Cutting control method, device, equipment and storage medium of liquid crystal screen

Abstract

The present invention relates to the technical field of liquid crystal screens, and discloses a cutting control method, device, equipment and storage medium for a liquid crystal screen. The method comprises: constructing a three-dimensional digital twin model of a liquid crystal screen and calculating thickness parameters and flatness parameters; optimizing and calculating the initial cutting planning parameters to obtain optimized cutting planning parameters; generating cutting path data through a path planning algorithm; performing real-time control and real-time feature extraction on a double-sided TFT cutting machine to obtain a coolant flow feature set and a cutting speed feature set; dynamically predicting and adjusting cutting parameters through an initial cutting process optimization model to obtain target cutting control parameters; obtaining a real-time cutting image and optimizing the initial cutting process optimization model to generate a target cutting process optimization model. The present invention can dynamically optimize cutting parameters according to the actual characteristics of the screen, improve cutting accuracy and efficiency, and reduce the defective product rate.

Description

Technical Field

[0001] The present invention relates to the technical field of liquid crystal screens, and in particular to a cutting control method, device, equipment and storage medium for a liquid crystal screen. Background Art

[0002] In the production process of LCD screens, cutting is one of the key processes, which directly affects the quality and yield of the product. The traditional LCD screen cutting method mainly relies on manual experience to set parameters, which is difficult to adapt to screens of different specifications and materials, and easily causes problems such as insufficient cutting accuracy and edge damage.

[0003] Due to the complex structure and diverse material properties of LCD screens, the cutting process involves multiple mutually coupled parameters, and the traditional fixed parameter cutting method is difficult to achieve optimal control. At the same time, the dynamic changes and uncertain factors in the cutting process, such as material stress and cooling effect, also bring challenges to the stability of cutting quality. Summary of the invention

[0004] The present invention provides a cutting control method, device, equipment and storage medium for a liquid crystal screen, which are used to dynamically optimize cutting parameters according to actual characteristics of the screen, improve cutting accuracy and efficiency, and reduce the defective product rate.

[0005] In a first aspect, the present invention provides a cutting control method for a liquid crystal screen, the cutting control method for a liquid crystal screen comprising:

[0006] Constructing a three-dimensional digital twin model of a liquid crystal screen, and calculating thickness parameters and flatness parameters of the liquid crystal screen;

[0007] Based on the thickness parameter and the flatness parameter, optimizing and calculating the initial cutting planning parameters of the liquid crystal screen to obtain optimized cutting planning parameters;

[0008] Generate cutting path data through a path planning algorithm according to the optimized cutting planning parameters and the structural characteristics of the liquid crystal screen;

[0009] Based on the cutting path data, the X-axis, Y-axis and Z-axis stepper motors of the double-sided TFT cutting machine are controlled in real time, and the coolant flow rate and cutting speed in the cutting process are analyzed and feature extracted in real time to obtain a coolant flow rate feature set and a cutting speed feature set;

[0010] Inputting the coolant flow characteristic set and the cutting speed characteristic set into a preset initial cutting process optimization model to dynamically predict and adjust cutting parameters to obtain target cutting control parameters;

[0011] A real-time cutting image corresponding to the target cutting control parameter is obtained, and the initial cutting process optimization model is optimized according to the real-time cutting image to generate a target cutting process optimization model.

[0012] In a second aspect, the present invention provides a cutting control device for a liquid crystal screen, the cutting control device for a liquid crystal screen comprising:

[0013] A construction module, used to construct a three-dimensional digital twin model of a liquid crystal screen and calculate thickness parameters and flatness parameters of the liquid crystal screen;

[0014] A calculation module, used for optimizing and calculating the initial cutting planning parameters of the liquid crystal screen based on the thickness parameter and the flatness parameter to obtain optimized cutting planning parameters;

[0015] A generating module, used for generating cutting path data through a path planning algorithm according to the optimized cutting planning parameters and the structural characteristics of the liquid crystal screen;

[0016] A control module is used to perform real-time control on the X-axis, Y-axis and Z-axis stepper motors of the double-sided TFT cutting machine based on the cutting path data, and to perform real-time analysis and feature extraction on the coolant flow rate and cutting speed during the cutting process to obtain a coolant flow rate feature set and a cutting speed feature set;

[0017] A prediction module, used for inputting the coolant flow characteristic set and the cutting speed characteristic set into a preset initial cutting process optimization model to dynamically predict and adjust cutting parameters to obtain target cutting control parameters;

[0018] The optimization module is used to obtain the real-time cutting image corresponding to the target cutting control parameter, and optimize the initial cutting process optimization model according to the real-time cutting image to generate a target cutting process optimization model.

[0019] The third aspect of the present invention provides a cutting control device for a liquid crystal screen, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the cutting control device for the liquid crystal screen executes the above-mentioned cutting control method for the liquid crystal screen.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned liquid crystal screen cutting control method.

[0021] In the technical solution provided by the present invention, by constructing a three-dimensional digital twin model of the liquid crystal screen, the accurate calculation of the thickness and flatness of the screen is realized. By adopting a cutting planning parameter optimization algorithm based on thickness parameters and flatness parameters, the cutting depth, speed and coolant flow rate can be dynamically adjusted according to the actual characteristics of the screen, thereby improving the adaptability and stability of the cutting process. The cutting path data is generated by the path planning algorithm, and the curve is smoothed to achieve the optimization of the cutting trajectory, reduce the stress concentration during the cutting process, and reduce the risk of screen damage. By real-time regulation of the X, Y, and Z axis stepper motors of the double-sided TFT cutting machine, combined with real-time analysis of the coolant flow rate and cutting speed, the precise control of the cutting process is achieved, and the consistency of the cutting quality is improved. The preset initial cutting process optimization model is introduced, and the cutting parameters are adjusted by dynamic prediction, so as to realize the intelligent control of the cutting process and improve the adaptive ability of the system. The efficient prediction and optimization of the cutting parameters are achieved, and the generalization ability and robustness of the model are improved. The continuous improvement of the cutting process is achieved through real-time cutting image feedback and model optimization mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic flow chart of a cutting control method for a liquid crystal screen provided in an embodiment of the present application;

[0024] Figure 2 A schematic block diagram of the structure of a cutting control device for a liquid crystal screen provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0027] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0028] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0029] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0030] See also Figure 1 , Figure 1 A schematic diagram of a flow chart of a cutting control method for a liquid crystal screen provided in an embodiment of the present application, such as Figure 1 As shown, the cutting control method of the liquid crystal screen provided in the embodiment of the present application includes steps S100 to S600.

[0031] Step S100, constructing a three-dimensional digital twin model of the LCD screen, and calculating thickness parameters and flatness parameters of the LCD screen;

[0032] It is understandable that the execution subject of the present invention may be a cutting control device of a liquid crystal screen, or may be a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0033] Specifically, the LCD screen is scanned at high resolution at multiple angles to obtain a point cloud data set of the LCD screen. The point cloud data set contains a large amount of spatial coordinate information on the surface of the LCD screen. Based on the point cloud data, a three-dimensional model of the LCD screen is preliminarily constructed. The initial three-dimensional model is meshed and topologically optimized. Through meshing, the three-dimensional model is discretized into smaller unit structures, which is convenient for analyzing the model. Topological optimization helps to remove redundant parts while maintaining structural integrity and accuracy, making the model lighter and more optimized. In this process, by calculating the stress distribution and deformation characteristics of the LCD screen, a stress distribution map and a deformation characteristic map are generated to reflect the specific conditions of the stress and deformation of the LCD screen in different regions. Based on the stress distribution map and the deformation characteristic map, the optimized three-dimensional mesh model is locally refined, and the mesh density is increased in the area with stress concentration or large deformation, so as to more accurately describe the geometric morphology and mechanical properties of the area. Through local refinement, a more accurate three-dimensional mesh model is obtained. Based on the refined three-dimensional mesh model, the height distribution of the surface of the LCD screen is calculated to obtain a height distribution matrix, which reflects the surface height difference of the LCD screen at various positions. In order to analyze the flatness of the LCD screen, the height distribution matrix is ​​statistically analyzed and fitted using the least squares method to calculate the flatness parameters of the LCD screen. The flatness parameter is an important indicator of the surface quality of the LCD screen, reflecting the smoothness of the screen surface relative to the reference plane. Based on the refined three-dimensional mesh model, the contour information of the LCD screen is extracted and the thickness parameter of the LCD screen is calculated. By analyzing the point cloud spacing at different positions in the three-dimensional model, the thickness distribution of the LCD screen is measured.

