A Smart Method for Detecting Droplet Volume in Inkjet Printing

Abstract

This invention belongs to the field of novel displays, specifically relating to an intelligent method and device for detecting droplet volume in inkjet printing. The method includes: performing a trial print to detect the set of deposited droplet features corresponding to each nozzle, using this as input parameters to establish an unlabeled classification sample set; labeling the unlabeled classification sample set to determine whether it is abnormal, establishing a labeled classification sample set, and training a random forest classification model using the labeled classification sample set; labeling normal samples in the labeled classification sample set using the results of the droplet volume measurement, establishing a regression sample set, and then training a BP neural network regression model using the regression dataset; and using the trained random forest classification model and BP neural network regression model to calculate the droplet volume corresponding to the set of deposited droplet features collected during the formal printing interval. This invention effectively improves the efficiency of arrayed droplet volume detection in large-area inkjet printing production lines while ensuring detection accuracy, achieving efficient manufacturing.

Description

Technical Field

[0001] This invention belongs to the field of novel displays, and more specifically, relates to a method and device for intelligent detection of inkjet printing droplet volume. Background Technology

[0002] Inkjet printing is an additive manufacturing technology that fabricates electronic devices without the need for templates. This technology relies on a printhead to eject specific functional droplets onto a substrate to achieve contactless printing. Compared to traditional manufacturing methods such as vapor deposition and screen printing, inkjet printing not only excels at large-area, high-precision manufacturing processes but also offers significant advantages in resource conservation and environmental protection. Given its advantages of low cost, large area coverage, and flexibility, inkjet printing technology has been successfully applied in high-tech fields such as graphic printing and microelectromechanical systems (MEMS), with the fabrication of novel display devices receiving the most attention.

[0003] Inkjet printing technology, as a manufacturing process for display panels, still faces many challenges, one of the key issues being the measurement of the print droplet volume. During the inkjet printing process for fabricating novel display devices, variations in printhead manufacturing, improper process parameter settings, and the relatively complex printing environment can lead to jetting defects such as nozzle clogging, satellite droplets, and abnormal droplet volume. These defects can cause deviations between the print droplet volume and the target value, ultimately affecting the quality of the final display device. Therefore, accurately measuring the droplet volume during the printing process is crucial for achieving high-quality and high-efficiency printing.

[0004] Existing methods for detecting the volume of inkjet printing droplets are mainly divided into flying droplet measurement methods and deposited droplet measurement methods. Flying droplet measurement methods primarily rely on visual measurement, with traditional visual measurement techniques largely based on the principle of stroboscopic imaging. However, with the continuous increase in inkjet printing resolution and printing area, the number of nozzles to be inspected has also increased significantly, making flying droplet observation for droplet volume detection too time-consuming and inefficient. Deposited droplet measurement methods mainly include white light interferometry, gravimetric analysis, and visual inspection. Among these, white light interferometry and gravimetric analysis have high space requirements and insufficient detection efficiency, and are often used for verifying detection results, while visual inspection is only used to detect the presence or absence of droplets.

[0005] Therefore, how to achieve efficient volume detection of arrayed droplets has become a key and challenging issue for the continued development of inkjet display technology. Summary of the Invention

[0006] In view of the shortcomings of existing technologies and the need for improvement, this invention provides an intelligent method and device for detecting the volume of inkjet printing droplets, the purpose of which is to achieve efficient volume detection of arrayed droplets.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for intelligent detection of inkjet printing droplet volume is provided, comprising:

[0008] During the formal printing interval, the nozzle is moved to the test printing area for test spraying; images of the deposition observation area are acquired to detect the feature set of deposition droplets corresponding to each nozzle, and input into a pre-trained machine learning classification model to obtain the judgment result of whether the nozzle is abnormal; based on the judgment result, the feature set of deposition droplets corresponding to each normal nozzle is input into a pre-trained machine learning regression model to obtain the volume of droplets ejected from each normal nozzle, thereby realizing the measurement of array droplet volume;

[0009] The machine learning classification model is constructed as follows: a trial print is performed, the feature set of deposited droplets corresponding to each nozzle is detected and labeled as to whether the nozzle is abnormal or not, to obtain a classification sample set; the classification sample set is used to train a machine learning classification model for judging whether the nozzle is abnormal based on the feature set of deposited droplets.

[0010] The machine learning regression model is constructed as follows: during trial printing, the volume of the flying droplets in each nozzle is measured; the flying droplet volume is labeled for the samples corresponding to the normal labels in the classification sample set to obtain a regression sample set; using the regression sample set, a machine learning regression model for calculating droplet volume based on the feature set of deposited droplets is trained.

[0011] Furthermore, the features included in the set of sedimentary droplet features are: the number of sedimentary droplets, the spreading diameter of sedimentary droplets, the x-axis landing point deviation of sedimentary droplets, and the y-axis landing point deviation of sedimentary droplets; wherein, if the number of sedimentary droplets is greater than or equal to two, then the spreading diameter, the x-axis landing point deviation, and the y-axis landing point deviation are all features of the main sedimentary droplet.

[0012] Furthermore, the criteria for determining whether the nozzle is abnormal are: the number of deposited droplets, the x-axis landing point deviation, and / or the y-axis landing point deviation; when there are no deposited droplets, the number of deposited droplets is greater than or equal to two, the x-axis landing point deviation exceeds a preset value, or the y-axis landing point deviation exceeds a preset value, the corresponding set of deposited droplet features is marked as an abnormal sample.

[0013] Furthermore, the machine learning classification model is a random forest classification model, which is obtained by combining multiple decision tree models based on a weighted voting method. In the process of constructing the random forest classification model, the weighted average of the classification accuracy of all decision tree models in the random forest classification model generated after each parameter adjustment is taken as the classification accuracy estimate of the random forest classification model.

