Industrial label printing adaptive optimization method based on multi-dimensional defect quantification

Abstract

The application discloses an industrial label printing self-adaptive optimization method based on multi-dimensional defect quantification. The method realizes intelligent management and control of label printing quality through task data preparation, image acquisition and parallel diagnosis, defect quantification and shunt decision, quality trend analysis and self-adaptive tuning, and predictive maintenance warning. The core of the method is to construct a defect severity index quantification model, combine the improved YOLOv8 algorithm to realize accurate classification and root diagnosis of defects, and realize real-time fine tuning of printing parameters through a fuzzy PID self-tuning tuning engine, and match a cache level real-time bit-level check to ensure data accuracy. The application breaks through the limitations of traditional passive detection, realizes active optimization and predictive maintenance, greatly improves production efficiency and reduces the rate of defective products, ensures stable label printing quality and accurate data, and is suitable for high-speed industrial label printing scenes.

Description

Technical Field

[0001] This invention belongs to the field of printer technology, and particularly relates to an adaptive optimization method for industrial label printing based on multidimensional defect quantification. Background Technology

[0002] In the industrial production sector, label printing is a crucial link in product identification, traceability, and anti-counterfeiting. Its printing quality directly affects product circulation, quality control, and brand reputation. As industrial production upgrades towards higher speed and intelligence, industrial label printing faces higher requirements for efficiency and accuracy. Existing industrial online barcode and printing defect detection systems are no longer sufficient to meet actual production needs.

[0003] The existing technology works by taking pictures with a high-speed camera and using template matching and barcode grading algorithms to determine whether the label is qualified. If a serious defect, such as an unreadable barcode, is detected, an alarm will sound and the machine will stop. However, this technology has several significant drawbacks: First, it can only achieve a simple binary judgment of "qualified / unqualified," failing to accurately distinguish the severity of defects. This can lead to minor and acceptable defects being mistakenly judged as unqualified, resulting in wasted consumables, and potentially overlooking moderate defects with potential risks. Second, once a defect is detected, it can only be handled by alarming and stopping the machine, requiring manual intervention to adjust printing parameters. This not only interrupts the production process and significantly increases unplanned downtime but also reduces the overall efficiency of industrial printing lines, making it unsuitable for high-speed continuous production. Third, it can only ensure the clear readability of variable data such as serial numbers, but cannot perform real-time comparison and verification of the accuracy of the data content itself. It is difficult to detect label errors caused by data transmission or processing errors, posing a significant risk to product traceability and anti-counterfeiting.

[0004] To address the problems of inaccurate defect quantification, lack of adaptive control capabilities, and incomplete data verification in existing technologies, and to realize the transformation of industrial label printing quality control from "passive detection" to "proactive, intelligent closed-loop adaptive optimization," there is an urgent need for an industrial label printing adaptive optimization method that can accurately quantify defects, automatically adjust printing parameters, and ensure data accuracy. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive optimization method for industrial label printing based on multidimensional defect quantification, so as to solve the problems mentioned in the background art.

[0006] In view of this, the present invention provides an adaptive optimization method for industrial label printing based on multidimensional defect quantification, comprising the following steps: S1: Task and data preparation. Receive print task instructions, preload the variable data to be printed and synchronize it to the print data cache. Variable data includes serial number and batch code. S2: Data acquisition and diagnosis. After the label is printed, the integrated vision acquisition unit acquires a full-frame high-definition image of the label during the high-speed movement of the label. The quality analysis processor runs the template comparison model, variable data OCR recognition model and deep learning defect classification model in parallel to perform diagnosis. At the same time, the OCR recognition results are compared with the original data in the printed data cache in real time at the bit level to complete the data verification. S3: Defect Quantification and Triage Decision. Based on the diagnostic results and data verification results from step S2, the defect quantification and diagnosis module calculates the defect severity index in real time, taking into account the defect type, area, and weight of the coverage rate of key information. The system makes a triage decision based on the defect severity index result. If the defect severity index is greater than the rejection threshold, a rejection operation is performed. If the defect severity index is less than the pass threshold, the label is judged as pass and enters trend analysis. If the defect severity index is between the rejection threshold and the pass threshold, the label is judged as pass and needs to enter trend analysis. S4: Quality Trend Analysis and Adaptive Tuning. The parameter adaptive tuning engine continuously analyzes the trend of the defect severity index and defect classification results of consecutive labels. If a systematic defect is found and the defect can be corrected by parameter adjustment, the minimum adjustment amount of the printer parameters is calculated and the adjustment instruction is sent to the printer to achieve real-time micro-parameter optimization without stopping the printer. A systematic defect is the trend of N consecutive labels showing the same type of defect. The printer parameters include printhead temperature, pressure and speed.

