Automated sealing and intelligent water testing system for expanding foam in curtain wall components

Abstract

This application relates to the field of automated inspection of glass curtain walls, and discloses an automated sealing and intelligent water testing system for expanding foam in curtain wall unit components. It aims to solve the problems in existing technologies where automated water testing equipment and material properties are independent, failing to form a process linkage. The system includes: expanding foam; an automated application module for identifying the structure and sealing path of different curtain wall unit components and generating corresponding application trajectories, and adjusting the application rate and sealing thickness; an automated water testing module for conducting water testing on the curtain wall unit components after application; a leakage monitoring and tracing module for identifying leakage defects in curtain wall unit components and determining the location of the corresponding leaking component; and a reporting and analysis module for generating leakage problem cause analysis and improvement suggestions based on the curtain wall unit component information. Through material innovation and system integration, the system significantly improves the economy, efficiency, and intelligence level of curtain wall unit component inspection.

Description

Technical Field

[0001] This application relates to the field of automated testing technology for glass curtain walls, and in particular to an automated sealing and intelligent water testing method for foamed adhesive in curtain wall unit components. Background Technology

[0002] In the production of glass curtain walls, the waterproof and sealing performance testing of curtain wall units is a critical step. The performance of the water sealing materials and the degree of automation in the testing process directly affect production efficiency and quality.

[0003] Traditional curtain wall unit sealing for water testing typically uses a 590ml tube of sealant priced at 26 yuan, requiring manual application using a caulking gun. This sealant has significant drawbacks: 1. High cost: The cost of sealing a single unit is approximately 20 yuan, resulting in high material costs and low utilization (only enough for about 8-10 meters of installation), making it difficult to meet the cost control requirements of large-scale production. 2. Low curing efficiency: The sealant takes over 12 hours to cure, and cannot be fully cured before water testing. This causes units to occupy workstations for extended periods while waiting for testing, severely impacting production line speed and leading to work surface congestion and production stockpiles. 3. Difficult cleaning: Incompletely cured sealant easily adheres to the surface of the unit, requiring time-consuming and labor-intensive manual cleaning. Incomplete cleaning may also affect the accuracy of subsequent testing. 4. Insufficient environmental friendliness: Residual sealant disposal is complex, failing to meet green production requirements. Furthermore, the thickness and uniformity of manual application are difficult to control, posing a risk of seal failure.

[0004] In terms of testing processes, the existing technology has a low degree of automation: although a few factories are equipped with automatic water testing modules, most still rely on manual hoisting, soaking or spraying for water testing, which is cumbersome and inefficient; there is a lack of a systematic mechanism for tracing the source of leakage defects, making it impossible to associate leakage problems with data such as adhesive application parameters and material characteristics, making it difficult to achieve targeted process optimization.

[0005] Therefore, a system is urgently needed to solve at least one of the above problems. Summary of the Invention

[0006] This application provides an automated sealing and intelligent water testing system for foamed adhesive in curtain wall unit components. It aims to solve the technical problem in the prior art where automated water testing equipment and material characteristics are independent of each other, and no process linkage is formed (such as the water testing cycle is not optimized according to the material curing time), nor is leakage data correlated with adhesive application parameters, resulting in the test results being unable to guide production optimization.

[0007] In a first aspect, embodiments of this application provide an automated sealing and intelligent water testing system for expanding foam in curtain wall components, comprising:

[0008] A foaming adhesive, the foaming adhesive comprising a white paste, the white paste comprising a modified polyurethane foaming agent and a fast-curing catalyst;

[0009] An automated adhesive application module is connected to the foaming adhesive. The automated adhesive application module includes a visual analysis system and a three-dimensional drawing parsing module, which are used to identify the structure and sealing path of different curtain wall unit components and generate corresponding adhesive application trajectories. The adhesive application rate and sealing thickness are adjusted according to the adhesive application trajectory to achieve uniform and complete adhesive application to the curtain wall unit components.

[0010] An automatic water testing module is used to conduct water testing on the curtain wall unit components after the sealant has been applied. The water testing includes mounting, pressurizing, testing, and removing the curtain wall unit components.

[0011] The leakage monitoring and tracing module is connected to the automatic water testing module. It is used to combine real-time image recognition and data comparison of curtain wall unit components to identify leakage defects in curtain wall unit components and determine the location of the corresponding leakage component.

[0012] The reporting and analysis module, connected to the leakage monitoring and tracing module, is used to import curtain wall unit information. The curtain wall unit information includes one or more of the following: foaming material information, application trajectory, dispensing rate, sealing thickness, test results corresponding to water testing, identification results corresponding to leakage defects, and location of leakage components. Based on the curtain wall unit information, the module generates a leakage problem cause analysis and improvement suggestions.

[0013] In some embodiments, the visual analysis system acquires multi-angle images of curtain wall unit components using an industrial camera, and extracts the contour features and sealing position marks of the unit components based on an image recognition algorithm; the three-dimensional drawing parsing module calls a preset three-dimensional model library of unit components, matches the acquired contour features with the three-dimensional model, and identifies the specification type and standard sealing area of ​​the unit components; based on the coordinate mapping relationship between the actual size of the unit components and the three-dimensional model, it generates adhesive application trajectory coordinate data including the adhesive application start point, path inflection point and end point.

[0014] In some embodiments, the automated dispensing module pre-stores standard sealing thickness parameters and matching rules for the curvature-rate of the dispensing trajectory corresponding to different unit component specifications. When the dispensing equipment corresponding to the automated dispensing module moves along the dispensing trajectory, it obtains the curvature radius of the current trajectory point in real time and automatically adjusts the opening degree of the dispensing valve according to the matching rules, so that the dispensing rate in the bending area with a curvature radius ≥ 50 mm decreases and the dispensing rate in the straight area with a curvature radius < 50 mm increases. The extrusion pressure of the foamed adhesive is monitored in real time by a pressure sensor, and the dispensing amount is dynamically compensated in combination with the preset sealing thickness parameters to maintain a uniform sealing thickness.

[0015] In some embodiments, the automatic water testing module uses a robotic arm to hoist the unit to the testing station and uses pneumatic clamps to fix the edge of the unit; it automatically adjusts the testing pressure according to the unit specifications; during the test, the pressure sensor monitors the pressure change in real time, and triggers a leakage alarm when the pressure drops below a preset threshold; after the test is completed, the robotic arm automatically moves the unit to the unloading area.

[0016] In some embodiments, the real-time image recognition uses a waterproof camera to capture surface images of the unit during the water testing process, and uses a deep learning algorithm to identify the water stain diffusion area and leakage point characteristics; the data comparison spatially matches the leakage location coordinates in the real-time image with the adhesive application trajectory coordinates and the joint position of the unit's three-dimensional model to locate the specific sealing defect area or component, and retrieves the corresponding adhesive application parameters and foam curing time data of the curtain wall unit to establish the correlation between the leakage location and the construction parameters.

[0017] In some embodiments, the white paste completely cures within 1 hour, forming a block structure after curing. 750ml of the expanding foam can be applied to a distance of 15-22 meters to cover the needs of various curtain wall unit components.

[0018] In some embodiments, the expanding foam and the automatic water testing module achieve process coordination through a time scheduling algorithm. By using a preset timing matching model, a production line cycle model is established based on the 1-hour curing time of the expanding foam and the 0.5-hour pressure holding time of the water testing. When multiple curtain wall units are continuously put into operation, the working time of each workstation is allocated through a dynamic scheduling algorithm to eliminate production line accumulation caused by waiting for curing and achieve seamless connection between the sealing and water testing processes.

[0019] In some embodiments, the automated adhesive application module integrates an adhesive application quality prediction model, which is trained using a machine learning algorithm based on historical adhesive application data and water leakage test data. Before each adhesive application, the specifications and adhesive application trajectory of the curtain wall unit are input into the adhesive application quality prediction model to generate the sealing reliability probability of the adhesive application. When the sealing reliability probability is lower than a preset threshold, an adhesive application parameter correction mechanism is triggered, and an early warning message is generated.

[0020] In some embodiments, the reporting and analysis module imports the collected curtain wall unit information into a data analysis model, uses an association rule algorithm to obtain the correlation between water leakage defects and insufficient adhesive thickness, trajectory deviation, and abnormal curing time; generates a list of main causes based on the correlation degree; and generates targeted improvement suggestions based on historical optimal parameter combinations and industry standards, and pushes them to a preset MES module for process parameter optimization.

[0021] In some embodiments, the report and analysis module is configured with a unit component information acquisition interface to read barcodes or QR codes on the surface of curtain wall unit components. The barcodes or QR codes contain unit component numbers, test times, and corresponding production line information. Based on the unit component information, water test results, and leakage problem cause analysis results, the report and analysis module generates a test report containing production traceability data and synchronizes the test report and improvement suggestions to the production scheduling unit of the MES module. The MES module adjusts the glue application trajectory planning strategy and process parameters of subsequent curtain wall unit components according to the received improvement suggestions, and transmits the optimized glue application trajectory, actual production data, and test feedback information back to the data analysis model of the report and analysis module to form a closed-loop optimization process for process data.

