Online detection system and method based on ink jet printing

Abstract

The invention discloses an online detection system and method based on ink-jet printing, and relates to the technical field of ink-jet printing, and the method comprises the steps: collecting the operation state data of ink-jet printing in real time, and transmitting the data to a deep learning model; predicting the track, drop point and shape change of the ink droplet through a deep learning model in combination with a fluid dynamics principle, generating a prediction result, and providing positioning information of an ink jet area; inputting the multispectral data into a convolutional neural network model, identifying color deviation and defect types in combination with a prediction result, and generating corresponding correction parameters; the correction parameters are transmitted to ink jet control equipment, and the jet parameters of the nozzle are adjusted in real time; according to the invention, the information of the boundary, the central point, the area, the coverage rate and the offset of the ink jet area is generated in real time, and a high-precision positioning basis is provided for subsequent multispectral detection and defect correction, so that efficient and accurate online detection and ink jet quality control are realized.

Description

Technical Field

[0001] The invention relates to the technical field of inkjet printing, in particular to an online detection system and method based on inkjet printing. Background Art

[0002] As a non-contact digital printing method, inkjet printing technology has been widely used in recent years. It has the advantages of high precision, strong flexibility, and applicability to a variety of substrates. It is widely used in industrial manufacturing, electronic device production, and high-resolution image output. Traditional inkjet printing technology relies on nozzles to accurately spray ink droplets onto the surface of the target substrate, and achieves high-quality pattern formation by controlling the spray angle, spray speed, and droplet size. However, with the continuous expansion of application scenarios and the continuous improvement of requirements for printing accuracy and stability, inkjet printing has put forward higher standards and challenges for droplet trajectory control, landing accuracy, and print quality detection. In recent years, in order to further improve the quality of inkjet printing, researchers have proposed some technical solutions based on online detection and feedback control, which improve inkjet accuracy and printing stability by real-time monitoring of droplet status and jetting behavior. However, the online detection methods of the existing technology are mostly limited to the monitoring of a single parameter, such as the droplet landing position or jetting speed, and lack a comprehensive evaluation of the dynamic behavior of the droplet and its final imaging effect.

[0003] Existing inkjet printing technology still has many shortcomings in real-time detection and feedback control. First, the jetting behavior of ink droplets is affected by the nozzle parameters (such as jetting pressure and jetting frequency), environmental conditions (such as temperature and humidity) and physical properties of ink droplets (such as viscosity and surface tension). Its motion trajectory and final landing point show highly nonlinear and dynamic change characteristics. Existing detection schemes often cannot fully capture the complex dynamic behavior of ink droplets during motion, and it is difficult to accurately predict the trajectory, landing point and shape change of ink droplets. Secondly, existing online detection methods usually rely only on a single optical or visual means, which cannot fully reflect the color deviation or defect type of the inkjet area, resulting in one-sided inkjet quality assessment. In addition, the lack of multi-spectral image acquisition and analysis capabilities in existing technologies also limits the comprehensive detection and accurate analysis of the color distribution, texture characteristics and spectral characteristics of the inkjet area. Especially under the requirements of high-resolution and high-precision inkjet, traditional detection technology is difficult to achieve real-time correction of the inkjet area, resulting in instability of inkjet quality. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] In a first aspect, the present application provides an online detection method based on inkjet printing, comprising:

[0006] Acquire operation status data during the inkjet printing process, and combine the operation status data to generate a time series data structure, wherein the operation status data includes nozzle ejection parameters and initial behavior data of ink droplets;

[0007] The time series data structure is transmitted to a deep learning model, and the time series data structure is preprocessed by the deep learning model to obtain a motion trajectory r of the ink droplet at the current time t, output a prediction result, and store the positioning information of the inkjet area;

[0008] According to the positioning information of the inkjet area, multispectral image acquisition is performed on the inkjet area to obtain multispectral data, wherein the multispectral data includes color distribution, texture and spectral characteristic data;

[0009] The multispectral data is input into the convolutional neural network model, and the color deviation and defect type are identified by combining the prediction results, and the corresponding correction parameters are generated;

[0010] The correction parameters are transferred to the inkjet control device to adjust the nozzle ejection parameters in real time.

[0011] Preferably, the performing data preprocessing on the time series data structure by the deep learning model comprises:

[0012] Decode the received time series data to obtain the feature vector of the ink drop;

[0013] Performing multi-modal feature fusion on the feature vector to obtain a fusion feature F of the ink drop;

[0014] The fusion feature F is used as the adjustment factor to calculate the dynamic adjustment fractional order α(t ′ ), its calculation expression is:

[0015]

[0016] Among them, t ′ represents the time parameter for dynamically adjusting the fractional order α, σ represents the Sigmoid activation function, represents the linear combination of the fused feature g, and b represents the bias term.

[0017] Preferably, the dynamically adjusted fractional order α(t ′ ), and also includes:

[0018] Introducing the time-varying order, the dynamic adjustment fractional order α(t ′ ) performs adaptive fractional derivative operation to capture the non-stationary and dynamic characteristics of ink droplet motion, and the expression is:

[0019]

[0020] in, represents the adaptive fractional derivative of the dynamic variable function f(t) of the ink drop, f(t) represents the dynamic variable function of the ink drop with the current time t, Γ represents the gamma function, 1-α(t ′ ) represents the input of the gamma function, dt represents the derivative of the current time t, t w is the window width at the current time t, tt w represents the integral operation, τ represents the past time, t-τ represents the time difference between the current time t and the past time τ, f(τ) is the function value of τ at a certain time in the past, dτ represents a small change in the integral variable, and d represents the differential operator;

[0021] Based on the adaptive fractional derivative, the motion trajectory r(t) of the ink droplet under gravity is calculated, and its calculation expression is:

[0022]

[0023] Among them, a(t ′ ) represents acceleration, t ′ Represents the time interval from the initial moment to the current moment. represents the viscous resistance, k is the damping coefficient, m is the mass of the ink drop, v(t ′ ) represents the ink drop at time t ′ The instantaneous speed, represents the external force acting on the ink droplet, represents the resistance of air to the high-speed moving ink droplets, β represents the air resistance coefficient, v(t) represents the instantaneous speed of the ink droplet at time t, v0 represents the initial speed of the ink droplet, represents the change in speed, represents the change in position, r(t) represents the motion trajectory of the ink droplet at the current time t, and r0 represents the initial position of the ink droplet.

[0024] Preferably, after obtaining the motion trajectory r(t) of the ink droplet under gravity and before outputting the prediction result, the method further includes:

[0025] According to the motion trajectory r(t) of the ink droplet at the current time t, combined with the jetting speed, gravity acceleration and jetting angle, the position of the ink droplet landing point is predicted;

[0026] The center point and distribution range of the ink droplet are calculated based on the location of the ink droplet and the motion trajectory r(t) of the ink droplet at the current time t.

