CN116383725A - Intelligent screening method and device for arrayed spray holes for ink-jet printing - Google Patents
Intelligent screening method and device for arrayed spray holes for ink-jet printing Download PDFInfo
- Publication number
- CN116383725A CN116383725A CN202310251962.3A CN202310251962A CN116383725A CN 116383725 A CN116383725 A CN 116383725A CN 202310251962 A CN202310251962 A CN 202310251962A CN 116383725 A CN116383725 A CN 116383725A
- Authority
- CN
- China
- Prior art keywords
- deposited
- state parameter
- jet printing
- ink
- support vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 239000007921 spray Substances 0.000 title claims abstract description 135
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000012216 screening Methods 0.000 title claims abstract description 70
- 238000007641 inkjet printing Methods 0.000 title claims abstract description 58
- 238000013145 classification model Methods 0.000 claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 41
- 238000005507 spraying Methods 0.000 claims abstract description 11
- 239000007788 liquid Substances 0.000 claims description 112
- 238000012706 support-vector machine Methods 0.000 claims description 62
- 230000008021 deposition Effects 0.000 claims description 23
- 230000005856 abnormality Effects 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 17
- 239000000758 substrate Substances 0.000 claims description 14
- 238000002347 injection Methods 0.000 claims description 4
- 239000007924 injection Substances 0.000 claims description 4
- 239000010410 layer Substances 0.000 description 60
- 230000002159 abnormal effect Effects 0.000 description 26
- 238000007639 printing Methods 0.000 description 18
- 230000006870 function Effects 0.000 description 14
- 238000013507 mapping Methods 0.000 description 14
- 238000013528 artificial neural network Methods 0.000 description 13
- 230000009286 beneficial effect Effects 0.000 description 12
- 238000002372 labelling Methods 0.000 description 10
- 238000010801 machine learning Methods 0.000 description 10
- 238000010276 construction Methods 0.000 description 9
- 238000001514 detection method Methods 0.000 description 8
- 238000004519 manufacturing process Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 7
- 230000004913 activation Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 238000003064 k means clustering Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 239000002699 waste material Substances 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000008020 evaporation Effects 0.000 description 2
- 238000001704 evaporation Methods 0.000 description 2
- 238000012886 linear function Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 239000011368 organic material Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012827 research and development Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 239000002346 layers by function Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011165 process development Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 239000002904 solvent Substances 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B41—PRINTING; LINING MACHINES; TYPEWRITERS; STAMPS
- B41J—TYPEWRITERS; SELECTIVE PRINTING MECHANISMS, i.e. MECHANISMS PRINTING OTHERWISE THAN FROM A FORME; CORRECTION OF TYPOGRAPHICAL ERRORS
- B41J2/00—Typewriters or selective printing mechanisms characterised by the printing or marking process for which they are designed
- B41J2/005—Typewriters or selective printing mechanisms characterised by the printing or marking process for which they are designed characterised by bringing liquid or particles selectively into contact with a printing material
- B41J2/01—Ink jet
- B41J2/07—Ink jet characterised by jet control
- B41J2/12—Ink jet characterised by jet control testing or correcting charge or deflection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Ink Jet (AREA)
Abstract
The invention belongs to the technical field of novel display ink-jet printing, and particularly relates to a method and a device for screening arrayed spray holes for ink-jet printing, wherein the method comprises the following steps: based on the deposited droplet feature set obtained by each spraying of each spraying hole, sequentially passing through a classification model and a regression model to obtain a state parameter set corresponding to normal flying droplets; randomly selecting a plurality of state parameter sets, and marking by using labels to train to obtain an initial classifier; respectively inputting the rest state parameter sets into an initial classifier to obtain corresponding classification probability, selecting a plurality of candidate state parameter sets according to the classification probability, clustering to obtain labels of whether the spray holes corresponding to each clustering center are normal or not, and further training the initial classifier; and respectively inputting each state parameter set corresponding to each spray hole into a final classifier to determine whether the spray hole can be used for formal ink-jet printing. The invention improves the screening efficiency and ensures the screening precision.
Description
Technical Field
The invention belongs to the technical field of novel display ink-jet printing, and particularly relates to a method and a device for screening arrayed spray holes for ink-jet printing.
Background
The inkjet printing technology is an emerging display panel manufacturing process, mainly using solvents to melt organic materials, and then directly inkjet-printing the materials on a substrate to form an organic functional layer. Compared with the traditional evaporation process, the method can print according to the required amount of organic materials, thereby saving 90% of raw materials. Meanwhile, a vacuum environment and an expensive precise mask plate are not needed, any large-size substrate is allowed to be manufactured, the method is considered to be a technical revolution for replacing an evaporation process, and research and development of high-precision and high-adaptability industrial-grade ink-jet printing display manufacturing equipment is a hot spot field of research of panel manufacturers and research and development institutions at home and abroad.
Inkjet printing technology still faces many challenges as a manufacturing process for display panels, one of which is the screening of abnormal orifices in printing. In the process of preparing the OLED display device by ink-jet printing, due to the reasons of difference in manufacturing of a spray head, improper setting of technological parameters, relatively complex printing environment and the like, spray defects such as spray hole blockage, satellite liquid drops, volume abnormality and the like can occur, and the quality of a final display device can be seriously influenced by the abnormal spray holes involved in a printing link. Therefore, how to detect abnormal spray holes in the spray printing process is crucial to achieving high-quality and high-efficiency spray printing.
The existing screening method of the jet orifices of the jet printing mainly utilizes the observation of ink drops to directly acquire the volume and the speed of the liquid drops for judgment. The ink drop observation is mainly based on visual measurement, most of traditional visual measurement technologies are based on the principle of stroboscopic imaging, but along with the continuous rising of jet printing resolution and printing area, the area of a substrate is also increased, the number of spray holes to be detected is also increased greatly, the time for detecting the liquid drop parameters by using the ink drop observation is too long, the efficiency is low, a large amount of process time is occupied, and the efficiency of ink jet printing is seriously reduced.
Therefore, how to realize the high-efficiency intelligent detection and screening of the arrayed spray holes becomes a key point and a difficult point for the continuous development of the spray printing display technology.
Disclosure of Invention
Aiming at the defects and improvement demands of the prior art, the invention provides an arrayed spray hole screening method and device for ink jet printing, which aim to solve the problem of low spray hole screening efficiency caused by low observation efficiency of flying liquid drops in the prior art and realize rapid and efficient screening of the arrayed spray holes.