[0022] This application provides an automated sealing and intelligent water testing system for expanding foam in curtain wall components. Through material innovation and system integration, it significantly improves the economy, efficiency, and intelligence of curtain wall component testing: the expanding foam fully cures in 1 hour, precisely matching the 1-hour pressure holding test time of the water testing equipment. A time scheduling algorithm achieves seamless integration of "applying adhesive - curing - water testing," eliminating waiting and accumulation at workstations, freeing up workshop production space, and shortening sealing and curing time compared to traditional solutions. After curing, the adhesive peels off in blocks, does not adhere to the surface of the component, eliminates the need for complex cleaning processes, and saves labor costs; improved material utilization reduces the pressure of residue disposal, meeting green production requirements. The automated adhesive application module generates precise adhesive application trajectories through visual analysis and 3D modeling, and dynamically compensates for the amount of adhesive dispensed by pressure sensors to ensure the uniformity and integrity of the sealing thickness. The leakage monitoring and tracing module achieves component-level location of leakage defects through image recognition and data comparison, and correlates adhesive application parameters and material data to provide data support for process optimization. The reporting and analysis module generates leakage causes and improvement suggestions based on association rule algorithms, and forms a data closed loop with the MES system to improve the level of intelligent production management.

[0023] In summary, this invention breaks through the bottleneck of the separation between materials, construction, and testing in the prior art. Through the deep synergy between material properties and automated systems, it achieves multi-dimensional improvements in cost, efficiency, and quality.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic block diagram of a structure for an automated sealing and intelligent water testing system for expanding foam in curtain wall components, provided in one embodiment of this application.

[0027] Figure 2 This is a schematic diagram of the glue application position corresponding to the foaming adhesive for curtain wall unit components provided in one embodiment of this application;

[0028] Figure 3 This is a schematic diagram of the structure of an automatic water testing module provided in one embodiment of this application;

[0029] Figure 4 This is a schematic diagram of the structure of an automated dispensing module provided in one embodiment of this application.

[0030] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0032] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0033] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0034] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0035] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0036] In glass curtain wall production, the waterproofing and sealing testing of curtain wall components is a crucial step. The performance of the sealing materials and the degree of automation in the testing process directly impact production efficiency and quality. Traditional technical solutions mainly suffer from the following problems:

[0037] Traditional curtain wall unit sealing for water testing typically uses a 590ml tube of sealant priced at 26 yuan, requiring manual application using a caulking gun. This sealant has significant drawbacks: High cost: Sealing a single unit costs approximately 20 yuan, resulting in high material costs and low utilization (only enough for about 8-10 meters of installation), making it difficult to meet the cost control requirements of large-scale production. Low curing efficiency: The sealant takes over 12 hours to cure, and cannot be fully cured before water testing, causing units to occupy workstations for extended periods while waiting for inspection, severely impacting production line speed and leading to work surface congestion and production stockpiles. Difficult cleaning: Incompletely cured sealant easily adheres to the surface of the unit, requiring time-consuming and labor-intensive manual cleaning, and incomplete cleaning may affect the accuracy of subsequent inspections. Insufficient environmental friendliness: Residual sealant disposal is complex, failing to meet green production requirements, and the thickness and uniformity of manual application are difficult to control, posing a risk of seal failure.

[0038] In terms of testing processes, the existing technology has a low degree of automation: although a few factories are equipped with automatic water testing modules, most still rely on manual hoisting, soaking or spraying for water testing, which is cumbersome and inefficient; there is a lack of a systematic mechanism for tracing the source of leakage defects, making it impossible to associate leakage problems with data such as adhesive application parameters and material characteristics, making it difficult to achieve targeted process optimization.

[0039] In existing technologies, improvements to sealing materials for water testing are limited to single performance optimizations (such as shortening curing time or reducing costs), but have not formed a systematic solution encompassing "material characteristics - automated construction - intelligent testing." For example, traditional technologies fail to recognize that the rapid curing (1 hour) and block peeling characteristics of foamed adhesive materials can be integrated with the standard testing time (1 hour) of automated water testing modules to form a streamlined process, completely solving the problem of production line accumulation. Simple material improvements cannot solve the problems of adhesive uniformity and traceability; they require the integration of automated technologies such as visual analysis and 3D modeling to achieve precise adhesive application, and the improvement of quality control through closed-loop data analysis.

[0040] Furthermore, in existing technologies, automated water testing equipment operates independently of material properties, failing to create process linkages (e.g., the water testing cycle is not optimized based on material curing time) and does not correlate leakage data with adhesive application parameters. Consequently, the test results cannot provide feedback for production optimization. Therefore, those skilled in the art will find it difficult to gain inspiration from existing technologies to combine material innovation with automated systems and intelligent algorithms to form a comprehensive solution covering the entire process of "sealing-detection-analysis."

[0041] To resolve the above issues, please refer to [link / reference]. Figures 1 to 4The provided system includes: a foaming adhesive, comprising a white paste containing a modified polyurethane foaming agent and a rapid-curing catalyst; an automated adhesive application module connected to the foaming adhesive, comprising a visual analysis system and a 3D drawing parsing module, used to identify the structure and sealing path of different curtain wall units and generate corresponding adhesive application trajectories, adjusting the adhesive dispensing rate and sealing thickness according to the application trajectory to achieve uniform and complete adhesive application to the curtain wall units; and an automatic water testing module for conducting water testing on the applied curtain wall units, including mounting, pressurizing, and... The system includes: a testing and removal module; a leakage monitoring and tracing module connected to the automatic water testing module, used to combine real-time image recognition and data comparison of curtain wall unit components to identify leakage defects in curtain wall unit components and determine the location of the corresponding leaking component; a reporting and analysis module connected to the leakage monitoring and tracing module, used to import curtain wall unit component information, including information on foaming adhesive material, adhesive application trajectory, adhesive dispensing rate, sealing thickness, test results corresponding to water testing, identification results corresponding to leakage defects, and one or more of the location of the leaking component; and generating a leakage problem cause analysis and improvement suggestions based on the curtain wall unit component information.

[0042] Specifically, the foam adhesive is a white paste-like foam containing modified polyurethane foaming agent and a fast-curing catalyst. Its core performance characteristics include: rapid curing: curing time is reduced to 1 hour, perfectly matching the standard testing time (1 hour) of the automatic water testing module, forming a streamlined "apply-curing-water testing" process. Block peeling characteristics: after curing, the adhesive has moderate adhesion to the surface of the unit component, and can be completely peeled off after water testing without residue, solving the problem of difficult cleaning of traditional sealants. High material utilization: a single tube (750ml) can cover approximately 15-22 meters, an improvement over traditional sealants, reducing material costs.

[0043] The automated adhesive application module scans the unit structure using an industrial camera and combines this data with BIM 3D drawings to automatically identify sealing locations (such as joints, bolt holes, etc.) and generate personalized adhesive application trajectories. Based on the trajectory curvature and gap width, it adjusts the adhesive dispensing rate and sealing thickness in real time to ensure uniform adhesive layer thickness and avoid the uneven thickness issues associated with manual application. Equipped with a multi-axis robotic arm, it supports precise adhesive application from multiple angles and is adaptable to different sizes of curtain wall units (size range: 0.5m×1m to 3m×4m).

[0044] The automatic water testing module integrates automatic unit loading (mechanical gripper), water / air pressure loading (adjustable pressure range 0-6kPa), spray / immersion testing (simulating heavy rain conditions), and automatic unloading functions, shortening the processing time per batch compared to manual operation. Combined with the foam adhesive curing time (1 hour), a fixed cycle of "1 hour of settling after adhesive application → 1 hour of water testing" is set to avoid production line waiting.

[0045] During the water testing process, the leakage monitoring and tracing module uses an infrared camera and machine vision algorithms to capture minute water seepage within 0.1 seconds (leakage volume ≥ 5ml / min is sufficient for identification), with a positioning accuracy down to the millimeter level. It establishes a real-time correlation between the leakage location and volume and adhesive application parameters (such as adhesive dispensing rate and thickness at that location) and material batch information (curing time and density), creating a "defect-parameter" mapping relationship. For example, if the leakage volume in a certain area exceeds the standard, the system automatically traces whether the adhesive dispensing rate of the corresponding application trajectory is lower than the standard value by 10%.

[0046] The reporting and analysis module generates a digital twin profile for each batch of unit components by inputting foam material parameters (curing time, tensile strength), application process data (trajectory coordinates, glue dispensing amount), and water test results (leakage point coordinates, pressure tolerance value). It uses machine learning algorithms to identify high-frequency defect patterns (such as "increased probability of leakage when glue thickness at corners is <2mm") and automatically generates process improvement suggestions (such as adjusting the glue dispensing rate at corners by 20%).

