An inkjet printing method and apparatus based on droplet volume regulation
By acquiring and expanding jetting data in inkjet printing technology, and utilizing similarity calculation and network model training, the problem of inaccurate droplet volume control caused by changes in ink properties was solved, achieving higher printing quality and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAZHONG UNIV OF SCI & TECH
- Filing Date
- 2026-02-06
- Publication Date
- 2026-06-05
Smart Images
Figure CN122143487A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of inkjet printing technology, and more specifically, relates to an inkjet printing method and apparatus based on droplet volume control. Background Technology
[0002] Inkjet printing technology, as a core process in next-generation display manufacturing, has shown great promise in the field of large-size displays due to its advantages such as high material utilization, large-area fabrication capability, and flexibility. This technology involves formulating functional materials into ink, using piezoelectric printheads to precisely deposit microliter or even picoliter-sized ink droplets into pixel pits on a substrate, and then drying and curing to form a functional thin film.
[0003] In OLED inkjet printing manufacturing, the physicochemical properties of the ink (such as viscosity, surface tension, and solids content) directly determine the droplet formation, trajectory, spreading behavior, and final film quality, making them key factors affecting printing accuracy and device performance. However, ink material properties (such as viscosity and surface tension) can shift with batch or formulation changes. This directly alters the mapping function between waveform parameters and droplet volume, disrupting the effectiveness of the jetting control strategy learned or set under the original conditions. Consequently, droplet volume cannot be accurately controlled within the target range, impacting print quality and stability. Therefore, a method is urgently needed to adjust the inkjet printing jetting control strategy based on ink variations, achieving more accurate control of the ejected droplet volume. Summary of the Invention
[0004] In view of the above-mentioned defects or improvement needs of the prior art, the present invention provides an inkjet printing method and apparatus based on droplet volume control, the purpose of which is to solve the technical problem that the droplet volume cannot be accurately controlled within the target range in existing inkjet printing, thus affecting the printing quality and stability.
[0005] To achieve the above objectives, according to one aspect of the present invention, an inkjet printing method based on droplet volume control is provided, comprising: S1: Obtain the first jetting data of manually jetting the target ink multiple times under the target jetting condition; the first jetting data includes the waveform parameters and droplet volume data corresponding to each jetting of the target ink; S2: Expand the data sample of the first jet data to obtain the second jet data of the target ink; S3: Calculate the feature similarity between the second jet data of the target ink and the historical jet data of the original ink in the database; S4: The historical jetting data of the original ink with a feature similarity greater than a set value are regarded as similar jetting samples of the target ink; S5: Expand the second jetting data based on the similar jetting samples of the target ink to obtain the target training data of the target ink, which includes the similar jetting samples and the second jetting data; S6: Use the target training data of the target ink to train and adjust the original network model used to control the original ink ejection volume, and obtain the target network model used to control the target ink ejection volume; S7: Use the target network model to control the target ink for jet printing.
[0006] Further, S1 includes: manually spraying the target ink multiple times under the target spraying condition to obtain the first spraying data of the target ink; if the currently acquired first spraying data meets a sufficient condition, then S2 is executed; wherein, the sufficient condition is related to the number of times the first spraying data is collected and the droplet volume of the target ink in each spray.
[0007] Furthermore, the sufficient condition is: the currently acquired first injection data satisfies... If , then it is considered to satisfy the sufficient condition; where, For droplet volume samples, Here, Var(x) represents the number of samples, and Var(x) represents the sample variance, used to reflect the stability of data fluctuations. This is the convergence threshold for the variance.
[0008] Further, S3 includes: using the formula Calculate the feature similarity between the second jet data and the historical jet data of the original ink stored in the database. ;
[0009] in, The target ink contains a dimensionless feature vector containing the second jet data, used to characterize the jet flow state; The database contains a dimensionless feature vector of historical jet data, used to characterize the jet flow state; Represents Euclidean distance, indicating feature differences; This refers to the bandwidth parameter of the similarity function; This represents an exponential function.
[0010] Further, S6 includes: determining the loss training weight of each data sample in the target training data; wherein the loss training weight of each data sample belonging to the second spray data is a preset value, and the loss training weight of each data sample belonging to the similar spray samples is calculated according to a preset weight algorithm; and training and adjusting the original network model according to each data sample in the target training data and its corresponding loss training weight to obtain the target network model.