[0007] In a further embodiment of the present invention, the integrated visual acquisition unit in step S2 includes a high-speed camera and a programmable array light source. The high-speed camera is a high-resolution, high-frame-rate industrial camera. The programmable array light source includes a ring illumination mode, a coaxial illumination mode, and a low-angle illumination mode. Its illumination parameters are automatically adjusted by the quality analysis processor according to the current label material or ink color requirements.

[0008] In a further embodiment of the present invention, the deep learning defect classification model in step S2 is used to identify and classify physical defects of the types of ink spots, scratches, broken lines and wrinkles.

[0009] In a further embodiment of the present invention, the rejection operation in step S3 is specifically triggered by the quality analysis processor to mark or physically remove the label by the defective product processing execution mechanism, which is a high-speed pneumatic rejection arm or an inkjet marking unit.

[0010] In a further embodiment of the present invention, when the parameter adaptive tuning engine calculates the printer parameter adjustment amount in step S4, it uses the defect root cause classification results and the trend of defect severity index changes as a basis to ensure that the adjustment amount is minimized and does not cause new defects.

[0011] In a further embodiment of the present invention, step S5 is also included: predictive maintenance early warning. The system records and analyzes the changing trend of defect data. When the defect severity index of a certain type of defect shows a slow upward trend, an early warning is issued to prompt the operator to replace or maintain consumables.

[0012] In a further embodiment of the present invention, the consumables include a printhead and a ribbon.

[0013] In a further embodiment of the present invention, during the calculation of the defect severity index in step S3, the weight of the defect's coverage rate of key information is higher than the weight of the defect area, and the weight of the defect type is set according to the degree of impact of the defect on the label's functionality.

[0014] In a further embodiment of the present invention, the real-time bit-level comparison in step S2 is implemented by the vision system directly performing data synchronization comparison with the printer's print data buffer, without relying on external files or databases.

[0015] In a further embodiment of the present invention, the delay in sending and executing the parameter adjustment command in step S4 is controlled at the millisecond level to ensure real-time synchronization with the label printing process.

[0016] The beneficial effects of this invention are: 1. This invention utilizes a parameter adaptive tuning engine to automatically fine-tune printer operating parameters such as temperature, pressure, and speed based on defect diagnosis results. For example, when defects such as insufficient ink or blurred edges are detected, the engine can calculate and issue a minimum adjustment command in real time, enabling the printer to have self-repair capabilities and continuously maintain print quality at its best. This significantly reduces reliance on manual intervention, effectively prevents defects from escalating, and significantly improves the stability and consistency of label printing quality.

[0017] 2. This invention introduces a defect severity index, which quantitatively calculates defects by comprehensively considering factors such as defect type, area, and coverage of key information. This replaces the traditional simple binary judgment of "pass / fail," enabling hierarchical management of defects. Simultaneously, by leveraging a deep learning defect classification model, it can accurately diagnose the root causes of defects, such as distinguishing between printhead blockage, insufficient heating, or ribbon slippage. This provides clear guidance for equipment maintenance, facilitating targeted repairs and maintenance by staff and reducing maintenance costs.

