Photovoltaic glass assembly assembly method and related device

Abstract

The present application provides an assembly method and related devices for photovoltaic glass components. The method: scan the stacking area to obtain the position information of the backplane glass in the stacking area; select the target backplane glass, and locate the target backplane glass based on the position information; use the planning model to plan the path of the located target backplane glass to obtain the initial motion path of the target backplane glass; during the movement, use the path optimization model to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator to obtain the optimized motion path of the target backplane glass; control the manipulator based on the initial motion path or the optimized motion path, so that the manipulator completes the assembly operation of the photovoltaic glass component. The embodiment of the present application realizes the automated lamination of the backplane glass, improves the production efficiency, and also ensures the accuracy and safety of the lamination operation.

Description

Technical Field

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

[0002] In the production of photovoltaic modules, photovoltaic glass needs to be transported and stacked to complete the assembly of photovoltaic glass.

[0003] In the related technologies, the traditional backplane glass assembly operation mainly relies on manual operation, which has problems such as low efficiency, poor precision and potential safety hazards. Manually transporting the backplane glass from the stacking area to the stacking area is not only time-consuming and labor-intensive, but also prone to glass breakage or uneven stacking due to position deviation or improper operation. In addition, manual path planning is not optimized enough, and it is difficult to deal with complex situations such as irregular stacking and tilted glass, and cannot meet the needs of modern and efficient production.

[0004] Therefore, it is urgent to design a new technical solution to solve at least one of the technical problems in the related art of low efficiency in manual handling of back panel glass, insufficient positioning accuracy, and easy damage to the glass. Summary of the invention

[0005] In order to solve or improve the above-mentioned problems, the embodiments of the present application provide a method for assembling a photovoltaic glass component and related devices to realize the automated lamination of back panel glass, thereby improving production efficiency and ensuring the accuracy and safety of the lamination operation.

[0006] In one embodiment of the present application, a method for assembling a photovoltaic glass assembly is provided. The method is applied to an assembly scenario of a photovoltaic glass assembly. The photovoltaic glass assembly is a plurality of backplane glasses in a stacking area, and includes:

[0007] Scan the stacking area to obtain the position information of the back glass in the stacking area; the position information at least includes: the glass size, position coordinates, and stacking area layout of each back glass;

[0008] Selecting a target backplane glass, and positioning the target backplane glass based on the position information;

[0009] Using a planning model, path planning is performed on the located target backplane glass to obtain an initial motion path of the target backplane glass; the initial motion path at least includes initial motion paths corresponding to a plurality of assembly operations of the manipulator, and each initial motion path includes a plurality of key nodes;

[0010] During the movement, a path optimization model is used to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator to obtain an optimized motion path for the target backplane glass;

[0011] The robot is controlled based on the initial motion path or the optimized motion path so that the robot completes the assembly operation of the photovoltaic glass component.

[0012] In another embodiment of the present application, a photovoltaic glass assembly assembly device is provided. The device is applied to an assembly scenario of a photovoltaic glass assembly. The photovoltaic glass assembly is a plurality of backplane glasses in a stacking area. The device comprises at least the following units:

[0013] A scanning unit, used for scanning the stacking area to obtain the position information of the back plate glass in the stacking area; the position information at least includes: the glass size, position coordinates and stacking area layout of each back plate glass;

[0014] A selection unit, used for selecting a target back panel glass and positioning the target back panel glass based on the position information;

[0015] A planning unit, configured to use a planning model to perform path planning on the located target backplane glass to obtain an initial motion path of the target backplane glass; the initial motion path at least includes initial motion paths corresponding to a plurality of assembly operations of the manipulator, and each initial motion path includes a plurality of key nodes;

[0016] An optimization unit, used to use a path optimization model to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator during the movement, so as to obtain an optimized motion path of the target backplane glass;

[0017] A control unit is used to control the manipulator based on the initial motion path or the optimized motion path so that the manipulator completes the assembly operation of the photovoltaic glass component.

[0018] In one embodiment of the present application, an electronic device includes:

[0019] at least one processor, memory, and input-output unit;

[0020] 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 method for assembling the photovoltaic glass component in the first embodiment.

[0021] 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 method for assembling the photovoltaic glass assembly in the first embodiment described above.

[0022] In the technical solution provided by the embodiment of the present application, the stacking area is scanned to obtain the position information of the backplane glass in the stacking area; the position information at least includes: the glass size, position coordinates, and stacking area layout of each backplane glass; the target backplane glass is selected, and the target backplane glass is positioned based on the position information; the planning model is used to plan the path of the positioned target backplane glass to obtain the initial action path of the target backplane glass; the initial action path at least includes the initial action paths corresponding to the multiple assembly operations of the manipulator, and each initial action path contains multiple key nodes; during the movement process, the path optimization model is used to perform path optimization on the initial action path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator to obtain the optimized action path of the target backplane glass; the manipulator is controlled based on the initial action path or the optimized action path, so that the manipulator completes the assembly operation of the photovoltaic glass assembly. The embodiment of the present application can realize the automatic lamination of the backplane glass, improve the production efficiency, and also ensure the accuracy and safety of the lamination operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] 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 paying any creative work.

[0024] Figure 1 A schematic diagram of the structure of an assembly device for a photovoltaic glass assembly provided in one embodiment of the present application;

[0025] Figure 2 A schematic diagram of a process for assembling a photovoltaic glass assembly provided in one embodiment of the present application;

[0026] Figure 3 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0027] To this end, 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 described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application.

[0028] 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 the order of precedence, nor do they limit "first" and "second" to be different types. In addition, the embodiments described below are only some embodiments of the present application, not all 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.

[0029] In the related technologies, the traditional backplane glass assembly operation mainly relies on manual operation, which has problems such as low efficiency, poor precision and potential safety hazards. Manually transporting the backplane glass from the stacking area to the stacking area is not only time-consuming and labor-intensive, but also prone to glass breakage or uneven stacking due to position deviation or improper operation. In addition, manual path planning is not optimized enough, and it is difficult to deal with complex situations such as irregular stacking and tilted glass, and cannot meet the needs of modern and efficient production.

[0030] Therefore, it is urgent to design a new technical solution to solve at least one of the technical problems in the related art of low efficiency in manual handling of back panel glass, insufficient positioning accuracy, and easy damage to the glass.

[0031] In order to solve the above technical problems, an assembly device of a photovoltaic glass assembly is provided in one embodiment of the present application. The assembly device of the photovoltaic glass assembly includes the following units:

[0032] A scanning unit, used for scanning the stacking area to obtain the position information of the back plate glass in the stacking area; the position information at least includes: the glass size, position coordinates and stacking area layout of each back plate glass;

[0033] A selection unit, used for selecting a target back panel glass and positioning the target back panel glass based on the position information;

[0034] A planning unit, configured to use a planning model to perform path planning on the located target backplane glass to obtain an initial motion path of the target backplane glass; the initial motion path at least includes initial motion paths corresponding to a plurality of assembly operations of the manipulator, and each initial motion path includes a plurality of key nodes;

[0035] An optimization unit, used to use a path optimization model to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator during the movement, so as to obtain an optimized motion path of the target backplane glass;

[0036] A control unit is used to control the manipulator based on the initial motion path or the optimized motion path so that the manipulator completes the assembly operation of the photovoltaic glass component.

[0037] In an embodiment of the present application, the above-mentioned device is applied to an assembly scenario of a photovoltaic glass assembly, and the photovoltaic glass assembly is a plurality of back panel glasses in a stacking area. Figure 1 The schematic diagram of the structure of the photovoltaic glass assembly assembly provided by one embodiment of the present application is shown. Figure 1 As shown, an embodiment of the present application proposes an assembly device for a photovoltaic glass component, which aims to realize the automated lamination of back panel glass, improve production efficiency, and ensure the accuracy and safety of lamination operations. Figure 1 In the figure, back glass pile 1, workbench 2, robot 3, industrial camera and laser sensor 4, suction cup 5, glass to be laminated 6.

[0038] In the embodiment of the present application, the manipulator uses a six-degree-of-freedom robot arm equipped with a precision vacuum suction cup and a force sensor. The manipulator sucks the glass through the suction cup, and the suction force and suction cup size can be automatically adjusted according to the size, thickness and surface conditions of the glass. The manipulator adjusts the suction force to ensure stable gripping of the back panel glass to avoid glass breakage.

[0039] In the embodiment of the present application, a visual guidance system can be further introduced, equipped with a high-precision industrial camera and a laser sensor, to obtain image data of the glass stacking area in real time. The visual system not only recognizes the shape, size and position of the glass, but also determines the stacking direction and surface state (such as smoothness or defects) of the back glass. At the same time, the system is also responsible for real-time monitoring of the specific position and posture of the glass to be laminated.

[0040] In the embodiment of the present application, the manipulator can be mounted on a workbench. The workbench can adjust the angle and position of the glass to accommodate back glass of different sizes and shapes. The back glass stacking area and the area to be laminated are set up separately to ensure smooth and interference-free operation. The stacking area is marked with clear coordinate points to facilitate the positioning of the manipulator.

[0041] In the embodiment of the present application, the embedded control system integrates the genetic algorithm and the path planning module. The system optimizes the path in real time according to the visual recognition data, so that the robot can accurately and efficiently grab the back glass and stack it to the target position.

[0042] Therefore, an embodiment of the present application proposes an assembly device for photovoltaic glass assemblies, which obtains detailed position information of the glass back panel by scanning the stacking area, including the glass size, position coordinates, and stacking area layout. This allows for greater accuracy in selecting and positioning the target glass back panel, avoiding assembly errors caused by inaccurate position information. For example, precise position coordinates allow the manipulator to accurately grasp the glass back panel, and understanding the glass size helps the manipulator adjust the grasping force and angle to ensure the stability of the glass back panel during the grasping and moving process. Here, the initial motion path generated by the planning model contains multiple key nodes, which clarify the various operating positions of the manipulator during the assembly process, making the manipulator's movements more precise, further improving the accuracy of the assembly of photovoltaic glass assemblies.

