Glass cleaning method and related device

Abstract

The invention provides a glass cleaning method and a related device. The method comprises the following steps: for a to-be-detected glass sample, collecting glass surface information corresponding to the glass sample; the glass surface information at least comprises glass surface images under a plurality of light source systems; inputting the glass surface information into the anomaly detection model, and carrying out anomaly detection on the glass sample; if the glass sample has abnormal scratches and the abnormal scratches exceed a set damage threshold value, stopping the cleaning process of the glass sample; if the glass sample does not have abnormal scratches or the abnormal scratches do not exceed a set damage threshold value, generating a cleaning strategy corresponding to the glass sample through a cleaning planning model; and controlling a wiping head of the wiping robot and a cleaning agent spraying system based on the cleaning strategy so as to execute a cleaning process on the glass sample. According to the embodiment of the invention, the wiping robot can be automatically controlled to perform cleaning, operation is performed according to an accurate strategy, human factor interference is reduced, and cleaning consistency, accuracy, stability and reliability are guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of curtain wall equipment, and in particular to a glass cleaning method and related devices. Background Art

[0002] Currently, in the glass screen printing industry, before printing production, operators need to first check the condition of the glass surface to see if there are obvious flaws, the degree of dust accumulation, and whether there are any residual chemicals, and clean it.

[0003] However, in the related art, factors such as the technical level and work attitude of different operators can lead to uneven glass wiping quality. For example, some operators may miss some areas during the wiping process due to fatigue, or fail to completely remove the detergent on the glass surface, resulting in water stains after the glass dries. As long as the program of the automated wiping robot system is correct and the equipment is operating normally, it can ensure that the wiping quality of each piece of glass is relatively stable. In addition, long-term glass wiping work will place a greater burden on the operator's body. For example, repeated wiping actions will cause fatigue and soreness in the arms, shoulders and other parts. Moreover, when cleaning large glass or a large amount of glass, the operator needs to frequently bend over, raise his hands and other actions, which can easily cause physical strain. In addition, the chemicals in the detergent may irritate the operator's skin and respiratory tract, and long-term exposure may be harmful to health.

[0004] Therefore, it is urgent to design a new technical solution to solve at least one of the above technical problems. Summary of the Invention

[0005] In order to solve or improve the above problems, the embodiments of the present application provide a glass cleaning method and related devices.

[0006] In one embodiment of the present application, a glass cleaning method is provided. The method is applied to a wiping robot. The wiping robot includes at least: a robot body, a robotic arm, a visual inspection system, a wiping head, a detergent spraying system, a base, and a main control computer. The method is executed based on the main control computer. The method includes: For a glass sample to be inspected, collecting glass surface information corresponding to the glass sample; the glass surface information at least includes glass surface images under multiple light source systems; the multiple light source systems respectively correspond to multiple specific illumination angles and multiple illumination intensities; Inputting the glass surface information into an anomaly detection model to perform anomaly detection on the glass sample; If the glass sample has abnormal scratches and the abnormal scratches exceed the set damage threshold, the cleaning process of the glass sample is stopped; If the glass sample does not have any abnormal scratches, or the abnormal scratches do not exceed a set damage threshold, a cleaning strategy corresponding to the glass sample is generated using a cleaning planning model. The cleaning planning model includes multiple planning prediction branches, each of which is used to adaptively construct a corresponding cleaning strategy based on the shape of the glass sample. The cleaning strategy includes at least: the motion trajectory, motion speed, angle, detergent spraying amount, and spraying angle of the wiping head. The wiping head and the detergent spraying system of the wiping robot are controlled based on the cleaning strategy to execute a cleaning process for the glass sample.

[0007] In another embodiment of the present application, a glass cleaning device is provided. The glass cleaning device is applied to a wiping robot. The wiping robot includes at least: a robot body, a robotic arm, a visual inspection system, a wiping head, a detergent spraying system, a base, and a main control computer. The glass cleaning device includes at least the following units: An acquisition unit is configured to acquire glass surface information corresponding to a glass sample to be inspected; the glass surface information includes at least glass surface images under multiple light source systems; the multiple light source systems respectively correspond to multiple specific illumination angles and multiple illumination intensities; a detection unit, configured to input the glass surface information into an anomaly detection model and perform anomaly detection on the glass sample; a stopping unit, configured to stop the cleaning process of the glass sample if the glass sample has abnormal scratches and the abnormal scratches exceed a set damage threshold; a generation unit configured to generate a cleaning strategy corresponding to the glass sample using a cleaning planning model if the glass sample does not have any abnormal scratches or the abnormal scratches do not exceed a set damage threshold; wherein the cleaning planning model includes multiple planning prediction branches, each planning prediction branch is configured to adaptively construct a corresponding cleaning strategy based on the shape of the glass sample; the cleaning strategy includes at least: a motion trajectory, motion speed, angle, detergent spraying amount, and spraying angle of the wiping head; An execution unit is used to control a wiping head and a detergent spraying system of a wiping robot based on the cleaning strategy to execute a cleaning process for the glass sample.

[0008] In one embodiment of the present application, an electronic device includes: at least one processor, memory, and input-output unit; The memory is used to store a computer program, and the processor is used to call the computer program stored in the memory to execute the glass cleaning method in the first embodiment.

[0009] In one embodiment of the present application, a computer-readable storage medium is provided, which includes instructions. When the instructions are executed on a computer, the computer executes the glass cleaning method in the first embodiment.

[0010] In the technical solution provided by the embodiment of the present application, for a glass sample to be inspected, glass surface information corresponding to the glass sample is collected; the glass surface information includes at least glass surface images under multiple light source systems; the glass surface information is input into an anomaly detection model to perform anomaly detection on the glass sample; if the glass sample has abnormal scratches, and the abnormal scratches exceed a set damage threshold, the cleaning process of the glass sample is stopped; if the glass sample does not have abnormal scratches, or the abnormal scratches do not exceed the set damage threshold, a cleaning strategy corresponding to the glass sample is generated through a cleaning planning model; based on the cleaning strategy, the wiping head and detergent spraying system of the wiping robot are controlled to execute the cleaning process for the glass sample. The embodiment of the present application can automatically control the wiping robot to perform cleaning, operate according to a precise strategy, reduce human interference, and ensure cleaning consistency, accuracy, stability, and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0012] Figure 1 A schematic structural diagram of a glass cleaning device provided in one embodiment of the present application; Figure 2 A schematic flow chart of a glass cleaning method provided in one embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0013] The present application provides the following embodiments to solve or improve at least some of the above problems. In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application.

[0014] In some features described in the specification, claims and the above-mentioned drawings of this application, descriptions such as "first" and "second" are used to distinguish different components, parts, modules, devices, etc., and do not represent a sequence, nor do they limit "first" and "second" to different types. In addition, the embodiments described below are only some of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0015] In order to solve the above technical problems, an embodiment of the present application provides a glass cleaning device, which includes the following units: For a glass sample to be inspected, collecting glass surface information corresponding to the glass sample; the glass surface information at least includes glass surface images under multiple light source systems; the multiple light source systems respectively correspond to multiple specific illumination angles and multiple illumination intensities; Inputting the glass surface information into an anomaly detection model to perform anomaly detection on the glass sample; If the glass sample has abnormal scratches and the abnormal scratches exceed the set damage threshold, the cleaning process of the glass sample is stopped; If the glass sample does not have any abnormal scratches, or the abnormal scratches do not exceed a set damage threshold, a cleaning strategy corresponding to the glass sample is generated using a cleaning planning model. The cleaning planning model includes multiple planning prediction branches, each of which is used to adaptively construct a corresponding cleaning strategy based on the shape of the glass sample. The cleaning strategy includes at least: the motion trajectory, motion speed, angle, detergent spraying amount, and spraying angle of the wiping head. The wiping head and the detergent spraying system of the wiping robot are controlled based on the cleaning strategy to execute a cleaning process for the glass sample.

[0016] In the embodiments of this application, the coordinated operation of a weight sensor, an ink sensing module, and an ink supply module enables precise control and real-time adjustment of ink volume. This solution not only improves the printing quality and production efficiency of enameled glass but also reduces material and labor costs by reducing ink waste, optimizing screen design, and improving operational convenience. Furthermore, the equipment's environmental adaptability and environmental performance have been significantly enhanced, providing strong support for process management and sustainable development in the screen printing industry.

[0017] Figure 1 The figure shows a schematic diagram of the structure of a glass cleaning device provided by an embodiment of the present application. The glass cleaning device is mounted on a wiping robot. Figure 1As shown, an embodiment of the present application proposes a glass cleaning device, which aims to automatically control a wiping robot to perform cleaning, operate according to a precise strategy, reduce interference from human factors, and ensure cleaning consistency, accuracy, stability and reliability. Figure 1 In the embodiment, the wiping robot comprises at least: a robot body 1, a robotic arm 2, a wiping head 3, a visual inspection system 4, a detergent spraying system 5, a base 6, and a main control computer 7.

[0018] The embodiment of the present application provides a glass cleaning device. The robot body is made of high-strength aluminum alloy, which not only ensures good rigidity and structural stability during operation, but also achieves lightweightness, making it easy to move and operate. The robotic arm has a multi-joint structure, with multiple servo motors driving each joint. It can achieve high-precision spatial motion positioning according to the control system instructions and accurately reach the specified position on the glass surface. The joints are equipped with high-precision harmonic reducers, which effectively reduce errors during movement, greatly improve the smoothness of the robotic arm's movement, and ensure the accuracy and stability of the wiping action.

[0019] The wiping head, mounted at the end of the robotic arm, is made of a replaceable, flexible material that automatically adjusts its contact area to the shape of the glass surface, ensuring a tight fit for both flat and curved glass, ensuring effective wiping. A pressure sensor within the wiping head monitors wiping pressure in real time and feeds this data back to the control system, which adjusts the robotic arm's movements and wiping parameters based on this pressure data to prevent excessive pressure from damaging the glass or insufficient pressure from causing poor cleaning results.

[0020] The detergent spray system is integrated with the wiping head. The reservoir is constructed of corrosion-resistant materials, and its capacity can be flexibly adjusted to accommodate the desired amount of detergent. A micropump, guided by the precise control system, precisely controls the detergent flow rate, ensuring the optimal amount is used. The atomizing nozzle evenly distributes the detergent across the glass surface, ensuring optimal coverage and enhancing cleaning effectiveness.

[0021] Industrial cameras utilize high-resolution, high-frame-rate charge-coupled device (CCD) cameras with lenses featuring adjustable focus and aperture. In different work scenarios, as the distance between the glass and the camera changes or lighting conditions vary, the camera can clearly capture images of the glass surface by adjusting the lens focus and aperture. The captured image data is rapidly transmitted via a data cable to the image processing unit, providing an accurate data basis for subsequent analysis of the glass surface condition. The lighting system, comprised of multiple LED light sources with adjustable brightness and angle, creates a uniform and stable lighting environment for the glass surface, minimizing image blur or loss of detail caused by uneven or excessively bright lighting. This ensures that the camera captures clear and complete images, enabling accurate assessment of surface stains, scratches, and other conditions.