In order to achieve the above object, according to an aspect of the present invention, there is provided an intelligent screening method of arrayed nozzle holes for inkjet printing, comprising:
Based on the characteristic set of the deposited liquid drop obtained by each spray hole spraying, determining whether the deposited liquid drop sprayed by the spray hole is normal or not by adopting a trained classification model, and inputting the characteristic set of each normal deposited liquid drop into a trained regression model to obtain a state parameter set of a corresponding flying liquid drop;
randomly selecting a plurality of state parameter sets, and acquiring a label of whether the spray holes corresponding to each state parameter set are normal or not so as to train and obtain an initial classifier; respectively inputting the rest state parameter sets into an initial classifier to obtain corresponding classification probability, and selecting a plurality of candidate state parameter sets by adopting a sample selection strategy; clustering the plurality of candidate state parameter sets to obtain labels of whether the spray holes corresponding to each clustering center are normal or not, wherein the labels are used for further training the initial classifier to obtain a final classifier;
and respectively inputting each state parameter set corresponding to each spray hole into the final classifier, and determining whether the spray hole can be used for formal inkjet printing according to each classification result.
The beneficial effects of the invention are as follows: firstly, constructing a state parameter set sample set of flying liquid drops of each spray hole, specifically, utilizing a feature set of deposited liquid drops obtained by one-time spraying of each spray hole, firstly, adopting a classification model to screen a feature set corresponding to normal deposited liquid drops, and then adopting a trained regression model based on the feature set corresponding to normal deposited liquid drops to obtain the state parameter set (which can comprise the volume and the speed of the flying liquid drops) of the flying liquid drops corresponding to the normal deposited liquid drops, thereby constructing a state parameter set sample set. Then, the invention proposes to construct a classifier capable of accurately predicting whether the spray hole is normal or not based on the state parameter set by utilizing the constructed state parameter set sample set through a series of specific operations such as pre-training, sample selection, clustering, retraining and the like. Finally, a classifier is constructed by utilizing a state parameter set of a plurality of flying liquid drops corresponding to each spray hole to be screened through a plurality of spray, a prediction result of whether the corresponding spray holes are normal or not is obtained, and whether the spray holes can be screened for formal ink-jet printing is judged according to the plurality of prediction results corresponding to the spray holes to be screened. According to the invention, three machine learning models are introduced, intelligent screening of the spray holes is realized cooperatively, and compared with the existing method, the screening efficiency is greatly improved, wherein a series of specific training operations such as pre-training, sample selection, clustering, retraining and the like are also provided during construction of the classifier, and the screening precision is ensured while the screening efficiency is improved.
Further, a sample selection strategy in active learning is employed as follows:
in the method, in the process of the invention,
representing the remaining one state parameter set sample x without labels using the initial classifier i Predicted as category->
Probability of->
And->
Respectively representing negative example category and positive exampleA category; x is x * And representing the state parameter set with the smallest difference between the probabilities of being respectively predicted as two types in the rest state parameter sets as a candidate state parameter set.
The invention has the further beneficial effects that: the sample selection strategy adopted by the invention preferentially considers sample data which are easily judged to be positive examples and negative examples and are positioned at the class boundary of the sample space, and the invention selects the samples as candidate state parameter sets to further train the classifier, thereby greatly improving the prediction performance of the classifier and further being beneficial to improving the screening precision of the spray holes. Wherein, the positive example is the data of the state parameter within the threshold value, and the negative example is the state parameter exceeding the threshold value.
Further, a k-means algorithm is used to cluster the plurality of candidate state parameter sets.
The invention has the further beneficial effects that: the invention adopts the k-means clustering method to realize an unsupervised clustering algorithm, reduces the redundancy of information, can effectively reduce the labeling of a large number of samples of the same type, avoids the slow waste of iteration cost due to the movement of a classification hyperplane, and effectively improves the convergence speed and the classification accuracy of the algorithm.
Further, the criteria for determining whether each nozzle hole is available for inkjet printing are:
and when the classification results output by the final classifier based on the state parameter sets corresponding to the spray holes are all normal, screening the spray holes as available for ink jet printing.
The beneficial effects of the invention are as follows: if the sampled samples are all positive examples, judging that the spray holes are normal spray holes, and if negative examples appear in the samples, judging that the spray holes are abnormal spray holes, so that the screening precision is further improved.
Further, the classification model is composed of a plurality of trained support vector machines, wherein each feature dimension in the feature set corresponds to a support vector machine for predicting the probability of the deposited droplet having abnormality in the feature dimension in the classification model, and the classification model also comprises a support vector machine for predicting the probability of the deposited droplet being normal;
the specific way of using the classification model is:
inputting each feature set into each support vector machine in the classification model, and predicting and outputting a probability by each support vector machine; and taking whether deposited liquid drops corresponding to the maximum probability are normal or not as an output result of the classification model.
The invention has the further beneficial effects that: aiming at the prediction of whether deposited liquid drops are normal or not, the invention adopts a multi-classification support vector machine. Each feature dimension in the feature set corresponds to a support vector machine for predicting the probability of the deposited droplet having an abnormality in the feature dimension and a support vector machine for predicting the probability of the deposited droplet being normal in the classification model. Each support vector machine is specially used for predicting the probability of abnormality of the deposited liquid drop in a certain dimension, instead of uniformly predicting the abnormal deposited liquid drop, the phenomenon that the abnormal deposited liquid drop is misjudged as a normal deposited liquid drop is avoided, the judging precision of whether the deposited liquid drop is normal or not is greatly improved, and the classifying precision is greatly improved.
Further, each support vector machine in the classification model is obtained by adopting the following training mode:
randomly selecting a plurality of feature set samples from a feature set sample set, and acquiring labels of whether deposited liquid drops corresponding to each feature set sample are normal or not or whether the deposited liquid drops have abnormality in a feature dimension according to a prediction target of the support vector machine so as to train and obtain an initial support vector machine; inputting the rest feature set samples into the initial support vector machine to obtain corresponding prediction probabilities so as to construct a plurality of candidate feature set samples by adopting a sample selection strategy; clustering a plurality of candidate feature set samples to determine each clustering center; and acquiring labels of whether the deposited liquid drops corresponding to each cluster center are normal or not or whether the deposited liquid drops have abnormality in a certain characteristic dimension according to the prediction targets, and further training the initial support vector machine to obtain a final support vector machine for realizing the prediction targets.
