Inkjet printer nozzle intelligent regulation component based on image quality online monitoring

Abstract

The present application relates to the inkjet printer nozzle intelligent regulation and control assembly based on image quality online monitoring, belongs to the technical field of additive manufacturing equipment, including: image acquisition module for collecting dynamic images; Multimodal quality analysis module is used for constructing three-dimensional quality evaluation model; Nozzle state positioning module is based on the three-dimensional quality evaluation model of printing defect and the preset nozzle-image pixel mapping, accurately positioning the coordinates and fault type of the fault nozzle; Adaptive control module outputs differential control instructions according to dynamic images and fault types, executes ink drop volume fine tuning, jet frequency correction or pre-cleaning operation on the fault nozzle, controls adjacent nozzles to carry out dynamic ink compensation at the same time, and realizes the real-time repair of printing defects; The beneficial effects of the present application are: through the adaptive control module, differential control instructions are output according to dynamic images and fault types, ink drop volume fine tuning, jet frequency correction or pre-cleaning operation is executed on the fault nozzle, and flexible adaptation of the control strategy is realized.

Description

Technical Field

[0001] This invention belongs to the field of additive manufacturing equipment technology, and specifically relates to an intelligent nozzle control component for inkjet printers based on online image quality monitoring. Background Technology

[0002] Inkjet printers use nozzles to spray ink onto the printing medium to form images or text. The working state of the nozzles directly determines the print quality. In actual printing, nozzles are prone to clogging, uneven ink flow, and ink jet angle deviation, resulting in defects such as broken lines, blurriness, and color differences in the printed image.

[0003] In existing technologies, nozzle control is mostly achieved through periodic manual inspection or preset fixed parameter control. Manual inspection is inefficient, cannot achieve real-time monitoring and control, and relies on the operator's experience, making it prone to misjudgment. Preset fixed parameter control cannot dynamically adjust according to changes in actual printed image quality, and is difficult to adapt to different printing materials, ink types, and dynamic changes in nozzle status during long-term printing, resulting in poor print quality stability and a high scrap rate.

[0004] In addition, although some existing technologies attempt to introduce image acquisition devices for quality monitoring, they suffer from low monitoring accuracy, large data processing delays, and limited control strategies. They are unable to quickly and accurately locate nozzle malfunctions and implement targeted controls, making it difficult to meet the demands for high-precision and high-efficiency printing. Summary of the Invention

[0005] This invention provides an intelligent nozzle control component for inkjet printers based on online image quality monitoring. It addresses the technical problem of low control precision in existing inkjet printer nozzle control methods. Through an adaptive control module, differentiated control commands are output based on dynamic images and fault types. This allows for fine-tuning of droplet volume, correction of ejection frequency, or pre-cleaning operations on faulty nozzles. Simultaneously, adjacent nozzles are dynamically compensated for ink volume, enabling real-time repair of printing defects. The feedback control mechanism of the adaptive control module ensures effective control, effectively repairs printing defects, improves print quality stability, and achieves flexible adaptation of control strategies.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: The inkjet printer nozzle intelligent control component based on online image quality monitoring includes: The image acquisition module is used to acquire images of the target medium and dynamic images of the nozzle ejection process in real time. The multimodal quality analysis module uses grayscale difference calculation, ink droplet distribution density statistics, and color uniformity analysis to construct a three-dimensional quality assessment model. The three-dimensional quality assessment model is used to identify printing defects such as missing prints, streaks, and color deviations. The nozzle status positioning module accurately locates the coordinates and fault type of faulty nozzles based on the printing defects in the three-dimensional quality assessment model and the preset nozzle-image pixel mapping relationship. The adaptive control module outputs differentiated control commands based on dynamic images and fault types, performing fine-tuning of ink droplet volume, correction of jet frequency, or pre-cleaning operations on faulty nozzles. At the same time, it controls adjacent nozzles to perform dynamic ink volume compensation, enabling real-time repair of printing defects.

[0007] Optionally, the image acquisition module includes a high frame rate industrial camera, a ring-shaped fill light unit, and an adaptive focus adjustment unit; High frame rate industrial cameras have a frame rate of no less than 200fps; The adaptive focus adjustment unit automatically adjusts the focal length based on the thickness information of the printing medium and the printing resolution; The ring-shaped fill light unit uses an RGB three-color adjustable light source, which dynamically matches the fill light parameters according to the background color of the printing medium and the ink color to eliminate reflection interference and improve image contrast.

[0008] Furthermore, the image acquisition module also includes a multi-view acquisition unit, which simultaneously acquires dynamic images of the printing medium surface and the nozzle ejection side using at least two industrial cameras at different angles. After the multi-view images are synthesized by an image fusion algorithm, the three-dimensional trajectory and landing pattern information of the ink droplet ejection are obtained, providing comprehensive data support for the adaptive control module to determine the cause of the fault and formulate control strategies.

[0009] Optionally, when constructing a 3D quality assessment model, the multimodal quality analysis module first performs Gaussian filtering for noise reduction and morphological enhancement preprocessing on the acquired images. Then, it obtains the gray-level difference matrix of adjacent pixels through gray-level difference calculation. Combining the ink droplet spatial distribution entropy obtained from ink droplet distribution density statistics and the CIELab color space deviation value obtained from color uniformity analysis, the gray-level difference matrix, ink droplet spatial distribution entropy, and CIELab color space deviation value are used as input features to construct a random forest classifier, thereby achieving accurate identification and classification of defects such as missing prints, streaks, and color deviation.