[0043] The above device can quickly obtain the initial motion path of the target glass back panel by using the path planning model, reducing the time for manual path planning. The manipulator can perform assembly operations in an orderly manner according to the preset path, avoiding unnecessary movements and waiting time. The path optimization model can optimize the initial motion path according to the glass size, stacking area layout and preset working range of the manipulator. For example, by optimizing the path, it is possible to avoid collisions between the manipulator and other objects in the stacking area during movement, reduce the time loss caused by collision detection and avoidance, and thus improve the efficiency of the entire assembly process.

[0044] The above device also takes into account factors such as the stacking area layout and the preset working range of the manipulator, so that the method can adapt to different working scenarios. Whether the layout of the stacking area changes or the manipulator of different specifications is replaced, the assembly strategy can be adjusted through scanning and path optimization. For glass back panels of different sizes, path planning and optimization can also be performed based on the glass size, ensuring that the manipulator can flexibly complete the assembly operation, expanding the scope of application of the assembly method and device.

[0045] During the path optimization process, the above device takes into account the preset working range of the manipulator, which can effectively prevent the manipulator from exceeding its safe working range or colliding with surrounding equipment and objects. This not only protects the manipulator itself and reduces the risk of equipment damage, but also ensures the smooth progress of the assembly process and improves the reliability of the entire assembly system. Accurate positioning and path planning also help reduce damage to the glass backplane caused by improper operation and improve product safety during the assembly of photovoltaic glass modules.

[0046] During operation, the robot will monitor changes in the environment in real time through sensors and vision systems. If an obstacle or emergency situation (such as glass tilt or position deviation) is encountered, the genetic algorithm will dynamically adjust the path, optimize the path order or change the direction of movement to avoid unnecessary collisions in the path. The genetic algorithm calculates the optimal obstacle avoidance path in real time to ensure that the robot can efficiently and seamlessly complete the grasping, moving and stacking of glass in complex environments. The self-learning ability of the genetic algorithm enables path optimization to be dynamically adjusted according to historical execution data, gradually improving the efficiency and stability of path selection. Specifically, the self-learning mechanism includes several aspects such as adjustment of fitness function, optimization of path selection, optimization of grasping and placement points, path stability and execution efficiency.

[0047] Each time a path is executed, the system records the execution data, including key parameters such as path length, time consumption, grasping success rate, and placement accuracy. The data will be used for subsequent optimization. Record the total distance the robot moves from the grasping point to the placement point. If the path is too long, it will be marked, and the system will automatically adjust the path planning to avoid ineffective long-distance movement. The time for each path execution includes the entire process from grasping to placement. Paths that consume too much time will be considered unsatisfactory, and the system will readjust the path planning to shorten the execution time and improve production efficiency. When grasping the backplane glass, the success rate of the robot is an important indicator of path quality. A low success rate may be due to poor path planning, which causes the robot to be unable to grasp accurately or slip, thereby affecting subsequent operations. The system will optimize the path planning of the grasping point based on the feedback data of each grasping to ensure that each grasping is successful. The placement accuracy is directly related to the accurate stacking of the backplane glass, avoiding the difficulty of subsequent operations caused by placement errors. When the accuracy is low, the system will adjust the path of the placement point to ensure that each placement is accurately docked to the specified position. By monitoring the accuracy of each path execution, the system can continuously improve the positioning of placement points and path planning to reduce errors.

[0048] The device provided in the embodiment of the present application can realize the automatic lamination of the back panel glass, improve the production efficiency, and ensure the accuracy and safety of the lamination operation.

[0049] It should be added here that the control logic of the relevant device will be introduced in detail in the method embodiment below.

[0050] An embodiment of the present application provides a method for assembling a photovoltaic glass assembly. The method of this embodiment can be performed by the device described in the above embodiment, which is implemented on the basis of the structure of the embodiment of the assembly device of the photovoltaic glass assembly. Figure 2 As shown, the method may include the following steps:

[0051] 201. Scan the stacking area to obtain position information of the backplane glass in the stacking area; the position information at least includes: glass size, position coordinates, and stacking area layout of each backplane glass;

[0052] 202. Select a target backplane glass, and locate the target backplane glass based on the position information;

[0053] 203. Using the planning model, perform path planning on the located target backplane glass to obtain an initial motion path of the target backplane glass; the initial motion path at least includes initial motion paths corresponding to a plurality of assembly operations of the manipulator, and each initial motion path includes a plurality of key nodes;

[0054] 204. During the movement process, a path optimization model is used to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator to obtain an optimized motion path of the target backplane glass;

[0055] 205. Control a robot arm based on the initial motion path or the optimized motion path, so that the robot arm completes the assembly operation of the photovoltaic glass component.

[0056] The present application provides a method for assembling a photovoltaic glass assembly, which obtains detailed position information of a glass back panel by scanning a stacking area, including glass size, position coordinates, and stacking area layout. This allows for greater precision in selecting and positioning the target glass back panel, avoiding assembly errors caused by inaccurate position information. For example, precise position coordinates allow a robot to accurately grasp the glass back panel, and understanding the glass size helps the robot adjust the grasping force and angle to ensure the stability of the glass back panel during grasping and movement. The initial motion path generated by the planning model contains multiple key nodes, which clarify the various operating positions of the robot during the assembly process, making the robot's movements more precise, further improving the accuracy of the assembly of photovoltaic glass assemblies.

[0057] The path planning model can be used to quickly obtain the initial motion path of the target glass back panel, reducing the time for manual path planning. The robot can perform assembly operations in an orderly manner according to the preset path, avoiding unnecessary movements and waiting time. The path optimization model can optimize the initial motion path according to the glass size, stacking area layout and the preset working range of the robot. For example, by optimizing the path, the robot can avoid collisions with other objects in the stacking area during movement, reduce the time loss caused by collision detection and avoidance, and thus improve the efficiency of the entire assembly process.

[0058] The method can adapt to different working scenarios by taking into account factors such as the stacking area layout and the preset working range of the manipulator. Whether the layout of the stacking area changes or the manipulator of different specifications is replaced, the assembly strategy can be adjusted through scanning and path optimization. For glass back panels of different sizes, path planning and optimization can also be performed based on the glass size to ensure that the manipulator can flexibly complete the assembly operation, expanding the scope of application of the assembly method and device.

[0059] During the path optimization process, the preset working range of the manipulator is taken into consideration, which can effectively prevent the manipulator from exceeding its safe working range or colliding with surrounding equipment and objects. This not only protects the manipulator itself and reduces the risk of equipment damage, but also ensures the smooth progress of the assembly process and improves the reliability of the entire assembly system. Accurate positioning and path planning also help reduce damage to the glass backplane caused by improper operation and improve product safety during the assembly process of photovoltaic glass modules.

[0060] As an optional embodiment, in 201, scanning the stacking area to obtain position information of the back plate glass in the stacking area includes:

[0061] Collect real-time image data of each back panel glass in the stacking area; based on the real-time image data, identify the position, angle, stacking direction of each back panel glass, and the relative position relationship between each back panel glass; the identification process at least includes: edge detection and object contour detection; based on the identification result, construct a corresponding three-dimensional space model of the stacking area; the three-dimensional space model of the stacking area includes the glass size, position coordinates, and stacking area layout of each back panel glass.

[0062] Specifically, in the above embodiments, image information of each backplane glass in the stacking area is obtained through an image acquisition device (such as a camera). These image data contain a variety of information such as the appearance characteristics and position distribution of the backplane glass, which is the basis for subsequent analysis. For example, the camera can shoot the stacking area from different angles to obtain comprehensive visual information, just like the human eye observes an object and records the scene as an image. In practical applications, the image acquisition device can be installed on a manipulator or in the surrounding environment.

[0063] Edge detection is a key step in identifying the boundaries of objects in an image. For example, by using algorithms (such as the Canny edge detection algorithm) to find areas in the image where the brightness or color changes sharply, these areas are usually the edges of the backplane glass. Object contour detection further determines the complete shape of the object based on edge detection. For the backplane glass, the four edges of the glass can be found through edge detection, and then these edges are connected through object contour detection to form a complete glass backplane contour. This helps to accurately determine the position and shape of each backplane glass in the image. Furthermore, based on the results of edge detection and object contour detection, the position of the backplane glass can be further analyzed. By converting the image coordinate system with the actual space coordinate system, the actual position coordinates of the backplane glass in the stacking area can be calculated. In addition, in some examples, for object contour detection, an algorithm based on shape context or a level set method can be used. These algorithms can accurately outline the complete contour of the glass backplane under complex backgrounds and stacking conditions, and improve the recognition accuracy of the shape and size of the glass backplane. Further optionally, multiple recognition algorithms are combined to process the same image, and then the accuracy is improved by comparison and verification. For example, when a pixel gradient-based algorithm and a region growing-based algorithm are used simultaneously to identify a glass back panel, the accuracy is higher when the results obtained by the two algorithms are consistent. If there is a difference, the cause of the difference can be further analyzed, and the correct recognition result can be determined through manual intervention or other auxiliary means (such as referring to the recognition results of adjacent glass back panels).

[0064] Among them, the angle information can be obtained by analyzing the direction of the backplane glass contour. For example, the angle is determined based on the angle between the major axis or minor axis of the glass backplane and the reference coordinate axis. The stacking direction can be determined by observing the upper and lower relationship between the backplane glasses. For example, the stacking direction can be determined by judging whether the edge of one glass backplane is partially blocked by another glass backplane and the position of the blockage. The relative position relationship between each backplane glass is determined by comparing their position coordinates and contours, such as calculating the distance and direction between the center positions of the two glass backplanes, so as to understand how they are arranged in the stacking area.