The invention has the further beneficial effects that: when each support vector machine is constructed, the invention provides a series of specific training operations such as pre-training, sample selection, clustering, retraining and the like, and improves the classification precision of the support vector machine.
Further, when training the support vector machine in the classification model, a sample selection strategy in the following active learning is adopted:
wherein X is a sample set of the remaining feature set samples, p (y B |x i ) And p (y) SB |x i ) For initial support vector machine pair sample x i The two highest probability values obtained are determined.
The invention has the further beneficial effects that: according to the sample selection strategy adopted by the invention, the sample data at the class boundary of the sample space is preferentially considered, and the samples are selected to serve as candidate deposited droplet feature sets to further train the support vector machine, so that the prediction performance of the support vector machine can be greatly improved.
Further, when the support vector machine in the classification model is trained, the clustering method adopted is a k-means algorithm.
The invention has the further beneficial effects that: the invention adopts the k-means clustering method to realize an unsupervised clustering algorithm, reduces the redundancy of information, can effectively reduce the labeling of a large number of samples of the same type, avoids the slow waste of iteration cost due to the movement of a classification hyperplane, and effectively improves the convergence speed and the classification accuracy of the algorithm.
Further, the feature dimensions included in the feature set are: whether the number of deposited droplets is 0, whether the number of deposited droplets is greater than 1, whether there is a landing deviation of deposited droplets, and whether the deposited droplet diameter exceeds a threshold.
The invention has the further beneficial effects that: according to the method, the number of the deposited droplets, the deviation of the landing points of the deposited droplets, whether the diameter of the deposited droplets exceeds a threshold value and other multidimensional degrees are adopted to mark the abnormal deposited ink droplets, so that the classification precision of a sample set can be improved, the influence of the abnormal deposited ink droplet parameters on the prediction precision of a regression model is avoided, and the screening of spray holes is finally influenced.
The invention also provides an intelligent screening device of the arrayed spray holes for ink-jet printing, which comprises the following steps: the device comprises an ink jet printing module, a liquid drop deposition substrate, a deposited liquid drop observation module, a movement module and a control module;
the control module is used for controlling the ink jet printing module to move above the droplet deposition substrate to perform trial injection through the motion module, controlling the deposition droplet observation module to move to a droplet deposition area through the motion module, collecting deposition droplet images sprayed by all spray holes and transmitting the deposition droplet images to the control module; the control module is also used for detecting and obtaining a characteristic set of deposited liquid drops corresponding to each spray hole based on the deposited liquid drop image, and executing the intelligent screening method of the arrayed spray holes for ink-jet printing.
The beneficial effects of the invention are as follows: the intelligent screening device for the arrayed spray holes for the ink-jet printing comprises an ink-jet printing module, a liquid drop deposition substrate, a deposited liquid drop observation module and a motion module, wherein the deposited liquid drop image acquisition is cooperatively realized, and the control module is used for processing the image to obtain a deposited liquid drop characteristic set, so that the control module can adopt the intelligent screening method for the arrayed spray holes for the ink-jet printing to realize high-precision and high-efficiency spray hole screening.
In general, through the above technical solutions conceived by the present invention, the following beneficial effects can be obtained:
according to the invention, three machine learning models are introduced, intelligent screening of the spray holes is realized cooperatively, and compared with the existing method, the screening efficiency is greatly improved. When the classifier is constructed, the invention also provides a series of specific training operations such as pre-training, sample selection, clustering, retraining and the like, and the screening precision is effectively ensured while the screening efficiency is improved.
Drawings
FIG. 1 is a flow chart of an intelligent screening method of an arrayed spray orifice for ink-jet printing according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of acquisition of a feature set of deposited droplets according to an embodiment of the present invention;
FIG. 3 is a process flow diagram of a nozzle screening stage in an intelligent screening method for arrayed nozzles of an inkjet printing system according to an embodiment of the present invention;
FIG. 4 is a flow chart of construction and application of a double-layer proxy model in an intelligent screening method for arrayed spray holes of an ink jet printing system according to an embodiment of the invention;
FIG. 5 is a block diagram of a machine learning regression model provided by an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
Example 1
An intelligent screening method of arrayed spray holes for ink-jet printing, as shown in figure 1, comprises the following steps:
based on the characteristic set of the deposited liquid drop obtained by each spray hole spraying, determining whether the deposited liquid drop sprayed by the spray hole is normal or not by adopting a trained classification model, and inputting the characteristic set of each normal deposited liquid drop into a trained regression model to obtain a state parameter set of a corresponding flying liquid drop;
Randomly selecting a plurality of state parameter sets, and acquiring a label of whether the spray holes corresponding to each state parameter set are normal or not so as to train and obtain an initial classifier; respectively inputting the rest state parameter sets into an initial classifier to obtain corresponding classification probability, and selecting a plurality of candidate state parameter sets by adopting a sample selection strategy; clustering the plurality of candidate state parameter sets, determining each clustering center, and acquiring a label of whether the spray holes corresponding to each clustering center are normal or not, so as to further train an initial classifier and obtain a final classifier;
and respectively inputting each state parameter set corresponding to each spray hole into the final classifier, and determining whether the spray hole can be used for formal inkjet printing according to each classification result.
It should be noted that the state parameter set may include a volume and a speed.
Firstly, a state parameter set sample set of flying liquid drops of each spray hole is constructed, specifically, a characteristic set of deposited liquid drops obtained by spraying each spray hole at one time is firstly adopted to screen a characteristic set corresponding to normal deposited liquid drops, and then a trained regression model is adopted to obtain the state parameter set (which can comprise the volume and the speed of the flying liquid drops) of the flying liquid drops corresponding to the normal deposited liquid drops based on the characteristic set corresponding to the normal deposited liquid drops, so that the state parameter set sample set is constructed. Then, the invention proposes to construct a classifier capable of accurately predicting whether the spray hole is normal or not based on the state parameter set by utilizing the constructed state parameter set sample set through a series of specific operations such as pre-training, sample selection, clustering, retraining and the like. Finally, a classifier is constructed by utilizing a state parameter set of a plurality of flying liquid drops corresponding to each spray hole to be screened through a plurality of spray, a prediction result of whether the corresponding spray holes are normal or not is obtained, and whether the spray holes can be screened for formal ink-jet printing is judged according to the plurality of prediction results corresponding to the spray holes to be screened. In the embodiment, three machine learning models are introduced to cooperatively realize intelligent screening of the spray holes, and compared with the existing method, the screening efficiency is greatly improved. The embodiment also provides a series of specific training operations such as pre-training, sample selection, clustering, retraining and the like when constructing the classifier, and the screening precision is ensured while the screening efficiency is improved.