[0010] Optionally, the nozzle status positioning module also has a fault trend prediction function. By statistically analyzing the frequency and level of defects of the same nozzle during continuous printing, and combining the cumulative working time of the nozzle, a nozzle life prediction model is constructed. When a serious nozzle failure is predicted, an early warning signal is output and preventive control operations are triggered to reduce the incidence of printing defects.

[0011] Furthermore, the nozzle-image pixel mapping relationship is pre-constructed through calibration experiments. During the calibration process, standard test patterns are printed, and machine vision technology is used to obtain the image pixel area corresponding to the ink droplets ejected by each nozzle. A three-dimensional mapping table of nozzle number, physical coordinates and image pixel coordinates is established. At the same time, a temperature compensation coefficient is introduced to dynamically correct the mapping relationship according to the printer's operating temperature, thereby improving the accuracy of faulty nozzle location.

[0012] Optionally, the adaptive control module includes a piezoelectric drive unit and an ink path pressure adjustment unit; When the fault type is minor missing print, the ink droplet volume of the faulty nozzle is finely adjusted to within 5% by the piezoelectric drive unit. When the fault type is stringing, the ink pressure in the nozzle cavity is adjusted by the ink path pressure regulating unit, and the spraying frequency is corrected to the preset optimal range. When the fault type is severe missing print or color deviation, first control the faulty nozzle to perform a pulsed pre-cleaning operation, and then achieve rapid repair of printing defects by compensating for the increase in ink droplet volume of adjacent nozzles and adjusting the jetting timing.

[0013] Furthermore, the adaptive control module is connected to an environmental sensing module, which is used to collect real-time data on the temperature, humidity, and air pressure of the printing environment. When constructing a 3D quality assessment model, the multimodal quality analysis module incorporates environmental perception data as auxiliary features into the model training. The adaptive control module dynamically adjusts the parameter thresholds of the control commands based on environmental parameters, enabling the control strategy to adapt to printing needs under different environmental conditions.

[0014] Furthermore, the adaptive control module adopts a closed-loop feedback control mechanism. After outputting the control command, the image acquisition module acquires the repaired printed image in real time, and the multimodal quality analysis module evaluates the repair effect. If the repaired image still has defects and the defect level does not meet the standard, the control parameters are iteratively optimized according to the evaluation results until the printing defects are eliminated or the preset quality standard is reached.

[0015] Optionally, the adaptive control module is connected to a data storage and traceability module. The data storage and traceability module is used to record image acquisition data, quality analysis results, faulty nozzle information, and control command parameters. It supports data query and traceability by printing task number and time range, and uses historical data to optimize the three-dimensional quality assessment model and control strategy parameters to achieve continuous iterative improvement of component performance.

[0016] The beneficial effects of this invention are: 1. The multimodal quality analysis module of the present invention uses grayscale difference calculation, ink droplet distribution density statistics and color uniformity analysis to construct a three-dimensional quality assessment model. The three-dimensional quality assessment model is used to identify printing defects such as missing prints, streaks and color deviations. It can capture printing image defects and nozzle jet dynamics in real time, accurately identify problems such as missing prints, streaks and color deviations, and significantly improve monitoring accuracy.

[0017] 2. The nozzle status positioning module of the present invention is based on the printing defects of the three-dimensional quality assessment model and the preset nozzle-image pixel mapping relationship to accurately locate the coordinates and fault type of the faulty nozzle. The nozzle status positioning module can quickly locate the coordinates and type of the faulty nozzle based on the pre-calibrated nozzle-image pixel mapping relationship and temperature compensation mechanism. At the same time, it has the fault trend prediction function to realize preventive control and reduce the defect incidence rate. The nozzle status positioning module can output differentiated control commands according to different fault types and environmental conditions.

[0018] 3. The adaptive control module of the present invention outputs differentiated control commands based on dynamic images and fault types, performs ink droplet volume fine-tuning, jet frequency correction or pre-cleaning operations on faulty nozzles, and simultaneously controls adjacent nozzles to perform dynamic ink volume compensation, realizing real-time repair of printing defects. The feedback control mechanism of the adaptive control module ensures the control effect, effectively repairs printing defects, improves print quality stability, and realizes flexible adaptation of control strategies. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the component system structure of the present invention; Figure 2 This is a schematic diagram of the component system workflow of the present invention. Detailed Implementation

[0021] The embodiments of this application will now be described in detail with reference to the accompanying drawings. Example

[0022] like Figure 1 As shown, this embodiment provides an intelligent nozzle control component for inkjet printers based on online image quality monitoring, including: an image acquisition module, a multimodal quality analysis module, a nozzle status positioning module, an adaptive control module, an environmental perception module, and a data storage and traceability module; the configuration of each module is as follows: The image acquisition module is used to acquire images of the target medium and dynamic images of the nozzle ejection process in real time. The multimodal quality analysis module uses grayscale difference calculation, ink droplet distribution density statistics, and color uniformity analysis to construct a three-dimensional quality assessment model. The three-dimensional quality assessment model is used to identify printing defects such as missing prints, streaks, and color deviations. The nozzle status positioning module accurately locates the coordinates and fault type of faulty nozzles based on the printing defects in the three-dimensional quality assessment model and the preset nozzle-image pixel mapping relationship. The adaptive control module outputs differentiated control commands based on dynamic images and fault types, performing fine-tuning of ink droplet volume, correction of jet frequency, or pre-cleaning operations on faulty nozzles. At the same time, it controls adjacent nozzles to perform dynamic ink volume compensation, enabling real-time repair of printing defects.

