Defect identification method and system for hand washing disc based on visual inspection

Abstract

The invention discloses a defect identification method and system for a hand washing basin based on visual inspection, and relates to the technical field of visual inspection. A plurality of defect characteristics of the hand washing basin are determined based on the form of each part, the working process of the hand washing basin and a corresponding panoramic image; and according to the feature positions of the plurality of defect features, the corresponding feature forms and the overall form of the wash basin, the defect identification system of the wash basin is determined, and the accuracy of the defect identification system of the wash basin is improved. Based on the spatial position of each part and the corresponding shipment quality grade, constructing a quality state diagram of the wash basin, determining a target defect area according to the quality state diagram, and determining a corresponding defect maintenance event according to the area position of the target defect area, the corresponding defect form and the maintenance system of the wash basin, and the shipment quality grade of each part is controlled, so that the accuracy of defect maintenance events is improved.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, and more particularly to a method and system for defect identification of handwashing basins based on visual inspection. Background Technology

[0002] With the development of technology, bathroom products are gradually being applied to people's lives. Washbasins are a type of bathroom product. Washbasins are produced on corresponding production lines and undergo corresponding visual inspection during the production process. In the existing technology, washbasins are produced on washbasin processing lines and undergo multiple processing steps. Each processing step involves processing content and marking corresponding washbasin images. Defects are determined based on the recognition of each washbasin image. However, this ignores the consideration of multiple defect features and the overall shape of the washbasin, affecting the accuracy of the defect recognition system and resulting in low accuracy of defect maintenance events. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a method and system for defect identification of handwashing basins based on visual inspection.

[0004] This invention provides a method for defect identification of handwashing basins based on visual detection, comprising: When the handwashing tray is in a dynamic processing state, mark multiple processing steps of the handwashing tray and perform visual inspection on the handwashing tray to determine the combination of handwashing tray images for each processing step; In each processing step, multiple key part images are determined based on the process content of the processing step and the corresponding handwashing basin image combination, and the corresponding component shape is determined based on the image recognition of each key part image; Based on the shape of each component, the working process of the handwashing tray, and the corresponding panoramic image, multiple defect features of the handwashing tray are determined. Based on the feature location, corresponding feature shape, and overall shape of the handwashing tray, a defect identification system for the handwashing tray is determined. The defect identification system defines the classification, severity level, and interrelationship of different defects. In the defect identification system, defect events in each part are identified based on the identification system. The outgoing quality grade of each part is determined according to the importance level of the part, the corresponding defect event, and the processing requirements of the handwashing basin. The outgoing quality grades include superior, qualified, rework, and scrap. A quality status diagram of the handwashing tray is constructed based on the spatial location of each part and the corresponding shipment quality level. The target defect area is determined based on the quality status diagram. The corresponding defect maintenance event is determined based on the location of the target defect area, the corresponding defect morphology, and the handwashing tray maintenance system. The defect maintenance event includes work order number, product serial number & component information, target area & defect description, maintenance steps, required materials and tools, estimated working hours, and safety instructions.

[0005] This invention provides a visual detection-based defect identification method for handwashing basins, which is applied to the aforementioned visual detection-based handwashing basin defect identification method.

[0006] Compared with the prior art, the beneficial effects of the present invention are: (1) When the handwashing basin is in a dynamic processing state, mark multiple processing steps of the handwashing basin and perform visual inspection on the handwashing basin to determine the handwashing basin image combination of each processing step; in each processing step, determine multiple key part images according to the process content of the processing step and the corresponding handwashing basin image combination, and determine the corresponding component shape based on the image recognition of each key part image; determine multiple defect features of the handwashing basin based on the shape of each component, the working process of the handwashing basin and the corresponding panoramic image, and determine the defect identification system of the handwashing basin according to the feature position of multiple defect features, the corresponding feature shape and the overall shape of the handwashing basin. The visual inspection of the handwashing basin is introduced to control multiple key part images, and the consideration of the feature position of multiple defect features, the corresponding feature shape and the overall shape of the handwashing basin is taken into account, which improves the accuracy of the defect identification system of the handwashing basin.

[0007] (2) In the defect identification system, the defect events of each part are determined based on the identification of the defect identification system. The shipment quality level of each part is determined according to the importance level of the part, the corresponding defect event and the processing requirements of the handwashing basin. The quality status map of the handwashing basin is constructed based on the spatial location of each part and the corresponding shipment quality level. The target defect area is determined according to the quality status map. The corresponding defect maintenance event is determined according to the regional location of the target defect area, the corresponding defect form and the handwashing basin maintenance system. The shipment quality level of each part is controlled, realizing the overall consideration of the regional location of the target defect area, the corresponding defect form and the handwashing basin maintenance system, and improving the accuracy of defect maintenance events. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the defect identification method for handwashing basins based on visual detection in an embodiment of the present invention. Figure 2This is a flowchart illustrating step S11 of the defect identification method for handwashing basins based on visual detection in an embodiment of the present invention. Figure 3 This is a flowchart illustrating step S12 of the defect identification method for handwashing basins based on visual detection in an embodiment of the present invention. Figure 4 This is a flowchart illustrating step S13 of the defect identification method for handwashing basins based on visual detection in an embodiment of the present invention. Figure 5 This is a flowchart illustrating step S14 of the defect identification method for handwashing basins based on visual detection in an embodiment of the present invention. Figure 6 This is a flowchart illustrating step S15 of the defect identification method for handwashing basins based on visual detection in an embodiment of the present invention. Figure 7 This is a schematic diagram of the structural composition of a visual detection-based handwashing basin for defect identification in an embodiment of the present invention. Detailed Implementation

[0009] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0010] Please see Figures 1 to 7 A defect identification method for handwashing basins based on visual inspection is proposed and applied to visual inspection scenarios. The defect identification method for handwashing basins based on visual inspection includes: Step S11: When the handwashing tray is in a dynamic processing state, mark multiple processing steps of the handwashing tray and perform visual inspection on the handwashing tray to determine the handwashing tray image combination of each processing step; Step S12: In each processing step, multiple key part images are determined based on the process content of the processing step and the corresponding handwashing basin image combination, and the corresponding component shape is determined based on the image recognition of each key part image; Step S13: Based on the shape of each component, the working process of the handwashing basin and the corresponding panoramic image, determine multiple defect features of the handwashing basin, and determine the defect identification system of the handwashing basin according to the feature position, corresponding feature shape and overall shape of the handwashing basin. Step S14: In the defect identification system, the defect events of each part are determined based on the identification of the defect identification system, and the shipment quality level of each part is determined according to the importance level of the part, the corresponding defect event and the processing requirements of the handwashing plate. Step S15: Construct a quality status map of the handwashing tray based on the spatial location of each part and the corresponding shipment quality level. Determine the target defect area based on the quality status map. Determine the corresponding defect maintenance event based on the location of the target defect area, the corresponding defect morphology, and the handwashing tray maintenance system.

[0011] refer to Figure 2 In step S11, the specific steps are as follows: S111: The handwashing tray moves dynamically in the handwashing tray processing line and goes through multiple processing steps. At this time, the handwashing tray is dynamically processed in each processing step and is visually detected by the corresponding camera during the processing to output multiple handwashing tray images of each processing step. S112: In multiple processing steps, multiple associated images of the processing step are determined based on the processing weight of the processing step, the corresponding process content, and multiple handwashing plate images. Based on the multiple associated images and the processing form of the handwashing plate in the processing step, the handwashing plate image combination of the processing step is determined, so as to determine the handwashing plate image combination of multiple processing steps.

[0012] In the embodiments of this application, the handwashing tray moves dynamically in the handwashing tray processing line and goes through multiple processing steps. At this time, the handwashing tray is dynamically processed in each processing step and is visually detected by the corresponding camera during the processing to output multiple handwashing tray images of each processing step, thus introducing the output of multiple handwashing tray images of each processing step.

[0013] In industrial production lines, the movement of workpieces is often complex and variable, with start-stop, jitter, and positional deviations being common occurrences. To ensure that the vision system accurately captures targets in dynamic environments, fixed-time triggering mechanisms must be abandoned in favor of event-based real-time positioning technology. By deploying photoelectric sensors, proximity switches, or laser rangefinders at key nodes in the production line, the system can perceive the workpiece position in real time. When the workpiece enters the predetermined field of view, the sensor immediately generates a digital signal, which is transmitted to the PLC or vision controller as a precise marker of the workpiece's positioning, providing a zero-point reference for subsequent image acquisition.