The sample selection strategy described above may employ sample selection strategies in active learning, including BvSB-based sample selection strategies and uncertainty-based sampling strategies. As a preferred embodiment, the present example employs the following BvSB based sample selection strategy:
in the method, in the process of the invention,
representing the remaining one state parameter set sample x without labels using the initial classifier i Predicted as category->
Probability of->
And->
Respectively representing a negative example category and a positive example category; x is x * And representing the state parameter set with the smallest difference between the probabilities of being respectively predicted as two types in the rest state parameter sets as a candidate state parameter set.
The adopted sample selection strategy prioritizes sample data which are easily judged as positive examples and negative examples and are positioned at the class boundary of the sample space, and the embodiment selects the samples as candidate state parameter sets to further train the classifier, so that the prediction performance of the classifier can be greatly improved, and further the improvement of the screening precision of the spray holes is facilitated. Wherein, the positive example is the data of the state parameter within the threshold value, and the negative example is the state parameter exceeding the threshold value.
As a preferred embodiment, a k-means algorithm is used to cluster a plurality of candidate state parameter sets.
Specifically, t unlabeled samples are selected from the unlabeled sample set X
As candidate sample set, then in candidate sample set +.>
The k (k) is initialized by the k-means algorithm<t) clustering centers, u j J=1, 2,..k, one cluster c for each cluster center j J=1, 2,..k. Calculating the respective samples x in each cluster currently i Distance to cluster center of each cluster, then x i Dividing into clusters closest to each other, determining the cluster center again after dividing all samples, repeating the above two steps until the distance between the samples in each cluster and the sample in the cluster center is not reduced, and returning to the final clustering result. For the k clustering results, sample submission labels of the clustering centers are selected to obtain labels for the clusters. Selecting unlabeled samples based on a difference criterion, and performing an unsupervised clustering algorithm>
Determining a cluster to which the sample belongs, wherein u j C is the cluster center (i) For sample x j The cluster closest to the k clusters.
A sample selection strategy based on BvSB is combined with a k-means clustering method, so that an unsupervised clustering algorithm is realized, and information redundancy is reduced. If a large number of samples of the same type are marked, the classification hyperplane moves slowly, the iteration cost is wasted, the convergence speed of the algorithm is slow, and the classification accuracy is affected.
As a preferred embodiment, the criteria for determining whether each nozzle hole is available for inkjet printing are:
and when the classification results output by the final classifier based on the state parameter sets corresponding to the spray holes are all normal, screening the spray holes as available for ink jet printing.
If the sampled samples are all positive examples, judging that the spray holes are normal spray holes, and if negative examples appear in the samples, judging that the spray holes are abnormal spray holes, so that the screening precision is further improved.
The classification model is composed of a plurality of trained support vector machines, wherein each feature dimension in the feature set corresponds to a support vector machine for predicting the probability of the deposited droplet having abnormality in the feature dimension in the classification model, and the classification model also comprises a support vector machine for predicting the probability of the deposited droplet being normal; the specific way of using the classification model is: inputting each feature set into each support vector machine in the classification model, and predicting and outputting a probability by each support vector machine; and taking whether deposited liquid drops corresponding to the maximum probability are normal or not as an output result of the classification model.
Aiming at the prediction of whether deposited liquid drops are normal or not, the invention adopts a multi-classification support vector machine. Each feature dimension in the feature set corresponds to a support vector machine for predicting the probability of the deposited droplet having an abnormality in the feature dimension and a support vector machine for predicting the probability of the deposited droplet being normal in the classification model. Each support vector machine is specially used for predicting the probability of abnormality of the deposited liquid drop in a certain dimension, instead of uniformly predicting the abnormal deposited liquid drop, the phenomenon that the abnormal deposited liquid drop is misjudged as a normal deposited liquid drop is avoided, the judging precision of whether the deposited liquid drop is normal or not is greatly improved, and the classifying precision is greatly improved.
The classification model adopted in the embodiment is an active learning classification model, and specifically includes: information entropy-based active learning model of improved strategy, wherein an initial model is a plurality of support vector machine models, modeling is carried out through five components, and A= (G, U) m ,T,Q,U n ) Wherein G represents a classifier model, Q represents a sample selection strategy, U m Representing marked sample set, U n And (5) representing unlabeled sample sets, and T representing a manual annotator. The initial model selected by the classification model as the support vector machine model can be expressed as: y (x) =sign (w) t Φ (x) +b); where Φ (x) is a kernel function, w is a function control coefficient, b is an unknown constant, and t is a coefficient control factor.
As a preferred embodiment, each support vector machine in the classification model is obtained by adopting the following training mode:
randomly selecting a plurality of feature set samples from a feature set sample set, and acquiring labels of whether deposited liquid drops corresponding to each feature set sample are normal or not or whether the deposited liquid drops have abnormality in a feature dimension according to a prediction target of the support vector machine so as to train and obtain an initial support vector machine; inputting the rest feature set samples into the initial support vector machine to obtain corresponding prediction probabilities so as to construct a plurality of candidate feature set samples by adopting a sample selection strategy; clustering a plurality of candidate feature set samples to determine each clustering center; and acquiring labels of whether the deposited liquid drops corresponding to each cluster center are normal or not or whether the deposited liquid drops have abnormality in a certain characteristic dimension according to the prediction targets, and further training the initial support vector machine to obtain a final support vector machine for realizing the prediction targets.
When each support vector machine is constructed, the embodiment provides a series of specific training operations such as pre-training, sample selection, clustering, retraining and the like, and can improve the classification precision of the support vector machine.
As a preferred embodiment, in training the support vector machine in the classification model, a sample selection strategy in active learning is adopted as follows:
wherein X is a sample set of the remaining feature set samples, p (y B |x i ) And p (y) SB |x i ) For initial support vector machine pair sample x i The two highest probability values obtained are determined.
The sample selection strategy is adopted to prioritize sample data at the class boundary of the sample space, and the invention selects the samples as candidate deposited droplet feature sets to further train the support vector machine, so that the prediction performance of the support vector machine can be greatly improved.