[0034] Step S200, optimizing and calculating the initial cutting planning parameters of the liquid crystal screen based on the thickness parameter and the flatness parameter to obtain optimized cutting planning parameters;

[0035] Specifically, based on the thickness parameter of the liquid crystal screen, the initial cutting depth is linearly mapped to obtain the initial depth mapping value of each area of ​​the liquid crystal screen. The thickness parameter reflects the thickness difference of different positions of the liquid crystal screen, and the initial depth mapping value is the basic data for adjusting the cutting depth according to the thickness difference. At the same time, according to the flatness parameter of the liquid crystal screen, the curvature distribution of its surface is calculated. The curvature distribution reflects the curvature degree of different areas of the screen surface. By calculating the curvature distribution matrix of the liquid crystal screen, the local undulation of the surface is understood. Based on the curvature distribution matrix, the initial depth mapping value is locally adjusted. According to the different degrees of curvature change, the cutting depth value is optimized and adjusted to ensure that the cutting process can adapt to the changes in the local undulations of the liquid crystal screen surface and avoid cutting too deep or too shallow. According to the adjusted cutting depth value, the pressure distribution of the cutting tool is calculated to obtain a pressure distribution curve, which reflects the force applied by the cutting tool to different areas during the cutting process, thereby preventing the screen from being damaged due to excessive local pressure. Based on the pressure distribution curve, the initial cutting speed is segmented and mapped to obtain the adjustment coefficient of the cutting speed. Since the surface flatness and thickness differences of the LCD screen will affect the rationality of the cutting speed, the cutting speed is corrected according to the pressure conditions in different areas to obtain the corrected speed adjustment coefficient to ensure that the cutting speed can still maintain accuracy on uneven surfaces. According to the speed adjustment coefficient, the initial cutting speed is corrected to obtain the optimized cutting speed value. Based on the optimized cutting speed value and the adjusted cutting depth value, the heat distribution during the cutting process is calculated. The heat distribution generated during the cutting process determines the flow demand of the coolant. Excessive heat may cause damage to the LCD screen or affect the cutting quality. Through the calculation of the heat distribution model, the temperature changes in different areas are understood, and the initial coolant flow rate is nonlinearly optimized accordingly to obtain the optimized coolant flow value, ensuring that the temperature of each area of ​​the LCD screen is kept within the appropriate range during the cutting process to avoid affecting the cutting quality due to overheating. The optimized cutting speed value, the adjusted cutting depth value and the optimized coolant flow value are combined to form the final optimized cutting planning parameters.

[0036] Step S300, generating cutting path data through a path planning algorithm according to the optimized cutting planning parameters and the structural characteristics of the liquid crystal screen;

[0037] Specifically, based on the structural characteristics of the liquid crystal screen, the liquid crystal screen is gridded. The cutting area of ​​the liquid crystal screen is divided into multiple grids to generate a grid matrix of the cutting area. The cutting boundary of the liquid crystal screen is extracted according to the cutting area grid matrix to obtain a coordinate set of the cutting boundary, which reflects the physical boundary of the liquid crystal screen. Based on the coordinate set of the cutting boundary, the initial cutting path is calculated to obtain a node sequence of the initial path, which reflects the key points passed by the cutting path on the surface of the liquid crystal screen. The initial path node sequence is locally optimized to consider factors such as local curvature and edge complexity to generate a smoother and more efficient optimized path node sequence. Based on the optimized path node sequence and the cutting speed value in the optimized cutting planning parameters, the movement time between each node is calculated to generate a time constraint matrix to reflect the time required for each path under different speed conditions, so as to ensure that the cutting tool maintains a uniform and reasonable speed during the movement. According to the time constraint matrix, the optimized path node sequence is time-synchronized and adjusted to obtain a synchronized path node sequence. Based on the synchronized path node sequence and the cutting depth value in the optimized cutting planning parameters, three-dimensional cutting path data is generated. The three-dimensional cutting path data is processed by curve smoothing, and a continuous and smooth motion trajectory is generated using a cubic spline interpolation algorithm. Through cubic spline interpolation, discrete path points are connected to obtain more continuous and smoothed cutting path data. Based on the smoothed cutting path data and the coolant flow value in the optimized cutting planning parameters, the coolant injection control parameters corresponding to each path segment are calculated, and a coolant control sequence is generated to ensure that the amount of coolant in each path segment can be accurately controlled according to the heat distribution and cutting speed during the cutting process, so as to keep the temperature of the LCD screen within a reasonable range and avoid damage caused by overheating. The smoothed cutting path data and the coolant control sequence are timestamped to ensure that the cutting path and coolant control operate synchronously in time to generate the final cutting path data.

[0038] Step S400: Based on the cutting path data, the X-axis, Y-axis and Z-axis stepper motors of the double-sided TFT cutting machine are controlled in real time, and the coolant flow rate and cutting speed in the cutting process are analyzed and feature extracted in real time to obtain a coolant flow rate feature set and a cutting speed feature set;

[0039] Specifically, the cutting path data is processed in time discretization to generate a set of coordinate points in a discrete time sequence, so that the cutting path can be divided into a series of specific time segments. In these time segments, each coordinate point corresponds to a specific position change on the LCD screen. Based on the coordinate point set, the displacement increment sequence of the X-axis, Y-axis and Z-axis is calculated to obtain a three-axis motion increment matrix. The matrix provides the displacement change information of the cutting machine on different axes. According to the three-axis motion increment matrix, a pulse frequency modulation algorithm is used to generate a control pulse sequence of the stepper motor. The control pulse sequence defines the movement instruction of the stepper motor at each time step, and drives the motor to move precisely through a pulse signal. After the pulse sequence is generated, the motor drive signal is digitally filtered to eliminate possible signal noise and irregular fluctuations, and a smoothed drive signal is obtained. The smoothed signal can ensure the stability of the stepper motor during operation, reduce errors and improve the accuracy of the cutting process. Based on the smoothed drive signal, the real-time position error of each axis is calculated. The real-time position error refers to the difference between the actual position of the stepper motor and the theoretical target position. The real-time position error is corrected by calculating the position error compensation amount. The position error compensation amount can be determined based on the sensor data in the real-time feedback system to ensure that the position of the cutting tool is highly consistent with the planned path. According to the position error compensation amount, the smoothed drive signal is corrected in real time to generate a corrected drive signal. Based on the corrected drive signal, the frequency domain characteristics of the cutting speed are extracted. By analyzing the characteristics of the cutting speed in the frequency dimension, a cutting speed feature set is generated to reflect the speed changes and laws during the cutting process. At the same time, for the control of the coolant flow, the flow data sampling method is used to generate a flow feature vector. The flow feature vector reveals the change pattern of the coolant during the cutting process by statistically analyzing the coolant flow at different time points. Based on the flow feature vector, the coolant flow feature set is extracted so that the coolant flow can be dynamically adjusted according to the temperature of the LCD screen and the cutting situation in actual operation.

[0040] Step S500, inputting the coolant flow characteristic set and the cutting speed characteristic set into a preset initial cutting process optimization model to dynamically predict and adjust the cutting parameters to obtain target cutting control parameters;