[0027] Record the shape change of the contact surface after the ink droplet lands, and calculate the spreading range of each ink droplet;

[0028] The spreading range of all ink droplets is integrated to define the boundary of the inkjet area and calculate the area of ​​the inkjet area;

[0029] Compare the overlapping area between the inkjet area and the target area, calculate the coverage, output the boundary, center point, area, coverage and offset of the inkjet area, and generate the inkjet area positioning information.

[0030] Preferably, the steps of collecting multispectral images of the inkjet area and acquiring multispectral data are as follows:

[0031] Load the positioning information of the inkjet area as the inkjet area range to be collected;

[0032] Align the field of view center of the multispectral camera to the center point of the inkjet area, calibrate the boundary of the inkjet area, and set the acquisition mode;

[0033] The inkjet area is scanned step by step to acquire multispectral image data.

[0034] Preferably, the multispectral data is input into the convolutional neural network model, the color deviation and defect type are identified in combination with the prediction results, and the corresponding correction parameters are generated. The specific steps are as follows:

[0035] Loading spectral image data into the data processing pipeline;

[0036] According to the inkjet area positioning information, the data of the large inkjet area is divided into small blocks, each of which corresponds to the data slice of the multi-spectral channel;

[0037] Process the multispectral data of each block and organize the processed multispectral data into an input tensor;

[0038] Add region meta information to each input tensor;

[0039] The input tensor is fed into a deep learning model based on a 3D convolutional neural network to extract the spectral-spatial joint features in the multispectral data and generate color deviation features and defect representations;

[0040] According to the color deviation characteristics and defect characterization, the color correction matrix is ​​calculated to dynamically adjust the color ratio and intensity when the ink droplets are ejected;

[0041] Based on the predicted droplet trajectory, landing point and shape change, the color correction matrix is ​​comprehensively analyzed with the boundary, center point, area, coverage and offset of the inkjet area to generate the inkjet correction parameters.

[0042] Preferably, the correction parameters are transmitted to the inkjet control device to adjust the nozzle jetting parameters in real time, and the specific steps are as follows:

[0043] The inkjet calibration parameters are standardized and packaged into data packets;

[0044] Using the message transmission protocol, the data packet is transmitted to the inkjet control device receiving end in real time and the correction parameters are decoded;

[0045] According to the decoded correction parameters, the injection pressure and injection frequency of the nozzle are dynamically adjusted;

[0046] Combined with the predicted drop point offset of ink drop motion, the jet angle is adjusted dynamically;

[0047] Dynamically adjust the color ratio of the nozzle according to the color correction matrix.

[0048] In a second aspect, the present invention provides an online detection system based on inkjet printing, comprising a data acquisition module, a motion prediction module, an image acquisition module, a defect recognition module and a real-time correction module;

[0049] The data acquisition module is used to collect the operating status data of inkjet printing in real time and transmit it to the deep learning model;

[0050] The motion prediction module is used to predict the trajectory, landing point and shape change of ink droplets by combining the deep learning model with the principle of fluid dynamics, generate prediction results and provide positioning information of the inkjet area;

[0051] The image acquisition module is used to acquire multispectral images of the inkjet area according to the positioning information of the inkjet area to obtain multispectral data;

[0052] The defect recognition module is used to input the multispectral data into the convolutional neural network model, identify the color deviation and defect type in combination with the prediction results, and generate corresponding correction parameters;

[0053] The real-time correction module is used to transmit the correction parameters to the inkjet control device to adjust the nozzle ejection parameters in real time.

[0054] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the online detection method based on inkjet printing as described in the first aspect of the present invention is implemented.

[0055] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the online detection method based on inkjet printing as described in the first aspect of the present invention is implemented.

[0056] The beneficial effects of the present invention are as follows: the present invention uses a deep learning model combined with the principles of fluid dynamics to accurately predict the trajectory, landing point and shape change of ink droplets, which can fully capture the dynamic characteristics and nonlinear behavior of ink droplet motion, and significantly improve the accuracy and reliability of inkjet area positioning; at the same time, the adaptive fractional-order derivative is introduced to model the complex dynamic behavior of ink droplet motion, and the calculation parameters can be dynamically adjusted to adapt to changes under different injection conditions, thereby optimizing the inkjet effect; in addition, by real-time generation of information on the boundaries, center points, areas, coverage and offsets of the inkjet area, a high-precision positioning basis is provided for subsequent multi-spectral detection and defect correction, thereby achieving efficient and accurate online detection and inkjet quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0058] Figure 1 This is a flow chart of the online detection method based on inkjet printing in Example 1.

[0059] Figure 2 This is a module diagram of the online detection system based on inkjet printing in Example 1. DETAILED DESCRIPTION

[0060] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0062] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides an online detection method based on inkjet printing, comprising the following steps:

[0063] S1. Collect the operating status data of inkjet printing in real time and transmit it to the deep learning model.

[0064] Furthermore, the operating status data includes nozzle ejection parameters, environmental data, and initial behavior data of ink droplets.

[0065] It should be noted that high-frequency real-time acquisition and analysis of operating status data can reflect the dynamic changes of the inkjet system, especially in complex environments, and can significantly improve the stability and efficiency of the inkjet system.

[0066] Specifically, high-precision sensors (such as pressure, flow, and temperature sensors) are installed near the nozzle to monitor the nozzle's injection parameters (injection pressure, flow rate, temperature, and other parameters);

[0067] Among them, the injection parameters collected by high-precision sensors can be used by deep learning models to dynamically adjust the working state of the nozzle, thereby reducing injection deviation and improving inkjet quality.

[0068] Deploy temperature and humidity sensors, air quality sensors, and vibration sensors to monitor the environmental data of the nozzles (ambient temperature, humidity, air cleanliness, and equipment vibration);

[0069] Record environmental status data in real time and integrate it with nozzle parameters synchronously;

[0070] Preferably, the synchronously integrated environmental data and nozzle parameters can provide instant feedback information for the inkjet system. For example, when the temperature and humidity change, the nozzle's injection pressure or temperature can be automatically adjusted by analyzing the environmental data; when abnormal vibration is detected, the injection frequency or injection angle can be adjusted to reduce the impact of vibration; this dynamic adjustment mechanism can significantly improve the adaptability of the inkjet system, enabling it to maintain high-quality printing effects under different environmental conditions.

[0071] It should also be noted that real-time monitoring and control of nozzle jetting parameters is the key to ensuring inkjet uniformity, accuracy and stability.

