Motion deformation prediction method and system in large-area substrate transfer process

Abstract

The invention discloses a motion deformation prediction method in a large-area substrate transfer process, and belongs to the technical field of display screen manufacturing, the deformation prediction method introduces an agent model method, and a training sample set is substituted into a first basic equation to obtain a first weight matrix; and substituting the first weight matrix into a first initial expression to obtain a first agent model. According to the agent model technology, a high-precision model matched with an original model is constructed by using a fitting and interpolation method, and the core idea of the agent model technology is that modeling is carried out on an agent with relatively low cost to replace an expensive accurate calculation process. Furthermore, the motion deformation condition of the substrate is predicted based on the RBF radial basis proxy model through the existing finite element simulation data, so that the operation time is greatly saved, the accuracy of the deformation data can be ensured, and the working efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of display screen manufacturing, and more particularly relates to a motion deformation prediction method and system in large-area substrate transfer. BACKGROUND

[0002] The flat panel display (FPD) industry is developing rapidly, and the demand for large-size display screens in the fields of advertising, entertainment, conference display, etc. is increasing, and the size of the glass substrate constituting the display screen has reached more than 10 generations. In the production process of the display panel, the glass substrate needs to be transported and placed in multiple stations for processing to meet the manufacturing process requirements. At the same time, due to the characteristics of large size, ultra-thin and easy bending of the glass substrate, the glass substrate operation surface of the production equipment is required to have high flatness and small deformation to ensure the safety of the glass substrate in the production process. In addition, the substrate will be subjected to inkjet printing and other operations during the production process, and the substrate resting position is required to have high accuracy to ensure that the printing is accurately processed to meet the process requirements.

[0003] In order to meet the stringent manufacturing conditions, improve production efficiency, reduce production cost and reduce substrate breakage rate, a glass substrate handling robot is applied to transfer the large-area substrate. The handling robot adsorbs the substrate by the suction cup on the fork, and due to the large size of the substrate and the large flexibility of the fork structure, residual vibration will occur in the fork due to acceleration change during the transfer process, and the substrate will be deformed. If the deformation amplitude of the substrate is large, it will cause the pixel pits on the substrate to be tilted and deformed, causing the liquid droplets printed in the pixel pits to flow out, affecting the uniformity of the film thickness and the clarity of the pattern, etc. Therefore, it is necessary to study the robot substrate transfer process, and then control the robot motion parameters to minimize the motion deformation of the substrate during the transfer process, thereby reducing the loss.

[0004] However, due to the complex manufacturing substrate environment, the use of physical prototypes for testing relies on empirical parameter adjustment, which not only has high cost and low efficiency, but also is difficult to predict the deformation of large-area substrates in high-precision dynamic environments. SUMMARY

[0005] In view of the above defects or improvement needs of the prior art, the present application provides a motion deformation prediction method and system in large-area substrate transfer process, which aims to solve the technical problem that the prior art is difficult to predict the deformation of large-area substrates in dynamic environments at low cost and high precision.

[0006] To achieve the above-mentioned purpose, according to one aspect of the present application, a motion deformation prediction method in large-area substrate transfer process is provided, comprising: S1: using the boundary conditions of each known fork motion data of each grid node d on the large-area substrate in the robot transfer process The corresponding substrate deformation amount Constructing the training sample set k represents the total number of various boundary conditions; the robot is used to transfer large-area substrates between various workstations; d identifies preset nodes on the substrate. n is the total number of d; S2: The training sample set Substitute into the first fundamental equation Obtain the first weight matrix ; ; S3: Adjust the first weight matrix... Substitute into the first initial expression Obtain the complete expression of the first proxy model ; These are radial basis functions; These are the current boundary conditions; Boundary conditions for a known substrate deformation; For the first The weights of the radial basis functions; S4: The current boundary corresponding to the robot's forklift motion data in the current working environment. Enter the complete expression of the first agent model. This yields the current substrate deformation prediction data.

[0007] Further, S1 includes: S11: Perform coarse meshing on the large-area substrate, export the XYZ coordinates of the mesh node data, and perform deduplication and numbering to obtain the preset node d; S12: Perform fine meshing and finite element analysis on the large-area substrate to find the N adjacent nodes that are most adjacent to the preset node d, and obtain the substrate deformation under different boundary conditions. S13: Utilizing different boundary conditions The following describes how the substrate deformation of N adjacent nodes determines different boundary conditions. The substrate deformation corresponding to the preset node d .

[0008] Furthermore, S13 includes: using the RBF interpolation method for each boundary condition. The boundary conditions are obtained by interpolating the substrate deformation of the next N adjacent nodes. Substrate deformation at the preset node d .