[0022] The laser ranging sensor uses a high-precision pulsed laser rangefinder, emitting laser light of a specific wavelength that effectively reduces ambient light interference. The sensor is mounted on an adjustable bracket at a specific location on the robotic arm. By adjusting the bracket's angle and position, the sensor can fully scan the glass surface for glass of varying sizes and shapes. The sensor emits laser light and receives reflected light, measuring distance data. This data is then transmitted to the data acquisition module, which in turn transmits it to the data processing unit. The data processing unit uses this distance data to construct a three-dimensional point cloud model of the glass surface. This model can be used to analyze information such as the shape and size of the glass, providing an important basis for the wiping robot to develop precise cleaning strategies. For example, parameters such as the motion trajectory and angle of the wiping head can be determined to meet the cleaning requirements of different glass types.

[0023] In this embodiment, the light source system should be set at a certain angle relative to the glass surface, typically between 30° and 60°. A 45° angle can be used as an initial test angle. This angle allows for a certain degree of reflection and refraction of light on the glass surface, thereby better highlighting the texture and imperfections, including minor scratches, and facilitating subsequent image analysis. In dimly lit indoor environments, the light source intensity can be increased to ensure sufficient illumination on the glass surface. However, excessive light intensity should be avoided, which can cause excessive reflections on the glass surface and affect image clarity. For rectangular glass, capture images from at least two perpendicular directions (e.g., along the long and short sides). For circular glass, the camera can capture images at equal angles around the center (e.g., every 45°). For irregularly shaped glass, select an angle that maximizes coverage of the glass surface and highlights key areas (areas with significant curvature changes) based on the shape. Furthermore, in bright light environments, low-angle lighting can be used to minimize the impact of reflected light on the camera image. In low light environments, auxiliary light sources can be added from the side or back to improve overall brightness uniformity and reduce shadows. LED light sources offer advantages such as high luminous efficiency, long life, and high adjustability. The camera's exposure time is dynamically adjusted based on the light intensity and the reflection on the glass surface. The camera's aperture is adjusted based on the shooting distance and desired depth of field.

[0024] It should be added here that the control logic of the control device will be introduced in detail in the method embodiment below.

[0025] An embodiment of the present application provides a glass cleaning method. The method described in this embodiment is implemented based on the structure of the above-mentioned glass cleaning device embodiment. Figure 2 As shown, the method may include the following steps: 101. For a glass sample to be tested, collect glass surface information corresponding to the glass sample; 102. Input the glass surface information into an anomaly detection model to perform anomaly detection on the glass sample; 103. If the glass sample has abnormal scratches and the abnormal scratches exceed a set damage threshold, the cleaning process of the glass sample is stopped; 104. If the glass sample does not have any abnormal scratches, or the abnormal scratches do not exceed a set damage threshold, generate a cleaning strategy corresponding to the glass sample using a cleaning planning model; 105. If the glass sample does not have any abnormal scratches, or the abnormal scratches do not exceed a set damage threshold, generate a cleaning strategy corresponding to the glass sample using a cleaning planning model.

[0026] Through the above steps, by capturing surface images of the glass sample under multiple light source systems with varying illumination angles and intensities, the glass surface condition can be comprehensively captured from multiple dimensions, resulting in richer and more detailed surface information. Inputting this comprehensive information into the anomaly detection model allows for more accurate detection of abnormal scratches on the glass sample, effectively avoiding missing subtle, hard-to-find damage and providing reliable evidence for subsequent cleaning procedures. If abnormal scratches are detected and exceed a set damage threshold, the cleaning process for the glass sample is terminated. This prevents cleaning of already severely damaged glass, as wiping and spraying during the cleaning process could further damage the glass. This protects the glass sample from further damage caused by improper cleaning. Using a cleaning planning model with multiple planning prediction branches, a cleaning strategy can be adaptively constructed based on the shape of the glass sample. Different glass shapes often require different cleaning strategies to achieve optimal cleaning results. For example, the wiper's trajectory, speed, and angle, as well as the amount and angle of the detergent spray, all need to be tailored to the specific shape of the glass. This adaptively generated cleaning strategy ensures that the cleaning process is more tailored to the characteristics of the glass sample, improving cleaning efficiency and quality, and enabling more targeted and optimized cleaning operations. Based on the generated cleaning strategy, the wiping robot's wiping head and detergent spraying system are controlled to execute the cleaning process, automating the entire cleaning process. Compared to manual cleaning, automation can operate according to precisely set parameters, ensuring the consistency and accuracy of each cleaning action, reducing the occurrence of uneven cleaning, inadequate cleaning, or accidental damage to the glass caused by human factors, and overall improving the stability and reliability of the cleaning process.

[0027] In an embodiment of the present application, the glass surface information includes at least glass surface images under multiple light source systems. Furthermore, the multiple light source systems correspond to multiple specific illumination angles and multiple illumination intensities, respectively.

[0028] When inspecting the condition of glass surfaces, relying solely on images from a single light source is often incomplete. By setting up multiple light source systems to illuminate the glass from different specific angles, with each light source having different light intensities, the glass surface can be rendered in a variety of different light and shadow effects.

[0029] For example, when a light source at a certain angle and intensity is illuminated, subtle scratches on the glass surface that are difficult to detect under normal viewing angles and lighting may be highlighted. However, changing the illumination angle and intensity can clearly reveal stains or other abnormalities hidden in specific textures, reflective areas, and other locations. In this way, multiple sets of glass surface images acquired from light sources of different angles and intensities together provide richer and more three-dimensional information about the glass surface, capable of capturing the characteristics of the glass surface in all aspects to the greatest extent possible, providing a sufficient and reliable data foundation for subsequent accurate analysis of the glass surface condition, such as determining the presence of scratches, the type of stains, and their distribution.

[0030] When these images of glass surfaces from multiple light sources are fed into the anomaly detection model, the rich image data helps the model identify anomalies from multiple angles. Because abnormalities (such as scratches, pits, and spots) appear differently under different lighting angles and intensities, the model can more accurately determine whether the glass sample has abnormal scratches or other defects based on these differences. This allows the model to determine whether to halt the cleaning process (if an anomaly exists and exceeds a set damage threshold) or continue generating a corresponding cleaning strategy, significantly improving the accuracy and scientific nature of the entire detection and decision-making process.

[0031] For example, in some high-precision glass production workshops, such as those used for back glass, even the smallest flaws on the glass surface can affect the final quality and performance of the product. By utilizing this multi-light system with multiple specific illumination angles and intensities to capture images of the glass surface, subtle defects that may be imperceptible to the naked eye can be promptly identified during the production process. These defects can then be promptly addressed, preventing problematic products from being passed on to the next production stage and ensuring overall product yield and quality.

[0032] In this embodiment of the present application, the cleaning planning model includes multiple planning prediction branches, each of which is used to adaptively construct a corresponding cleaning strategy based on the shape of the glass sample. Furthermore, the cleaning strategy includes at least: the motion trajectory, speed, angle, amount of detergent sprayed, and spray angle of the wiping head.

[0033] The cleaning planning model incorporates multiple prediction branches to flexibly address the cleaning needs of glass samples of varying shapes. Different glass shapes, such as square, round, and irregular, require different wiper motions and detergent spraying methods. Each prediction branch is a cleaning plan generator tailored to a specific glass type or shape. Based on the input glass sample shape information, it invokes corresponding algorithms and rules to adaptively construct a cleaning strategy optimally suited for that shape.

[0034] It is understandable that the diversity of glass shapes determines that a unified cleaning mode cannot achieve the ideal cleaning effect. For example, for round glass, the wiping head may need to make circular motions along the circumference, and the parameters such as the movement speed and angle should be reasonably set according to factors such as its radius. The spray angle of the detergent should also fit the arc of the circle to ensure comprehensive and uniform cleaning. For glass with complex and irregular shapes, the movement trajectory of the wiping head should be flexibly adjusted according to its contour characteristics to avoid missing cleaning areas or damaging the glass due to unreasonable movement. By constructing a cleaning strategy based on shape adaptation, it can be ensured that when facing glass samples of various shapes, cleaning work can be carried out in a targeted manner, maximizing cleaning efficiency and cleaning quality, and ensuring that the cleanliness of the glass after cleaning meets the expected standards.

[0035] Among them, the movement trajectory of the wiping head clearly defines the route of the wiping head moving on the glass surface, determines which areas can be wiped and the order of wiping, etc., and is a key factor in ensuring comprehensive cleaning of the glass. The reasonable trajectories corresponding to glass of different shapes are different, which is related to the integrity of the cleaning.

[0036] The appropriate speed of movement affects the wiping effect and cleaning efficiency. Too fast a speed may result in incomplete cleaning and missed stains; too slow a speed will take too much time and reduce overall work efficiency. It needs to be determined based on the material of the glass, the stain condition, and the shape of the stain.

[0037] Angle: The angle between the wiping head and the glass surface will affect the pressure distribution during wiping and the cleaning coverage. A reasonable angle can make the wiping head fit the glass better and enhance the cleaning power. Angle adjustment is especially important for glass with curved or irregular surfaces.

[0038] The amount of detergent sprayed: too little may not fully dissolve the stains, affecting the cleaning effect; too much may cause a waste of detergent and may also leave residue, affecting the appearance of the glass. It needs to be precisely controlled based on the glass area, the degree of stains, etc.

[0039] The spraying angle determines the coverage and distribution of the detergent on the glass surface. It cooperates with the movement trajectory of the wiping head to ensure that the detergent is evenly distributed on the glass, so that the subsequent wiping action of the wiping head can better remove stains.

[0040] In summary, the multi-branch structure of the cleaning planning model and the generated cleaning strategy containing these key elements together provide a strong guarantee for achieving efficient, high-quality and targeted glass cleaning work.

[0041] Further optionally, for each glass sample to be inspected, before collecting the glass surface information corresponding to the glass sample, the shape of the glass sample is identified. Furthermore, a shooting mode for the glass sample is set based on the shape of the glass sample, and shooting parameters within the shooting mode are fine-tuned based on the size information of the glass sample.

[0042] Among them, the shapes of glass samples include rectangular, circular, and irregular; for rectangular glass samples, the shooting angles in the shooting mode are at least two vertical directions, and the at least two vertical directions include the long side direction and the short side direction; for circular glass samples, the shooting positions in the shooting mode include at least multiple shooting points distributed at equal angles around the center of the circle; for irregular glass samples, the shooting angles in the shooting mode are angles that cover the glass surface to the greatest extent and include key parts, and the key parts include at least areas with large curvature changes.

[0043] In principle, the shape of the glass sample being inspected is determined through image processing technology or pre-set geometric shape recognition algorithms. For example, for regularly shaped glass (rectangular and circular), identification can be performed by examining the number of straight line segments and their angular relationship (rectangular), or by examining the curvature of the boundary (circular). For irregularly shaped glass, the complexity of the boundary curve and characteristic points, such as areas of significant curvature variation, are analyzed to determine whether it is an irregular shape.

[0044] Rectangular glass samples: Shooting from both the long and short sides is crucial because these two perpendicular directions are most effective for capturing information about the rectangular glass surface. Shooting from the long side allows for coverage of the larger surface area, while shooting from the short side provides additional vertical detail, ensuring that evenness, scratches, and stains are clearly visible across the entire rectangular surface.

[0045] Circular glass samples: The use of evenly spaced shooting points around the center of the circle is based on the geometric symmetry of the circle. This shooting mode ensures uniform image capture of the glass surface from all directions, avoiding potential defects that could be missed due to uneven shooting angles. It also better captures changes in the curvature of the circular glass surface.