[0072] Install a high-speed industrial camera at the nozzle outlet to capture the shape, speed, direction and trajectory of ink droplets when they are generated, and use image processing algorithms to extract the initial behavior characteristics of ink droplets;

[0073] Furthermore, the collected nozzle jetting parameters, environmental data, and initial behavior data of ink droplets are packaged into a time series data structure to provide multimodal input for subsequent deep learning models;

[0074] Through the communication protocol, the time series data structure is transmitted to the server side of the deep learning model to ensure low latency and high integrity of the data, providing support for the rapid response of the deep learning model.

[0075] It should be noted that before the time series data structure is transmitted to the server side of the deep learning model, the deep learning model should be trained with a large amount of historical inkjet printing process data. The process data includes the nozzle injection parameters, ink droplet behavior data, motion trajectory and related environmental condition data. The supervised learning method is used to train the deep learning model through labeled training data (such as the relationship between the actual motion trajectory of the ink droplet and the inkjet parameters) to ensure that the deep learning model can accurately process the real-time input time series data structure and effectively predict the motion trajectory of the ink droplet.

[0076] The server decodes the received time series data and outputs the input feature vector, providing high-quality input for the training and reasoning of the deep learning model.

[0077] S2. Through deep learning models combined with the principles of fluid dynamics, the trajectory, landing point and shape change of ink droplets are predicted, the prediction results are generated and the positioning information of the inkjet area is provided.

[0078] Furthermore, the received time series data is decoded to obtain the feature vector of the ink drop;

[0079] Specifically, the decoding process is to process the received time series data and convert the original time series into a feature vector that can describe the behavior of ink droplets. In this process, the time series data is first preprocessed (such as denoising, standardization, etc.) to remove irrelevant noise and outliers, and then the ink droplet behavior data (such as position, speed, injection angle, etc.) at each time point is extracted to generate a set of representative values. Then, the dynamic changes of the data are converted into a set of stable eigenvalues ​​using sliding window technology or other feature extraction methods (such as Fourier transform, waveform analysis, etc.). Finally, by weighting and normalizing these eigenvalues, a high-dimensional feature vector is obtained. This feature vector contains the motion characteristics and behavior trends of the ink droplets in the entire time series, which can be used for subsequent deep learning model training and prediction.

[0080] Input the feature vector into the deep learning model, perform multimodal feature fusion, and output the fused feature F;

[0081] Among them, multimodal feature fusion refers to the joint modeling of multiple features from different sources (nozzle parameters, environmental data, ink droplet behavior, etc.) to generate fused features with stronger semantic expression capabilities.

[0082] The fusion feature F is used as the adjustment factor to dynamically adjust the fractional order α(t ′ ), the value range is usually 0<α(t ′ )<1, which determines the strength of the memory effect of the fractional derivative and is a time-dependent dynamic variable;

[0083] Dynamically adjusted fractional order α(t ′ ), the expression is:

[0084]

[0085] Among them, t ′ represents the time parameter for dynamically adjusting the fractional order α, σ represents the Sigmoid activation function, represents a linear combination of the input feature g, and b represents the bias term;

[0086] Preferably, the fusion features can integrate multimodal information such as nozzle parameters, environmental status and ink droplet behavior to fully display the operating status and dynamic characteristics; by driving the dynamic adjustment of the fractional order with the fusion features, accurate modeling of the nonlinear and non-stationary characteristics of the inkjet system can be achieved, so that the fractional order model can flexibly adapt to the changes in the complexity of ink droplet motion, thereby capturing the impact of small disturbances or changing environments on ink droplet trajectories; this method not only improves the prediction accuracy of ink droplet motion trajectories and landing point distribution, but also provides data support for real-time correction and high-precision control during the inkjet process, significantly improving the stability and adaptability of the inkjet system.

[0087] The time-varying order is introduced and the adaptive fractional derivative is defined to capture the non-stationary and dynamic characteristics of the ink droplet motion. The expression is:

[0088]

[0089] in, It represents the fractional derivative operation on the dynamic variable function f(t), and the order is determined by α(t ′ ), f(t) represents the dynamic variable function of the ink droplet at the current time t, Γ represents the gamma function, 1-α(t ′ ) represents the input of the gamma function, which depends on the fractional order α(t ′ ), dt represents the derivative of the current time t, which represents the instantaneous rate of change, t w is the window width at the current time t, indicating the historical range of the integration, tt w Indicates the integral operation, the interval is from tt w To the current time t, τ represents the past time, t-τ represents the time difference between the current time t and the past time τ, f(τ) is the function value at a certain time τ in the past, representing the historical state of the function, dτ represents a small change in the integral variable, and d represents the differential operator;

[0090] Preferably, the time-varying order is introduced and the adaptive fractional-order derivative is defined. Compared with the traditional integer-order or fixed fractional-order method, it can more flexibly capture the non-stationarity and dynamic characteristics of the ink droplet motion, especially in the inkjet process under complex environments or multiple forces. The traditional integer-order method is usually used to deal with linear and stable scenes, and it is difficult to accurately describe the nonlinear changes of the ink droplets under the combined action of damping, gravity, jet force, environmental interference and other factors during the jetting process.

[0091] For example, in inkjet printing, the trajectory of ink droplets may experience viscosity fluctuations due to changes in ambient temperature. Traditional methods cannot dynamically adjust model parameters, resulting in large deviations in trajectory prediction. Although the fixed fractional order method improves nonlinear modeling to a certain extent, its order cannot change dynamically over time, making it difficult to capture changes in transient characteristics.

[0092] In contrast, adaptive fractional derivatives introduce time-varying orders, which enable the orders to be adjusted in real time based on nozzle parameters, environmental data, and droplet characteristics.

[0093] For example, when the ink droplets are dominated by gravity and jet force, the time-varying order will be adjusted to better capture the stable motion law; when the environmental vibration or wind speed interference causes the ink droplet trajectory to deviate, the time-varying order can respond quickly to capture the non-stationary characteristics. This dynamic adjustment capability not only improves the accuracy of ink droplet trajectory prediction, but also can identify and correct possible deviations in real time to ensure high-precision matching between the inkjet area and the target area. Therefore, compared with traditional methods, adaptive fractional-order derivatives can more accurately describe the complex and dynamic physical behavior in the actual inkjet process, significantly enhance robustness and adaptability, and show obvious advantages, especially in high-precision inkjet printing and industrial applications in complex environments.