[0009] Further, S13 includes: The three-dimensional coordinates of the N adjacent nodes a three-dimensional coordinate of a preset node d and a substrate deformation amount of N adjacent nodes substitute the second basic equation , to obtain a second weight matrix , ; is a radial basis function; m is the total number of adjacent nodes; substitute the second weight matrix into the second initial expression to obtain a complete expression of the second surrogate model of the preset node d ; for each preset node, repeat the above operation to obtain the second surrogate model of each preset node under each boundary condition; input the substrate deformation amount of the N adjacent nodes into the complete expression of the second surrogate model to obtain the substrate deformation amount of the corresponding preset node d under the specified boundary condition .

[0010] Further, the obtaining of the corresponding substrate deformation amount under different boundary conditions comprises: performing finite element simulation analysis on the deformation of the large-area substrate under different acceleration excitations to obtain the substrate deformation amount of the N adjacent nodes under different boundary conditions.

[0011] According to another aspect of the present application, a motion deformation prediction system for a large-area substrate during transportation is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the motion deformation prediction method.

[0012] According to another aspect of the present application, a large-area substrate transportation system is provided, comprising: a physical equipment system; a digital twin system, comprising: a data acquisition layer, configured to acquire various data and simulation analysis data in the physical equipment system, including: boundary conditions corresponding to various working environment fork movement data and corresponding substrate deformation amounts, substrate geometric structure data, and robot movement data; a deformation prediction layer, connected to the data acquisition layer, configured to execute the steps of the motion deformation prediction method according to the acquired various data to obtain substrate deformation prediction data; a virtual equipment layer, connected to the data acquisition layer and the deformation prediction layer, configured to simulate the physical equipment system according to the substrate geometric structure data and the robot movement data, and configured to visually display the deformation prediction data of the large-area substrate.

[0013] Further, the virtual equipment layer can also provide an interactive interface for parameter debugging; the debugged parameters are transmitted to the first proxy model to make it update the current substrate deformation data and feedback to the virtual equipment layer and show to the user.

[0014] Overall, compared with the prior art, the above technical solutions conceived by the present application can achieve the following beneficial effects: (1) The present application provides a large-area substrate motion deformation prediction method during transfer, which takes into account the boundary conditions corresponding to different robot fork motion parameters in the finite element analysis process, and the deformation data of the large-area substrate need to be solved again, which requires a lot of time and is inefficient, and is not practical in actual operation. The proxy model technology uses fitting and interpolation methods to construct a high-precision model matching the original model, and the training sample set is substituted into the first basic equation to obtain the first weight matrix, and the first weight matrix is substituted into the first initial expression to obtain the first proxy model. The core idea is to replace the expensive precise calculation process with a lower-cost proxy model. Further, the existing finite element simulation data is used to predict the motion deformation of the substrate based on the RBF radial basis proxy model, which greatly saves the calculation time and also ensures the accuracy of the deformation data, improving the work efficiency.

[0015] (2) The present application uses different boundary conditions to determine the substrate deformation of the N adjacent nodes under different boundary conditions , and selects the nodes on the substrate to obtain the corresponding substrate deformation, which can control the training sample size while ensuring the training accuracy.

[0016] (3) The present application uses RBF interpolation method to interpolate the substrate deformation of the N adjacent nodes under each boundary condition to obtain the substrate deformation of the preset node d under the boundary condition , which is simple to operate, low in calculation complexity, and can ensure a certain accuracy.

[0017] (4) The present application substitutes the three-dimensional coordinates of the N adjacent nodes , the three-dimensional coordinates of the preset node d , and the substrate deformation of the N adjacent nodes into the second basic equation, which combines the proxy model technology with simulation data to realize the downscaling mapping from the high-dimensional finite element model to the low-dimensional proxy model.

[0018] (5) The scheme carries out finite element simulation analysis on the deformation of the large-area substrate under different acceleration excitations, is simple to operate, has low calculation complexity, and can quickly obtain the substrate deformation of the N adjacent nodes of the preset node d under different boundary conditions.