[0046] Shaped glass samples: Shooting at angles that maximize surface coverage and include critical areas takes into account the irregularities of shaped glass. Critical areas, such as areas with significant changes in curvature, are often prone to stress concentration, manufacturing defects, or the accumulation of stains. Focusing on these areas allows for more accurate assessment of glass quality and surface condition.

[0047] Fine-tuning the shooting parameters based on the size of the glass sample is necessary because different glass sizes require different focal lengths, apertures, and other parameters to ensure image clarity and detail integrity at the same shooting angle. For example, larger glass may require a smaller aperture to increase depth of field, ensuring the entire glass surface is in focus. At the same time, the focal length may need to be adjusted to ensure the camera has an appropriate field of view to fully capture the glass surface.

[0048] By setting appropriate shooting modes and fine-tuning shooting parameters for glass samples of different shapes, more comprehensive and detailed information about the glass surface can be collected. This approach effectively reduces blind spots and maximizes the detection of surface abnormalities such as scratches, stains, and bubbles, regardless of whether the glass is regular or irregular in shape, thereby improving detection accuracy.

[0049] Because the shooting pattern is specifically set according to the shape, it can avoid unnecessary shooting angles and repeated shooting while ensuring inspection quality, thereby improving inspection efficiency. For example, for round glass, shooting at shooting points spaced at equal angles can obtain sufficient information within a limited number of shots, eliminating the need for the numerous redundant shooting operations required by random shooting.

[0050] This method is well adapted to glass samples of various shapes and sizes, whether they are common rectangular or round glass or complex special-shaped glass, and can be appropriately inspected based on their characteristics. This makes the entire inspection system more flexible and versatile, able to meet the needs of different glass production or processing scenarios.

[0051] A rectangle has four right angles and parallel, equal sides. This shape of glass is commonly found in architectural windows, picture frames, and other structures. Its long and short sides require photographing from these two perpendicular directions to fully capture the glass surface.

[0052] A circle is a geometric shape in which every point on its boundary is equidistant from its center. Round glass is often used in optical instruments, decorative items, and other applications. By photographing the glass surface at multiple points spaced at equal angles around the center, we can exploit its symmetry and capture a uniform image of the glass surface.

[0053] Irregularly shaped glass refers to glass with complex boundary curves and no fixed geometric pattern. This type of glass may be designed to meet special design requirements or be used in non-standard applications. When photographing this type of glass, it's important to focus on key areas, such as those with significant changes in curvature, to ensure that any quality issues or surface anomalies are detected.

[0054] Based on the above, calculating the curvature of the boundary of shaped glass is an optional method for identifying critical locations. Mathematical algorithms are used to detect changes in the curvature of the boundary curve. Areas with significant curvature changes are often critical locations. For example, in computer-aided design (CAD) software or specialized image processing software, the principles of differential geometry are used to calculate the second-order derivative of the curve to obtain curvature information. Locations where the absolute value of the curvature suddenly increases or decreases are likely locations where the glass was subjected to uneven stress, deformed, or prone to accumulation of stains during the manufacturing process.

[0055] From a physical perspective, areas with large curvature may experience greater process difficulty during the molding process. For example, in the glass hot bending process, areas with drastic changes in curvature are more likely to have problems such as uneven thickness and uneven surface.

[0056] Due to its irregular shape, special-shaped glass can generate stress concentrations in certain areas when subjected to external forces or internal stresses. Finite element analysis (FEA) or photoelastic testing can be used to detect these stress concentration points. In practice, while a full FEA analysis can be complex, simplified simulation methods can roughly identify stress concentration areas. For example, based on the shape and support structure of the special-shaped glass, and considering the stress distribution under gravity, points where stress is significantly higher than other areas are critical locations.

[0057] Photoelastic testing is an experimental method that introduces polarized light into a glass model and observes changes in the light interference fringes to identify areas of stress concentration. These stress concentration points are more likely to crack or break in actual use, so they require special attention during testing.

[0058] Determine critical areas based on the manufacturing process of the shaped glass. For shaped glass produced through cutting, microcracks or surface roughness variations may occur along the cut edge, near the cut, and in areas adjacent to the cutting path. These areas are considered critical areas. For shaped glass produced through splicing, the sealing of the joints and the interface between the splicing material and the glass are all key areas requiring focused inspection.

[0059] For example, in the production of some complex special-shaped artistic glass, colored glass may be used for splicing or decorative materials may be added to the glass surface. The joints between these decorative materials and the glass are prone to peeling, gaps and other problems, so they are critical parts.

[0060] Critical areas should be determined based on the actual functional use of the shaped glass. If the shaped glass is used as a lens in an optical device, then the central area where light passes and the optical correction areas at the edges are critical areas, as the quality of these areas directly affects optical performance. If the shaped glass is used for architectural decoration, such as as part of a building facade, then those areas directly exposed to the external environment and susceptible to rain, wind, sand, or direct sunlight are critical areas.

[0061] For example, in some special-shaped glass dome buildings with special lighting functions, the top of the glass and the surrounding edge areas are key areas for inspection because they are in direct contact with the outside world and are subject to greater environmental pressure.

[0062] Among the above parameters, calculating the curvature of the boundary of special-shaped glass is an important method for determining key areas. Mathematical algorithms are used to detect changes in the curvature of the boundary curve. Areas with large curvature changes are usually key areas. For example, in computer-aided design (CAD) software or specialized image processing software, the principles of differential geometry are used to calculate the second-order derivative of the curve to obtain curvature information. When the absolute value of the curvature suddenly increases or decreases, these locations are likely to be places where the glass is subjected to uneven stress, deformed, or prone to accumulation of stains during the manufacturing process. From a physical perspective, areas with large curvature may experience greater process difficulties during the molding process. For example, in the glass hot bending process, areas with drastic changes in curvature are more likely to have problems such as uneven thickness and uneven surface.

[0063] Due to its irregular shape, special-shaped glass can experience stress concentrations in certain areas when subjected to external forces or internal stresses. Finite element analysis (FEA) or photoelastic testing can be used to detect these stress concentration points. In practice, while a full FEA analysis can be complex, simplified simulation methods can provide a rough estimate of stress concentration areas. For example, based on the shape and support structure of the special-shaped glass, and considering the stress distribution under gravity, points with significantly higher stress than other areas are considered critical locations. Photoelastic testing is an experimental method that introduces polarized light into a glass model and observes changes in the light interference fringes to identify stress concentration areas. These stress concentration points are more likely to crack or break in actual use and therefore require special attention during testing.

[0064] Determine the key areas based on the manufacturing process of the shaped glass. If the shaped glass is made through a cutting process, microcracks or surface roughness changes may occur at the cutting edge, near the incision, and in the area adjacent to the cutting path. These areas are key areas. For shaped glass made through a splicing process, the sealing condition of the splicing joints, the bonding area between the splicing materials and the glass, etc. are all key areas that require intensive inspection. For example, in the production of some complex shaped artistic glass, colored glass splicing may be used or decorative materials may be added to the glass surface. The bonding area between these decorative materials and the glass is prone to peeling, gaps, and other problems, so it is a key area.

[0065] Critical areas should be determined based on the actual functional use of the shaped glass. If the shaped glass is used as a lens in an optical device, then the central area where light passes and the optical correction areas at the edges are critical areas, as the quality of these areas directly affects optical performance. If the shaped glass is used for architectural decoration, such as as part of a building facade, then those areas directly exposed to the external environment and susceptible to rain, wind, sand, or direct sunlight are critical areas.

[0066] For example, in some special-shaped glass dome buildings with special lighting functions, the top of the glass and the surrounding edge areas are key areas for inspection because they are in direct contact with the outside world and are subject to greater environmental pressure.

[0067] Further optionally, after setting the shooting mode of the glass sample based on the shape of the glass sample, multiple light source systems matching the glass sample can also be set; specific illumination angles between the light sources in the multiple light source systems and the glass surface of the glass sample are configured; wherein the specific illumination angle ranges from 30° to 60°; based on the environment in which the glass sample is located, the illumination intensity is set to match the background light intensity of the environment; wherein the illumination intensity is inversely proportional to the background light intensity.

[0068] Specifically, professional light intensity measurement equipment (such as a lux meter) is first needed to determine the background light intensity of the glass sample's environment. This equipment accurately measures ambient light levels, typically in lux. For example, indoors, light intensity may be higher near windows, while lower in corners or shaded areas. Outdoors, light intensity may exceed 100,000 lux on a sunny day and range from 1,000 to 10,000 lux on a cloudy day. The measured background light intensity data provides a basis for subsequent adjustments to the lighting system's light intensity.

[0069] For example, modern light source systems (such as LED light sources) are equipped with controllers that adjust brightness. After determining the required light intensity, I, the controller adjusts the light source system. This adjustment can be manual, with the operator adjusting the brightness knob on the controller based on the calculated light intensity value. Alternatively, it can be automatic, with a light intensity sensor connected to the light source system's controller. A control program can be written to automatically adjust the light intensity based on the ambient light intensity feedback from the sensor to achieve the set matching value. Furthermore, the light source system's power limitations and light intensity uniformity must be considered during the adjustment process. Ensure that the light source operates stably while adjusting the light intensity, and that the light emitted is evenly distributed across the glass surface, avoiding areas of excessive brightness or darkness. Light intensity uniformity can be improved by placing a light intensity distribution test board on the glass surface or by using a combination of multiple small light sources.

[0070] Furthermore, since rectangular glass has distinct long and short sides, to fully capture surface information, the shooting angle should encompass at least the long and short sides. Shooting along the long side covers a wider range, allowing observation of scratches, stains, or flatness issues along that direction. Shooting along the short side provides additional vertical detail, ensuring the entire rectangular surface is fully inspected. For example, in industrial inspection, shooting along the long side of rectangular glass used in large buildings may reveal long scratches, while shooting along the short side helps identify defects along the edges. In addition to the two vertical directions, consider shooting at oblique angles. Oblique angles can highlight subtle undulations or unevenness on the glass surface. Under oblique lighting, these subtle variations create shadows or reflection differences, making them easier to capture. Generally, an oblique angle of approximately 45° from the long or short side is ideal. This captures more surface detail without losing primary directional information.

[0071] Circular glass is highly symmetrical, so capturing images at evenly spaced intervals around the center is ideal. For example, the 360° circumference can be divided into several equal intervals, such as 8, 12, or 16, and images taken at these equally spaced intervals. This ensures uniform image capture of the glass surface from all directions, fully utilizing the geometric properties of a circle and avoiding potential defects that might be missed due to uneven shooting angles. Furthermore, images captured in this manner can be easily stitched together to form a complete image of the circular glass surface for subsequent analysis. In addition to capturing images at the circumference, capturing images at the center of the circular glass is also crucial. Shooting from the center provides information about the central region of the glass. This is particularly important for circular glass used in optical applications, where the quality of the central region (such as transparency and flatness) is often more critical. Furthermore, images captured at the center can serve as a reference, combined with images captured at the circumference, for a more comprehensive assessment of the overall condition of the circular glass.