[0065] Then, after obtaining information such as the position, angle, stacking direction, and relative position relationship of each backplane glass, this information is used to construct a three-dimensional spatial model of the stacking area. This model is similar to a virtual copy of the stacking area, in which each backplane glass has its corresponding spatial attributes, including glass size (determined by information such as length and width obtained by contour detection), position coordinates (obtained through position recognition), and stacking area layout (determined by the relative position relationship of each backplane glass). In this way, the two-dimensional image information is converted into three-dimensional spatial information that is more intuitive and more conducive to path planning and assembly operations.

[0066] In this way, the position and shape of the back glass can be accurately determined through detailed edge detection and object contour detection. Compared with simple estimation, this method can provide more accurate glass size information, for example, accurate to the millimeter level, thus providing more accurate data support for the robot's grasping and assembly operations, reducing unstable grasping or assembly errors caused by dimensional errors.

[0067] Next, the constructed three-dimensional spatial model includes information such as the relative position relationship and stacking direction between the various backplane glasses, so that the system can fully understand the status of the stacking area. This helps to better consider the situation of the surrounding glass backplanes when selecting the target glass backplane, avoid colliding with other backplane glasses during the grabbing process, and improve the safety and accuracy of the entire assembly process.

[0068] Regardless of whether the back panel glass is stacked neatly or messily, the method in this embodiment can accurately obtain its position information through image acquisition and analysis. For example, when the stacking angle and direction of the back panel glass are irregular, the state of each glass back panel can still be determined through edge detection and contour detection, thereby constructing an effective three-dimensional space model, so that the assembly method can adapt to the complex and changeable actual stacking situation. If the layout of the stacking area changes, such as adding other equipment or adjusting the stacking position of the back panel glass, you only need to re-capture the image and analyze it to quickly update the three-dimensional space model. This flexibility enables the assembly system to better adapt to the dynamic changes of the production environment and ensure the continuous and efficient progress of the assembly work.

[0069] The three-dimensional space model of the stacking area is a virtual model that digitally reproduces the actual situation of the stacking area. It mainly contains the key information of each backplane glass in the stacking area, such as the glass size (including length, width, thickness, etc.). These size information is an important factor in determining the physical form and space occupancy of the glass backplane. The position coordinates are used to accurately represent the position of each backplane glass in the three-dimensional space of the stacking area, usually based on a pre-set spatial coordinate system, which can be based on the boundary of the stacking area or a fixed reference point. For example, with a corner of the stacking area as the origin, the horizontal direction is the x-axis, the vertical direction is the y-axis, and the height direction is the z-axis. The x, y, and z values ​​corresponding to the center position of each backplane glass are its position coordinates. The stacking area layout describes the arrangement and relationship of the backplane glass in the entire stacking area, including how the various glass backplanes are stacked, whether they are neatly arranged or staggered, and the spacing between them.

[0070] In terms of path planning, when assembling photovoltaic glass modules, the robot needs to grab the target glass backplane from the stacking area and move it to the assembly position. The position coordinates and stacking area layout information in the three-dimensional space model can help plan a reasonable grabbing path to avoid collisions between the robot and other glass backplanes or obstacles in the stacking area. For example, the model can calculate the shortest collision-free path from the starting position to the target glass backplane, while taking into account the position and stacking method of the surrounding glass backplanes, and reasonably adjust the movement amplitude and direction of the robot.

[0071] In terms of target selection decisions, the system can quickly determine the most suitable target glass back panel to be grasped based on the stacking area layout and the position coordinates of the glass back panel. For example, the system will prioritize targets located on the top layer of the stacking area, which are easy to grasp and will not interfere with the positions of other glass back panels, thereby improving assembly efficiency.

[0072] In practical applications, the 3D spatial model provides an intuitive tool for operators to understand the spatial cognition of the stacking area. Through the visual model, the distribution of the glass back panels in the stacking area can be clearly seen, just like seeing a virtual copy of the stacking area. This helps operators better understand the entire assembly environment and discover potential problems in advance, such as whether some glass back panels are stacked too high and may cause instability. This model is also very useful when debugging and optimizing assembly processes. Different assembly strategies and robot motion paths can be simulated on the model to observe possible situations without repeated trials in the actual stacking area, thus saving time and resources.

[0073] For example, the three-dimensional model of the stacking area can be used as a data integration platform to integrate information about the stacking area from different sources. In addition to the basic information of the glass backplane, information such as ambient temperature and humidity that may affect the glass performance or assembly process can also be integrated.

[0074] At the same time, this model facilitates data sharing between different automation system components. For example, the robot control system, inventory management system, etc. can access the data in this model, allowing various systems to work better together. The robot control system can accurately grasp the glass back panel according to the position of the glass back panel in the model, and the inventory management system can update the inventory status of the glass back panel in real time according to the model.

[0075] As an optional embodiment, in the above steps, after constructing the corresponding three-dimensional space model of the stacking area based on the recognition result, the following steps are further included:

[0076] Control the laser sensor and the depth camera to obtain the relative position relationship and surface morphology data of each back panel glass by multi-angle scanning; set the scanning frequency to a target scanning frequency that matches the operating speed of the back panel glass assembly line; determine the thickness of each back panel glass based on the relative position relationship of each back panel glass; perform surface texture analysis based on the surface morphology data to obtain the surface flatness of each back panel glass; perform defect analysis based on the surface flatness and the surface morphology data to identify whether each back panel glass has a preset type of defect; the preset type includes at least: scratches, cracks, bubbles.

[0077] Specifically, the laser sensor and the depth camera can provide spatial information about the backplane glass. By scanning from multiple angles, a more comprehensive relative position relationship of each backplane glass can be obtained. For example, the laser sensor calculates the distance by emitting a laser beam and receiving the reflected light, and the depth camera can directly obtain the depth information of the surface of the object. Their collaborative work can construct more accurate three-dimensional spatial data. The scanning frequency matches the operating speed of the backplane glass assembly line to ensure that the data of each backplane glass can be obtained in time during the operation of the line. For example, if the line speed is fast, the scanning frequency is also increased accordingly to ensure that the information of any backplane glass passing by is not missed. Further, according to the relative position relationship of each backplane glass obtained, the thickness of each glass is determined by measuring the distance difference between adjacent glass surfaces. For example, in stacked glass, the outer surface position coordinates of two adjacent glasses are known, and the thickness of the glass can be obtained by a simple subtraction operation. The surface morphology data obtained by the depth camera is used to judge the surface flatness by analyzing the fluctuation of the data. If the data of a certain area of ​​the surface changes greatly, it means that the area may be uneven. For example, surface flatness can be quantified by calculating statistical indicators such as the variance or standard deviation of surface height data.

[0078] For the identification of scratches, the linear abnormal areas in the surface morphology data are analyzed to judge. These abnormal areas are shown as sudden changes in depth or height in the data, and have certain length and width characteristics. The recognition principle of cracks is similar, but cracks usually cause more obvious depth changes and may present irregular branching shapes in morphology. The identification of bubbles is mainly through finding local raised areas in the surface morphology data. The shapes of these areas are usually approximately circular, and there are obvious differences from the surrounding areas in the depth data. The interior may be hollow features, thereby identifying bubbles. In this way, the thickness of the backplane glass can be accurately measured to ensure that the thickness of each piece of glass meets the production standards. For photovoltaic glass components, the consistency of thickness is an important factor in ensuring its performance and quality. Accurate thickness detection helps to screen out glass that does not meet the thickness requirements and avoid quality problems caused by thickness differences after assembly. Through surface flatness analysis and defect identification, backplane glass with scratches, cracks or bubbles on the surface can be effectively screened out. These defects may affect the optical properties, mechanical strength, etc. of photovoltaic glass components. Detecting and eliminating defective glass in advance can greatly improve the quality and reliability of the final product.

[0079] After obtaining the thickness and surface condition of the back glass, the assembly process can be optimized. For example, for glass with slightly different thicknesses, the gripping force and assembly position of the robot can be adjusted to ensure assembly accuracy and stability. For glass with uneven surfaces, special assembly measures can be taken, such as adding buffer materials or adjusting the assembly angle, to increase the success rate of assembly.

[0080] The data obtained from these tests can be accumulated as production data and used to analyze the quality trend of the entire production process. For example, by counting the frequency and location of scratches, cracks and bubbles over a period of time, the production links that may cause these defects can be found, so as to make targeted improvements to the production process. At the same time, these data can also be used for quality traceability. If problems are found during subsequent product inspections or use, these test data can be traced back to find the source of the problematic glass and possible problematic links.

[0081] As an optional embodiment, in the above steps, defect analysis is performed based on the surface flatness and the surface morphology data to identify whether each back panel glass has a preset type of defect, including:

[0082] For the current backplane glass, extracting a high-resolution image from the surface morphology data;

[0083] Through the first branch of the defect detection model, texture analysis is performed on the high-resolution image in combination with the surface flatness to extract corresponding texture features; the texture features are used to indicate the texture changes at various locations on the current backplane glass surface;

[0084] Comparing the texture features with a pre-recorded standard texture change trend to identify whether there are scratches on the current back panel glass;

[0085] Extracting candidate bubble positions with a height variation greater than a set threshold from the high-resolution image through the second branch of the defect detection model, and performing depth image analysis on the candidate bubble positions to identify bubbles existing on the surface or inside of the current backplane glass;

[0086] Through the third branch of the defect detection model, Fourier transform analysis is performed on the high-resolution image to obtain the frequency spectrum characteristics corresponding to the high-resolution image, and based on the frequency spectrum characteristics, it is identified whether there are cracks in the current back panel glass.