[0014] The vision system also needs to understand the specific tasks of the current process. By communicating with the MES or PLC, the system can obtain process flow information, such as drilling or grinding. This context directly determines the detection algorithm, parameter settings (exposure time, light source type), and expected image features loaded by the vision system, ensuring that the detection logic is highly matched with the process requirements.

[0015] Detection during dynamic processing must be precisely synchronized with the equipment's movements; for example, during the drilling process, detection is triggered before the drill bit contacts the hole, during drilling, and after the drill bit retracts; the I / O signals output by the PLC or equipment controller (such as spindle start and processing completion) become the trigger source for the vision system, ensuring that image capture is strictly aligned with the critical processing time points.

[0016] A single camera cannot cover complex curved surfaces and various types of defects, so a multimodal approach is required: a multi-camera layout (top-down, oblique, and side views) to eliminate occlusion; multi-light source technology (coaxial light, dark field light, and colored ring light) to optimize imaging for different defects; a high-speed global shutter camera to ensure clear images without motion blur in dynamic environments; raw images need to be structured and labeled to generate authentication information containing metadata such as timestamps, workpiece serial numbers, process IDs, and camera parameters; data is written to a database or message queue in real time to ensure traceability and provide contextually clear material for subsequent analysis.

[0017] Specifically, in front of the drilling station on the handwashing basin production line, a through-beam photoelectric sensor is installed under the conveyor belt. When the handwashing basin completely blocks the light beam, the sensor sends a high-level signal to the PLC, which then stops the conveyor belt to ensure that the workpiece is still at the moment of shooting. At the same time, the workpiece serial number is recorded as data for subsequent processes.

[0018] After receiving the workpiece positioning signal, the PLC queries the process status table to confirm that the current process is drilling. Then, it sends a command packet to the vision master controller via the industrial Ethernet, which includes the task ID (such as DRILLING) and parameters (top exposure 5000μs, ring light source on). The vision system automatically switches to the drilling detection mode to optimize the imaging conditions.

[0019] The drilling robot controller sends signals at three points in time: the drill bit starts rotating (Drill_Start), drilling is completed (Drill_Complete), and the robot returns to its home position (Arm_Home). The vision system is programmed to take a picture the instant the Drill_Complete signal is triggered, at which point the hole has been formed but metal shavings have not been removed, which is the best time to detect burrs.

[0020] When the Drill_Complete signal is triggered, the three cameras work synchronously: the top camera works with coaxial light to detect the hole diameter; the 45-degree angle camera uses ring light to capture chipped edges; and the side camera uses dark field light to highlight drilling stress cracks. The three sets of images are packaged into a data record, including timestamp, workpiece SN (A-WS-20251008-0145), process (DRILLING), and trigger event (Drill_Complete). The image file name is associated with metadata to form a complete visual snapshot.

[0021] Furthermore, in multiple processing steps, multiple associated images for a processing step are determined based on the processing weight of the processing step, the corresponding process content, and multiple handwashing basin images. Based on the multiple associated images and the processing form of the handwashing basin in the processing step, the combination of handwashing basin images for the processing step is determined. This method takes into account the overall consideration of multiple associated images and the processing form of the handwashing basin in the processing step, ensuring the accuracy of the combination of handwashing basin images for the processing steps.

[0022] At this point, before processing the image, the system needs to understand the importance of the current process. By querying the MES or process database, it loads context information containing processing weights. Processing weights are quantitative indicators (0-1) that are pre-set by process engineers based on the impact of the process on product performance, safety, and cost. For example, the drilling process, which affects sealing, has a much higher weight than the grinding process, which only affects aesthetics. This weight, as an attention coefficient, directly affects the sensitivity of subsequent image screening and the threshold for defect judgment.

[0023] The process of filtering related images from a massive amount of raw images is carried out in two stages: quality pre-screening: IQA algorithms (such as Laplacian variance, signal-to-noise ratio) are used to evaluate image sharpness, brightness, etc., and low-quality images caused by motion blur, underexposure, or water mist interference are removed; correlation analysis: key features of the process (such as hole edges in the drilling process) are identified through lightweight deep learning models (such as MobileNet-SSD) or template matching; only images containing high-confidence key features are identified as related images.

[0024] It provides evaluation criteria for image combinations. The processing form is a digital description of the ideal state of the part after the process is completed, based on CAD models and engineering drawings. It includes geometric tolerances (such as position and roundness), surface quality (roughness Ra) and appearance standards (color and gloss). The process context, associated image set and processing form benchmark are encapsulated into structured data units (such as JSON / XML) to form a handwashing basin image combination. This data package contains its own context and evaluation criteria, eliminating the need for subsequent modules to repeatedly query the database, thus improving processing efficiency.

[0025] Specifically, after the handwashing basin enters the drilling process, the vision system obtains the context packet from the MES: {"Process_ID":"P003","Process_Name":"Drilling","Weight":0.95,"Criticism_Level":"Critical"}; the system resolves that the drilling weight is 0.95, which is the highest level, meaning that any minor anomaly is considered high risk, and the subsequent screening criteria will be extremely strict.

[0026] S111 outputs 20 drilling process images: Quality pre-screening removes 5 blurry images (clarity < threshold), leaving 15; the YOLO model infers from the remaining images, identifying 10 clear images of the faucet mounting hole (confidence > 95%), while the other 5 are not detected due to coolant obstruction; finally, the 10 images are marked as relevant images. The machining morphology benchmark for the drilling process is loaded as follows: {"Feature":"MountingHole","Nominal_Diameter":"35.0mm","Tolerance"} clearly defines the quantitative standard for the perfect faucet mounting hole; the system integrates the previous results to generate a drilling process data package, which clearly instructs the subsequent system: This is high-weight process (drilling) data. Please analyze 10 selected images according to the defined standard to ensure accurate and efficient information transmission.

[0027] refer to Figure 3 In step S12, the specific steps are as follows: S121: Real-time monitoring of each processing step of the handwashing tray, marking the process content of each processing step, determining the corresponding processing part based on the identification of the process content, and determining multiple key part images based on the processing part, the overall shape of the handwashing tray, and the combination of the handwashing tray image. S122: In multiple key part images, multiple key regions are determined based on the recognition of each key part image, and multiple corresponding morphological features are determined based on the detection of each key region. The corresponding component shape is determined based on the multiple morphological features, the corresponding processing parts, and the overall shape of the handwashing basin.

[0028] In the embodiments of this application, each processing step of the handwashing basin is monitored in real time, and the process content of each processing step is marked. The corresponding processing part is determined based on the identification of the process content. Multiple key part images are determined based on the processing part, the overall shape of the handwashing basin, and the combination of handwashing basin images. This approach takes into account the overall consideration of the processing part, the overall shape of the handwashing basin, and the combination of handwashing basin images, ensuring the accuracy of multiple key part images.

[0029] At this point, the vision system needs to understand the current task, rather than analyzing images in isolation; it communicates with the MES / PLC in real time via OPC-UA or ModbusTCP protocol to obtain the workpiece's process label (such as DRILLING); the system maintains a process-content knowledge base internally, mapping the label to a recognizable visual task description; for example, drilling is defined as creating a circular through hole of a specific size and edge quality at a specified location, thus providing a clear parameterized target for visual analysis.

[0030] The abstract process content is transformed into specific processing parts in the image; two technical approaches are adopted: loading a standard CAD model, aligning the model to the actual image through SIFT / SURF feature point or edge contour matching, and directly obtaining the component coordinates; at the same time, using U-Net or DeepLab network, the component is identified at the pixel level and a mask is generated (e.g., the basin is marked as 1, the table is marked as 2), and the accurate contour is output.

[0031] Based on the previous results, key parts of the image are intelligently cropped from the original image. By edge detection and ellipse fitting, it is determined whether the workpiece has rotation or perspective distortion. The affine / perspective transformation matrix is ​​calculated for correction to ensure posture consistency. On the corrected image, a closely fitting polygon or minimum bounding rectangle ROI (extended edge pixels) is generated based on the part contour to accurately crop the local image.