[0094] Based on the adaptive fractional derivative, the motion trajectory r(t) of the ink droplet under gravity is calculated, and its calculation expression is:

[0095]

[0096] Among them, a(t ′ ) represents acceleration, t ′ Represents the time interval from the initial moment to the current moment. represents the viscous resistance, k is the damping coefficient, m is the mass of the ink drop, v(t ′ ) represents the ink drop at time t ′ The instantaneous speed, represents the external force acting on the ink droplet, represents the resistance of air to the high-speed moving ink droplets, β represents the air resistance coefficient, v(t) represents the instantaneous speed of the ink droplet at time t, v0 represents the initial speed of the ink droplet, represents the change in speed, represents the change in position, r(t) represents the motion trajectory of the ink droplet at the current time t, and r0 represents the initial position of the ink droplet.

[0097] It should be noted that f(t) represents dynamic physical quantities such as the position, velocity or acceleration of the ink droplet;

[0098] Assume that f(t) represents the position function r(t) of the ink drop, that is: f(t) = r(t);

[0099] Among them, r(t) represents the spatial position of the ink droplet at time t (it may be in the form of three-dimensional coordinates, such as [x(t), y(t), z(t)]). In addition, f(t) may also represent the instantaneous velocity v(t) or acceleration a(t ′ ), depending on how the dynamic variable is defined in the model. Therefore, f(t) can be different physical quantities that describe the motion of the ink droplet, and the specific function form depends on the requirements of the method in practical applications.

[0100] For example, in different inkjet printing systems, the behavior and control requirements of ink droplets may be different. If the goal is to precisely control the position of the ink droplet, f(t) may represent the position function r(t) of the ink droplet, that is, the spatial coordinates of the ink droplet will change over time, similar to f(t) = r(t) = [x(t), y(t), z(t)]. At this point, the deep learning model may need to predict the trajectory and ejection results of the ink droplet by calculating the change in the position of the ink droplet for subsequent adjustments and corrections. If the goal is to adjust the ejection speed of the ink droplet in real time, then f(t) may choose a function v(t) representing the speed, that is, the speed of the ink droplet changes over time, similar to f(t) = v(t) = [vx(t), vy(t), vz(t)], which will help adjust the ejection parameters of the inkjet device according to the speed. If the actual application focuses more on the acceleration change of the ink droplet, such as in the case of rapid ejection, the acceleration of the ink droplet has a greater impact on the final effect, f(t) may also represent the acceleration function a(t ′ ), that is, the acceleration of the ink droplet changes with time, similar to f(t) = a(t ′ )=[ax(t),ay(t),az(t)]. Therefore, the specific function form of f(t) depends on the goals and control requirements in practical applications. Whether it is position, velocity or acceleration, the most appropriate description method will be selected according to the specific physical model, so as to make accurate dynamic adjustments in the process of deep learning and multispectral image processing.

[0101] In summary, f(t) is a dynamic physical quantity function used to describe the motion characteristics of the ink droplet, which may be the position, velocity or acceleration of the ink droplet, depending on the subsequent physical model and mathematical expression.

[0102] It should also be noted that the motion trajectory of the ink droplet is calculated by modeling the dynamic characteristics of the ink droplet (such as acceleration, velocity, etc.) using the fractional derivative method. This process focuses on the instantaneous motion and physical properties of the ink droplet, such as the effects of gravity, viscous resistance, etc. on the motion of the ink droplet. It is worth noting that the injection angle will also be used as one of the initial behavioral data of the ink droplet, and the initial velocity and direction are already implicit in the injection parameters. Therefore, in the motion trajectory calculation, the injection angle is not explicitly considered separately, because it has been indirectly reflected in the initial conditions of the ink droplet (such as initial velocity and direction).

[0103] Specifically, the motion acceleration of the ink droplet a(t ′ ) is determined by multiple factors, including nonlinear dynamic behavior, fluid resistance, external forces and other physical effects. By introducing adaptive fractional derivatives This term effectively captures the non-stationarity and history dependence of ink droplet motion. Fractional derivatives are an effective tool for dealing with memory effects and nonlinear phenomena in complex dynamic systems. They are especially widely used in modeling liquid motion and can truly reflect the complex mechanical effects on ink droplets during inkjet. ′ )|v(t ′ ) and -(β / m)|v(t ′ )|v(t ′ ) respectively simulates the viscous resistance and air resistance of ink droplets moving in the air, which conforms to the classic resistance law in fluid mechanics. The resistance is proportional to the square of the speed, and the deceleration effect on ink droplets is particularly significant under high-speed motion. At the same time, the external force term F ext / m introduces the influence of external factors such as gravity and wind on the motion of ink droplets, ensuring the comprehensiveness of the motion equation. The following speed formula Based on the classic Newton's law of motion, it describes the law of velocity change over time. By integrating the acceleration, the instantaneous velocity of the ink droplet at any time can be calculated, ensuring the dynamics and accuracy of the model. Next, the position formula The spatial position change of the ink droplet is further derived, which is the standard form of the relationship between position and velocity in classical physics. In summary, the model of acceleration, velocity and position is theoretically reasonable and consistent with the principles of classical mechanics. In particular, the introduction of fractional derivatives adds an accurate description of non-stationary dynamics and complex resistance to the motion of the ink droplet, enhances the applicability and accuracy of the model, and thus provides a more accurate prediction and correction basis for the inkjet control system.

[0104] It should be noted that by analyzing the dynamic behavior of ink droplets based on adaptive fractional-order derivatives and describing their motion trajectories in combination with the force conditions of the ink droplets, it is possible to accurately capture the complex nonlinear and non-stationary characteristics of ink droplets, dynamically adapt to environmental changes and injection conditions, and significantly improve the prediction accuracy of ink droplet trajectories and landing points. This process can accurately predict ink droplet trajectories and landing points, provide support for dynamically adjusting injection parameters, improving inkjet coverage accuracy, and reducing printing deviations, while effectively improving the overall performance and adaptability of inkjet printing.

[0105] According to the motion trajectory r(t) of the ink droplet at the current time t, combined with the jetting speed, gravity acceleration and jetting angle, the position of the ink droplet landing point is predicted;

[0106] Specifically, by collecting the nozzle's jet parameters (such as jet speed and jet angle) and the initial position of the ink droplet in real time, a motion model of the ink droplet during the jetting process is established; in the motion model, the effect of gravity acceleration on the vertical direction and the influence of air resistance in the horizontal and vertical directions are considered, and the motion equations of the ink droplet in the horizontal and vertical components are established respectively through Newton's second law; then, by solving the motion equation, the motion trajectory equation of the ink droplet is obtained, and combined with the fractional derivative tool, the non-stationary and complex dynamic characteristics in the ink droplet motion are captured to further improve the accuracy of trajectory prediction; then, by calculating the time point when the ink droplet reaches the target surface and substituting it into the trajectory equation, the landing point position of the ink droplet is accurately predicted; finally, combined with the spreading behavior of the ink droplet, the landing point distribution range and center point of the ink droplet are calculated, and the prediction results are corrected using the real-time collected environmental parameters (such as wind speed, temperature, etc.) to ensure the accuracy and robustness of the prediction.

[0107] The center point and distribution range of the ink droplet are calculated based on the location of the ink droplet and the motion trajectory r(t) of the ink droplet at the current time t.

[0108] Specifically, according to the predicted ink droplet motion trajectory r(t) and landing point position, the actual landing point coordinates of all ink droplets are recorded through multiple real-time monitoring to form a landing point data set;

[0109] The geometric center point of the set of landing points is calculated using statistical methods;

[0110] Among them, the geometric center point reflects the overall distribution center of the ink droplet landing point and is the core reference point of the inkjet area.

[0111] It should also be noted that in the ink droplet motion trajectory calculation step of the first part, the predicted ink droplet landing position is based on the known ink droplet motion trajectory and other physical parameters (such as gravity, injection speed, etc.) to infer the final landing point. Although the injection angle is not explicitly used in the motion trajectory calculation in the third part, the injection angle can still be used as an important factor to adjust and accurately calculate the final position of the ink droplet in the subsequent landing point prediction. In this step, the injection angle can actually be indirectly reflected by the aforementioned motion trajectory and initial behavior data (such as initial velocity and direction). Therefore, the landing point prediction calculation step in the second part focuses more on accurate landing point estimation based on the existing motion trajectory data.

[0112] Furthermore, in order to describe the discrete degree of ink droplet landing points, the standard deviation method is used to calculate the distribution range;

[0113] At the same time, the distribution range is modified according to the spreading behavior of the ink droplet after it contacts the surface (for example, the relationship between the spreading diameter and the ink droplet momentum, surface tension and contact angle).

[0114] Use regression analysis or machine learning algorithms (such as support vector regression SVR or random forest regression RF) to analyze the impact of environmental factors (such as wind speed, humidity, temperature, etc.) on the distribution range of ink droplets;

[0115] After the model is trained with historical data, the center point and distribution range of the landing point can be corrected in real time to further improve the accuracy of the calculation;

[0116] The calculated geometric center point and the corrected distribution range (standard deviation or maximum radius) are output as the core information of the inkjet area, providing a basis for the subsequent boundary delineation, area calculation and coverage evaluation of the inkjet area.

[0117] Record the shape change of the contact surface after the ink droplet lands, and calculate the spreading range of each ink droplet;

[0118] Specifically, the spreading process of the ink droplet contacting the surface can be recorded in real time. High-speed camera technology or contact angle measuring instrument can be used to capture the dynamic process of the ink droplet from the landing point to the stable spreading, and the equipment can be ensured to have a sufficiently high frame rate (≥1000FPS) to accurately record the details of the spreading process; if the shape of the ink droplet is static after the spreading is stable, a microscope or high-resolution imaging equipment (such as an industrial camera) can be used to capture its final spreading shape; through image processing technology, the collected image is pre-processed (such as grayscale, denoising and threshold segmentation) to clarify the boundary of the ink droplet; then, edge detection algorithm (such as Sobel algorithm) is used to detect the ink droplet. The boundary information of the ink droplet is extracted by using the geometric measurement algorithm, and the spreading diameter or spreading area is calculated by the geometric measurement algorithm (for non-circular ink droplets, the specific value can be obtained by pixel area statistics); then, the spreading range is related to the physical parameters such as the initial momentum (mass, speed), surface tension and contact angle of the ink droplet, and it is verified whether it conforms to the spreading model; if the model cannot be directly established, the prediction formula of the spreading range can be generated by fitting the experimental data (such as multivariate regression analysis); finally, the spreading range of each ink droplet is recorded, including the spreading diameter, spreading area and spreading boundary information, and a data set of the spreading range is generated.

[0119] The spreading range of all ink droplets is integrated to define the boundary of the inkjet area and calculate the area of ​​the inkjet area;

[0120] Specifically, after integrating the spreading ranges of all ink droplets, a geometric merging algorithm (such as a convex hull algorithm) may be used to connect the spreading boundaries of all ink droplets to form an overall boundary of the inkjet area.

[0121] Furthermore, if it is necessary to characterize the irregular area more accurately, the AlphaShape algorithm can be used to generate a non-convex boundary, and the total area of ​​the inkjet area can be calculated based on the delineated boundary using the polygon area formula.

[0122] Compare the overlapping area between the inkjet area and the target area, calculate the coverage, output the boundary, center point, area, coverage and offset of the inkjet area, and generate the inkjet area positioning information.

[0123] It should be noted that in order to obtain the center point of the inkjet area, the drop point position of the inkjet area is first predicted based on the motion trajectory r(t) of the ink droplet, the jetting velocity, the gravitational acceleration and the jetting angle; then, based on the motion trajectory of all ink droplets, the spreading range of each ink droplet on the contact surface is recorded (i.e., the area where the shape of each ink droplet changes after landing). This spreading range is actually the "printing trace" of the ink droplet on the contact surface; by integrating the spreading ranges of all ink droplets and based on the distribution of these spreading ranges, the geometric center of the entire inkjet area is calculated. This geometric center is the "center point of the inkjet area". Specifically, the overall center point of the inkjet area can be obtained by taking a weighted average of the geometric centers of all spreading areas.

[0124] The offset is determined in the following way: First, the actual final landing position of the ink drop is determined based on the droplet landing prediction (by calculating the motion trajectory r(t) of the ink droplet). These actual landing points constitute the boundaries and coverage area of ​​the inkjet area; the target area is usually the preset, desired inkjet location, which is specifically defined by the design file, target pattern or other reference standard; by comparing the center point, boundary, area, etc. of the actual inkjet area with the corresponding position of the target inkjet area, the spatial deviation or position offset between them is calculated. The offset can be expressed as the distance between the actual center point position and the target center point position, or it can be quantified by calculating the difference between the geometric overlap of the actual inkjet area and the target area. In summary, the offset is the difference obtained by comparing the position, shape, size, etc. of the actual inkjet area with the target area, and it is a key parameter for measuring the inkjet accuracy deviation.