[0019] (6) The large-area substrate transfer system provided by the present application not only solves the problems of high calculation cost and poor timeliness of finite element simulation, but also effectively promotes the combination of digital twin physics and virtual data, and realizes faster and more efficient mapping between physical space and twin space. The large-area substrate transfer system can perform virtual-real interaction modules, for example, the data connection and information interaction between the entity equipment system and the digital twin system can be realized through TCP Socket communication, the substrate three-dimensional model is reconstructed and colored by using the substrate geometric structure data, deformation data and robot tongs motion data, the visualization of the substrate motion deformation is realized, and the deformation of the substrate under different acceleration conditions during the transfer process is predicted in real time by calling the surrogate model according to the acceleration data obtained by the sensor, thereby solving the problem that the substrate motion deformation during the robot tongs substrate transfer process is difficult to monitor. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The flowchart for constructing the substrate deformation prediction algorithm model is provided for the embodiment 1 of the present application. Figure 2 The architecture schematic diagram of the large-area substrate transfer process motion deformation prediction system based on digital twin is provided for the embodiment 3 of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0022] Embodiment 1 Figure 1 The flowchart for constructing the substrate deformation prediction algorithm model is provided for the embodiment 1 of the present application. The embodiment provides a motion deformation prediction method in the large-area substrate transfer process, which comprises S1-S4.

[0023] S1: using the boundary conditions of each known tong motion data of each grid node d on the large-area substrate in the robot transfer process corresponding substrate deformation constitute a training sample set k is the total number of various boundary conditions. The robot is used to transfer large area substrates between various stations. d identifies preset nodes on the substrate, n is the total number of d.

[0024] Wherein, the robot fork hands are adsorbed to the substrate by the suction cup. In order to reduce the deformation in the movement process in the early stage, the acceleration of each joint and connecting rod of the robot can be controlled by combining the production rhythm, and the acceleration curve is shaped and optimized by using the zero vibration zero derivative (ZVD) input shaper, so as to reduce the residual vibration in the substrate transfer process and reduce the deformation of the substrate.

[0025] The following gives a specific example. The acceleration of each joint of the robot is planned by using an S-shaped curve in combination with the actual production rhythm, so as to ensure that the acceleration curve is relatively smooth and does not exist sharp mutation. The modal analysis is carried out on each joint of the robot to obtain the natural frequency and damping ratio of each joint, which are used for the design of the input shaper. The shaper (for example, the ZVD shaper) is selected. The shaper is a pulse sequence composed of n pulses, which can be expressed as:

[0026] Wherein, is the amplitude of the i-th pulse, is the delay of the i-th pulse. The target of input shaping is to make the system residual vibration zero, and the constraint condition is:

[0027] Wherein, is the percentage of system residual vibration. The parameters of the input shaper are obtained by solving:

[0028] Wherein, , , and the time delay is .

[0029] As an optional implementation, for each motion unit of the robot, after designing the corresponding shaper, the pulse sequence obtained and the input signal of the system, that is, the acceleration curve, are convolved, the shaping is completed, and the convolved signal is input to the system, so as to control the movement of the robot, suppress the residual vibration, and ensure that the deformation of the substrate is small during the transfer process.

[0030] S2: Substitute the training sample set into the first basic equation to obtain the first weight matrix . .

[0031] S3: input the first weight matrix into the first initial expression to obtain the complete expression of the first surrogate model . is a radial basis function. is the current boundary condition. is the boundary condition of the known substrate deformation amount. is the weight of the th radial basis function.

[0032] S4: input the current boundary condition corresponding to the fork hand motion data of the robot in the current working environment into the complete expression of the first surrogate model to obtain the current deformation prediction data.

[0033] As an optional implementation, the S1 comprises: S11: perform rough grid division on the large-area substrate, derive grid node data XYZ coordinates and perform de-duplication and numbering processing to obtain the preset node d. Based on Ansys, the deformation of the substrate under different boundary conditions can be analyzed to derive the grid node data and deformation data of the substrate, and the de-duplication and numbering processing can be performed in PyCharm to obtain the non-repeated grid node (denoted as the preset node d) and the node index file. In addition, the substrate can be modeled in SolidWorks and imported into Ansys. First, the substrate is roughly meshed, the grid node data XYZ coordinates are derived, and the de-duplication and numbering processing is performed to obtain the “non-repeated node” and “node index” files.

[0034] S12: perform fine grid division on the large-area substrate, find N adjacent nodes most adjacent to the preset node d, and obtain the corresponding substrate deformation amount under different boundary conditions. As an optional implementation, the obtaining of the corresponding substrate deformation amount under different boundary conditions comprises: performing finite element simulation analysis on the deformation of the large-area substrate under different acceleration excitations to obtain the substrate deformation amount of the N adjacent nodes under different boundary conditions.

[0035] S13: determine the substrate deformation amount of the preset node d under different boundary conditions using the substrate deformation amount of the N adjacent nodes under different boundary conditions .

[0036] Specifically, a plurality of nodes most adjacent to a specified grid node in the “non-repeated node” are found using the KNN algorithm as input, and the deformation amount of the specified node is solved as output.