[0011] Furthermore, the weighting algorithm employs the following formula: ; in, This represents the adaptive weight of the i-th source domain sample during training, used to reflect the similarity between the sample and the target domain sample distribution; This represents the feature vector of the i-th data sample in the source domain, used to characterize the input parameters of that sample in the source domain; This represents the feature vector of the t-th sample in the target domain, which is the input sample collected under the target working condition and is used to reflect the distribution differences. A distance metric function representing the distance between samples in the source domain and samples in the target domain; This represents the temperature adjustment coefficient, used to control the smoothness of the sample weight distribution. When the difference in weights is large, the difference is amplified; when the difference is small, the weights tend to be smoothed out.
[0012] Furthermore, the total loss function of the target network model in S62 is: ; in, Indicates the total number of samples of the target ink; This represents the adaptive weight corresponding to the i-th target domain sample; This represents the sample-level loss function, used to measure the error between the model's predicted output and the actual droplet volume; and The i-th training sample is input and the true output are given respectively; This represents the set of trainable parameters for the model in the current target domain. This represents the model parameters pre-trained in the source domain, used to provide the initial structure for cross-domain transfer; This represents the parameter regularization coefficient, which is used to constrain the magnitude of parameter updates during the fine-tuning phase of the model, preventing deviation from the optimal solution in the source domain. This represents the parameter offset penalty term, used to prevent the model from overfitting under small sample conditions.
[0013] According to another aspect of the present invention, an inkjet printing apparatus based on droplet volume control is provided, comprising: The first data expansion module is used to acquire first jetting data of the target ink being manually jetted multiple times under the target jetting condition; the first jetting data includes waveform parameters and droplet volume data corresponding to each jetting of the target ink; it is also used to expand the first jetting data with data samples to obtain second jetting data of the target ink; The second data augmentation module is used to calculate the feature similarity between the second jet data of the target ink and the historical jet data of the original ink in the database; it is also used to regard the historical jet data of the original ink with a feature similarity greater than a set value as similar jet samples of the target ink; and it is also used to augment the second jet data according to the similar jet samples of the target ink to obtain target training data of the target ink, which includes the similar jet samples and the second jet data. The model adjustment module is used to train and adjust the original network model for controlling the jet volume of the original ink using the target training data of the target ink, so as to obtain a target network model for controlling the jet volume of the target ink; and then to use the target network model to control the target ink for jet printing.
[0014] According to another aspect of the present invention, a printing apparatus is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the inkjet printing method.
[0015] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the inkjet printing method.
[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: The inkjet printing method based on droplet volume control provided in this application determines the target ink jetting model step by step at both the data acquisition level and the model training level. In terms of data acquisition, firstly, after manually spraying the target ink under the target spraying conditions and collecting a certain amount of first spraying data, the data is expanded based on the actual first spraying data while ensuring the continuity of data distribution. This allows for the acquisition of a larger amount of sample data that conforms to the actual situation using a smaller amount of actual spraying data, i.e., second spraying data. Secondly, through a database including the historical spraying data of the original ink, similar spraying samples with certain feature similarities to the second spraying data are further compared. These similar spraying samples are then used to further expand the second spraying data. Finally, through the existing small amount of actual spraying data of the target ink and the historical spraying data of the original ink, the target training data that the target ink can satisfy for model training is reasonably derived. This ensures that the data samples are sufficient while the obtained data samples are more consistent with the actual situation of the target ink during spraying. At the model training level, after obtaining the target ink ejection data, the original network model that controls the original ink ejection volume can be trained to obtain the target network model corresponding to the target ink. Then, the target network model is used to automatically control the ejection of the target ink, ensuring that the ejected droplet volume meets the requirements when using the target ink, and can also control the droplet volume within the target range, thereby improving printing quality and stability. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the overall process of an embodiment of the inkjet printing method based on droplet volume control in this application; Figure 2 This is a schematic diagram of the functional modules in one embodiment of the inkjet printing device based on droplet volume control of this application; Figure 3 This is a schematic diagram of the hardware structure of an inkjet printing device based on droplet volume control involved in the embodiments of this application. Detailed Implementation
[0018] 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.