[0023] The image acquisition module in this embodiment can achieve real-time and accurate monitoring. That is, through the high frame rate and multi-view image acquisition module, combined with the multimodal quality analysis model, it can capture printing image defects and nozzle jet dynamics in real time, accurately identify problems such as missing prints, streaks, and color deviations, and significantly improve monitoring accuracy. The nozzle status positioning module can quickly locate the coordinates and type of a faulty nozzle based on a pre-calibrated nozzle-image pixel mapping relationship and temperature compensation mechanism. It also has a fault trend prediction function to achieve preventive control and reduce the defect incidence rate. The nozzle status positioning module can output differentiated control commands based on different fault types and environmental conditions. Combined with the feedback control mechanism of the adaptive control module, it ensures the control effect, effectively repairs printing defects, improves printing quality stability, and achieves flexible adaptation of control strategies.

[0024] The data storage and traceability module records data throughout the entire process, supports querying and tracing, and utilizes historical data to iteratively optimize models and control parameters, thereby achieving continuous improvement in component performance. Data is traceable and performance is continuously optimized. Example

[0025] Based on Embodiment 1, the image acquisition module includes a high frame rate industrial camera, a ring-shaped fill light unit, and an adaptive focus adjustment unit. High frame rate industrial cameras have a frame rate of no less than 200fps; The adaptive focus adjustment unit automatically adjusts the focal length based on the thickness information of the printing medium and the printing resolution; The ring-shaped fill light unit uses an RGB three-color adjustable light source, which dynamically matches the fill light parameters according to the background color of the printing medium and the ink color to eliminate reflection interference and improve image contrast.

[0026] After the printing task is started, the high frame rate industrial camera acquires the thickness data of the printing medium in real time and transmits it to the adaptive focus adjustment unit. The adaptive focus adjustment unit, combined with the resolution set by the printing task (e.g., 1200dpi, 2400dpi), automatically adjusts the focal length of the focusing lens using an electric translation stage to ensure that the camera acquires a clear image. At the same time, the ring light unit receives the background color and ink color parameters of the printing medium (synchronously transmitted by the printer driver) through a serial port and dynamically adjusts the brightness ratio of the RGB three-color light source. For example, when printing black ink on white photo paper, the brightness of the blue channel is increased and the brightness of the red and green channels is decreased to eliminate the interference of paper reflection on image acquisition. Under the synchronous control of the image acquisition card, the two industrial cameras at different perspectives acquire images synchronously at a frame rate of 220fps. The front-view camera acquires a 2cm×2cm area on the printing medium in each frame, while the side-view camera captures the complete trajectory of the ink droplets from the nozzle to the landing on the medium surface in each frame. The acquired image data is transmitted in real time to the multimodal quality analysis module via the GigE interface and PCIe interface, with the transmission delay controlled within 50ms.

[0027] In addition, the image acquisition module also includes a multi-view acquisition unit, which simultaneously acquires dynamic images of the printing medium surface and the nozzle ejection side using at least two industrial cameras at different angles. After the multi-view images are synthesized by an image fusion algorithm, the three-dimensional trajectory and landing pattern information of the ink droplet ejection are obtained, providing comprehensive data support for the adaptive control module to determine the cause of the fault and formulate control strategies. Example

[0028] Based on Example 1, when constructing a three-dimensional quality assessment model, the multimodal quality analysis module first performs Gaussian filtering for noise reduction and morphological enhancement preprocessing on the acquired images. Then, it obtains the gray-level difference matrix of adjacent pixels through gray-level difference calculation. Combining the ink droplet spatial distribution entropy obtained from ink droplet distribution density statistics and the CIELab color space deviation value obtained from color uniformity analysis, the gray-level difference matrix, ink droplet spatial distribution entropy, and CIELab color space deviation value are used as input features to construct a random forest classifier, thereby achieving accurate identification and classification of defects such as missing prints, streaks, and color deviation.

[0029] The multimodal quality analysis module uses the OpenCV library to implement Gaussian filtering for noise reduction. The filter kernel size is set to 5×5, and the standard deviation σ=1.2, which effectively removes random noise in the image. Morphological enhancement uses an opening operation of dilation followed by erosion, with a 3×3 rectangular structuring element to enhance the contour features of the ink droplet region and suppress small noise points. The gray-level difference calculation for the multimodal quality analysis module is as follows: the gray-level difference between adjacent pixels is calculated by dividing the preprocessed image by row and column, and an M×N gray-level difference matrix is ​​constructed (M and N are the height and width of the image in pixels). The matrix element values ​​are the absolute differences between the gray-level values ​​of adjacent pixels. Ink droplet distribution density statistics are performed by extracting ink droplet regions from an image using a connected component analysis algorithm, and then calculating the area of ​​each connected region. Specifically, the connected component analysis algorithm is defined by the following formula: ; in, The spatial distribution entropy of ink droplets (unit: bit / symbol), with a value range of [value missing]. , The total number of connected regions of ink droplets in the image (extracted by a morphological connected region analysis algorithm, with the 8-neighborhood rule used for connectivity determination). The base of the logarithm is 2 by default (which conforms to the classic definition of information entropy, ensuring that the entropy value is in bits for easy quantitative analysis). It can also be the natural constant e or 10 as needed, and must be kept consistent during model training.