[0072] For special-shaped glass, the first thing to do is to identify its key areas, such as areas with large changes in curvature, stress concentration points, manufacturing process feature points, or functional key areas. The shooting angle should be based on the principle of maximizing coverage of the glass surface and including these key areas. For example, for special-shaped glass with complex curved shapes, focus on shooting the inner and outer surfaces of the curved area, as these areas are prone to thickness changes, surface quality issues, or stress concentration. Due to the irregular shape of special-shaped glass, it needs to be photographed from multiple different angles to ensure that there are no blind spots in the inspection. You can first determine several main observation directions from the overall perspective. These directions should be able to roughly outline the shape of the special-shaped glass. Then, in each main direction, adjust the angle to shoot within a small range to obtain more detailed surface information. During the shooting process, 3D modeling software or image processing algorithms can be used to mark and splice the captured images to better construct a complete surface model of the special-shaped glass.

[0073] Further optionally, inputting the glass surface information into an anomaly detection model to perform anomaly detection on the glass sample includes: The glass surface information is projected into a feature space to obtain a feature space image corresponding to the glass surface information; candidate abnormal features are extracted from the feature space image; based on the attribute information of the glass sample, a matching standard abnormal database is selected; the attribute information includes at least: glass material type, shooting environment information, and glass material characteristics; the candidate abnormal feature is compared with the standard abnormal features preset in the standard abnormal database to obtain a comparison evaluation score corresponding to the candidate abnormal feature; if the comparison evaluation score reaches a set threshold, it is determined that the candidate abnormal feature is an abnormal scratch.

[0074] For thin glass samples (e.g., less than 3mm thick), light refraction and reflection are relatively weak. To capture clear surface details, the camera lens should be held as perpendicular to the glass surface as possible, and the shooting distance should be kept close. This reduces image distortion caused by refraction and ensures that even tiny surface defects such as scratches and bubbles can be clearly captured. For example, when photographing thin glass samples such as mobile phone screens, the camera should be kept approximately 10-20cm from the glass surface to obtain high-resolution images.

[0075] Thin glass is susceptible to background and reflected light, necessitating the use of multiple light sources illuminated from various angles. The light source angle can be adjusted from 30° to 60° relative to the glass surface to create the appropriate light and shadow effects, highlighting surface details. For example, during filming, two or three LED light sources can be used to illuminate the glass from different directions. By adjusting the intensity and angle of the light sources, surface defects such as scratches can be more clearly seen against the contrast between light and shadow. Furthermore, to prevent reflected light from interfering with camera imaging, a polarizing filter can be used to reduce the intensity of light reflected from the glass surface.

[0076] For thin glass where internal structure observation is required, such as glass with multiple coatings or internal inclusions, a layered imaging method can be used. By adjusting the camera's focal length, images are taken at different depths of the glass. With each shot, the focus position is precisely controlled to focus on a specific layer within the glass, capturing clear images of each layer. This method, similar to tomography in medicine, can help inspectors detect potential minor defects or impurities within the glass.

[0077] For medium-thick glass samples (e.g., 3mm to 10mm), in addition to focusing on surface details, the side profiles also need to be considered. Multiple angles can be used for photography, including perpendicular and parallel to the side profiles. Shooting perpendicular to the surface is useful for observing the quality of the front and back surfaces of the glass, such as scratches and stains. Shooting parallel to the side profiles can be used to inspect the interior of the glass for bubbles, inclusions, and uniform thickness. For example, when photographing medium-thick glass such as automobile windshields, shooting from a side angle can reveal internal delamination or tiny bubbles, which may not be easily visible when photographed perpendicularly. For medium-thick glass, the light source intensity can be appropriately increased to ensure that light penetrates the glass and creates sufficient internal scattering for easy observation of the internal structure. The light source angle can be adjusted between 40° and 70°, which helps create appropriate light and shadow effects on the glass surface and interior. At the same time, care should be taken to avoid excessive reflections from the glass surface due to excessive light intensity, which can affect the observation of surface details. Reflections can be controlled by using a lens hood or adjusting the distance between the light source and the glass.

[0078] When photographing medium-thick glass, macro mode is recommended for capturing surface details. Using a macro lens or a macro attachment to a regular lens allows for high-magnification images that clearly reveal microscopic surface features. Furthermore, long-range photography is necessary to observe the overall shape of the glass and its surface quality over a wide area. Long-range photography provides a comprehensive view of the glass, making it easier to detect large stains, scratches, or irregularities.

[0079] For thick glass samples (e.g., those thicker than 10mm), observing the internal structure is even more important, as they may contain numerous bubbles, impurities, or stress concentration points. Transmitted light photography can be used, with the light source positioned on one side of the glass and the camera positioned on the other. This allows for a clear view of the light path through the glass, allowing for the detection of internal defects. For example, when photographing thick glass curtain walls used in buildings, transmitted light photography can help inspectors detect internal concretions (impurities produced during the glass manufacturing process), which cast distinct shadows under transmitted light. Due to the thickness of thick glass, a single angle alone is insufficient to fully capture its quality. Multi-angle stereo photography can be used, with multiple cameras positioned around the glass to capture images from different directions. These cameras can capture images simultaneously or sequentially. Computer image processing techniques are then used to stitch these images together into a 3D model, enabling a comprehensive analysis of the surface and internal conditions of the thick glass. During the photography process, careful attention must be paid to the accuracy of each camera's shooting angle and position to ensure accurate reconstruction of the thick glass's 3D structure. Thick glass can cause significant light scattering and refraction, so you need to consider these factors when adjusting your shooting parameters. For example, you can increase the exposure time to compensate for light loss due to light scattering, and adjust the camera's aperture and focal length to reduce image blur caused by refraction. You can also use light compensation tools or software algorithms to correct for light scattering and refraction in the captured image, improving image quality and allowing you to better observe the various features of thick glass.

[0080] For example, through specific mathematical transformation methods (such as principal component analysis and linear discriminant analysis), the collected glass surface information (including multi-dimensional data such as images from multiple light source systems) is projected into a feature space to form a feature space image. This process aims to reduce the dimensionality of the data, extract the data that best represents the key characteristics of the glass surface, and remove some redundant and noisy information, making subsequent analysis and processing more efficient and focusing on valuable feature representations. For example, after principal component analysis, the originally complex image pixel information is transformed into a feature space image composed of a few principal components. These principal components can comprehensively reflect the texture, light and shadow, and other characteristics of the glass surface.

[0081] Based on the feature-space image, image recognition and feature extraction algorithms (such as those based on edge detection and texture analysis) are used to identify features that may indicate anomalies. For example, a scratch on the glass surface may appear in the feature-space image as a discontinuity in the texture or a sudden change in the edge. These unique features can be captured and extracted by the corresponding algorithm as candidate anomaly features, which provide the basis for subsequent judgment of whether the scratch is truly abnormal.

[0082] Because the appearance of abnormal features varies depending on the glass material type, shooting environment, and other properties, a standard anomaly database must be selected based on these properties of the glass sample. For example, the appearance of scratched tempered glass and ordinary float glass in an image may differ due to differences in material hardness, transparency, and other properties. Similarly, different shooting environments, such as lighting and background, can cause abnormal features to appear differently. Only with an accurate matching database can a reasonable and effective comparison be made.

[0083] When the extracted candidate abnormality features are compared with the standard abnormality features preset in the standard abnormality database, a comparison evaluation score is assigned based on factors such as feature similarity using a pre-set quantitative algorithm. This score comprehensively considers the degree of match in multiple aspects such as feature shape, position, and intensity, thereby measuring the closeness of the candidate abnormality features to the standard abnormality features and determining whether they are abnormal scratches.

[0084] This approach considers multiple aspects of the glass sample's properties and compares them against a standard database, avoiding the one-sidedness of relying solely on visual image analysis. It can more accurately identify true abnormal scratches and reduce false positives. For example, some glass surfaces, due to their unique texture, may appear scratched under certain lighting angles. However, by comparing against a corresponding standard, this false positive can be accurately eliminated. This system is adaptable to testing glass samples of varying materials and in diverse shooting environments. Whether it's high-precision glass used in optical instruments or ordinary glass used in buildings, and regardless of the shooting conditions, whether strong or low light, the sample information can be accurately compared based on its properties, enabling effective anomaly detection and expanding the applicability of the entire inspection process. By using comparative evaluation scores to determine abnormal scratches, the system provides a quantitative measure of the test results, moving beyond a simple "yes" or "no" judgment. This facilitates further analysis of the severity of the anomaly based on the score, and facilitates optimization of the inspection system, improving algorithms and parameter settings based on the degree of match between the score and the actual situation.

[0085] For example, suppose we need to inspect a sample of x-glass used in curtain wall applications (material type is known). The image is captured indoors under uniform lighting and against a white background (clear information about the shooting environment), and the material has properties such as high hardness and transparency (known material properties). First, the captured glass surface information is projected into a feature space to generate a corresponding feature space image. From this image, candidate anomaly features, such as texture discontinuities and localized light and shadow abrupt changes, are extracted. Based on these attribute information, a standard anomaly database for curtain wall glass in indoor environments is selected and compared with these candidate anomaly features. For example, if the extracted texture discontinuity feature is highly similar in shape, location, and relative intensity to a scratch-related standard anomaly feature in the standard anomaly database, and the comparison score reaches a set threshold, then the candidate anomaly feature is determined to be an abnormal scratch, indicating that scratches are present on the glass sample surface. Further cleaning or other treatment procedures are required based on the severity of the scratches.

[0086] As an optional embodiment, generating a cleaning strategy corresponding to the glass sample by using a cleaning planning model includes: The glass sample is comprehensively scanned by a laser scanning module in a visual inspection system to obtain depth feature information of the glass sample; a three-dimensional point cloud model of the glass sample is constructed based on the depth feature information; and a planning prediction branch in a cleaning planning model that matches the three-dimensional point cloud model is used to generate a cleaning strategy corresponding to the glass sample.

[0087] The planning prediction branch matches the shape and size parameters of the glass sample in the 3D point cloud model. The cleaning planning model includes multiple planning prediction branches, each of which is used to adaptively construct a corresponding cleaning strategy based on the glass sample shape. The cleaning strategy includes at least the motion trajectory, speed, angle, detergent spray volume, and spray angle of the wiping head.

[0088] Specifically, the laser scanning module in the visual inspection system works based on the principle of laser ranging. It emits a laser beam toward the surface of the glass sample. The laser beam is reflected upon contact with the glass surface. The sensor in the laser scanning module receives the reflected laser signal and calculates the distance from each scanning point to the sensor based on parameters such as the laser propagation time and the known laser propagation speed. By scanning the surface of the glass sample point by point, a large amount of distance data at different locations can be obtained. This data reflects the depth of each location on the glass surface, that is, the depth feature information. For example, for a glass surface with a certain curvature or unevenness, the distance measured at different locations will be different, and these different data constitute part of the depth feature information.

[0089] It can accurately obtain information related to the three-dimensional morphology of the glass sample surface. Compared with the information obtained only through two-dimensional images, the depth feature information can more comprehensively display the actual shape of the glass. Whether it is a flat surface, a curved surface or a special-shaped surface with complex undulations, it can be recorded in detail, providing basic data for the subsequent construction of an accurate three-dimensional model, avoiding the omission of surface morphological details that have an important impact on the cleaning strategy due to relying solely on a two-dimensional perspective.