[0087] In the embodiment of the present application, the high-resolution image is extracted from the surface morphology data of the back glass. It is an image that can show the microscopic features of the glass surface in detail, with high pixel density and clarity. For example, the number of pixels is sufficient to distinguish the small texture changes, flaws and other details on the glass surface. Generally, its resolution is higher than that of ordinary images used to observe the general appearance, which may reach thousands of pixels per inch (dpi) or even higher, and such a resolution helps to accurately analyze the situation of the glass surface. The high-resolution image is collected from the surface morphology data of the back glass. These data may be obtained through a depth camera or other high-precision surface detection equipment. These devices can measure the depth, height and other information of each point on the glass surface, and then convert these data into image form. For example, taking depth information as an example, points of different depths are represented by different grayscale values ​​or colors, points with large depths may be represented by darker colors, and points with small depths are represented by brighter colors, thereby forming an image that can reflect the surface morphology.

[0088] High-resolution images play a key role in defect analysis. They provide a detailed visual basis for subsequent texture analysis, bubble and crack identification. For texture analysis, high-resolution images can clearly present information such as the direction and degree of change of the glass surface texture. For example, when identifying scratches, it is possible to accurately see whether the texture is damaged and the degree and shape of the damage. When identifying bubbles and cracks, high-resolution images can help locate areas where defects may exist. For example, areas with local color or grayscale abnormalities may be the location of bubbles or cracks, providing specific target areas for further deep image analysis and spectral feature analysis.

[0089] Specifically, after extracting a high-resolution image from the surface morphology data, the first branch of the defect detection model is used for texture analysis. Texture is an inherent property of the surface of an object, and a normal photovoltaic glass backplane has its own specific texture change trend. Surface flatness information plays an auxiliary role here, because scratches will change the flatness of the surface, thereby affecting the texture characteristics. For example, the local binary pattern (LBP) method is used to extract texture features. LBP generates a binary code to describe the local texture structure by comparing the grayscale values ​​of the central pixel with its neighboring pixels. When there are scratches on the glass surface, the LBP features of the scratched area will be significantly different from those of the normal texture area.

[0090] Then, the extracted texture features are compared with the pre-recorded standard texture change trend. The standard texture change trend is established by analyzing the texture of a large number of scratch-free high-quality glass backplanes. If the texture features of the current backplane glass deviate significantly from the standard texture, such as the continuity of the texture is destroyed, the texture direction or density changes suddenly, etc., it can be judged that there are scratches.

[0091] Alternatively, the second branch of the defect detection model can be used to extract locations with height changes greater than a set threshold from high-resolution images as candidate bubble locations. This is because bubbles will cause local height changes whether they are on the glass surface or inside. For example, when a depth camera collects surface morphology data, the depth value (height value) of the bubble area will be different from that of the surrounding normal glass area. By setting a reasonable height change threshold, areas where bubbles may exist can be screened out.

[0092] Perform depth image analysis on candidate bubble locations. Since bubbles will form specific shapes and grayscale change patterns on the depth image, these patterns can be analyzed to further determine whether they are bubbles. For example, bubbles may appear as local convex or concave areas that are approximately circular in shape on the depth image, and their grayscale value changes also have certain regularities. By analyzing these features, bubbles can be accurately identified.

[0093] Alternatively, the third branch of the defect detection model can be used to perform Fourier transform analysis on the high-resolution image. Fourier transform can convert the image from the spatial domain to the frequency domain, and cracks have specific frequency characteristics in the frequency domain. The presence of cracks will change the frequency components of the image, for example, cracks usually produce changes in high-frequency components.

[0094] Cracks are identified by analyzing the spectral features corresponding to high-resolution images. In the frequency domain, cracks may appear as enhancements or abnormal distributions of certain specific frequency components. Compared with the spectral features of a normal glass backplane, if there is a significant difference in a certain frequency range or certain frequency ranges, it can be inferred that there are cracks. At the same time, the morphological features of cracks can also be learned by training a convolutional neural network (CNN) model. The CNN model can automatically extract morphological information such as the length, width, and direction of the cracks, further improving the accuracy of crack detection. The embodiment of the present application combines multiple methods such as texture analysis, deep image analysis, and Fourier transform analysis to detect defects in backplane glass from different angles. For three different types of defects, scratches, bubbles, and cracks, targeted detection methods are used to avoid the limitations of a single detection method and to more accurately identify various defects. In scratch detection, the combined effect of scratches on surface texture and flatness is considered in combination with the trend of surface flatness and standard texture changes, reducing the possibility of misjudgment. For example, some slight stains may affect the texture but do not change the surface flatness, and this comprehensive judgment can distinguish them from scratches.

[0095] By constructing three branches of the defect detection model, an automated defect detection process is realized. Once there is a backplane glass that needs to be inspected, the model can automatically extract high-resolution images and perform texture analysis, candidate bubble position extraction and depth image analysis, Fourier transform analysis and other operations according to the preset process, which greatly improves the detection efficiency. It is especially suitable for photovoltaic glass backplane inspection on large-scale production lines. This automated detection method can adapt to different production environments and glass types. As long as the standard texture change trend is established in advance and the parameters such as the height change threshold are adjusted according to the actual situation, defects can be detected on photovoltaic glass backplanes of different sizes and materials, which has strong versatility.

[0096] As an optional embodiment, in 203, a planning model is used to perform path planning on the located target backplane glass to obtain an initial motion path of the target backplane glass, including:

[0097] Based on the position information, determine the position, surface state, candidate grasping angles, and relative position relationship between the target back-panel glass and other back-panel glasses; based on the position, surface state, candidate grasping angles, and relative position relationship between the target back-panel glass and other back-panel glasses, simulate the assembly operation to be performed when the target back-panel glass moves from the current position to the target placement point in the area to be laminated, and generate multiple path segment combinations containing the assembly operation; optimize the multiple path segment combinations through a path fitness function to obtain an initial motion path of the target back-panel glass; the encoding information of the initial motion path includes at least: position encoding of multiple key nodes, obstacle encoding, and target placement point encoding in the area to be laminated.

[0098] Specifically, based on the previously obtained position information, the position of the target backplane glass is determined, which is the starting point of the planned path. At the same time, its surface condition, such as whether it is flat and whether there are defects, is considered, because these will affect the gripping method and stability of the manipulator. The candidate gripping angles are determined based on the shape of the backplane glass, the surrounding environment, and the structure of the manipulator. Different gripping angles may have different effects on subsequent movement and placement operations. In addition, the relative position relationship between the target backplane glass and other backplane glasses is clarified, which helps avoid collisions with other glasses during gripping and movement.

[0099] According to the starting position, surface state, gripping angle and relative position of the surrounding glass of the target backplane glass, the assembly operation of moving it from the current position to the target placement point in the area to be laminated is simulated in the virtual environment. This process generates multiple possible path segment combinations. Each path segment combination represents a way of moving from the starting point to the end point, which may include different moving directions, height changes, and rotation operations. For example, during the simulation, consider moving the robot to a safe height above the target backplane glass, then vertically descending to a suitable gripping position, then rotating and lifting the glass at a certain angle, and finally moving the glass to the target placement point along a route that avoids other glass and obstacles.

[0100] Use the path fitness function to evaluate and optimize multiple path segment combinations. This function takes into account multiple factors, such as path length, whether to avoid obstacles, and smoothness of movement. The shorter the path length, the less time and energy may be consumed during the movement; avoiding obstacles can ensure the safety of operation; and smoothness of movement is very important to prevent glass damage. By scoring each path segment combination, the combination with the highest score (i.e. the best fitness) is selected as the initial motion path of the target backplane glass. The encoding information of this initial motion path contains the position encoding of multiple key nodes, which can be important locations such as grasping points, rotation points, and obstacle avoidance points; obstacle encoding is used to accurately identify and avoid obstacles during movement; and the target placement point encoding in the area to be laminated clarifies the final placement position.

[0101] Furthermore, by comprehensively considering factors such as the position, surface state, gripping angle, and surrounding environment of the target backplane glass, the generated initial motion path is more in line with the actual situation. This ensures that the manipulator can operate in the best way during the gripping and moving process, improving assembly efficiency. For example, choosing a suitable gripping angle can make it easier for the manipulator to grip the glass stably and reduce the probability of gripping failure.

[0102] After optimization using the path fitness function, the initial motion path achieved a good balance in many aspects. It can shorten the path length as much as possible and reduce the assembly time while avoiding obstacles and other glass. At the same time, considering the smoothness of movement helps protect the glass back panel and reduces the risk of glass damage during movement.

[0103] In this way, the coded information of the initial motion path provides clear guidance for the control of the robot. The position coding of key nodes enables the robot to accurately perform each important operation step, such as grabbing, rotating or placing at a specified location. Obstacle coding can help the robot identify and avoid possible obstacles in real time, improving the safety of operation. The target placement point coding ensures that the glass can be accurately placed at the specified position of the area to be laminated, improving the accuracy of assembly.

[0104] Exemplarily, the path fitness function is expressed as the following formula:

[0105]

[0106] in, is the weight parameter of each sub-item, L represents the total length of the path from the current position to the target placement point in the area to be stacked under the combination of path segments, It is used to measure the degree of smoothness in the combination of path segments. It represents the minimum distance between the path segment combination and the obstacle, and is used to indicate the obstacle avoidance capability of the path segment combination. The grasping and placement accuracy P is used to measure the error between the current position of the manipulator and the target placement point.

[0107] In the above formula, path length is an important consideration in the path fitness function. The total length L of the path from the current position of the manipulator to the target placement point directly affects the efficiency of the assembly operation. In actual operation, a shorter path length means that the manipulator can complete the task faster, reducing the time consumption of the entire assembly process. For example, in an industrial production line, time is productivity, and shortening the path length can improve production efficiency. The smoothness part is used to measure the smoothness in the path segment combination. During the movement of the manipulator, a smooth path can reduce the vibration and impact of the manipulator arm, which is crucial for the stability of the grasped photovoltaic glass backplane. If the path is not smooth, the manipulator may suddenly accelerate, decelerate or turn during the movement, which will cause damage to the glass backplane, especially for products such as photovoltaic glass that have high requirements for surface quality and structural integrity. The obstacle avoidance ability part represents the minimum distance between the path segment combination and the obstacle, which is used to indicate the obstacle avoidance ability of the path segment combination. During the assembly process, there may be obstacles such as other backplane glass or other equipment in the stacking area. The manipulator must be able to effectively avoid these obstacles to prevent collision accidents. This parameter ensures that the path planning can ensure that the robot can bypass obstacles at a safe distance to avoid damage to itself and the surrounding environment. The grasping and placement accuracy is used to measure the error between the robot's current position and the target placement point. When grasping the target backplane glass, the robot needs to accurately locate and grasp it, and also needs to accurately place the glass to the target placement point during the placement process. High-precision grasping and placement can ensure the assembly quality of photovoltaic glass modules and avoid assembly failures or product quality problems caused by position deviations.