[0030] For the first The area percentage of each ink droplet connected region Index of a connected region of a single ink droplet ( );Right now , For the first The pixel area of ​​the connected region of each ink droplet (unit: pixel²); The total pixel area of ​​all connected regions of ink droplets ( (Unit: pixels²) The addition of a negative sign is for summation. Summation counts the contribution of all ink droplet regions, and the negative sign converts the result to a positive number (positive entropy is more intuitive). For the area ratio Take the logarithm, when The larger the droplet area (the more dominant it is). Approaching 0; when The smaller the size (the ink droplet area becomes secondary). The absolute value of a negative number increases. The negative sign is used to convert the summation result into a positive value, making the physical meaning of the entropy value more intuitive (the larger the entropy value, the more disordered the distribution). Satisfy normalization conditions This ensures the rationality of entropy calculation; If there are no ink droplets in the image ( Then define If all ink droplets are concentrated in a single connected region ( ),but , This is the most uniform state.

[0031] Ink droplet spatial distribution entropy The uniformity of ink droplet distribution in a printed image is used to quantify the characteristics of ink droplets. It is one of the core features for identifying defects such as missing prints and streaks. Essentially, it describes the degree of disorder in the spatial distribution of ink droplets through information entropy. The higher the entropy value, the more dispersed and uneven the ink droplet distribution is, and the higher the probability of printing defects. The lower the entropy value, the more concentrated and uniform the ink droplet distribution is, and the better the printing quality is.

[0032] Color uniformity analysis involves converting an RGB image to the CIELab color space and calculating the deviations of each pixel's L* (luminance), a* (red-green deviation), and b* (blue-yellow deviation) values ​​from the standard color values, thus obtaining the CIELab color space deviation value. : ; in, This is the total deviation value for the CIELab color space, without units; the larger the value, the more severe the color deviation. The deviation is invisible to the naked eye. (for obvious deviations) For the Lightness channel: the value ranges from 0 to 100, where 0 represents pure black and 100 represents pure white; Positive values ​​indicate that the actual color is brighter, while negative values ​​indicate that the actual color is darker. For the red-green channel: the value range is -128 to +127, with positive values ​​leaning towards red and negative values ​​leaning towards green; Positive values ​​indicate that the color is actually redder, while negative values ​​indicate that the color is more greener; For the blue-yellow channel: the value range is -128 to +127, positive values ​​are biased towards yellow, and negative values ​​are biased towards blue; Positive values ​​indicate a more yellow color, while negative values ​​indicate a more blue color.

[0033] Calculate for all pixels in the same color region of the image , and Substitute the average value into the formula to obtain the total deviation value of the CIELab color space for that region. As a basis for grading color deviation defects (e.g.: The color deviation was determined to be slight. The color deviation is determined to be moderate. (Judged as severe color deviation).

[0034] The CIELab color space is a standardized color model that simulates human color perception. Its deviation value... It can accurately describe the distance between the actual printed color and the target color in the three-dimensional color space. CIELab color space total deviation value. The calculation combines the entire process of image acquisition, preprocessing, and color conversion to ensure the feature input of the 3D quality assessment model.

[0035] The random forest classifier is built using the Scikit-learn library. It consists of 100 decision trees, each with a maximum depth of 15, a minimum number of splits of 2, and a minimum number of leaf nodes of 1. The input features are the mean, variance, and droplet spatial distribution entropy of the gray-level difference matrix. and CIELab color space deviation value The output is the defect type (missing print, streaks, and color deviation) and the defect level (level 1-5, level 1 is a minor defect and level 5 is a serious defect).

[0036] After the multimodal quality analysis module receives the image data transmitted by the image acquisition module: First, start the preprocessing process, and control the processing time of a single frame image to within 20ms; Next, feature extraction is performed. The extraction time for the three types of features is less than 30ms in total. The extracted feature vectors are then input into a pre-trained random forest classifier. The classifier uses CUDA to accelerate inference, and the defect recognition time for a single frame image is less than 10ms. The temperature, humidity, and air pressure data transmitted by the environmental perception module are normalized (normalized to the [0,1] interval) and incorporated into the classifier's reasoning process as auxiliary features. The classification results are adjusted by weighted summation, with weight coefficients of 0.3 for temperature, 0.4 for humidity, and 0.3 for air pressure, thereby improving the model's recognition accuracy in different environments. Finally, the quality assessment results of the defect type, defect level, and pixel coordinates of the defect area are output, with an accuracy rate of no less than 95% and a defect level judgment error of no more than 1 level. Example

[0037] Based on Example 1, the nozzle-image pixel mapping relationship is pre-constructed through calibration experiments. During the calibration process, standard test patterns are printed, and machine vision technology is used to obtain the image pixel area corresponding to each nozzle's ink droplet. A three-dimensional mapping table of nozzle number, physical coordinates, and image pixel coordinates is established. At the same time, a temperature compensation coefficient is introduced to dynamically correct the mapping relationship according to the printer's operating temperature, thereby improving the accuracy of fault nozzle location.

[0038] The specific process of constructing the nozzle-image pixel mapping relationship is as follows: The test pattern was used, which consisted of a 10×10 grid of ink dots, each dot having a diameter of 0.1 mm and a dot spacing of 0.5 mm. The print resolution was set to 2400 dpi. The printed test pattern is placed in the image acquisition area, and the image of the test pattern is acquired through the image acquisition module. Using the template matching algorithm in machine vision technology, the pixel coordinates of each ink dot are accurately located. Combined with the physical arrangement coordinates of the printer nozzles (the nozzle spacing is 0.2mm, there are 32 nozzles arranged in one column), a three-dimensional mapping table of nozzle number (1-32), physical coordinates (X,Y) and image pixel coordinates (u,v) is established for calibration experiments.