[0090] The numerous depth feature information acquired by the laser scanning module (the 3D coordinates of each scanned point) is then integrated to form discrete points. These points, when gathered in space, form a 3D point cloud model. Each point contains its positional coordinates in 3D space (typically represented by X, Y, and Z coordinates). These point clouds provide a macroscopic view of the glass sample's overall shape and surface details. For example, for a circular glass dome, laser scanning generates point clouds at various locations. The model shows the arcuate shape, with its high center and low edges, as well as the distribution of point clouds corresponding to any surface irregularities.

[0091] This intuitively displays the three-dimensional form of the glass sample in a visual, digital form, facilitating the subsequent accurate extraction and analysis of key parameters such as the glass's shape and size. Furthermore, the 3D point cloud model can be easily integrated with computer-aided design (CAD) software and computer vision algorithms, providing a standardized, universal data format for further processing and decision-making (such as matching cleaning plan branches and simulating the cleaning process), thereby improving the automation and intelligence of the entire cleaning strategy generation process.

[0092] The cleaning planning model pre-configures multiple prediction branches, each with a specific cleaning strategy for glass samples with different shapes and dimensions. After constructing a 3D point cloud model of the glass sample, the system analyzes the model to extract the sample's shape (e.g., circular, rectangular, or irregular) and specific dimensions (length, width, height, radius, etc.). The cleaning planning model then searches for a matching prediction branch. For example, if the 3D point cloud model indicates a long, curved, irregular glass sample, the system will find a prediction branch specifically tailored to this shape and size. This branch, based on pre-set algorithms and rules and considering the specific conditions of the glass surface, determines specific cleaning strategy parameters, including the wiper's trajectory (e.g., following the curved path, focusing on the ends of the strip), speed (adjusted to the surface's cleaning difficulty and material), angle (for contact with curved or irregular edges), detergent spray volume (determined by surface size and soiling severity), and spray angle (to ensure even coverage).

[0093] This enables adaptive generation of cleaning strategies, enabling the development of highly tailored cleaning plans for glass samples of varying shapes and sizes, significantly improving cleaning efficiency and quality. Compared to using a fixed, unified cleaning strategy, this strategy based on matching branch generation avoids issues such as incomplete cleaning (e.g., missing complex corners of irregularly shaped glass), surface damage (e.g., scratches caused by wiping at an inappropriate angle), and waste of cleaning agents caused by improper cleaning methods. This makes the entire cleaning process more scientific, precise, and efficient, meeting the cleaning needs of diverse glass samples.

[0094] In summary, through the above series of steps, using the laser scanning module, building a 3D point cloud model, and matching the planning prediction branch, a complete and intelligent glass sample cleaning strategy generation system has been formed, which helps to improve the overall level of glass cleaning work.

[0095] Optionally, each planning prediction branch in the cleaning planning model is categorized and designed based on the shape (e.g., circular, rectangular, triangular, polygonal, and various irregular shapes) and size (including specific parameters such as length, width, height, and radius) of the glass sample. For example, the planning prediction branch for circular glass samples uses algorithms and rules based on the geometric properties of circles. It takes into account the symmetry of the circle's circumference. When generating a cleaning strategy, the wiper's trajectory may follow a circular motion around the circle's center. The speed is set based on the circle's radius and the estimated cleaning time, and the amount of detergent sprayed is calculated based on the circle's area. These branches are typically trained using a large amount of sample data. During the training process, data from successful cleaning cases of glass samples of various shapes and sizes is collected, including the corresponding cleaning strategy parameters (wiping head trajectory, speed, angle, detergent spray amount, and angle) and post-cleaning performance evaluations. Machine learning algorithms, such as neural networks and decision trees, are used to learn from this data, enabling each planning prediction branch to predict an appropriate cleaning strategy based on the input glass sample shape and size information.

[0096] In addition to being data-driven, the planning and prediction branch also incorporates knowledge from physics and chemistry. For example, from a physical perspective, it considers the friction between the wiper head and the glass surface, as well as the flow and diffusion characteristics of the cleaning agent on the glass surface. From a chemical perspective, it determines the appropriate cleaning agent composition and spray amount based on the glass material and stain type (oil, dust, water, etc.). This knowledge is embedded in the branch's algorithms and rules to ensure that the generated cleaning strategy is both efficient and safe in real-world applications.

[0097] The circular glass branch primarily moves in a circular motion around the center of the circle to ensure uniform cleaning across the entire circular surface. The radius of the motion path can be adjusted based on the radius of the glass, and multiple circular motions may be configured to meet cleaning requirements. For example, a spiral motion may be performed from the center outward, followed by a reverse spiral motion from the outside back to the center to ensure thorough cleaning. Given the characteristics of circular motion, the speed is relatively stable. The speed setting is related to the radius of the circle. Generally, a larger radius results in a faster speed, but the glass material and the degree of contamination should also be considered. For example, for optical lenses with larger radii, due to their high surface quality requirements, a moderate speed is used to avoid incomplete cleaning or damage to the lens surface caused by excessive speed. The wiping head is maintained at a perpendicular or near-perpendicular angle to the glass surface to ensure even distribution of wiping pressure during the circular motion. Precise angle control is particularly important for circular glasses with special optical coatings to prevent damage. The spray volume is calculated based on the area of ​​the circle to ensure uniform coverage of the entire circular surface. The spraying angle is generally perpendicular to the glass surface, or tilted to a certain extent according to the direction of circular motion, so that the detergent is evenly distributed under the drive of circular motion to avoid local accumulation.

[0098] For rectangular glass, the motion trajectory includes linear motion along both the long and short sides. Multiple reciprocating wiping strokes may be performed along the long side, followed by additional wiping strokes along the short sides to ensure that all corners and edges are cleaned. For large rectangular glass, a diagonal wiping trajectory may also be designed to address potential stain buildup or difficult-to-clean areas. The speeds along the long and short sides can differ. For longer long sides, the speed may be slightly faster to improve cleaning efficiency, but will slow down appropriately near edges and corners. The speed along the short sides is adjusted based on the width of the rectangle and the amount of stains. For example, for large rectangular glass curtain walls used in buildings, the wiping speed along the long side can be optimized based on the curtain wall's height and the capacity of the cleaning equipment, while the speed along the short side will slow down near the window frame to ensure a high-quality clean at the junction of the frame and glass. During the wiping process, the angle of the wiping head to the glass surface is maintained within an appropriate range to ensure even and effective wiping pressure. At the edges and corners of the rectangle, the angle is adjusted appropriately, allowing the wiper to better conform to the corners and avoid missing stains. The spray volume is calculated based on the area of ​​the rectangle, taking into account that edges and corners may require more detergent to tackle stubborn stains, so appropriate additional spray is applied to these areas. The spray angle can be perpendicular to the glass surface or tilted according to the direction of the wiper's movement to ensure that the detergent is evenly distributed along the wiping direction, improving cleaning results.

[0099] For shaped glass, the wiper head's trajectory can be configured based on the specific shape of the glass and key areas (such as areas with significant curvature changes, concave or convex sections). The trajectory focuses on key areas with multiple circular or contoured wiping. For irregular edges and complex curves, flexible trajectories are employed, such as traversing the curve's tangents or performing a reciprocating in-and-out motion in concave areas, to ensure that all surface areas are cleaned. Speeds vary significantly across different areas of shaped glass. Speeds can be increased for flatter, easier-to-clean areas; however, speeds are significantly reduced for key areas, particularly those with complex shapes and prone to damage. For example, for shaped artistic glass with intricate carvings or textures, the wiping speed is very slow in these areas to ensure cleaning quality and surface integrity. The wiper head's angle is continuously adjusted based on the curvature and shape of the shaped glass surface. On convex surfaces, the wiper head angle is tilted outward to conform to the convex shape; on concave surfaces, the angle is tilted inward. When transitioning between areas of varying curvature, the angle should be adjusted smoothly to avoid damaging the glass surface. The spray volume is adjusted based on the size of the shaped glass and the cleaning difficulty of each area. For recessed areas, gaps, or complex curved surfaces where stains easily accumulate, the amount of detergent sprayed is increased. The spray angle is adjusted based on the shape of the shaped glass and the movement of the wiper head to ensure that the detergent reaches all areas that need cleaning, especially those that are hidden or difficult to reach.

[0100] Furthermore, during the actual cleaning process, feedback is collected after the cleaning strategy is executed, such as cleaning effectiveness evaluation (e.g., residual stains, damage to the glass surface), cleaning time, and detergent usage. This feedback data is fed back to the corresponding planning and prediction branch, where the parameters and rules within the branch are updated and optimized using a machine learning algorithm. For example, if a certain shape of glass is found to have a high level of residual stains after cleaning using the current cleaning strategy, the system analyzes the problem in the wiper head's trajectory, speed, detergent application, and other aspects, and then adjusts the relevant parameters in the planning and prediction branch accordingly. When encountering new glass shapes or sizes, or new cleaning requirements and conditions, this new sample data is collected and added to the training dataset, retraining the planning and prediction branch of the cleaning planning model. This continuously expands the model's applicability and accuracy, enabling it to handle a wider range of glass cleaning tasks. For example, as new, special-shaped glass becomes increasingly common in architectural decoration or industrial products, these new sample data can be incorporated into the training set to update the rules and parameters of the special-shaped glass branch to better meet cleaning needs.

[0101] Further optionally, the adopting a strategy branch in the cleaning planning model that matches the three-dimensional point cloud model to generate a cleaning strategy corresponding to the glass sample includes: If the shape of the glass sample of the three-dimensional point cloud model is a rectangle, any corner point in the three-dimensional point cloud model is obtained through the first planning prediction branch; with the corner point as the starting point, the cleaning strategy is set as follows: along the long side direction of the three-dimensional point cloud model, a linear movement is performed at a constant movement speed, and an adaptive wiping pressure and a cleaning agent spraying amount are used for wiping operations; when the movement reaches the end point of the long side, the movement is turned and a linear movement is performed along the short side direction of the three-dimensional point cloud model, and an adaptive wiping pressure and a cleaning agent spraying amount are used for wiping operations; when the movement reaches the end point of the short side, the movement is repeated along the long side direction of the three-dimensional point cloud model, a linear movement is performed at a constant movement speed, and an adaptive wiping pressure and a cleaning agent spraying amount are used for wiping operations, until the cleaning process of the glass sample is completed; wherein, the wiping pressure and the cleaning agent spraying amount are adaptively configured based on the degree of dirtiness of the glass surface detected in real time during the movement.

[0102] For rectangular glass samples, select any corner point in the 3D point cloud model as the starting point. This is because rectangles have four right angles, and starting from a corner allows for a smooth, regular cleaning along the edges of the rectangle. A corner point is the intersection of two edges, and using it as a starting point naturally guides the wiper head along the long and short sides, ensuring the entire rectangular surface is covered.

[0103] The linear motion is simple and efficient, moving at a constant speed along the long side. Since the long side is relatively long, maintaining a constant speed can ensure the stability and predictability of the cleaning process, making it easier to control the cleaning time and quality. During the movement, the wiping pressure and the amount of detergent sprayed are adjusted according to the real-time detection of the degree of dirt on the glass surface. If a certain area is detected to be heavily dirty, the system will automatically increase the wiping pressure and the amount of detergent sprayed to ensure that the stains can be effectively removed. For example, when stubborn oil stains are detected on the glass surface, appropriately increasing the wiping pressure can enable the wiping head to better contact the glass surface, while increasing the amount of detergent sprayed can help dissolve and remove the oil stains. This is based on physical and chemical principles, and overcomes the adhesion between the stains and the glass surface by applying appropriate external force and sufficient detergent.