[0047] The rapid curing properties of expanding foam provide a foundation for production line cycle design, while the precision of automated application reduces the risk of seal failure. Leakage tracing data allows for the optimization of material formulations (such as adjusting catalyst ratios to improve edge adhesion), forming a positive optimization and reverse feedback loop of "materials → process → quality." By digitizing all stages from 3D drawing analysis (input) to defect cause analysis (output), the reliance on experience from traditional manual operations is avoided, enabling traceability and quantitative analysis of quality control.

[0048] For example, material preparation and system initialization include: Foaming adhesive loading: Loading buckets of foaming adhesive (590ml / tube) into the pressure supply tank of the automated adhesive application equipment, and setting basic parameters (initial dispensing rate 5ml / s, default thickness 3mm). Unit component data import: Importing the BIM model of the unit component to be inspected into the MES system, and the 3D drawing parsing module automatically marks the sealing area (such as profile splicing seams, 5mm range around mounting holes).

[0049] The automated adhesive application process includes: Visual positioning: An industrial camera scans the unit component from multiple angles, generating point cloud data, which is then matched with the BIM model to confirm actual installation deviations (e.g., gap width error ±1mm). Trajectory planning: Based on the coordinates of the sealing area and the gap width, a spiral or linear adhesive application trajectory is generated (e.g., a spiral trajectory is used at corners to ensure complete adhesive coverage). Dynamic adhesive application: The robotic arm moves along the trajectory, and the vision system monitors the adhesive thickness in real time. A PID algorithm dynamically adjusts the opening of the dispensing valve (response time <0.2 seconds) to ensure uniform thickness.

[0050] The automated water testing includes: Unit component mounting: A mechanical gripper secures the glued unit components to the water testing fixture, automatically connecting them to the water tightness testing interface (air pressure / water pressure interface). Pressure loading and testing: Pressurize to the design value (e.g., 5 kPa) according to standard procedures, maintain the pressure for 30 minutes, and simultaneously start the spray system (flow rate 80 L / (m³)). 2 (*min) Simulates a rainstorm), with an infrared camera scanning the surface in real time. The leakage monitoring module records the first leakage time, leakage point coordinates, and leakage volume, and simultaneously marks the parameters of the corresponding adhesive application trajectory (e.g., adhesive application rate of 4.5ml / s and thickness of 2.8mm at this point).

[0051] Leakage tracing and analysis includes: Defect identification: Differentiating between genuine leaks (continuous seepage) and false leaks (surface water stains) using image recognition algorithms. Data association: Mapping the coordinates of the leak point to specific nodes on the adhesive application trajectory, retrieving the adhesive application rate, thickness, and material batch number of that node, and combining this with the test water pressure value, using a preset rule engine (e.g., "High risk of leakage when thickness < 2.5mm and pressure > 4kPa"). Report generation: Generating a "Leakage Defect Analysis Report" by batch, including the top 3 defect types, corresponding process parameter deviations, and improvement suggestions (e.g., increasing the adhesive thickness to 3.5mm for the upper left corner gap of a certain model unit).

[0052] The process of cleaning and recycling the adhesive includes: After the water test is completed, due to the blocky peeling characteristics of the foam adhesive, the adhesive can be directly peeled off by hand or robotic arm, which takes less time than traditional sealant cleaning and leaves no residual pollution.

[0053] This system breaks through the limitations of traditional "single material improvement" or "equipment upgrade," constructing a full-chain collaborative system of "matching material properties to equipment cycle time → automated construction to ensure quality → intelligent detection to optimize processes," providing a replicable intelligent transformation template for glass curtain wall production. Through the deep integration of material innovation (rapid-curing foam), equipment upgrade (visual-guided adhesive application robotic arm), and intelligent algorithms (leakage tracing model), it systematically solves the problems of high cost, low efficiency, and crude quality control in traditional technologies, achieving automation and intelligence throughout the entire "sealing-detection-analysis" process.

[0054] In some embodiments, the visual analysis system acquires multi-angle images of curtain wall unit components using an industrial camera, and extracts the contour features and sealing position marks of the unit components based on an image recognition algorithm; the three-dimensional drawing parsing module calls a preset three-dimensional model library of unit components, matches the acquired contour features with the three-dimensional model, and identifies the specification type and standard sealing area of ​​the unit components; based on the coordinate mapping relationship between the actual size of the unit components and the three-dimensional model, it generates adhesive application trajectory coordinate data including the adhesive application start point, path inflection point and end point.

[0055] Image-driven contour recognition uses industrial cameras to acquire multi-angle images (≥3 viewpoints) of unit components, and uses edge detection algorithms (such as the Canny operator) to extract contour features and identify sealing location marks such as splicing seams and bolt holes.

[0056] 3D model matching and specification identification involves calling a pre-set BIM 3D model library (containing multiple curtain wall unit models) to perform ICP (Iterative Closest Point) registration between the contour features and the model, thereby determining the unit specifications (such as an L-shaped curtain wall panel with dimensions of 2m×3m) and standard sealing areas (such as a 5mm range at the edge of the profile joint).

[0057] Coordinate mapping and trajectory generation are based on the coordinate alignment of the actual contour and the 3D model (accuracy ±0.5mm), generating 3D adhesive application trajectory coordinate data (format in XYZ coordinate system + attitude angle) that includes the adhesive application start point (such as the end point of the joint), the path inflection point (such as the midpoint of the 90° corner), and the end point (the other end point of the joint).

[0058] Image acquisition involves placing a unit on a rotating platform, and an industrial camera (resolution ≥ 12 million pixels) captures images from three angles: left, right, and directly above, to obtain RGB images and depth maps (using structured light technology).

[0059] Contour feature extraction involves converting the image to grayscale and reducing noise, then using Hough transform to detect straight line / circular features and marking key sealing areas such as seam edges and hole locations.

[0060] The 3D model matching process involves matching the extracted contour feature point cloud with the corresponding model in the 3D model library. The standard sealing area CAD drawing for that model is automatically retrieved by model number (e.g., marked "adhesive application is required for transverse joints, but not for longitudinal joints").

[0061] The trajectory generation is based on the deviation between the actual contour and the model (e.g., the error between the measured value and the model value of the joint length is ≤2mm). The actual adhesive application trajectory is generated by offsetting the model coordinates to ensure that each sealing point is fully covered (e.g., extending 10mm at each end of the joint).

[0062] In some embodiments, the automated dispensing module pre-stores standard sealing thickness parameters and matching rules for the curvature-rate of the dispensing trajectory corresponding to different unit component specifications. When the dispensing equipment corresponding to the automated dispensing module moves along the dispensing trajectory, it obtains the curvature radius of the current trajectory point in real time and automatically adjusts the opening degree of the dispensing valve according to the matching rules, so that the dispensing rate in the bending area with a curvature radius ≥ 50 mm decreases and the dispensing rate in the straight area with a curvature radius < 50 mm increases. The extrusion pressure of the foamed adhesive is monitored in real time by a pressure sensor, and the dispensing amount is dynamically compensated in combination with the preset sealing thickness parameters to maintain a uniform sealing thickness.

[0063] The pre-stored process rule library includes built-in standard sealing thicknesses corresponding to different unit component specifications (e.g., 3mm for flat joints, 4mm for rounded corners) and a "radius of curvature - dispensing rate" matching table (e.g., 4ml / s for curvature radius < 50mm, 6ml / s for curvature radius ≥ 50mm). Real-time curvature calculation is performed by using trajectory coordinate data to calculate the radius of curvature at the current point (using a three-point fitted circle algorithm) as the dispensing robot moves, with an accuracy of ±2mm. Dynamic compensation control monitors the extrusion pressure of the foam adhesive in real time using a pressure sensor, and, combined with preset thickness parameters, adjusts the dispensing valve opening using a PID algorithm to ensure uniform adhesive layer thickness in different curvature areas.

[0064] For example, the rule base is initialized by pre-setting sealing thickness parameters (e.g., the thickness of the straight joint of unit A is 3mm, and the thickness of the arc corner is 4mm) and curvature-rate mapping rules (for every 10mm decrease in curvature, the rate decreases by 1ml / s).

[0065] The adhesive application process is controlled by a robotic arm moving along the application trajectory. Every 5mm of movement, the coordinates of the current trajectory point are collected, and the radius of curvature is calculated (e.g., at a 90° corner, the radius of curvature is automatically identified as 30mm). Based on the radius of curvature, a matching rule is retrieved, and the opening of the dispensing valve is automatically adjusted (e.g., when the curvature is 30mm, the valve opening decreases from 50% to 40%, and the dispensing rate decreases from 6ml / s to 4ml / s). A pressure sensor provides real-time feedback on the extrusion pressure. If the actual pressure is 10% lower than the preset value (indicating a thinner adhesive layer), the valve opening is automatically increased by 5% to compensate for the reduced dispensing volume.

[0066] In some embodiments, the automatic water testing module uses a robotic arm to hoist the unit to the testing station and uses pneumatic clamps to fix the edge of the unit; it automatically adjusts the testing pressure according to the unit specifications; during the test, the pressure sensor monitors the pressure change in real time, and triggers a leakage alarm when the pressure drops below a preset threshold; after the test is completed, the robotic arm automatically moves the unit to the unloading area.