[0039] The operating temperature is collected by a PT100 temperature sensor (measurement range -50℃ to 200℃, accuracy ±0.1℃) installed in the printer nozzle cavity. The temperature compensation coefficient K(T) is obtained by piecewise linear fitting, K(T) = 1 + 0.002 × (T - 25), where T is the actual operating temperature (℃). When the temperature deviates from the standard calibration temperature (25℃), the mapping relationship between pixel coordinates and physical coordinates is corrected by K(T). The correction formula is u' = u × K(T) and v' = v × K(T), to obtain the corrected image pixel coordinates (u', v'). This ensures that the faulty nozzle positioning error does not exceed 0.05mm within the operating temperature range of 15℃ to 35℃.

[0040] In addition, the nozzle status positioning module also has a fault trend prediction function. By statistically analyzing the frequency of defects and the changes in defect level of the same nozzle during continuous printing, and combining the cumulative working time of the nozzle, a nozzle life prediction model is constructed. When a serious nozzle failure is predicted, an early warning signal is output and preventive control operations are triggered to reduce the incidence of printing defects.

[0041] The specific process for fault trend prediction and location is as follows: The fault trend prediction is achieved by the nozzle status positioning module, which statistically analyzes the frequency and level of defects of the same nozzle in 100 consecutive frames of images in real time. Combined with the cumulative working time of the nozzle (starting from the printer startup and accurate to the second), a nozzle life prediction model is constructed. This model is implemented using an LSTM neural network. The inputs are the frequency of defects, the average level of defects, and the increment of cumulative working time in the past 10 times. The output is the probability of a serious fault (level 4-5 defect) occurring in the next 5000 sprays. When the probability exceeds 80%, a warning signal is output. Fault localization involves receiving the pixel coordinates of the defect area from the multimodal quality analysis module, then searching for the corresponding nozzle number and physical coordinates using a 3D mapping table. For example, if the pixel coordinates of the defect area are (u0, v0), after temperature compensation correction, they become (u0', v0'). The mapping table is then consulted to determine the corresponding nozzle number as B and the physical coordinates as (Xn, Yn). Simultaneously, the fault type is determined by combining the defect type (i.e., missing print corresponds to nozzle blockage, string marks correspond to abnormal ink pressure, and color deviation corresponds to changes in ink characteristics or nozzle ejection angle offset). The localization result is transmitted to the adaptive control module, with a localization time of no more than 20ms. Example

[0042] Based on Embodiment 1, the adaptive control module includes a piezoelectric drive unit and an ink path pressure adjustment unit; When the fault type is minor missing print, the ink droplet volume of the faulty nozzle is finely adjusted to within 5% by the piezoelectric drive unit. When the fault type is stringing, the ink pressure in the nozzle cavity is adjusted by the ink path pressure regulating unit, and the spraying frequency is corrected to the preset optimal range. When the fault type is severe missing print or color deviation, first control the faulty nozzle to perform a pulsed pre-cleaning operation, and then achieve rapid repair of printing defects by compensating for the increase in ink droplet volume of adjacent nozzles and adjusting the jetting timing.

[0043] Specifically, the control strategy and process of the adaptive control module are as follows: Minor print defects (level 1-2 defects): The adaptive control module outputs instructions based on the fault type, and the piezoelectric drive unit adjusts the drive voltage to fine-tune the droplet volume. The droplet volume is linearly related to the drive voltage, and the voltage adjustment step is 0.1V. After each adjustment, 10 frames of images are acquired by the image acquisition module, and the multimodal quality analysis module evaluates the repair effect until the droplet volume fine-tuning accuracy is controlled within 5%. For example, if the standard droplet volume is 10pL, the adjusted droplet volume is between 9.5pL and 10.5pL. Leading defects (level 2-3 defects): The ink pressure regulating unit adjusts the ink pressure in the nozzle cavity to the preset optimal range. The optimal pressure range varies for different ink types: 0.2MPa-0.25MPa for water-based inks, 0.25MPa-0.3MPa for oil-based inks, and 0.3MPa-0.35MPa for UV inks. At the same time, it corrects the jetting frequency. The jetting frequency adjustment range is 1kHz-10kHz, with an adjustment step of 100Hz. For example, if the original jetting frequency is 5kHz, it will be corrected to 4.8kHz or 5.2kHz depending on the defect, to ensure uniform ink droplet ejection and eliminate leading defects. For severe printing defects or color deviations (level 4-5 defects): First, initiate a pulse-type pre-cleaning operation. The electromagnetic reversing valve switches on and off at a preset frequency (10Hz), and the high-pressure air pump introduces pulsed high-pressure gas into the nozzle cavity. The pulse pressure is 0.3MPa-0.6MPa, and the pulse duration is 50ms-200ms, adjusted according to the defect level. Level 5 defects correspond to 0.6MPa pressure and 200ms duration, and level 4 defects correspond to 0.4MPa pressure and 100ms duration. After pre-cleaning, control adjacent nozzles to compensate for the increase in ink droplet volume, with an increase ratio of 10%-20%. At the same time, adjust the ejection sequence of adjacent nozzles, with a staggered time of 5μs-10μs to avoid ink droplet superposition interference and achieve rapid repair of defective areas. Furthermore, the adaptive control module is connected to an environmental sensing module, which is used to collect real-time data on the temperature, humidity, and air pressure of the printing environment. When constructing a 3D quality assessment model, the multimodal quality analysis module incorporates environmental perception data as auxiliary features into the model training. The adaptive control module dynamically adjusts the parameter thresholds of the control commands based on environmental parameters, enabling the control strategy to adapt to printing needs under different environmental conditions.