[0108] The formula of this path fitness function can comprehensively consider multiple key factors in path planning and calculate a comprehensive fitness value. This makes path optimization no longer rely solely on a single factor, but can balance and optimize between path length, smoothness, obstacle avoidance, and grasping and placement accuracy. For example, a path may be short, but if it is not smooth or has weak obstacle avoidance, its fitness value may not be high and it will not be selected as the optimal path.

[0109] Due to the existence of weight parameters, the path fitness function can be easily adapted to different production demand scenarios. By adjusting the weights, path planning can focus more on different aspects such as efficiency, accuracy or safety. For example, in the early stages of production, when the feasibility of the assembly process needs to be quickly verified, the weight of the path length can be appropriately increased to quickly generate a feasible path; and in the mass production stage where product quality requirements are higher, the weights of the grasping and placement accuracy and obstacle avoidance capabilities can be increased to ensure high-quality assembly operations.

[0110] In this way, by considering the degree of smoothness and the accuracy of gripping and placement, the path fitness function helps to improve the assembly quality. A smooth path can prevent the glass backplane from being damaged during movement, while high gripping and placement accuracy can ensure that the backplane is accurately placed at the target position, reducing assembly errors and thus improving the overall quality of photovoltaic glass modules. Considering the path length enables the generated path to shorten the manipulator's movement path and reduce movement time as much as possible while ensuring other quality factors. At the same time, the optimization of obstacle avoidance also avoids time waste caused by unexpected situations such as collisions, thereby effectively improving assembly efficiency.

[0111] Further, based on the above formula, the calculation formula for the i-th path length L is:

[0112]

[0113] Among them, the coordinates of the current position are ( , , ), the coordinates of the target placement point are ( , , ), is the rotation radius of the manipulator, is the angle between adjacent path segments;

[0114] in, Used to measure the smoothness of the path segment combination, the path stability of the i-th path segment The calculation formula is:

[0115] ;in is the angle between adjacent path segments;

[0116] Among them, the grabbing and placing accuracy The calculation formula is ; The grasping and placing accuracy P is used to measure the error between the grasping point and the placing point of the robot. The higher the grasping and placing accuracy, the higher the accuracy of grasping and placing. is the error between the actual coordinates and the target coordinates.

[0117] In the path length calculation formula, the coordinate difference part is based on the idea of ​​the distance formula between two points. In three-dimensional space, the spatial distance from the starting point to the end point is preliminarily measured by calculating the sum of the squares of the coordinate differences between the target placement point and the current position in the three coordinate axis directions of x, y, and z. This is similar to the calculation of Euclidean distance and is the basis for determining the straight-line distance between two points.

[0118] The addition of the rotation radius and angle takes into account the rotation of the robot during movement. is the rotation radius of the robot and is the angle between adjacent path segments. When the robot rotates in the path, its motion trajectory will add additional length due to the rotation. By calculating the product of the rotation radius and the angle and adding it to the path length, the total distance actually moved by the robot in the entire path segment combination can be more accurately calculated. This calculation method takes into account the actual situation of the robot's movement, not just the simple straight-line distance, but also includes the path extension caused by rotation.

[0119] In this way, the actual length of the path can be more accurately evaluated through a detailed path length calculation formula. Factors such as the rotation radius and angle are taken into account, so that the calculation of the path length is more consistent with the motion trajectory of the manipulator in actual operation, avoiding underestimation of the path length, and providing more accurate length information for path planning. The calculation method of path smoothness can effectively evaluate the smoothness of the path during movement. This is very important for protecting the photovoltaic glass backplane, because a smooth path can reduce the vibration and impact of the glass during movement, thereby reducing the risk of glass damage. The calculation of grasping and placing accuracy can intuitively reflect the accuracy of the manipulator at key operating points (grasping points and placement points). This helps to ensure the assembly quality of photovoltaic glass modules, because high-precision grasping and placement are key factors in ensuring the structural integrity and performance of the modules. These precise calculation formulas provide specific quantitative indicators for the path fitness function, so that the path fitness function can more accurately evaluate the advantages and disadvantages of different path segment combinations. In the path planning process, multiple path segment combinations are compared and optimized according to these indicators, and an initial motion path that better meets actual needs can be generated. For example, when comparing two paths, the better path can be selected based on a comprehensive evaluation of multiple indicators such as path length, smoothness, and pick-and-place accuracy, thereby improving assembly efficiency and quality. By optimizing path length, assembly time can be reduced and production efficiency can be improved. Optimizing path smoothness can protect the glass backplane and reduce the defective rate. Improving pick-and-place accuracy can ensure assembly accuracy and improve product quality. These factors combined can significantly improve the overall performance of photovoltaic glass module assembly, including production efficiency, product quality, and equipment stability.

[0120] As an optional embodiment, the optimization objective of the path fitness function in the planning model includes at least one of the following: shortest path length, lowest collision risk, no repeated sections in the path segment combination or the proportion of repeated sections is lower than a set threshold, and the manipulator has the highest placement accuracy.

[0121] In the assembly process, the path length is directly related to the assembly efficiency. The shorter the path length, the shorter the time required for the manipulator to complete an assembly operation. From a kinematic point of view, the movement speed of the manipulator is limited to a certain extent. A shorter path means that the back glass can be moved from the starting position to the target placement point faster at the same speed. For example, in three-dimensional space, if the starting position and the target position are regarded as two points, the optimization algorithm is used to find the shortest path connecting the two points, which is like drawing a straight line between the two points (taking into account the actual constraints), which can reduce unnecessary movement time. The shortening of the path length may have a certain impact on the risk of collision. Generally speaking, a shorter path may reduce the chance of contact with obstacles, but this also requires comprehensive consideration of other factors, such as the direction of the path and the complexity of the surrounding environment. At the same time, the optimization of the path length does not necessarily conflict with the placement accuracy. Under reasonable planning, the shortest path length can be achieved while ensuring the placement accuracy.

[0122] Due to the presence of multiple back glass and other equipment in the stacking area, there is a risk of collision when the robot moves. To reduce the risk of collision, the planning model needs to consider factors such as the location and shape of obstacles, the range of motion of the robot, and the size and shape of the back glass. By building a virtual workspace model, the robot's motion trajectory is collided during the path planning process to avoid contact with obstacles. For example, the workspace is divided into multiple small areas using a spatial segmentation algorithm, and the areas where obstacles are located are marked. When planning the path, ensure that the robot's motion trajectory does not pass through these dangerous areas. Reducing the risk of collision may result in an increase in path length. For example, in order to avoid an obstacle, the robot may need to take a detour, making the path longer. However, this trade-off is necessary because collisions may damage the robot and the back glass, resulting in production interruptions and reduced product quality. In terms of the relationship with placement accuracy, the reduction in collision risk helps ensure that the robot can stably transport the back glass to the target location, which indirectly helps to improve placement accuracy.

[0123] It is understandable that the path segment combination does not contain repeated sections or the proportion of repeated sections is lower than the set threshold, so avoiding repeated sections can further improve assembly efficiency. If the manipulator has repeated movements in the path, it will waste time and energy. From the perspective of path planning, this is an unreasonable way of movement. By setting constraints in the planning model to prohibit or restrict the appearance of repeated sections, the movement of the manipulator can be made more direct and efficient. For example, in the algorithm for searching the path, the path segments that have been traversed are recorded, and these segments are avoided from being selected again in the subsequent path generation process, thereby optimizing the continuity and efficiency of the path. In this way, reducing repeated sections is usually consistent with the goal of shortest path length. Paths that do not contain repeated sections are often shorter, which helps to increase assembly speed. At the same time, this can also reduce the risk of collision to a certain extent, because complex paths containing a large number of repeated sections may increase the probability of collision with obstacles. For placement accuracy, simple and direct paths are more conducive to the manipulator accurately controlling the direction and position of movement, thereby improving placement accuracy.

[0124] Placement accuracy is critical to the assembly quality of photovoltaic glass modules. High-precision placement can ensure the accurate position of the backplane glass after assembly, which is beneficial to subsequent production processes and product performance. In the path planning process, considering the placement accuracy requires precise control of the positioning of the manipulator at the target placement point. This involves the precise solution of the manipulator kinematic model, including the calculation of joint angles and the control of the end effector position. For example, through the inverse kinematics algorithm, the angles of each joint of the manipulator are calculated according to the coordinates of the target placement point, so as to accurately place the backplane glass at the target position. Improved placement accuracy may require the selection of a more stable and accurate motion path during path planning, which may have a certain impact on the path length and collision risk. For example, in order to improve placement accuracy, it may be necessary to select a relatively gentle path that avoids more interference factors, which may increase the path length. At the same time, high-precision placement also requires full consideration of collision risks during path planning, because any collision may cause the manipulator position to shift, thereby affecting placement accuracy.

[0125] By optimizing the path length and avoiding repeated sections, the robot can complete assembly tasks faster. For example, in actual production, if the path length of each assembly operation can be shortened by a certain percentage, or the time of repeated movements can be reduced, the overall assembly time will be significantly reduced in the case of a large number of assembly tasks, thereby increasing the output per unit time. A shorter and more efficient path means that the robot's motor and other power components work for a shorter time and consume less energy accordingly. This not only reduces production costs, but also helps save energy and protect the environment.