[0019] Example 1 This embodiment provides an inkjet printing method based on droplet volume control, such as... Figure 1 As shown, the steps include: S1-S7.
[0020] S1: Obtain the first jet data of manually jetting the target ink multiple times under the target jetting condition; the first jet data includes the waveform parameters and droplet volume data corresponding to each jetting of the target ink.
[0021] Specifically, S1 requires manual operation under target ejection conditions to eject the target ink, thereby collecting a small amount of actual data, the first ejection data. To avoid redundant sampling and ensure data representativeness, the sufficiency of sampling can be automatically determined based on variance. Furthermore, in some preferred embodiments, a preset sufficiency sampling algorithm is used to determine whether the acquired first ejection data is sufficient. The sufficiency sampling algorithm is related to the number of times the first ejection data is collected and the droplet volume of each ejection of the target ink. The preset sufficiency sampling algorithm is as follows:
[0022] In the formula, For droplet volume samples; ε1 is the number of samples; Var(x) is the sample variance, used to reflect the stability of data fluctuations; ε1 is the convergence threshold of the variance, which is preferably 0.01 to 0.05 in this embodiment. It should be noted that the smaller this value is, the more samples are obtained, and the longer it takes.
[0023] Because it's difficult to obtain a large amount of valid waveform-volume sample data in a short period after switching to a new ink, and because the droplet generation process in inkjet printing involves multiple complex processes requiring a large amount of data to fit a suitable control strategy model, but new target inks often make it difficult to obtain a large amount of new data in a short time, subsequent steps are needed to expand the data at the data level.
[0024] S2: Expand the data sample of the first jet data to obtain the second jet data of the target ink.
[0025] Specifically, this embodiment employs the Mixup method to enhance the collected limited sample data. This method is used to increase the number and distribution continuity of samples, alleviate the problem of insufficient data volume with new ink, smooth the loss surface, and stabilize the learning and training process. This also simulates the smooth response of continuous waveform changes to droplet volume in a real printing device. Specifically, it is expressed as follows:
[0026] in, The enhanced input sample is the new input sample vector obtained by interpolation of the original sample, representing the generated virtual waveform parameters. This indicates an enhanced output sample, which is a new sample output obtained by interpolation of the original output sample, representing the predicted droplet volume under the corresponding interpolated waveform. , This represents the original input sample, which consists of the i-th and j-th groups of waveform parameter input samples already collected in the dataset. A typical form includes the driving voltage. Pulse width T, voltage slope, and number of pulses wait. , This represents the original output sample, corresponding to the input sample. , The measured value of the droplet volume. This represents the interpolation coefficient, used to control the mixing ratio of the two samples, and its value ranges from [0,1]. "Time" indicates that the two samples are mixed with equal weight.
[0027] S3: Calculate the feature similarity between the second jet data of the target ink and the historical jet data of the original ink in the database.
[0028] Specifically, the physical information feature similarity between the target ink and the original ink data in the database is calculated. When the similarity is high, the data from this ink can be used as supplementary training data. The similarity calculation uses the following formula:
[0029] in: This represents a dimensionless eigenvector of the new ink and the ink data in the database, used to characterize the jet flow state, primarily containing the Reynolds number. Weber numbers and Eunersen number ; Represents Euclidean distance, indicating feature differences; This represents the bandwidth parameter of the similarity function, which mainly controls the rate at which distance diminishes the similarity score. A larger value indicates a smaller impact of sample differences on similarity; a smaller value indicates greater sensitivity to similarity changes. After normalization, the feature value is typically set to 0.3–0.8. If it's just a change in ink batch or the same type of ink, 0.7 or 0.8 can be used. If the ink differences are significant (e.g., RGB and TFE inks), 0.3 or 0.4 is generally preferred. middle,
[0030] The density of the liquid is (kg / m³). The velocity of the ejected droplets is (m / s). The nozzle diameter or droplet characteristic diameter (m); The dynamic viscosity of the liquid is (Pa·s). The surface tension of the liquid is (N / m).