[0032] Specifically, after the handwashing basin's serial number is scanned by RFID, the PLC sends a data packet; the vision system queries the knowledge base and parses out the polishing process details. {"Target_Areas":["Basin_Surface","Ledge_Surface"],"Goal":"Achieve_Smoothness","Defects_to_Find":["Scratches","Orange_Peel","Pits"]}, the task is to focus on the basin and countertop and detect smoothness defects.

[0033] The system loads a panoramic image of the polishing process and starts a pre-trained semantic segmentation model. In the output mask image, the inner surface of the basin is blue, the countertop is green, and the overflow outlet and faucet hole are red and yellow, respectively. By analyzing the color connected components, the system obtains the pixel-level contour coordinates of the basin and the countertop.

[0034] Upon detecting a 1.5-degree clockwise rotation of the washbasin, the system calculates a counterclockwise transformation matrix and corrects the image to align the edges with the coordinate axes. It generates a minimum bounding rectangle (ROI) (expanded by 10 pixels) for the basin (blue mask) and countertop (green mask), cropping out two local images: key part images_basin.jpg and key part images_countertop.jpg. This provides a highly consistent and content-focused data foundation for the S122's fine-grained shape recognition.

[0035] Furthermore, in multiple key part images, multiple key regions are determined based on the recognition of each key part image, and multiple corresponding morphological features are determined based on the detection of each key region. The corresponding component shape is determined based on the multiple morphological features, the corresponding processing parts, and the overall shape of the handwashing basin. This comprehensive consideration of multiple morphological features, the corresponding processing parts, and the overall shape of the handwashing basin ensures the accuracy of the corresponding component shape.

[0036] At this point, a part (such as a tabletop) may contain sub-regions with different functions or risk levels, such as the visual focus area and the vulnerable area. This semantic division based on prior knowledge enables subsequent detection to be tailored to local conditions and adopt differentiated algorithms and standards. Regions are defined according to geometric features, such as a 10mm annular area at the edge of a hole as a high-stress area and a 5mm strip at the edge of a component as a vulnerable area. A fine segmentation model is used to further distinguish semantic labels such as the central area and the edge area within the tabletop.

[0037] Multi-dimensional physical examinations are performed in parallel for each key area to extract quantified morphological features: Geometric features: through sub-pixel edge detection (such as Canny) and contour analysis, size, position, roundness, etc. are measured to detect structural defects; Surface texture features: roughness is quantified using Gabor filters or GLCM, and anomaly detection algorithms such as Autoencoder are combined to identify scratches and pits; Optical features: the average color difference (ΔE) is calculated in the CIELAB color space to detect color difference or stains.

[0038] By fusing scattered morphological features with contextual information, structured component morphological data is generated: scattered features in the same region are merged (e.g., fractured scratch segments are merged into complete defects); local features are compensated or their weights are adjusted by referring to the overall morphological assessment (e.g., distinguishing between local defects and overall deformation); and the data is output in JSON format, including component name, overall status, and a list of key regional features.

[0039] Specifically, the system performs geometric rule division on the key part image_countertop.jpg: it identifies the faucet mounting holes and overflow outlets, generates a circular area with a radius of 15mm centered on the hole center, and defines it as the key area_hole peri-stress zone; it extracts the outer contour of the countertop and offsets it inward by 5mm to form an annular area, which is defined as the key area_countertop edge zone; the remaining part is defined as the key area_countertop center zone; the countertop is divided into three sub-regions with clear engineering significance.

[0040] The system performs parallel analysis on three key areas: Central area: A surface defect classification network detects bubbles, quantified as {defect type: bubble, area: 1.5 mm², grayscale contrast: 45}; Edge area: After edge enhancement using the Gaussian Laplacian operator, Hough transform identifies scratches, quantified as {defect type: scratch, length: 12 mm, average width: 0.07 mm, orientation angle: 25°}; Peripheral stress area: Subpixel edge detection fits the hole contour, radial deviation is calculated, quantified as {defect type: chipped edge, maximum radial deviation: 0.3 mm, number of chipped edges: 2}; The system integrates the detection results, refers to the overall good flatness assessment, confirms the defect as a localized problem, and finally generates a data package of the tabletop component's shape.

[0041] refer to Figure 4In step S13, the specific steps are as follows: S131: Collect the working history of the handwashing tray, determine the processing events of the handwashing tray in multiple processing steps based on the detection of the working history of the handwashing tray, and at the same time, collect panoramic images of the handwashing tray, and determine the first-level defect factors based on the working history of the handwashing tray and the corresponding panoramic images. S132: Determine the second level of defect factors based on the shape of each component and the panoramic image of the handwashing basin, determine multiple defect features of the handwashing basin based on the first and second level of defect factors, and mark the feature positions of the multiple defect features. S133: Determine the feature shape of the defect feature based on the identification of each defect feature, determine the corresponding defect identification framework of the handwashing basin based on the feature position of multiple defect features and the overall shape of the handwashing basin, and determine the defect identification system of the handwashing basin based on the defect identification framework and the feature shape of multiple defect features.

[0042] In the embodiments of this application, the working process of the handwashing basin is collected, and the processing events of the handwashing basin in multiple processing steps are determined based on the detection of the working process of the handwashing basin. At the same time, a panoramic image of the handwashing basin is collected, and the first-level defect factor is determined according to the working process of the handwashing basin and the corresponding panoramic image. This approach takes into account both the working process of the handwashing basin and the corresponding panoramic image, ensuring the accuracy of the first-level defect factor.

[0043] At this point, to construct high-precision digital information for each workpiece, it is necessary to simultaneously integrate multi-source heterogeneous data; through an industrial IoT gateway, the following data are collected in real time: equipment status data (axis position, motor speed, valve status), with high frequency and high reliability; production management data such as process instructions, material batches, and operator information; physical process simulation quantities (such as injection molding machine pressure / temperature, drilling machine torque); all data streams are aligned with timestamps through PTP (Precision Time Protocol) to form a multi-dimensional time series database that corresponds one-to-one with the physical workpiece.

[0044] To identify potential risky processing events (operations deviating from the process window or abnormal patterns) from massive amounts of work history data, a combination of multiple algorithms is employed: Rule-based threshold detection: setting hard upper and lower limits for key parameters (e.g., drilling pressure ≤12MPa), exceeding the limit immediately triggers an event; Statistical process control (SPC): monitoring parameter mean and fluctuation through control charts, even if the parameter is within the threshold, continuous drift or increased fluctuation will trigger a process runaway event; Machine learning anomaly detection: using models such as isolated forest and LSTM to identify subtle anomalies in complex nonlinear relationships.

[0045] Establish a causal link between processing events and potential defects, and generate the first-level defect factor with causal assumptions: Based on the event timestamp (T_event), retrieve panoramic images within the time window from T_event to T_event+500ms (considering physical effect delays); according to the process to which the event belongs, query the knowledge base to obtain the main affected processing parts (e.g., drilling pressure events affect faucet holes); run a fast defect pre-detection algorithm (edge ​​detection, spot detection) in the corresponding area, and if a suspected defect is found, generate the first-level defect factor.

[0046] Specifically, after the handwashing basin's serial number is scanned by RFID, its operational data collection is initiated: Injection molding process: Records {mold temperature: 82°C, injection pressure: 95MPa, holding pressure: 75MPa, cooling time: 15.2s} at a frequency of 100Hz; Drilling process: Records {drilling speed: 2500rpm, feed rate: 50mm / min, spindle load: 45%, hydraulic pressure: 9.5MPa}; The data is uniformly timestamped and bound to the serial number for storage in digital information.

[0047] System analysis of handwashing basin drilling process data: At 14:30:15, the drilling hydraulic pressure was 11.8MPa (below the 12MPa threshold), but the SPC control chart showed that 5 consecutive sample points exceeded the +2σ control upper limit; SPC module triggered event: {"Event ID":"E001","Timestamp":"2025-10-08T14:30:15.500Z","Process":"Drilling","Event Type":"Process Out of Control","Description":"Drilling pressure remains high","Parameter value":"11.8MPa"}.

[0048] Inferences were made based on the event of persistently high drilling pressure (E001): Time alignment: retrieved the panoramic image taken at 14:30:15.750Z; Spatial positioning: consulted the knowledge base to confirm that the drilling process affected the faucet hole and overflow outlet; Visual verification: ran edge integrity detection in the faucet hole area and found minor breakpoints; generated the first-level defect factor: {"Factor ID":"F001","Inferred cause":"Persistently high drilling pressure (E001)","Potential defect type":"Broken edge","Potential defect location":"Famous hole edge (x=320,y=250)","Associated image":"IMG_Full_0145_003.jpg","Confidence":"High"}.