[0126] In this way, reducing the risk of collision can prevent the back glass from being damaged during transportation and placement. Since photovoltaic glass is fragile, any collision may cause defects such as cracks and scratches on the glass, affecting the optical performance and mechanical strength of the product. By optimizing path planning and ensuring that the manipulator can safely transport the glass, the product qualification rate can be improved. Improving placement accuracy can ensure that the back glass is accurately positioned after assembly, which is beneficial to subsequent processes such as component packaging. Accurate assembly position can ensure good electrical connection and stable structure of photovoltaic modules, thereby improving product quality and performance.

[0127] The optimized path planning makes the robot's movement more reasonable and stable. It avoids complex and error-prone movement paths, and reduces system failures and downtime caused by unreasonable path planning. For example, reducing repeated sections and reducing the risk of collision can reduce the probability of unexpected situations during the robot's movement and improve the reliability of the system. The optimization of placement accuracy requires a more precise robot control strategy, which helps to improve the control accuracy of the entire assembly system. By continuously optimizing path planning and control algorithms, the system can complete assembly tasks more stably and reliably, and adapt to long-term, high-intensity production needs.

[0128] As an optional embodiment, in 204, during the movement process, a path optimization model is used to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator to obtain an optimized motion path of the target backplane glass, including:

[0129] During the movement, real-time image information of the target backplane glass is collected in real time; through the path optimization model, according to the pre-stored glass size of the target backplane glass, the stacking area layout, the preset working range of the robot, and the real-time collected real-time image information, the target backplane glass is subjected to cross-operation processing to recombine multiple path segment combinations; the path fitness function is used to calculate the path fitness of each regenerated path segment combination, and the path with a path fitness higher than the set fitness threshold is selected as the optimized action path; the optimized action path is subjected to mutation operation processing to change the moving order of key nodes in the optimized action path, the current position, obstacles, and the real-time values ​​of the target placement points to avoid the occurrence of local optimal solutions.

[0130] Specifically, new paths are generated through crossover and mutation operations in the genetic algorithm. The crossover operation selects paths with high fitness and reorganizes the path segments; the mutation operation changes the order of the paths or adjusts the coordinates of the grab points and placement points to increase the diversity of the paths to avoid falling into the local optimal solution.

[0131] During the movement, real-time image information of the target backplane glass is collected in real time. This step is equivalent to providing real-time "on-site feedback" for subsequent path optimization. The actual situation of the target backplane glass's current environment is captured by image acquisition equipment (such as cameras, etc.), including its relative position with surrounding objects, its own posture and other information, so that path optimization can be adjusted according to the latest and most realistic state, avoiding the problem of not being consistent with the actual situation due to relying solely on the initial setting information.

[0132] The path optimization model performs cross-operation processing on the target backplane glass based on the relatively fixed basic parameters such as the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator stored in advance, combined with the real-time image information just collected. The cross-operation draws on the idea of ​​genetic algorithms, which is similar to the gene cross-recombination in biological genetics. It recombines some elements in different path segment combinations (such as the direction of some paths, the setting of key nodes, etc.) to create multiple new path segment combinations. This method can make full use of the existing path information, explore more possible path planning solutions, break the limitations of the original path, and explore the possibility of a better path.

[0133] The path fitness function is used to calculate the path fitness of each regenerated path segment combination. The path fitness function comprehensively considers multiple factors such as path length, collision risk, stability, placement accuracy, etc. (as described in the previous article, its specific composition and calculation method), and through the quantitative evaluation of these factors, a fitness "score" is given to each path segment combination. Then, the path whose path fitness is higher than the set fitness threshold is selected, which means selecting those relatively better paths that meet certain standards in comprehensive performance as the optimized action path, ensuring that the selected path can meet the requirements of the assembly operation in terms of multiple key indicators, and ensuring the efficiency and accuracy of the assembly process.

[0134] The mutation operation is performed on the optimized action path, which also originates from the concept of genetic algorithm. The mutation operation introduces a certain degree of randomness and variability by changing the real-time values ​​of the moving order, current position, obstacles, target placement points, etc. of the key nodes in the selected optimized action path. In the process of path optimization, if we only rely on crossover operation and fitness-based screening, we are likely to fall into the local optimal solution, that is, we only find the path that seems to be the best in the current search range but is not the global optimal solution. The mutation operation can make the path jump out of this local optimal "trap" and continue to explore other possible better paths, so that the path optimization can continue to evolve towards the true global optimal direction.

[0135] By collecting image information in real time, path optimization can closely follow the actual state of the target back panel glass during the movement process, and promptly respond to temporary changes that may occur in the stacking area (such as the movement of other objects, unexpected obstacles, etc.), so that the optimized path can always adapt to the ever-changing actual working conditions, ensuring that the assembly work can proceed smoothly without problems such as inoperability or collision due to environmental changes.

[0136] Cross-operation processing can fully tap the potential of various path combinations and generate new solutions from many possible path planning, greatly increasing the chances of finding a better path. Compared with relying only on a single or a few path planning ideas that are initially set, this method broadens the scope of path search and helps to find paths that may have been overlooked but actually have better performance, thereby further improving assembly efficiency and reducing collision risks.

[0137] The path fitness function is used to screen the path, ensuring that the final selected optimized action path reaches a high level in multiple key performance indicators, balancing the requirements of path length, stability, placement accuracy, etc. This enables the robot to quickly move the target backplane glass to the target placement point when performing assembly operations according to the optimized action path, while ensuring a smooth movement process, avoiding collisions, and placing it accurately, which comprehensively improves the quality and efficiency of assembly.

[0138] The mutation operation processing effectively solves the problem that path optimization is prone to fall into the local optimal solution, making the path optimization process more dynamic and exploratory. By continuously introducing changes, the path is constantly adjusted to a better direction, as close as possible to the global optimal path, and the assembly process is optimized to the maximum extent, improving the overall performance and benefits of the entire photovoltaic glass assembly system.

[0139] In this way, after correction by the genetic algorithm, a new path will be selected and the movements of the robot will be adjusted to ensure that collisions are avoided and unnecessary movements are reduced.

[0140] Further optionally, in the above steps, the path optimization model is expressed as the following formula:

[0141]

[0142] in, is the initial action path, It represents the total length of the path from the current position to the target placement point in the area to be stacked under the initial action path P. represents the minimum distance between the path segment combination and the obstacle, Indicates the minimum distance between a path segment combination and an obstacle The first minimum optimization distance after optimization correction, Indicates the minimum distance between a path segment combination and an obstacle The second minimum optimized distance obtained by obstacle avoidance correction, Represents the path dataset corresponding to the initial action path, is the real-time tilt angle of the back glass. is the optimized action path corrected by the path optimization model. is the weight parameter of each sub-item. The sum of the weight parameters is 1.

[0143] Path length is an important consideration in the path optimization model. It represents the total length of the path that the robot moves from the current position to the target placement point in the area to be stacked under the initial action path. Path length directly affects the efficiency of assembly. A shorter path means that the robot can complete the task faster and reduce assembly time. This part is similar to the path length factor in the previous path fitness function, and is one of the basic indicators for measuring the quality of the path. For example, the fitness function of the path will comprehensively evaluate the following factors: path length: the shorter the path, the higher the fitness; path smoothness: paths with smaller turning angles and smoother speed changes in the path are preferred; obstacle avoidance ability: the fewer obstacle avoidance points in the path, the higher the fitness; grasping accuracy: the smaller the error between the grasping point and the placement point, the higher the fitness.

[0144] Specifically, Indicates the minimum distance between a path segment combination and an obstacle The first minimum optimized distance after optimization correction. During the assembly process, it is crucial to ensure that the manipulator maintains a safe distance from obstacles in the stacking area. This parameter reflects the obstacle avoidance capability of the path, preventing the manipulator from colliding with other backplane glass or equipment during movement, thereby ensuring the safety of the assembly operation and the integrity of the photovoltaic glass backplane.

[0145] is the real-time tilt angle of the backplane glass. This factor is taken into account because the tilt angle of the photovoltaic glass backplane during movement will affect its stability and assembly accuracy. If the backplane glass tilt angle is too large, it may cause the glass to fall, collide with other objects, or fail to be accurately placed in the target position. Therefore, by monitoring and controlling this angle, the success rate and quality of assembly can be improved.

[0146] The weight parameters of each sub-item play a role in balancing the importance of different factors in the path optimization model. Different production scenarios and assembly requirements may have different emphases on path length, obstacle avoidance capability, and back glass tilt angle. For example, in scenarios with high assembly speed requirements, the weight of path length may be set higher; while in environments with narrow space and more obstacles, the weight of obstacle avoidance distance may account for a larger proportion. By adjusting the weight parameters, the path optimization model can be optimized according to actual needs, so that the generated optimized action path is more in line with specific production requirements.

[0147] The formula of the entire path optimization model comprehensively considers the three key factors of path length, obstacle avoidance distance and back glass tilt angle, and combines the corresponding weight parameters to correct the initial action path to obtain the optimized action path. This correction process is a trade-off and optimization, aiming to find a path solution that maximizes assembly efficiency while ensuring assembly safety and accuracy.

[0148] The path optimization model can comprehensively consider multiple key factors such as path length, obstacle avoidance ability and back glass tilt angle, and obtain the optimized action path through calculation. This makes path optimization no longer rely solely on a single factor, but can balance the relationship between various factors on the basis of ensuring safety and accuracy, and find a better path solution. For example, a path may be short in length, but if its obstacle avoidance ability is poor or it is easy to cause the back glass to tilt too much, it may not be selected as the optimized action path after calculation by this model. Due to the existence of weight parameters, the path optimization model can be easily adapted to different production demand scenarios. By adjusting the weights, path optimization can focus more on different aspects such as efficiency, safety or accuracy. For example, in the early stage of production, when it is necessary to quickly verify the feasibility of the assembly process, the weight of the path length can be appropriately increased to quickly generate a feasible path; in the mass production stage where product quality requirements are high, the weights of obstacle avoidance distance and back glass tilt angle can be increased to ensure high-quality assembly operations.