[0018] 3. Unlike existing technologies that require alarm shutdowns and manual adjustments after defects are discovered, this invention can perform adaptive fine-tuning online, optimizing printing parameters without shutting down the machine. This effectively solves the problem of unplanned downtime, maximizes the continuous operation of the production line, and significantly improves the overall efficiency of industrial printing lines.

[0019] 4. This invention uses a vision system to perform real-time bit-level comparison directly with the printer's data buffer area, without relying on external files or databases. This not only ensures the clear readability of variable data such as serial numbers and batch codes, but also ensures that the printed content is completely consistent with the source data. This completely eliminates label content errors caused by data transmission or processing errors, and significantly improves the reliability of product traceability and anti-counterfeiting.

[0020] 5. By finely classifying defects through the severity index, it is possible to avoid incorrectly rejecting labels due to minor or acceptable defects, thus reducing material waste. At the same time, the parameter adaptive tuning mechanism can correct defects in a timely manner before they escalate to the point of requiring a shutdown, reducing the risk of producing defective products in batches, thereby reducing the cost of handling defective products and achieving more economical and efficient quality control.

[0021] 6. The system continuously records and analyzes the changing trends of defect data. When the severity index of a certain type of defect shows a slow upward trend, it can issue an early warning prompt, informing the operator to replace or maintain consumables such as printheads and ribbons in a timely manner. This realizes the upgrade from "post-event processing" to "pre-event prevention", effectively reducing the equipment failure rate, extending the service life of equipment, and ensuring the smooth operation of the production process. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the steps of the method of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0024] In the description of this application, it should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. For ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0025] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, the first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0026] It should be noted that in the description of this application, the directional terms such as "front, back, up, down, left, right", "horizontal, vertical, horizontal" and "top, bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description. Unless otherwise stated, these directional terms do not indicate or imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the scope of protection of this application. The directional terms "inner" and "outer" refer to the inner and outer contours relative to the outline of each component itself.

[0027] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0028] This embodiment provides an adaptive optimization method for industrial label printing based on multidimensional defect quantification. This method addresses the problems of existing technologies that only offer a simple binary judgment of "pass / fail," lack adaptive control capabilities, and suffer from incomplete data verification. It achieves a fundamental shift in industrial label printing quality control from "passive detection" to "active, intelligent closed-loop adaptive optimization." Specifically, it includes the following steps: S1: Task and Data Preparation. Upon receiving the print task instruction, the variable data to be printed is pre-loaded and synchronized to the print data cache. This variable data includes the serial number and batch code. A high-speed data link is directly established between the print data cache and the printer data interface, employing a dual-cache synchronization mechanism. While the current print task data is loaded into the main cache for real-time verification and comparison, the next batch of data to be printed is synchronously pre-loaded into the backup cache. Once the current batch is printed, the cache is immediately switched, ensuring zero-delay data loading and meeting the real-time data supply requirements of high-speed continuous printing scenarios. This step, by completing data synchronization and caching in advance, lays the data foundation for subsequent real-time bit-level verification, avoiding verification failures due to data transmission delays. It ensures the accuracy and timeliness of data verification from the source, significantly improving data acquisition efficiency and reliability compared to existing technologies that rely on external files or databases for data comparison.