[0049] Furthermore, a second layer of defect factors is determined based on the shape of each component and the panoramic image of the handwashing basin. Based on the first and second layer of defect factors, multiple defect features of the handwashing basin are determined, and the feature positions of multiple defect features are marked. This approach takes into account both the first and second layer of defect factors, ensuring the accuracy of the multiple defect features of the handwashing basin.

[0050] At this point, the quantized component shape data packet output by S12 is directly parsed, and the defect description (such as {defect type: scratch, length: 15mm}) is directly used as the second defect factor; large-scale analysis is performed on the panoramic image provided by S112 to capture defects that may be ignored by local focusing; global color and gloss analysis: calculate the color histogram, average color difference (ΔE) and gloss distribution to detect overall color difference or gloss unevenness; global morphology analysis: generate a 3D point cloud through structured light or binocular vision, compare it with the CAD model, and quantify the overall warping or depression; global texture analysis: use Fourier transform or wavelet transform to identify large-area texture anomalies (such as flow lines).

[0051] Compare the potential location of the first-level factor with the actual location of the second-level factor, calculate the centroid distance or intersection-over-union (IoU) ratio, and determine whether they point to the same location; compare the potential type of the first-level factor with the actual type of the second-level factor (e.g., whether the edge collapse is consistent); defects that match both spatially and semantically are upgraded to high-confidence defects, and the inferred cause changes from potential to strongly correlated; first-level factors without matching are retained, but the inferred cause is marked as unknown, pointing to random factors or unmonitored process fluctuations; first-level factors without matching are retained, but the inferred cause is marked as unknown, pointing to random factors or unmonitored process fluctuations.

[0052] The merged list of defect features is standardized and labeled to establish a unified reference system: a global handwashing basin coordinate system is established based on the CAD model or physical benchmark (such as the center of the positioning hole), and all defect locations are converted to coordinates in this coordinate system to ensure uniqueness and traceability; the precise location of each defect is recorded, in the form of a point (center of the bubble), a line (start and end points of the scratch), or a polygon (outer edge contour); it is encapsulated into a structured list, with each entry containing defect ID, type, coordinates, quantitative features, confidence level, and inferred cause.

[0053] Specifically, the system performs parallel analysis on the handwashing basin: Component morphology analysis: S12 data shows that the countertop has {defect: scratch, length: 12mm, width: 0.07mm}, the faucet hole edge has {defect: chipped edge, maximum radial deviation: 0.25mm}, and the basin body has {defect: air bubble, area: 1.8mm²}; Panoramic image global analysis: It was found that the average brightness of the basin bottom area is 5% lower, and the ΔE value reaches 2.5 (exceeding the 2.0 threshold), generating a second defect factor: {defect type: color difference, location: basin bottom area, ΔE: 2.5}.

[0054] The system performs cross-validation: F001 in S131 ({potential cause: excessive drilling pressure, location: edge of faucet hole, type: edge chipping}) is a perfect spatial and semantic match with the edge chipping observed in S12, and is confirmed as a high-confidence defect, with the cause upgraded to excessive drilling pressure; scratches, bubbles, and color differences have no corresponding first-level factors and are confirmed as valid defects, with the inferred cause marked as unknown; the output contains a list of four confirmed defects (edge ​​chipping, scratches, bubbles, and color differences).

[0055] The system marks four defects: with the center of the faucet hole as the origin (0,0), the long side of the countertop as the X-axis, and the short side as the Y-axis; the marking positions are: D001 (chipped edge): the polygonal outline of the edge of the faucet hole, coordinates {(x1,y1),(x2,y2),...}; D002 (scratches): line segments, starting point (-50,120), ending point (-38,132); D003 (bubbles): point (25,-80); D004 (color difference): polygonal area of ​​the basin bottom, coordinates {...}; the final output is a list of defect features containing the four complete markings, each feature is accurately located in the standard coordinate system, providing a data foundation for S13 to build a defect identification system.

[0056] Therefore, based on the identification of each defect feature, the feature shape of the defect feature is determined. Based on the feature location of multiple defect features and the overall shape of the handwashing basin, a corresponding defect identification framework for the handwashing basin is determined. Based on this defect identification framework and the feature shape of multiple defect features, a defect identification system for the handwashing basin is determined. This system takes into account both the defect identification framework and the feature shape of multiple defect features, ensuring the accuracy of the defect identification system for the handwashing basin. At the same time, visual inspection of the handwashing basin is introduced to control the images of multiple key parts, taking into account the feature location, corresponding feature shape, and overall shape of the handwashing basin, thus improving the accuracy of the defect identification system for the handwashing basin.

[0057] At this point, the original morphological features (such as pixel length and grayscale value) are converted into real engineering units (millimeters, ΔE value, roughness Ra) through calibration parameters; according to the preset quality standards, the morphological parameters of the defects are assigned a severity score (such as a scratch with a length > 10mm gets 8 / 10 points), which serves as a key input for subsequent decisions; the inferred cause in S131 is finally bound to the confirmed defect in S132; if there is a match, it is marked as strongly correlated, otherwise it is unknown, so that the defect carries complete causal information.

[0058] Construct a framework to describe the interrelationships between defects, revealing their spatial and functional correlations, and assessing the cumulative effects and systemic risks: Spatial relationship analysis: Visualize all defect locations on a 2D / 3D model, and determine whether a defect is an isolated event or a systemic problem by calculating distance, distribution density, and clustering patterns; Functional impact analysis: Overlay defect locations with predefined functional area diagrams (such as sealing surfaces, mounting holes, and high-aesthetic areas) to assess the potential impact on product functionality; Structural impact analysis: Combine overall morphology (such as FEA stress distribution diagrams) to determine whether a defect is located in a high-stress concentration area and assess its potential hazards.

[0059] By combining the framework and form with the quality rule base, a defect identification system that can be invoked by S14 is formed. The defect identification system, as authoritative content, loads a multi-level decision tree or expert system rule base. The defect identification system includes enterprise quality standards (e.g., if the severity score of a critical functional area is >8, it is scrapped). The defect form and framework are used as input, and reasoning and matching are performed in the rule base to generate preliminary judgments and suggestions. The framework, defect details, and judgment results are encapsulated into a standardized data package, which not only describes the defect but also explains its meaning and handling suggestions. Therefore, the defect identification system defines the classification, severity level, and interrelationships of different defects.

[0060] Specifically, the system makes the final determination of four defects in the handwashing basin: D001 (chipped edge): {Type: chipped edge, size: 2.5mm, severity score: 9 / 10, inferred cause: excessive drilling pressure (strong correlation)}; D002 (scratches): {Type: scratches, length: 12mm, severity score: 7 / 10, inferred cause: unknown}; D003 (bubbles): {Type: bubbles, area: 1.8mm², severity score: 3 / 10, inferred cause: unknown}; D004 (color difference): {Type: color difference, ΔE: 2.5, severity score: 6 / 10, inferred cause: unknown}.

[0061] The system constructs a framework based on the location of defects and the overall shape (the edge of the faucet hole is a high-stress area): Spatial relationship: The four defects are spatially dispersed and are judged as multiple isolated defects; Functional impact: D001 (edge ​​chipping) is located in the critical functional area of ​​the faucet mounting hole, affecting the sealing performance; D002 (scratches) is located at the edge of the countertop, affecting the feel; D003 / D004 are located at the bottom of the basin, affecting the aesthetics; Structural impact: D001 is located in a high-stress area and there is a risk of expansion.

[0062] Framework output: {"Overall assessment":"Multiple isolated defects exist","Key risk points":"D001","Spatial relationship":"Dispersed","Functional impact summary":"D001 affects core functions, D002 affects user experience, D003 / D004 affects appearance","Structural risk summary":"D001 is located in a high-stress area and has the risk of expansion"}.

[0063] The system loads the high-end series of quality rule library and reasons: Rule IF Defect.Location = 'Critical Functional Area' AND Severity Score > 8 THEN Decision = 'Scrap' is triggered by D001 (Break, Score 9) and generates a defect identification system. This defect identification system provides multi-dimensional decision-making basis for the quality level assessment of S14.