[0041] Specifically, the coolant flow feature set and the cutting speed feature set are normalized. The input features are converted into a unified scale by standardization, and the dimension differences between different features are eliminated to obtain standardized input features. The standardized input features are input into the feature extractor in the initial cutting process optimization model. The feature extractor adopts the ResNet structure, which consists of 4 residual blocks. Each residual block contains 3 convolutional layers and a short-circuit connection. BatchNormalization and ReLU activation functions are used between the convolutional layers. Through this structure, the feature extractor can efficiently extract useful high-level features from the input data while maintaining the flow of information, avoiding the gradient vanishing problem. After multiple nonlinear transformations of the residual block, the extracted high-level features are integrated into a feature mapping matrix, which represents the abstract representation of the coolant flow and cutting speed features. The feature mapping matrix is ​​subjected to global average pooling to convert it into a fixed-dimensional feature vector. By compressing the matrix, the number of parameters is reduced while retaining important feature information, thereby improving the generalization ability of the model. The obtained feature vector is input into the parameter predictor in the initial cutting process optimization model. The parameter predictor adopts a multi-layer perceptron structure and consists of 3 fully connected layers. Each layer uses Dropout and LeakyReLU activation functions to reduce the risk of overfitting and increase the nonlinear expression ability of the model. The initial cutting parameter prediction value is obtained through the nonlinear transformation of the fully connected layer. Based on the initial prediction value, time series modeling is performed in combination with historical cutting data. Time series modeling uses a preset LSTM (long short-term memory) network to capture the dependencies in the time dimension. Through the processing of the LSTM network, the time series feature sequence of the cutting process is obtained, reflecting the dynamic changes and trends of different cutting parameters in time. The time series feature sequence is input into the optimization generator in the initial cutting process optimization model to generate the optimized cutting parameter candidate values. The optimization generator adopts two layers of LSTM units, each layer contains 128 hidden nodes, and uses the Tanh activation function to ensure the continuity and interpretability of the output. The optimization discriminator used in conjunction with it uses a one-dimensional convolutional neural network structure, through three layers of one-dimensional convolutional layers and MaxPooling operations, combined with the LeakyReLU activation function, to determine the validity of the generated parameters. In order to ensure the feasibility of the cutting parameters, the feasibility of the optimized cutting parameter candidate values ​​is verified. Through the preset constraint check module, the effective parameter set that meets the actual cutting process requirements is screened out. This constraint check process ensures that the generated parameters meet the physical limitations and process requirements of LCD screen cutting. The screened effective parameter set is input into the optimization discriminator, which calculates the optimization score of each group of parameters to form a parameter score list. The score list is sorted according to the actual optimization effect of each group of parameters. According to the parameter score list, the parameter combination with the highest score is selected through the sorting algorithm.The parameter combination that has been screened and optimized multiple times is used as the final target cutting control parameter to guide the actual cutting process of the LCD screen and achieve the optimal balance between cutting efficiency and quality.

[0042] Step S600: acquiring a real-time cutting image corresponding to a target cutting control parameter, and optimizing an initial cutting process optimization model according to the real-time cutting image to generate a target cutting process optimization model.

[0043] Specifically, a real-time cutting image corresponding to the target cutting control parameter is obtained. Real-time images of the LCD screen cutting process are captured by a camera device, and these images are preprocessed. The preprocessing steps include operations such as denoising, contrast enhancement and edge enhancement, so as to obtain an enhanced cutting image. Based on the enhanced cutting image, the cutting edge features of the LCD screen are extracted to generate an edge feature map. Through image processing algorithms, such as Sobel operator or Canny edge detection, the boundary of the cutting area can be accurately captured. Morphological operations are performed on the edge feature map, including expansion, corrosion, opening operation and closing operation, etc., to optimize the connectivity and accuracy of the edge, and obtain an optimized edge feature map. Based on the optimized edge feature map, the straight line segment is detected by the Hough transform algorithm to obtain a cutting path straight line parameter set. Hough transform is an image processing algorithm used to detect geometric shapes in an image. In this process, the direction and angle of the LCD screen cutting path can be accurately identified. Cluster analysis is performed on the extracted cutting path straight line parameter set to optimize the cutting path information, and a cutting direction feature vector is extracted from it to reflect the overall trend and directionality of the cutting path. Based on the extracted cutting direction feature vector and the target cutting control parameter, a cutting quality evaluation score is constructed. The cutting quality evaluation score combines indicators such as the straightness of the cutting path, angle accuracy, and edge smoothness to measure the actual quality of the cutting process. The cutting quality score is compared with the preset threshold, and the corresponding model optimization strategy is generated through the discriminant function. If the evaluation score is lower than the threshold, it means that the cutting quality does not meet the ideal standard and the initial cutting process optimization model needs to be adjusted. Based on the generated model optimization strategy, the weight of the initial cutting process optimization model is fine-tuned. By adjusting the weight parameters of each layer in the model, the model can better fit the actual cutting situation and improve the control accuracy of the cutting process. After fine-tuning the weights, the parameters of the model are regularized to prevent overfitting. Regularization can ensure that the model maintains good generalization ability and avoids the model from over-relying on specific features when facing new data. The regularized model parameters are updated to the initial cutting process optimization model, and the target cutting process optimization model is generated by reorganizing the model structure.

[0044] In the embodiment of the present invention, by constructing a three-dimensional digital twin model of the liquid crystal screen, the thickness and flatness of the screen are accurately calculated. By using a cutting planning parameter optimization algorithm based on thickness parameters and flatness parameters, the cutting depth, speed and coolant flow rate can be dynamically adjusted according to the actual characteristics of the screen, thereby improving the adaptability and stability of the cutting process. The cutting path data is generated by the path planning algorithm, and the curve is smoothed to optimize the cutting trajectory, reduce stress concentration during the cutting process, and reduce the risk of screen damage. By real-time regulation of the X, Y, and Z axis stepper motors of the double-sided TFT cutting machine, combined with real-time analysis of coolant flow and cutting speed, accurate control of the cutting process is achieved, and the consistency of cutting quality is improved. A preset initial cutting process optimization model is introduced, and the cutting parameters are adjusted by dynamic prediction, so that intelligent control of the cutting process is achieved, and the adaptive ability of the system is improved. Efficient prediction and optimization of cutting parameters are achieved, and the generalization ability and robustness of the model are improved. Through real-time cutting image feedback and model optimization mechanism, continuous improvement of the cutting process is achieved.

[0045] In a specific embodiment, the process of executing step S100 may specifically include the following steps:

[0046] Perform high-resolution multi-angle scanning on the LCD screen to obtain a point cloud data set of the LCD screen, and construct an initial three-dimensional model of the LCD screen based on the point cloud data set;

[0047] Performing mesh division and topology optimization on the initial three-dimensional model to obtain an optimized three-dimensional mesh model, and calculating the stress distribution and deformation characteristics of the liquid crystal screen based on the optimized three-dimensional mesh model to obtain a stress distribution map and a deformation characteristic map;

[0048] According to the stress distribution map and the deformation characteristic map, the optimized three-dimensional mesh model is locally refined to obtain a refined three-dimensional mesh model;

[0049] Based on the refined three-dimensional grid model, the height distribution of the LCD screen surface is calculated to obtain the height distribution matrix, and the height distribution matrix is ​​statistically analyzed to obtain the flatness parameters of the LCD screen through least squares fitting;

[0050] Based on the refined three-dimensional mesh model, the contour information of the LCD screen is extracted, and the thickness parameters of the LCD screen are calculated.

[0051] Specifically, the LCD screen is scanned at multiple angles with high resolution. The LCD screen is scanned from multiple angles by a high-precision scanner to obtain a point cloud data set of the LCD screen, and the positions of each point on the surface of the LCD screen are recorded to form a spatial distribution representation method. Based on the point cloud data set, the initial three-dimensional model of the LCD screen is constructed. The point cloud is processed by a triangular meshing algorithm. The commonly used algorithm is Delaunay triangulation, which connects the points in the point cloud to form a polygonal mesh composed of triangles. The meshing process organizes the scattered point cloud information into a geometric structure. The initial three-dimensional model can intuitively display the overall shape and size of the LCD screen. The initial three-dimensional model is meshed to discretize the three-dimensional model into a series of small units (usually triangles or quadrilaterals). At the same time, according to the shape characteristics of the LCD screen, meshes of different densities are used. Topological optimization is performed to reduce redundant meshes, maintain the integrity of the geometric structure and computational efficiency, and obtain an optimized three-dimensional mesh model. Based on the optimized three-dimensional mesh model, the stress distribution and deformation characteristics of the LCD screen are calculated. The stress distribution reflects the stress borne by different areas of the LCD screen under stress, while the deformation characteristics show the deformation of the screen after being stressed. The calculation of stress distribution usually adopts the finite element analysis method, and the stress values ​​in different grid units are obtained by solving the stress field equation. Based on the basic equation of elastic mechanics, the stress is calculated by the following formula:

[0052] σ=E·ε;