[0104] When the wiping head reaches the end of the long side, it turns and moves in a straight line along the short side. This steering mechanism ensures that the cleaning path covers the entire rectangular surface. Short-side cleaning also uses adaptive wiping pressure and detergent spray volume, following the same principle as long-side cleaning: effective cleaning based on soiling conditions.

[0105] By repeating the cleaning steps in the long and short sides directions until the entire glass sample is cleaned, this cycle ensures that the entire rectangular glass surface is thoroughly and evenly cleaned, avoiding missing any area.

[0106] This cleaning strategy ensures that every part of the rectangular glass surface is cleaned. Starting from the corners, a circular cleaning path along the long and short sides covers the entire rectangular area, including edges and corners. Compared to random cleaning paths or irregular cleaning methods, this sequential cleaning strategy effectively reduces blind spots and improves cleaning completeness.

[0107] The constant speed of the long-side motion helps improve cleaning efficiency. Combined with adaptive wiping pressure and detergent spray volume, it can flexibly adjust to the actual level of dirt, ensuring cleaning quality. In lightly soiled areas, it avoids over-cleaning and wasting time and detergent. In heavily soiled areas, it can take effective cleaning measures to ensure thorough removal.

[0108] Adaptive wiping pressure protects both the wiping device and the glass surface. When cleaning requires minimal pressure, excessive pressure won't damage the wiping head or scratch the glass. When increased pressure is needed for stubborn stains, the device maintains a reasonable adjustment range, preventing excessive wear and tear on the device and damage to the glass surface.

[0109] As an optional embodiment, the step of using a strategy branch in the cleaning planning model that matches the three-dimensional point cloud model to generate a cleaning strategy corresponding to the glass sample includes: If the glass sample shape of the three-dimensional point cloud model is circular, the center position of the three-dimensional point cloud model is determined through the second planning prediction branch; with the center position as the starting point, the cleaning strategy is set as follows: wiping operation is performed outward along the path of the Archimedean spiral; during the wiping process, the wiping pressure is adjusted according to the radius length, and the amount of detergent sprayed increases according to a preset functional relationship with the radius length.

[0110] For circular glass samples, determining the center of the circle is a key starting point for the cleaning strategy. This is based on the geometric properties of a circle, where the center is the central symmetric point of the circle. By performing geometric analysis on the 3D point cloud model, an algorithm is used to precisely locate this center. For example, by calculating the distance from each point in the point cloud model to the model boundary and finding a set of points with equal distances, the center of the circle is determined. This center position provides the central reference for the subsequent spiral cleaning path.

[0111] The Archimedean spiral was chosen as the wiping path because it can cover the entire circular surface in a uniform and continuous manner. The Archimedean spiral is characterized by starting from the center of the circle and increasing its radius at a fixed rate as the angle increases. This ensures that the wiping head can sequentially pass through each part of the circular surface as it expands outward, avoiding duplication or omission of the cleaning path. In a polar coordinate system, during the cleaning process, the wiping head moves according to this path equation. As the angle continues to change, the radius gradually increases, achieving comprehensive cleaning from the center to the circumference.

[0112] Adjusting wiping pressure based on radius length takes into account that on circular surfaces, as the radius increases, the linear velocity of the wiping head increases (assuming the angular velocity of the wiping head is constant). Based on physical principles, wiping pressure needs to be adjusted appropriately to ensure consistent cleaning results. In areas with a smaller radius, where the wiping head moves relatively slowly, a lower wiping pressure can be used. As the radius increases and the speed increases, the wiping pressure should be increased appropriately to ensure that the friction between the wiping head and the glass surface effectively removes stains.

[0113] As the radius lengthens, the amount of detergent sprayed increases according to a preset function. This is because the area of ​​a circle is proportional to the square of the radius. During the cleaning process, as the radius increases, the area to be cleaned also increases. To ensure that the detergent can evenly cover the glass surface, the spraying amount needs to be increased accordingly. For example, the preset function can be a linear relationship (such as the spraying amount is proportional to the radius) or a quadratic function (such as the spraying amount is proportional to the square of the radius). The specific functional relationship can be determined based on factors such as the properties of the detergent, the material of the glass surface, and the type of stain.

[0114] In this way, the wiping path of the Archimedean spiral can efficiently cover the entire surface of the circular glass, avoiding cleaning dead corners. Compared with other irregular or discontinuous cleaning paths, this path can complete the cleaning of the circular glass in one continuous movement, greatly improving the cleaning efficiency. By adjusting the wiping pressure according to the length of the radius, the cleaning effect on the entire circular surface is more consistent. Whether in the area close to the center of the circle or the circumference, it can ensure that the wiping head has appropriate friction with the glass surface, effectively remove stains, and improve the stability of the cleaning quality. Increasing the amount of detergent sprayed with the radius according to a preset functional relationship can ensure that the detergent is evenly distributed on the surface of the circular glass. This not only avoids the waste caused by spraying too much detergent in the center area, but also ensures that there is enough detergent to clean stains in large areas such as the circumference, thereby realizing the rational use of detergent.

[0115] Understandably, different glass materials have varying adsorption and reactivity to cleaning agents. For example, due to differences in chemical composition and microstructure, the spread and penetration of cleaning agents on the surfaces of ordinary soda-lime glass and tempered glass will differ. For glass materials with stronger adsorption, the increase in the cleaning agent spray volume with radius in the preset function relationship may be relatively small, as the cleaning agent spreads better on the surface. For glass materials with weaker adsorption, the spray volume may need to be increased more quickly to ensure effective cleaning.

[0116] The surface roughness of the glass also affects the amount of detergent required. If the glass surface is rough, stains tend to hide in pits and crevices, requiring more detergent to fully reach and dissolve the stains. In this case, the slope of the detergent spray rate as the radius increases in the preset function relationship may be larger to accommodate the demand for detergent on rough surfaces. For example, frosted glass requires more detergent than smooth glass. The function relationship may be adjusted based on surface roughness test data, such as using a quadratic function relationship with a larger quadratic coefficient.

[0117] Different types of stains require different amounts of detergent to remove. For example, oil stains typically require more detergent to dissolve because oil adheres strongly to the glass surface; dry stains like dust are relatively easy to remove and require less detergent. If the glass sample is primarily contaminated with oil stains, the preset function relationship will increase the amount of detergent sprayed with a larger increase in radius. An exponential function relationship may be used to ensure that there is enough detergent to break down the oil stains. For glass contaminated with dust, the function relationship may be more gentle, such as a linear relationship with a smaller slope.

[0118] The degree of soiling is also an important factor in determining the preset functional relationship. If the glass surface is heavily soiled, more detergent will be required for a thorough clean. By assessing the degree of soiling on the glass surface (using methods such as optical inspection and image analysis), the functional relationship can be adjusted. For example, in heavily soiled areas, the rate at which the detergent spray rate increases with radius can be increased, such as by increasing the slope of a linear function or adjusting the coefficients of a quadratic function, to ensure that the detergent is adequately effective against the soiling.

[0119] The performance of the cleaning equipment's detergent spray system can limit the preset functional relationship. For example, factors such as the spray pressure, nozzle type, and spray pattern can affect the actual spray effect of the detergent. If the spray pressure is low, the detergent droplets from the nozzle are larger, and the functional relationship may need to be adjusted so that the detergent spray volume increases more with increasing radius to ensure uniform coverage of the glass surface.

[0120] Parameters such as the wiper's speed and angle also affect the distribution and efficiency of the cleaning agent. If the wiper's speed is high, the cleaning agent needs to cover the glass surface more quickly. In this case, the rate at which the cleaning agent spray rate increases with radius in the pre-set function relationship may need to be increased to avoid situations where the cleaning area is not covered by the cleaning agent in time. Furthermore, the angle between the wiper and the glass surface also affects the flow direction and distribution of the cleaning agent, which also needs to be considered when determining the function relationship.

[0121] Before actual cleaning, experiments can be conducted on glass samples of the same material and shape, simulating different staining conditions. The amount of detergent sprayed at different radii can be recorded to achieve the desired cleaning effect. Through multiple experiments, data can be collected and analyzed to fit a suitable functional relationship. For example, a circular glass sample can be manually smeared with varying degrees of staining. Cleaning can be performed using a set wiping path and speed, with the amount of detergent sprayed adjusted each time. The cleaning effect can then be observed. Based on the experimental results, a functional relationship between the amount of detergent sprayed and the radius can be constructed.

[0122] Using computer simulation software, a physical model of glass cleaning is established. By inputting information such as the glass surface characteristics, the type and extent of stains, and cleaning equipment parameters, the system simulates the distribution of detergent on the glass surface and the cleaning process. By continuously adjusting the parameters in the pre-set functional relationship and observing the simulation results, the optimal functional relationship is found, enabling the detergent to achieve excellent cleaning results on the glass surface in the most economical and efficient manner. For example, by setting different functional relationships in the simulation software to simulate the spraying and flow of detergent, the cleaning results and detergent usage under different settings are compared to determine the most appropriate pre-set functional relationship.

[0123] As an optional embodiment, in the specific method of increasing the amount of detergent sprayed according to a preset functional relationship with the radius length, different amounts of detergent are required for optimal cleaning of glass surfaces at different levels of dirtiness. By categorizing the dirtiness level into three ranges: light (0-30), moderate (30-70), and heavy (70-100), the amount of detergent sprayed can be more precisely controlled. Within each range, a linear function is used to describe the relationship between the amount of detergent sprayed and the degree of dirtiness because, within a certain range of dirtiness, the amount of detergent used and the degree of dirtiness are roughly linearly related. For example, in the light dirtiness range, a small amount of detergent may be sufficient to clean the glass surface. As the degree of dirtiness increases, the amount of detergent sprayed increases linearly accordingly (Q = k1D + b1). Here, k1 and b1 are coefficients determined experimentally or empirically. k1 represents the rate at which the amount of detergent sprayed increases with the degree of dirtiness, and b1 may be a baseline amount of detergent used to address the minimum amount of detergent required even for light dirtiness.

[0124] In the moderately soiled range (Q = k2D + b2) and heavily soiled range (Q = k3D + b3), the effects of k2, k3, and b2, b3 are similar to those in the lightly soiled range, but the coefficients vary depending on the degree of soiling. k2 and k3 are typically greater than k1 because, as soiling increases, more detergent is required to ensure effective cleaning. The values ​​of b2 and b3 may be adjusted based on actual conditions. For example, to account for surface properties of moderately and heavily soiled glass, such as stronger stain adhesion, an additional base spray volume may be required to initiate the cleaning process. These coefficients can be determined experimentally. For example, in a laboratory setting, multiple circular glass samples of the same material and shape are prepared and soiled to varying degrees (e.g., by applying known amounts of dust, oil, etc.). A wiping robot is then used to clean along a programmed path. During the cleaning process, the detergent spray volume is adjusted. Data on the spray volume required to achieve optimal cleaning results at different soiling levels are recorded, and the coefficients for each range are determined through data fitting.