[0064] refer to Figure 5 In step S14, the specific steps are as follows: S141: Real-time monitoring of the defect identification system, dynamic identification of the defect identification system, identification of multiple defect items during the identification process, and identification of the defect events of the corresponding parts based on the multiple defect items, the corresponding scope of influence and the corresponding parts, so as to mark the defect events of each part; S142: Determine the function of the part based on its spatial position relative to the handwashing basin, determine the importance level of the part based on the detection of its function, and determine the first shipment quality coefficient based on the importance level of the part and the corresponding defect event. S143: Determine the second shipment quality coefficient based on the processing requirements of the handwashing basin and the corresponding defect events. Determine the shipment quality level of the part based on the mapping table of the first shipment quality coefficient, the second shipment quality coefficient and the shipment quality level, so as to mark the shipment quality level of each part.

[0065] In the embodiments of this application, a real-time monitoring defect identification system is used to dynamically identify defects. During the identification process, multiple defect items are identified, and the defect events of each part are determined based on the multiple defect items, their corresponding impact ranges, and their corresponding locations. This process marks the defect events of each part, taking into account the overall consideration of multiple defect items, their corresponding impact ranges, and their corresponding locations, thus ensuring the accuracy of the defect events of each part.

[0066] At this point, the S141 service subscribes in real time to the defect identification system data stream published by S13 via ESB or message queue (such as Kafka); it checks the data packet structure, field types, and required items against the predefined JSONSchema or XSD to prevent format errors or missing fields from causing processing crashes; after verification, it parses the data packet into a hierarchical object model (such as the DefectIdentificationSystem object containing the DefectFramework and Detailed_Defects list) for easy program operation.

[0067] The flattened defect list is intelligently aggregated by location and impact range to form preliminary defect events. At this point, all DefectFeature objects are traversed, and their associated processing parts (such as faucet holes) are read and stored in a hash table (such as Map[countertop]) with the part name as the key. The spatial impact range is calculated according to the defect morphology: linear defects (scratches): length × width; planar defects (bubbles, color difference): area; point / edge defects (chipped edges): maximum radial depth or perimeter. When grouping, the number of defects is counted and the total impact range is accumulated to intuitively reflect the degree of erosion of the location.

[0068] Generate a unique ID (e.g., UUID) for each part to ensure global uniqueness (e.g., E001-54f3-4a8b-b2c1); calculate a summary for each part, including: part_name: part name; defect_count: total number of defects; most_severe_defect_type: type of defect with the highest severity score; max_severity_score: highest severity score; total_impact_scope: total impact scope; encapsulate the event ID, summary, and original defect ID list into a defect event data object.

[0069] Specifically, after S13 completes the analysis, it publishes a JSON data packet, which is immediately consumed by the S141 service: it checks the JSON schema to confirm that it contains necessary fields such as Overall_Verdict and Detailed_Defects, and that each defect object has an ID and Morphology; it parses the JSON into an A_Washbasin_Defect_Report object in memory, whose defect list contains four DefectFeature objects D001-D004, each carrying complete attributes.

[0070] The system aggregates four defect objects: Location grouping: D001 (chipped edge) → Map [faucet hole]; D002 (scratches) → Map [countertop]; D003 (bubbles), D004 (color difference) → Map [basin]; Impact range aggregation: Faucet hole: D001 impact range 0.25mm → total impact range 0.25mm; Countertop: D002 impact range 12mm×0.07mm=0.84mm² → total impact range 0.84mm²; Basin: D003 (1.8mm²) + D004 (50mm²) → total impact range 51.8mm².

[0071] Faucet hole (Event IDE001): {Location: Faucet hole, Number of defects: 1, Severe type: Chipping, Maximum severity: 9, Total affected area: 0.25mm}; Countertop (Event IDE002): {Location: Countertop, Number of defects: 1, Severe type: Scratches, Maximum severity: 7, Total affected area: 0.84mm²}; Basin (Event IDE003): {Location: Basin, Number of defects: 2, Severe type: Color difference, Maximum severity: 6, Total affected area: 51.8mm²}; The final output contains a list of three defect event objects, each clearly marked and with a complete summary.

[0072] Furthermore, the function of the part is determined based on its spatial position relative to the handwashing basin, and the importance level of the part is determined based on the detection of its function. The first shipment quality coefficient is determined based on the importance level of the part and the corresponding defect event, which takes into account the overall consideration of the importance level of the part and the corresponding defect event, and ensures the accuracy of the first shipment quality coefficient.

[0073] At this point, each defect event output by S141 is assigned a clear engineering function to its location, realizing a mapping from geometric space to functional semantics. This is based on a static location-function knowledge base jointly created by product engineers and designers. The system directly queries the knowledge base to obtain the core function description by parsing the location name in the event, ensuring the consistency of evaluation standards and engineering accuracy.

[0074] After defining the functions, the importance of each function needs to be quantified to form an importance level. This level comprehensively reflects the impact of component failure on product safety, durability, core performance, and brand reputation. It is usually developed by cross-functional teams (quality, R&D, marketing) and fixed in the configuration database. The assessment can use the AHP or MCDA model, and finally outputs a value of 1-10.

[0075] The first shipment quality coefficient is calculated by combining the inherent severity of the defect with the importance of the location. The classic multiplicative weighted model is adopted: First shipment quality coefficient = highest severity score of defect event × importance level of location. The logic of this model is clear: if a high-severity defect occurs in a minor location, the risk coefficient may not be high; conversely, if a medium-severity defect occurs in a critical location, the risk coefficient will increase sharply.

[0076] Specifically, the system receives three defect events (E001 faucet hole, E002 countertop, E003 basin) output by S141 and queries the knowledge base: E001 (faucet hole): returns {"Part":"faucet hole","core function":"sealing and connection","functional description":"ensuring a leak-free physical connection with the faucet assembly is a prerequisite for the product to achieve its basic functions;"}; E002 (countertop): returns {"Part":"countertop","core function":"load-bearing and aesthetics","functional description":"provides a flat surface for placing items and constitutes the main area of ​​visual contact for users, affecting user experience and brand perception;"}; E003 (basin): returns {"Part":"basin","core function":"water holding and cleaning","functional description":"as the main body for holding water, its surface condition directly affects the water storage function and the convenience of daily cleaning;"}.

[0077] The system queries the importance level configuration table according to function: E001 (faucet hole): function is sealing and connection, failure will lead to serious accidents such as water leakage, importance level is rated 10 (highest level); E002 (countertop): function is load-bearing and aesthetics, although it does not affect the core function, it seriously affects the user experience and is a vulnerable area, importance level is rated 7; E003 (basin): function is water holding and cleaning, affects the core experience but minor defects will not cause functional loss, importance level is rated 8 (slightly higher than countertop).

[0078] The system calculates three defect events: E001 (faucet hole): highest severity score = 9 (chipped edge), importance level = 10 → coefficient = 9 × 10 = 90; E002 (countertop): highest severity score = 7 (scratches), importance level = 7 → coefficient = 7 × 7 = 49; E003 (basin): highest severity score = 6 (color difference), importance level = 8 → coefficient = 6 × 8 = 48; Result analysis: Although the physical severity of the countertop scratch (7 points) is higher than that of the basin color difference (6 points), the risk coefficients of the two are close because the importance level of the basin is higher; the chipped edge of the faucet hole is critical in function, and its risk coefficient (90) is much higher than that of the others.

[0079] Therefore, a second shipment quality coefficient is determined based on the processing requirements of the handwashing basin and the corresponding defect events. The shipment quality level of this part is determined based on the mapping table of the first shipment quality coefficient, the second shipment quality coefficient, and the shipment quality level, so as to mark the shipment quality level of each part. This takes into account the overall consideration of the mapping table of the first shipment quality coefficient, the second shipment quality coefficient, and the shipment quality level, and ensures the accuracy of the shipment quality level of the part.

[0080] At this point, the defect event is evaluated from the perspective of standard compliance, and the deviation between the actual state and the ideal standard is measured. The core logic is that the acceptability of the same defect is completely different for products of different quality grades or customer requirements. According to the production work order or customer order of the current batch, the corresponding processing requirements specifications are dynamically loaded from the PDM or ERP system to define the defect acceptance standards (AQL) for different quality grades (such as superior products and qualified products).