[0053] Among them, σ is the stress tensor, E is the elastic modulus of the material, and ε is the strain tensor. The stress distribution diagram shows the stress state of the LCD screen in different areas, which is used to analyze which parts are prone to cracking or deformation. At the same time, through similar finite element method analysis, the deformation characteristic diagram of the LCD screen is calculated. The deformation characteristic diagram shows the deformation amount of each area of ​​the LCD screen under different stress conditions. According to the stress distribution diagram and the deformation characteristic diagram, the optimized three-dimensional mesh model is locally refined, and the mesh density is increased in areas with stress concentration or large deformation to more accurately simulate the stress and deformation of these areas and generate a refined three-dimensional mesh model. Based on the refined three-dimensional mesh model, the height distribution of the LCD screen surface is calculated. The height distribution refers to the height change of each point on the surface of the LCD screen relative to a certain reference plane. By selecting a suitable reference plane on the model surface, the height value of each mesh unit is calculated to generate a height distribution matrix. Assume that the three-dimensional coordinates of a point are (x i ,y i ,z i), then the height value of the point can be expressed as its z coordinate, and each element in the matrix represents the height difference of the point. Statistical analysis is performed on the height distribution matrix, and the least squares method is used to fit the local data to obtain the half-degree parameter of the LCD screen. The least squares method is a commonly used fitting method, and its goal is to find a set of parameters that minimizes the sum of squared errors between the fitting curve and the observed data. For the fitting of the flatness parameter, the least squares expression can be written as:

[0054]

[0055] Where F(h) is the sum of squared errors, z i represents the height of the observation point, f(x i ,y i ) represents the fitting function, and n is the number of observation points. By minimizing the sum of squared errors, the height variation trend of the LCD screen is fitted and its flatness parameter is obtained. The flatness parameter can reflect the smoothness of the LCD screen surface relative to the reference plane. A lower flatness parameter indicates that the screen surface is very smooth, while a higher flatness parameter indicates that there are large irregularities on the surface. Based on the refined three-dimensional grid model, the contour information of the LCD screen is extracted, and the thickness parameter of the LCD screen is calculated. The extraction of contour information is usually carried out by analyzing the points on the edge of the grid to construct a two-dimensional projection contour of the LCD screen. By measuring the distance of contour points at different positions, the thickness of the LCD screen in different areas is calculated. The thickness parameter reflects the overall thickness distribution of the LCD screen structure and is an important indicator for determining the cutting accuracy.

[0056] In a specific embodiment, the process of executing step S200 may specifically include the following steps:

[0057] Based on the thickness parameter, the initial cutting depth is linearly mapped to obtain an initial depth mapping value, and the curvature distribution of the surface of the LCD screen is calculated according to the flatness parameter to obtain a curvature distribution matrix;

[0058] Based on the curvature distribution matrix, the initial depth mapping value is locally adjusted to obtain an adjusted cutting depth value, and according to the adjusted cutting depth value, the pressure distribution of the cutting tool is calculated to obtain a pressure distribution curve;

[0059] Based on the pressure distribution curve, the initial cutting speed is mapped by piecewise function to obtain the speed adjustment coefficient, and the initial cutting speed is corrected according to the speed adjustment coefficient to obtain the optimized cutting speed value;

[0060] Based on the optimized cutting speed value and the adjusted cutting depth value, the heat distribution in the cutting process is calculated to obtain a heat distribution model, and according to the heat distribution model, the initial coolant flow rate is nonlinearly optimized to obtain an optimized coolant flow rate value;

[0061] The optimized cutting speed value, the adjusted cutting depth value, and the optimized coolant flow value are combined into optimized cutting planning parameters.

[0062] Specifically, the thickness parameters of the LCD screen are processed to determine the initial cutting depth. The thickness distribution of the LCD screen varies at different locations, and the cutting depth is adjusted according to these differences to ensure the cutting accuracy. The initial cutting depth is linearly mapped based on the thickness parameters. The linear mapping can be expressed by the following formula:

[0063] d i = k·h i ;

[0064] Among them, d i represents the initial cutting depth at position i, h i Represents the thickness of the LCD screen at this position, and k is the mapping coefficient, which determines the linear relationship between the cutting depth and the thickness. Through this formula, the initial depth mapping values ​​at different positions are obtained. According to the flatness parameter, the curvature distribution of the LCD screen surface is calculated to obtain the curvature distribution matrix. The flatness parameter reflects the smoothness of the LCD screen surface, and the curvature distribution describes the curvature of the surface in different areas. In order to calculate the curvature distribution, the second-order derivative of the surface is calculated by fitting a small range of the screen surface, and the curvature distribution matrix is ​​generated. The curvature distribution matrix represents the local curvature of different areas. The larger the curvature, the more obvious the surface curvature. Based on the curvature distribution matrix, the initial depth mapping value is locally adjusted. Areas with larger curvatures are usually more likely to produce stress concentration during the cutting process, resulting in increased cutting difficulty. Therefore, in these areas, the cutting depth should be appropriately reduced. By combining the curvature distribution matrix, the initial depth mapping value is adjusted to obtain a more accurate cutting depth value. The local adjustment can be expressed by the following formula:

[0065] d′ i =d i ·(1-α·C i );

[0066] Among them, d′ i Indicates the adjusted cutting depth, C i represents the curvature value at position i in the curvature distribution matrix, and α is an adjustment coefficient used to control the effect of curvature on cutting depth. In areas with large curvature, the cutting depth is appropriately reduced to reduce the stress risk during the cutting process. After obtaining the adjusted cutting depth value, the pressure distribution of the cutting tool is calculated. The pressure distribution determines the force applied by the tool to different areas during the cutting process. The magnitude of the pressure depends not only on the cutting depth, but also on the hardness of the material and the physical properties of the tool. The pressure distribution can be calculated using a mechanical model, as shown below:

[0067] P i =F / A i ;

[0068] Among them, P i is the cutting pressure at position i, F is the cutting force, A i Indicates the contact area at that position. By calculating the pressure of the tool at each position, a pressure distribution curve is generated to reflect the pressure changes applied by the cutting tool in different areas of the LCD screen. Based on the pressure distribution curve, the initial cutting speed is mapped by piecewise function. The cutting speed is closely related to the pressure of the tool. In areas with higher pressure, the cutting speed needs to be slowed down to avoid damaging the LCD screen. During the mapping process, the speed is adjusted according to the pressure to obtain the speed adjustment coefficient. The adjustment coefficient can be expressed by the following function:

[0069] v′ i =v0(1-β·P i );

[0070] Among them, v′ i is the adjusted cutting speed, v0 is the initial cutting speed, and β is the pressure influence coefficient. The cutting speed will decrease accordingly at the location with greater pressure, and vice versa, the speed will remain higher at the location with less pressure. By adjusting the cutting speed, excessive pressure can be avoided from damaging the LCD screen. Based on the optimized cutting speed value and the adjusted cutting depth value, the heat distribution during the cutting process is calculated. The heat distribution depends on the cutting speed, cutting depth and thermal conductivity of the material. The heat distribution is calculated using the heat conduction equation:

[0071] Q i =k t ·v′ i ·d′ i ;

[0072] Among them, Q i is the heat generated at position i, k t is the thermal conductivity of the material, v′ i and d″ i are the cutting speed and cutting depth at that position respectively. The heat distribution model can reflect the temperature changes in each area during the cutting process. Based on the heat distribution model, the initial coolant flow rate is nonlinearly optimized to ensure that the temperature of the LCD screen remains within a reasonable range during the cutting process. The coolant flow rate should be proportional to the heat generated, and the flow optimization can be achieved through the nonlinear equation:

[0073] Q i =γ·f i ;

[0074] Among them, f irepresents the coolant flow rate, and γ is the heat dissipation efficiency coefficient of the coolant. By adjusting the flow value, ensure that the temperature of each area is not too high to avoid damage to the screen material due to overheating. The optimized coolant flow value can ensure the effectiveness of temperature control during the cutting process. The optimized cutting speed value, the adjusted cutting depth value and the optimized coolant flow value are combined together to form the optimized cutting planning parameters.

[0075] In a specific embodiment, the process of executing step S300 may specifically include the following steps:

[0076] Based on the structural characteristics of the liquid crystal screen, the liquid crystal screen is gridded to obtain a cutting area grid matrix, and the cutting boundary of the liquid crystal screen is extracted according to the cutting area grid matrix to obtain a cutting boundary coordinate set;

[0077] Based on the cutting boundary coordinate set, the initial cutting path is calculated to obtain the initial path node sequence, and the initial path node sequence is locally optimized to obtain the optimized path node sequence;

[0078] Based on the optimized path node sequence and the cutting speed value in the optimized cutting planning parameters, the movement time between each node is calculated to obtain the time constraint matrix;

[0079] According to the time constraint matrix, the optimized path node sequence is time-synchronized to obtain a synchronized path node sequence, and three-dimensional cutting path data is generated based on the synchronized path node sequence and the cutting depth value in the optimized cutting planning parameters;

[0080] Perform curve smoothing on the three-dimensional cutting path data, generate a continuous and smooth motion trajectory through the cubic spline interpolation algorithm, and obtain smoothed cutting path data;

[0081] Based on the smoothed cutting path data and the coolant flow value in the optimized cutting planning parameters, the coolant injection control parameters of each path segment are calculated to obtain a coolant control sequence;

[0082] The smoothed cutting path data and the coolant control sequence are time stamped and aligned to generate the final cutting path data.