[0031] After calculating the feature similarity, the following steps can be used to determine whether it can be used as a similar jetting sample to expand the target ink data.
[0032] S4: Consider the historical jetting data of the original ink with a feature similarity greater than the set value as similar jetting samples of the target ink.
[0033] Specifically, the similarity is calculated using a formula. hour, This threshold is used to determine whether a sample from the source domain is sufficiently close to a sample from the target domain. When the calculated similarity is greater than this threshold, the sample is included in the augmented dataset. The typical value is between 0.75 and 0.95. When rigorously filtering highly similar samples is desired, a value of 0.9 to 0.95 can be used; if the target domain data volume is small and it is desired to expand the sample coverage, a value of 0.75 to 0.8 can be used.
[0034] S5: Expand the second jet data based on similar jet samples of the target ink to obtain target training data of the target ink, which includes similar jet samples and second jet data.
[0035] Specifically, since the similar spray samples in the target training data are samples initially screened through physical information, each similar ink may only exhibit similar characteristics to the new ink in a small portion of the sample data. Therefore, it is necessary to further enhance the learning contribution of samples more similar to the new ink environment at the data layer and suppress interference from irrelevant samples. The specific approach is to add a training weight to the data samples, so that the network will pay more attention to samples with stronger relevance during training. In this embodiment, the training weight of the second spray data in the target training data is preset to a fixed value of 1 because it directly comes from the actual spray data of the target ink, while the training weights corresponding to the similar spray samples are calculated using a preset weighting algorithm.
[0036] S6: Use the target training data of the target ink to train and adjust the original network model used to control the original ink ejection volume, and obtain the target network model used to control the target ink ejection volume.
[0037] Specifically, during training, if the training weight is 1, 100% of the individual loss for that sample is included in the loss function; if it's 0.1, only 10% is included in the loss function. Specifically, this includes the following steps: S61: Determine the loss training weight for each data sample in the target training data; wherein, the loss training weight for each data sample belonging to the second spray data is a preset value of 1, and the weight for each data sample belonging to the similar spray samples is calculated according to a preset weight algorithm. The weight algorithm adopts the following formula:
[0038] in: Indicates the first The adaptive weights of each source domain sample during training are used to reflect the similarity between the distribution of the sample and the target domain sample. Represents the source domain. The feature vector of each sample is used to characterize the input parameters of that sample; The feature vector representing the target domain sample is the input sample collected under the target working condition, used to calculate the distribution difference; : This represents the distance metric function between samples in the source domain and samples in the target domain. Depending on the specific application, it can be selected as Euclidean distance, maximum mean difference (MMD), KL divergence, or cosine similarity, etc. : Represents the temperature adjustment coefficient, used to control the smoothness of the sample weight distribution. When the difference in weights is large, the difference is amplified; when it is small, the weights tend to smooth out. Since the distance D is calculated normally between 0 and 1, to ensure the effective gradient range of the exponential function, α is recommended to be 3 to 10. When the difference between new and old inks is large, 8-10 is recommended because this can filter out more similar samples under different inks, avoiding abnormal model training. When the difference between new and old inks is small, 3-5 is recommended because this allows more samples to help the model.
[0039] S62: Based on each data sample in the target training data and its corresponding loss training weights, train and adjust the original network model to obtain the target network model. Wherein, the original network model M0:
[0040] in: The volume of the droplet; A parameter vector representing the current injection waveform; This represents the set of parameters for the regulation model.
[0041] Specifically, during the training and adjustment process, in the model fine-tuning stage, the pre-trained base model is first used. As the initial network structure and parameter basis, the model. The low-level network parameters (such as feature extraction layers, convolutional layers, or the first two fully connected layers) are frozen during fine-tuning and do not participate in gradient updates to prevent general features from being destroyed by small sample perturbations in the target domain. Subsequently, only the parameters of the high-level feature layers and output layers of the model are updated, enabling the model to adaptively adjust to new changes (new ink) in the target domain.