As a preferred embodiment, the clustering method used in training the support vector machine in the classification model is a k-means algorithm.
By adopting a method based on BvSB strategy and combining k-means clustering, an unsupervised clustering algorithm is realized, information redundancy is reduced, labeling of a large number of samples of the same type can be effectively reduced, slow waste of iteration cost due to super-planar movement of classification is avoided, and the convergence speed and classification accuracy of the algorithm are effectively improved.
As a preferred embodiment, the feature dimensions included in the feature set include: whether the number of deposited droplets is 0, whether the number of deposited droplets is greater than 1, whether there is a landing deviation of deposited droplets, and whether the deposited droplet diameter exceeds a threshold.
The abnormal deposited ink drops are marked by adopting multiple dimensions, so that the classification precision of a sample set can be improved, the influence of abnormal deposited ink drop parameters on the prediction precision of a regression model is avoided, and the screening of spray holes is finally influenced.
Here, regarding the situation that the volume and the speed of the flying droplet are mapped according to the feature set of the deposited droplet in the regression model, the regression construction process is shown by using an approximate formula:
the process of volume mapping is exemplified by the same speed, the actual mapping of the volume is the area of deposited ink drops, the speed corresponds to the deviation of the landing points of the deposited ink drops, and the mapping relation corresponds to the following formula:
y＝xA 1 A 2 ；
wherein x is a vector with length of 1×4, and represents an input deposited droplet feature set, which is an input sample; a is that 1 A matrix of 4 x 9 representing a set of input layer to hidden layer mapping weights; a is that 2 A 9×1 vector representing a set of hidden layer-to-output layer mapping weights; y is the predicted output of the regression model on the sample x, v i,j Representing that the ith node of the input layer maps weight values to the jth node of the hidden layer; omega p,1 Representing hidden layer p-th node to output layer node mappingA weight; i represents an integer of 1 to 4; j. p represents an integer of 1 to 9.
For more clear description of the method of the present invention, the intelligent screening method of this embodiment may be described as including: the method comprises a model construction stage, a droplet parameter detection stage and a spray hole parameter acquisition and screening stage. The model building stage includes operations S1 to S6, the droplet parameter detecting stage includes operation S7, and the nozzle parameter obtaining and screening stage includes operation S8.
Model construction stage:
the method comprises the following steps of S1, performing trial printing, detecting a deposited droplet feature set corresponding to each spray hole, and establishing a label-free classification data set by taking the deposited droplet feature set as an input parameter;
s2, labeling samples in the label-free classified data set according to the abnormal number and the abnormal position of the liquid drops, and establishing a label-free classified data set;
s3, training a machine learning classification model by using the labeled classification data set to obtain the machine learning classification model for judging whether the liquid drop abnormality exists or not based on the deposited liquid drop feature set;
s4, marking normal samples in the labeled classified data set by using the flying droplet volume and speed measurement results, and establishing a regression data set;
S5, training a machine learning regression model by using the regression data set to obtain a machine learning regression model for calculating the volume of the liquid drop based on the deposited liquid drop characteristic set;
s6, establishing a double-layer proxy model by using the established classification model and regression model;
a liquid drop parameter detection stage:
and S7, before formal printing, moving the spray head to a test printing area to perform test printing and spraying, detecting a deposited liquid drop characteristic set corresponding to each spray hole, inputting the deposited liquid drop characteristic set into the double-layer proxy model, and calculating the liquid drop volume and the liquid drop speed corresponding to the deposited liquid drop characteristic set.
And S8, obtaining the speed and volume parameters of the liquid drops, and obtaining the overall state of each spray hole by using a sampling method based on active learning to screen the spray holes.
The BP neural network regression model can be used as the regression model, and specifically includes: an input layer, a hidden layer and an output layer, wherein the input layer comprises a node number P 1 M, where M is the number of deposited droplet features, the input layer does not use an activation function; the hidden layer contains the node number P 2 ＝2P 1 +1, the activation function is:
wherein x is p For the p-th input sample, c i For the i-th center point, h is the number of nodes of the hidden layer, and n is the number of output samples. The output layer contains the node number P 3 The activation function is a linear function, and the function formula is f (x) =x. The predicted output of the RBF neural network regression model for sample x is:
wherein x is p For the p-th input sample, c i For the i-th center point, h is the number of nodes of the hidden layer, and n is the number of output samples.
After the flight droplet state parameter set is obtained by using the double-layer proxy model, a large number of parameter sets corresponding to each spray hole are required to be processed so as to predict the condition of the spray hole, and a sampling method based on active learning is adopted, and the method specifically comprises the following steps: training a classifier C by using a labeled classification sample set T, using a random forest classifier, evaluating the importance of unlabeled samples by using a query strategy in active learning, selecting the most valuable samples, and further training the classifier.
That is, the classifier training method adopted in this embodiment is: the method is characterized in that an active learning mode is adopted, the active learning process is a continuous iterative training process of a classifier, a sample set U1 is a marked sample, a sample set U2 is an unmarked sample, a model G is firstly trained by the U1, the model G is utilized to select the U2, the most suitable sample is selected according to a sample selection strategy and is delivered to expert marks, and each time, the sample which can improve the performance of the model is selected.
In general, the intelligent screening method for the spray holes of the embodiment includes: performing trial printing, detecting a deposited droplet feature set corresponding to each spray hole, and establishing a label-free classification data set by taking the deposited droplet feature set as an input parameter; labeling the unlabeled classified data sets based on the condition that whether liquid drops exist, the quantity, the positions are abnormal and the diameters are abnormal, establishing the labeled classified data sets, training an active learning model by using the labeled classified data sets, selecting unlabeled sample sets by using the model, finding out the most suitable sample according to a sample selection strategy, and marking by an expert, wherein each time the sample which can improve the performance of the model is selected for model training, and the expert can obtain a better model through a small amount of work; labeling normal samples in the labeled classified data set by using the flying droplet volume and speed measurement result, establishing a regression data set, and training a BP neural network regression model by using the regression data set; and establishing a double-layer agent model by using the trained active learning classification model and the BP neural network regression model, and calculating the droplet volume and the droplet speed corresponding to any deposited droplet feature set. The invention constructs a framework for applying a machine learning model to spray state prediction of spray holes and spray hole screening, and obtains a prediction model with better performance by applying the active learning method with smaller data marking cost, thereby effectively improving the efficiency of array spray hole screening in a large-area spray printing display production line and realizing high-efficiency intelligent production and manufacture under the condition of ensuring detection precision.