[0067] The automated loading and unloading system uses a six-axis robotic arm (load ≥ 50kg) equipped with a vacuum adsorption gripper to automatically lift unit parts from the buffer area to the testing station (positioning accuracy ± 2mm). Pneumatic clamps (pressure adjustable range 0-10kPa) fix the edges of the unit parts to prevent displacement during testing.

[0068] The pressure parameters are adaptive and automatically retrieve preset pressure parameters based on the unit specifications (e.g., the larger the size, the higher the test pressure). The test pressure is 4 kPa for a 1m×1m unit and 5 kPa for a 2m×3m unit. The pressure holding time is fixed at 0.5 hours.

[0069] Pressure anomaly monitoring uses a high-precision pressure sensor (accuracy ±0.05kPa) to monitor the pressure in the test chamber in real time. When the pressure drop rate is >0.1kPa / min, an audible and visual alarm is triggered, and the time of leakage is recorded simultaneously.

[0070] For example, the hoisting and fixing of unit components includes: the robotic arm identifies the position of the unit component in the buffer area through a vision positioning system, the vacuum gripper picks up the component and moves it to the test station, and the pneumatic clamp automatically clamps the four sides of the unit component.

[0071] The test parameters are set by automatically applying the test pressure (e.g., 5 kPa for model B) based on the unit component model (identified by the specifications in the above embodiment), and the pressure holding time is set to, for example, 0.5 hours.

[0072] Pressure monitoring and alarms include the following: During the water test, the pressure sensor collects data every 5 seconds. If the pressure drop exceeds the threshold three times consecutively (e.g., a cumulative drop of 0.3 kPa), the system marks it as a leak, stops the test, and records the initial location of the leak (calculated based on the rate of pressure drop). Automatic unloading includes the following: After the test is completed (regardless of whether there is a leak), the robotic arm automatically moves the unit to the unloading area. Qualified parts proceed to the next process, while leaking parts are moved to the rework area.

[0073] In some embodiments, the real-time image recognition uses a waterproof camera to capture surface images of the unit during the water testing process, and uses a deep learning algorithm to identify the water stain diffusion area and leakage point characteristics; the data comparison spatially matches the leakage location coordinates in the real-time image with the adhesive application trajectory coordinates and the joint position of the unit's three-dimensional model to locate the specific sealing defect area or component, and retrieves the corresponding adhesive application parameters and foam curing time data of the curtain wall unit to establish the correlation between the leakage location and the construction parameters.

[0074] The intelligent leakage feature identification system uses a waterproof camera to capture images of the water test process and a U-Net deep learning model to identify the water stain diffusion pattern (distinguishing between dripping, linear, and surface leaks) and locate the coordinates of the leakage point.

[0075] Spatial coordinate matching converts the pixel coordinates of the leakage point into the three-dimensional coordinates of the unit component (based on camera calibration parameters), and then spatially superimposes them with the adhesive application trajectory coordinates and the joint position of the three-dimensional model (such as the coordinates of the profile splicing joint) to lock the specific sealing defect area (such as the middle 10cm section of a certain joint).

[0076] Multi-dimensional data association is achieved by retrieving the corresponding adhesive application parameters (adhesive dispensing rate, thickness) and foam adhesive batch data (curing time, density) for the area, and establishing a ternary association table of "leakage location - construction parameters - material characteristics". For example, "the leakage in the middle section of the joint exceeds the standard, which corresponds to an adhesive dispensing rate that is 15% lower than the standard value during adhesive application".

[0077] Image acquisition and leakage identification: During the water test, a waterproof camera captures images of the unit surface from the front and 45° side. The images are input into the AI ​​model in real time, and the model detects dynamic water seepage by changing pixel grayscale (the threshold is set to the appearance of newly wetted areas in 3 consecutive frames).

[0078] Spatial coordinate transformation is achieved by calibrating the camera's intrinsic and extrinsic parameters, converting the pixel coordinates (u,v) of the leakage point in the image into three-dimensional coordinates (X,Y,Z) in the unit component coordinate system with an error ≤1mm. At the same time, it is matched to the specific component of the three-dimensional model (such as marking "between the 2nd and 3rd bolt holes of the left column joint").

[0079] Data correlation analysis automatically retrieves the adhesive application trajectory node corresponding to the coordinate by the system, and obtains the adhesive dispensing rate at that time (e.g., standard value 5ml / s, actual value 4.2ml / s), sealing thickness (standard 3mm, actual measurement 2.5mm), and curing time of the batch of foam adhesive (e.g., actual measurement 1.2 hours, exceeding the standard by 1 hour).

[0080] By using a preset rule engine (such as "thickness < 2.8mm and curing time > 1 hour, the probability of leakage increases by 60%), a preliminary conclusion on the cause of leakage is generated (such as "insufficient local adhesive thickness + delayed curing leads to insufficient adhesion").

[0081] In some embodiments, the white paste completely cures within 1 hour, forming a block structure after curing. 750ml of the expanding foam can be applied to a distance of 15-22 meters to cover the needs of various curtain wall unit components.

[0082] The white paste-like foam adhesive is mainly composed of modified polyurethane prepolymer (60%), water-based foaming agent (20%), fast-curing catalyst (5%, containing organotin compounds) and filler (15%). It achieves complete curing in 1 hour (under standard temperature of 25℃ and humidity of 50%) through molecular structure design.

[0083] After curing, the adhesion strength between the adhesive and the substrate (aluminum alloy, glass, stone) is controlled at 0.3-0.5 MPa (significantly lower than the 1.2 MPa of traditional sealants). It can be peeled off in whole blocks with a residue rate of <0.1%. The density of a single 750ml tube of expanding foam is 0.8 g / cm³. 3 After curing, the volume expansion rate is 150%, and the construction length is 15-22 meters (dynamically adjusted according to the gap width of 2-5mm), which can meet the sealing requirements of more than 90% of curtain wall unit components.

[0084] Material preparation: Prepolymer preparation: Polyether polyol and diisocyanate were reacted at 80℃ for 2 hours to obtain an end-NCO-based prepolymer. Mixing process: The prepolymer, foaming agent, catalyst, and filler were added sequentially in a vacuum mixer and stirred at 2000 rpm for 3 minutes. The mixture was then filled into 750ml aluminum tubes (store away from light; shelf life 6 months). Application equipment: Foaming adhesive was extracted using a pressure pump and extruded through a 2mm diameter nozzle. Upon contact with air at room temperature, foaming began, with initial shaping within 10 minutes and complete curing to form a closed-cell foam structure (density 30-40 kg / m³). 3 ).

[0085] In some embodiments, the expanding foam and the automatic water testing module achieve process coordination through a time scheduling algorithm. By using a preset timing matching model, a production line cycle model is established based on the 1-hour curing time of the expanding foam and the 0.5-hour pressure holding time of the water testing. When multiple curtain wall units are continuously put into operation, the working time of each workstation is allocated through a dynamic scheduling algorithm to eliminate production line accumulation caused by waiting for curing and achieve seamless connection between the sealing and water testing processes.

[0086] The preset timing matching model construction includes: defining process time constraints: the curing of the foam adhesive is a rigid constraint (it must wait 1 hour before water testing), and the water testing pressure holding time is 0.5 hours.

[0087] Establish a mathematical model for the production line cycle time: T cycle time = max(T application, T curing wait, T test); where T curing wait = max(0, 1 hour - (T start time of next unit application - T end time of this unit application)) to ensure that the curing time is not interrupted.

[0088] Each curtain wall unit is assigned an independent curing time window (e.g., the curing window for unit A is 9:00-10:00). The water testing process can only be started within 0.5 hours after the end of the time window to avoid data distortion caused by premature testing.

[0089] The dynamic scheduling algorithm design employs an improved EDD (Earliest Delivery Date) algorithm combined with buffer time compensation: when N units are continuously produced, the theoretical available time for each station is calculated (e.g., the gluing station completes 1 unit every 30 minutes, and the curing area has 5 independent curing bays), and the curing bays and test equipment numbers are dynamically allocated. A "back pressure control" mechanism is introduced: when the curing area bay occupancy rate is >80%, the speed of the gluing station is automatically reduced by 10% (extended to 22 minutes / unit) to avoid production line backlog.

[0090] The cross-process data synchronization mechanism includes: developing a real-time data platform to synchronize the glue application end time (accuracy ±5 seconds), curing chamber temperature (25±2℃, affecting the actual curing speed), and test equipment load (current number of queued unit parts), and updating the scheduling plan every 10 seconds.

[0091] A dynamic compensation function is established to automatically adjust the starting point of the time window for subsequent unit components (to start curing 0.2 hours earlier) based on the measured curing time (e.g., when the room temperature is 18℃ in winter, the curing time is extended to 1.2 hours).

[0092] The implementation of the preset timing matching model includes the following steps:

[0093] Step 1: Initialize the curing time window; For the first unit to be put into production, the adhesive application end time is t0, the curing time window is set to [t0, t0+1 hours], the water test start time is forced to t0+1 hours, and the pressure holding end time is t0+1.5 hours.