[0042] As an optional implementation, a weighted loss function for the target domain samples is introduced, and the samples are adaptively weighted. Enhance the learning contribution of samples with distributions similar to the target domain, and weaken the influence of irrelevant samples. Model parameters During training, the source domain parameters are used to determine the parameters. Gradually update to the optimal parameters of the target domain Its change process is affected by the parameter regularization term. The constraints are designed to prevent excessive deviation from the source domain solution, specifically expressed as follows:
[0043] In the formula: This represents the total loss function during the model fine-tuning phase, used to constrain the convergence direction and stability of the model in the target domain; Indicates the total number of samples of the target ink; This represents the adaptive weight corresponding to the i-th target domain sample; This represents the sample-level loss function, used to measure the error between the model's predicted output and the actual droplet volume; , Indicates the first Each training sample input and the actual output; This represents the set of trainable parameters for the model in the current target domain. This represents the model parameters pre-trained in the source domain, used to provide the initial structure for cross-domain transfer; : Represents the parameter regularization coefficient, used to constrain the magnitude of parameter updates during the fine-tuning phase of the model, preventing deviation from the optimal solution in the source domain; in actual training process, The value range is usually 0.001 to 0.05. When the number of samples in the target domain is large and the distribution is similar to that of the source domain, a smaller value of 0.005 can be used. When the number of samples in the target domain is small or the distribution is significantly different, the value can be increased to around 0.05. : Represents the parameter offset penalty term, used to prevent the model from overfitting under small sample conditions.
[0044] The reason for freezing the low-level parameters in the above process is that the low-level network has learned the general jet dynamics and waveform structure characteristics; the high-level parameters mainly reflect the volume deviation mapping relationship of specific nozzles, inks or environments; fine-tuning only the high-level parameters can retain the general features, avoid overfitting on small samples, and significantly reduce the amount of computation and training time, thereby ensuring transfer stability and accuracy.
[0045] Furthermore, S7: Uses a target network model to control the target ink for jet printing.
[0046] Example 2 This embodiment provides an inkjet printing device based on droplet volume control, including: a first data expansion module, a second data expansion module, and a model adjustment module.
[0047] The first data expansion module is used to acquire first jetting data of the target ink being manually jetted multiple times under the target jetting condition; the first jetting data includes waveform parameters and droplet volume data corresponding to each jetting of the target ink; it is also used to expand the first jetting data with data samples to obtain second jetting data of the target ink.
[0048] The second data augmentation module is used to calculate the feature similarity between the second jet data of the target ink and the historical jet data of the original ink in the database; it is also used to regard the historical jet data of the original ink with a feature similarity greater than a set value as similar jet samples of the target ink; and it is also used to augment the second jet data according to the similar jet samples of the target ink to obtain target training data of the target ink, which includes the similar jet samples and the second jet data.
[0049] The model adjustment module is used to train and adjust the original network model for controlling the jet volume of the original ink using the target training data of the target ink, so as to obtain a target network model for controlling the jet volume of the target ink; and then to use the target network model to control the target ink for jet printing.
[0050] Example 3 This embodiment provides a printing device, such as... Figure 3 As shown, it includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the inkjet printing method.
[0051] Example 4 This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the inkjet printing method.
[0052] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.
[0053] Example 5 This invention provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the method described in the above embodiments of this invention.
[0054] The technical features of the embodiments described above can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. It should be noted that the terms "in one embodiment," "for example," and "again" in this invention are intended to illustrate the invention and are not intended to limit the invention.
[0055] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.
Claims
1. An inkjet printing method based on droplet volume control, characterized in that, include: S1: Obtain the first jet data of manually jetting the target ink multiple times under the target jetting condition; The first jetting data includes waveform parameters and droplet volume data corresponding to each jetting of the target ink; S2: Expand the data sample of the first jet data to obtain the second jet data of the target ink; S3: Calculate the feature similarity between the second jet data of the target ink and the historical jet data of the original ink in the database; S4: The historical jetting data of the original ink with a feature similarity greater than a set value are regarded as similar jetting samples of the target ink; S5: Expand the second jetting data based on the similar jetting samples of the target ink to obtain the target training data of the target ink, which includes the similar jetting samples and the second jetting data; S6: Use the target training data of the target ink to train and adjust the original network model used to control the original ink ejection volume, and obtain the target network model used to control the target ink ejection volume; S7: Use the target network model to control the target ink for jet printing.