[0014] Furthermore, the classification accuracy estimate is expressed as:

[0015]

[0016]

[0017] Where P is the classification accuracy estimate, m represents the number of decision tree models, and p i TP represents the classification accuracy of the i-th decision tree model. i TN represents the number of normal samples correctly predicted by the i-th decision tree model. i P represents the number of outlier samples correctly predicted by the i-th decision tree model. i N represents the number of normal samples. i Indicates the number of abnormal samples.

[0018] Furthermore, the prediction output of the random forest classification model for sample x is:

[0019]

[0020]

[0021] Where x represents a sample of features of a sedimentary droplet; sgn() is the step function; m is the number of decision tree models; and w i It is the i-th decision tree model h i The weight, h i (x) is the i-th decision tree model h i For the predicted output of x, i represents an integer from 1 to m.

[0022] Furthermore, the machine learning regression model is a BP neural network regression model, comprising: an input layer, a hidden layer, and an output layer. The input layer contains P1 = M nodes, where M is the number of feature values ​​of the sediment droplets in the feature set, and no activation function is used in the input layer. The hidden layer contains P2 = 2P1 + 1 nodes, and the activation function is the ReLU function. The output layer contains P3 = 1 node, and the activation function is a linear function with the expression f(z) = z, where z is the input variable.

[0023] The present invention also provides an intelligent detection device for inkjet printing droplet volume, for performing an intelligent detection method for inkjet printing droplet volume as described above, comprising: an inkjet printing module, a first droplet deposition substrate, a deposited droplet observation module, a motion module, and a control module;

[0024] The control module is used to control the inkjet printing module to move above the first droplet deposition substrate for test spraying via the motion module, and to control the deposition droplet observation module to move to the droplet deposition area via the motion module, to acquire images of the deposition droplets ejected from each nozzle and transmit them to the control module; the control module is also used to detect the deposition droplet feature set corresponding to each nozzle based on the deposition droplet images, and input it into a pre-trained machine learning classification model to obtain a judgment result on whether the nozzle is abnormal or not, and according to the judgment result, input the deposition droplet feature set corresponding to each normal nozzle into a pre-trained machine learning regression model to obtain the droplet volume ejected from each normal nozzle.

[0025] Furthermore, it also includes a flying droplet observation module and a second droplet deposition substrate;

[0026] The control module is also used to control the inkjet printing module to perform trial printing on the second droplet deposition substrate, control the flying droplet observation module to acquire flying droplet images and transmit them to the control module, and control the deposition droplet observation module to acquire deposition droplet images on the second droplet deposition substrate and transmit them to the control module. The control module is also used to calculate the flying droplet volume of multiple nozzles based on the flying droplet images, detect the deposition droplet feature sets corresponding to the multiple nozzles based on the deposition droplet images, train the machine learning classification model using a classification sample set, and train the machine learning regression model using a regression sample set. The classification sample set is obtained by labeling multiple deposition droplet feature sets as abnormal or not, and the regression sample set is obtained by labeling the flying droplet volume of the deposition droplet feature sets labeled as normal in the classification sample set.

[0027] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0028] (1) The method of the present invention establishes a learning model between the volume of flying droplets and the characteristics of deposited droplets, including a mapping model from the characteristics of deposited droplets to the anomaly of the nozzle and a mapping model from the characteristics of deposited droplets of normal nozzles to the volume of flying droplets. Finally, the mapping from the characteristics of deposited droplets to the volume of droplets is realized. This process can replace the inefficient observation of flying droplets with the efficient observation of deposited droplets. Since the observation of deposited droplets can observe a large number of droplets at once, it solves the problem of low detection efficiency caused by the observation of flying droplets, thereby realizing the efficient volume detection of arrayed droplets.

[0029] (2) The method of this invention proposes to use a random forest classification model to classify the abnormal situation of sediment droplets, and realizes the determination of whether there is an anomaly in the sediment droplets by the feature set of sediment droplets. Compared with the traditional manual anomaly identification method, it effectively reduces the detection cost and improves the detection efficiency of droplet anomalies.

[0030] (3) The method of this invention proposes to use a BP neural network regression model to calculate the volume of the inkjet droplets, which realizes the calculation of the corresponding droplet volume through the characteristics of the deposited droplets. Compared with the traditional formula calculation and simulation method, it avoids the complexity of fluid characteristics and physical dynamics, significantly reduces the calculation cost and improves the calculation efficiency. Attached Figure Description

[0031] Figure 1 This is an overall flowchart of the intelligent detection method for inkjet printing droplet volume provided in an embodiment of the present invention;

[0032] Figure 2 This is a schematic diagram of the intelligent inkjet printing droplet volume detection system provided in an embodiment of the present invention;

[0033] Figure 3 A schematic diagram illustrating an embodiment of the intelligent inkjet printing droplet volume detection system and method provided in this invention.

[0034] Figure 4 This is a process flow diagram of the model building stage in the intelligent detection method for inkjet printing droplet volume provided in an embodiment of the present invention;

[0035] Figure 5 This is a process flow diagram of the intelligent volume detection stage in the inkjet printing droplet volume intelligent detection method provided in the embodiments of the present invention;

[0036] Figure 6 A structural diagram of a machine learning regression model provided in an embodiment of the present invention;

[0037] Figure 7 This is a schematic diagram illustrating the acquisition of the feature set of deposited droplets provided in an embodiment of the present invention.

[0038] In all the accompanying drawings, the same reference numerals are used to denote the same elements or structures, wherein:

[0039] 1 is the inkjet printing module; 21 is the side-view observation camera; 22 is the stroboscopic light source; 23 is the waste liquid tank; 31 is the downward-view observation camera; 32 is the droplet deposition substrate; 41 is the printhead moving device; 43 is the substrate moving device; 5 is the control module. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0041] Example 1

[0042] A method for intelligent detection of inkjet printing droplet volume includes:

[0043] During the formal printing interval, the nozzle is moved to the test printing area for test spraying; images of the deposition observation area are acquired to detect the feature set of deposition droplets corresponding to each nozzle, and input into a pre-trained machine learning classification model to obtain the judgment result of whether the nozzle is abnormal; based on the judgment result, the feature set of deposition droplets corresponding to each normal nozzle is input into a pre-trained machine learning regression model to obtain the volume of droplets ejected from each normal nozzle, thereby realizing the measurement of array droplet volume;

[0044] The machine learning classification model was constructed as follows: a trial print was performed, the feature set of the deposited droplets corresponding to each nozzle was detected and the nozzle was labeled as abnormal or not, and a classification sample set was obtained; the classification sample set was used to train a machine learning classification model for judging the nozzle abnormality based on the feature set of deposited droplets.

[0045] The machine learning regression model was constructed as follows: during trial printing, the volume of the flying droplets at each nozzle was measured; the flying droplet volume was labeled for the samples corresponding to the normal labels in the classification sample set to obtain a regression sample set; using the regression sample set, a machine learning regression model for calculating droplet volume based on the feature set of deposited droplets was trained.

[0046] It should be noted that each nozzle corresponds to a set of deposited droplet characteristics. In this embodiment, in order to increase the volume difference of deposited droplets between each nozzle, multiple droplets are continuously ejected from all nozzles of the nozzle head. Since each nozzle should normally eject only one deposited droplet at a time, and the droplets ejected multiple times should be superimposed on the same deposited droplet, when a certain nozzle has an anomaly, the droplets ejected from that nozzle may have multiple deposited droplets on the substrate, and the landing position may also deviate from the theoretical landing position.

[0047] Therefore, as a preferred embodiment, the features included in the above-mentioned set of sedimentary droplet features are: the number of sedimentary droplets, the spreading diameter of sedimentary droplets, the x-axis landing point deviation of sedimentary droplets, and the y-axis landing point deviation of sedimentary droplets; wherein, if the number of sedimentary droplets is greater than or equal to two, the spreading diameter, the x-axis landing point deviation, and the y-axis landing point deviation are all features of the main sedimentary droplets.

[0048] As a further preferred implementation, the criteria for determining whether the above-mentioned nozzle is abnormal are: the number of deposited droplets, the x-axis landing point deviation, and / or the y-axis landing point deviation; when there are no deposited droplets, the number of deposited droplets is greater than or equal to two, the x-axis landing point deviation exceeds a preset value, or the y-axis landing point deviation exceeds a preset value, then the corresponding set of deposited droplet features is marked as an abnormal sample. That is, the above-mentioned criteria for marking abnormalities in the number and position of deposited droplets are determined by the number of deposited droplets, the x-axis landing point deviation, and the y-axis landing point deviation.

[0049] This embodiment of the method establishes a learning model between the volume of flying droplets and the characteristics of deposited droplets, including a mapping model from the characteristics of deposited droplets to the anomaly status of the nozzle and a mapping model from the characteristics of deposited droplets in a normal nozzle to the volume of flying droplets. Ultimately, it achieves the mapping from the characteristics of deposited droplets to the volume of droplets. This process can replace the inefficient observation of flying droplets with the efficient observation of deposited droplets. Since the observation of deposited droplets can observe a large number of droplets at once, it solves the problem of low detection efficiency caused by the observation of flying droplets, thereby achieving efficient volume detection of arrayed droplets.

[0050] In other words, the method in this embodiment constructs a mapping model between the volume of flying droplets and the characteristics of deposited droplets during the printing process, realizing the acquisition of deposited droplet characteristics and subsequent calculation of droplet volume through array-based deposited droplet detection. Traditional visual detection methods for flying droplets detect a very small number of droplets per scan, leading to an increased number of acquisition attempts. Furthermore, the stroboscopic principle limits the detection time per scan, resulting in low efficiency. The detection method proposed in this embodiment shifts the detection target from flying droplets to deposited droplets, significantly reducing the difficulty of detection while increasing the number of droplets detected per scan and shortening the detection time, thereby greatly improving the efficiency of droplet volume detection.

[0051] This embodiment proposes using a random forest classification model to classify anomalies in sedimentary droplets. This allows for the determination of anomalies by analyzing the feature set of sedimentary droplets, effectively reducing detection costs and improving detection efficiency compared to traditional manual anomaly identification methods. Furthermore, this embodiment proposes using a backpropagation (BP) neural network regression model to calculate the volume of printed droplets. This allows for the calculation of corresponding droplet volumes based on sedimentary droplet features. Compared to traditional formula calculations and simulation methods, this avoids the complexity of fluid characteristic characterization and physical dynamics, significantly reducing computational costs and improving computational efficiency.

[0052] Preferably, the above-mentioned machine learning classification model is a random forest classification model, which is obtained by combining multiple decision tree models based on weighted voting. In the process of constructing the random forest classification model, the weighted average of the classification accuracy of all decision tree models in the random forest classification model generated after each parameter adjustment is taken as the classification accuracy estimate of the random forest classification model.

[0053] The specific parameters for the random forest classification model are as follows: the number of decision trees is m; the node splitting criterion for the decision tree model is the "Gini coefficient"; and the maximum depth of the decision tree model is N. d The minimum number of samples required for a decision tree model to split a node is S. s The minimum number of samples at a leaf node in a decision tree model is S. l .

[0054] The method for selecting the parameters of the above random forest classification model is as follows:

[0055] 1) According to the preset ratio, the labeled classification sample set is divided into a training set and a test set. Samples are drawn from the training set using random sampling with replacement and used as training data for the decision tree model. This results in a random forest classification model composed of multiple decision tree models.

[0056] 2) Gradually adjust the model parameters such as the number of decision tree models, the maximum depth of the decision tree model, the minimum number of samples required for split nodes of the decision tree model, and the minimum number of samples at leaf nodes of the decision tree model. Repeat step 1), and take the classification accuracy p of all decision tree models in the random forest classification model generated after each adjustment. i The weighted average is used as the classification accuracy estimate P of the random forest classification model. The model parameters corresponding to the highest classification accuracy P are selected as the optimal model parameters. P is calculated by the following formula:

[0057]

[0058]

[0059] Where m represents the number of decision tree models, p i TP represents the classification accuracy of the i-th decision tree model. i TN represents the number of correctly predicted normal samples. i P represents the number of correctly predicted outliers. i N represents the number of normal samples. i Indicates the number of abnormal samples.

[0060] The random forest classification model's prediction output for sample x is:

[0061]

[0062]

[0063] Where x represents the set of sediment droplet features of the sample; sgn() is the step function, m is the number of decision tree models, and w i It is the i-th decision tree model h i The weight, h i (x) is the i-th decision tree model h i For the predicted output of sample x, i represents an integer from 1 to m.

[0064] Preferably, the above-mentioned machine learning regression model is a BP neural network regression model, including: an input layer, a hidden layer and an output layer, wherein the input layer contains P1 = M nodes, where M is the number of feature values ​​of the sediment droplets in the feature set, and no activation function is used in the input layer; the hidden layer contains P2 = 2P1 + 1 nodes, and the activation function is the ReLU function; the output layer contains P3 = 1 node, and the activation function is a linear function with the function f(z) = z, where z is the input variable.

[0065] The predicted output of the above BP neural network regression model for sample x is:

[0066] y = xA1A2;

[0067] Where x is a vector of length 1×M, representing the set of features of the deposited droplets, where M is the number of features in the sample; A1 is a matrix of length P1×P2, representing the set of weights mapped from the input layer to the hidden layer; A2 is a matrix of length P2×P3, representing the set of weights mapped from the hidden layer to the output layer; y is the predicted output of the regression model for sample x.

[0068] The mapping weight matrices A1 and A2 are obtained by the following formula:

[0069]

[0070]

[0071] Among them, v i,j ω represents the weights mapped from the i-th node in the input layer to the j-th node in the hidden layer; p,1 This represents the weights mapped from the p-th node in the hidden layer to the nodes in the output layer; i represents an integer from 1 to P1; j and p represent integers from 1 to P2.

[0072] In summary, please refer to Figure 1The method for intelligent detection of inkjet printing droplet volume provided in this embodiment includes a model building stage and a volume detection stage. The model building stage includes the following operations S1 to S5, and the volume detection stage includes operation S6:

[0073] Model building phase:

[0074] Operation S1 is used to perform a trial print, detect the feature set of the deposited droplets corresponding to each nozzle, and use it as input parameters to build an unlabeled classification dataset.

[0075] Operation S2 is used to label the samples in the unlabeled classification dataset based on whether there are anomalies in the number and position of droplets, thus creating a labeled classification dataset;

[0076] Operation S3: Use the labeled classification dataset to train a machine learning classification model to obtain a machine learning classification model for determining whether there is droplet anomaly based on the feature set of deposited droplets.

[0077] Operation S4: Use the flight droplet volume measurement results to label the normal samples in the labeled classification dataset and establish a regression dataset;

[0078] Operation S5: Use the regression dataset to train a machine learning regression model to obtain a machine learning regression model that calculates droplet volume based on the feature set of deposited droplets.

[0079] Volume detection stage:

[0080] In operation S6, during the formal printing interval, the printhead is moved to the test printing area for test spraying, the set of deposited droplet features corresponding to each nozzle is detected, and the machine learning classification model and machine learning regression model are input sequentially to calculate the droplet volume corresponding to any set of deposited droplet features.

[0081] In this embodiment, the flying droplets and deposited droplets belong to the same droplet group and have the same volume. Given fixed substrate and solution properties, there must be a corresponding relationship between the characteristics of the deposited droplets and the volume of the flying droplets. However, this relationship can be relatively complex and difficult to obtain experimentally. Machine learning methods are less costly and more effective in handling such mapping problems. The number of droplets that can be detected in a single pass for deposited droplet detection is far greater than that for flying droplet detection. Therefore, replacing flying droplet detection with deposited droplet detection can significantly improve the efficiency of droplet volume detection.

[0082] Example 2

[0083] An intelligent detection device for inkjet printing droplet volume, used to perform an intelligent detection method for inkjet printing droplet volume as described above, includes: an inkjet printing module, a first droplet deposition substrate, a deposited droplet observation module, a motion module, and a control module.

[0084] The control module controls the inkjet printing module to move above the first droplet deposition substrate for test spraying via the motion module, and controls the deposition droplet observation module to move to the droplet deposition area via the motion module, acquiring images of the deposition droplets ejected from each nozzle and transmitting them to the control module. The control module is also used to detect the deposition droplet feature set corresponding to each nozzle based on the deposition droplet images, and input it into a pre-trained machine learning classification model to obtain a judgment result on whether the nozzle is abnormal. Based on the judgment result, the deposition droplet feature set corresponding to each normal nozzle is input into a pre-trained machine learning regression model to obtain the droplet volume ejected from each normal nozzle.

[0085] It should be noted that the movement of the inkjet printing module, the deposition droplet observation module, and the first droplet deposition substrate is all achieved by the control module controlling the movement of the motion module.

[0086] Preferably, the above-mentioned intelligent detection device for inkjet printing droplet volume further includes a flying droplet observation module; the control module is also used to control the inkjet printing module to perform trial printing on the second droplet deposition substrate, control the flying droplet observation module to acquire flying droplet images and transmit them to the control module, control the deposition droplet observation module to acquire deposition droplet images on the second droplet deposition substrate and transmit them to the control module; the control module is also used to calculate the flying droplet volume of multiple nozzles based on the flying droplet images, detect the deposition droplet feature sets corresponding to the multiple nozzles based on the deposition droplet images, train the machine learning classification model using a classification sample set, and train the machine learning regression model using a regression sample set, wherein the classification sample set is obtained by labeling the multiple deposition droplet feature sets as abnormal or not, and the regression sample set is obtained by labeling the deposition droplet feature sets labeled as normal in the classification sample set with flying droplet volume.

[0087] It should be noted that the printhead in the inkjet printing module described above may or may not be the same during the model building stage and the actual printing stage. Similarly, the first and second droplet deposition substrates may or may not be the same. However, the printing solution used for model building and the material of the second droplet deposition substrate are the same as those used in the actual printing stage. If the printing solution and droplet deposition substrate material change during the actual printing stage, the model needs to be rebuilt.

[0088] Specifically, the flying droplet detection module includes: a side-view observation camera, a stroboscopic light source, and a waste liquid tank; it uses backlighting for illumination and stroboscopic imaging for measuring the volume of flying droplets. The deposition droplet observation module is generally a downward-view observation camera, using coaxial illumination for acquiring images of deposition droplets. The motion module includes: a nozzle moving device and a substrate moving device, used for controlling each motion axis and adjusting the relative positions of each module.

[0089] The measurement of the volume of the flying droplets and the acquisition of images of the deposited droplets can be carried out through the following operation: First, move the nozzle to the position of the flying droplet detection module, measure and record the volume of the droplet corresponding to each nozzle, then move the nozzle above the droplet deposition substrate for test spraying, and then move the downward observation camera above the droplet deposition substrate to collect images of the deposited droplets.

[0090] To better illustrate Embodiment 1 and Embodiment 2, a specific detection system is now provided, such as... Figure 2 As shown, the system for intelligent detection of droplet volume in inkjet printing includes an inkjet printing module, a flying droplet observation module, a deposited droplet observation module, a motion module, and a control module. Each module will be explained in detail below:

[0091] The inkjet printing module 1 is used to generate droplets of a certain volume and velocity as required;

[0092] The flying droplet observation module 2 includes a side-viewing camera 21, a stroboscopic light source 22, and a waste liquid tank 23. The side-viewing camera 21 and the stroboscopic light source 22 are located on the same horizontal line and are used to acquire images of flying droplets and transmit them to the control module 5 in real time to calculate and obtain the actual volume data of the droplets. The waste liquid tank 23 is used to collect waste ink generated during the flying droplet observation process to prevent contamination of the equipment.

[0093] The droplet deposition observation module 3 includes a downward-looking observation camera 31 and a droplet deposition substrate 32. The downward-looking observation camera 31 is used to acquire images of the array of deposited droplets and transmit them to the control module 5 in real time to calculate and obtain the feature set of the deposited droplets; the droplet deposition substrate 32 is used to hold the deposited droplets.

[0094] The motion module includes a printhead moving device 41 and a substrate moving device 42. The printhead moving device 41 is used to adjust the position of the inkjet printing module 1 and the downward observation camera 31 to achieve flight observation, inkjet printing, or droplet deposition observation functions; the substrate moving device 42 is used to adjust the position of the droplet deposition substrate 32 to ensure that all inkjet droplets are deposited on the substrate.

[0095] The control module 5 is used to control the use of the side-view observation camera 21, the stroboscopic light source 22 and the downward-view observation camera 31 to acquire clear droplet images; it is also used to store and process the image data transmitted in real time by the flying droplet observation module 2 and the deposited droplet observation module 3 to obtain the corresponding droplet parameters; at the same time, it is also used to control the movement of the nozzle moving device 41 and the substrate moving device 42, and adjust the relative positions between the modules to carry out the detection process normally.

[0096] The inkjet printing droplet volume detection system described above can be used to detect the volume of inkjet printing droplets. The overall process is as follows: Figure 3 As shown:

[0097] (1) This system can first be applied to the trial printing stage before actual printing. By collecting observation data of flying droplets and sedimented droplets, a random forest classification model and a BP neural network regression model can be established sequentially, such as... Figure 4 As shown.

[0098] During the observation of flying droplets, the inkjet printing module 1 is moved to the position of the flying droplet detection module 2, specifically, to the horizontal line where the side-view observation camera 21 and the stroboscopic light source 22 are located. The nozzle flashes and the observation is performed. The side-view observation camera 21 and the stroboscopic light source 22 acquire images of flying droplets and transmit them to the calculation and control module 5. The volume of droplets corresponding to each nozzle is calculated and recorded. During droplet deposition observation, the inkjet printing module 1 is moved above the droplet deposition substrate 32 for trial printing, and the theoretical landing position of the droplets ejected from each nozzle is recorded. Then, the downward observation camera 31 is moved above the droplet deposition substrate 32 for observation, and the images of the deposited droplets are acquired and transmitted to the calculation control module 5. The feature set (D,n,Δx,Δy) of the deposited droplets corresponding to each nozzle is calculated, where D represents the spreading diameter of the deposited droplet, n represents the number of deposited droplets, Δx represents the x-axis landing point deviation of the deposited droplet, and Δy represents the y-axis landing point deviation of the deposited droplet. A certain number of samples are extracted from all the feature sets of the deposited droplets and used as input parameters to establish an unlabeled classification dataset θ.

[0099] According to a preferred embodiment of the present invention, the capacity of the unlabeled classification dataset θ is N. h =2000, the constructed unlabeled classification dataset can be represented as:

[0100]

[0101] Where (D,n,Δx,Δy) represents the set of sediment droplet features for each sample group.

[0102] The unlabeled classification dataset θ is labeled based on the presence of abnormal droplet numbers and positions. Samples in the unlabeled classification dataset are labeled as abnormal if there are no deposited droplets, the number of deposited droplets is greater than or equal to two, or the x / y direction droplet deviation exceeds a preset value; otherwise, they are labeled as normal samples. A labeled classification sample set D is then established. c :

[0103]

[0104] Where y represents the sample label, which takes the value of 0 or 1; when y = 0, it means that the sediment droplets corresponding to the sample are abnormal and the sample is an abnormal sample; when y = 1, it means that the sediment droplets corresponding to the sample are not abnormal and the sample is a normal sample.

[0105] Using labeled classification sample set D c The random forest classification model is trained with the set of features of sediment droplets as input and the output of whether the sediment droplets are abnormal, that is, whether the nozzles are abnormal.

[0106] More specifically, the parameters of the random forest classification model include: the number of decision trees in the model is m; the node splitting criterion for the decision tree model is the "Gini coefficient"; and the maximum depth of the decision tree model is N. d The minimum number of samples required for a decision tree model to split a node is S. s The minimum number of samples at a leaf node in a decision tree model is S. l ;

[0107] According to a preferred embodiment of the present invention, the training process of the random forest classification model is as follows:

[0108] Step 1: Divide the labeled classification dataset D into groups of 4:1. c Divided into training set D ce With test set D ct Random sampling with replacement is used to sample from the training set D. ce A subset of 200 samples is extracted from the dataset, and this process is repeated m times to obtain the subset {D}. ce1 D ce2 ,…,D cem};

[0109] Step 2: For each subset of data, train the decision tree model sequentially using unlabeled classification dataset samples as input parameters and sample labels as output parameters, obtaining a decision tree model set {h1, h2, ..., h...}. m};

[0110] Step 3: Take the classification accuracy p of all decision tree models in the random forest classification model generated after each adjustment. iThe weighted average is used as an estimate of the classification accuracy P of the random forest classification model:

[0111]

[0112]

[0113] Where m represents the number of decision tree models, p i TP represents the classification accuracy of the i-th decision tree model. i TN represents the number of normal samples correctly predicted in the subset of data. i P represents the number of outlier samples correctly predicted in the subset of data. i N represents the total number of normal samples in the subset. i This represents the total number of outlier samples in the subset;

[0114] Step 4: Gradually adjust the number of decision tree models (m) and the maximum depth of the decision tree model (N). d The minimum number of samples S required for splitting nodes in a decision tree model s The minimum number of samples S at the leaf nodes of the decision tree model l Once the model parameters are equal, repeat steps one through three, plot the relationship curves between the relevant parameters and the classification accuracy, and select the model parameters corresponding to the highest classification accuracy P as the optimal model parameters.

[0115] Step 5: Based on the obtained optimal model parameters, use the decision tree model set obtained in Steps 1 and 2, and construct a random forest classification model using a weighted voting method. The prediction output of the obtained random forest classification model for sample x is:

[0116]

[0117]

[0118] Where H(x) represents the prediction output of the random forest classification model on sample x; sgn() is the step function, m is the number of decision tree models, and w i It is the i-th decision tree model h i The weight, h i (x) is the i-th decision tree model h i For the predicted output of sample x, i represents an integer from 1 to m;

[0119] Step 6: Use test set D ct Test the classification performance of the random forest classification model and record the final classification accuracy to complete the construction of the random forest classification model.

[0120] Using the volume of flying droplets to classify labeled dataset D cNormal samples were labeled to establish a droplet volume regression sample set D. r :

[0121]

[0122] Where V represents the sample label, specifically the actual volume of the droplet; P h This represents a labeled classification dataset D. c The total number of normal samples.

[0123] Using droplet volume regression sample set D r Train a BP neural network regression model, with the set of features of the deposited droplets as the model input and the droplet volume as the model output;

[0124] More specifically, Figure 6 This is a schematic diagram of the BP neural network regression model structure provided by the present invention, including: an input layer, a hidden layer and an output layer, wherein: the input layer contains P1 = M nodes, where M is the number of feature values ​​of the deposited droplet, and the input layer does not use an activation function; the hidden layer contains P2 = 2P1 + 1 nodes, and the activation function is the ReLU function; the output layer contains P3 = 1 node, and the activation function is a linear function with the function f(z) = z.

[0125] According to a preferred embodiment of the present invention, the training process of the BP neural network regression model is as follows:

[0126] Step 1: Regress the droplet volume into the sample set D at a ratio of 4:1. r Divided into training set D re With test set D rt The learning rate γ = 0.1, the number of iterations n = 200, the number of nodes in the input layer P1 = 4, the number of nodes in the hidden layer P2 = 9, the number of nodes in the output layer P3 = 1, and the loss function is the mean squared error function.

[0127]

[0128] In the formula, k is the number of samples in the training set. For the predicted output of the i-th sample group, y [i] This is the actual output for the i-th sample group.

[0129] Step 2: Initialize all connection weights and biases in the network to 0.

[0130] Step 3: Add training set D re Using the sample features as input parameters, the forward propagation algorithm is applied to calculate the output of the j-th node in the hidden layer under the current weight conditions, according to the following formula. With the output α of the output layer node [2] :

[0131]

[0132]

[0133] Where, x i ν represents the input feature vector of the i-th node in the input layer. i,j ω represents the weights mapped from the i-th node in the input layer to the j-th node in the hidden layer; p,1 This represents the mapping weight from the p-th node in the hidden layer to the output layer node, where i represents an integer from 1 to 4; j and p represent integers from 1 to 9; ο [1] This represents the deviation in the mapping from the input layer to the hidden layer. [2] This represents the deviation in the mapping from the hidden layer to the output layer.

[0134] Step 4: Calculate the propagation error δ at all nodes under the current weight conditions using the following formula:

[0135]

[0136]

[0137]

[0138] in, y represents the propagation error of the output layer nodes, and y represents the actual droplet volume in the sample. ω represents the propagation error at the p-th node in the hidden layer. i,1 This represents the mapping weight from the i-th node in the hidden layer to the node in the output layer; ν represents the propagation error at the i-th node in the input layer. i,j This represents the mapping weight from the i-th node in the input layer to the j-th node in the hidden layer.

[0139] Step 5: Update all connection weights and biases according to the following formula:

[0140]

[0141]

[0142]

[0143]

[0144] Where γ represents the learning rate, This represents the gradient term that maps the weights from the i-th node in the input layer to the j-th node in the hidden layer. The gradient term represents the deviation in the mapping from the input layer to the hidden layer. This represents the gradient term of the weights mapped from the p-th node in the hidden layer to the output layer nodes. This represents the gradient term representing the deviation in the mapping from the hidden layer to the output layer.

[0145] Step Six: Repeat steps Three through Five above, continuously updating the connection weights and biases until the iteration termination condition is met, thus completing the construction of the BP neural network regression model. The predicted output of the obtained BP neural network regression model for sample x is:

[0146] y = xA1A2;

[0147]

[0148]

[0149] Where x is a 1×4 vector representing the set of input sediment droplet features; A1 is a 4×9 matrix representing the set of weights mapped from the input layer to the hidden layer; A2 is a 9×1 vector representing the set of weights mapped from the hidden layer to the output layer; and y is the predicted output of the regression model for sample x.

[0150] Step 7: Use a BP neural network regression model on the test set D rt Perform prediction verification and record the final prediction error range to complete the construction of the BP neural network regression model.

[0151] (2) After completing the construction of the random forest classification model and the BP neural network regression model, the system is applied to the printing interval stage in the actual printing process. After the test spray, the deposited droplets are detected to obtain the volume of the droplets, which provides a basis for the real-time adjustment of subsequent printing parameters.

[0152] As an optional process flow, such as Figure 5 The process flow for volume detection in the inkjet printing droplet volume intelligent detection method shown includes the following steps:

[0153] 1) Move the inkjet printing module 1 above the droplet deposition substrate 32 for test printing and record the theoretical landing position of the droplets ejected from each nozzle; then move the downward observation camera 31 above the deposited droplet corresponding to the first nozzle to be detected for observation, acquire the image of the deposited droplet and transmit it to the control module 5, and calculate the feature set of the deposited droplet corresponding to the current nozzle.

[0154] 2) Input the obtained set of sediment droplet features into the random forest classification model and output the droplet anomaly information. If there is anomaly in the sediment droplets, mark the current nozzle as an abnormal nozzle, directly proceed to the detection of the next nozzle and jump to step 1); if there is no anomaly in the sediment droplets, proceed to the next step.

[0155] 3) Input the obtained set of sediment droplet features into the BP neural network regression model, output the droplet volume and record it, and then perform the next nozzle detection;

[0156] 4) Repeat steps 1) to 3) until all nozzles are detected, output the status of all nozzles and the volume of the corresponding droplets, and complete the detection of inkjet printing droplet volume.

[0157] It should be noted that the following steps can be used to obtain the feature set of sedimentary droplets:

[0158] (1) Detection region segmentation: The detection region ROI of the i-th deposited droplet is divided according to the theoretical landing position of the droplet. i Perform coarse image segmentation;

[0159] (2) Image denoising: Due to environmental factors, there is some noise in the image, so image denoising algorithms are needed to denoise the ROI. i Noise reduction processing is performed;

[0160] (3) Subpixel image segmentation: Segmenting the denoised detection region ROI i Perform subpixel segmentation to obtain the droplet region Area i To achieve precise image segmentation;

[0161] (4) Calculate the droplet feature set: Use the corresponding algorithm to calculate the feature set of deposited droplets, such as the spreading diameter, number of droplets, and landing point deviation;

[0162] More specifically, the calculation method for the set of sedimentary droplet features in point (4) above is as follows: Figure 7 As shown, the data includes the spreading diameter D of the sedimentary droplets, the number of sedimentary droplets n, and the x-axis and y-axis deviations of the sedimentary droplets:

[0163] D = R1 + R2; n = num(Area) i ); Δx=x a -x o ; Δy = y a -y o ;

[0164] Where R1 and R2 represent the droplet regions Area, respectively. i The maximum inscribed circle radius and the minimum circumscribed circle radius; num(Area) i ) represents the droplet region Area i The number of x a y a These represent the droplet regions Area. i The x-coordinates and y-coordinates of the centroid; x o y oThese represent the x-coordinate and y-coordinate of the theoretical droplet landing point, respectively.

[0165] In summary, this invention proposes a method and system for intelligent detection of droplet volume in inkjet printing, comprising: performing a trial print to detect the feature set of deposited droplets corresponding to each nozzle, and using this as input parameters to establish an unlabeled classification dataset; labeling the unlabeled classification dataset based on the presence of abnormal droplet quantity and position to establish a labeled classification dataset, and training a random forest classification model using the labeled classification dataset; labeling normal samples in the labeled classification dataset using the flight droplet volume measurement results to establish a regression dataset, and then training a BP neural network regression model using the regression dataset; and using the trained random forest classification model and BP neural network regression model to calculate the droplet volume corresponding to any set of deposited droplet features. This invention effectively improves the efficiency of arrayed droplet volume detection in large-area inkjet printing production lines while ensuring detection accuracy, achieving efficient manufacturing.