[0149] By considering the obstacle avoidance distance and the tilt angle of the back glass, the path optimization model helps to improve the assembly quality. A reasonable obstacle avoidance distance can prevent the robot from colliding with obstacles and ensure the safe transportation of the back glass. Controlling the tilt angle of the back glass can ensure the stability of the glass during movement, reduce glass damage or inaccurate placement caused by tilt, and thus improve the overall quality of photovoltaic glass components.

[0150] Taking the path length into consideration enables the generated optimized motion path to shorten the robot's motion path and reduce motion time as much as possible while ensuring other quality factors. At the same time, the optimization of obstacle avoidance also avoids time waste caused by unexpected situations such as collisions, thereby effectively improving assembly efficiency.

[0151] For example, the total path length needs to comprehensively consider the straight length of the path segment, the additional path caused by the turning angle, and the actual size of the robot. For the decomposed path segment, the path of the robot from the starting point to the end point is decomposed into multiple path segments, each segment is represented by two adjacent key points, and the coordinates of each key point are . Calculate the straight-line distance based on the coordinates of adjacent key points:

[0152]

[0153] Additional paths that take into account turning angles and calculate turning angles. For path segments and , calculate the turning angle of the robot :

[0154]

[0155] in, and is the direction vector of the path segment.

[0156] According to the motion characteristics of the robot, the turning angle will introduce additional path length. Assume that the turning radius of the robot is , the turning path length can be estimated as:

[0157]

[0158] If the robot is longer (e.g. end extension), the path needs to be adjusted:

[0159]

[0160] It is the additional path adjustment introduced by the length of the robot and needs to be calculated in combination with specific geometric parameters.

[0161] Then, the straight length of all path segments and the additional length introduced by the turning angle are summed up to get the total path length

[0162]

[0163] Therefore, the final path length calculation formula is:

[0164]

[0165] Exemplarily, a simplified path stability evaluation method may be further adopted. Specifically, the turning angle in the path may be directly calculated as the angle between each two path segments. The stability is evaluated by calculating the absolute value of each turning angle and taking the average value thereof.

[0166]

[0167] in, is the i-th turning angle, is the number of turns in the path.

[0168] The speed variation on a path can be measured by calculating the speed change between path segments. If the speed change is too large, the path is not smooth. Calculate the difference in speed change between each path segment and take the average.

[0169]

[0170] in, is the velocity difference between the path segments, is the number of path segments.

[0171] Combining the smoothness of the turning angle and speed, we can get the overall smoothness score of the path. A simple method is to take the weighted average of the two parts:

[0172]

[0173] in, and is the weight, controlling the influence of the two parts.

[0174] Further optionally, in 205, after controlling the manipulator based on the initial motion path or the optimized motion path, the manipulator control data of the target backplane glass can also be collected; the manipulator control data at least includes: actual path execution time, actual path length, grasping success rate, and placement accuracy. Then, the manipulator control data is used to iterate the planning model and the path optimization model to improve the model performance.

[0175] The actual path execution time refers to the time it takes for the robot to actually complete the entire operation process of grabbing, moving and placing the target backplane glass according to the initial action path or the optimized action path. It directly reflects the efficiency of the current planned path during actual execution, and is affected by many factors such as path length, robot movement speed, and intermediate pauses. For example, if there are many complex turns and avoidance operations in the path, or the key nodes are not smoothly connected, the actual path execution time may be longer.

[0176] The actual path length is the actual path length that the robot walks in actual operation. By comparing it with the planned path length, we can see the accuracy of the planning model and path optimization model in estimating the path. Since there may be some factors that are not considered in advance in the actual working conditions (such as temporary small obstacles, slight changes in the position of other glass, etc.), the actual path length may be different from the theoretical planning.

[0177] The grasping success rate reflects the success rate of the robot grasping the target backplane glass, which depends on many factors, including whether the grasping angle determined in the path planning is reasonable, the accuracy of the robot when reaching the grasping position, and the surface condition of the target backplane glass. If the grasping angle is not good, the robot may not be able to grasp the glass stably, thus affecting the entire assembly process.

[0178] The placement accuracy is used to measure the accuracy of the robot in placing the backplane glass at the target placement point, and is a key indicator for evaluating assembly quality. It is closely related to the positioning accuracy of the path end point and the stability of the robot's placement action. Low placement accuracy may cause the backplane glass to fail to be accurately placed at the predetermined position, affecting the subsequent assembly quality of photovoltaic glass components.

[0179] After collecting these actual manipulator control data, analyze the differences between them and the expected output of the planning model to specifically adjust the relevant parameters of the planning model or improve its algorithm logic. For example, if it is found that the actual path execution time is generally long, it may be that the planning model does not reasonably weigh factors such as path length and smoothness when generating the initial action path. In this case, the path generation algorithm can be optimized, or the weight parameters of each sub-item in the path fitness function can be adjusted to make it more inclined to generate a more efficient path. In the case of a low grasping success rate, the module for determining candidate grasping angles in the planning model can be further optimized to integrate more actual factors and improve the grasping success rate.

[0180] The path optimization model is also improved based on these control data. If the actual path length is much longer than the expected length after optimization, it means that the path optimization model is insufficient in considering the actual working conditions (such as obstacles, real-time status of the back glass, etc.). Based on this, the model can be improved in terms of real-time image information utilization, cross operation, mutation operation and other links, so that it can optimize the initial action path more accurately, better balance key factors such as path length, obstacle avoidance, and back glass tilt angle, so as to output an optimized action path that is more in line with reality and has better performance. At the same time, by analyzing the data related to placement accuracy, adjust the parts of the path optimization model that affect the final placement accuracy, such as the processing of key nodes and the optimization of parameters related to the target placement point.

[0181] Therefore, by continuously iterating the planning model and path optimization model with actual robot control data, the model can be made more in line with the real assembly environment and operation conditions. The model will gradually adapt to different stacking area layouts, back glass states, and various complex working conditions, reduce the gap between theoretical planning and actual operation, make the generated initial motion path and optimized motion path more in line with actual needs, and improve the accuracy of path planning. With the iterative optimization of the model, the output path will be more conducive to the efficient execution of assembly operations by the robot. For example, shortening the actual path execution time, improving the grasping success rate, and enhancing the placement accuracy, all of which directly help to improve the efficiency of the entire photovoltaic glass assembly assembly, reduce the time waste caused by operational errors, unreasonable paths, etc., and also improve the assembly quality of the product, ensure that the structure and performance of the photovoltaic glass assembly meet the requirements, and reduce defective products caused by assembly accuracy problems. After multiple iterations of the planning model and path optimization model, it can better cope with various possible situations and make the operation of the robot more stable and reliable. Reduce the frequency of abnormal situations such as accidental collisions and grasping failures caused by poor path planning, improve the stability of the entire assembly system during long-term operation, reduce maintenance costs and the risk of production interruptions, and ensure the continuity and stability of production.

[0182] In the embodiment of the present application, by introducing a path optimization system based on a genetic algorithm and a self-learning feedback mechanism, the automated and efficient lamination of the backplane glass is achieved. Traditional manual operation or simple robot arm control methods often fail to optimize the path, resulting in a long movement distance of the robot, a large time consumption, and poor operation accuracy. The embodiment of the present application optimizes the movement path of the robot through a genetic algorithm, significantly shortens the path length, reduces unnecessary movement, and greatly improves production efficiency. Compared with traditional manual operation, the embodiment of the present application can ensure the accurate grasping and placement of glass through intelligent path planning. By optimizing the path, the risk of grasping failure is reduced, the accuracy of glass stacking is improved, and subsequent problems caused by operational errors are avoided. Traditional path planning usually lacks an effective obstacle avoidance mechanism, which makes the robot prone to collision in a complex environment. The embodiment of the present application effectively prevents collisions with other glass or equipment through path optimization and the setting of real-time obstacle avoidance points, ensuring the safety and reliability of operation. By accumulating historical execution data, the embodiment of the present application has self-learning and dynamic optimization functions. After each path execution, the system adjusts the path planning according to actual feedback to improve the accuracy and stability of path selection. As the number of operations increases, the path optimization effect gradually improves, and the system continues to improve efficiency and accuracy in long-term operation. The automated path optimization and feedback mechanism of the embodiment of the present application greatly reduces the reliance on manual operation and realizes an efficient and accurate fully automatic lamination process. Only simple monitoring and adjustment of the system is required, reducing the complexity and labor intensity of manual intervention. The embodiment of the present application can realize the automated lamination of the back panel glass, improve production efficiency, and ensure the accuracy and safety of the lamination operation.

[0183] The present application provides an assembly device for a photovoltaic glass assembly, which is applied to an assembly scenario of a photovoltaic glass assembly, wherein the photovoltaic glass assembly is a plurality of backplane glasses in a stacking area, and the device comprises at least the following units:

[0184] A scanning unit, used for scanning the stacking area to obtain the position information of the back plate glass in the stacking area; the position information at least includes: the glass size, position coordinates and stacking area layout of each back plate glass;

[0185] A selection unit, used for selecting a target back panel glass and positioning the target back panel glass based on the position information;

[0186] A planning unit, configured to use a planning model to perform path planning on the located target backplane glass to obtain an initial motion path of the target backplane glass; the initial motion path at least includes initial motion paths corresponding to a plurality of assembly operations of the manipulator, and each initial motion path includes a plurality of key nodes;

[0187] An optimization unit, used to use a path optimization model to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator during the movement, so as to obtain an optimized motion path of the target backplane glass;

[0188] A control unit is used to control the manipulator based on the initial motion path or the optimized motion path so that the manipulator completes the assembly operation of the photovoltaic glass component.

[0189] The specific implementation of the device can be found in the related introduction in the previous embodiments, which will not be expanded here.