[0044] By receiving temperature, humidity, and air pressure data from the environmental sensing module, the threshold values ​​of the control parameters are dynamically adjusted. For example, when the temperature rises by 10°C, the optimal range of ink path pressure is lowered by 0.02 MPa, and the spray frequency is increased by 500 Hz; when the humidity rises by 20% RH, the pulse pressure of pre-cleaning is increased by 0.05 MPa, ensuring that the control strategy adapts to different environmental conditions.

[0045] Furthermore, the adaptive control module adopts a closed-loop feedback control mechanism. After outputting the control command, the image acquisition module acquires the repaired printed image in real time, and the multimodal quality analysis module evaluates the repair effect. If the repaired image still has defects and the defect level does not meet the standard, the control parameters are iteratively optimized according to the evaluation results until the printing defects are eliminated or the preset quality standard is reached.

[0046] Closed-loop feedback control involves the image acquisition module acquiring the repaired printed image in real time after the control command is output. One frame is transmitted to the multimodal quality analysis module every 50ms. If the defect level of the repaired image still does not meet the standard (not reduced to level 1 or below), the control parameters are iteratively optimized based on the evaluation results. The number of iterations does not exceed 5, and the parameter adjustment range of each iteration is 50% of the previous one, until the printing defect is eliminated or the preset quality standard is reached. The entire closed-loop control process takes no more than 300ms.

[0047] The environmental perception module works as follows: it collects temperature, humidity and air pressure data of the printing environment in real time, transmits them to the computing platform via I2C interface with a data transmission delay of less than 10ms, and transmits the collected data to the multimodal quality analysis module for model training and the adaptive control module for parameter adjustment, while storing it in the data storage and traceability module.

[0048] The adaptive control module is also connected to a data storage and traceability module. The data storage and traceability module is used to record image acquisition data, quality analysis results, faulty nozzle information, and control command parameters. It supports data query and traceability by printing task number and time range, and uses historical data to optimize the three-dimensional quality assessment model and control strategy parameters to achieve continuous iterative improvement of component performance.

[0049] The data storage and traceability module stores data categorized by print task number and timestamp. The stored content includes: raw image data from the image acquisition module (stored frame by frame, named in the format "task number-timestamp-frame number.jpg"), quality assessment results from the multimodal quality analysis module (JSON format, including defect type, level, area coordinates, etc.), faulty nozzle information from the nozzle status positioning module (CSV format, including nozzle number, physical coordinates, fault type, warning status, etc.), and control command parameters from the adaptive control module (XML format, including drive voltage, ink path pressure, spray frequency, pre-cleaning parameters, etc.). It supports data queries by print task number (e.g., "Task_20250601_001") or time range (e.g., "2025-06-01 08:00:00 to 2025-06-01 18:00:00"), with a query response time of no more than 1 second. Every 50 historical print task data are accumulated, the model optimization program is automatically started, using the gradient descent algorithm to update the parameters of the 3D quality assessment model and the threshold of the adaptive control strategy, so that the defect identification accuracy and control success rate of components are continuously improved. With each iteration, the identification accuracy is improved by at least 1%, and the control success rate is improved by at least 2%. Example

[0050] Based on Examples 1-5, such as Figure 2 As shown, the collaborative working process of the components in this embodiment in a real printing scenario is as follows: When a print job is started, the user selects the printing material (e.g., photo paper), ink type (e.g., water-based ink), and print resolution (e.g., 2400 dpi) through the printer control panel. The printer driver then synchronizes the relevant parameters to each module. The image acquisition module is activated. The adaptive focus adjustment unit adjusts the focal length to 50mm based on the paper thickness acquired by the laser thickness sensor, such as 0.2mm and 2400dpi resolution. The ring fill light unit adjusts the RGB light source brightness ratio to R:G:B=3:3:4 to eliminate paper reflection. The industrial cameras with two perspectives acquire images simultaneously and transmit them to the multimodal quality analysis module in real time. The environmental sensing module collects temperature (e.g., 28℃), humidity (e.g., 60%RH), and air pressure (e.g., 1013hPa) data in real time and transmits them to the multimodal quality analysis module and the adaptive control module. The multimodal quality analysis module preprocesses and extracts features from the acquired images, inputs them into a random forest classifier, identifies a level 2 missing print defect in a certain area, and integrates environmental data to optimize the analysis results, confirming that the defect is caused by nozzle blockage. Based on the pixel coordinates of the defect area and after temperature compensation correction, the nozzle status positioning module located the faulty nozzle as No. 12, with physical coordinates of (5.2mm, 3.8mm). At the same time, it counted the defect frequency of this nozzle in the last 100 frames of images as 3 times, with a cumulative working time of 1200s. Based on the life prediction model, the probability of a serious failure in the future is judged to be 30%, and no warning signal is output for the time being. After receiving the fault information, the adaptive control module controls the piezoelectric drive unit to adjust the drive voltage for the level 2 missing print defect, and fine-tunes the ink droplet volume of nozzle No. 12 from 10pL to 10.3pL with a fine-tuning accuracy of 3%. At the same time, it controls the ink droplet volume increment of adjacent nozzles No. 11 and No. 13 by 5% for dynamic compensation. Closed-loop feedback control is initiated. The image acquisition module acquires the repaired image, and the multimodal quality analysis module assesses and displays that the defect level has dropped to below level 1, meeting the preset quality standard. The control is then complete. The data storage and traceability module records the image data, quality analysis results, fault information of nozzle No. 12, and control parameters of this printing task, forming complete traceability data; After the printing task is completed, the data storage and traceability module accumulates the data for this task. Once 50 task data are accumulated, the data will automatically participate in the optimization of the model and control strategy.