[0083] Specifically, the liquid crystal screen is meshed, and the local geometric shape of the surface is described by dividing the surface of the liquid crystal screen into multiple small rectangular or triangular units. Each unit has a clear boundary and vertex. By constructing these units into a cutting area grid matrix, the geometric complexity of the liquid crystal screen is converted into a mathematical model that can be calculated and processed. Based on the cutting area grid matrix, the cutting boundary of the liquid crystal screen is extracted. After the boundary is extracted, each point on the boundary will correspond to a coordinate to form a cutting boundary coordinate set. Based on the cutting boundary coordinate set, the initial cutting path is calculated. The initial cutting path refers to the connection trajectory from one boundary point to another boundary point. The initial path node sequence is calculated by the nearest neighbor path algorithm, and each node in the sequence represents a key position on the cutting path. The order of the node sequence determines the order of cutting and the shape of the trajectory. The initial path node sequence is locally optimized to improve the cutting accuracy. Local optimization adjusts the inflection points in the path by analyzing the geometric relationship between the nodes to ensure that the cutting path is smoother and more efficient. Based on the optimized path node sequence and the cutting speed value in the optimized cutting planning parameters, the movement time between each node is calculated. According to the cutting speed value, the movement time between every two nodes is calculated using the following formula:

[0084]

[0085] Among them, t i,j represents the movement time from node i to node j, d i,j is the distance between the two nodes, v i,j is the corresponding cutting speed. Through this formula, the time constraint matrix between all nodes is obtained. The time constraint matrix is ​​a two-dimensional matrix, in which each element represents the time required between adjacent nodes. According to the calculated time constraint matrix, the optimized path node sequence is adjusted for time synchronization to ensure that the movement of the cutting tool between different nodes is coordinated and consistent, avoiding path deviation caused by inconsistent speed. By adjusting the time interval of the nodes, the speed and time in the cutting process are coordinated to obtain a synchronized path node sequence. Based on the synchronized path node sequence and the cutting depth value in the optimized cutting planning parameters, three-dimensional cutting path data is generated. In order to optimize the cutting path, the three-dimensional cutting path data is smoothed. Discrete nodes in the cutting path may cause discontinuity in the motion trajectory or unnecessary jitter. A continuous and smooth motion trajectory is generated by the cubic spline interpolation algorithm. Cubic spline interpolation achieves smooth connection by generating a cubic polynomial curve for each path segment. Assuming interpolation between nodes i and i+1, the cubic spline curve can be expressed as:

[0086] S(x)=a i x 3 +b ix 2 +c i x+d i ;

[0087] Among them, S(x) is the interpolation function, a i 、b i 、c i and d i It is a coefficient that needs to be determined by node values ​​and derivative conditions. The interpolated path is smoother, eliminating unnecessary motion mutations and generating smoothed cutting path data. Based on the smoothed cutting path data, the coolant injection control parameters of each path are calculated. The coolant flow directly affects the temperature of the cutting tool, especially when the cutting speed and depth change, the coolant flow needs to be adjusted dynamically. By combining the speed and depth information in the cutting path data, the coolant demand in each path is calculated to generate a coolant control sequence. The coolant control sequence reflects the coolant injection pattern during the cutting process to ensure that the LCD screen will not be damaged due to excessive heat generated by cutting. The smoothed cutting path data and the coolant control sequence are timestamped to generate the final cutting path data. By matching the time and coolant flow of each path point, coordinated control during the cutting process can be achieved.

[0088] In a specific embodiment, the process of executing step S400 may specifically include the following steps:

[0089] The cutting path data is processed in time discretization to obtain a coordinate point set in a discrete time series, and based on the coordinate point set, the displacement increment sequence of the X-axis, Y-axis and Z-axis is calculated to obtain a three-axis motion increment matrix;

[0090] According to the three-axis motion increment matrix, a control pulse sequence of the stepper motor is generated through a pulse frequency modulation algorithm to obtain a motor drive signal, and the motor drive signal is digitally filtered to obtain a smoothed drive signal;

[0091] Based on the smoothed drive signal, the real-time position error of each axis is calculated to obtain the position error compensation amount, and the smoothed drive signal is corrected in real time according to the position error compensation amount to obtain a corrected drive signal;

[0092] Based on the corrected driving signal, the frequency domain characteristics of the cutting speed are extracted to obtain a cutting speed feature set, and the coolant flow data during the cutting process is sampled to obtain a flow feature vector. Based on the flow feature vector, the coolant flow feature set is extracted.

[0093] Specifically, the cutting path data is processed in time discretization, and the continuous cutting path is decomposed into a series of coordinate point sets at discrete time points. By setting a fixed time step Δt, the corresponding spatial coordinates are extracted at each time step to form a coordinate point set in a discrete time series. For example, assuming that the parameter equation of the cutting path at time t is r(t), at discrete time point t n =nΔt, the coordinate point is r(t n ), the coordinate point set is {r(t0),r(t1),…,r(t N )}. Based on the coordinate point set, calculate the displacement increment sequence of the X-axis, Y-axis and Z-axis. For two adjacent time points t n and t n+1 , the corresponding displacement increment is:

[0094] Δx n =x n+1 -x n ,Δy n =y n+1 -y n ,Δz n =z n+1 -z n ;

[0095] Among them, x n ,y n ,z n The time t n The X, Y, and Z coordinates at the moment. Arrange all displacement increments to form a three-axis motion increment matrix:

[0096]

[0097] According to the three-axis motion increment matrix, the control pulse sequence of the stepper motor is generated by the pulse frequency modulation algorithm. The displacement of the stepper motor is achieved by receiving a certain number of pulse signals. The frequency and number of pulses determine the speed and displacement of the motor. For each axis, the required number of pulses can be calculated as:

[0098]

[0099] Among them, Δr nis the displacement increment at the nth time step on a certain axis, and s is the step size of the stepper motor (the displacement corresponding to each pulse). By adjusting the frequency of the pulse, the speed of the motor is controlled to meet different cutting speed requirements. The generated motor drive signal may be affected by quantization error and signal noise, and needs to be digitally filtered. Commonly used filters include low-pass filters, which can smooth the signal, eliminate high-frequency noise, and obtain a smoothed drive signal. Based on the smoothed drive signal, the real-time position error of each axis is calculated. In actual operation, the stepper motor may have position deviations due to load changes, mechanical wear, etc. The actual position R of the motor is monitored in real time through a position sensor or encoder. actual (t), and the expected position R desired (t) and the position error is obtained:

[0100] E(t)=R desired (t)-R actual (t);

[0101] The correction signal that needs to be applied is calculated based on the error to obtain the position error compensation. The proportional integral differential controller is used to adjust the drive signal according to the error and its rate of change. The corrected drive signal can be expressed as:

[0102]

[0103] Among them, S smoothed (t) is the smoothed driving signal, K p ,K i ,K d It is the proportional, integral and differential coefficients of the PID controller. Through real-time correction, the position error is significantly reduced and the cutting accuracy is improved. The frequency domain characteristics of the cutting speed are extracted based on the corrected drive signal. By Fourier transforming the drive signal, the frequency domain characteristics are obtained, revealing the periodicity and volatility of the speed change during the cutting process. The frequency domain feature set includes indicators such as main frequency, harmonic components, and spectrum energy distribution. At the same time, the coolant flow data during the cutting process is sampled to obtain the flow feature vector. Changes in coolant flow will also affect the cutting effect and the temperature control of the equipment. By recording the coolant flow at fixed time intervals, time series data is formed. The flow feature vector is analyzed to extract the coolant flow feature set. These features may include average flow, flow fluctuation amplitude, frequency components, etc.