[0094] Step 2: Calculate the continuous production cycle time; When the adhesive application end time of the second unit is t1, if t1-t0<1 hour, then the curing waiting time of the second unit is 1 hour-(t1-t0), that is, its water testing start time is t1+1+(1 hour-(t1-t0))=t0+2 hours, which coincides with the water testing end time of the first unit. At this time, the scheduling algorithm automatically allocates different water testing equipment (such as equipment 1 to process the first unit and equipment 2 to process the second unit).

[0095] The dynamic scheduling algorithm execution includes: Scenario A: Continuous production of unit parts of the same model (fixed gluing time 20 minutes / piece, 5 curing bays): Piece 1: gluing 9:00-9:20, curing bay 1 (9:20-10:20), testing equipment 1 (10:20-10:50); Piece 2: gluing 9:20-9:40, since t1-t0=20 minutes<1 hour, curing waiting time is 40 minutes, curing bay 2 (9:40+40 minutes=10:20-11:20), testing equipment 2 (11:20-11:50); and so on, ensuring that 1 piece is produced every 20 minutes, and the utilization rate of curing bays and testing equipment reaches 100%.

[0096] Scenario B: Mixed model production (unit X adhesive application 30 minutes / piece, unit Y adhesive application 15 minutes / piece); when unit X (adhesive application 9:00-9:30) and unit Y (adhesive application 9:30-9:45) are continuously put into production, the curing waiting time of Y is 1 hour - 15 minutes = 45 minutes, that is, the trial start time of Y is 9:45 + 45 minutes = 10:30, which coincides with the trial start time of X at 10:30. The scheduling algorithm prioritizes the allocation of idle equipment (if equipment 1 has completed the X trial at 10:20, then equipment 1 can process Y to avoid equipment idleness).

[0097] If the adhesive application time for a certain unit exceeds 30 minutes (e.g., 9:00-9:30), the system immediately calculates the subsequent impact: the curing time window is postponed to 9:30-10:30, the water test start time is 10:30, which causes the time window of subsequent unit components to be shifted later as a whole.

[0098] In some embodiments, the automated adhesive application module integrates an adhesive application quality prediction model, which is trained using a machine learning algorithm based on historical adhesive application data and water leakage test data. Before each adhesive application, the specifications and adhesive application trajectory of the curtain wall unit are input into the adhesive application quality prediction model to generate the sealing reliability probability of the adhesive application. When the sealing reliability probability is lower than a preset threshold, an adhesive application parameter correction mechanism is triggered, and an early warning message is generated.

[0099] The model training data is based on historical glue application records (including glue application rate, thickness, and trajectory curvature) and corresponding water test results (leakage / pass). A random forest algorithm is used to train a binary classification model to predict the probability of sealing reliability (output range 0-100%).

[0100] The prediction-correction closed loop includes: Input parameters: unit component specifications (such as size, joint type), adhesive application trajectory (inflection point coordinates, curvature distribution). Output response: When the prediction probability is <85%, parameter correction is triggered (such as automatically increasing the adhesive thickness at the corner by 0.5mm), and an early warning is pushed to the operator (such as "high risk of sealing the left joint, manual review is recommended").

[0101] Model training phase: Data preprocessing involved standardizing historical data (expansion rate normalized to 0-1, thickness 3-5mm mapped to 0.6-1), dividing the data into a 70% training set and a 30% validation set. Algorithm optimization involved determining the optimal parameters for the random forest using grid search (tree depth 15, minimum number of samples per leaf node 5), achieving a model accuracy of 92% and an AUC-ROC area of ​​0.95.

[0102] The real-time prediction application automatically retrieves the BIM model parameters of the current unit component 5 minutes before adhesive application and inputs them into the prediction model to generate a reliability probability (e.g., the prediction probability of unit component C is 88%, which is higher than the threshold of 85%, allowing direct adhesive application). If the probability is less than 85% (e.g., the prediction probability of unit component D is 82%), the system automatically initiates parameter correction: increasing the adhesive dispensing rate from 5ml / s to 6ml / s and the thickness from 3mm to 3.5mm for all areas of curvature <40mm of that unit component.

[0103] In some embodiments, the reporting and analysis module imports the collected curtain wall unit information into a data analysis model, uses an association rule algorithm to obtain the correlation between water leakage defects and insufficient adhesive thickness, trajectory deviation, and abnormal curing time; generates a list of main causes based on the correlation degree; and generates targeted improvement suggestions based on historical optimal parameter combinations and industry standards, and pushes them to a preset MES module for process parameter optimization.

[0104] The data input dimensions include collecting unit component information (model, size), adhesive application parameters (thickness, rate, trajectory deviation), material data (curing time, density), and water test results (leak point coordinates, leakage volume), totaling 20+ fields. Association rule mining uses the Apriori algorithm to calculate the support and confidence of each parameter with the leakage defect. For example, it was found that "adhesive thickness < 2.5mm ∧ curing time > 1.2h" has a support of 15% and a confidence of 90%, indicating that this combination is a high-frequency cause of leakage. Improvement suggestions are generated based on association ranking (top 3 causes), combined with historically optimal parameters (e.g., the lowest leakage rate at 3mm thickness) and industry standards, to generate specific improvement solutions (e.g., "force adhesive re-applying to all areas with a thickness < 2.5mm").

[0105] Data cleaning and preprocessing remove outliers (such as invalid data with curing time > 2h) and discretize continuous variables (thickness is divided into three levels: < 2.5mm, 2.5-3.5mm, and > 3.5mm).

[0106] Association rule calculation is performed by setting a minimum support of 10% and a minimum confidence of 80%, running the Apriori algorithm, and outputting a list of rules (e.g., "Trajectory deviation > 2mm → Leakage" with a confidence of 85%).

[0107] Report generation and push: A daily "Leakage Cause Analysis Report" is automatically generated every morning, sorting the top 3 issues by relevance (e.g., ① Insufficient thickness (support rate 20%), ② Trajectory deviation (15%), ③ Abnormal curing time (10%)). Improvement suggestions are pushed to the MES module via API (e.g., suggesting increasing the minimum adhesive application thickness from 2.5mm to 2.8mm).

[0108] In some embodiments, the report and analysis module is configured with a unit component information acquisition interface to read barcodes or QR codes on the surface of curtain wall unit components. The barcodes or QR codes contain unit component numbers, test times, and corresponding production line information. Based on the unit component information, water test results, and leakage problem cause analysis results, the report and analysis module generates a test report containing production traceability data and synchronizes the test report and improvement suggestions to the production scheduling unit of the MES module. The MES module adjusts the glue application trajectory planning strategy and process parameters of subsequent curtain wall unit components according to the received improvement suggestions, and transmits the optimized glue application trajectory, actual production data, and test feedback information back to the data analysis model of the report and analysis module to form a closed-loop optimization process for process data.

[0109] The unit component information acquisition interface is implemented by integrating an industrial-grade barcode scanning device (such as a fixed QR code reader or a handheld barcode scanner) into the report and analysis module, corresponding to the unique barcode / QR code on the surface of the curtain wall unit component (usually affixed to the edge of the unit component or a pre-defined marking area). The scanning device connects to the information processing unit of the report and analysis module via RS-485, Ethernet, or a wireless communication module (such as Wi-Fi or Bluetooth).

[0110] The barcode / QR code pre-encoded with at least the following information: Unit component number: a unique identifier for each curtain wall unit component (e.g., "WL-20250731-001"); Test time: the start time of the associated water testing (accurate to the second); Corresponding production line information: the production line number (e.g., "Line-A-03") or equipment number that produced the unit component. When the unit component enters the water testing station or unloading area, the barcode scanning device automatically reads the QR code information and transmits it to the database of the reporting and analysis module via an interface, establishing a unique association with the unit component's foaming material information, application trajectory, water testing results, and other data.

[0111] The data analysis engine in the reporting and analysis module automatically integrates the following data to generate inspection reports based on preset templates (such as Excel or custom formats): Unit component basic information: number, production line, specification type, 3D model version; Construction parameters: foam adhesive type, adhesive application trajectory coordinates, adhesive dispensing rate, sealing thickness; Test results: test water pressure value, pressure holding time, whether there is leakage, and coordinates of the leakage location; Cause analysis: leakage causes output by association rule algorithms (such as "insufficient adhesive thickness" and "trajectory inflection point deviation") and their correlation ranking; Improvement suggestions: specific adjustment schemes generated based on historical best parameters (such as "increase adhesive dispensing rate in the inflection point area by 15%). Reports support automatic naming and storage on the server, while also generating searchable index files.

[0112] Inspection reports and improvement suggestions are pushed to the production scheduling unit of the MES system in real time through standardized data interfaces (such as OPC UA and REST API). The receiving end of the MES module is configured with a data verification mechanism to ensure that the unit part number uniquely matches the production order and process parameters.