[0166] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent detection of inkjet printing droplet volume, characterized in that, include: During the regular printing interval, move the printhead to the test printing area for a test print; Images of the sedimentation observation area are acquired to detect the sediment droplet feature set corresponding to each nozzle, and then input into a pre-trained machine learning classification model to obtain the judgment result of whether the nozzle is abnormal or not. Based on the judgment result, the sediment droplet feature set corresponding to each normal nozzle is input into a pre-trained machine learning regression model to obtain the droplet volume ejected from each normal nozzle, thereby realizing the array droplet volume measurement. The machine learning classification model is constructed as follows: a trial print is performed, the feature set of deposited droplets corresponding to each nozzle is detected and labeled as to whether the nozzle is abnormal or not, to obtain a classification sample set; the classification sample set is used to train a machine learning classification model for judging whether the nozzle is abnormal based on the feature set of deposited droplets. The machine learning regression model is constructed as follows: during trial printing, the volume of the flying droplets in each nozzle is measured; the flying droplet volume is labeled for the samples corresponding to the normal labels in the classification sample set to obtain a regression sample set; using the regression sample set, a machine learning regression model for calculating droplet volume based on the feature set of deposited droplets is trained. The machine learning classification model is a random forest classification model, which is obtained by combining multiple decision tree models based on a weighted voting method. In the process of constructing the random forest classification model, the weighted average of the classification accuracy of all decision tree models in the random forest classification model generated after each parameter adjustment is taken as the classification accuracy estimate of the random forest classification model. The random forest classification model for samples The predicted output is: ； ； in, This represents a sample of the feature set of a sedimentary droplet; It is a step function. It is the number of decision tree models. It is the first A decision tree model The weight, It is the first A decision tree model for The predicted output, Indicates taking Integers.

2. The intelligent detection method for inkjet printing droplet volume according to claim 1, characterized in that, The features included in the set of sedimentary droplet features are: the number of sedimentary droplets, the spreading diameter of sedimentary droplets, and the size of sedimentary droplets. Towards the point of impact, and the deposition of droplets Towards the landing point deviation; wherein, if the number of the deposited droplets is greater than or equal to two, then the spreading diameter, Towards landing point deviation and The deviation in the point of impact is a characteristic of the main sedimentary droplets.

3. The intelligent detection method for inkjet printing droplet volume according to claim 2, characterized in that, The criterion for determining whether the nozzle is abnormal is the number of deposited droplets. Towards landing point deviation and / or Deviation from the landing point; When there are no deposited droplets, or the number of deposited droplets is greater than or equal to two, The deviation from the landing point exceeds the preset value or If the deviation from the landing point exceeds a preset value, the corresponding set of sedimentary droplet features is marked as an abnormal sample.

4. The intelligent detection method for inkjet printing droplet volume according to claim 1, characterized in that, The classification accuracy estimate is expressed as: ； ； in, For the classification accuracy estimate, Indicates the number of decision tree models. Indicates the first The classification accuracy of a decision tree model. Indicates the first The number of normal samples correctly predicted by a decision tree model Indicates the first The number of outliers correctly predicted by each decision tree model Indicates the number of normal samples. Indicates the number of abnormal samples.

5. The intelligent detection method for inkjet printing droplet volume according to claim 1, characterized in that, The machine learning regression model is a backpropagation (BP) neural network regression model, comprising: an input layer, a hidden layer, and an output layer, wherein the input layer contains the number of nodes. In the formula The set of features of the sedimentary droplets contains the number of feature values ​​of the sedimentary droplets; no activation function is used in the input layer; the hidden layer contains the number of nodes. The activation function is Function; number of nodes in the output layer The activation function is a linear function, and its function expression is: In the formula For input variables.

6. An intelligent detection device for inkjet printing droplet volume, characterized in that, The method for performing an intelligent detection method for inkjet printing droplet volume as described in any one of claims 1 to 5 includes: an inkjet printing module, a first droplet deposition substrate, a deposited droplet observation module, a motion module, and a control module; The control module is used to control the inkjet printing module to move above the first droplet deposition substrate for test spraying via the motion module, and to control the deposition droplet observation module to move to the droplet deposition area via the motion module, to acquire images of the deposition droplets ejected from each nozzle and transmit them to the control module; the control module is also used to detect the deposition droplet feature set corresponding to each nozzle based on the deposition droplet images, and input it into a pre-trained machine learning classification model to obtain a judgment result on whether the nozzle is abnormal or not, and according to the judgment result, input the deposition droplet feature set corresponding to each normal nozzle into a pre-trained machine learning regression model to obtain the droplet volume ejected from each normal nozzle.

7. The intelligent inkjet printing droplet volume detection device according to claim 6, characterized in that, It also includes a flying droplet observation module and a second droplet deposition substrate; The control module is also used to control the inkjet printing module to perform trial printing on the second droplet deposition substrate, control the flying droplet observation module to acquire flying droplet images and transmit them to the control module, and control the deposition droplet observation module to acquire deposition droplet images on the second droplet deposition substrate and transmit them to the control module. The control module is also used to calculate the flying droplet volume of multiple nozzles based on the flying droplet images, detect the deposition droplet feature sets corresponding to the multiple nozzles based on the deposition droplet images, train the machine learning classification model using a classification sample set, and train the machine learning regression model using a regression sample set. The classification sample set is obtained by labeling multiple deposition droplet feature sets as abnormal or not, and the regression sample set is obtained by labeling the flying droplet volume of the deposition droplet feature sets labeled as normal in the classification sample set.

Images