[0081] Qualitative processing requirements are transformed into quantitative requirement weights to reflect the strictness of the standards; for example, the weight of the faucet hole in a high-end hotel order may be 10 (zero tolerance), while it is only 7 in an economy order; a weighted model is used, where the weight is the requirement weight rather than the importance level; the formula is: Second Shipment Quality Coefficient = Highest Severity Score of Defect Event × Requirement Weight; the higher the coefficient, the less the defect meets the current shipment standards.

[0082] Using the first shipment quality coefficient (functional risk) and the second shipment quality coefficient (standard deviation) as inputs, a predefined shipment quality level mapping table is consulted to determine the final level. This mapping table is a multi-dimensional decision matrix reflecting the company's quality strategy. A decision matrix set by quality experts is loaded, and nested IF-THEN-ELSE logic or a two-dimensional lookup table is used to define a combined judgment logic based on the two coefficients. The two dimensions are ORed, meaning that if either coefficient exceeds a threshold, a more stringent judgment is triggered, reflecting the principle of balancing functional criticality and standard compliance. The shipment quality level mapping table is shown in Table 1. Table 1 Shipment Quality Grade Mapping Table

[0083] The decision results are formally marked and outputs that can be directly used by executable systems (such as MES and WMS) to close the entire intelligent inspection process; the shipment quality level is written back to the defect event data structure, and a final_quality_grade field is added; disposal instructions are automatically associated according to the level, such as scrapping is associated with transfer to the scrap warehouse, and qualified is associated with transfer to the packaging line; a complete data package containing all component-level quality levels and disposal instructions is generated and sent to downstream systems through the industrial network.

[0084] Specifically, assuming the current batch is a high-end order from a five-star hotel with stringent standards: Loading requirements: Load specifications from ERP, define requirement weights: faucet hole: 10, countertop: 8, basin: 9; Coefficient calculation: E001 (faucet hole): 9×10=90; E002 (countertop): 7×8=56; E003 (basin): 6×9=54.

[0085] The system makes decisions on three events: E001 (faucet hole): First coefficient 90, second coefficient 90 → meets the scrapping conditions → grade: scrap; E002 (countertop): First coefficient 49, second coefficient 56 → not exceeding 60 but higher than 30 → grade: qualified; E003 (basin): First coefficient 48, second coefficient 54 → same as E002 → grade: qualified.

[0086] The system marks events and generates final instructions: Event marking: E001: {"Event ID":"E001",...,"final_quality_grade":"Scrap","Disposal Instruction":"Transfer to scrap warehouse"}; E002: {"Event ID":"E002",...,"final_quality_grade":"Qualified","Disposal Instruction":"Transfer to packaging line"}; E003: {"Event ID":"E003",...,"final_quality_grade":"Qualified","Disposal Instruction":"Transfer to packaging line"}; Final output: Due to the scrapping of the critical component faucet hole, the system triggers the overall judgment logic and generates the final instruction: {"Part_SN":"A-WS-XXX","Overall_Verdict":"Scrap","Reason":"Critical functional component (faucet hole) has a fatal defect"}. This instruction is sent to MES, and the product is automatically removed from the production line.

[0087] refer to Figure 6 In step S15, the specific steps are as follows: S151: Mark the spatial location of each part, and construct a quality status diagram of the handwashing basin based on the spatial location of each part, the corresponding shipment quality grade, and the overall shape of the handwashing basin. This quality status diagram presents the shipment quality grade of each part in a three-dimensional form. S152: Based on the identification of the quality status diagram, multiple defect factors are determined. Based on the factor type, corresponding defect priority, and previous critical parts of the handwashing basin, the target defect area is determined and the location of the target defect area is marked. S153: Based on the identification of the area location, determine the corresponding defect morphology. At the same time, collect the handwashing tray maintenance system. Based on the handwashing tray maintenance system and the area location of the target defect area, determine the first level of defect maintenance content. Based on the handwashing tray maintenance system and the defect morphology of the target defect area, determine the second level of defect maintenance content. Based on the first level of defect maintenance content and the second level of defect maintenance content, determine the corresponding defect maintenance event.

[0088] In the embodiments of this application, the spatial location of each part is marked, and a quality status diagram of the handwashing basin is constructed based on the spatial location of each part, the corresponding shipment quality grade, and the overall shape of the handwashing basin. This quality status diagram presents the shipment quality grade of each part in a three-dimensional form, taking into account the overall consideration of the spatial location of each part, the corresponding shipment quality grade, and the overall shape of the handwashing basin, thus ensuring the accuracy of the quality status diagram of the handwashing basin.

[0089] At this point, the standard 3D CAD model (digital twin baseline) of the current batch of products is retrieved from the PDM system, containing precise geometric dimensions, topology, and assembly relationships; the parts identified by S141 (such as faucet holes) are used as semantic tags and mapped to the corresponding geometric entities in the digital twin model; at this point, a predefined naming selection set (such as Set_Faucet_Hole) is defined in the CAD, and the system directly associates it; the algorithm automatically identifies features such as holes and planes in the model and matches them with part names; a global 3D coordinate system is established with the CAD model design origin as the reference to ensure that all positional information is unique and traceable.

[0090] The shipment quality grades assessed by S14 are visually encoded and presented intuitively on the digital twin model. At this point, the final shipment quality grades assessed by S143 for each part are read. The system maintains a grade-color mapping table (pseudo-color rendering strategy), which is formulated by ergonomics and quality experts to ensure intuitiveness and unambiguity. Typical mapping table: Superior product → Green (#00FF00); Qualified product → Blue (#0070FF); Reworked product → Yellow (#FFFF00); Scrapped product → Red (#FF0000). Through the built-in rendering engine (such as OpenGL / DirectX), all marked parts are traversed, and the corresponding color is queried according to its grade and dynamically applied as material or texture.

[0091] The visualization window allows users to construct 3D scenes, including rendered models, virtual light sources, cameras, and reference meshes, enhancing realism and spatial awareness. It integrates a standard 3D interactive controller, allowing users to perform the following actions via mouse or touchscreen: Rotate: Freely rotate the model 360 degrees; Zoom: Zoom in / out to view the global or local area; Pan: Move the model within the viewport; Section: Cut the model with a sectioning plane to observe the interior or occluded parts; When the mouse hovers over a specific area, a pop-up information box displays key information such as the area name, shipment quality grade, and main defect types, making the visualization both aesthetically pleasing and practical.

[0092] Specifically, the system performs spatial marking for the washbasin: Loading the model: Loading the high-precision 3D CAD file A_Washbasin_v2.3.step; Semantic marking: The faucet hole label is bound to the inner wall and surrounding curved surface of the cylindrical hole with a diameter of 35mm on the model; the countertop label is bound to the top horizontal plane area; the basin label is bound to the internal concave curved surface; with the center of the lower surface of the model as the origin (0,0,0), the long side is the X-axis, the short side is the Y-axis, and the vertical upward is the Z-axis.

[0093] The system renders the digital twin model of the handwashing basin: Data acquisition: S143 results are faucet hole → scrap, countertop → qualified, basin → qualified; Color mapping: scrap → red, qualified → blue; Rendering application: The curved surface material of the faucet hole area is set to opaque high-gloss red; the countertop and basin areas are set to semi-transparent ceramic texture blue; other undetected parts (such as the bottom) remain in default gray.

[0094] The system displays the final quality status diagram on the touchscreen at the quality inspection station: Scene presentation: The screen displays a realistic 3D model, with the faucet hole in red and the countertop and basin in blue; Interactive operation: The quality inspector can pinch and zoom in on the faucet hole area with their finger to observe the red chipped edge location; Rotate the model to observe from inside the basin upwards to confirm that there are no other problems at the bottom of the basin; Information linkage: When the red faucet hole is clicked, a pop-up window appears: Location: Faucet hole | Quality level: Scrap | Major defect: Chipped edge (Severity: 9); Final effect: The overall quality status of the product is fully conveyed through a three-dimensional, interactive chart.

[0095] Furthermore, based on the identification of the quality status map, multiple defect factors are determined. Based on the factor types, corresponding defect priorities, and previous key parts of the handwashing basin, the target defect area is determined and the location of the target defect area is marked. This approach takes into account the factor types, corresponding defect priorities, and previous key parts of the handwashing basin to ensure the accuracy of the target defect area.