[0125] Dirt level sensors can be implemented using a variety of technologies. One common approach is optical sensors, which determine the degree of dirt on a glass surface by emitting light of a specific wavelength and detecting the intensity and spectral characteristics of the reflected light. When the glass surface is contaminated, the intensity and spectrum of the reflected light change, and the sensor quantifies the degree of dirtiness based on these changes. For example, for clear glass, the intensity of reflected light is low on a clean surface, while it increases on a contaminated surface. The spectrum of reflected light also varies depending on the type of dirt. By analyzing these parameters, the degree of dirtiness can be determined.

[0126] Another possible approach is to utilize a capacitive sensor. The presence of dirt changes the dielectric constant of the glass surface, thereby affecting the capacitance value of the capacitive sensor. By measuring this change in capacitance, the degree of dirt on the glass surface can be inferred. Upon receiving data from the dirt level sensor, the wiping robot's control system first determines the range of dirt level D (light, moderate, or heavy). It then calculates the detergent spray volume Q based on the corresponding piecewise linear function. For example, if the sensor detects a dirt level D of 40, the control system determines that it is in the moderate dirt level range and substitutes D = 40 into Q = k²D + b² to calculate the required detergent spray volume. Based on this calculation, the control system adjusts the parameters of the detergent spray system, such as the micropump flow control, in real time to ensure precise detergent spraying. This ensures that the detergent spray volume consistently matches the actual dirt level on the glass surface during the cleaning process.

[0127] In this way, by adjusting the amount of detergent sprayed in real time according to the degree of dirtiness, it is possible to avoid incomplete cleaning due to insufficient detergent use, or waste and residue due to excessive detergent use. In lightly dirty areas, only an appropriate amount of detergent is used, which not only effectively cleans the glass but also saves detergent. In heavily dirty areas, sufficient detergent can be provided to deal with stubborn stains and ensure cleaning quality. Since the amount of detergent sprayed is closely matched to the degree of dirtiness, the cleaning process is more scientific and reasonable. For areas with different degrees of dirtiness, the most suitable cleaning conditions can be provided, thereby improving the cleaning quality of the entire glass surface, enabling the glass to achieve better cleaning effects and meet different cleaning needs.

[0128] This piecewise linear function control method maximizes cleaning agent utilization. Compared to cleaning methods using a fixed spray rate, it precisely distributes cleaning agent according to actual needs, reducing unnecessary waste. This can significantly reduce cleaning agent costs, especially when cleaning large glass areas or large numbers of glass samples. Precise cleaning agent application helps improve cleaning efficiency. With the appropriate amount of cleaning agent, stains are dissolved and removed more quickly, resulting in more effective wiping from the wiper head, thus reducing cleaning time. Furthermore, it avoids the post-cleaning residual processing time caused by excessive cleaning agent, further improving overall cleaning efficiency.

[0129] As an optional embodiment, the step of using a strategy branch in the cleaning planning model that matches the three-dimensional point cloud model to generate a cleaning strategy corresponding to the glass sample includes: If the glass sample shape of the three-dimensional point cloud model is irregular, the contour of the three-dimensional point cloud model is segmented into multiple line segments and / or curve segments through the third planning prediction branch; a cleaning sequence corresponding to the three-dimensional point cloud model is predicted; and a wiping operation is performed along the multiple line segments and / or curve segments based on the cleaning sequence.

[0130] For special-shaped glass samples, their shapes are complex and irregular. By performing shape analysis on the three-dimensional point cloud model, the contour is divided into multiple line segments and / or curve segments using a geometric algorithm. This process is similar to decomposing a complex figure, breaking down the boundary of the special shape into basic geometric elements that are easier to handle. For example, for special-shaped glass with a combination of curved edges and straight edges, by identifying the points where the curvature changes on the boundary, the continuous contour is divided at the point where the curvature suddenly changes, obtaining straight segments and curve segments. In terms of algorithm implementation, a method based on differential geometry can be used to calculate the curvature of each point on the contour. When the curvature exceeds a certain threshold, it is determined to be the starting or ending point of the curve segment, thereby dividing different line segments and curve segments. These line segments and curve segments can more accurately describe the contour characteristics of the special-shaped glass, providing a basis for the subsequent formulation of cleaning strategies.

[0131] When predicting the cleaning order, it's important to consider both the shape and cleaning ease of the shaped glass. Generally, priority is given to easily accessible, larger areas to improve cleaning efficiency. For example, if the glass has concave or convex sections, the convex sections might be cleaned first because they are more accessible to the wiper and are less likely to accumulate contaminants like dirty water after cleaning.

[0132] At the same time, the continuity of the cleaning path and the avoidance of repeated cleaning must also be considered. By analyzing the connections between individual line segments and curves, an optimal cleaning sequence is determined, allowing the wiping head to transition naturally from one section to another during movement, reducing unnecessary path reversals and jumps. For example, based on the topological structure of shaped glass, adjacent line segments and curves are arranged in a specific order, allowing the cleaning process to proceed along a continuous path, similar to planning the most reasonable travel route on a map.

[0133] Based on a defined cleaning sequence, the wiping robot's control system directs the wiping head to sequentially wipe along multiple lines and curves. When wiping along lines, the wiping head's motion trajectory is relatively simple, primarily linear. Controlling the robot's arm joints ensures the wiping head maintains stable movement along the straight lines. Wiping along curved lines requires more complex motion control.

[0134] The control system adjusts the speed and angle of the wiper head based on the curvature and length of the curved section. For example, on a curve with a large curvature, the wiper head's speed will be appropriately slowed to ensure it can closely follow the curve for cleaning. At the same time, the wiper head's angle is continuously adjusted based on the tangent direction of the curve, ensuring that the wiper head always maintains the appropriate contact angle with the glass surface, effectively removing stains.

[0135] By dividing the contour into line and curve segments and wiping them in a reasonable cleaning order, we can ensure that every part of the special-shaped glass is cleaned. This method effectively overcomes the cleaning difficulties caused by the irregular shape of special-shaped glass. Whether it is glass with complex depressions, protrusions or twists, this sophisticated strategy can achieve comprehensive cleaning and reduce blind spots.

[0136] Because the wiping operation closely follows the contours of the shaped glass and is performed in a reasonable order, stains can be removed more thoroughly. During the cleaning process, the targeted wiping of each line and curve segment, as well as the orderly transition between different sections, help improve the cleaning quality and achieve better cleaning results for shaped glass.

[0137] A reasonable cleaning sequence prediction can reduce the length of the wiper head's motion path on irregular glass surfaces. This improves cleaning efficiency by avoiding unnecessary path reversals and repeated cleaning. For example, when cleaning irregular glass with multiple branches, following the optimal cleaning sequence allows the wiper head to cover the majority of the area in a single continuous motion, rather than shuttling back and forth between branches, saving cleaning time. The optimized cleaning path and motion control method can reduce frequent starts, stops, and sudden turns of the wiper head and robotic arm. This smooth motion process helps reduce equipment wear, extend the life of the wiping robot, and reduce equipment maintenance costs.

[0138] As an optional embodiment, wiping operations are performed along multiple line segments and / or curve segments based on the cleaning sequence, including: when wiping a line segment, moving in a straight line at a constant motion speed, and using adaptive wiping pressure and detergent spraying amount to perform a straight line wiping operation; and / or, when wiping a curve segment, adjusting the motion trajectory, speed and angle of the wiping head according to the curvature and torsion of the curve segment through a motion control algorithm.

[0139] Moving in a straight line at a constant speed makes the cleaning process predictable and stable. When wiping a line segment, the robot can move forward steadily at a set speed, which helps to plan the cleaning time. For example, when cleaning a certain number of special-shaped glasses on an industrial production line, a constant speed allows operators to accurately estimate the cleaning time of each glass, thereby better arranging the production process. This stable speed also helps to improve cleaning efficiency. Because there is no need to adjust the speed frequently, the wiping head can move continuously on the glass surface, reducing the pause time that may be caused by speed changes. Moreover, during the straight-line movement, the continuity of the cleaning path can be guaranteed, avoiding uneven cleaning or missing areas due to speed fluctuations.

[0140] Adaptive wiping pressure allows for flexible adjustments based on the actual level of dirt on the glass surface and the material properties of the glass. For dirtier areas or glass made of harder materials, appropriately increasing the wiping pressure can better remove stains. For example, when cleaning glass segments with stubborn oil stains, the adaptive system will increase the wiping pressure based on the dirt information fed back by the sensor, allowing the wiping head to more effectively contact the stains, thereby improving the cleaning effect. At the same time, adaptive wiping pressure also protects the glass surface. When cleaning cleaner areas or glass made of more fragile materials, the pressure automatically decreases to avoid scratches caused by excessive pressure. This flexible adjustment based on actual conditions ensures high-quality cleaning under varying conditions.

[0141] Adaptive detergent application ensures that the amount of detergent used matches the degree of dirt on the glass surface during straight-line wiping. In lightly soiled areas, reducing the amount of detergent applied avoids waste and reduces residue after cleaning. In heavily soiled areas, increasing the amount of detergent applied effectively dissolves and removes stains. For example, a moderate amount of detergent sprayed on a dusty line is sufficient; however, in areas with sticky stains, the system automatically increases the amount of detergent applied to ensure thorough removal.

[0142] By adjusting the motion trajectory of the wiping head based on the curvature and torsion of the curved segment, the wiping head can closely follow the curved shape of the glass surface. Curvature reflects the degree of curvature of the curve, while torsion describes the degree of distortion of the curve in three-dimensional space. Through motion control algorithms, the wiping head can precisely move along the actual shape of the curve, ensuring that the entire curved segment is cleaned, avoiding the situation where stains are missed due to inability to conform to the curve. For example, for a curved segment of special-shaped glass with a complex twist, by accurately considering the curvature and torsion, the wiping head can move along the curve as accurately as along a track, achieving close contact regardless of whether the curve is convex or concave, thereby effectively cleaning stains from the curved surface.

[0143] Adjusting the wiper head speed based on the curvature and torsion of a curved section ensures excellent cleaning results across curves of varying degrees. On curves with greater curvature or higher torsion, the speed is appropriately reduced to allow the wiper head sufficient time and appropriate conditions to clean stains. These complex curves can be more challenging to remove and require more precise motion control. For example, on a sharply curved glass section, slowing the wiper head speed prevents excessive speed from causing the wiper head to lift off the curved surface or ineffectively clean stains. This speed adjustment helps improve cleaning quality, ensuring consistent cleaning standards across the entire curved section. Adjusting the wiper head angle allows it to better align with the tangent of the curve, ensuring the wiper head maintains an optimal contact angle with the glass surface. A suitable contact angle evenly distributes wiping pressure across the curve, resulting in more effective stain removal. For example, on a concave curve, the wiper head angle is adjusted inward, allowing the wiper head to penetrate deeper into the curve; on a convex curve, the angle is adjusted outward. This dynamic adjustment of the angle can avoid incomplete cleaning or damage to the glass surface due to improper angle, and also improves the stability of the wiping head in the curved section.