[0190] Based on the same implementation principle, the embodiment of the present application also provides an electronic device, Figure 3 is a structural block diagram of the electronic device 300, such as Figure 3 As 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 via 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 implement corresponding processing functions; the communication bus 304 is used for communication between the electronic device 300 and other devices; and the controller 305 is used to implement the assembly method of the photovoltaic glass component described in the above method embodiment.

[0191] 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 only one type of bus.

[0192] The memory 302 may include a random access memory (RAM) or a non-volatile memory (non-volatile memory), such as at least one disk memory. Optionally, the memory may also be at least one storage device located far away from the processor 301 .

[0193] 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, and the specific form can be determined according to actual needs.

[0194] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed, can implement each step that can be executed by an electronic device in the above method embodiment.

[0195] 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 it. 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 method for assembling a photovoltaic glass assembly, characterized in that: The method is applied to the assembly scenario of a photovoltaic glass assembly, wherein the photovoltaic glass assembly is a plurality of back plate glasses in a stacking area, and the method comprises: Scan the stacking area to obtain the position information of the back glass in the stacking area; the position information at least includes: the glass size, position coordinates, and stacking area layout of each back glass; Selecting a target backplane glass, and positioning the target backplane glass based on the position information; Using a planning model, path planning is performed on the located target backplane glass to obtain an initial motion path of the target backplane glass; the initial motion path at least includes initial motion paths corresponding to a plurality of assembly operations of the manipulator, and each initial motion path includes a plurality of key nodes; During the movement, a path optimization model is used to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator to obtain an optimized motion path for the target backplane glass; Controlling the manipulator based on the initial motion path or the optimized motion path so that the manipulator completes the assembly operation of the photovoltaic glass assembly; The adopting of the planning model to perform path planning on the located target backplane glass to obtain the initial motion path of the target backplane glass includes: Based on the position information, determine the position, surface state, candidate grasping angles, and relative positional relationship between the target backplane glass and other backplane glasses; Based on the position, surface state, candidate gripping angles, and relative positional relationship between the target backplane glass and other backplane glasses, simulate the assembly operation to be performed when the target backplane glass moves from the current position to the target placement point in the area to be laminated, and generate multiple path segment combinations including the assembly operation; The combination of the multiple path segments is optimized through a path fitness function to obtain an initial motion path of the target backplane glass; the coding information of the initial motion path includes at least: position coding of multiple key nodes, obstacle coding, and target placement point coding in the area to be laminated.

2. The method according to claim 1, characterized in that The scanning of the stacking area to obtain position information of the back plate glass in the stacking area includes: Collecting real-time image data of each back panel glass in the stacking area; Based on the real-time image data, the position, angle, stacking direction of each backplane glass, and the relative position relationship between each backplane glass are identified; the identification process at least includes: edge detection and object contour detection; A corresponding three-dimensional space model of the stacking area is constructed based on the recognition result; the three-dimensional space model of the stacking area includes the glass size, position coordinates, and stacking area layout of each back panel glass.

3. The method according to claim 2, characterized in that After the corresponding three-dimensional space model of the stacking area is constructed based on the recognition result, the method further includes: Control the laser sensor and the depth camera to obtain the relative position relationship and surface morphology data of each back panel glass by multi-angle scanning; the scanning frequency is set to a target scanning frequency that matches the operating speed of the back panel glass assembly line; Determine the thickness of each back panel glass based on the relative position relationship of each back panel glass; Performing surface texture analysis based on the surface morphology data to obtain the surface flatness of each backplane glass; Defect analysis is performed based on the surface flatness and the surface morphology data to identify whether each back panel glass has a preset type of defect; the preset type includes at least: scratches, cracks, and bubbles.

4. The method according to claim 3, characterized in that The defect analysis based on the surface flatness and the surface morphology data to identify whether each back panel glass has a preset type of defect includes: For the current backplane glass, extracting a high-resolution image from the surface morphology data; Through the first branch of the defect detection model, texture analysis is performed on the high-resolution image in combination with the surface flatness to extract corresponding texture features; the texture features are used to indicate the texture changes at various locations on the current backplane glass surface; Comparing the texture features with a pre-recorded standard texture change trend to identify whether there are scratches on the current back panel glass; Extracting candidate bubble positions with a height variation greater than a set threshold from the high-resolution image through the second branch of the defect detection model, and performing depth image analysis on the candidate bubble positions to identify bubbles existing on the surface or inside of the current backplane glass; Through the third branch of the defect detection model, Fourier transform analysis is performed on the high-resolution image to obtain the frequency spectrum characteristics corresponding to the high-resolution image, and based on the frequency spectrum characteristics, it is identified whether there are cracks in the current back panel glass.

5. The method according to claim 1, characterized in that: The path fitness function is expressed as the following formula: ； in, is the weight parameter of each sub-item, L represents the total length of the path from the current position to the target placement point in the area to be stacked under the combination of path segments, It is used to measure the degree of smoothness in the combination of path segments. It represents the minimum distance between the path segment combination and the obstacle, and is used to indicate the obstacle avoidance capability of the path segment combination. The grasping and placement accuracy P is used to measure the error between the current position of the manipulator and the target placement point.

6. The method according to claim 5, characterized in that The calculation formula for the i-th path length L is: Among them, the coordinates of the current position are ( , , ), the coordinates of the target placement point are ( , , ), is the rotation radius of the manipulator, is the angle between adjacent path segments; in, Used to measure the smoothness of the path segment combination, the path stability of the i-th path segment The calculation formula is: ;in is the angle between adjacent path segments; Among them, the grabbing and placing accuracy The calculation formula is ; The grasping and placing accuracy P is used to measure the error between the grasping point and the placing point of the robot. The higher the grasping and placing accuracy, the higher the accuracy of grasping and placing. is the error between the actual coordinates and the target coordinates.

7. The method according to claim 5, characterized in that The optimization objective of the path fitness function includes at least one of the following: The path length is the shortest, the collision risk is the lowest, the path segment combination does not contain repeated sections or the proportion of repeated sections is lower than the set threshold, and the placement accuracy of the robot is the highest.

8. The method according to claim 1, characterized in that: During the movement, a path optimization model is used to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator to obtain an optimized motion path of the target backplane glass, including: During the movement, real-time image information of the target backplane glass is collected in real time; Through the path optimization model, according to the glass size of the target backplane glass stored in advance, the stacking area layout, the preset working range of the robot, and the real-time image information collected in real time, the target backplane glass is subjected to cross-operation processing to re-combine to obtain multiple path segment combinations; Calculating the path fitness of each regenerated path segment combination by using a path fitness function, and selecting a path whose path fitness is higher than a set fitness threshold as the optimization action path; A mutation operation is performed on the optimized action path to change the movement order, current position, obstacles, and real-time values ​​of target placement points of key nodes in the optimized action path to avoid a local optimal solution.

9. The method according to claim 8, characterized in that The path optimization model is expressed as the following formula: in, is the initial action path, It represents the total length of the path from the current position to the target placement point in the area to be stacked under the initial action path P. represents the minimum distance between the path segment combination and the obstacle, Indicates the minimum distance between a path segment combination and an obstacle The first minimum optimization distance after optimization correction, Indicates the minimum distance between a path segment combination and an obstacle The second minimum optimized distance obtained by obstacle avoidance correction, Represents the path dataset corresponding to the initial action path, is the real-time tilt angle of the back glass. is the optimized action path corrected by the path optimization model. is the weight parameter of each item.

10. The method according to claim 1, characterized in that After controlling the manipulator based on the initial motion path or the optimized motion path, the method further includes: Collecting robot control data of the target backplane glass; the robot control data at least includes: actual path execution time, actual path length, grasping success rate, and placement accuracy; The manipulator control data is used to iterate the planning model and the path optimization model to improve the model performance.

11. An assembly device for a photovoltaic glass assembly, characterized in that: The device is applied to the assembly scene of photovoltaic glass components, wherein the photovoltaic glass components are a plurality of back plate glasses in a stacking area, and the device comprises at least the following units, wherein A scanning unit, used for scanning the stacking area to obtain position information of the back plate glass in the stacking area; The position information at least includes: glass size, position coordinates, and stacking area layout of each back panel glass; A selection unit, used for selecting a target backplane glass and positioning the target backplane glass based on the position information; A planning unit, configured to use a planning model to perform path planning on the located target backplane glass to obtain an initial motion path of the target backplane glass; the initial motion path at least includes initial motion paths corresponding to a plurality of assembly operations of the manipulator, and each initial motion path includes a plurality of key nodes; An optimization unit, used to use a path optimization model to perform path optimization on the initial motion path based on the glass size of the target backplane glass, the stacking area layout, and the preset working range of the manipulator during the movement, so as to obtain an optimized motion path of the target backplane glass; A control unit, configured to control the manipulator based on the initial motion path or the optimized motion path, so that the manipulator completes the assembly operation of the photovoltaic glass assembly; Among them, the planning unit, using a planning model, performs path planning on the located target backplane glass to obtain the initial motion path of the target backplane glass, and is specifically used to: determine the position, surface state, candidate grasping angles, and relative position relationship between the target backplane glass and other backplane glasses based on the position, surface state, candidate grasping angles, and relative position relationship between the target backplane glass and other backplane glasses; simulate the assembly operation to be performed when the target backplane glass moves from the current position to the target placement point in the area to be laminated based on the position, surface state, candidate grasping angles, and relative position relationship between the target backplane glass and other backplane glasses, and generate multiple path segment combinations containing the assembly operations; optimize the multiple path segment combinations through a path fitness function to obtain the initial motion path of the target backplane glass; the encoding information of the initial motion path includes at least: position encoding of multiple key nodes, obstacle encoding, and target placement point encoding in the area to be laminated.

12. A non-transitory computer-readable storage medium, characterized in that: The storage medium stores a computer software program, and when the computer software program is executed by a processor, the method for assembling a photovoltaic glass assembly according to any one of claims 1 to 10 is implemented.

Images