[0125] Specifically, a geometric Boolean operation (such as an intersection algorithm) is used to calculate the overlap between the inkjet area and the target area, and the boundary and shape of the overlap area are extracted;

[0126] Applying the polygon area formula to the boundary of the overlapping area, calculating the overlapping area, and calculating the utilization rate of the inkjet area according to the ratio of the overlapping area to the total area of ​​the inkjet area;

[0127] The geometric center point of the inkjet area is further calculated, and the offset is calculated in combination with the center point of the target area; finally, the positioning information of the boundary, center point, area, coverage and offset of the inkjet area is output;

[0128] It should also be noted that the prediction result refers to the prediction of the ink droplet motion trajectory by the deep learning model based on the input time series data (such as the jetting parameters during the inkjet process, the initial behavior data of the ink droplets, etc.). This prediction result usually refers to the prediction of the motion trajectory of the ink droplets over time (i.e., dynamic parameters such as position and speed). Specifically, this result includes:

[0129] The change in the position of the ink droplet: that is, the expected position of the ink droplet at a certain point in time; dynamic characteristics such as speed and acceleration: these are also key parameters that affect the trajectory of the ink droplet;

[0130] In short, the prediction result is usually the output given by a prediction model, which is a prediction of the future behavior of the ink droplets, so as to facilitate the subsequent inkjet area analysis, correction and adjustment.

[0131] Furthermore, the relationship between the motion trajectory and the positioning information is as follows: the motion trajectory reflects the actual motion path of the ink droplets over time, while the positioning information of the inkjet area is used to determine whether the ink droplets accurately cover the target area. Specifically, the positioning information describes the boundaries and center of the inkjet area, while the motion trajectory of the ink droplets determines the distribution of the ink droplets in the area and its deviation.

[0132] Combining the relationship between the prediction results and the positioning information: The prediction results provide the motion trajectory of the ink droplets, while the positioning information helps confirm whether the ink droplets fall within the predetermined inkjet area. For example, if the drop point offset of the ink droplets is predicted by the motion trajectory, combined with the positioning information of the inkjet area, the parameters of the inkjet device (such as the spray angle, speed, etc.) can be adjusted to ensure that the ink droplets fall correctly into the predetermined area.

[0133] S3. According to the positioning information of the inkjet area, multispectral image acquisition is performed on the inkjet area to obtain multispectral data.

[0134] Furthermore, the positioning information of the inkjet area is loaded as the inkjet area range to be collected. After the positioning information is accurately loaded, the collection time can be reduced and the waste of computing resources can be reduced.

[0135] Specifically, the positioning information of the inkjet area is loaded, including the boundary, center point, area, coverage and offset.

[0136] Furthermore, the positioning information is loaded into the image acquisition controller using a standardized interface;

[0137] According to the boundary information of the inkjet area, the optimal field of view of the multispectral camera is calculated, such as the ratio of the field of view width to height, the focal length adjustment range, etc.

[0138] The positioning information of the inkjet area is matched with the calibration parameters of the multispectral camera to determine the range of the inkjet area that needs to be collected.

[0139] Align the field of view center of the multispectral camera to the center point of the inkjet area, calibrate the boundary of the inkjet area, and set the acquisition mode;

[0140] Among them, for the setting of the acquisition mode, it is necessary to set the multi-spectral acquisition mode according to the shape and area of ​​the inkjet area. The specific modes are as follows:

[0141] Single-point acquisition mode: suitable for small inkjet areas, collecting data for the entire area at one time.

[0142] Block scanning mode: suitable for large inkjet areas, the inkjet area is divided into several small blocks, and multispectral data is collected block by block.

[0143] Dynamic adjustment mode: dynamically adjust the acquisition range and blocking strategy according to the real-time inkjet area offset.

[0144] The inkjet area is scanned step by step to acquire multispectral image data.

[0145] Among them, gradually scanning the inkjet area refers to the camera movement route designed within the inkjet area to ensure that the multispectral camera can cover the target area without blind spots.

[0146] S4. Input the multispectral data into the convolutional neural network model, combine the prediction results to identify the color deviation and defect type, and generate the corresponding correction parameters.

[0147] Furthermore, the spectral image data is loaded into the data processing pipeline;

[0148] According to the inkjet area positioning information, the data of the large inkjet area is divided into small blocks, each of which corresponds to the data slice of the multi-spectral channel;

[0149] Process the multispectral data of each block and organize the processed multispectral data into an input tensor;

[0150] Specifically, each block data is subjected to noise filtering (such as based on median filtering or Gaussian filtering) to remove interference signals in the spectral image;

[0151] Normalize the multi-spectral channel data and map the value range of each channel to the interval [0,1] to enhance the convergence performance of the model;

[0152] Convert each block of data into a three-dimensional tensor format (such as [number of channels, height, width]) and organize it into input tensors in channel order.

[0153] Add region metadata to each input tensor to ensure that the block data has geometric location correlation;

[0154] The input tensor is fed into a deep learning model based on a 3D convolutional neural network to extract the spectral-spatial joint features in the multispectral data and generate color deviation features and defect representations;

[0155] It should be noted that before the input tensor is input into the deep learning model based on the 3D convolutional neural network, the convolutional neural network model needs to be trained with multispectral image data, and the training data should include multispectral images of different inkjet quality, color distribution, texture and defect types. The convolutional neural network is trained using labeled image data (such as images with defect types and color deviations marked) using supervised learning methods to ensure that the network can automatically identify defects and color deviations in images and generate accurate correction parameters.

[0156] Among them, the spectral-spatial joint features specifically refer to the features that combine spectral channel information and spatial texture information, which are used to analyze color deviation and defect type.

[0157] It should be noted that the specific steps for generating color deviation features and defect representations are as follows: batch loading the input tensors into the deep learning model, and extracting spectral-spatial joint features through the 3D convolution layer; in the convolution layer, the spectral features are extracted using the correlation between multi-spectral channels, and the texture features are extracted by combining spatial information; the feature vector is output, which contains key information such as color deviation features and defect representations.

[0158] According to the color deviation characteristics and defect characterization, the color correction matrix is ​​calculated to dynamically adjust the color ratio and intensity when the ink droplets are ejected;

[0159] Among them, the color correction matrix is ​​calculated, specifically, the parameters of the color correction matrix, including color ratio and intensity adjustment values, are calculated using a regression model. Specifically, the color correction matrix is ​​calculated based on the input color deviation characteristics and defect characterizations, and its core purpose is to dynamically adjust the color output according to actual conditions during the inkjet process. The role of the regression model here is to predict and calculate the color ratio (jetting weight of each color) and intensity (jetting amount of each color droplet) that need to be adjusted through mathematical modeling based on historical data and current color deviations (such as color unevenness, color distortion, etc.) and defect characteristics of ink droplets (such as ink droplet shape, size, quality, etc.).

[0160] It should also be noted that color deviation characteristics and defect characterization are two important factors that affect color output during the inkjet process: Color deviation characteristics: usually refers to the difference between the actual color output and the expected value during the printing process. For example, some areas may have too much red or lack of blue, or the color may be uneven. Defect characterization: refers to defects that may occur in the ink droplet jetting process, such as irregular droplet shape, droplet size deviation, or inaccurate droplet jetting angle. These important factors are analyzed by the deep learning model and combined with the dynamic parameters of the inkjet system (such as jet pressure, jet frequency, etc.), and finally the jetting ratio and intensity of each color are adjusted through the regression model. Through this method, accurate color correction can be achieved to ensure the stability and consistency of the printing effect.

