Virtual liquid drop closed-loop regulation and control method and system based on digital twinning

Abstract

The invention belongs to the related technical field of jet printing manufacturing, and particularly relates to a virtual liquid drop closed-loop regulation and control method and system based on digital twinning, and the method comprises the steps: collecting a current working condition parameter in real time, and uploading the current working condition parameter to a digital twinning platform; through the digital twin platform, based on the current working condition parameters, waveform parameters are optimized and fed back to the entity nozzle module; a virtual equipment model of an entity nozzle module in a virtual space is configured in the twin platform, the virtual equipment model comprises a liquid drop state parameter hybrid prediction model and a reverse waveform prediction model, and the liquid drop state parameter hybrid prediction model is used for predicting jet liquid drop state information based on multi-dimensional working condition parameters; and the reverse waveform prediction model is used for reversely reasoning nozzle waveform parameters as optimized waveform parameters based on predicted jet liquid drop state information and target liquid drop state parameters set by a user. The problems that the existing ink-jet printing liquid drop state is difficult to perceive in real time and the process debugging depends on experience trial and error can be solved, and the process debugging efficiency of the nozzle is improved.

Description

Technical Field

[0001] This invention belongs to the technical field of inkjet printing manufacturing, and more specifically, relates to a virtual droplet closed-loop control method and system based on digital twins. Background Technology

[0002] Inkjet printing technology, with its advantages of not requiring masks, high material utilization, and non-contact operation, is more suitable for OLED display panel manufacturing than traditional vapor deposition processes. The development of inkjet printing equipment for large-size OLED display panel manufacturing is gradually becoming a focus of industry attention.

[0003] Large-format inkjet printing equipment is evolving towards higher precision, higher yield, and rapid iteration. As a key component of printing equipment, the printhead's performance often fluctuates during actual processes, making it difficult to predict and maintain stable performance over long periods. This can lead to printhead slippage, eventually resulting in nozzle clogging or uncontrollable waveform parameter adjustments. The time-varying nature of the printhead poses a significant challenge to the efficient control of waveform parameters by process engineers, substantially impacting printing efficiency and accuracy, and reducing the stability and reliability of inkjet printing equipment.

[0004] Currently, in the actual panel printing process, it is impossible to perceive the printhead ejection status in real time during the inkjet printing droplet ejection process. Furthermore, the most common method used in actual process debugging is parameter trial and error. Therefore, the printing performance of inkjet printing equipment and the efficiency of printhead process debugging are greatly restricted. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a virtual droplet closed-loop control method and system based on digital twins. Its purpose is to solve the problem that the droplet state in inkjet printing is difficult to perceive in real time and the process debugging relies on experience trial and error, thereby improving the efficiency of printhead process debugging.

[0006] To achieve the above objectives, according to one aspect of the present invention, a virtual droplet closed-loop control method based on digital twins is provided, comprising: Real-time acquisition of current operating parameters is uploaded to the digital twin platform; based on the current operating parameters, the waveform parameters are optimized and fed back to the physical nozzle module to achieve real-time control of the physical nozzle. The digital twin platform is equipped with a high-fidelity virtual equipment model of the physical printhead module in virtual space. It includes a droplet state parameter hybrid prediction model and a reverse waveform prediction model. The droplet state parameter hybrid prediction model is used to predict the state information of the ejected droplets based on multi-dimensional operating parameters including printhead waveform parameters, ink path back pressure, and ink cavity temperature. The reverse waveform prediction model is used to inversely deduce the printhead waveform parameters that meet the ejection requirements based on the predicted ejected droplet state information and the target droplet state parameters set by the user, as optimized waveform parameters.

[0007] Furthermore, the digital twin platform is also equipped with a three-dimensional visualization interface for multi-angle dynamic interaction and visualization of droplet state. The multi-angle dynamic interaction includes the input of process parameters and the feedback of reverse reasoning results.