[0104] In a specific embodiment, the process of executing step S500 may specifically include the following steps:

[0105] Normalizing the coolant flow feature set and the cutting speed feature set to obtain standardized input features;

[0106] The standardized input features are input into the feature extractor in the initial cutting process optimization model, and the high-level features are extracted through the residual block to obtain the feature mapping matrix; the feature extractor adopts the ResNet structure, and the feature extractor includes 4 residual blocks, each of which contains 3 convolutional layers and a short-circuit connection, and uses BatchNormalization and ReLU activation functions;

[0107] Perform global average pooling on the feature map matrix to obtain a feature vector of fixed dimension;

[0108] The feature vector is input into the parameter predictor in the initial cutting process optimization model, and nonlinear transformation is performed through a multi-layer perceptron to obtain the initial cutting parameter prediction value; the parameter predictor adopts a multi-layer perceptron structure; the parameter predictor includes 3 fully connected layers, and each layer uses Dropout and LeakyReLU activation functions;

[0109] Based on the initial cutting parameter prediction value and historical cutting data, time series modeling is performed through the preset LSTM network to obtain the time series feature sequence;

[0110] The time series feature sequence is input into the optimization generator in the initial cutting process optimization model to generate the optimized cutting parameter candidate values; the optimization generator adopts the LSTM structure, and the optimization discriminator adopts the one-dimensional convolutional neural network structure; the optimization generator contains 2 layers of LSTM units, each layer has 128 hidden nodes, and uses the Tanh activation function; the optimization discriminator contains 3 one-dimensional convolutional layers, each layer uses the LeakyReLU activation function and the MaxPooling operation;

[0111] The feasibility of the optimized cutting parameter candidate values ​​is verified, and the effective parameters are screened through the preset constraint check module to obtain the effective parameter set, and the effective parameter set is input into the optimization discriminator in the initial cutting process optimization model, and the optimization score of each set of parameters is calculated to obtain the parameter score list;

[0112] According to the parameter scoring list, the parameter combination with the highest score is selected through a sorting algorithm as the target cutting control parameter.

[0113] Specifically, the coolant flow feature set and the cutting speed feature set are normalized, and the feature data of different dimensions are converted to the same scale so that they are comparable in subsequent models and standardized input features are obtained. The standardized input features are input into the feature extractor in the initial cutting process optimization model. The feature extractor adopts the ResNet structure, including 4 residual blocks, each of which consists of 3 convolutional layers and contains a short-circuit connection to ensure that the input of the previous layer can be directly passed to the next layer. Each convolutional layer is followed by a BatchNormalization layer and a ReLU activation function. BatchNormalization can accelerate network convergence, and ReLU introduces nonlinearity to enable the network to learn more complex features. After the input features are processed by the residual block, the extractor extracts high-level features and generates a feature mapping matrix containing important information in the coolant flow and cutting speed features. The feature mapping matrix is ​​subjected to global average pooling. Global average pooling is a commonly used dimensionality reduction method. By performing average pooling on each channel in the matrix, the entire matrix is ​​compressed into a feature vector of fixed dimension. Global average pooling can retain the global information of features and reduce the number of model parameters to avoid overfitting. The fixed-dimensional feature vector is input into the parameter predictor in the initial cutting process optimization model. The parameter predictor adopts a multi-layer perceptron structure to predict the initial cutting parameters through nonlinear transformation. The multi-layer perceptron includes three fully connected layers, each of which uses Dropout technology and LeakyReLU activation function. Dropout is a common method to prevent overfitting. By randomly discarding a part of neurons, the model is made more robust during training. Compared with the traditional ReLU activation function, the LeakyReLU activation function allows a small range of negative outputs, thereby alleviating the "death" problem of ReLU. After the nonlinear transformation of the multi-layer perceptron, the initial cutting parameter prediction value is generated. In order to capture the changing characteristics in the time dimension, based on the initial cutting parameter prediction value and historical cutting data, time series modeling is performed through the preset LSTM network. LSTM is a neural network suitable for processing time series data. Through its built-in memory unit and gating mechanism, it can effectively capture the long-term dependency in the sequence. Through the processing of the LSTM network, the time series feature sequence in the cutting process is obtained. The time series feature sequence is input into the optimization generator in the initial cutting process optimization model to generate optimized cutting parameter candidate values. The optimization generator uses two layers of LSTM units, each layer contains 128 hidden nodes, and uses the Tanh activation function. The Tanh function limits the output to the range of -1 to 1, which can effectively control the range of the model output. Through the LSTM unit processing, the generated optimized cutting parameter candidate values ​​reflect the optimal parameter settings for the cutting process at different time points. At the same time, the optimization discriminator evaluates the generated cutting parameters through a one-dimensional convolutional neural network (CNN) structure.The optimization discriminator contains three one-dimensional convolutional layers, each of which uses the LeakyReLU activation function and downsamples the features through the MaxPooling operation. The one-dimensional convolutional layer can effectively extract local features in the sequence data, while MaxPooling reduces the feature dimension and improves the computational efficiency. The main task of the optimization discriminator is to evaluate whether the generated cutting parameters are valid and assign optimization scores to them. The feasibility of the optimized cutting parameter candidate values ​​is verified. Through the preset constraint check module, the model screens the generated parameters according to the actual process requirements to ensure that they meet the physical constraints and process standards in the LCD screen cutting process. Through the constraint check, the valid parameter set is screened out. The valid parameter set is input into the optimization discriminator to calculate the optimization score of each set of parameters. The optimization score reflects the effect of each set of parameters in actual application. The higher the score, the more the parameter combination meets the cutting requirements. The optimization discriminator generates a parameter score list by scoring each set of parameters. According to the parameter score list, the parameter combination with the highest score is selected as the target cutting control parameter through the sorting algorithm.

[0114] In a specific embodiment, the process of executing step S600 may specifically include the following steps:

[0115] Acquire a real-time cutting image corresponding to a target cutting control parameter, and preprocess the real-time cutting image to obtain an enhanced cutting image;

[0116] Based on the enhanced cutting image, the cutting edge features are extracted to obtain an edge feature map, and the edge feature map is subjected to morphological operations to obtain an optimized edge feature map;

[0117] Based on the optimized edge feature map, the straight line segment is detected by the Hough transform algorithm to obtain the cutting path straight line parameter set, and the cutting path straight line parameter set is clustered and analyzed to obtain the cutting direction feature vector;

[0118] Based on the cutting direction feature vector and the target cutting control parameters, a cutting quality evaluation score is constructed, and the cutting quality score is compared with the preset threshold, and a model optimization strategy is generated through a discriminant function;

[0119] Based on the model optimization strategy, the weights of the initial cutting process optimization model are fine-tuned to obtain updated model parameters, and the updated model parameters are regularized to obtain regularized model parameters;

[0120] The regularized model parameters are updated to the initial cutting process optimization model, and the target cutting process optimization model is generated by reorganizing the model structure.

[0121] Specifically, a real-time cutting image corresponding to the target cutting control parameter is obtained. The image is preprocessed, including image denoising, contrast enhancement and edge enhancement. Common methods include Gaussian filter denoising and adaptive histogram equalization. After preprocessing, an enhanced cutting image is obtained. Based on the enhanced cutting image, cutting edge feature extraction is performed. Edge features refer to the boundaries between the material and the uncut part on the cutting path. These boundaries are usually manifested as significant changes in brightness or color in the image. These edge features are extracted using the Canny edge detection algorithm or the Sobel operator to obtain an edge feature map. Morphological operations are performed on the edge feature map. Morphological operations refer to the expansion and contraction of structural elements in the image to optimize the edge feature map. Common morphological operations include erosion and dilation. The erosion operation can remove small noise points, while the dilation operation helps to enhance the connectivity of the edge. After morphological processing, the edge feature map becomes more coherent and smooth, and an optimized edge feature map is obtained. Based on the optimized edge feature map, the straight line segment is detected by the Hough transform algorithm to obtain a cutting path straight line parameter set. Hough transform is an algorithm for detecting straight lines. It maps edge points in an image from coordinate space to parameter space and finds a set of points that meet the straight line characteristics through an accumulator. The equation of a straight line can be expressed as:

[0122] ρ=xcosθ+ysinθ;

[0123] Among them, ρ is the vertical distance from the straight line to the origin of the coordinate system, and θ is the angle between the straight line and the x-axis. Through the Hough transform algorithm, multiple straight line segments in the cutting path are detected, and the straight line parameter set of the cutting path is obtained. Cluster analysis is performed based on the straight line parameter set of the cutting path. Straight line segments with similar directions and positions are classified into one category to simplify the description of the cutting path. Commonly used algorithms for cluster analysis include K-means and DBSCAN. Through cluster analysis, the representative direction and position of each type of straight line are obtained, and finally a cutting direction feature vector is generated to describe the main trend of the cutting path, which can reflect the overall direction and stability of the cutting process. Based on the cutting direction feature vector and the target cutting control parameters, a cutting quality evaluation score is constructed. The evaluation score mainly depends on the smoothness, straightness and deviation of the cutting path from the predetermined cutting path. By calculating the error between the cutting path and the ideal path, a quantitative evaluation of the cutting quality is obtained. The formula for the evaluation score can be expressed as follows:

[0124]

[0125] Where Q represents the cutting quality evaluation score, d i represents the distance difference between the i-th actual cutting path and the ideal path, d idealis the predetermined ideal distance, and N is the total number of path points. The lower the score, the higher the cutting quality. Compare the cutting quality evaluation score with the preset threshold. If the evaluation score is lower than the threshold, it means that the cutting quality is within an acceptable range; if the evaluation score is higher than the threshold, it indicates that there are deviations or defects in the cutting, and the cutting process needs to be optimized. Through the discriminant function, a model optimization strategy is generated. The discriminant function can be a simple binary classifier used to determine whether the model weights need to be adjusted. Based on the generated model optimization strategy, the weights of the initial cutting process optimization model are fine-tuned. The fine-tuning process can be performed through the back propagation algorithm. The error in the model will be fed back to the weights of each layer of the model according to the evaluation score, and the model parameters will be adjusted to improve the cutting accuracy. The fine-tuned model parameters need to be regularized to prevent the model from overfitting. Commonly used regularization methods include L2 regularization or weight decay. The formula is:

[0126]

[0127] Among them, L regularization is the regularization loss, in is the regularization coefficient, w i is the weight parameter in the model. Regularization imposes penalties on weights to avoid excessive weight values, thereby enhancing the generalization ability of the model. After the regularization process is completed, the regularized model parameters are updated to the initial cutting process optimization model. The target cutting process optimization model is generated by reorganizing the model structure. The model reorganization process can include adding or deleting layers in the neural network, adjusting the size of the convolution kernel, or changing the activation function. The reorganized model is more suitable for the current cutting task and can provide better performance in the cutting process.

[0128] See also Figure 2 , Figure 2 A schematic block diagram of the structure of a cutting control device 200 for a liquid crystal screen provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the cutting control device 200 of the liquid crystal screen includes:

[0129] A construction module 210 is used to construct a three-dimensional digital twin model of the liquid crystal screen and calculate thickness parameters and flatness parameters of the liquid crystal screen;

[0130] A calculation module 220, configured to optimize and calculate the initial cutting planning parameters of the liquid crystal screen based on the thickness parameter and the flatness parameter to obtain the optimized cutting planning parameters;

[0131] A generating module 230, for generating cutting path data through a path planning algorithm according to the optimized cutting planning parameters and the structural characteristics of the liquid crystal screen;

[0132] The control module 240 is used to control the X-axis, Y-axis and Z-axis stepper motors of the double-sided TFT cutting machine in real time based on the cutting path data, and to perform real-time analysis and feature extraction on the coolant flow rate and cutting speed during the cutting process to obtain a coolant flow rate feature set and a cutting speed feature set;

[0133] Prediction module 250, used for inputting the coolant flow characteristic set and the cutting speed characteristic set into a preset initial cutting process optimization model to dynamically predict and adjust the cutting parameters to obtain target cutting control parameters;

[0134] The optimization module 260 is used to obtain a real-time cutting image corresponding to the target cutting control parameters, and optimize the initial cutting process optimization model according to the real-time cutting image to generate a target cutting process optimization model.

[0135] Through the collaboration of the above components, by building a three-dimensional digital twin model of the LCD screen, the thickness and flatness of the screen can be accurately calculated. The cutting planning parameter optimization algorithm based on thickness parameters and flatness parameters can dynamically adjust the cutting depth, speed and coolant flow according to the actual characteristics of the screen, improving the adaptability and stability of the cutting process. The cutting path data is generated by the path planning algorithm, and the curve is smoothed to optimize the cutting trajectory, reduce stress concentration during the cutting process, and reduce the risk of screen damage. By real-time regulation of the X, Y, and Z axis stepper motors of the double-sided TFT cutting machine, combined with real-time analysis of coolant flow and cutting speed, the cutting process is accurately controlled and the consistency of cutting quality is improved. The preset initial cutting process optimization model is introduced, and the cutting parameters are adjusted by dynamic prediction, which realizes the intelligent control of the cutting process and improves the adaptive ability of the system. The efficient prediction and optimization of cutting parameters are achieved, and the generalization ability and robustness of the model are improved. The continuous improvement of the cutting process is achieved through real-time cutting image feedback and model optimization mechanism.

[0136] The present application also provides a cutting control device for a liquid crystal screen, the cutting control device for the liquid crystal screen comprising a memory and a processor, the memory storing computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the cutting control method for the liquid crystal screen in the above-mentioned embodiments.

[0137] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein instructions are stored in the computer, and when the instructions are executed on a computer, the computer executes the steps of the cutting control method of the liquid crystal screen.

[0138] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0139] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0140] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A cutting control method for a liquid crystal screen, characterized in that: include: Constructing a three-dimensional digital twin model of a liquid crystal screen, and calculating thickness parameters and flatness parameters of the liquid crystal screen; Based on the thickness parameter and the flatness parameter, the initial cutting planning parameters of the liquid crystal screen are optimized and calculated to obtain the optimized cutting planning parameters; specifically, the method includes: based on the thickness parameter, linearly mapping the initial cutting depth to obtain the initial depth mapping value, and calculating the curvature distribution of the surface of the liquid crystal screen according to the flatness parameter to obtain the curvature distribution matrix; based on the curvature distribution matrix, locally adjusting the initial depth mapping value to obtain the adjusted cutting depth value, and calculating the pressure distribution of the cutting tool according to the adjusted cutting depth value to obtain the pressure distribution curve; based on the pressure distribution curve, performing piecewise function mapping on the initial cutting speed to obtain the speed adjustment coefficient, and correcting the initial cutting speed according to the speed adjustment coefficient to obtain the optimized cutting speed value; based on the optimized cutting speed value and the adjusted cutting depth value, calculating the heat distribution in the cutting process to obtain the heat distribution model, and performing nonlinear optimization on the initial coolant flow according to the heat distribution model to obtain the optimized coolant flow value; combining the optimized cutting speed value, the adjusted cutting depth value and the optimized coolant flow value into the optimized cutting planning parameters; Generate cutting path data through a path planning algorithm according to the optimized cutting planning parameters and the structural characteristics of the liquid crystal screen; Based on the cutting path data, the X-axis, Y-axis and Z-axis stepper motors of the double-sided TFT cutting machine are controlled in real time, and the coolant flow rate and cutting speed in the cutting process are analyzed and feature extracted in real time to obtain a coolant flow rate feature set and a cutting speed feature set; Inputting the coolant flow characteristic set and the cutting speed characteristic set into a preset initial cutting process optimization model to dynamically predict and adjust cutting parameters to obtain target cutting control parameters; A real-time cutting image corresponding to the target cutting control parameter is obtained, and the initial cutting process optimization model is optimized according to the real-time cutting image to generate a target cutting process optimization model.

2. The cutting control method of the liquid crystal screen according to claim 1, characterized in that: The constructing of the three-dimensional digital twin model of the liquid crystal screen and calculating the thickness parameter and flatness parameter of the liquid crystal screen includes: Performing multi-angle high-resolution scanning on the LCD screen to obtain a point cloud data set of the LCD screen, and constructing an initial three-dimensional model of the LCD screen based on the point cloud data set; Performing mesh division and topological optimization on the initial three-dimensional model to obtain an optimized three-dimensional mesh model, and calculating stress distribution and deformation characteristics of the liquid crystal screen based on the optimized three-dimensional mesh model to obtain a stress distribution map and a deformation characteristic map; According to the stress distribution map and the deformation characteristic map, locally refine the optimized three-dimensional mesh model to obtain a refined three-dimensional mesh model; Based on the refined three-dimensional grid model, the height distribution of the surface of the liquid crystal screen is calculated to obtain a height distribution matrix, and the height distribution matrix is ​​statistically analyzed to obtain the flatness parameter of the liquid crystal screen by least squares fitting; Based on the refined three-dimensional grid model, the contour information of the liquid crystal screen is extracted, and the thickness parameter of the liquid crystal screen is calculated.