[0113] Based on received improvement suggestions, the MES module automatically or manually adjusts the following key process parameters: Adhesive application trajectory planning strategy: For trajectory deviation issues corresponding to leaking unit parts, the 3D drawing analysis module is invoked to recalculate the coordinates of the adhesive application start point, inflection point, and end point (e.g., coordinate mapping relationships), generating corrected trajectory data; Process parameters: Based on the cause analysis results (e.g., "insufficient adhesive dispensing rate in areas with curvature radius ≥ 50mm"), the preset rules of the automated adhesive application module are adjusted (e.g., "curvature-rate matching rules"), for example, increasing the opening degree of the adhesive dispensing valve in the bending area of ​​a certain specification unit part from 50% to 60%. Production scheduling linkage: The adjusted parameters are transmitted in real time to the control system of the automated adhesive application equipment via industrial bus (e.g., Profinet, EtherCAT), and the equipment automatically loads the new parameters during subsequent unit part production. The MES module synchronously updates the production plan, marks the affected unit part batches, and ensures that the adjusted process parameters take effect on the designated production line.

[0114] The adhesive application equipment and water testing equipment record actual adhesive application parameters (adhesive dispensing rate, sealing thickness) and water testing results (pressure curve, leakage marker) in real time through sensors (such as pressure sensors and position encoders), and upload them to the reporting and analysis module via an IoT gateway. The data format is uniformly standardized (such as JSON) and includes fields such as timestamp, equipment number, actual parameter values, and test result status.

[0115] The machine learning model in the reporting and analysis module (such as the adhesive application quality prediction model provided in the above embodiment) receives optimized process parameters and actual production data, and updates the model parameters through incremental learning algorithms. For example, if the leakage rate of a batch of unit parts decreases after adopting the corrected adhesive application trajectory, the model automatically strengthens the weight of that trajectory parameter. The model verification process is triggered periodically (e.g., daily / weekly) to compare the prediction accuracy before and after optimization, ensuring the effectiveness of closed-loop optimization.

[0116] In some embodiments, a 1:1 high-precision digital twin model of the curtain wall unit (accuracy ±0.1mm) is constructed, integrating BIM geometric data, material mechanical properties (such as the elastic modulus of the substrate), and adhesive application process parameters (adhesive dispensing rate, nozzle pressure). A virtual adhesive application environment is developed to simulate the flow and diffusion behavior of the foamed adhesive on different curved surfaces after extrusion (based on fluid dynamics simulation using the Navier-Stokes equations), predicting the thickness uniformity of the adhesive layer after molding.

[0117] The reinforcement learning trajectory optimization is achieved by designing a reward function: "the proportion of areas with adhesive layer thickness error < 0.2mm" is used as a positive reward, and "material waste" is used as a negative reward. The robot arm's motion strategy is trained through the PPO (proximal policy optimization) algorithm to generate the optimal adhesive application trajectory for complex curved surfaces (such as hyperboloid curtain wall panels).

[0118] The virtual-real data closed loop collects real-time data on the position of the robotic arm (accuracy ±0.3mm) and the thickness of the adhesive layer (accuracy ±0.1mm) from the sensor, and then corrects the simulation parameters of the digital twin model (such as the expansion coefficient of the foam adhesive) in reverse, forming a dynamic optimization closed loop of "simulation-execution-feedback-iteration".

[0119] The twin model was constructed by importing the unit component BIM model into the digital twin platform, dividing the adhesive application area into a mesh (minimum mesh size 2mm×2mm), and presetting the initial adhesive application trajectory (based on the contour recognition results of Example 1). The rheological parameters of the foam adhesive (viscosity 0.8Pa*s, surface tension 35mN / m) were input to simulate the adhesive layer diffusion morphology at a nozzle moving speed of 50mm / s, marking areas with potentially insufficient thickness (such as the center of a convex surface with curvature > 80mm).

[0120] Reinforcement learning training uses trajectory nodes in the virtual environment as the state space (including coordinates, curvature, and preset thickness), and the action space is defined as the XYZ coordinate offset of the trajectory points (discretized within ±2mm). After every 100 virtual application iterations, the reward value is calculated and the policy network is updated until the reward value fluctuation is less than 5% for 5 consecutive iterations, at which point the optimal trajectory is output (e.g., automatically adding 3 intermediate application points in the convex region).

[0121] The virtual-real collaborative optimization involves importing the measured adhesive layer thickness data (collected via a laser thickness gauge at 10mm intervals) into the twin model after each unit component is completed during actual adhesive application. This corrects the foam expansion rate parameter (e.g., if the measured expansion rate is 140%, the model's preset value is updated to 150%). For new unit components (such as the first-time production of hyperboloid panels), the optimized trajectory generated by the twin model is prioritized to avoid the limitations of relying on historical data.

[0122] In some embodiments, a ResNet-based model for predicting adhesive application parameters is constructed. Common features (such as seam type and curvature distribution patterns) of historical models (≥100 types) are extracted as shared layers. For new models (sample size <5 pieces), only the parameters of the last 3 network layers are fine-tuned. A Domain Adversarial Network is designed to reduce the distribution differences between old and new models in the adhesive application trajectory feature space, avoiding the negative transfer problem of "old knowledge interfering with the learning of new knowledge".

[0123] The small-sample rapid calibration mechanism requires only three measured data points (glue application trajectory + water test results) for newly produced unit parts. It then uses a meta-learning algorithm to quickly generate adaptation parameters (such as glue application rate correction coefficients and corner thickness compensation values). A "model similarity evaluation model" is established, automatically matching the most similar historical model (similarity > 80%), reusing 90% of its basic parameters, and adjusting only the differences (such as trajectory length scaling due to size differences).

[0124] The transfer learning model training passed the following stages: Pre-training phase: The shared feature extraction layer was trained using over 100,000 historical data points, enabling the model to recognize common features such as "straight seams" and "rounded corners" with an accuracy of >95%.

[0125] Fine-tuning phase: When a new model unit (such as a curtain wall panel with ventilation louvers) is launched, input the actual glue application trajectory data of 3 pieces, freeze the first 10 shared layers, train only the last 3 fully connected layers, and complete the model adaptation within 2 hours.

[0126] Similarity matching and parameter reuse are achieved by calculating the geometric similarity between the new model and historical models (based on the Hausdorff distance of the contour point cloud). If a historical model A with 85% similarity is found, the "curvature-rate" matching rule of A is directly reused, and the trajectory coordinates are adjusted only according to the size difference (e.g., a length scaling factor of 0.8). For the difference areas (e.g., the newly added triangular sealing area at the edge of the louver), temporary adhesive application parameters (e.g., preset thickness of 3.5mm, rate of 4ml / s) are generated through a meta-learning algorithm. After verification by one trial production, the adhesive is cured onto the model.

[0127] Negative migration suppression, by incorporating a domain adversarial loss function during training, forces the model to be unable to distinguish whether the input data comes from a historical model or a new model, ensuring the universality of shared features and avoiding interference from special parameters of old models (such as unconventional glue) with the adaptation of new models.

[0128] In some embodiments, a "generator-discriminator" architecture is constructed: the generator takes the surface geometry data of the unit component (STL format) as input and outputs the 3D point cloud shape of the ideal adhesive layer (including thickness and continuity parameters); the discriminator distinguishes the differences between the generated shape and historical high-quality adhesive layers and optimizes the generator parameters. Conditional constraints are introduced (such as "adhesive layer thickness ≥ 3mm within 5mm of the joint edge" and "adhesive layer around bolt holes must be completely covered") to ensure that the generated shape meets the process standards.

[0129] The real-time visual feedback closed-loop system acquires point clouds of the adhesive layer surface in real time during the application process using a line laser scanner (accuracy ±0.2mm). These points are compared with the ideal shape generated by the GAN to calculate a deviation matrix (e.g., if the thickness of a certain area is less than 0.3mm). Based on the deviation matrix, the robot arm's motion parameters are dynamically adjusted (e.g., reducing the moving speed by 10% and increasing the adhesive dispensing pressure by 5%) to achieve closed-loop control of "generating the ideal shape → real-time correction of deviations → approximating the optimal result".

[0130] GAN model training includes: Dataset construction: Collecting adhesive layer point cloud data (0.5mm resolution) from 1000+ historical high-quality unit parts, and labeling qualified areas (thickness 2.5-3.5mm, no breaks) as real samples. Generator design: Using a 3D convolutional neural network, the input is the triangular mesh data of the unit part surface, and the output is the XYZ coordinates and thickness values ​​of the adhesive layer point cloud. The discriminator (binary classification CNN) judges whether it is close to the real sample, and iterative training is carried out until the qualified rate of the generated samples is >98%.

[0131] Real-time morphology control: Before applying adhesive, the surface data of the current unit is input into the GAN model to generate a point cloud template of the ideal adhesive layer (e.g., for a wavy curtain wall panel, a variable thickness adhesive layer that follows the surface undulations is generated). Every 10mm of adhesive is applied, a line laser scanner collects the surface data of the current adhesive layer, compares it with the template, and generates a correction command: if the actual thickness is 0.4mm thinner than the template, the opening of the adhesive dispensing valve is immediately adjusted from 40% to 45% for 2 seconds (response delay < 0.1 seconds).