[0037] ​As an optional implementation, the S13 comprises: using a RBF interpolation method to interpolate the substrate deformation of each boundary condition The substrate deformation of the next N adjacent nodes is interpolated to obtain the boundary condition The substrate deformation of the preset node d .

[0038] As an optional implementation, the S13 comprises: The three-dimensional coordinates of the N adjacent nodes , the three-dimensional coordinates of the preset node d , and the substrate deformation of the N adjacent nodes are substituted into the second basic equation to obtain a second weight matrix , . is a radial basis function. m is the total number of adjacent nodes.

[0039] The second weight matrix is substituted into the second initial expression to obtain the complete expression of the second surrogate model of the preset node d . For each preset node, repeat the above operation, that is, the second surrogate model of each preset node under each boundary condition can be obtained. The substrate deformation of the N adjacent nodes is input into the complete expression of the second surrogate model to obtain the substrate deformation of the corresponding preset node d under the specified boundary condition .

[0040] Embodiment 2 The embodiment provides a motion deformation prediction system in a large-area substrate transfer process, comprising a memory and a processor, the memory stores a computer program, characterized in that the processor executes the computer program to realize the steps of the motion deformation prediction method.

[0041] Embodiment 3 As shown in Figure 2 , the embodiment provides a large-area substrate transfer system, comprising: A physical equipment system (i.e., a physical equipment layer).

[0042] A digital twin system, comprising: The data acquisition layer is used for collecting various data and simulation analysis data in the entity equipment system, including boundary conditions corresponding to various working environment fork movement data and corresponding substrate deformation, substrate geometric structure data, and robot movement data. Python program can be used to read and process various data, and the geometric structure data and deformation data of the substrate are transmitted to the digital twin system through the TCP Socket communication protocol for bidirectional data transmission.

[0043] The deformation prediction layer is connected with the data acquisition layer, and is used for performing the steps of the movement deformation prediction method according to the collected various data to obtain substrate deformation prediction data. The core of the substrate deformation prediction algorithm model is a proxy model technology, which is trained offline based on a substrate deformation data set through an RBF neural network algorithm to obtain a proxy model for substrate deformation prediction.

[0044] The virtual equipment layer is connected with the data acquisition layer and the deformation prediction layer, and is used for simulating the entity equipment system according to the substrate geometric structure data and the robot movement data, and is also used for visualizing the deformation prediction data of the large-area substrate.

[0045] As an optional implementation, the virtual equipment layer can also provide an interactive interface for parameter debugging. The debugged parameters are transmitted to the first proxy model to update the current substrate deformation data and fed back to the virtual equipment layer and displayed to the user. Specifically, a virtual scene of substrate transfer can be constructed based on Unity3D visualization software, supporting multi-angle interaction, and the movement parameter debugging and substrate transfer motion picture and substrate deformation display can be performed, realizing high-fidelity dynamic simulation and real-time interaction of the substrate transfer process.

[0046] Specifically, the entity equipment layer (i.e., the entity equipment system) mainly includes a robot body, a robot fork and suction cup assembly, and a sensor, etc. The data acquisition layer collects the movement data of the robot fork in the substrate transfer process. The data acquisition layer performs finite element simulation on the deformation of the substrate under different boundary conditions, processes and trains the simulation data, constructs a substrate deformation data set, and sends the sensor data and the substrate geometric structure data to the virtual equipment layer. The deformation prediction layer mainly constructs a proxy model through a neural network algorithm, which is used for predicting the substrate deformation and sending the prediction result to the virtual equipment layer. The virtual equipment layer integrates the digital mockup of the equipment, the robot motion control, and the substrate deformation visualization, and is used for displaying the substrate deformation prediction result, realizing the monitoring of the substrate transfer process, and corresponding visualization of the prediction system.

[0047] Regarding data communication, PyCharm and Unity3D realize data connection and information interaction through TCP Socket communication protocol. Socket script is mounted on the main camera. When Unity3D runs the script, it will send the substrate geometry data, deformation data and robot motion data to the digital twin system.

[0048] Regarding substrate deformation visualization, the substrate model is reconstructed in Unity3D according to the received geometry data, and the deformation data is mapped to the color interval according to the formula The deformation amount is converted, where max and min are the maximum and minimum values in the data, respectively, and r is the predicted deformation amount. The converted data is the Hue value of the HSV color space, and then it is converted to RGB color. According to the deformation amount of each grid node, the grid is colored, representing the substrate deformation size, realizing the substrate deformation visualization.