3. The cutting control method of the liquid crystal screen according to claim 1, characterized in that: The step of generating cutting path data by a path planning algorithm according to the optimized cutting planning parameters and the structural features of the liquid crystal screen includes: Based on the structural characteristics of the liquid crystal screen, the liquid crystal screen is gridded to obtain a cutting area grid matrix, and the cutting boundary of the liquid crystal screen is extracted according to the cutting area grid matrix to obtain a cutting boundary coordinate set; Based on the cutting boundary coordinate set, an initial cutting path is calculated to obtain an initial path node sequence, and the initial path node sequence is locally optimized to obtain an optimized path node sequence; Based on the optimized path node sequence and the cutting speed value in the optimized cutting planning parameters, the movement time between each node is calculated to obtain a time constraint matrix; According to the time constraint matrix, the optimized path node sequence is time-synchronized to obtain a synchronized path node sequence, and three-dimensional cutting path data is generated based on the synchronized path node sequence and the cutting depth value in the optimized cutting planning parameters; Performing curve smoothing processing on the three-dimensional cutting path data, generating a continuous and smooth motion trajectory through a cubic spline interpolation algorithm, and obtaining smoothed cutting path data; Based on the smoothed cutting path data and the coolant flow value in the optimized cutting planning parameters, the coolant injection control parameters of each path segment are calculated to obtain a coolant control sequence; The smoothed cutting path data and the coolant control sequence are time stamp aligned to generate final cutting path data.

4. The cutting control method of the liquid crystal screen according to claim 1, characterized in that: Based on the cutting path data, the X-axis, Y-axis and Z-axis stepper motors of the double-sided TFT cutting machine are controlled in real time, and the coolant flow rate and cutting speed in the cutting process are analyzed and feature extracted in real time to obtain a coolant flow rate feature set and a cutting speed feature set, including: Performing time discretization processing on the cutting path data to obtain a coordinate point set in a discrete time sequence, and based on the coordinate point set, calculating the displacement increment sequence of the X-axis, Y-axis and Z-axis to obtain a three-axis motion increment matrix; According to the three-axis motion increment matrix, a control pulse sequence of the stepping motor is generated by a pulse frequency modulation algorithm to obtain a motor drive signal, and the motor drive signal is digitally filtered to obtain a smoothed drive signal; Based on the smoothed drive signal, the real-time position error of each axis is calculated to obtain a position error compensation amount, and the smoothed drive signal is corrected in real time according to the position error compensation amount to obtain a corrected drive signal; Based on the corrected driving signal, the frequency domain characteristics of the cutting speed are extracted to obtain a cutting speed feature set, and the coolant flow data during the cutting process is sampled to obtain a flow feature vector. Based on the flow feature vector, the coolant flow feature set is extracted.

5. The cutting control method of the liquid crystal screen according to claim 1, characterized in that: The step of inputting the coolant flow characteristic set and the cutting speed characteristic set into a preset initial cutting process optimization model to dynamically predict and adjust cutting parameters to obtain target cutting control parameters includes: Normalizing the coolant flow feature set and the cutting speed feature set to obtain standardized input features; The standardized input features are input into a feature extractor in an initial cutting process optimization model, and high-level features are extracted through a residual block to obtain a feature mapping matrix; the feature extractor adopts a ResNet structure, and the feature extractor includes 4 residual blocks, each of which includes 3 convolutional layers and a short-circuit connection, and uses BatchNormalization and ReLU activation functions; Performing a global average pooling operation on the feature mapping matrix to obtain a feature vector of fixed dimension; The feature vector is input into the parameter predictor in the initial cutting process optimization model, and a nonlinear transformation is performed through a multi-layer perceptron to obtain an initial cutting parameter prediction value; the parameter predictor adopts a multi-layer perceptron structure; the parameter predictor includes 3 fully connected layers, and each layer uses Dropout and LeakyReLU activation functions; Based on the initial cutting parameter prediction value and the historical cutting data, time series modeling is performed through a preset LSTM network to obtain a time series feature sequence; The temporal feature sequence is input into the optimization generator in the initial cutting process optimization model to generate optimized cutting parameter candidate values; the optimization generator adopts an LSTM structure, and the optimization discriminator adopts a one-dimensional convolutional neural network structure; the optimization generator includes 2 layers of LSTM units, each layer has 128 hidden nodes, and uses a Tanh activation function; the optimization discriminator includes 3 one-dimensional convolutional layers, each layer uses a LeakyReLU activation function and a MaxPooling operation; Performing feasibility verification on the optimized cutting parameter candidate values, screening valid parameters through a preset constraint check module to obtain a valid parameter set, and inputting the valid parameter set into the optimization discriminator in the initial cutting process optimization model, calculating the optimization score of each group of parameters, and obtaining a parameter score list; According to the parameter scoring list, a parameter combination with the highest score is selected through a sorting algorithm as the target cutting control parameter.

6. The cutting control method of the liquid crystal screen according to claim 1, characterized in that: The acquiring of the real-time cutting image corresponding to the target cutting control parameter, and optimizing the initial cutting process optimization model according to the real-time cutting image to generate the target cutting process optimization model, comprises: Acquire a real-time cutting image corresponding to the target cutting control parameter, and preprocess the real-time cutting image to obtain an enhanced cutting image; Based on the enhanced cutting image, cutting edge features are extracted to obtain an edge feature map, and morphological operations are performed on the edge feature map to obtain an optimized edge feature map; Based on the optimized edge feature map, the straight line segments are detected by using the Hough transform algorithm to obtain a cutting path straight line parameter set, and the cutting path straight line parameter set is clustered and analyzed to obtain a cutting direction feature vector; Based on the cutting direction feature vector and the target cutting control parameter, a cutting quality evaluation score is constructed, and the cutting quality score is compared with a preset threshold value, and a model optimization strategy is generated through a discriminant function; Based on the model optimization strategy, the weights of the initial cutting process optimization model are fine-tuned to obtain updated model parameters, and the updated model parameters are regularized to obtain regularized model parameters; The regularized model parameters are updated to the initial cutting process optimization model, and a target cutting process optimization model is generated by reorganizing the model structure.

7. A cutting control device for a liquid crystal screen, characterized in that: A method for controlling the cutting of a liquid crystal screen according to any one of claims 1 to 6, comprising: A construction module, used to construct a three-dimensional digital twin model of a liquid crystal screen and calculate thickness parameters and flatness parameters of the liquid crystal screen; A calculation module is used to optimize and calculate the initial cutting planning parameters of the liquid crystal screen based on the thickness parameter and the flatness parameter to obtain the optimized cutting planning parameters; specifically comprising: linearly mapping the initial cutting depth based on the thickness parameter to obtain an initial depth mapping value, and calculating the curvature distribution of the surface of the liquid crystal screen based on the flatness parameter to obtain a curvature distribution matrix; locally adjusting the initial depth mapping value based on the curvature distribution matrix to obtain an adjusted cutting depth value, and calculating the pressure distribution of the cutting tool based on the adjusted cutting depth value to obtain a pressure distribution curve; The pressure distribution curve is used to perform piecewise function mapping on the initial cutting speed to obtain a speed adjustment coefficient, and the initial cutting speed is corrected according to the speed adjustment coefficient to obtain an optimized cutting speed value; based on the optimized cutting speed value and the adjusted cutting depth value, the heat distribution in the cutting process is calculated to obtain a heat distribution model, and according to the heat distribution model, the initial coolant flow rate is nonlinearly optimized to obtain an optimized coolant flow rate value; the optimized cutting speed value, the adjusted cutting depth value and the optimized coolant flow rate value are combined into optimized cutting planning parameters; A generating module, used for generating cutting path data through a path planning algorithm according to the optimized cutting planning parameters and the structural characteristics of the liquid crystal screen; A control module is used to perform real-time control on the X-axis, Y-axis and Z-axis stepper motors of the double-sided TFT cutting machine based on the cutting path data, and to perform real-time analysis and feature extraction on the coolant flow rate and cutting speed during the cutting process to obtain a coolant flow rate feature set and a cutting speed feature set; A prediction module, used for inputting the coolant flow characteristic set and the cutting speed characteristic set into a preset initial cutting process optimization model to dynamically predict and adjust cutting parameters to obtain target cutting control parameters; The optimization module is used to obtain the real-time cutting image corresponding to the target cutting control parameter, and optimize the initial cutting process optimization model according to the real-time cutting image to generate a target cutting process optimization model.

8. A cutting control device for a liquid crystal screen, characterized in that: The cutting control device of the liquid crystal screen comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instruction in the memory to enable the cutting control device of the liquid crystal screen to execute the cutting control method of the liquid crystal screen according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the cutting control method of the liquid crystal screen as described in any one of claims 1 to 6 is implemented.

Images