Example two
An intelligent screening device for arrayed spray holes for ink-jet printing, comprising: the device comprises an ink jet printing module, a liquid drop deposition substrate, a deposited liquid drop observation module, a movement module and a control module;
the control module is used for controlling the ink jet printing module to move to the position above the droplet deposition substrate to perform trial injection through the movement module, controlling the deposition droplet observation module to move to the droplet deposition area through the movement module, collecting deposition droplet images sprayed by all spray holes and transmitting the deposition droplet images to the control module; the control module is further configured to detect and obtain a feature set of the deposited droplet corresponding to each nozzle based on the deposited droplet image, and execute an intelligent screening method of the arrayed nozzle for inkjet printing according to the first embodiment.
The intelligent screening device for the arrayed spray holes for the ink-jet printing comprises an ink-jet printing module, a liquid drop deposition substrate, a deposited liquid drop observation module and a motion module, wherein the deposited liquid drop image acquisition is cooperatively realized, and the control module is used for processing the image to obtain a deposited liquid drop characteristic set, so that the control module can adopt the intelligent screening method for the arrayed spray holes for the ink-jet printing to realize high-precision and high-efficiency spray hole screening.
For a better illustration of the present invention, the following specific detection system will be given to further describe the detection procedure described in embodiment one:
fig. 2 is a schematic diagram of acquiring a feature set of a deposited droplet, which clearly shows the deposition condition of a droplet on a substrate, and has a droplet that is normally deposited and a droplet that is abnormally deposited, including undeposited and abnormally deposited, and selects a suitable droplet for measuring relevant parameters, including the diameter of the deposited droplet and the deviation of the deposited droplet from a theoretical landing point, to acquire the relevant parameters of the feature set of the deposited droplet.
Further, fig. 3 is a process flow chart of volume and speed detection in the intelligent screening method for spray holes according to the embodiment of the invention, which specifically includes the following steps:
(1) Performing trial injection, recording the theoretical drop point position of the injected liquid drop of each spray hole, acquiring a deposited liquid drop image through a downward-looking observation camera, and acquiring a characteristic set of the deposited liquid drop corresponding to the current spray hole;
(2) Inputting the obtained deposited droplet feature set into an active learning classification model, outputting droplet abnormality information, marking the current spray hole as an abnormal spray hole if the deposited droplet is abnormal, directly entering the detection of the next spray hole, and jumping to the step (1); if the deposited liquid drop is not abnormal, entering the next step;
(3) Inputting the obtained deposited droplet feature set into a BP neural network regression model, outputting and recording the volume and the speed of the droplet, judging whether the spray hole is abnormal or not by using a sampling detection method based on active learning, and detecting the next spray hole;
(4) And (3) repeating the steps (1) to (3) until all the spray holes are detected, outputting the volumes and the speeds of the liquid drops corresponding to all the spray holes, judging the state of the spray holes, and finishing the screening of the spray holes for ink-jet printing.
Further, fig. 4 is a flow chart of construction and application of a double-layer proxy model in the intelligent screening method for spray holes, which specifically includes the following steps:
(1) And moving the spray head module to an ink drop observation position, carrying out flash spraying on the spray head, observing, acquiring a flying liquid drop image through a side-looking observation camera, calculating to obtain the liquid drop volume and the speed corresponding to each spray hole, and recording.
(2) And (3) moving the spray head module to the upper part of the liquid drop deposition substrate for test spraying, recording the theoretical landing point position of the liquid drop sprayed by each spray hole, moving the downward-looking observation camera to the upper part of the liquid drop deposition substrate for observation, acquiring a deposited liquid drop image, and calculating to acquire a deposited liquid drop feature set (d, m, delta x and delta y) corresponding to each spray hole, wherein d represents the spreading diameter of the deposited liquid drop, m represents the number of the deposited liquid drops, delta x represents the x-direction landing point deviation of the deposited liquid drop, delta y represents the y-direction landing point deviation of the deposited liquid drop, extracting a certain number of samples from all the deposited liquid drop feature sets, and establishing a label-free classification data set theta by taking the sample as an input parameter.
According to a preferred embodiment of the invention, the unlabeled classified data set θ has a capacity of N H =3000, the constructed unlabeled classification dataset can be expressed as:
where (d, m, Δx, Δy) represents the deposited droplet feature set for each group of samples.
(3) Labeling the sample in the label-free classified data set theta based on the abnormal number and the abnormal position of the liquid drops,labeling the sample of the unlabeled classified data set as an abnormal sample when the number of the unlabeled droplets or the deposited droplets is more than or equal to two or the deviation of the x/y falling points exceeds a preset value, otherwise labeling the sample of the unlabeled classified data set as a normal sample, and establishing a labeled classified data set D c ：
Wherein y represents a sample label, and the value is 0 or 1; when y=0, it indicates that there is an abnormality in the deposited droplet corresponding to the sample, and the sample belongs to an abnormal sample; when y=1, it indicates that there is no abnormality in deposited droplets corresponding to the sample, and the sample belongs to a normal sample.
(4) Classifying data sets D using labeled c Training a random active learning classification model, inputting the model into a deposited droplet feature set, and outputting the model into whether deposited droplets are abnormal or not;
further, the machine learning classification model is an active learning classification model, and specifically includes: an active learning model of an improvement strategy based on information entropy, wherein an initial model is a support vector machine model, and modeling is carried out through five components:
A＝(G,U m ,T,Q,U n )；
Wherein G represents a classifier model, Q represents a sample selection strategy, U m Representing marked sample set, U n And (5) representing unlabeled sample sets, and T representing a manual annotator.
The initial model selected by the active learning classification model is a support vector machine model, and can be expressed as:
y(x)＝sign(w t Φ(x)+b)；
where Φ (x) is a kernel function, w is a function control coefficient, b is an unknown constant, and t is a coefficient control factor.
Further, according to the active learning classification model, the sample strategy selection method comprises the following steps:
wherein X is an unlabeled sample set, p (y B |x i ) And p (y) SB |x i ) For classifier on sample x i Judging the two highest probability values, if the uncertainty of one sample is higher, the C (x i ) The smaller the value.