[0051] The components in this embodiment, through the precise cooperation of each module, achieve real-time monitoring of nozzle status, accurate fault location, and adaptive control, effectively solving the problems of low control accuracy, slow response, and poor adaptability in the prior art. It can be widely used in various high-precision printing scenarios such as document printing, photo printing, and industrial parts printing, significantly improving print quality stability and printing efficiency, and reducing scrap rate.

[0052] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope described in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent nozzle control component for inkjet printers based on online image quality monitoring, characterized in that, include: The image acquisition module is used to acquire images of the target medium and dynamic images of the nozzle ejection process in real time. The multimodal quality analysis module uses grayscale difference calculation, ink droplet distribution density statistics, and color uniformity analysis to construct a three-dimensional quality assessment model. The three-dimensional quality assessment model is used to identify printing defects such as missing prints, streaks, and color deviations. In the construction of the three-dimensional quality assessment model, the multimodal quality analysis module first performs Gaussian filtering for noise reduction and morphological enhancement preprocessing on the acquired image. Then, it obtains the gray-level difference matrix of adjacent pixels through gray-level difference calculation. Combining the ink droplet spatial distribution entropy obtained from ink droplet distribution density statistics and the CIELab color space deviation value obtained from color uniformity analysis, the gray-level difference matrix, ink droplet spatial distribution entropy, and CIELab color space deviation value are used as input features to construct a random forest classifier, thereby achieving accurate identification and classification of defects such as missing prints, streaks, and color deviation. The nozzle status positioning module accurately locates the coordinates and fault type of faulty nozzles based on the printing defects in the three-dimensional quality assessment model and the preset nozzle-image pixel mapping relationship. The nozzle-image pixel mapping relationship is pre-constructed through calibration experiments. During the calibration process, standard test patterns are printed, and machine vision technology is used to obtain the image pixel area corresponding to each nozzle's ink droplet. A three-dimensional mapping table of nozzle number, physical coordinates, and image pixel coordinates is established. At the same time, a temperature compensation coefficient is introduced to dynamically correct the mapping relationship according to the printer's operating temperature, thereby improving the accuracy of fault nozzle location. The adaptive control module outputs differentiated control commands based on dynamic images and fault types, performing fine-tuning of ink droplet volume, correction of jet frequency, or pre-cleaning operations on faulty nozzles, while controlling adjacent nozzles to perform dynamic ink volume compensation, thereby achieving real-time repair of printing defects. The gray-level difference calculation is as follows: calculate the gray-level difference between adjacent pixels in the preprocessed image by row and column respectively, and construct an M×N gray-level difference matrix, where M and N are the height and width of the image in pixels, and the matrix element value is the absolute difference between the gray-level values ​​of adjacent pixels. The ink droplet distribution density statistics are as follows: ink droplet regions in the image are extracted using a connected component analysis algorithm. The connectivity determination of the connected component analysis algorithm adopts the 8-neighborhood rule. The pixel area of ​​the connected region of each ink droplet is counted, and the area ratio of the connected region of each ink droplet is calculated. , For the first The ratio of the pixel area of ​​a connected region of an ink droplet to the total pixel area of ​​all connected regions of ink droplets. Let be the total number of connected regions of the ink droplets, and calculate the spatial distribution entropy of the ink droplets using the following formula. : The base of the logarithm is 2; The color uniformity analysis involves converting the RGB image to the CIELab color space and calculating the color uniformity of each pixel. , and The deviation between the value and the standard color value is averaged over all pixels in the same color region of the image to obtain the total deviation value of the CIELab color space for that region. The input features of the random forest classifier are the mean and variance of the gray-level difference matrix, the spatial distribution entropy of the ink droplets, and the total deviation value of the CIELab color space. The output is the defect type and defect level. The defect type is missing print, streaks, and color deviation. The defect level is 1-5. The adaptive control module includes a piezoelectric drive unit and an ink path pressure adjustment unit; When the fault type is level 1-2 missing print, the driving voltage is adjusted by the piezoelectric drive unit to fine-tune the ink droplet volume. The voltage adjustment step is 0.1V. After each adjustment, 10 frames of images are acquired by the image acquisition module, and the multimodal quality analysis module evaluates the repair effect until the fine-tuning accuracy of the ink droplet volume of the faulty nozzle is controlled within 5%. When the fault type is level 2-3, the ink pressure in the nozzle cavity is adjusted to the preset optimal range through the ink path pressure adjustment unit. The optimal range is: 0.2MPa-0.25MPa for water-based ink, 0.25MPa-0.3MPa for oil-based ink, and 0.3MPa-0.35MPa for UV ink. The jetting frequency is also corrected. The jetting frequency adjustment range is 1kHz-10kHz, and the adjustment step is 100Hz. When the fault type is level 4-5 (missing print or color deviation), first control the faulty nozzle to perform a pulsed pre-cleaning operation. The electromagnetic reversing valve switches at a preset frequency of 10Hz, and the high-pressure air pump introduces pulsed high-pressure gas into the nozzle cavity. The pulse pressure is 0.3MPa-0.6MPa, and the pulse duration is 50ms-200ms, which is adjusted according to the defect level. Level 5 defects correspond to 0.6MPa pressure and 200ms duration, and level 4 defects correspond to 0.4MPa pressure and 100ms duration. After the pre-cleaning is completed, control the adjacent nozzles to perform ink droplet volume increment compensation, with an increment ratio of 10%-20%. At the same time, adjust the ejection sequence of the adjacent nozzles, with a staggered time of 5μs-10μs, to achieve rapid repair of printing defects.