2. The inkjet printing method based on droplet volume control as described in claim 1, characterized in that, S1 includes: manually spraying the target ink multiple times under the target spraying condition to obtain the first spraying data of the target ink; if the currently obtained first spraying data meets the sufficient condition, then S2 is executed; The sufficient condition is related to the number of times the first jet data is collected and the droplet volume of the target ink in each jet.
3. The inkjet printing method based on droplet volume control as described in claim 2, characterized in that, The sufficient condition is that the currently acquired first injection data satisfies... If , then it is considered to satisfy the sufficient condition; where, For droplet volume samples, Here, Var(x) represents the number of samples, and Var(x) represents the sample variance, used to reflect the stability of data fluctuations. This is the convergence threshold for the variance.
4. The inkjet printing method based on droplet volume control as described in claim 1, characterized in that, S3 includes: using the formula Calculate the feature similarity between the second jet data and the historical jet data of the original ink stored in the database. ; in, The target ink contains a dimensionless feature vector containing the second jet data, used to characterize the jet flow state; The database contains a dimensionless feature vector of historical jet data, used to characterize the jet flow state; Represents Euclidean distance, indicating feature differences; This refers to the bandwidth parameter of the similarity function; This represents an exponential function.
5. The inkjet printing method based on droplet volume control as described in claim 1, characterized in that, S6 includes: S61: Determine the loss training weight of each data sample in the target training data; wherein, the loss training weight of each data sample belonging to the second spray data is a preset value, and the loss training weight of each data sample belonging to the similar spray samples is calculated according to a preset weight algorithm; S62: Based on each data sample in the target training data and its corresponding loss training weights, the original network model is trained and adjusted to obtain the target network model.
6. The inkjet printing method based on droplet volume control as described in claim 5, characterized in that, The weighting algorithm uses the following formula: ； in, This represents the adaptive weight of the i-th source domain sample during training, used to reflect the similarity between the sample and the target domain sample distribution; This represents the feature vector of the i-th data sample in the source domain, used to characterize the input parameters of that sample in the source domain; This represents the feature vector of the t-th sample in the target domain, which is the input sample collected under the target working condition and is used to reflect the distribution differences. A distance metric function representing the distance between samples in the source domain and samples in the target domain; This represents the temperature adjustment coefficient, used to control the smoothness of the sample weight distribution. When the difference in weights is large, the difference is amplified; when the difference is small, the weights tend to be smoothed out.
7. The inkjet printing method based on droplet volume control as described in claim 5, characterized in that, The total loss function of the target network model in S62 is: ； in, This represents the total number of samples of the target ink. This represents the adaptive weight corresponding to the i-th target domain sample; This represents the sample-level loss function, used to measure the error between the model's predicted output and the actual droplet volume; and The i-th training sample is input and the true output are given respectively; This represents the set of trainable parameters for the model in the current target domain. This represents the model parameters pre-trained in the source domain, used to provide the initial structure for cross-domain transfer; This represents the parameter regularization coefficient, which is used to constrain the magnitude of parameter updates during the fine-tuning phase of the model, preventing deviation from the optimal solution in the source domain. This represents the parameter offset penalty term, used to prevent the model from overfitting under small sample conditions.
8. An inkjet printing device based on droplet volume control, characterized in that, include: The first data expansion module is used to acquire the first jetting data of manually jetting the target ink multiple times under the target jetting condition; The first jetting data includes waveform parameters and droplet volume data corresponding to each jetting of the target ink; it is also used to expand the first jetting data with data samples to obtain the second jetting data of the target ink. The second data augmentation module is used to calculate the feature similarity between the second jet data of the target ink and the historical jet data of the original ink in the database; it is also used to regard the historical jet data of the original ink with a feature similarity greater than a set value as similar jet samples of the target ink; and it is also used to augment the second jet data according to the similar jet samples of the target ink to obtain target training data of the target ink, which includes the similar jet samples and the second jet data. The model adjustment module is used to train and adjust the original network model for controlling the jet volume of the original ink using the target training data of the target ink, so as to obtain a target network model for controlling the jet volume of the target ink; and then to use the target network model to control the target ink for jet printing.
9. A printing apparatus comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the inkjet printing method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the inkjet printing method according to any one of claims 1 to 7.