According to a preferred embodiment of the invention, the training process of the active learning classification model is as follows:
selecting t unlabeled samples from the unlabeled sample set X
As candidate sample set, then in candidate sample set +.>
The k (k) is initialized by the k-means algorithm<t) clustering centers, u j J=1, 2,..k, one cluster c for each cluster center j J=1, 2,..k. D samples x for each candidate sample set i Calculate its distance to each cluster center and then let x i Dividing into clusters with the smallest cluster, calculating the closest cluster in the sample and k clusters, determining the cluster center again after dividing all samples, repeating the two steps until the distance between the sample in each cluster and the sample in the cluster center is not reduced, and returning to the final clustering result. For the k clustering results, a sample submission annotation in which the clustering center is selected.
(5) Classifying data sets D with labels by flying drop volume and speed c Labeling normal samples in the liquid drop volume regression data set D is established r ：
Where V represents the actual volume of the sample label, in particular the droplet; v (V) e Representing sample labels, particularly dropActual speed; p (P) h Representing a labeled classification dataset D c Is a total number of normal samples.
(6) Regression data set D using drop volumes r And training a BP neural network regression model, wherein the model is input into a deposited droplet characteristic set, and the model is output into droplet volume and velocity.
More specifically, fig. 5 is a schematic structural diagram of a BP neural network regression model according to an embodiment of the present invention, including: an input layer, a hidden layer, and an output layer, wherein: the input layer contains the node number P 1 M, where M is the number of deposited droplet features, the input layer does not use an activation function; the hidden layer contains the node number P 2 ＝2P 1 +1, the activation function is:
wherein x is p For the p-th input sample, c i For the i-th center point, h is the number of nodes of the hidden layer, and n is the number of output samples.
The output layer contains the node number P 3 The activation function is a linear function, and the function formula is f (x) =x.
Further, the prediction output of the BP neural network regression model for the sample x is:
wherein x is p For the p-th input sample, c i For the i-th center point, h is the number of nodes of the hidden layer, and n is the number of output samples.
According to a preferred embodiment of the invention, the training process of the BP neural network regression model is as follows:
step one, according to 5:1 to regress drop volumes to data set D r Is divided into training sets D re And test set D rt The learning rate gamma=0.1, the iteration number n=300, and the input layer contains the node number P 1 =5, the hidden layer contains the number P of nodes 2 The output layer contains the node number P =11 3 =1, the loss function is a mean squared error function:
where k is the number of training set samples,
predicting output for the ith group of samples, y [i] The actual output is for the i-th set of samples.
And step two, initializing all connection weights and deviations in the network to 0.
Step three, training set D re The sample characteristics in (a) are used as input parameters, and the output of the j-th node in the hidden layer under the current weight condition is calculated by using a forward propagation algorithm according to the following formula
Output alpha from output layer node [2] ：
Wherein x is i Input feature vector, v, representing the i-th node of the input layer i,j Representing that the ith node of the input layer maps weight values to the jth node of the hidden layer; omega p,1 Mapping weights from the p-th node of the hidden layer to the nodes of the output layer, wherein i represents an integer from 1 to 5; j. p represents an integer of 1 to 11; o (O) [1] Representing the deviation of the mapping of the input layer to the hidden layer, O [2] Indicating the deviation of the hidden layer to output layer mapping.
Step four, calculating propagation errors delta at all nodes under the current weight condition according to the following formula:
wherein, the liquid crystal display device comprises a liquid crystal display device,
representing the propagation error of the output layer node, and y represents the actual liquid drop volume and speed in the sample; />
Representing propagation error, ω, at the p-th node in the hidden layer i,1 Representing the mapping weight value from the ith node of the hidden layer to the node of the output layer; />
Representing propagation error, v at the i-th node in the input layer i,j And the mapping weight value of the ith node of the input layer to the jth node of the hidden layer is represented.
Updating all connection weights and deviations according to the following formula:
Wherein, gamma represents the learning rate,
gradient item representing the mapping weight of the ith node in the input layer to the jth node in the hidden layer,/->
Gradient term representing the deviation of the mapping of the input layer to the hidden layer,/->
Gradient item representing mapping weight of p-th node in hidden layer to output layer node, ++>
Gradient terms representing the bias of the hidden layer mapping to the output layer.
And step six, repeating the steps three to five, and continuously updating the connection weight and the deviation until the iteration termination condition is met, thereby completing the construction of the BP neural network regression model.
Step seven, using BP neural network regression model to test set D rt And carrying out prediction verification, recording a final prediction error range, and completing construction of a BP neural network regression model.
After the construction of the active learning classification model and the BP neural network regression model is completed, the system acts on a printing interval stage (test printing stage) in the actual printing process, and deposited liquid drops are detected after test spraying to obtain the volumes and the speeds of the liquid drops for spray hole screening.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. An intelligent screening method of arrayed spray holes for ink-jet printing is characterized by comprising the following steps:
based on the characteristic set of the deposited liquid drop obtained by each spray hole spraying, determining whether the deposited liquid drop sprayed by the spray hole is normal or not by adopting a trained classification model, and inputting the characteristic set of each normal deposited liquid drop into a trained regression model to obtain a state parameter set of a corresponding flying liquid drop;
randomly selecting a plurality of state parameter sets, and acquiring a label of whether the spray holes corresponding to each state parameter set are normal or not so as to train and obtain an initial classifier; respectively inputting the rest state parameter sets into an initial classifier to obtain corresponding classification probability, and selecting a plurality of candidate state parameter sets by adopting a sample selection strategy; clustering the plurality of candidate state parameter sets to obtain labels of whether the spray holes corresponding to each clustering center are normal or not so as to further train an initial classifier and obtain a final classifier;
and respectively inputting each state parameter set corresponding to each spray hole into the final classifier, and determining whether the spray hole can be used for formal inkjet printing according to each classification result.
2. The intelligent screening method of arrayed spray holes for ink-jet printing according to claim 1, wherein the following sample selection strategy in active learning is adopted:
In the method, in the process of the invention,
representing the remaining one state parameter set sample x without labels using the initial classifier i Predicted as category->
Probability of->
And->
Respectively representing a negative example category and a positive example category; x is x * And representing the state parameter set with the smallest difference between the probabilities of being respectively predicted as two types in the rest state parameter sets as a candidate state parameter set.
3. The intelligent screening method of arrayed spray orifices for ink-jet printing according to claim 1, wherein a k-means algorithm is adopted to cluster a plurality of candidate state parameter sets.