[0008] Furthermore, the method for constructing the droplet state parameter hybrid prediction model is as follows: A neural network structure was trained based on historical operating data to obtain a basic prediction model for droplet state parameters. By conducting trial printing with physical nozzle modules, a small sample working condition dataset was collected. This small sample working condition dataset was then used to perform residual calibration on the basic model of the droplet state parameter prediction algorithm, forming a hybrid prediction model for droplet state parameters. The residual calibration method is as follows: Using a droplet state parameter-based prediction model for a small sample dataset of operating conditions Make predictions and calculate the prediction residuals: In the formula, To predict residuals, For small sample working condition datasets The measured droplet state information in the data. For the basic prediction model of droplet state parameters The predicted output is the droplet state information. This indicates multi-dimensional operating parameters including waveform parameters, ink path back pressure, and ink cavity temperature under the current operating conditions. K Represents a small sample working condition dataset The number of samples; by As supervised samples, residual calibration models are constructed through training. The residual calibration model is linearly combined with the basic prediction model of droplet state parameters to form a hybrid prediction model of droplet state parameters.

[0009] Furthermore, both the basic prediction model for droplet state parameters and the residual calibration model use MLP networks as their network structures.

[0010] Furthermore, the reverse waveform prediction model is trained using a multi-objective optimization method based on historical operating data.

[0011] Furthermore, the configuration method for the high-fidelity virtual equipment model is as follows: Geometric mapping: The physical coordinates and virtual coordinates of the physical nozzle module are aligned using a rigid-scale variation alignment formula to achieve geometric mapping; the rigid-scale variation alignment formula is as follows:

[0012] In the formula, This represents the positioning anchor point in the virtual equipment model, that is, the display coordinates determined by the virtual equipment model. This represents the coordinates of the corresponding points on the physical nozzle model that correspond to the virtual equipment model. Represents the global scale factor. , R Indicates coordinate rotation transformation. It is a three-dimensional rotation matrix. t Indicates coordinate translation changes. It is a translation vector; Behavior mapping: A combination of prior prediction and measured correction is used to iterate the isomorphic state vector of the physical nozzle module and synchronously map it to the behavior of the physical nozzle module. The prior prediction and measured correction methods are as follows:

[0013] In the formula, For the solid nozzle model in discrete control cycle isomorphic state vectors under the following conditions For the solid nozzle model in discrete control cycle isomorphic state vectors under; To provide a priori prediction model for the state of the next control cycle under the same time base. For the observation correction model, the deviation between the prior prediction model and the actual observation results is considered. Furthermore, the isomorphic state of the solid nozzle model is mapped and corrected. This represents the actual observed droplet state information. This represents the droplet state information calculated by the prior prediction model.

[0014] According to another aspect of the present invention, a virtual droplet closed-loop control system based on digital twin is provided for executing the virtual droplet closed-loop control method as described above, comprising: a nozzle module, a virtual-real interaction module, and a digital twin platform; The nozzle module is used for hardware operations, controlling the droplet spraying function, and collecting operating parameters in real time. The virtual-real interaction module is used to acquire operating parameters and convert the data into a unified format before uploading it to the digital twin platform; The digital twin platform is used to optimize waveform parameters based on current operating conditions and feed them back to the physical printhead module, enabling real-time control of the physical printhead. The digital twin platform includes a high-fidelity virtual equipment model of the physical printhead module in virtual space, comprising a droplet state parameter hybrid prediction model and a reverse waveform prediction model. The droplet state parameter hybrid prediction model predicts the ejected droplet state information based on multi-dimensional operating conditions including printhead waveform parameters, ink path back pressure, and ink cavity temperature. The reverse waveform prediction model, based on the predicted ejected droplet state information and the user-defined target droplet state parameters, inversely deduces the printhead waveform parameters that meet the ejection requirements, serving as the optimized waveform parameters.

[0015] According to another aspect of the present invention, a debugging method for the virtual droplet closed-loop control system as described above is provided, comprising: By conducting test printing, the system verifies whether the mixed prediction model of droplet state parameters in the digital twin platform is distorted. If not, the debugging process ends. If so, a small sample dataset of operating conditions is collected from the test printing, and the basic prediction model of droplet state parameters is used to analyze the small sample dataset. Make predictions and calculate the prediction residuals: In the formula, To predict residuals, For small sample working condition datasets The measured droplet state information in the data. For the basic prediction model of droplet state parameters The predicted output is the droplet state information. This indicates multi-dimensional operating parameters including waveform parameters, ink path back pressure, and ink cavity temperature under the current operating conditions. K Represents a small sample working condition dataset The number of samples; by As supervised samples, residual calibration models are constructed through training. The residual calibration model and the basic prediction model of droplet state parameters are linearly combined to form a new hybrid prediction model of droplet state parameters, and the system debugging is completed.

[0016] In summary, compared with the prior art, the technical solutions conceived by this invention have the following main advantages: 1. This invention proposes a virtual droplet closed-loop control method based on digital twins. Through the interaction between the physical equipment layer and the digital twin layer, it fully utilizes high-speed camera and sensor data, historical state data, and algorithm modeling. In the digital twin layer, a hybrid prediction model for droplet state parameters and a reverse waveform prediction model are integrated. The hybrid prediction model predicts the ejected droplet state information based on multi-dimensional operating parameters including printhead waveform parameters, ink path back pressure, and ink cavity temperature. The reverse waveform prediction model infers the printhead waveform parameters that meet the ejection requirements based on the predicted ejected droplet state information and the user-defined target droplet state parameters. These optimized waveform parameters are then fed back to the physical equipment layer, achieving virtual droplet closed-loop control. This method, introduced through a digital twin platform, can efficiently predict and control key droplet state parameters such as droplet volume, velocity, and ejection angle, effectively improving the efficiency and accuracy of inkjet printing process debugging. It solves the problem in existing technologies where real-time perception of droplet state is difficult and process debugging relies on trial and error, thus improving printhead process debugging efficiency.

[0017] 2. Building upon point 1, this invention further proposes a configuration method for a high-fidelity virtual equipment model. It introduces a rigid-scale variation alignment method to align the physical coordinates of the physical printhead module with the virtual coordinates, and introduces prior prediction and measured correction methods to achieve state isomorphism and behavior mapping. This concretizes and standardizes the digital twin system in the inkjet printing field. It should be noted that this configuration method for the high-fidelity virtual equipment model is merely one preferred embodiment of this invention and should not be construed as the sole limitation of the invention. In practical applications of the technical solution corresponding to point 1, other methods can also be used to configure the high-fidelity virtual equipment model.

[0018] 3. This invention also proposes another technical solution: a debugging method for a virtual droplet closed-loop control system. Addressing the problem of model prediction inaccuracies caused by nozzle state drift over time, it proposes a small-sample-driven residual calibration mechanism. This mechanism utilizes small samples collected during the trial printing phase to efficiently correct the basic model output in real time, thus constructing a hybrid model for droplet state parameter prediction algorithms. Simultaneously, it is used in conjunction with the waveform prediction algorithm model. This method generates multiple optimized schemes to assist process engineers in parameter selection and enables adaptive control through automatic system feedback, effectively improving model robustness and debugging efficiency, and significantly reducing process debugging time and material costs. It constructs a "prediction-calibration-control-feedback" closed loop, enabling high-fidelity debugging and state awareness of the inkjet printing process in virtual space. This effectively enhances the monitoring capability of the inkjet printing process and significantly improves the efficiency of inkjet printing process debugging, solving pain points in existing technologies such as difficulty in real-time perception of inkjet printing droplet state, reliance on experience-based trial and error for process debugging, and model inaccuracies caused by printhead degradation. Attached Figure Description

[0019] Figure 1 A schematic diagram of a virtual droplet closed-loop control method based on digital twin provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a debugging method for a virtual droplet closed-loop control system based on digital twins, provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0021] Example 1 A closed-loop control method for virtual droplets based on digital twins, comprising: Real-time acquisition of current operating parameters is uploaded to the digital twin platform; based on the current operating parameters, the waveform parameters are optimized and fed back to the physical nozzle module to achieve real-time control of the physical nozzle. The digital twin platform is equipped with a high-fidelity virtual equipment model of the physical printhead module in virtual space. It includes a droplet state parameter hybrid prediction model and a reverse waveform prediction model. The droplet state parameter hybrid prediction model is used to predict the state information of the ejected droplets based on multi-dimensional operating parameters including printhead waveform parameters, ink path back pressure, and ink cavity temperature. The reverse waveform prediction model is used to inversely deduce the printhead waveform parameters that meet the ejection requirements based on the predicted ejected droplet state information and the target droplet state parameters set by the user, as optimized waveform parameters.

[0022] This embodiment proposes a virtual droplet closed-loop control method based on digital twins. Through the interaction between the physical equipment layer and the digital twin layer, it fully utilizes high-speed camera and sensor data, historical state data, and algorithm modeling. In the digital twin layer, a hybrid prediction model for droplet state parameters and a reverse waveform prediction model are integrated. The hybrid prediction model predicts the ejected droplet state information based on multi-dimensional operating parameters including printhead waveform parameters, ink path back pressure, and ink cavity temperature. The reverse waveform prediction model, based on the predicted ejected droplet state information and the user-defined target droplet state parameters, inversely deduces the printhead waveform parameters that meet the ejection requirements. These optimized waveform parameters are then fed back to the physical equipment layer, achieving virtual droplet closed-loop control. This method, introduced through a digital twin platform, can efficiently predict and control key droplet state parameters such as droplet volume, velocity, and ejection angle. It effectively improves the efficiency and accuracy of inkjet printing process debugging, solving the problem in existing technologies where real-time perception of droplet state is difficult and process debugging relies on trial and error, thus improving printhead process debugging efficiency.

[0023] As a priority, the digital twin platform is also equipped with a three-dimensional visualization interface for multi-angle dynamic interaction and visualization of droplet state. The multi-angle dynamic interaction includes the input of process parameters and the feedback of reverse reasoning results.

[0024] The system architecture corresponding to the method in this embodiment is as follows: Figure 1 As shown, a virtual-real fusion system consisting of a physical equipment layer and a digital twin layer has been established. The two layers interact through multi-channel data and model linkage to form an inkjet process support platform with high-fidelity mapping and closed-loop control capabilities.

[0025] As a preferred implementation method, the droplet state parameter mixing prediction model is constructed as follows: (1) Train the neural network structure based on historical working condition data to obtain the basic prediction model of droplet state parameters; For example, it includes the following sub-steps: S11: Construct each set of samples in the historical dataset into an input-output pair. x i ,y i ), where input

[0026] Specifically, Includes waveform parameters, ink path back pressure, ink cavity temperature, etc. d 3D variable; output parameter is droplet volume droplet velocity and injection tilt angle :

[0027] S12: Define the basic model of the prediction algorithm For having L Layered MLP network structure:

[0028] Specifically, Indicates the output parameters. This indicates parameters including waveform parameters, ink path back pressure, and ink cavity temperature. d dimensional variables, This represents the weight parameters of the MLP network layer.

[0029] The forward propagation method of the network is as follows:

[0030]

[0031]

[0032]

[0033] in, For predicting the base model k Layer weights, For predicting the base model, the first k Layer bias vector, for Relu The activation function, with the number of hidden layer nodes denoted as . (h 1 ,h 2 ,h 3 ) .

[0034] S13: Construct a loss function to measure the error between the predicted value and the actual observed value. Taking the root mean square as an example, the definition is as follows:

[0035] S14: Iteratively train the model using historical datasets, minimize the loss function using optimization algorithms such as gradient descent, and finally obtain a stable and convergent basic model for droplet volume prediction.

[0036] (2) By printing a test printout using a physical printhead module, a small sample working condition dataset is collected. The small sample working condition dataset is then used to perform residual calibration on the basic model of the droplet state parameter prediction algorithm to form a hybrid prediction model for droplet state parameters. The residual calibration method is as follows: S21: Collect a small sample dataset of test prints under the current printhead operating condition, and record... ,in This represents the input feature vector under the current operating conditions. These are the measured droplet state parameters; S22: Use the basic prediction model to predict the results of the trial print samples and calculate the prediction residuals:

[0037] in, This represents the input feature vector (including waveform parameters, ink path back pressure, ink cavity temperature, etc.) under the current operating conditions. For measured droplet state parameters (including droplet volume) droplet velocity and injection tilt angle ); S23: with As training samples, a residual calibration model is constructed, with an MLP network structure. Specifically, an MLP residual network with one hidden layer is established. , for input Output residual prediction:

[0038] in, To predict the residual vector, the droplet state parameters are... The output error is estimated. The set of all trainable parameters for the residual network includes: For hidden layer weights, This is the hidden layer bias vector. For ReLU activation functions, For output layer weights, This is the output layer bias vector.

[0039] S24: Linearly combine the residual calibration model with the basic model to form a hybrid prediction model. :

[0040] in, This is the input feature vector under the current operating conditions (including waveform parameters, ink path back pressure, ink cavity temperature, etc.). The output value of the mixed prediction model for droplet state parameters. The output values ​​of the basic prediction model for droplet state parameters. This is used to calibrate the output value of the residual calibration network.

[0041] As a preferred implementation, the reverse waveform prediction model is trained using a multi-objective optimization method based on historical operating data.

[0042] As a preferred implementation method, the high-fidelity virtual equipment model can be configured as follows: The overall state (geometric and behavioral actions) of the nozzle module is stably and verifiably mapped to the virtual equipment model established by the digital twin system, supporting subsequent closed-loop control and traceability within the digital twin system, including: Scene alignment and geometry mapping require the physical coordinates of the nozzle module. Virtual coordinates of the twin nozzle module established in the visualization module Establish a verifiable and repeatable mapping relationship between them. Align the physical and virtual coordinates using a rigid-scale change alignment method:

[0043] Specifically, This represents the corresponding point (anchor point) in the virtual model, i.e., the display coordinates determined by the digital twin layer model. This represents the coordinates of the corresponding points of the physical equipment in the digital twin layer model. This represents the global scale factor, taking into account the simplification and scale adjustment required for the twin model in the digital twin layer. That is, the dimensions of the twin equipment are scaled down relative to the corresponding physical equipment. R Indicates coordinate rotation transformation. It is a three-dimensional rotation matrix. t Indicates coordinate translation changes. It is a translation vector.

[0044] State isomorphism and behavior mapping: The virtual-real interaction module needs to map the measurable / calculable low-dimensional isomorphic spray state of the nozzle module to the twin model within a single control cycle, maintaining stable alignment even in the presence of noise. A method combining prior prediction and experimental correction is used to iterate the isomorphic state vector of the twin equipment while maintaining its behavior mapping with the physical equipment.

[0045] In the formula, For the solid nozzle model in discrete control cycle isomorphic state vectors under the following conditions For the solid nozzle model in discrete control cycle isomorphic state vectors under; To provide a priori prediction model for the state of the next control cycle under the same time base. For the observation correction model, the deviation between the prior prediction model and the actual observation results is considered. Furthermore, the isomorphic state of the solid nozzle model is mapped and corrected. This represents the actual observed droplet state information. This represents the droplet state information calculated by the prior prediction model.

[0046] This embodiment utilizes the interaction between the physical equipment layer and the digital twin layer to fully leverage high-speed camera and sensor data, historical status data, and algorithm modeling. This enables high-fidelity debugging and status awareness of the inkjet printing process in virtual space, effectively enhancing the monitoring capabilities of the inkjet printing process and significantly improving the efficiency of inkjet printing process debugging.

[0047] Example 2 A virtual droplet closed-loop control system based on digital twins is used for closed-loop control and visual monitoring of droplets in inkjet printing processes. The system includes: a printhead module, a virtual-real interaction module, and a digital twin platform. The nozzle module is used for hardware operations, controlling the droplet ejection function and acquiring real-time data such as waveform parameters, droplet volume, flight velocity, and ejection angle. As an example, the nozzle module includes: a nozzle and control board, a droplet observation module, a DMU detection module, and a multi-axis motion system, enabling multi-dimensional observation and data acquisition of the droplet state. The nozzle and control board control the droplet ejection behavior through waveform drive; the droplet observation module and DMU detection module acquire key process state parameters such as the actual droplet volume, velocity, and ejection angle; and the multi-axis motion system adjusts the observation angle and provides precise positioning.

[0048] The virtual-real interaction module is used to acquire various types of data (including waveform parameters, droplet velocity, and spray angle) from the nozzle module's control of droplet ejection, and converts the data into a unified format for easy extraction and use in the digital twin system. As an example, the virtual-real interaction module communicates with the nozzle module via the Modbus TCP communication protocol, and stores droplet ejection images and video data in the digital twin system database via FTP sharing. The optimized waveform parameters calculated by the platform are fed back to the nozzle module through the control interface, achieving bidirectional transmission of real-time nozzle data.

[0049] The digital twin platform is used to optimize waveform parameters based on current operating conditions and feed them back to the physical printhead module, enabling real-time control of the physical printhead. The digital twin platform includes a high-fidelity virtual equipment model of the physical printhead module in virtual space, comprising a droplet state parameter hybrid prediction model and a reverse waveform prediction model. The droplet state parameter hybrid prediction model predicts the ejected droplet state information based on multi-dimensional operating conditions including printhead waveform parameters, ink path back pressure, and ink cavity temperature. The reverse waveform prediction model, based on the predicted ejected droplet state information and the user-defined target droplet state parameters, inversely deduces the printhead waveform parameters that meet the ejection requirements, serving as the optimized waveform parameters to achieve virtual control of the printhead module's waveform parameters.

[0050] In addition, the system may include a visualization module to display visual animations of droplet flight, parameter responses, and prediction feedback results, and to provide an interactive interface for process debugging. As an example, the visualization module uses 3D visualization software such as Unity 3D to construct a virtual scene of the nozzle module, supporting multi-angle interaction. It includes a process parameter input area, an experimental feedback area, and a droplet flight animation display area, achieving high-fidelity dynamic simulation and real-time interaction of the droplet ejection process.

[0051] The above system can be constructed according to the following steps: S1, Hardware system setup, assembly of nozzle module, including nozzle and control board, droplet observation module, DMU detection module and multi-axis motion platform, to ensure the acquisition of key data such as waveform parameters, droplet volume, velocity and tilt angle; S2, the virtual-real interaction module is built, which transmits formatted waveform-droplet parameters via Modbus TCP and shares real-time images / videos via FTP to realize bidirectional data interaction between the physical equipment layer and the digital twin layer; S3, a visualization module, builds a virtual scene of the nozzle within Unity 3D, and sets up a process parameter input area, an experimental feedback area, and a droplet flight animation display area, supporting multi-angle interaction; S4, Historical dataset construction: Starting from the initial state of the printhead, waveforms, voltages, ink cavity temperatures, ink path back pressures, droplet volumes and velocities are collected in a time-series manner to form a traceable dataset. S5, Training the basic prediction algorithm model for droplet state parameters: Using the historical data, the multilayer perceptron (MLP) network is trained offline to obtain the basic model for the droplet state parameter prediction algorithm. S6, Inverse waveform prediction algorithm model training: Based on the target droplet volume, velocity and tilt angle requirements, the inverse waveform prediction algorithm model is trained using historical data and jet mechanism knowledge, and a multi-objective optimization method is adopted. S7 integrates system functions, combining and debugging the basic model of droplet state parameter prediction algorithm, the reverse waveform prediction algorithm model, and the three-dimensional visualization interface in the digital twin system to form a virtual control system with a "prediction-visualization-control" closed loop.

[0052] This embodiment presents a digital twin-based virtual droplet closed-loop control system, which integrates processes such as historical dataset construction, basic model training, small-sample calibration, and waveform optimization. In practical applications, a debugging phase is involved before actual printing, which checks whether the algorithm model in the twin platform is suitable for the current printing task. If it is not suitable, such as... Figure 2 As shown, the debugging method may include the following steps: During the trial printing phase, a small sample dataset was collected, and the residuals of the basic model for predicting droplet state parameters were recalibrated using this small sample data to form a hybrid model for predicting droplet state parameters. Based on the droplet state parameter prediction hybrid model and the inverse waveform prediction model algorithm, the desired droplet parameters are input to generate multiple sets of optimized waveform parameters, which assists process engineers in adjusting the nozzle waveform parameters. If the actual spraying deviation is large, the new sample data is automatically fed back to the residual calibration module for further iterative optimization.

[0053] This debugging method reflects that the control system in this embodiment can not only achieve high-fidelity digital twin mapping of the nozzle process, but also realize closed-loop virtual debugging function.

[0054] The relevant technical solutions are the same as above, and will not be repeated here.

[0055] In summary, this invention involves mapping real-world data collected from a physical printhead module, such as waveform parameters, droplet volume, droplet velocity, and droplet tilt angle, to a digital twin platform under the same time base and isomorphic state. The platform integrates a hybrid prediction model for droplet state parameters and an inverse waveform prediction algorithm, constructing a virtual scene of the printhead module based on a visualization module. The prediction model and algorithm are trained offline using historical datasets, enabling virtual closed-loop control and testing of the droplets. Considering the time-varying nature of the printhead, during the trial printing phase, small sample data is used to online correct the droplet state parameter prediction model to ensure its robustness under the current printhead state. In actual printing, this invention effectively improves the visualization and monitoring capabilities of the manufacturing process, balancing prediction accuracy and real-time performance, and significantly improving the efficiency of process debugging.

[0056] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A digital-twin-based virtual droplet closed-loop regulation method, characterized in that, The method comprises: real-time acquisition of current working condition parameters and uploading to a digital twin platform; optimization of waveform parameters based on the current working condition parameters through the digital twin platform and feedback to the physical printhead module to realize real-time control of the physical printhead; wherein the digital twin platform is configured with a high-fidelity virtual equipment model of the physical printhead module in a virtual space, which comprises a droplet state parameter hybrid prediction model and a reverse waveform prediction model, the droplet state parameter hybrid prediction model being used to predict droplet state information based on multi-dimensional working condition parameters including printhead waveform parameters, ink path back pressure and ink cavity temperature, and the reverse waveform prediction model being used to infer printhead waveform parameters meeting the requirements of droplet ejection as optimized waveform parameters based on the predicted droplet state information and target droplet state parameters set by a user.

2. The method of claim 1, wherein, The digital twin platform is also configured with a three-dimensional visualization interface for multi-angle dynamic interaction and visualization of droplet states, wherein the multi-angle dynamic interaction includes input of process parameters and feedback of reverse inference results.

3. The method of claim 1, wherein, The droplet state parameter hybrid prediction model is constructed in the following manner: training a neural network structure based on historical working condition data to obtain a droplet state parameter basic prediction model; collecting a small sample working condition data set through test printing of the physical printhead module, and using the small sample working condition data set to calibrate the residual error of the droplet state parameter basic prediction model to form a droplet state parameter hybrid prediction model; wherein the residual error calibration is performed in the following manner: Using a droplet state parameter-based prediction model for a small sample dataset of operating conditions Make predictions and calculate the prediction residuals: In the formula, To predict residuals, For small sample working condition datasets The measured droplet state information in the data. For the basic prediction model of droplet state parameters The predicted output is the droplet state information. This indicates multi-dimensional operating parameters including waveform parameters, ink path back pressure, and ink cavity temperature under the current operating conditions. K Represents a small sample working condition dataset The number of samples; With As a supervision sample, a residual calibration model is constructed by training; linearly combining the residual error calibration model and the droplet state parameter basic prediction model to form the droplet state parameter hybrid prediction model.

4. The method of claim 3, wherein, The network structures of the droplet state parameter basic prediction model and the residual error calibration model are both MLP networks.

5. The method of claim 1, wherein, The reverse waveform prediction model is trained in a multi-objective optimization manner based on historical working condition data.

6. The method of claim 1, wherein, The high-fidelity virtual equipment model is configured in the following manner: geometric mapping: aligning the physical coordinates and virtual coordinates of the physical printhead module through a rigid-scale change alignment formula to realize geometric mapping; wherein the rigid-scale change alignment formula is: wherein represents a positioning anchor point in the virtual equipment model, i.e. a display coordinate determined by the virtual equipment model, represents a same-name point coordinate of the entity nozzle model corresponding to the virtual equipment model, represents a global scale factor, , R represents a coordinate rotation change, is a three-dimensional rotation matrix, t represents a coordinate translation change, is a translation vector; behavior mapping: iteratively predicting and correcting the isomorphic state vector of the physical printhead module, and synchronously mapping the behavior of the physical printhead module; wherein the prediction and correction method is: wherein is the homogeneous state vector of the physical injector model at the discrete control period is the homogeneous state vector of the physical injector model at the discrete control period is the homogeneous state vector of the physical injector model at the discrete control period is the homogeneous state vector of the physical injector model at the discrete control period is the prior prediction model giving the state of the next control period under the same time base is the observation correction model considering the deviation between the prior prediction model and the actual observation and mapping and correcting the homogeneous state of the physical injector model represents the droplet state information actually observed represents the droplet state information calculated by the prior prediction model 7. A virtual droplet closed-loop regulation system based on digital twinning, characterized in that, a virtual droplet closed-loop control method as claimed in any one of claims 1 to 6, comprising a printhead module, a virtual-real interaction module, and a digital twin platform; the printhead module is used for hardware entity operation and control of droplet ejection function, and real-time acquisition of working condition parameters; the virtual-real interaction module is used for obtaining working condition parameters and converting the data into a unified format before uploading to the digital twin platform; The digital twin platform is used to optimize the waveform parameters based on the current working condition parameters and feed them back to the physical nozzle module to realize real-time control of the physical nozzle; wherein the digital twin platform is configured with a high-fidelity virtual equipment model of the physical nozzle module in a virtual space, which includes a droplet state parameter hybrid prediction model and a reverse waveform prediction model, the droplet state parameter hybrid prediction model is used to predict the jet droplet state information based on the multi-dimensional working condition parameters including the nozzle waveform parameters, the ink path back pressure and the ink cavity temperature; the reverse waveform prediction model is used to infer the nozzle waveform parameters meeting the jet requirements as the optimized waveform parameters based on the predicted jet droplet state information and the target droplet state parameters set by the user.

8. The method of claim 7, wherein the method further comprises: determining the number of virtual droplets in the virtual droplet closed-loop control system; and determining the number of virtual droplets in the virtual droplet closed-loop control system based on the number of virtual droplets in the virtual droplet closed-loop control system. Comprise: By conducting test printing, the system verifies whether the mixed prediction model of droplet state parameters in the digital twin platform is distorted. If not, the debugging process ends. If so, a small sample dataset of operating conditions is collected from the test printing, and the basic prediction model of droplet state parameters is used to analyze the small sample dataset. Make predictions and calculate the prediction residuals: In the formula, To predict residuals, For small sample working condition datasets The measured droplet state information in the data. For the basic prediction model of droplet state parameters The predicted output is the droplet state information. This indicates multi-dimensional operating parameters including waveform parameters, ink path back pressure, and ink cavity temperature under the current operating conditions. K Represents a small sample working condition dataset The number of samples; With As a supervision sample, a residual calibration model is constructed by training; Linearly combine the residual calibration model with the droplet state parameter basic prediction model to form a new droplet state parameter hybrid prediction model, and complete system debugging.

Images