2. The intelligent nozzle control component for inkjet printers based on online image quality monitoring according to claim 1, characterized in that, The Gaussian filtering denoising kernel size is set to 5×5, and the standard deviation σ=1.2; The morphological enhancement employs an opening operation that first dilates and then erodes, with the structuring element being a 3×3 rectangular structuring element. The random forest classifier contains 100 decision trees, with a maximum depth of 15 for each decision tree, a minimum number of sample splits of 2, and a minimum number of sample leaf nodes of 1. The preprocessing time for a single frame image is controlled within 20ms, the extraction time of the three types of features is not more than 30ms in total, the random forest classifier takes no more than 10ms to identify defects in a single frame image, the recognition accuracy is not less than 95%, and the defect level judgment error is not more than 1 level.

3. The intelligent nozzle control component for inkjet printers based on online image quality monitoring according to claim 2, characterized in that, The image acquisition module includes a high frame rate industrial camera, a ring-shaped fill light unit, and an adaptive focus adjustment unit. The frame rate of the high frame rate industrial camera is no less than 200fps; The adaptive focus adjustment unit automatically adjusts the focal length based on the thickness information of the printing medium and the printing resolution; The ring-shaped fill light unit uses an RGB three-color adjustable light source, which dynamically matches the fill light parameters according to the background color of the printing medium and the ink color to eliminate reflection interference and improve image contrast.

4. The inkjet printer nozzle intelligent control component based on online image quality monitoring according to claim 3, characterized in that, The image acquisition module also includes a multi-view acquisition unit, which simultaneously acquires dynamic images of the printing medium surface and the nozzle ejection side using at least two industrial cameras at different angles. After the multi-view images are synthesized by an image fusion algorithm, the three-dimensional trajectory and landing pattern information of the ink droplet ejection are obtained, providing comprehensive data support for the adaptive control module to determine the cause of the fault and formulate control strategies.

5. The inkjet printer nozzle intelligent control component based on online image quality monitoring according to claim 1, characterized in that, The specific process for constructing the nozzle-image pixel mapping relationship is as follows: The standard test pattern consists of a 10×10 grid-like ink dot array, with each ink dot having a diameter of 0.1 mm and a dot spacing of 0.5 mm, and the printing resolution is set to 2400 dpi. Using template matching algorithms in machine vision technology, the pixel coordinates of each ink dot are accurately located. Combined with the physical arrangement coordinates of the printer nozzles, a three-dimensional mapping table of nozzle number, physical coordinates and image pixel coordinates is established. The operating temperature is collected by a PT100 temperature sensor installed in the printer nozzle cavity. The temperature compensation coefficient K(T) = 1 + 0.002 × (T - 25), where T is the actual operating temperature. K(T) is used to correct the mapping relationship between pixel coordinates and physical coordinates. The correction formula is u' = u × K(T) and v' = v × K(T), so that the positioning error of the faulty nozzle does not exceed 0.05mm within the operating temperature range of 15℃-35℃.

6. The inkjet printer nozzle intelligent control component based on online image quality monitoring according to claim 5, characterized in that, The nozzle status positioning module also has a fault trend prediction function, which counts the frequency of defects and the change in defect level of the same nozzle in 100 consecutive frames of images in real time. Combined with the cumulative working time of the nozzle, a nozzle life prediction model is constructed. The nozzle life prediction model is implemented using an LSTM neural network. The input is the frequency of defects, the average defect level, and the increment of cumulative working time in the past 10 times. The output is the probability of a level 4-5 defect occurring in the next 5000 sprays. When the probability exceeds 80%, an early warning signal is output and a preventive control operation is triggered.

7. The inkjet printer nozzle intelligent control component based on online image quality monitoring according to claim 1, characterized in that, The adaptive control module is connected to an environmental sensing module, which is used to collect real-time temperature, humidity and air pressure data of the printing environment. The temperature, humidity, and air pressure data, after being normalized to the [0,1] interval, are incorporated as auxiliary features into the inference process of the random forest classifier. The classification results are adjusted by weighted summation, with weight coefficients of 0.3 for temperature, 0.4 for humidity, and 0.3 for air pressure. The adaptive control module dynamically adjusts the parameter thresholds of the control command according to environmental parameters: when the temperature rises by 10°C, the optimal range of ink path pressure is lowered by 0.02MPa and the spray frequency is increased by 500Hz; when the humidity rises by 20%RH, the pulse pressure of pre-cleaning is increased by 0.05MPa.

8. The inkjet printer nozzle intelligent control component based on online image quality monitoring according to claim 7, characterized in that, The adaptive control module adopts a closed-loop feedback control mechanism. After outputting the control command, the image acquisition module acquires the repaired printed image in real time and transmits one frame to the multimodal quality analysis module every 50ms. If the defect level of the repaired image does not meet the standard, the control parameters are iteratively optimized according to the evaluation results. The number of iterations does not exceed 5, and the parameter adjustment range of each iteration is 50% of the previous one, until the printing defect is eliminated or the preset quality standard is reached. The entire closed-loop control process takes no more than 300ms.

9. The inkjet printer nozzle intelligent control component based on online image quality monitoring according to claim 1, characterized in that, The adaptive control module is connected to a data storage and traceability module. The data storage and traceability module stores image acquisition data, quality analysis results, faulty nozzle information, and control command parameters according to the printing task number and timestamp. It supports data query by printing task number or time range, and the query response time does not exceed 1 second. Every 50 historical printing tasks, the model optimization program is automatically started, using the gradient descent algorithm to update the parameters of the 3D quality assessment model and the threshold of the adaptive control strategy.

Images