[0029] S2: Data Acquisition and Diagnosis. After label printing, an integrated vision acquisition unit acquires a full-frame high-definition image of the label during its high-speed movement. A quality analysis processor runs in parallel a template comparison model, a variable data OCR recognition model, and a deep learning defect classification model for diagnosis. Simultaneously, the OCR recognition results are compared in real-time at the bit-level with the original data in the printed data cache to complete data verification. The integrated vision acquisition unit includes a high-speed camera and a programmable array light source. The high-speed camera is an industrial camera with a resolution of at least 5 megapixels and a frame rate of at least 120fps. Its lens focal length can be automatically adjusted according to the label size, ensuring that even in scenarios where the label movement speed is at least 300mm / s, a blur-free and distortion-free full-frame image can still be acquired. The programmable array light source includes a ring light... The system features three illumination modes: bright mode, coaxial illumination mode, and low-angle illumination mode. The illumination intensity can be steplessly adjusted within the range of 0-1000 lux, and the illumination color temperature can be switched between 3000K-6500K. The quality analysis processor analyzes the acquired image contrast, brightness uniformity, and other parameters in real time, and automatically adjusts the illumination parameters based on a preset image quality evaluation function. The image quality evaluation function is: Q=α×C+β×U+γ×S, where Q is the overall image quality score, C is the image contrast, U is the brightness uniformity, S is the signal-to-noise ratio, and α, β, and γ are weighting coefficients, with α+β+γ=1. Through extensive experiments, α=0.4, β=0.3, and γ=0.3 are calibrated to ensure optimal image acquisition results under different label materials (such as coated paper, synthetic paper, and PET film) and different ink concentrations. The deep learning defect classification model adopts an improved YOLOv8 algorithm. The optimization process is as follows: First, the backbone feature extraction part of the original YOLOv8 network is improved by introducing a C2f module to replace the original C3 module. The C2f module improves the depth of feature extraction and gradient propagation efficiency by adding branch convolutional layers and residual connections, enabling the model to better capture the subtle features of defects. Second, an adaptive feature pyramid (AFPN) structure is added to the neck feature fusion part. By dynamically adjusting the weight distribution of feature maps at different scales, the feature representation of small-sized defects (such as tiny ink spots and thin short lines) is enhanced. Finally, a decoupled detection head design is adopted in the detection head part, separating the classification task and the regression task, and using independent convolutional layers for feature mapping to reduce interference between tasks.The model was trained using an industrial label defect dataset containing over 100,000 samples. This dataset covers eight common physical defects, including ink spots, scratches, broken lines, and wrinkles, with each defect category containing samples of varying sizes, locations, and severity. The training process employed data augmentation techniques such as random cropping, flipping, rotation, and brightness / contrast adjustment to enhance sample diversity. An adaptive moment estimation (AdamW) optimizer was used for parameter updates, with an initial learning rate of 0.001. A cosine annealing learning rate scheduling strategy was used to dynamically adjust the learning rate, and the training iterations were set to 300 rounds. The final model achieved an accuracy of at least 99.2% and a recognition speed of at least 50 frames per second, accurately identifying and classifying physical defects such as ink spots, scratches, broken lines, and wrinkles. The template matching model uses a template matching algorithm based on normalized cross-correlation (NCC). First, a standard label image without defects is selected as the template image. The acquired image to be detected is then matched with the template image in blocks, and the NCC coefficient for each sub-block is calculated. The NCC coefficient calculation formula is: [Formula omitted for brevity]. Where T(x,y) is the pixel value of the template image at position (x,y), Tˉ is the average pixel value of the template image, I(x+u,y+v) is the pixel value of the image to be detected at position (x+u,y+v), Iˉ is the average pixel value of the corresponding sub-block of the image to be detected, and NCC(u,v) ranges from [-1,1]. When NCC(u,v) is greater than the set threshold (0.95), the sub-block is determined to match the template, that is, the fixed content is without defects; when NCC(u,v) is less than the threshold, the fixed content is determined to have defects. The variable data OCR recognition model employs the CRNN (Convolutional Recurrent Neural Network) algorithm, which consists of three parts: convolutional layers, recurrent layers, and transcription layers. The convolutional layers use a CNN network to extract visual features from the input image, mapping two-dimensional image features into a one-dimensional feature sequence. The recurrent layers use a bidirectional LSTM (Long Short-Term Memory) network to perform temporal modeling of the one-dimensional feature sequence, capturing the contextual dependencies between features. The transcription layer uses the CTC (Connection-Temporal Classification) algorithm to convert the feature sequence output by the recurrent layers into a character sequence, achieving end-to-end text recognition without the need for manual character position annotation. The model is specifically trained for numbers, letters, and some special characters in industrial labels, achieving a recognition accuracy of no less than 99.5%, and can quickly and accurately recognize variable data such as serial numbers and batch codes. The real-time bit-level comparison is achieved by the vision system directly synchronizing and comparing data with the printer's print data buffer, without relying on external files or databases. The comparison process is performed bit by bit, byte by byte. The specific steps are as follows: First, extract the binary data sequence D1 of the OCR recognition result, and simultaneously read the corresponding binary data sequence D2 of the original data from the print data buffer. Then, calculate the XOR value of the two data sequences. If the XOR result is all 0, the data is considered consistent. If there are non-zero bits, record the position and value of the non-zero bits and mark them as data errors. The response time of this comparison method is no more than 1ms, ensuring real-time data verification in high-speed printing scenarios and completely eliminating label content errors caused by data transmission or processing errors. Compared with the existing technology, which can only guarantee clear and readable data, this method significantly improves data accuracy.

[0030] S3: Defect Quantification and Triage Decision. The defect quantification and diagnosis module, based on the diagnostic results and data verification results from step S2, and considering the weights of defect type, area, and coverage rate of key information, calculates the defect severity index in real time. The system makes a triage decision based on the defect severity index result. If the defect severity index is greater than the rejection threshold, a rejection operation is performed; if the defect severity index is less than the acceptance threshold, the label is judged as acceptable and enters trend analysis; if the defect severity index is between the rejection threshold and the acceptance threshold, the label is judged as acceptable and enters trend analysis. The Defect Severity Index (DSI) is... The calculation uses a weighted summation formula: DSI = w1 × T + w2 × A + w3 × C, where T is the defect type weight coefficient, A is the defect area proportion coefficient, and C is the defect coverage coefficient for key information. The defect type weight T is set according to the degree of impact of the defect on the label's functionality, with data error defects T = 10.0 (most severe), unreadable barcode defects T = 8.5, broken lines defects T = 6.0, ink spot defects T = 3.0, scratch defects T = 2.5, wrinkles defects T = 2.0, and other minor defects T = 1.0. The defect area proportion coefficient A is calculated as A = S_defec t / S_label, where S_defect is the pixel area of ​​a single defect, calculated using a connected component labeling algorithm. The specific steps are: scanning the binarized defect image, labeling the pixel coordinates of each defect region, and counting the number of pixels in each region to obtain S_defect. S_label is the total pixel area of ​​the label, calculated from the physical size of the label and the image resolution. The value of A ranges from 0 to 1; when A > 0.05, A is calculated as 0.05 (i.e., the maximum contribution value is 0.05 × w²). The defect coverage coefficient C is calculated as C... =S_cover / S_key, where S_cover is the pixel area of ​​the key information region obscured by the defect, obtained by intersecting the defect region with the key information region (the pre-labeled barcode, serial number, and batch code area), and S_key is the total pixel area of ​​the key information region. The value of C ranges from 0 to 1. The weight coefficients w1, w2, and w3 are determined using the Analytic Hierarchy Process (AHP). First, a judgment matrix is ​​constructed, and five industry experts are invited to compare the importance of the three evaluation indicators (defect type, defect area, and key information obscuration rate) pairwise. The resulting judgment matrix is ​​as follows: Then, the maximum eigenvalue λ_max and the corresponding eigenvector of the judgment matrix were calculated. λ_max = 3.0092, and the normalized eigenvectors were [0.5404, 0.2973, 0.1623]. Passing the consistency test (consistency ratio CR = 0.0082 < 0.1), the judgment matrix showed satisfactory consistency. Finally, w1 = 0.5, w2 = 0.2, and w3 = 0.3 (rounded to one decimal place), with the weight of defect coverage of key information (w3) higher than the weight of defect area (w2), ensuring that defects related to key information are given priority consideration. The rejection and acceptance thresholds were determined using statistical process control (SPC) methods. Based on the defect severity index distribution of 10,000 acceptable labels in historical production data, a normal distribution was used to fit the DSI, which follows N(0.8, 0.2). 2 The defect index follows a normal distribution. Based on the 3σ principle, the acceptable threshold is set at DSI = 1.2 (μ + 2σ), indicating that 97.72% of acceptable labels have a DSI less than this value. The rejection threshold is set at DSI = 4.5 (μ + 18.5σ), ensuring that only severely defective labels are rejected. The rejection operation involves the quality analysis processor triggering a defective product processing mechanism to mark or physically remove the labels. The defective product processing mechanism is either a high-speed pneumatic rejection arm or an inkjet marking unit. The high-speed pneumatic rejection arm has a response time of no more than 5ms and a rejection accuracy of no less than 99.5%. The inkjet marking unit uses fast-drying ink, and the marking speed is synchronized with the printing speed, ensuring that defective products are accurately marked without affecting subsequent production processes. This step, through refined defect quantification calculations, replaces the simple binary judgment of existing technologies, achieving hierarchical management of defects. This avoids waste of consumables due to accidental rejection of minor defects and accurately identifies and promptly handles severe defects, improving the accuracy and economy of quality control.

[0031] S4: Quality Trend Analysis and Adaptive Tuning. The parameter adaptive tuning engine continuously analyzes the trend of defect severity index changes and defect classification results of consecutive labels. If a systematic defect is found and it can be corrected by parameter adjustment, the minimum adjustment amount of the printer parameters is calculated, and the adjustment command is sent to the printer to achieve real-time micro-parameter optimization without stopping the printer. A systematic defect is defined as the trend of N consecutive labels showing the same type of defect. The value of N is determined experimentally. Combining production efficiency and defect early warning sensitivity, N=5 is set (i.e., if 5 consecutive labels show the same type of defect, it is judged as a systematic defect). Printer parameters include printhead temperature, pressure, and speed. Among them, the parameter adaptive tuning engine... The system employs an improved tuning model based on the PID algorithm. The inputs to this model are the defect root cause classification results and the defect severity index (DSI) change trend (calculated by linear regression to determine the DSI slope k). The output is the adjustment amount for the printer parameters. First, a mapping database between defect types and printer parameters is established. For example, insufficient ink defects correspond to adjustments in printhead temperature and pressure parameters, fringed edges correspond to adjustments in print speed and pressure parameters, and scratches correspond to adjustments in printhead height and speed parameters. Then, the target parameters to be adjusted are determined based on the defect root cause classification results. The adjustment direction and initial adjustment amount are calculated based on the DSI slope k, and then iterative optimization is performed using the PID algorithm. The formula for the PID algorithm is: Where ΔP is the parameter adjustment amount, Kp is the proportional coefficient, Ki is the integral coefficient, Kd is the derivative coefficient, e(k) is the deviation between the current DSI and the target DSI (set to a passing threshold of 1.2), e(i) is the cumulative historical deviation, and e(k-1) is the deviation at the previous moment. To improve the stability and speed of tuning, a fuzzy PID algorithm is used to perform online self-tuning of Kp, Ki, and Kd. Based on the magnitude of the deviation e(k) and the rate of change of deviation ec(k) (ec(k) = e(k) - e(k-1)), it is divided into 5 fuzzy subsets (negative large, negative small, zero, positive small, positive large), and a fuzzy rule table is established. Through fuzzy inference and declarative processing, the values ​​of Kp, Ki, and Kd are adjusted in real time. For example, when e(k) is positive and ec(k) is negative, Kp is increased, Ki is decreased, and Kd is increased to speed up the response and reduce overshoot. When calculating printer parameter adjustments, the adaptive tuning engine uses the defect root cause classification results and the defect severity index trend as a basis. Through multiple iterations, it ensures that the adjustment amount is minimized and does not trigger new defects. Each parameter adjustment does not exceed 5% of the current parameter value. For example, if the printhead temperature is currently 120℃, the single adjustment will not exceed 6℃, avoiding print quality fluctuations due to sudden parameter changes. The sending and execution delay of parameter adjustment commands is controlled at the millisecond level (not exceeding 3ms), ensuring real-time synchronization with the label printing process and achieving real-time micro-parameter optimization without downtime. This step solves the drawback of existing technologies requiring manual parameter adjustments after defect detection. Through real-time adaptive tuning, the printer has self-repair capabilities, continuously maintaining optimal print quality, significantly reducing unplanned downtime and improving production efficiency.

[0032] In a further embodiment of this invention, step S5 is also included: predictive maintenance early warning. The system records and analyzes the changing trend of defect data, and smooths the defect severity index of 100 consecutive tags using a sliding window algorithm. The size of the sliding window is set to 10, that is, the average DSI value of the most recent 10 tags is calculated as the window output value each time. The calculation formula is as follows: Where DSIj is the mean DSI value of the j-th window, and DSI_i is the DSI value of the i-th label; then, the slope of the DSI mean change for each defect type is calculated, and linear regression is performed using the least squares method. The regression equation is: Where 'a' is the slope of change and 'b' is the intercept, when the severity index of a certain type of defect shows a slow upward trend (i.e., the slope of the average DSI value change over three consecutive sliding windows, 'a' > 0.02), an early warning is issued, prompting the operator to replace or maintain consumables. Consumables include printheads and ribbons. The system simultaneously records the cumulative usage time and number of prints of consumables, and establishes a consumable life prediction model based on defect trend data. The prediction model uses a multiple linear regression algorithm, and the regression equation is: Y = c0 + c1 × T + c2 × N + c3 × DSI, where Y is the remaining life of the consumable (hours), T is the cumulative usage time of the consumable (hours), N is the cumulative number of prints (sheets), DSI is the average DSI value of the most recent 100 labels, and c_0, c_1, c_2, and c_3 are regression coefficients obtained by fitting historical data. The model's prediction accuracy is no less than 90%, and it can issue consumable replacement warnings 5-8 hours in advance, achieving an upgrade from "post-event processing" to "pre-event prevention," reducing equipment failure rate and defect rate.

[0033] This embodiment achieves dynamic and precise matching between defect quantification and parameter tuning. By constructing a multi-dimensional defect severity index calculation model, combined with an improved YOLOv8 algorithm for defect classification and a fuzzy PID self-tuning tuning engine, a closed-loop control of "defect diagnosis - quantitative assessment - precise tuning" is formed, enabling millimeter-level precise correspondence between printer parameter adjustments and defect severity and root causes. Compared to traditional manual parameter tuning methods, the accuracy of parameter adjustments is improved by more than 85%, and the consistency error of label printing quality is controlled within ±2%, far exceeding the industry standard of ±5% error. At the same time, the defect repair response time is shortened to less than 10ms, achieving immediate defect repair.

[0034] This technology overcomes the bottleneck of the contradiction between high-speed printing and high-precision quality control. Through an integrated visual acquisition unit with adaptive lighting adjustment algorithms, a multi-model parallel computing architecture (simultaneous template comparison, OCR recognition, and deep learning classification), and a millisecond-level data verification mechanism, it achieves a defect identification accuracy rate of over 99.2% and a data verification accuracy rate of 100% while increasing label printing speed to 1.5 times that of traditional equipment (i.e., a maximum printing speed of 450mm / s). This technical solution, through the collaborative optimization of hardware modules and the parallel acceleration of algorithms, solves the long-standing industry problem of "speeding up inevitably leads to a decrease in quality," achieving simultaneous improvement in "high-speed production" and "high-quality control," resulting in an overall efficiency (OEE) improvement of over 40% for industrial label printing lines.

[0035] A comprehensive intelligent quality control system covering the entire product lifecycle has been established, integrating real-time defect handling, adaptive parameter optimization, and predictive maintenance early warning functions to form a full-chain control model of "in-process repair - pre-event prevention - post-event traceability." Through the predictive maintenance early warning model, the accuracy of consumable replacement timing has been improved to over 95%, unplanned equipment downtime has been reduced by 70%, and consumable waste rate has been reduced by 35%. Simultaneously, the system automatically records defect data, parameter adjustment records, quality judgment results, and corresponding production time and equipment status information for each label, forming an immutable quality archive. This provides data-driven decision support for production process optimization, driving a leapfrog upgrade in industrial label printing from "passive quality inspection" to "proactive intelligent control," reducing enterprise quality control costs by over 50% and improving the reliability and efficiency of product traceability by over 80%.

[0036] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An adaptive optimization method for industrial label printing based on multidimensional defect quantification, characterized in that, Includes the following steps: S1: Task and data preparation. Receive print task instructions, preload the variable data to be printed and synchronize it to the print data cache. Variable data includes serial number and batch code. S2: Data acquisition and diagnosis. After the label is printed, the integrated vision acquisition unit acquires a full-frame high-definition image of the label during the high-speed movement of the label. The quality analysis processor runs the template comparison model, variable data OCR recognition model and deep learning defect classification model in parallel to perform diagnosis. At the same time, the OCR recognition results are compared with the original data in the printed data cache in real time at the bit level to complete the data verification. S3: Defect Quantification and Triage Decision. The defect quantification and diagnosis module calculates the defect severity index in real time based on the diagnosis results and data verification results of step S2, combined with the defect type, area, and weight of the coverage rate of key information. The system makes a triage decision based on the defect severity index results. If the defect severity index is greater than the rejection threshold, then a rejection operation is performed; If the defect severity index is less than the acceptable threshold, the label is judged as acceptable and enters the trend analysis; If the defect severity index is between the rejection threshold and the acceptance threshold, the label is deemed acceptable and must proceed to trend analysis. S4: Quality trend analysis and adaptive tuning. The parameter adaptive tuning engine continuously analyzes the trend of the defect severity index and the defect classification results of continuous labels. If a systematic defect is found and the defect can be corrected by parameter adjustment, the minimum adjustment amount of the printer parameters is calculated and the adjustment instruction is sent to the printer to achieve real-time micro-parameter optimization without stopping the printer. Systematic defects are characterized by a trend of the same type of defect appearing on N consecutive labels. Printer parameters include printhead temperature, pressure, and speed.

2. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, The integrated vision acquisition unit in step S2 includes a high-speed camera and a programmable array light source. The high-speed camera is a high-resolution, high-frame-rate industrial camera. The programmable array light source includes ring illumination mode, coaxial illumination mode and low-angle illumination mode. Its illumination parameters are automatically adjusted by the quality analysis processor according to the current label material or ink color requirements.

3. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, The deep learning defect classification model in step S2 is used to identify and classify physical defects such as ink spots, scratches, broken lines, and wrinkles.

4. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, The rejection operation in step S3 specifically involves the quality analysis processor triggering the defective product processing actuator to mark or physically remove the label. The defective product processing actuator is either a high-speed pneumatic rejection arm or an inkjet marking unit.

5. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, In step S4, when the parameter adaptive tuning engine calculates the printer parameter adjustment amount, it uses the defect root cause classification results and the trend of defect severity index changes as a basis to ensure that the adjustment amount is minimized and does not cause new defects.

6. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, It also includes step S5: predictive maintenance early warning. The system records and analyzes the changing trend of defect data. When the defect severity index of a certain type of defect shows a slow upward trend, an early warning is issued to prompt the operator to replace or maintain consumables.

7. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 6, characterized in that, Consumables include printheads and ribbons.

8. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, In the calculation of the defect severity index in step S3, the weight of the defect's coverage of key information is higher than the weight of the defect area, and the weight of the defect type is set according to the degree of impact of the defect on the label's functionality.

9. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, The real-time bit-level comparison in step S2 is achieved by the vision system directly synchronizing and comparing data with the printer's print data buffer, without relying on external files or databases.

10. The adaptive optimization method for industrial label printing based on multidimensional defect quantization according to claim 1, characterized in that, The delay in sending and executing the parameter adjustment command in step S4 is controlled at the millisecond level to ensure real-time synchronization with the label printing process.

Images