[0096] At this point, the 3D rendering engine API is used to traverse all semantically marked parts in the digital twin model; the rendering material or color attribute of each part is queried, and any area that does not match the qualified product reference color (such as blue) is identified as a candidate defect factor; for each candidate area, the data cache of S13 / S141 is reverse-queried and the associated detailed defect information is pulled, and encapsulated into a standardized defect factor data object, which includes visual anomalies and their engineering details.

[0097] From all defect factors, the most priority target defect areas are selected through a multi-dimensional evaluation model. This is essentially an application of Multi-Criterion Decision Analysis (MCDA). Three key dimensions are loaded: Factor Type (Quality Level): assigning basic priority weights to different levels (e.g., {Scrap: 10, Rework: 6, Acceptable: 1}); Defect Priority (Functional Importance): extracting the importance level of this part from S142, reflecting its criticality to product function; Historical Key Parts (Historical Risk): connecting to the historical quality database (or CMMS) to query the FMEA data of this part, with high-frequency failure points obtaining higher historical risk scores.

[0098] A weighted summation model is adopted: Comprehensive priority score = (quality level weight × α) + (functional importance × β) + (historical risk score × γ); where α, β, and γ are configurable weight coefficients (e.g., α = 0.5, β = 0.4, γ = 0.1), reflecting the company's quality strategy; the scores of all defect factors are ranked, and the area with the highest score is determined as the target defect area.

[0099] On the S151 quality status diagram, additional dynamic visual markers are applied to the target defect area to distinguish it from the static rendering; a flashing or flowing high-brightness border is rendered on the area outline; the 3D view camera is smoothly moved to the optimal viewing angle of the area; at the same time, the precise location of the area is recorded, using a dual description: semantic location: readable description (such as the edge of the faucet hole); geometric location: precise coordinates in the global coordinate system (point, outline polygon).

[0100] Specifically, the system analyzes the quality status diagram of the handwashing basin: it traverses the three parts: faucet hole, countertop, and basin; it finds that the countertop and basin are blue (qualified) and ignores them; the faucet hole is red (scrap) and is marked as a candidate defect factor; based on the faucet hole label, it queries the S141 event (E001) and S13 defect (D001) and generates an object: {"Factor ID":"F001","Related Part":"Faucet Hole","Visual Status":"Red (Scrap)","Detailed Defect":{"ID":"D001","Type":"Cracked Edge","Severity":9}}.

[0101] The system evaluates the unique defect factor F001. Factor type: quality grade scrap → weight 10; defect priority: importance of faucet hole function (from S142) is 10; past key parts: query the historical database, faucet hole is not a high frequency failure point → historical risk score 2; score calculation: (10×0.5)+(10×0.4)+(2×0.1)=5+4+0.2=9.2; F001 is determined to be the highest priority with a score of 9.2, and the faucet hole area where it is located is officially determined as the target defect area.

[0102] The system marks the target area of ​​the faucet hole: On the 3D view of the quality inspection station, a bright, continuously flashing yellow border appears around the edge of the red faucet hole, along with an arrow and text label: Target defect area: Faucet hole (scrap); Record location: {semantic location: "faucet hole edge", geometric location: {type: "polygon outline", coordinate point set: [(x1,y1,z1),(x2,y2,z2),...]}}. This precise location data will be transmitted to S153 to generate accurate maintenance instructions.

[0103] Therefore, based on the identification of the regional location, the corresponding defect morphology is determined. Simultaneously, the handwashing tray maintenance system is collected. Based on the handwashing tray maintenance system and the regional location of the target defect area, the first level of defect maintenance content is determined. Based on the handwashing tray maintenance system and the defect morphology of the target defect area, the second level of defect maintenance content is determined. Based on the first and second level of defect maintenance content, the corresponding defect maintenance event is determined. This approach integrates the overall consideration of the first and second level of defect maintenance content, ensuring the accuracy of the corresponding defect maintenance events. Furthermore, the shipment quality level of each part is controlled, achieving a holistic consideration of the regional location, corresponding defect morphology, and handwashing tray maintenance system of the target defect area, thus improving the accuracy of defect maintenance events.

[0104] At this point, using the location of the target defect area determined in S152 as an index, the detailed defect data stored in S13 or S141 is directly queried to ensure that the maintenance instructions target the most accurate defect description after full-process analysis. A connection is established with the CMMS / EAM system via API, and the relevant maintenance system knowledge base is dynamically queried and retrieved based on the product model and target area. This knowledge base is a structured database containing: Standard Operating Procedures (SOPs): repair steps for different components and defect types; Bill of Materials (BOM): required spare parts and consumables (such as glue and sandpaper); Tool List: dedicated or general equipment; Skill Requirements: skill levels of personnel performing maintenance; Safety Procedures: relevant safety precautions and protective measures.

[0105] The first level of defect maintenance (location-based general procedures): Based on the location of the target defect area (e.g., faucet hole), keyword matching is performed in the knowledge base to retrieve general and preventative maintenance procedures related to the component, ensuring overall functional checks and calibrations. The second level of defect maintenance (morphology-based specific processes): Based on the defect morphology of the target defect area (e.g., chipped edges), a second keyword matching is performed to retrieve repair processes specifically for that defect type. The two parts are logically integrated, with the first level serving as a preliminary inspection step and the second level serving as the core repair step, forming a complete sequence.

[0106] Create a data structure for defect maintenance events. Each defect maintenance event contains all the information elements required to perform maintenance, such as: work order number (globally unique identifier); product serial number & component information (clearly defining the maintenance object); target area & defect description (clearly defining the problem); maintenance steps (the first and second layers of content after merging); required materials and tools (automatically generated from the BOM and tool list); estimated working hours (calculated based on a standard working hour library); safety instructions (associated safety procedures); populate all the information into the work order structure, serialize it into a standard format (such as JSON), and push it to the designated execution terminal (such as a maintenance technician's tablet or a workshop electronic dashboard) through an industrial IoT platform or MES system.

[0107] Specifically, the system prepares for the target defect area of ​​the handwashing basin: S152 outputs the target area as the faucet hole, and the system queries the final defect morphology: {Type: chipped edge, severity: 9, size: 2.5mm}; it sends an API request to CMMS GET / api / maintenance_knowledge?product_model=A-WS&component=faucet_hole, which returns a JSON object containing the handwashing basin faucet hole related maintenance SOP, materials and tools.

[0108] The system generates maintenance content from the CMMS knowledge base: First level of content (location: faucet hole): General SOP retrieved: [1. Thoroughly clean the faucet hole area with a lint-free cloth and detergent, 2. Check the sealing surface for scratches with a magnifying glass, 3. Check the installation threads for damage or foreign objects by touch]; Second level of content (morphology: chipped edge): Specific repair process retrieved: [1. Fill with 808 high-strength AB glue, 2. Allow to cure vertically at room temperature for 2 hours, 3. Initially sand smooth with 400-grit sandpaper, 4. Finely sand with 1200-grit sandpaper, 5. Restore gloss with ceramic polishing compound]; The two are combined into a complete maintenance step sequence with a specific order.

[0109] The system generates a final maintenance work order: Work Order No.: WO-AWS-20251008-001; Product Serial No.: A-WS-XXX; Target Area: Faucet Hole; Defect Description: Edge chipping, approximately 2.5mm in size; Maintenance Steps: [1. Clean the faucet hole area..., 2. Inspect the sealing surface..., 3. Fill with 808 AB glue..., 4....]; Required Materials and Tools: ["lint-free cloth", "neutral detergent", "magnifying glass", "808 AB glue and hardener", "400 / 1200 grit sandpaper", "polishing paste", "polishing machine"]; Estimated Time: 45 minutes; The work order is packaged into JSON format and sent to the mobile device of the maintenance team leader via the network; Upon receiving the push notification, the leader can click to view the illustrated maintenance guide and scan the code to start the task; At this point, the intelligent closed loop from detection to maintenance is completely formed.

[0110] Please see Figure 7 , Figure 7 This is a schematic diagram of the structural composition of a visual detection-based handwashing basin defect identification system according to an embodiment of the present invention; the visual detection-based handwashing basin defect identification includes: The handwashing tray image combination module 21 is used to mark multiple processing steps of the handwashing tray when the handwashing tray is in a dynamic processing state, and to perform visual inspection on the handwashing tray to determine the handwashing tray image combination of each processing step; The component shape module 22 is used to determine multiple key part images in each processing step based on the process content of the processing step and the corresponding handwashing plate image combination, and to determine the corresponding component shape based on the image recognition of each key part image. The defect identification system module 23 is used to determine multiple defect features of the handwashing basin based on the shape of each component, the working process of the handwashing basin and the corresponding panoramic image, and to determine the defect identification system of the handwashing basin based on the feature position, corresponding feature shape and overall shape of the handwashing basin. The shipment quality grade module 24 is used to determine the defect events of each part based on the identification of the defect identification system in the defect identification system, and to determine the shipment quality grade of each part according to the importance level of the part, the corresponding defect event and the processing requirements of the handwashing plate. The defect maintenance event module 25 is used to construct a quality status map of the handwashing tray based on the spatial location of each part and the corresponding shipment quality level, determine the target defect area based on the quality status map, and determine the corresponding defect maintenance event based on the regional location of the target defect area, the corresponding defect morphology and the handwashing tray maintenance system.

[0111] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for defect identification of handwashing basins based on visual inspection, characterized in that, include: When the handwashing tray is in a dynamic processing state, mark multiple processing steps of the handwashing tray and perform visual inspection on the handwashing tray to determine the combination of handwashing tray images for each processing step; In each processing step, multiple key part images are determined based on the process content of the processing step and the corresponding handwashing basin image combination, and the corresponding component shape is determined based on the image recognition of each key part image; Based on the shape of each component, the working process of the handwashing tray and the corresponding panoramic image, multiple defect features of the handwashing tray are determined. Based on the feature location, corresponding feature shape and overall shape of the handwashing tray, a defect identification system for the handwashing tray is determined. The defect identification system defines the classification, severity level, and interrelationships of different defects; In the defect identification system, defect events in each part are identified based on the identification system. The outgoing quality grade of each part is determined according to the importance level of the part, the corresponding defect event, and the processing requirements of the handwashing basin. The outgoing quality grades include superior, qualified, rework, and scrap. A quality status diagram of the handwashing tray is constructed based on the spatial location of each part and the corresponding shipment quality level. The target defect area is determined based on the quality status diagram. The corresponding defect maintenance event is determined based on the location of the target defect area, the corresponding defect morphology, and the handwashing tray maintenance system. The defect maintenance event includes work order number, product serial number & component information, target area & defect description, maintenance steps, required materials and tools, estimated working hours, and safety instructions.

2. The defect identification method for handwashing basins based on visual inspection according to claim 1, characterized in that, The process of marking multiple processing steps of the handwashing tray while it is in a dynamic processing state, and performing visual inspection on the handwashing tray to determine the combination of handwashing tray images for each processing step, includes: The handwashing tray moves dynamically in the handwashing tray processing line and goes through multiple processing steps. At this time, the handwashing tray is dynamically processed in each processing step and is visually detected by the corresponding camera during the processing to output multiple handwashing tray images of each processing step. In multiple processing steps, multiple associated images of a processing step are determined based on the processing weight of the processing step, the corresponding process content, and multiple handwashing tray images. Based on the multiple associated images and the processing form of the handwashing tray in the processing step, the handwashing tray image combination of the processing step is determined, thereby determining the handwashing tray image combination of multiple processing steps.

3. The defect identification method for handwashing basins based on visual inspection according to claim 1, characterized in that, In each processing step, multiple key component images are determined based on the process content of that processing step and the corresponding handwashing basin image combination. The corresponding component shape is determined based on image recognition of each key component image, including: Real-time monitoring of each processing step of the handwashing tray, marking the process content of each processing step, determining the corresponding processing part based on the identification of the process content, and determining multiple key part images based on the processing part, the overall shape of the handwashing tray, and the combination of handwashing tray images; In multiple key part images, multiple key regions are determined based on the recognition of each key part image, and multiple corresponding morphological features are determined based on the detection of each key region. The corresponding component shape is determined based on the multiple morphological features, the corresponding processing parts, and the overall shape of the handwashing basin.

4. The defect identification method for handwashing basins based on visual inspection according to claim 1, characterized in that, The system determines multiple defect features of the handwashing basin based on the shape of each component, its working process, and the corresponding panoramic image. Based on the feature location, corresponding feature shape, and overall shape of the handwashing basin, a defect identification system is established, including: The working process of the handwashing tray is collected, and the processing events of the handwashing tray in multiple processing steps are determined based on the detection of the working process of the handwashing tray. At the same time, panoramic images of the handwashing tray are collected, and the first-level defect factors are determined based on the working process of the handwashing tray and the corresponding panoramic images. Based on the shape of each component and the panoramic image of the handwashing basin, a second level of defect factors is determined. Based on the first and second level of defect factors, multiple defect features of the handwashing basin are determined, and the feature locations of multiple defect features are marked.

5. The defect identification method for handwashing basins based on visual detection according to claim 4, characterized in that, The method for determining multiple defect features of the handwashing basin based on the shape of each component, its working process, and corresponding panoramic images, and for determining a defect identification system for the handwashing basin based on the feature location, corresponding feature shape, and overall shape of the handwashing basin, further includes: Based on the identification of each defect feature, the characteristic shape of the defect feature is determined. Based on the characteristic location of multiple defect features and the overall shape of the handwashing basin, the corresponding defect identification framework of the handwashing basin is determined. Based on the defect identification framework and the characteristic shape of multiple defect features, the defect identification system of the handwashing basin is determined.

6. The defect identification method for handwashing basins based on visual inspection according to claim 1, characterized in that, In the defect identification system, defect events in each part are determined based on the identification by the system. The shipment quality level of each part is determined according to its importance level, the corresponding defect event, and the processing requirements of the handwashing basin. This includes: The system monitors the defect identification system in real time, performs dynamic identification of defects, identifies multiple defect items during the identification process, and determines the defect events of each part based on the multiple defect items, their corresponding impact range, and their corresponding locations, thereby marking the defect events of each part.

7. The defect identification method for handwashing basins based on visual detection according to claim 6, characterized in that, The defect identification system, based on the identification of defects in each part, determines the defect events of each part, and determines the shipment quality level of each part according to the importance level of the part, the corresponding defect event, and the processing requirements of the handwashing basin. It also includes: The function of the part is determined based on its spatial position relative to the handwashing basin, and the importance level of the part is determined based on the detection of its function. The first shipment quality coefficient is determined based on the importance level of the part and the corresponding defect event. The second shipment quality coefficient is determined based on the processing requirements of the handwashing basin and the corresponding defect events. The shipment quality grade of the part is determined based on the mapping table of the first shipment quality coefficient, the second shipment quality coefficient and the shipment quality grade, so as to mark the shipment quality grade of each part.

8. The defect identification method for handwashing basins based on visual detection according to claim 1, characterized in that, The process involves constructing a quality status map for the handwashing basin based on the spatial location of each part and its corresponding shipment quality level. Based on this quality status map, target defect areas are identified. Then, based on the location of these target defect areas, their corresponding defect morphology, and the handwashing basin's maintenance system, corresponding defect maintenance events are determined, including: Mark the spatial location of each part, and construct a quality status diagram of the handwashing basin based on the spatial location of each part, the corresponding shipment quality grade, and the overall shape of the handwashing basin. This quality status diagram presents the shipment quality grade of each part in a three-dimensional form. Based on the identification of the quality status map, multiple defect factors are determined. Based on the factor type, corresponding defect priority, and previous critical parts of the handwashing basin, the target defect area is determined and the location of the target defect area is marked.

9. The defect identification method for handwashing basins based on visual detection according to claim 8, characterized in that, The process of constructing a quality status map for the handwashing basin based on the spatial location of each part and its corresponding shipment quality level, determining target defect areas based on this quality status map, and determining corresponding defect maintenance events based on the location of the target defect areas, their corresponding defect morphology, and the handwashing basin's maintenance system, also includes: Based on the identification of the regional location, the corresponding defect morphology is determined. At the same time, the maintenance system of the handwashing tray is collected. Based on the maintenance system of the handwashing tray and the regional location of the target defect area, the first level of defect maintenance content is determined. Based on the maintenance system of the handwashing tray and the defect morphology of the target defect area, the second level of defect maintenance content is determined. Based on the first level of defect maintenance content and the second level of defect maintenance content, the corresponding defect maintenance event is determined.

10. A defect identification method for handwashing basins based on visual detection, characterized in that, The defect identification of the handwashing basin based on visual detection is applied to the defect identification method of the handwashing basin based on visual detection as described in any one of claims 1-9.

Images