[0049] Regarding substrate deformation prediction, when the actual equipment is running, the digital twin system performs corresponding motion and receives the robot fork hand acceleration data, sends a request to call the agent model to PyCharm, takes the acceleration as input, and gets the output as the substrate deformation data. After mapping to the color interval, coloring is performed to complete the substrate deformation prediction.

[0050] The technical features of the above-described embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as the combination of the technical features does not exist, it should be considered as the scope of the present disclosure. It should be noted that the "in an embodiment of the present application", "for example", "such as" and the like are intended to illustrate the present application, but not to limit the present application.

[0051] The above-described embodiments only express several embodiments of the present application, which are described in detail and in detail. However, it should not be construed as limiting the scope of the application. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application.

Claims

1. A method for predicting motion deformation during the transport of a large-area substrate, characterized in that, Comprising: S1: using boundary conditions of each known fork movement data of each grid node d on a large-area substrate in a robot transfer process corresponding substrate deformation amount constitute a training sample set k is the total number of various boundary conditions; the robot is used to transfer a large-area substrate between each station; d identifies a preset node on the substrate, n is the total number of d; S2: obtaining a training sample set Substitute the first basic equation Get the first weight matrix ; ; S3: Adjust the first weight matrix... Substitute into the first initial expression Obtain the complete expression of the first proxy model ; These are radial basis functions; These are the current boundary conditions; Boundary conditions for a known substrate deformation; For the first The weights of the radial basis functions; S4: obtaining the current boundary corresponding to the fork hand motion data of the robot in the current working environment inputting the complete expression of the first proxy model , to obtain the current substrate deformation prediction data.

2. The method of claim 1, wherein the method further comprises: The S1 comprises: S11: Roughly grid dividing the large-area substrate, deriving grid node data XYZ coordinates and performing de-duplication and numbering processing to obtain the preset node d; S12: Fine grid dividing the large-area substrate and performing finite element analysis, finding N adjacent nodes most adjacent to the preset node d, and obtaining corresponding substrate deformation amounts under different boundary conditions; S13: using different boundary conditions The substrate deformation of the N adjacent nodes described below determines the different boundary conditions The substrate deformation of the corresponding preset node d below .

3. The method of claim 2, wherein the motion deformation prediction is performed during the large area substrate transfer process. The S13 comprises: using a RBF interpolation method to interpolate the substrate deformation of each boundary condition The substrate deformation of the next N adjacent nodes is interpolated to obtain the boundary condition The substrate deformation of the preset node d .

4. The method of claim 3, wherein the motion deformation prediction is performed during the large area substrate transfer process. The S13 comprises: the three-dimensional coordinates of the N adjacent nodes , the three-dimensional coordinates of the preset node d , and the substrate deformation amount of the N adjacent nodes Substitute the second basic equation , to obtain the second weight matrix , ; is a radial basis function; m is the total number of adjacent nodes; the second weight matrix substitute the second initial expression obtain the complete expression of the second surrogate model of the preset node d ; repeat the above operation for each preset node to obtain the second surrogate model of each preset node under each boundary condition; inputting the substrate deformation amount of the N adjacent nodes into a complete expression of the second proxy model , to obtain the substrate deformation amount of the preset node d under the specified boundary condition . .

5. The method for predicting motion deformation during the transport of a large-area substrate as described in claim 2, characterized in that, The obtaining of the substrate deformation amounts under different boundary conditions comprises: performing finite element simulation analysis on the deformation of the large-area substrate under different acceleration excitations to obtain the substrate deformation amounts of the N adjacent nodes under different boundary conditions.

6. A motion-induced distortion prediction system for large area substrate handling, the system comprising: Comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method of any one of claims 1 to 5 when executing the computer program.

7. A large area substrate handling system, characterized by, Comprising: Physical equipment system; Digital twin system, comprising: A data acquisition layer for acquiring various data and simulation analysis data in the physical equipment system, including: boundary conditions corresponding to various working environment fork movement data and corresponding substrate deformation amounts, substrate geometric structure data, and robot movement data; A deformation prediction layer connected with the data acquisition layer, configured to obtain substrate deformation prediction data by performing the steps of the method of any one of claims 1 to 5 according to the acquired various data; A virtual equipment layer connected with the data acquisition layer and the deformation prediction layer, configured to simulate the physical equipment system according to the substrate geometric structure data and the robot movement data, and to visually display the deformation prediction data of the large-area substrate.

8. The large area substrate handling system of claim 7, wherein, The virtual equipment layer can also provide an interactive interface for parameter debugging; the debugged parameters are transmitted to the first proxy model to update the current substrate deformation data and fed back to the virtual equipment layer, and displayed to the user.

Images