[0144] Furthermore, in order to make the wiping head move smoothly on the curve segments with different curvatures and torsions, while taking into account factors such as cleaning effect and equipment operation safety, a speed scaling factor is introduced. The speed scaling factor of the wiping head is expressed as: .in, 、 is the weight parameter used to adjust the curvature and torsion The impact on speed. is the minimum speed. Curvature reflects the degree of curvature of a curve, while torsion describes the degree of distortion of a curve in three-dimensional space. When the curvature or torsion is large, it means that the shape of the curve changes more dramatically, and the wiping head needs to be more careful when moving on such a curve. 、 Used to adjust the effect of curvature and torsion on speed. When the curvature or torsion is large, or The value of will increase, so that the speed scaling factor This will reduce the speed of the wiping head. This will prevent the wiping head from moving unsteadily or deviating from the curve trajectory due to rapid curve changes. It is used to ensure that the wiping head has a minimum movement speed under any circumstances to ensure that the cleaning process can continue.

[0145] When curvature and torsion are small, the velocity scaling factor When the curvature or torsion is large, the speed scaling factor The width of the wiper head will become smaller, and the speed of the wiper head will decrease accordingly, ensuring that the wiper head can move smoothly along the curve, ensuring the cleaning effect and safe operation of the equipment.

[0146] In an embodiment of the present application, for a glass sample to be inspected, glass surface information corresponding to the glass sample is collected; the glass surface information includes at least glass surface images under multiple light source systems; the glass surface information is input into an anomaly detection model to perform anomaly detection on the glass sample; if the glass sample has abnormal scratches, and the abnormal scratches exceed a set damage threshold, the cleaning process of the glass sample is stopped; if the glass sample does not have abnormal scratches, or the abnormal scratches do not exceed the set damage threshold, a cleaning strategy corresponding to the glass sample is generated through a cleaning planning model; based on the cleaning strategy, the wiping head and detergent spraying system of the wiping robot are controlled to execute the cleaning process for the glass sample. The embodiment of the present application can automatically control the wiping robot to perform cleaning, operate according to a precise strategy, reduce human interference, and ensure cleaning consistency, accuracy, stability, and reliability.

[0147] Based on the same implementation principle, an embodiment of the present application further provides an electronic device, Figure 3 is a structural block diagram of the electronic device 300, such as Figure 3As shown, the electronic device 300 includes: a processor 301, a memory 302, a communication interface 303, a communication bus 304 and a controller 305; wherein the processor 301, the memory 302 and the communication interface 303 communicate with each other through the communication bus 304; the memory 302 is used to store computer programs; the processor 301 is used to execute the programs stored in the memory 302 to realize the corresponding processing functions; the communication bus 304 is used for communication between the electronic device 300 and other devices; the controller 305 is used to implement the glass cleaning method described in the above method embodiment.

[0148] In the embodiment of the present application, the communication bus 304 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 304 may be divided into an address bus, a data bus, a control bus, etc., and the specific form is not limited. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0149] The memory 302 may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor 301.

[0150] The processor 301 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other editable logic devices, discrete gate or transistor logic devices, discrete hardware components. The specific form can be determined according to actual needs.

[0151] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement the steps that can be performed by the electronic device in the above method embodiment.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A glass cleaning method, characterized in that: The method is applied to a wiping robot, which at least includes: a robot body, a mechanical arm, a visual detection system, a wiping head, a detergent spraying system, a base, and a main control computer. The method is executed based on the main control computer; the method includes: For a glass sample to be tested, collecting glass surface information corresponding to the glass sample; the glass surface information at least includes glass surface images under multiple light source systems; the multiple light source systems correspond to multiple specific illumination angles and multiple illumination intensities, respectively; Inputting the glass surface information into an anomaly detection model to perform anomaly detection on the glass sample; If the glass sample has abnormal scratches, and the abnormal scratches exceed the set damage threshold, the cleaning process of the glass sample is stopped; If the glass sample has no abnormal scratches, or the abnormal scratches do not exceed the set damage threshold, a cleaning strategy corresponding to the glass sample is generated through a cleaning planning model; wherein the cleaning planning model includes a plurality of planning prediction branches, each planning prediction branch is used to adaptively construct a corresponding cleaning strategy based on the shape of the glass sample; the cleaning strategy includes at least: the motion trajectory, motion speed, angle, detergent spraying amount, and spraying angle of the wiping head; The wiping head and the cleaning agent spraying system of the wiping robot are controlled based on the cleaning strategy to execute a cleaning process for the glass sample.

2. The method according to claim 1, characterized in that Before collecting the glass surface information corresponding to the glass sample to be tested, the method further includes: For each newly added glass sample, identifying the shape of the glass sample; Setting a shooting mode of the glass sample based on the shape of the glass sample, and fine-tuning shooting parameters in the shooting mode based on size information of the glass sample; Among them, the shapes of glass samples include rectangular, circular, and irregular; for rectangular glass samples, the shooting angles in the shooting mode are at least two vertical directions, and the at least two vertical directions include the long side direction and the short side direction; for circular glass samples, the shooting positions in the shooting mode include at least a plurality of shooting points distributed at equal angles around the center of the circle; for irregular glass samples, the shooting angles in the shooting mode are angles that cover the glass surface to the greatest extent and include key parts, and the key parts include at least areas with large curvature changes.

3. The method according to claim 2, characterized in that After setting the shooting mode of the glass sample based on the shape of the glass sample, the method further includes: Setting a plurality of light source systems matching the glass samples; Configuring a specific irradiation angle between a light source in a plurality of light source systems and a glass surface of the glass sample; wherein the specific irradiation angle ranges from 30° to 60°; Based on the environment in which the glass sample is located, the illumination intensity is set to match the background light intensity of the environment; wherein the illumination intensity is inversely proportional to the background light intensity.

4. The method according to claim 1, characterized in that: The step of inputting the glass surface information into an abnormality detection model and performing abnormality detection on the glass sample comprises: Performing feature space projection on the glass surface information to obtain a feature space image corresponding to the glass surface information; extracting candidate abnormal features in the feature space image; Based on the attribute information of the glass sample, a matching standard anomaly database is selected; the attribute information at least includes: glass material type, shooting environment information, and glass material characteristics; Comparing the candidate abnormal feature with the standard abnormal feature preset in the standard abnormal database to obtain a comparison evaluation score corresponding to the candidate abnormal feature; If the comparison evaluation score reaches a set threshold, it is determined that the candidate abnormal feature belongs to an abnormal scratch.

5. The method according to claim 1, characterized in that: Generating a cleaning strategy corresponding to the glass sample by using a cleaning planning model includes: Performing a comprehensive scan on the glass sample by using a laser scanning module in a visual inspection system to obtain depth feature information of the glass sample; Based on the depth feature information, construct a three-dimensional point cloud model of the glass sample; A planning prediction branch in a cleaning planning model that matches the three-dimensional point cloud model is used to generate a cleaning strategy corresponding to the glass sample; wherein the planning prediction branch matches the shape and size parameters of the glass sample in the three-dimensional point cloud model, the cleaning planning model includes multiple planning prediction branches, and each planning prediction branch is used to adaptively construct a corresponding cleaning strategy based on the shape of the glass sample; the cleaning strategy includes at least: the motion trajectory, motion speed, angle, detergent spraying amount, and spraying angle of the wiping head.

6. The method according to claim 5, characterized in that The step of adopting a strategy branch in the cleaning planning model that matches the three-dimensional point cloud model to generate a cleaning strategy corresponding to the glass sample includes: If the shape of the glass sample in the three-dimensional point cloud model is a rectangle, obtaining any corner point in the three-dimensional point cloud model through the first planning prediction branch; Taking the corner point as the starting point, the cleaning strategy is set as follows: moving in a straight line at a constant speed along the long side of the three-dimensional point cloud model, and using adaptive wiping pressure and detergent spraying amount to perform wiping operations; When moving to the end point of the long side, turning and moving in a straight line along the short side direction of the three-dimensional point cloud model, and using adaptive wiping pressure and detergent spraying amount to perform wiping operations; When the movement reaches the end point of the short side, repeat the steps of moving in a straight line at a constant speed along the long side direction of the three-dimensional point cloud model and performing the wiping operation with an adaptive wiping pressure and a spraying amount of a cleaning agent until the cleaning process of the glass sample is completed; The wiping pressure and the amount of detergent sprayed are adaptively configured based on the degree of dirt on the glass surface detected in real time during movement.

7. The method according to claim 5, characterized in that The step of adopting a strategy branch in the cleaning planning model that matches the three-dimensional point cloud model to generate a cleaning strategy corresponding to the glass sample includes: If the shape of the glass sample of the three-dimensional point cloud model is circular, determining the center position of the circle of the three-dimensional point cloud model through the second planning prediction branch; Taking the center of the circle as the starting point, the cleaning strategy is set as follows: wiping is performed outward along the path of the Archimedean spiral; During the wiping process, the wiping pressure is adjusted according to the radius length, and the amount of detergent sprayed is increased according to a preset functional relationship with the radius length.

8. The method according to claim 5, characterized in that The step of adopting a strategy branch in the cleaning planning model that matches the three-dimensional point cloud model to generate a cleaning strategy corresponding to the glass sample includes: If the glass sample shape of the three-dimensional point cloud model is irregular, segmenting the contour of the three-dimensional point cloud model into a plurality of line segments and / or curve segments through a third planning prediction branch; Predicting a cleaning sequence corresponding to the three-dimensional point cloud model; The wiping operation is performed sequentially along a plurality of line segments and / or curve segments based on the cleaning sequence.

9. The method according to claim 8, characterized in that The wiping operation is performed sequentially along a plurality of line segments and / or curve segments based on the cleaning sequence, comprising: When wiping a line segment, the linear movement is performed at a constant speed and the linear wiping operation is performed with adaptive wiping pressure and detergent spraying amount; and / or When wiping a curve segment, the motion trajectory, speed and angle of the wiping head are adjusted by a motion control algorithm according to the curvature and torsion of the curve segment; wherein the speed scaling factor of the wiping head is expressed as: ; in, , is the weight parameter used to adjust the curvature and torsion The impact on speed, is the minimum speed.

10. A glass cleaning device, characterized in that: The glass cleaning device is applied to a wiping robot, which at least includes: a robot body, a mechanical arm, a visual detection system, a wiping head, a detergent spraying system, a base, and a main control computer. The glass cleaning device at least includes the following units, wherein A collection unit, for collecting glass surface information corresponding to a glass sample to be tested; the glass surface information at least includes glass surface images under multiple light source systems; the multiple light source systems correspond to multiple specific illumination angles and multiple illumination intensities, respectively; A detection unit, used for inputting the glass surface information into an abnormality detection model to perform abnormality detection on the glass sample; A stopping unit, used for stopping the cleaning process of the glass sample if the glass sample has abnormal scratches and the abnormal scratches exceed a set damage threshold; A generation unit, for generating a cleaning strategy corresponding to the glass sample through a cleaning planning model if the glass sample has no abnormal scratches or the abnormal scratches do not exceed a set damage threshold; wherein the cleaning planning model includes a plurality of planning prediction branches, each planning prediction branch is used to adaptively construct a corresponding cleaning strategy based on the shape of the glass sample; the cleaning strategy includes at least: a motion trajectory, a motion speed, an angle, a spray amount of a cleaning agent, and a spray angle of a wiping head; An execution unit is used to control a wiping head and a cleaning agent spraying system of a wiping robot based on the cleaning strategy to execute a cleaning process for the glass sample.

Images