[0161] Furthermore, the intensity here refers to the force of ink droplet ejection or the number of ink droplets, which is usually controlled by parameters such as ejection pressure and frequency. In the process of ink droplet ejection, the intensity of ink droplets directly affects the projection distance, shape and final color distribution of ink droplets. Therefore, the calculation of the color correction matrix not only considers the color deviation, but also adjusts the intensity of the ejection (i.e., the ejection amount of each color ink droplet) to correct any possible deviation or unevenness.

[0162] Based on the predicted droplet trajectory, landing point and shape change, the color correction matrix is ​​comprehensively analyzed with the boundary, center point, area, coverage and offset of the inkjet area to generate the inkjet correction parameters.

[0163] It should be noted that "comprehensive analysis" refers to integrating multiple parameters (such as color correction matrix, ink drop trajectory, landing point, shape change, and geometric information of the inkjet area) through algorithms and models, comprehensively considering the correlation and influence between each parameter, so as to generate inkjet correction parameters.

[0164] S5. The correction parameters are transmitted to the inkjet control device to adjust the nozzle ejection parameters in real time.

[0165] Furthermore, the inkjet correction parameters are standardized and packaged into data packets;

[0166] Among them, "standardization processing" specifically refers to the processing of inkjet correction parameters to unify the format, quantify the data, normalize the range (such as normalizing to a specified range), add metadata and verification information to ensure the consistency, accuracy and parseability of the data during transmission and use.

[0167] Using the message transmission protocol, the data packet is transmitted to the inkjet control device receiving end in real time and the correction parameters are decoded;

[0168] Among them, decoding the correction parameters specifically refers to parsing the received data packet using a decoding method that matches the data packet format (such as a parser based on JSON or XML format, or using a protocol buffer parser for binary data packets, such as ProtocolBuffers), extracting the correction parameters and verifying their integrity and validity.

[0169] According to the decoded correction parameters, the injection pressure and injection frequency of the nozzle are dynamically adjusted;

[0170] It should be noted that the color correction matrix, inkjet correction parameters, jetting pressure and jetting frequency values ​​are interrelated key factors. First, the color deviation and defects of the inkjet area are analyzed through multispectral image data and deep learning models to generate a color correction matrix, which is used to adjust the color ratio and intensity of each ink droplet to ensure the color consistency of the inkjet output. Subsequently, based on the prediction of the ink droplet motion trajectory, landing point and shape change, combined with information such as the boundary, area and coverage of the inkjet area, a comprehensive inkjet correction parameter is generated. These inkjet correction parameters include not only the adjustment of the color correction matrix, but also the correction of the jetting position and jetting angle, as well as the dynamic adjustment of the jetting pressure and jetting frequency. Specifically, the jetting pressure determines the initial velocity and jetting direction of the ink droplet, which directly affects the flight trajectory and position accuracy of the ink droplet; while the jetting frequency value controls the jetting density of the ink droplet, affecting the uniformity and coverage of the inkjet area. By transmitting and decoding the generated inkjet correction parameters in real time, the inkjet control device can accurately adjust the jetting pressure and jetting frequency of the nozzle, thereby optimizing the distribution, shape and color output of the ink droplet. The entire process ensures high precision, uniformity and color consistency of inkjet printing, avoiding color deviations and defects.

[0171] Combined with the predicted drop point offset of ink drop motion, the jet angle is adjusted dynamically;

[0172] Dynamically adjust the color ratio of the nozzle according to the color correction matrix.

[0173] Specifically, the nozzle's injection pressure, frequency, injection angle and color ratio are adjusted in a closed loop according to the decoded correction parameters: the correction parameters are first input into the nozzle control module, the injection pressure and frequency are adjusted through a micro solenoid valve to adjust the injection power, the injection angle is adjusted through a servo motor or a stepper motor to adjust the spatial position of the nozzle, and the color ratio is dynamically allocated by a multi-channel ink droplet controller according to the color correction matrix to allocate the injection ratio of different color ink droplets; at the same time, the injection status data (such as pressure, frequency, landing point position and color output) is collected in real time, the adjustment effect is verified through sensors and motion prediction modules, and the parameters are further optimized to ensure that the ink droplet trajectory, landing point and color output are consistent with the target.

[0174] The present embodiment also provides an online detection system based on inkjet printing, including: a data acquisition module, a motion prediction module, an image acquisition module, a defect recognition module and a real-time correction module; the data acquisition module is used to collect the operating status data of inkjet printing in real time and transmit it to the deep learning model; the motion prediction module is used to predict the trajectory, landing point and shape change of ink droplets through the deep learning model combined with the principle of fluid dynamics, generate prediction results and provide positioning information of the inkjet area; the image acquisition module is used to perform multispectral image acquisition on the inkjet area according to the positioning information of the inkjet area to obtain multispectral data; the defect recognition module is used to input the multispectral data into the convolutional neural network model, identify the color deviation and defect type in combination with the prediction results, and generate corresponding correction parameters; the real-time correction module is used to transmit the correction parameters to the inkjet control device to adjust the nozzle injection parameters in real time.

[0175] This embodiment also provides a computer device, which is suitable for the case of an online detection method based on inkjet printing, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the online detection method based on inkjet printing proposed in the above embodiment.

[0176] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0177] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the online detection method based on inkjet printing proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0178] In summary, the present invention combines the deep learning model with the principles of fluid dynamics to accurately predict the trajectory, landing point and shape change of ink droplets, which can fully capture the dynamic characteristics and nonlinear behavior of ink droplet motion, and significantly improve the accuracy and reliability of inkjet area positioning; at the same time, the adaptive fractional-order derivative is introduced to model the complex dynamic behavior of ink droplet motion, and the calculation parameters can be dynamically adjusted to adapt to changes under different injection conditions, thereby optimizing the inkjet effect; in addition, by generating information on the boundary, center point, area, coverage and offset of the inkjet area in real time, a high-precision positioning basis is provided for subsequent multi-spectral detection and defect correction, thereby achieving efficient and accurate online detection and inkjet quality control.

[0179] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An online detection method based on inkjet printing, characterized in that: include, Acquire operation status data during the inkjet printing process, and combine the operation status data to generate a time series data structure, wherein the operation status data includes nozzle ejection parameters and initial behavior data of ink droplets; The time series data structure is transmitted to a deep learning model, and the time series data structure is preprocessed by the deep learning model to obtain a motion trajectory r of the ink droplet at the current time t, output a prediction result, and store the positioning information of the inkjet area; According to the positioning information of the inkjet area, multispectral image acquisition is performed on the inkjet area to obtain multispectral data, wherein the multispectral data includes color distribution, texture and spectral characteristic data; The multispectral data is input into the convolutional neural network model, and the color deviation and defect type are identified by combining the prediction results, and the corresponding correction parameters are generated; The correction parameters are transferred to the inkjet control device to adjust the nozzle ejection parameters in real time.

2. The online detection method based on inkjet printing according to claim 1, characterized in that: The performing data preprocessing on the time series data structure by using the deep learning model includes: Decode the received time series data to obtain the feature vector of the ink drop; Performing multi-modal feature fusion on the feature vector to obtain a fusion feature F of the ink drop; Taking the fusion feature F as the adjustment factor, the dynamic adjustment fractional order α(t′) of the ink droplet is calculated, and its calculation expression is: α(t′)=σ(W T g+b); Among them, t′ represents the time parameter for dynamically adjusting the fractional order α, σ represents the Sigmoid activation function, and W T g represents the linear combination of the fused features g, and b represents the bias term.

3. The online detection method based on inkjet printing according to claim 2, characterized in that: After obtaining the dynamic adjustment fractional order α(t′) of the ink droplet, the method further includes: The time-varying order is introduced to perform adaptive fractional derivative operation on the dynamically adjusted fractional order α(t′) to capture the non-stationary and dynamic characteristics of the ink droplet motion. The expression is: in, represents the fractional derivative operation of the dynamic variable function f(t), f(t) represents the function of the ink droplet with a dynamic variable at the current time t, Γ represents the gamma function, 1-α(t′) represents the input of the gamma function, dt represents the derivative at the current time t, t w is the window width at the current time t, tt w represents the integral operation, τ represents the past time, t-τ represents the time difference between the current time t and the past time τ, f(τ) is the function value of τ at a certain time in the past, dτ represents a small change in the integral variable, and d represents the differential operator; Based on the adaptive fractional derivative, the motion trajectory r(t) of the ink droplet under gravity is calculated, and its calculation expression is: Among them, a(t′) represents acceleration, t′ represents the time interval from the initial moment to the current moment, represents the viscous resistance, k is the damping coefficient, m is the mass of the ink drop, v(t′) represents the instantaneous velocity of the ink drop at time t′, represents the external force acting on the ink droplet, represents the resistance of air to the high-speed moving ink droplets, β represents the air resistance coefficient, v(t) represents the instantaneous speed of the ink droplet at time t, v0 represents the initial speed of the ink droplet, represents the change in speed, represents the change in position, r(t) represents the motion trajectory of the ink droplet at the current time t, and r0 represents the initial position of the ink droplet.

4. The online detection method based on inkjet printing according to claim 3, characterized in that: After obtaining the motion trajectory r(t) of the ink droplet under gravity and before outputting the prediction result, the method further includes: According to the motion trajectory r(t) of the ink droplet at the current time t, combined with the jetting speed, gravity acceleration and jetting angle, the position of the ink droplet landing point is predicted; The center point and distribution range of the ink droplet are calculated based on the location of the ink droplet and the motion trajectory r(t) of the ink droplet at the current time t. Record the shape change of the contact surface after the ink droplet lands, and calculate the spreading range of each ink droplet; The spreading range of all ink droplets is integrated to define the boundary of the inkjet area and calculate the area of ​​the inkjet area; Compare the overlapping area between the inkjet area and the target area, calculate the coverage, output the boundary, center point, area, coverage and offset of the inkjet area, and generate the inkjet area positioning information.

5. The online detection method based on inkjet printing according to claim 1, characterized in that: According to the positioning information of the inkjet area, multispectral image acquisition is performed on the inkjet area to obtain multispectral data. The specific steps are as follows: Load the positioning information of the inkjet area as the inkjet area range to be collected; Align the field of view center of the multispectral camera to the center point of the inkjet area, calibrate the boundary of the inkjet area, and set the acquisition mode; The inkjet area is scanned step by step to acquire multispectral image data.

6. The online detection method based on inkjet printing according to claim 5, characterized in that: The multispectral data is input into the convolutional neural network model, the color deviation and defect type are identified in combination with the prediction results, and the corresponding correction parameters are generated. The specific steps are as follows: Loading spectral image data into the data processing pipeline; According to the inkjet area positioning information, the data of the large inkjet area is divided into small blocks, each of which corresponds to the data slice of the multi-spectral channel; Process the multispectral data of each block and organize the processed multispectral data into an input tensor; Add region meta information to each input tensor; The input tensor is fed into a deep learning model based on a 3D convolutional neural network to extract the spectral-spatial joint features in the multispectral data and generate color deviation features and defect representations; According to the color deviation characteristics and defect characterization, the color correction matrix is ​​calculated to dynamically adjust the color ratio and intensity when the ink droplets are ejected; Based on the predicted ink drop trajectory, landing point and shape change, the color correction matrix is ​​comprehensively analyzed with the boundary, center point, area, coverage and offset of the inkjet area to generate the inkjet correction parameters.

7. The online detection method based on inkjet printing according to claim 6, characterized in that: The correction parameters are transmitted to the inkjet control device to adjust the nozzle jetting parameters in real time. The specific steps are as follows: The inkjet calibration parameters are standardized and packaged into data packets; Using the message transmission protocol, the data packet is transmitted to the inkjet control device receiving end in real time and the correction parameters are decoded; According to the decoded correction parameters, the injection pressure and injection frequency of the nozzle are dynamically adjusted; Combined with the predicted drop point offset of ink drop motion, the jet angle is adjusted dynamically; Dynamically adjust the color ratio of the nozzle according to the color correction matrix.

8. An online detection system based on inkjet printing, used to perform the method according to any one of claims 1 to 7, characterized in that: The system includes: a data acquisition module, a motion prediction module, an image acquisition module, a defect recognition module and a real-time correction module; The data acquisition module is used to collect the operating status data of inkjet printing in real time and transmit it to the deep learning model; The motion prediction module is used to predict the trajectory, landing point and shape change of ink droplets by combining the deep learning model with the principle of fluid dynamics, generate prediction results and provide positioning information of the inkjet area; The image acquisition module is used to acquire multispectral images of the inkjet area according to the positioning information of the inkjet area to obtain multispectral data; The defect recognition module is used to input the multispectral data into the convolutional neural network model, identify the color deviation and defect type in combination with the prediction results, and generate corresponding correction parameters; The real-time correction module is used to transmit the correction parameters to the inkjet control device to adjust the nozzle ejection parameters in real time.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the online detection method based on inkjet printing according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the online detection method based on inkjet printing described in any one of claims 1 to 7 are implemented.

Images