection station to ensure that the sample position is consistent with the positioning reference during actual inspection. Suitable light source types and combinations are selected, and the light source angle, brightness output ratio, and camera exposure parameters of each light source are adjusted. Iterative optimization is carried out according to the defect imaging quality assessment standard. When the image of a certain type of defect is blurred, the light source intensity in the corresponding direction is increased. When there is reflection interference, the light source angle is adjusted or the exposure time is reduced. When all target defects of the typical sample can be identified in the image without interference, a candidate light source template for the material is formed; when different typical samples of the same material are replaced, the candidate light source template is called for imaging verification. When the stability of the defects meets the detection requirements, it is confirmed to be valid and saved as a light source template.

3. The automated visual inspection method for exterior defects of automotive interior parts according to claim 1, characterized in that, The defined regions of interest on the left and right sides of the interior trim include: Based on the structural characteristics of automotive interior parts and the locations where defects are prone to occur, and in conjunction with product design drawings, the areas that need to be focused on for inspection are determined, while background or non-critical areas that do not need to be inspected are excluded. At the inspection station, the placement reference of the interior parts is determined by positioning fixtures. Based on the placement reference, the boundary of the area to be inspected and the corresponding specific physical location range are marked, forming a closed area outline in the image. By capturing images of interior parts, it is determined whether the area contained by the camera coordinate boundary completely covers the target detection area. When there is a deviation, the boundary marker is adjusted, and the process is repeated and verified until the region contour matches the detection requirements. The left and right regions of interest are saved respectively.

4. The automated visual inspection method for exterior defects of automotive interior parts according to claim 1, characterized in that, The step of calling the light source template to capture images of interior parts in the left and right regions of interest includes: The system identifies the material type of the current interior trim component, retrieves the corresponding light source template based on the material type, sends a command to the light source controller, and activates the light source according to the light source template for preheating. After the light source stabilizes, a positioning command is sent to the inspection robot. When the interior trim component is in a fixed state, the first robot, carrying a camera, moves to the preset fixed inspection position on the right, and the second robot, carrying a camera, moves to the preset fixed inspection position on the left. When the interior trim component is in a moving state, the robot monitors the position of the interior trim component in real time through a vision sensor, dynamically adjusting the X-axis and Y-axis displacement and rotation angle to ensure that the inspection posture matches the position of the interior trim component. The robot arms are placed with their working ranges staggered. The system ensures that the camera lens axis is aligned with the center of the corresponding region of interest on the left and right sides, and captures images of the interior trim component in the region of interest on both sides.

5. The automated visual inspection method for exterior defects of automotive interior parts according to claim 1, characterized in that, For interior trim parts with unconventional defects, directed edges are established between nodes based on the material properties of the trim parts and the physical properties of the defects, using each defect as a node. Identify the type, location, and corresponding interior trim material of unconventional defects. Treat each defect as a node in the graph, and associate its location with the material information and physical properties of the defect itself in the node attributes. Analyze the material properties of the interior trim and the physical properties of the defects at the locations corresponding to each node to determine the interaction between different materials and defect properties. Perform pairwise comparisons on all defect nodes and determine the influence relationship based on the material properties and physical properties of the defects. When the characteristics of defect P, combined with the material properties of its location, directly or indirectly cause the generation or deterioration of defect Q, it is determined that P has an influence on Q. For defect nodes that have an impact relationship, a directed edge is established from P to Q, and the edge attributes are labeled according to the impact, forming a graph structure that includes all non-standard defect nodes and the impact relationships between nodes.

6. An automated visual inspection system for exterior defects of automotive interior parts, using the automated visual inspection method for exterior defects of automotive interior parts according to any one of claims 1-5, characterized in that, include: Region of Interest (ROI) Definition Module: Includes a light source template generation unit and an ROI definition unit; wherein, the light source template generation unit adjusts the light source combination, sets the exposure parameters and brightness ratio, and saves it as a light source template; the ROI definition unit defines the left and right ROIs of the interior parts; The defect detection module includes an image acquisition unit, a defect detection unit, and a defect information aggregation unit. The image acquisition unit controls the conveyor line to move interior trim parts to the defect detection area, using a combination of fixed and moving conveyor methods. It calls a light source template to capture images of the interior trim parts in the left and right regions of interest. The defect detection unit preprocesses the interior trim part images and uses a defect detection algorithm to detect different types of defects. The defect information aggregation unit aggregates the defect information detected on the left and right sides to obtain the defect detection results. The routine defect judgment module includes: an interior part acceptance judgment unit, an NG re-inspection clamping unit, an association establishment unit, and a routine defect judgment unit. The interior part acceptance judgment unit determines whether an interior part is acceptable based on defect detection results. When an interior part is unacceptable, the NG re-inspection clamping unit moves it to the NG product unloading area, adjusts the clamping angle according to the defect location, and clamps it to the NG re-inspection line. The association establishment unit reads the markings on the surface of the interior part and establishes a unique association with the corresponding automotive parts. The routine defect judgment unit extracts defect features from the defect detection results, calls a pre-trained routine defect judgment model, and outputs whether the defect is a routine defect. The solution generation module includes a directed edge establishment unit, an initial defect location screening unit, and a solution generation unit. The directed edge establishment unit, for interior parts with unconventional defects, establishes directed edges between nodes based on the material properties of the interior parts and the physical properties of the defects, using each defect as a node. The initial defect location screening unit selects the earliest defect node in the graph structure to form an initial defect location set. The solution generation unit starts from the initial defect node, traverses along the directed edges of the graph structure to reach subsequent nodes, forming an influence path chain, distinguishing between root defects and derived defects, and generating a solution.

Images