[0132] Embedded process standards: Add constraint terms to the loss function of the generator to force compliance with industry standards (such as the sealant thickness ≥3mm specified in JGJ102). If the thickness of a certain area in the generated form is <2.8mm, automatically add the application trajectory points for that area.

[0133] In some embodiments, constructing an adhesive application knowledge graph includes: integrating multi-source data to construct triples (entity-relationship-attribute): Entity: unit component model, adhesive application equipment, foaming adhesive batch, leakage pattern (e.g., bolt hole leakage, joint mid-section leakage). Relationship: "Bolt hole leakage of model A corresponds to adhesive application trajectory deviation" "Pressure sensor failure of equipment B causes abnormal adhesive dispensing rate". Attribute: equipment fault code (e.g., E001 = pressure sensor timeout), material failure threshold (e.g., adhesion decreases by 50% when curing time > 1.5h).

[0134] Develop a graph neural network (GNN) inference engine to support multi-layered reasoning from "abnormal phenomenon → potential cause → solution" (e.g., "drop in test pressure → historical fault records of pressure sensors in related equipment → recommended sensor calibration").

[0135] The real-time anomaly diagnosis mechanism collects real-time equipment status data (such as robotic arm joint angle error > 0.5°, dispensing temperature > 35℃) and process data (thickness < 2.5mm at 3 consecutive points), and matches it with anomaly patterns in a knowledge graph (with 100+ preset fault templates). It provides tiered diagnostic results including: primary anomalies automatically trigger equipment self-checks (such as restarting a pressure sensor); intermediate anomalies are pushed to the operator (with 3 optimal handling solutions); and advanced anomalies (such as systemic parameter drift) trigger the MES to suspend the production line.

[0136] The knowledge graph construction process includes: Data extraction: parsing historical fault reports and process documents through natural language processing to extract entity relationships (e.g., extracting "Model C - Association - Trajectory Algorithm Error" from "On [Date], Model C caused water leakage due to trajectory planning algorithm error"). Graph storage uses the Neo4j graph database, with more than 5000 nodes and more than 30000 relationships, supporting millisecond-level complex queries (e.g., "Query all water leakage events caused by sensor failure").

[0137] When the system detects a "test water pressure drop rate of 0.15 kPa / min for unit D," it first matches the "Leakage Mode - Bolt Hole Leakage" node in the knowledge graph, linking it to "Bolt Hole Leakage - Common Causes - Trajectory Does Not Cover Bolt Hole Edge," and retrieves the following solutions: ① Check if the trajectory planning for this model includes a 5mm area around the bolt hole; ② If not, automatically generate a glue-applying trajectory and mark this model as "To be optimized." For equipment anomalies (such as a robotic arm TCP accuracy alarm), the system links it to "TCP Deviation > 0.5mm → leading to trajectory coordinate offset → increasing leakage risk by 60%" through the knowledge graph, immediately triggering the equipment calibration process (automatic calibration using the calibration board, taking 5 minutes).

[0138] The knowledge graph iteration involves injecting the solution (such as "adding a new edge adhesive compensation rule for unit components with ventilation holes") back into the graph after each manual handling of anomalies, thereby enabling the autonomous evolution of diagnostic knowledge (10-15 new effective rules are added each month).

[0139] In some embodiments, a multi-objective optimization model is constructed, including: defining optimization objectives: ① minimizing production line energy consumption (robot arm motor power, heating system energy consumption); ② maximizing production efficiency (unit processing cycle); ③ maximizing material utilization (foam waste rate < 5%). The NSGA-II (non-dominated sorting genetic algorithm) is used to solve for the Pareto optimal solution set, considering constraints (e.g., fixed foam curing time of 1 hour, single processing time at the trial station ≥ 45 minutes). Dynamic adaptive working conditions are implemented: real-time collection of workshop environmental data (temperature 20-30℃, humidity 40-60%, affecting foam curing speed) and equipment load (current work-in-process quantity 0-5 pieces) dynamically adjusts parameters at each station (e.g., shortening the curing buffer waiting time by 5 minutes when the temperature rises). An energy consumption-efficiency mapping table is established: when orders are urgent (delivery time < 24 hours), the most efficient solution is prioritized (sacrificing 10% energy consumption); when orders are less frequent, the most energy-efficient solution is selected (extending the cycle by 15 minutes, saving 20% ​​electricity).

[0140] Model parameters are defined as follows: Objective function: Efficiency: 1 / total processing cycle of unit (cycle = application time + curing time + test time) Energy consumption: Σ (robotic arm power × running time + heating system power × start time) Material utilization rate: 1 - (waste of foaming adhesive / total usage) Decision variables: robotic arm movement speed (50-100mm / s), curing buffer temperature (20-25℃, affecting curing speed), number of parallel processing units at the test station (1-3 units).

[0141] Real-time scheduling execution includes: During early shift production (with sufficient personnel), the system selects the optimal solution of "efficiency + material utilization": robotic arm speed 80mm / s (balancing accuracy and speed), curing zone temperature 25℃ (accelerating curing), and two pieces are processed in parallel at the test station. During night shift production (unattended operation), the system selects the optimal solution of "energy consumption + reliability": speed 60mm / s (reducing motor load), curing zone temperature 20℃ (extending curing time but saving electricity), and single-batch processing at the test station.

[0142] The Pareto solution set update includes: weekly adjustments to the objective function weights (increasing the material utilization weight to 40%) based on actual production data (such as finding that the material waste rate exceeds 10% during high-speed movement), retraining the optimization model, and ensuring that the solution set conforms to the latest process constraints.

[0143] In some embodiments, this invention discloses an automated process for sealing and testing curtain wall unit components with expanding foam. It innovatively utilizes low-cost, high-performance expanding foam sealing materials adapted to automated production lines and achieves precise application of the foam using visual analysis and 3D drawing fusion technology. Through an automatic testing module, a leakage monitoring and traceability platform, and an MES system, it constructs a fully automated solution from automatic sealing and unmanned testing to intelligent analysis, effectively solving problems such as high cost, low efficiency, difficult cleaning, and unintelligent detection in traditional processes, and significantly improving production quality and management efficiency.

[0144] Material composition: The new foam adhesive is a white paste, the main component of which is modified polyurethane foaming agent; a fast curing catalyst is added to ensure complete curing within 1 hour; the foaming ratio is optimized, and a single bottle (750ml) can cover 15-22 meters, covering the needs of multiple specifications of unit components.

[0145] Implementation Process: Automated Adhesive Application Module: An automated adhesive application device is introduced at the front end. This device integrates a visual analysis system and a 3D drawing parsing module to automatically identify different unit component structures and standard / non-standard sealing paths, generating adhesive application trajectories in real time. The control system precisely adjusts the adhesive dispensing rate and sealing thickness to ensure uniform and complete application. Curing and Cleaning Mechanism: This foaming adhesive material can cure within one hour, forming a blocky structure with strong peelability and non-adhesion, allowing for direct manual peeling, saving cleaning time and costs. Water Testing and Traceability: An automated water testing module is configured in the mid-stage testing phase, automating the entire process from shelving, pressurization, testing, to removal. An integrated leakage monitoring and traceability system, combined with real-time image recognition and data comparison functions, enables automatic defect identification and component-level traceability. Reporting and Analysis System: The water testing platform is integrated with the MES system, supporting one-click import of unit component information, automatically generating test records and data reports, and combining historical data to analyze the causes of leakage problems and push improvement suggestions.

[0146] Based on the pain points that traditional sealants bring to our company (traditional sealants are expensive; the curing time is as long as 12 hours, which leads to crowded work surfaces and affects production capacity; and the incomplete curing of materials leads to time-consuming and laborious cleaning), we learned from the experience of our peers and, after trying many process and material replacements, we accidentally discovered that foaming adhesive can cure and peel off in blocks in 1 hour.

[0147] After completing testing and verification on over 100 unit components and successfully implementing it across the manufacturing plant for one month, the new foam adhesive material effectively meets the requirements for cost reduction and efficiency improvement: overall costs are reduced by 90% (the cost of sealing a single unit component has decreased from 26 yuan to 1 yuan), resulting in annual cost savings of 150,000 yuan (based on our company's trial production of over 7,000 units per year); it does not occupy glue guns on the production line, is ready to use immediately, cures quickly, and simplifies the operation process; sealing and curing time is shortened by 80%, and block peeling completely solves the problems of time-consuming glue removal and incomplete cleaning, freeing up production work surfaces in the workshop. The sealing and waterproofing time is up to 24 hours, and the sealing performance far meets the industry's dynamic negative pressure test requirements (US standard 25 minutes, 100mm subsidence).

[0148] Compared to the original sealant, the new material significantly reduces the curing time from 12 hours to 1 hour. This curing time closely matches the standard testing time (1 hour) of existing unit component testing equipment, enabling seamless assembly line operations. This effectively solves the problem of insufficient curing leading to untimely testing of units and excessive stacking occupying space, eliminating production bottlenecks and potential work surface congestion. Simultaneously, this improvement optimizes the efficiency of the front-end unit component sealing and adhesive removal processes, enhancing the overall test pass rate.

[0149] This application provides an automated sealing and intelligent water testing system for expanding foam in curtain wall components. Through material innovation and system integration, it significantly improves the economy, efficiency, and intelligence of curtain wall component testing: the expanding foam fully cures in 1 hour, precisely matching the 1-hour pressure holding test time of the water testing equipment. A time scheduling algorithm achieves seamless integration of "applying adhesive - curing - water testing," eliminating waiting and accumulation at workstations, freeing up workshop production space, and shortening sealing and curing time compared to traditional solutions. After curing, the adhesive peels off in blocks, does not adhere to the surface of the component, eliminates the need for complex cleaning processes, and saves labor costs; improved material utilization reduces the pressure of residue disposal, meeting green production requirements. The automated adhesive application module generates precise adhesive application trajectories through visual analysis and 3D modeling, and dynamically compensates for the amount of adhesive dispensed by pressure sensors to ensure the uniformity and integrity of the sealing thickness. The leakage monitoring and tracing module achieves component-level location of leakage defects through image recognition and data comparison, and correlates adhesive application parameters and material data to provide data support for process optimization. The reporting and analysis module generates leakage causes and improvement suggestions based on association rule algorithms, and forms a data closed loop with the MES system to improve the level of intelligent production management.

[0150] In summary, this invention breaks through the bottleneck of the separation between materials, construction, and testing in the prior art. Through the deep synergy between material properties and automated systems, it achieves multi-dimensional improvements in cost, efficiency, and quality.

[0151] It should be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. It should be understood that when an element or layer is referred to as “on,” “adjacent to,” “connected to,” or “coupled to” other elements or layers, it may be directly on, adjacent to, connected to, or coupled to other elements or layers, or there may be intervening elements or layers. Conversely, when an element is referred to as “directly on,” “directly adjacent to,” “directly connected to,” or “directly coupled to” other elements or layers, there are no intervening elements or layers. It should be understood that although the terms first, second, third, etc., may be used to describe various elements, components, areas, layers, and / or portions, these elements, components, areas, layers, and / or portions should not be limited by these terms. These terms are merely used to distinguish one element, component, area, layer, or portion from another element, component, area, layer, or portion. Therefore, without departing from the teachings of this application, the first element, component, area, layer, or portion discussed below may be referred to as a second element, component, area, layer, or portion.

[0152] Spatial relation terms such as “below,” “under,” “below,” “under,” “above,” “above,” etc., are used herein for convenience of description to describe the relationship between one element or feature shown in the figure and other elements or features. It should be understood that, in addition to the orientation shown in the figure, spatial relation terms are intended to also include different orientations of the device in use and operation. For example, if the device in the figure is flipped, then the element or feature described as “below,” “under,” or “below” other elements or features will be oriented “above” other elements or features. Therefore, the exemplary terms “below” and “under” can include both above and below orientations. The device may be otherwise oriented (rotated 90 degrees or otherwise) and the spatial descriptive terms used herein will be interpreted accordingly.

[0153] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising” and / or “including,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0154] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0155] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An automated sealing and intelligent water testing system for expanding foam in curtain wall unit components, characterized in that, include: A foaming adhesive, the foaming adhesive comprising a white paste, the white paste comprising a modified polyurethane foaming agent and a fast-curing catalyst; An automated adhesive application module is connected to the foaming adhesive. The automated adhesive application module includes a visual analysis system and a three-dimensional drawing parsing module, which are used to identify the structure and sealing path of different curtain wall unit components and generate corresponding adhesive application trajectories. The adhesive application rate and sealing thickness are adjusted according to the adhesive application trajectory to achieve uniform and complete adhesive application to the curtain wall unit components. An automatic water testing module is used to conduct water testing on the curtain wall unit components after the sealant has been applied. The water testing includes mounting, pressurizing, testing, and removing the curtain wall unit components. The leakage monitoring and tracing module is connected to the automatic water testing module. It is used to combine real-time image recognition and data comparison of curtain wall unit components to identify leakage defects in curtain wall unit components and determine the location of the corresponding leakage component. The reporting and analysis module, connected to the leakage monitoring and tracing module, is used to import curtain wall unit information. This information includes one or more of the following: foam material information, application trajectory, dispensing rate, sealing thickness, test results corresponding to the water test, identification results corresponding to leakage defects, and the location of the leaking component. Based on this information, the module generates a leakage problem cause analysis and improvement suggestions. The white paste completely cures within one hour, forming a block structure. 750ml of the foam can cover 15-22 meters of curtain wall units to meet the needs of various specifications. The foam and the automatic water test module achieve process coordination through a time scheduling algorithm. A production line cycle model is established based on the 1-hour curing time of the foam and the 0.5-hour pressure holding time of the water test, using a preset timing matching model. When multiple curtain wall units are continuously added to the line, a dynamic scheduling algorithm allocates the working time of each workstation, eliminating production line backlog caused by waiting for curing and achieving seamless connection between the sealing and water testing processes.

2. The system according to claim 1, characterized in that, The visual analysis system acquires multi-angle images of curtain wall unit components using industrial cameras, and extracts the contour features and sealing position marks of the unit components based on image recognition algorithms; The three-dimensional drawing parsing module calls the preset unit component three-dimensional model library, matches the collected contour features with the three-dimensional model, and identifies the specification type and standard sealing area of ​​the unit component. Based on the actual dimensions of the unit component and the coordinate mapping relationship with the 3D model, glue application trajectory coordinate data including the glue application start point, path inflection point and end point is generated.

3. The system according to claim 1, characterized in that, The automated dispensing module pre-stores standard sealing thickness parameters and dispensing trajectory curvature-rate matching rules for different unit specifications; When the glue application equipment corresponding to the automated glue application module moves along the glue application trajectory, the radius of curvature of the current trajectory point is obtained in real time, and the opening degree of the glue dispensing valve is automatically adjusted according to the matching rules, so that the glue dispensing rate in the bending area with a radius of curvature ≥ 50 mm is reduced and the glue dispensing rate in the straight area with a radius of curvature < 50 mm is increased. The extrusion pressure of the foam is monitored in real time by a pressure sensor, and the amount of foam dispensed is dynamically compensated in combination with preset sealing thickness parameters to maintain a uniform sealing thickness.

4. The system according to claim 1, characterized in that, The automatic water testing module uses a robotic arm to hoist the unit component to the testing station and employs pneumatic clamps to fix the edge of the unit component; it also automatically adjusts the testing pressure according to the unit component specifications. During the test, pressure changes are monitored in real time by a pressure sensor, and a leak alarm is triggered when the pressure drops below a preset threshold. After the test is completed, the robotic arm automatically moves the unit to the unloading area.

5. The system according to claim 1, characterized in that, The real-time image recognition uses a waterproof camera to capture surface images of the unit during the water test process, and uses deep learning algorithms to identify the water stain diffusion area and leakage point characteristics. The data comparison spatially matches the leakage location coordinates in the real-time image with the adhesive application trajectory coordinates and the joint position of the unit component's 3D model to locate the specific sealing defect area or component, and retrieves the corresponding adhesive application parameters and foam curing time data of the curtain wall unit component to establish the correlation between the leakage location and construction parameters.

6. The system according to claim 1, characterized in that, The automated adhesive application module integrates an adhesive application quality prediction model, which is trained and obtained through machine learning algorithms based on historical adhesive application data and water leakage test data. Before each application of adhesive, the specifications and application trajectory of the curtain wall unit are input into the adhesive application quality prediction model to generate the sealing reliability probability of the adhesive application. When the probability of sealing reliability is lower than the preset threshold, the adhesive application parameter correction mechanism is triggered and an early warning message is generated.

7. The system according to claim 1, characterized in that, The report and analysis module imports the collected curtain wall unit information into the data analysis model and uses an association rule algorithm to obtain the correlation between water leakage defects and insufficient adhesive thickness, trajectory deviation, and abnormal curing time. A list of main causes is generated based on the correlation degree; targeted improvement suggestions are generated based on the historical best parameter combinations and industry standards, and pushed to the preset MES module for process parameter optimization.

8. The system according to claim 7, characterized in that, The report and analysis module is configured with a unit component information collection interface, which is used to read the barcode or QR code on the surface of the curtain wall unit component. The barcode or QR code contains the unit component number, test time and corresponding production line information. The report and analysis module generates a detection report containing production traceability data based on the unit component information, water test results, and leakage problem cause analysis results, and synchronizes the detection report and improvement suggestions to the production scheduling unit of the MES module. The MES module adjusts the adhesive application trajectory planning strategy and process parameters of subsequent curtain wall unit components based on the received improvement suggestions, and transmits the optimized adhesive application trajectory, actual production data and test feedback information back to the data analysis model of the report and analysis module to form a closed-loop optimization process for process data.

Images