4. The intelligent screening method of arrayed nozzle orifices for ink jet printing according to claim 1, wherein the criteria for determining whether each nozzle orifice is available for ink jet printing is:
and when the classification results output by the final classifier based on the state parameter sets corresponding to the spray holes are all normal, screening the spray holes as available for ink jet printing.
5. The intelligent screening method of the arrayed spray holes for the ink-jet printing according to claim 1, wherein the classification model is composed of a plurality of trained support vector machines, wherein each feature dimension in the feature set corresponds to a support vector machine for predicting the probability of the deposited liquid drop having abnormality on the feature dimension in the classification model, and the classification model also comprises a support vector machine for predicting the probability of the deposited liquid drop being normal;
The specific way of using the classification model is:
inputting each feature set into each support vector machine in the classification model, and predicting and outputting a probability by each support vector machine; and taking whether deposited liquid drops corresponding to the maximum probability are normal or not as an output result of the classification model.
6. The intelligent screening method of the arrayed spray holes for the ink-jet printing according to claim 5, wherein each support vector machine in the classification model is obtained by adopting the following training mode:
randomly selecting a plurality of feature set samples from a feature set sample set, and acquiring labels of whether deposited liquid drops corresponding to each feature set sample are normal or not or whether the deposited liquid drops have abnormality in a feature dimension according to a prediction target of the support vector machine so as to train and obtain an initial support vector machine; inputting the rest feature set samples into the initial support vector machine to obtain corresponding prediction probabilities so as to construct a plurality of candidate feature set samples by adopting a sample selection strategy; clustering a plurality of candidate feature set samples to determine each clustering center; and acquiring labels of whether the deposited liquid drops corresponding to each cluster center are normal or not or whether the deposited liquid drops have abnormality in a certain characteristic dimension according to the prediction targets, and further training the initial support vector machine to obtain a final support vector machine for realizing the prediction targets.
7. The intelligent screening method of arrayed spray orifices for ink-jet printing according to claim 6, wherein the sample selection strategy in the following active learning is adopted when training a support vector machine in the classification model:
wherein X is a sample set of the remaining feature set samples, p (y B |x i ) And p (y) SB |x i ) For initial support vector machine pair sample x i The two highest probability values obtained are determined.
8. The intelligent screening method of arrayed spray orifices for ink-jet printing according to claim 6, wherein the clustering method adopted in training the support vector machine in the classification model is a k-means algorithm.
9. The intelligent screening method of arrayed spray holes for ink-jet printing according to claim 1, wherein the feature dimensions included in the feature set are as follows: whether the number of deposited droplets is 0, whether the number of deposited droplets is greater than 1, whether there is a landing deviation of deposited droplets, and whether the deposited droplet diameter exceeds a threshold.
10. An intelligent screening device of array spray holes for ink-jet printing, which is characterized by comprising: the device comprises an ink jet printing module, a liquid drop deposition substrate, a deposited liquid drop observation module, a movement module and a control module;
The control module is used for controlling the ink jet printing module to move above the droplet deposition substrate to perform trial injection through the motion module, controlling the deposition droplet observation module to move to a droplet deposition area through the motion module, collecting deposition droplet images sprayed by all spray holes and transmitting the deposition droplet images to the control module; the control module is further configured to detect and obtain a feature set of the deposited droplet corresponding to each nozzle based on the deposited droplet image, and execute an intelligent screening method of the arrayed nozzle for inkjet printing according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310251962.3A CN116383725A (en) | 2023-03-13 | 2023-03-13 | Intelligent screening method and device for arrayed spray holes for ink-jet printing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310251962.3A CN116383725A (en) | 2023-03-13 | 2023-03-13 | Intelligent screening method and device for arrayed spray holes for ink-jet printing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116383725A true CN116383725A (en) | 2023-07-04 |
Family
ID=86977953
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310251962.3A Pending CN116383725A (en) | 2023-03-13 | 2023-03-13 | Intelligent screening method and device for arrayed spray holes for ink-jet printing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116383725A (en) |
-
2023
- 2023-03-13 CN CN202310251962.3A patent/CN116383725A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108594788B (en) | Airplane actuator fault detection and diagnosis method based on deep random forest algorithm | |
| CN109823050B (en) | Multi-stage positioning error compensation method and device for liquid drop ejection for ink-jet printing | |
| CN109823051B (en) | Liquid drop injection fusion overall process volume control method and system and printer | |
| CN106156401B (en) | Multi-combination classifier based data driving system state model online identification method | |
| CN109409252A (en) | A kind of traffic multi-target detection method based on modified SSD network | |
| CN111079602A (en) | Vehicle fine granularity identification method and device based on multi-scale regional feature constraint | |
| CN111461292B (en) | Unmanned aerial vehicle real-time track prediction method | |
| CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
| Ai et al. | Diagnosis of sensor faults in hypersonic vehicles using wavelet packet translation based support vector regressive classifier | |
| CN110033017A (en) | A kind of more radar track substep Interconnected Fuzzy clustering algorithms | |
| CN107871156B (en) | Ant colony multi-cell tracking system based on pheromone prediction | |
| CN111582373A (en) | Radiation source identification method based on weighted migration extreme learning machine algorithm | |
| CN116383725A (en) | Intelligent screening method and device for arrayed spray holes for ink-jet printing | |
| CN108495263A (en) | A kind of indoor orientation method based on broad sense multi-kernel function | |
| CN111239137A (en) | Grain quality detection method based on transfer learning and adaptive deep convolution neural network | |
| CN109359664A (en) | The efficient Checking model construction method and system of self-teaching update training sample | |
| CN116330845A (en) | Novel display spray printing manufactured droplet volume self-adaptive intelligent regulation and control method | |
| Zhuang et al. | Discovering implicit classes achieves open set domain adaptation | |
| CN115965911A (en) | Intelligent detection method for droplet volume of ink-jet printing | |
| Yue et al. | Intelligent control system for droplet volume in inkjet printing based on stochastic state transition soft actor–critic DRL algorithm | |
| CN116452854A (en) | Adaptive image classification method based on width learning and random sensitivity | |
| CN114462149B (en) | Aircraft pneumatic parameter identification method based on pre-training and incremental learning | |
| Ferreira et al. | Deep learning image analysis for angular measurements in wind tunnels | |
| Zhang et al. | Multi-dimensional local weighted regression ship motion identification modeling based on particle swarm optimization | |
| CN114781512A (en) | Electromagnetic target track fitting method based on multi-source heterogeneous data fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |