Digital printer control system based on artificial intelligence and control method thereof

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a digital printer control system based on artificial intelligence and a control method thereof. The system comprises a data collection simulation unit used for collecting historical printing data, generating ink diffusion simulation data and actual ink diffusion images under different conditions, and establishing a defect database at the same time; the prediction model building unit is used for building a space-time prediction model of ink diffusion, training the space-time prediction model by using ink diffusion simulation data, and finally generating a total loss function; and the feedback optimization unit is used for monitoring the change trend of the total loss function in real time. According to the digital printing machine control system based on artificial intelligence and the control method thereof, the prediction precision and the control effect of printing quality can be improved, fine ink diffusion differences caused by paper adsorbability changes can be effectively captured, and therefore accurate prediction of various defects possibly occurring in the printing process is achieved; the printing efficiency and the product quality are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based digital printing machine control system and a control method thereof. Background Art

[0002] In the field of modern digital printing, as market demands become increasingly diversified and personalized, the requirements for printing quality are also getting higher and higher. Customers not only expect to obtain prints with bright colors and clear details, but also require the ability to quickly respond to small batches and customized needs. However, existing printing control systems face many challenges in handling these complex and changeable printing tasks. In order to meet these high standards, every link in the printing process needs to be precisely controlled, especially the interaction between ink and paper. There are many types of paper, from ordinary copy paper to high-quality art paper, each with its own unique physical and chemical properties, especially significant differences in adsorption. This means that the same ink may exhibit completely different diffusion behaviors on different types of paper, which in turn affects the final printing effect. However, traditional printing control systems are usually based on fixed parameter settings and do not fully take these variables into account, making it difficult to achieve ideal printing quality in actual operation.

[0003] Because each paper type has its own unique adsorption characteristics, and the existing printing control systems often adopt a "one-size-fits-all" approach, fail to make personalized adjustments for specific papers, and lack understanding of the ink diffusion process, resulting in frequent manual corrections, reducing production efficiency, and also affecting product stability. Therefore, an artificial intelligence-based digital printing machine control system and a control method thereof are provided. Summary of the invention

[0004] The purpose of the present invention is to provide an artificial intelligence-based digital printing machine control system and a control method thereof, so as to solve the problem that the existing printing control systems proposed in the above background technology often adopt a "one-size-fits-all" approach, fail to make personalized adjustments for specific papers, and have insufficient understanding of the ink diffusion process, resulting in frequent manual corrections and reduced production efficiency.

[0005] To achieve the above objectives, on the one hand, the present invention aims to provide a digital printing machine control system based on artificial intelligence, comprising: A data collection simulation unit, the data collection simulation unit is used to collect historical printing data, generate ink diffusion simulation data and actual ink diffusion images under different conditions, and at the same time, mark defects in the actual ink diffusion images to establish a defect database; A prediction model building unit, wherein the prediction model building unit is used to combine the fluid mechanics equation with the generative adversarial network to build a spatiotemporal prediction model of ink diffusion, and train the spatiotemporal prediction model using ink diffusion simulation data to finally generate a total loss function; The feedback optimization unit is used to monitor the changing trend of the total loss function in real time and dynamically adjust the parameters of the generator and the discriminator according to the evaluation result of the total loss function.

[0006] As a further improvement of the technical solution, the historical printing data includes ink diffusion data, ink physical properties and paper property data.

[0007] As a further improvement of the technical solution, the prediction model building unit includes a physical neural operator module, a generative adversarial network module and a defect detection module; The physical neural operator module builds a spatiotemporal prediction model of ink diffusion based on the multi-layer perceptron, and embeds the fluid mechanics equation into the multi-layer perceptron to simulate the diffusion process of ink under different conditions, thereby obtaining the key prediction value of the ink diffusion trend, and transmitting the key prediction value to the generative adversarial network module; The generative adversarial network module takes the key prediction values ​​and historical printing data as input based on the generator, and outputs the predicted image of ink diffusion. The discriminator is then used to evaluate the predicted image of ink diffusion and compare it with the actual ink diffusion image, and the error is fed back to the generator. Finally, the parameters of the generator and discriminator are adjusted by generating a total loss function. The defect detection module is used to extract the predicted image of ink diffusion output by the generator, determine whether the predicted image of ink diffusion has known types of defects based on the defect database, use the convolutional neural network to classify the defect features, and feed the classified defect features back to the generator to adjust the parameters. At the same time, the classified defect features are fed back to the discriminator to evaluate whether the image has defects.

[0008] As a further improvement of the technical solution, the key prediction values ​​include velocity field, pressure field and concentration field.

[0009] As a further improvement of the technical solution, the specific steps of the physical neural operator module to obtain the key prediction value of the ink diffusion trend are as follows: S211. Embed the fluid mechanics equation into the multi-layer perceptron. In the physical neural operator, the fluid mechanics equation is the constraint condition of the multi-layer perceptron and is embedded through the loss function: ; In the formula, is the loss function; is the velocity field of the fluid; is the pressure field of the fluid; is the fluid density; is the kinematic viscosity of the fluid; is the external force term; For time; is the rate of change of the fluid velocity field over time; is the divergence of the velocity field; S212, Approximating the velocity field of the fluid by multi-layer perceptron and pressure field , simulate the diffusion behavior of ink under different conditions; S213. Combining the predicted value with the actual observed data, constructing the fluid loss function: ; In the formula, is the fluid loss function; is the regularization parameter; It is the real ink speed data; It is the real ink pressure data; is the output of the velocity field of the fluid approximated by the multi-layer perceptron; is the output of the pressure field of the fluid approximated by a multi-layer perceptron; is the spatial position vector; are the parameters of the multilayer perceptron; S214. Physical constraints on ink diffusion process: ; In the formula, is the ink concentration field; is the diffusion coefficient; S215, using multi-layer perceptron to approximate ink concentration field ; S216. Combining physical constraints with actual observation data through ink diffusion loss function: ; In the formula, is the ink diffusion loss function; It is the real ink concentration data; is the output of the concentration field of the fluid approximated by a multi-layer perceptron.

[0010] As a further improvement of the technical solution, in the ink diffusion loss function, when the ink is applied to the paper, it is absorbed into the pores of the paper by capillary action. The adsorption of the paper and the change in the viscosity of the ink lead to uneven distribution of the ink. Therefore, in the process of predicting the ink trend, the influence of the adsorption between the ink and the paper is considered, and then the ink diffusion loss function is optimized, as follows: Obtain the adsorption coefficient based on ink diffusion data, ink physical properties, and paper properties data: ; In the formula, is the adsorption coefficient; is the basic adsorption rate constant; is the density of active sites on the paper surface; is the paper surface roughness factor; The water absorption of paper; is the ink viscosity; Temperature dependence factor; Physical constraints on the ink diffusion process after considering the influence of adsorption between ink and paper: ; in, It is the concentration loss of ink due to paper adsorption; Use multi-layer perceptron and introduce adsorption coefficient to approximate ink concentration field ,generate ; Finally, the physical constraints are combined with the actual observation data through the ink diffusion loss function: ; in, is the optimized ink diffusion loss function; is the ink concentration field after the adsorption coefficient is introduced; It is the output of the concentration field of the approximated fluid after considering the adsorption coefficient through the multi-layer perceptron.

[0011] As a further improvement of the technical solution, the specific steps of the generative adversarial network module adjusting the parameters of the generator and the discriminator by generating a total loss function are as follows: S221, inputting the key prediction value provided by the physical neural operator module and the historical printing data provided by the data collection simulation unit into the generator, and outputting the predicted image of ink diffusion; S222, inputting the predicted image of ink diffusion and the actual ink diffusion image provided by the data collection simulation unit into the discriminator, and outputting a physical consistency score; S223. A total loss function is established based on the physical consistency score and the defect features transmitted from the defect detection module. The total loss function includes a fluid loss function, an adversarial loss function, a defect loss function, and a diffusion loss function.

[0012] As a further improvement of the technical solution, the specific steps of S222 are as follows: Output the image authenticity probability through the convolutional network: ; in, Score authenticity; is the input image; is the convolutional neural network used in the discriminator Parameter set; is the Sigmoid function; It is a convolutional neural network; Extract velocity field features, pressure field features, and concentration field features from the image, compare them with the predicted values ​​of the physical neural operator module, and calculate the comprehensive physical residual: ; ; ; ; in, To extract velocity field features from images using convolutional networks; is the velocity field residual; To extract pressure field features in images using convolutional networks; is the pressure field residual; To extract concentration field features in images using convolutional networks; is the concentration field residual; is the comprehensive physical residual; is the weight of the velocity field residual; is the weight of the pressure field residual; is the weight of the concentration field residual; Convert the integrated residuals to physically consistent probabilities: ; in, score for physical consistency; Weighted Overall Authenticity and Physics Rating: ; in, For comprehensive rating; the weights used to score authenticity; The weight with which to score the physical consistency.

[0013] As a further improvement of the technical solution, the specific steps of S223 are as follows: The adversarial loss function is: ; in, is the actual ink diffusion image; For all real images Expected value; For the real input image The results of the rating; is the predicted image of ink diffusion; To generate all images Expected value; The generated image for the discriminator to input The results of the rating; The total loss function is: ; in, is the total loss function; is the hyperparameter of the diffusion loss function; is the hyperparameter of the fluid loss function; is the hyperparameter of the defect loss function; is the defect loss function; for .

[0014] On the other hand, the present invention provides a digital printing machine control method based on artificial intelligence, which is used in the above-mentioned digital printing machine control system based on artificial intelligence, and comprises the following steps: S1. Collect historical printing data, generate ink diffusion simulation data and actual ink diffusion images under different conditions, and mark defects in the actual ink diffusion images to establish a defect database; S2, the physical neural operator module embeds the fluid mechanics equation into the multi-layer perceptron, simulates the diffusion process of ink under different conditions, and constructs a spatiotemporal prediction model of ink diffusion, thereby obtaining the key prediction value of the ink diffusion trend, and transmitting the key prediction value to the generative adversarial network module; S3, the generative adversarial network module uses the key prediction values ​​provided by the generator based on the physical neural operator module and the historical printing data as the input of the spatiotemporal prediction model of ink diffusion and outputs the predicted image of ink diffusion; S4, the generative adversarial network module uses the discriminator to evaluate the predicted image of ink diffusion and compares it with the actual ink diffusion image to generate a physical consistency score, and feeds the physical consistency score back to the generator; S5, the defect detection module recognizes and analyzes the predicted image of ink diffusion output by the generator, determines whether there are known types of defects according to the defect database, classifies the defect features using a convolutional neural network, and feeds the classified defect features back to the generator and the discriminator; S6. Finally, construct the total loss function, monitor the changing trend of the total loss function in real time, and dynamically adjust the parameters of the generator and discriminator according to the evaluation results of the total loss function.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. In the artificial intelligence-based digital printing machine control system and its control method, the adsorption effect between ink and paper is taken into account in the process of predicting ink trends, and the ink diffusion loss function is optimized accordingly, which can improve the prediction accuracy and control effect of printing quality, and can also effectively capture the subtle ink diffusion differences caused by changes in paper adsorption, thereby achieving accurate prediction and prevention of various defects that may occur in the printing process, and improving printing efficiency and product quality.

[0016] 2. In the artificial intelligence-based digital printing machine control system and control method thereof, the defect detection module feeds back the classified defect features to the generator to adjust the parameters. At the same time, the classified defect features are fed back to the discriminator to evaluate whether the image has defects. This can significantly reduce the defects in the generated image, improve the quality and consistency of the printed products, and enhance the automation level and adaptability of the system, so that the system can identify potential problems at an early stage and make automatic adjustments, thereby greatly improving production efficiency and the qualified rate of finished products, and ensuring the high accuracy and stability of the final printing effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is the overall flow chart of the present invention; Figure 2 is a flow chart of the overall method of the present invention; The meaning of each number in the figure is: 1. Data collection simulation unit; 2. Prediction model building unit; 21. Physical neural operator module; 22. Generative adversarial network module; 23. Defect detection module; 3. Feedback optimization unit. DETAILED DESCRIPTION

[0018] 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 only 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.

[0019] Example 1: Please refer to Figure 1 As shown, a digital printing machine control system based on artificial intelligence is provided, including a data collection simulation unit 1, a prediction model building unit 2 and a feedback optimization unit 3; The data collection simulation unit 1 is used to collect historical printing data, generate ink diffusion simulation data and actual ink diffusion images under different conditions, and mark defects in the actual ink diffusion images to establish a defect database. The historical printing data includes ink diffusion data, ink physical properties and paper property data; The prediction model building unit 2 is used to combine the fluid mechanics equation with the generative adversarial network to build a spatiotemporal prediction model of ink diffusion, and use the ink diffusion simulation data to train the spatiotemporal prediction model, and finally generate a total loss function; The prediction model building unit 2 includes a physical neural operator module 21, a generative adversarial network module 22 and a defect detection module 23; The physical neural operator module 21 constructs a spatiotemporal prediction model of ink diffusion based on a multi-layer perceptron, and embeds the fluid mechanics equation into the multi-layer perceptron to simulate the diffusion process of ink under different conditions, thereby obtaining key prediction values ​​of ink diffusion trend, and transmits the key prediction values ​​to the generative adversarial network module 22; wherein the fluid mechanics equation is the Navier-Stokes equation; the key prediction values ​​include velocity field, pressure field and concentration field; The specific steps of the physical neural operator module 21 to obtain the key prediction value of the ink diffusion trend are as follows: S211. Embed the fluid mechanics equation into the multi-layer perceptron. In the physical neural operator, the fluid mechanics equation is the constraint condition of the multi-layer perceptron and is embedded through the loss function: ; In the formula, is the loss function; is the velocity field of the fluid; is the pressure field of the fluid; is the fluid density; is the kinematic viscosity of the fluid; is the external force term; For time; is the rate of change of the fluid velocity field over time; is the divergence of the velocity field, which should be zero to satisfy the condition of incompressible fluid; this loss function measures whether the velocity and pressure fields output by the multilayer perceptron satisfy the Navier-Stokes equations.

[0020] S212, Approximating the velocity field of the fluid by multi-layer perceptron and pressure field , simulate the diffusion behavior of ink under different conditions; ; ; The above multi-layer perceptron and , respectively accepting spatial coordinates and time As input, and output the corresponding velocity and pressure prediction values; S213. Combining the predicted value with the actual observed data, constructing the fluid loss function: ; In the formula, is the fluid loss function; is the regularization parameter, balancing the weight between physical constraints and data fitting; It is the real ink speed data; It is the real ink pressure data; is the output of the velocity field of the fluid approximated by the multi-layer perceptron; is the output of the pressure field of the fluid approximated by a multi-layer perceptron; is a spatial position vector, usually a variable containing the spatial coordinates of the fluid location or ink diffusion area; are the parameters of the multilayer perceptron; by minimizing this comprehensive loss function, we can ensure that the model conforms to the laws of physics and is as close to the actual observed data as possible.

[0021] S214. Physical constraints on ink diffusion process: ; In the formula, is the ink concentration field; is the diffusion coefficient; S215, using multi-layer perceptron to approximate ink concentration field ; ; S216. Combining physical constraints with actual observation data through ink diffusion loss function: ; In the formula, is the ink diffusion loss function; It is the real ink concentration data; is the output of the concentration field of the fluid approximated by the multilayer perceptron; by minimizing this loss function, the multilayer perceptron can accurately simulate the diffusion behavior of the ink.

[0022] When ink is applied to paper, it is absorbed into the pores of the paper through capillary action. The adsorption of the paper and the change in the viscosity of the ink cause the ink to be unevenly distributed. Therefore, the influence of the adsorption between the ink and the paper is considered in the process of predicting the ink trend, and then the ink diffusion loss function is optimized, as follows: Obtain the adsorption coefficient based on ink diffusion data, ink physical properties, and paper properties data: ; In the formula, is the adsorption coefficient; is the basic adsorption rate constant; is the density of active sites on the paper surface; is the paper surface roughness factor; The water absorption of paper; is the ink viscosity; Temperature dependence factor; The surface active site density indicates the number of active sites on the paper surface that can be used to adsorb ink, which is related to the chemical composition of the paper (such as cellulose content) and the coating material; the surface roughness factor describes the smoothness of the paper surface. , rough surface The rough surface will increase the contact area between ink and paper, but may reduce the adsorption efficiency; water absorption refers to the ability of paper to absorb ink, which is related to the porosity, fiber structure and coating water absorption of the paper; the higher the viscosity of the ink, the slower the diffusion and adsorption rate; the ink density affects its fluidity and diffusion behavior; the temperature dependence factor describes the effect of temperature on the adsorption rate.

[0023] ; Apparent activation energy (J⋅mol -1 ), fitted by temperature swing adsorption experiment; is the ideal gas constant (8.314 J⋅mol -1 ⋅K -1 ); is the current temperature, is the reference temperature (298K in this embodiment).

[0024] Physical constraints on the ink diffusion process after considering the influence of adsorption between ink and paper: ; in, It is the concentration loss of ink due to paper adsorption; Use multi-layer perceptron and introduce adsorption coefficient to approximate ink concentration field ,generate ; ,Will As the input variable of the neural network, it is pre-determined for different papers and the value is dynamically input during the printing process.

[0025] Finally, the physical constraints are combined with the actual observation data through the ink diffusion loss function: ; in, is the optimized ink diffusion loss function; is the ink concentration field after the adsorption coefficient is introduced; The output of the concentration field of the approximated fluid after considering the adsorption coefficient through the multi-layer perceptron; Considering the influence of the adsorption between ink and paper in the process of predicting ink trends and optimizing the ink diffusion loss function accordingly can improve the prediction accuracy and control effect of printing quality. Specifically, by incorporating the adsorption coefficient into the model, the diffusion behavior of ink on different papers can be simulated more accurately. The coefficient combines the influence of key parameters such as the density of active sites on the paper surface, roughness factor, water absorption and ink viscosity, and takes into account the influence of temperature on the adsorption process. This enables the model to not only reflect physical reality, but also dynamically adapt to different printing conditions. Introducing a method based on a multi-layer perceptron to approximate the ink concentration field and constructing an optimized ink diffusion loss function in combination with physical constraints can minimize the difference between the predicted value and the actual observed data while ensuring that the numerical solution satisfies the basic physical laws. The biggest advantage of this method is that it can effectively capture the subtle ink diffusion differences caused by changes in paper adsorption, thereby achieving accurate prediction and prevention of various defects that may occur in the printing process (such as dot enlargement or uneven color), and ultimately achieving the purpose of improving printing efficiency and product quality. In addition, this strategy also supports personalized adjustments and optimization for specific paper types and environmental conditions, further enhancing its application value.

[0026] The generative adversarial network module 22 takes the key prediction value and the historical printing data as input based on the generator, thereby outputting the predicted image of ink diffusion, and then using the discriminator to evaluate the predicted image of ink diffusion and compare it with the actual ink diffusion image, and feeds back the error to the generator, and finally adjusts the parameters of the generator and the discriminator by generating a total loss function, so that the generated ink diffusion image is as close to the actual situation as possible; The specific steps of the generative adversarial network module 22 adjusting the parameters of the generator and the discriminator by generating a total loss function are as follows: S221, inputting the key prediction value provided by the physical neural operator module 21 and the historical printing data provided by the data collection simulation unit 1 into the generator, and outputting the predicted image of ink diffusion; S222, inputting the predicted image of ink diffusion and the actual ink diffusion image provided by the data collection simulation unit 1 into the discriminator, and outputting a physical consistency score; S223, based on the physical consistency score and the defect feature quantification image transmitted by the defect detection module 23, a total loss function is constructed, wherein the comprehensive loss function includes a fluid loss function, a confrontation loss function, a defect loss function, and a diffusion loss function; S222 specific steps are as follows: Output the image authenticity probability through the convolutional network: ; in, Score authenticity; is the input image, any image input to the discriminator, including both real images and predicted images; is the convolutional neural network used in the discriminator Parameter set; is the Sigmoid function; It is a convolutional neural network that extracts image features and Calculate a scalar value, which is passed through the Sigmoid function Converted into a probability between 0 and 1; this process aims to use deep learning technology to automatically identify and distinguish generated images from actual observed images, providing a basis for subsequent optimization. Specifically, in the context of digital printing control systems, this step helps improve the system's recognition accuracy of ink diffusion patterns, thereby more accurately adjusting printing parameters to match the expected effect.

[0027] The velocity field features, pressure field features and concentration field features are extracted from the image, compared with the predicted values ​​of the physical neural operator module 21, and the comprehensive physical residual is calculated: ; ; ; ; in, To extract velocity field features from images using convolutional networks; is the velocity field residual; To extract pressure field features in images using convolutional networks; is the pressure field residual; To extract concentration field features in images using convolutional networks; is the concentration field residual; is the comprehensive physical residual; is the weight of the velocity field residual; is the weight of the pressure field residual; is the weight of the concentration field residual. This ensures that the generated ink diffusion image not only looks realistic, but also strictly follows the basic principles of fluid mechanics. For digital printing, this means that even under complex or changing printing conditions, highly consistent and predictable ink diffusion behavior can be maintained, thereby improving printing quality and efficiency.

[0028] Convert the integrated residuals to physically consistent probabilities: ; in, A score is assigned for physical consistency; a higher score indicates that the image conforms more closely to the laws of physics. A simple mathematical transformation is used so that smaller residuals correspond to higher scores. By quantifying how closely an image conforms to the laws of physics, this transformation provides the system with a direct measure of the physical accuracy of the generated image. In digital printing applications, this helps to quickly screen out the ink diffusion solution that best matches physical properties, reducing trial and error costs and speeding up the production process.

[0029] Weighted Overall Authenticity and Physics Rating: ; in, For comprehensive rating; the weights used to score authenticity; The weight of the physical consistency score is integrated with the realism and physical accuracy of the image to ensure that the final output is both realistic and scientifically reasonable, and to achieve more refined operational control, which not only improves the quality of the finished product, but also enhances the system's ability to cope with diverse printing needs.

[0030] S223 The specific steps are as follows: The adversarial loss function is: ; in, is the actual ink diffusion image; For all real images The expected value here is actually the average of the results obtained after applying the discriminator to all real images. It reflects the ability of the discriminator to correctly identify real images; For the real input image The result of the scoring indicates the probability that the image is real. Ideally, for all , Should be close to 1; is the predicted image of ink diffusion; To generate all images The expected value here is actually the average of the results obtained after applying the discriminator D to all generated images. It reflects the ability of the discriminator to correctly identify generated images (non-real images); The generated image for the discriminator to input The result of the scoring indicates the probability that the image is real. Ideally, for all , Should be close to 0; The adversarial loss function aims to improve the quality of generated images by optimizing the discriminator and the generator. Specifically, it improves the performance of the generator by minimizing the gap between the generated images and the real images, while promoting this process by enhancing the discriminator's ability to distinguish between real and fake images. In this process, a reference of authenticity is provided, which is the goal that the generator strives to approach. Ultimately, this mutually adversarial learning mechanism promotes the progress of both parties, making the generated images closer and closer to the real situation.

[0031] The total loss function is: ; in, is the total loss function; is the hyperparameter of the diffusion loss function; is the hyperparameter of the fluid loss function; is the hyperparameter of the defect loss function; is the defect loss function; for .

[0032] The fluid loss function and diffusion loss function are used to ensure that the generated ink diffusion image complies with the basic physical laws, including the consistency of velocity field, pressure field and concentration field and the satisfaction of diffusion equation. This helps to generate more realistic simulation results; The adversarial loss function is used to measure the difference between the images generated by the generator and the real images, and at the same time train the discriminator's ability to distinguish between real and fake images. This can further improve the quality of the generated images, making them more difficult to identify as synthetic; The defect loss function analyzes the generated image through a convolutional neural network to identify and quantify known types of defects. By minimizing this loss function, the generated model can learn how to avoid these defects, thereby improving the overall quality of the generated image; The combined application of the above methods can significantly improve the quality and efficiency of digital printing, reduce various problems caused by ink diffusion, such as color distortion and blur, and also reduce the workload required for manual inspection and correction.

[0033] The defect detection module 23 is used to extract the predicted image of ink diffusion output by the generator, determine whether the predicted image of ink diffusion has known types of defects based on the defect database, use a convolutional neural network to classify defect features, and feed the classified defect features back to the generator to adjust the parameters. At the same time, the classified defect features are fed back to the discriminator to evaluate whether the image has defects.

[0034] This can not only significantly reduce defects in the generated images and improve the quality and consistency of printed products, but also enhance the system's automation level and adaptability, enabling the system to identify potential problems at an early stage and make automatic adjustments, thereby greatly improving production efficiency and finished product qualification rate, ensuring the final printing effect is highly accurate and stable.

[0035] The feedback optimization unit 3 is used to monitor the changing trend of the total loss function in real time, and dynamically adjust the parameters of the generator and the discriminator according to the evaluation results of the total loss function.

[0036] The learning rate and other hyperparameters of the generator and discriminator are dynamically adjusted according to the evaluation results of the comprehensive loss function to accelerate convergence and improve the stability and efficiency of model training. By real-time monitoring of the changing trend of the loss function, the feedback optimization unit can intelligently adjust the key parameters in the training process to ensure that the system can still quickly and accurately generate high-quality ink diffusion images when facing complex printing tasks, while reducing the risk of overfitting and improving the robustness and adaptability of the overall system. This can make the control system of digital printing machines based on artificial intelligence more intelligent and efficient, and meet diverse printing needs.

[0037] Example 2: Please refer to Figure 2 As shown, a digital printing machine control method based on artificial intelligence is provided, which is used for the above-mentioned digital printing machine control system based on artificial intelligence, and includes the following steps: S1. Collect historical printing data, generate ink diffusion simulation data and actual ink diffusion images under different conditions, and mark defects in the actual ink diffusion images to establish a defect database; S2, the physical neural operator module 21 embeds the fluid mechanics equation into the multi-layer perceptron, simulates the diffusion process of ink under different conditions, and constructs a spatiotemporal prediction model of ink diffusion, thereby obtaining the key prediction value of the ink diffusion trend, and transmits the key prediction value to the generative adversarial network module 22; S3, the generative adversarial network module 22 uses the key prediction values ​​provided by the generator based on the physical neural operator module 21 and the historical printing data as the input of the spatiotemporal prediction model of ink diffusion, and outputs the predicted image of ink diffusion; S4, the generative adversarial network module 22 uses the discriminator to evaluate the predicted image of ink diffusion and compares it with the actual ink diffusion image, generates a physical consistency score, and feeds the physical consistency score back to the generator; S5, the defect detection module 23 recognizes and analyzes the predicted image of ink diffusion output by the generator, determines whether there are known types of defects according to the defect database, classifies defect features using a convolutional neural network, and feeds the classified defect features back to the generator and the discriminator; S6. Finally, construct the total loss function, monitor the changing trend of the total loss function in real time, and dynamically adjust the parameters of the generator and discriminator according to the evaluation results of the total loss function.

[0038] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A digital printing machine control system based on artificial intelligence, characterized in that: include: A data collection simulation unit (1), the data collection simulation unit (1) is used to collect historical printing data, generate ink diffusion simulation data and actual ink diffusion images under different conditions, and at the same time, mark defects in the actual ink diffusion images to establish a defect database; A prediction model building unit (2), the prediction model building unit (2) is used to combine the fluid mechanics equation with the generative adversarial network to build a spatiotemporal prediction model for ink diffusion, and train the spatiotemporal prediction model using ink diffusion simulation data, and finally generate a total loss function; A feedback optimization unit (3), wherein the feedback optimization unit (3) is used to monitor the change trend of the total loss function in real time, and dynamically adjust the parameters of the generator and the discriminator according to the evaluation result of the total loss function.

2. The digital printing machine control system based on artificial intelligence according to claim 1 is characterized in that: The historical printing data includes ink diffusion data, ink physical properties and paper property data.

3. The digital printing machine control system based on artificial intelligence according to claim 2 is characterized in that: The prediction model building unit (2) includes a physical neural operator module (21), a generative adversarial network module (22) and a defect detection module (23); The physical neural operator module (21) constructs a spatiotemporal prediction model of ink diffusion based on a multi-layer perceptron, and embeds the fluid mechanics equation into the multi-layer perceptron to simulate the diffusion process of ink under different conditions, thereby obtaining key prediction values ​​of ink diffusion trends, and transmitting the key prediction values ​​to the generative adversarial network module (22); The generative adversarial network module (22) takes the key prediction value and the historical printing data as input based on the generator, thereby outputting the predicted image of ink diffusion, and then using the discriminator to evaluate the predicted image of ink diffusion and compare it with the actual ink diffusion image, and feeds back the error to the generator, and finally adjusts the parameters of the generator and the discriminator by generating a total loss function; The defect detection module (23) is used to extract the predicted image of ink diffusion output by the generator, determine whether the predicted image of ink diffusion has known types of defects based on the defect database, use a convolutional neural network to classify defect features, feed back the classified defect features to the generator, and then adjust the parameters. At the same time, feed back the classified defect features to the discriminator to evaluate whether the image has defects.

4. The digital printing machine control system based on artificial intelligence according to claim 3 is characterized in that: The key prediction values ​​include velocity field, pressure field and concentration field.

5. The artificial intelligence-based digital printing machine control system according to claim 4, characterized in that: The specific steps of the physical neural operator module (21) to obtain the key prediction value of the ink diffusion trend are as follows: S211. Embed the fluid mechanics equation into the multi-layer perceptron. In the physical neural operator, the fluid mechanics equation is the constraint condition of the multi-layer perceptron and is embedded through the loss function: ； In the formula, is the loss function; is the velocity field of the fluid; is the pressure field of the fluid; is the fluid density; is the kinematic viscosity of the fluid; is the external force term; For time; is the rate of change of the fluid velocity field over time; is the divergence of the velocity field; S212, Approximating the velocity field of the fluid by multi-layer perceptron and pressure field , simulate the diffusion behavior of ink under different conditions; S213. Combining the predicted value with the actual observed data, constructing the fluid loss function: ； In the formula, is the fluid loss function; is the regularization parameter; It is the real ink speed data; It is the real ink pressure data; is the output of the velocity field of the fluid approximated by the multi-layer perceptron; is the output of the pressure field of the fluid approximated by a multi-layer perceptron; is the spatial position vector; are the parameters of the multilayer perceptron; S214. Physical constraints on ink diffusion process: ； In the formula, is the ink concentration field; is the diffusion coefficient; S215, using multi-layer perceptron to approximate ink concentration field ; S216. Combining physical constraints with actual observation data through ink diffusion loss function: ； In the formula, is the ink diffusion loss function; It is the real ink concentration data; is the output of the concentration field of the fluid approximated by a multi-layer perceptron.

6. The artificial intelligence-based digital printing machine control system according to claim 5, characterized in that: In the ink diffusion loss function, when the ink is applied to the paper, it is absorbed into the pores of the paper by capillary action. The adsorption of the paper and the change in the viscosity of the ink lead to uneven distribution of the ink. Therefore, the influence of the adsorption between the ink and the paper is considered in the process of predicting the ink trend, and then the ink diffusion loss function is optimized, as follows: Obtain the adsorption coefficient based on ink diffusion data, ink physical properties, and paper properties data: ； In the formula, is the adsorption coefficient; is the basic adsorption rate constant; is the density of active sites on the paper surface; is the paper surface roughness factor; The water absorption of paper; is the ink viscosity; Temperature dependence factor; Physical constraints on the ink diffusion process after considering the influence of adsorption between ink and paper: ； in, It is the concentration loss of ink due to paper adsorption; Use multi-layer perceptron and introduce adsorption coefficient to approximate ink concentration field ,generate ; Finally, the physical constraints are combined with the actual observation data through the ink diffusion loss function: ； in, is the optimized ink diffusion loss function; is the ink concentration field after the adsorption coefficient is introduced; It is the output of the concentration field of the approximated fluid after considering the adsorption coefficient through the multi-layer perceptron.

7. The artificial intelligence-based digital printing machine control system according to claim 6, characterized in that: The specific steps of the generative adversarial network module (22) adjusting the parameters of the generator and the discriminator by generating a total loss function are as follows: S221, inputting the key prediction value provided by the physical neural operator module (21) and the historical printing data provided by the data collection simulation unit (1) into the generator, and outputting a predicted image of ink diffusion; S222, inputting the predicted image of ink diffusion and the actual ink diffusion image provided by the data collection simulation unit (1) into the discriminator, and outputting a physical consistency score; S223, constructing a total loss function based on the physical consistency score and the defect features transmitted from the defect detection module (23), wherein the total loss function includes a fluid loss function, an adversarial loss function, a defect loss function and a diffusion loss function.

8. The artificial intelligence-based digital printing machine control system according to claim 7, characterized in that: The specific steps of S222 are as follows: Output the image authenticity probability through the convolutional network: ； in, Score authenticity; is the input image; is the convolutional neural network used in the discriminator Parameter set; is the Sigmoid function; It is a convolutional neural network; The velocity field features, pressure field features and concentration field features are extracted from the image, compared with the predicted values ​​of the physical neural operator module (21), and the comprehensive physical residual is calculated: ； ； ； ； in, To extract velocity field features from images using convolutional networks; is the velocity field residual; To extract pressure field features in images using convolutional networks; is the pressure field residual; To extract concentration field features in images using convolutional networks; is the concentration field residual; is the comprehensive physical residual; is the weight of the velocity field residual; is the weight of the pressure field residual; is the weight of the concentration field residual; Convert the integrated residuals to physically consistent probabilities: ； in, score for physical consistency; Weighted Overall Authenticity and Physics Rating: ； in, For comprehensive rating; the weights used to score authenticity; The weight with which to score the physical consistency.

9. The artificial intelligence-based digital printing machine control system according to claim 8, characterized in that: The specific steps of S223 are as follows: The adversarial loss function is: ； in, is the actual ink diffusion image; For all real images Expected value; For the real input image The results of the rating; is the predicted image of ink diffusion; To generate all images Expected value; The generated image for the discriminator to input The results of the rating; The total loss function is: ； in, is the total loss function; is the hyperparameter of the diffusion loss function; is the hyperparameter of the fluid loss function; is the hyperparameter of the defect loss function; is the defect loss function; for .

10. A digital printing press control method based on artificial intelligence, used in the digital printing press control system based on artificial intelligence as claimed in any one of claims 1 to 9, characterized in that: The steps include: S1. Collect historical printing data, generate ink diffusion simulation data and actual ink diffusion images under different conditions, and mark defects in the actual ink diffusion images to establish a defect database; S2, the physical neural operator module (21) embeds the fluid mechanics equation into the multi-layer perceptron, simulates the diffusion process of ink under different conditions, and constructs a spatiotemporal prediction model of ink diffusion, thereby obtaining the key prediction value of the ink diffusion trend, and transmits the key prediction value to the generative adversarial network module (22); S3, the generative adversarial network module (22) uses the key prediction values ​​provided by the generator based on the physical neural operator module (21) and the historical printing data as the input of the spatiotemporal prediction model of ink diffusion, and outputs a predicted image of ink diffusion; S4, the generative adversarial network module (22) further utilizes the discriminator to evaluate the predicted image of ink diffusion and compares it with the actual ink diffusion image, generates a physical consistency score, and feeds the physical consistency score back to the generator; S5, the defect detection module (23) recognizes and analyzes the predicted image of ink diffusion output by the generator, determines whether there are known types of defects according to the defect database, classifies defect features using a convolutional neural network, and feeds the classified defect features back to the generator and the discriminator; S6. Finally, construct the total loss function, monitor the changing trend of the total loss function in real time, and dynamically adjust the parameters of the generator and discriminator according to the evaluation results of the total loss function.

Images