A High-Precision Positioning Method for Screen Printing Based on Machine Vision

Abstract

This invention relates to the field of printing machinery technology, and particularly to a high-precision positioning method for screen printing based on machine vision, comprising: Step 1: Constructing a collaborative calibration and marking system; Step 2: Dynamic acquisition of deformation data: Within a set time window after the squeegee leaves the screen, images of the first type of marker points and the second type of target are acquired simultaneously; the center of the marker is located using a sub-pixel edge detection algorithm, and the difference between the actual spacing and the nominal spacing is calculated; Step 3: Generation of thermo-mechanical coupling compensation parameters; Step 4: Multi-degree-of-freedom collaborative positioning: Motion control commands for the substrate support platform and the screen tensioning mechanism are generated based on the translation compensation amount and rotation compensation angle; Positioning adjustment is completed and the marker overlap is verified before the start of the next printing cycle; Step 5: Iterative optimization of closed-loop error. This method achieves efficient perception, accurate modeling, and real-time compensation of multi-source errors such as thermal deformation, force-induced deformation, environmental disturbances, and material properties during the screen printing process.

Description

Technical Field

[0001] This invention relates to the field of printing machinery technology, and in particular to a high-precision positioning method for screen printing based on machine vision. Background Technology

[0002] Screen printing technology is widely used in solder paste printing processes in the microelectronics packaging field, and its positioning accuracy directly affects the chip packaging yield. With the popularization of micro-components such as 0201 and 01005, the pad size has been reduced to below 100μm, requiring a positioning accuracy of ±5μm between the substrate and the screen. However, existing positioning technologies face the following inherent defects:

[0003] There is a significant difference in the coefficient of thermal expansion (CTE) between ceramic substrates (such as Al2O3 and AlN) and stainless steel mesh used in microelectronic packaging. The CTE range of ceramic substrates is 4.5-7.5ppm / ℃, while that of stainless steel mesh is 16-18ppm / ℃.

[0004] When the ambient temperature fluctuates by ±3℃ (typical temperature control deviation in an SMT workshop), a 100mm×100mm substrate and stencil will experience a relative deformation of 8-12μm. Existing mechanical positioning systems rely on rigid clamps to fix the substrate and cannot detect non-uniform expansion caused by local temperature gradients (such as curvature changes caused by temperature differences between the substrate edge and center). This problem is particularly prominent in multilayer wiring structures on ceramic substrates, as the internal metal layers exacerbate thermal deformation anisotropy.

[0005] In addition, under the action of squeegee pressure (typical value 0.3-0.5MPa), the screen undergoes elastic tensile deformation, which causes the printed pattern to shift relative to the substrate pads. The deformation in the central area of ​​the screen can reach 15-25μm, and the deformation in the squeegee travel direction is non-linearly distributed.

[0006] Traditional static positioning schemes only perform position calibration once before printing, which cannot correct dynamic deformation caused by squeegee pressure. More seriously, the screen rebound hysteresis effect causes the deformation to persist after the squeegee is lifted, resulting in drift of the positioning reference of adjacent substrates during continuous printing.

[0007] Therefore, there is an urgent need for a high-precision positioning method for screen printing based on machine vision to solve the above problems. Summary of the Invention

[0008] To achieve the above objectives, the present invention provides a high-precision positioning method for screen printing based on machine vision, comprising:

[0009] Step 1: Construct a collaborative calibration and labeling system:

[0010] At least three asymmetrically distributed first-type markers are etched in the non-printing area of ​​the screen printing plate, and second-type targets corresponding to the first-type markers are processed at the edge of the printed substrate.

[0011] The design value of the center distance between the first type of marker point and the target is greater than the set multiple of the maximum expected thermal deformation of the substrate;

[0012] Step 2: Dynamic acquisition of deformation data:

[0013] Within a set time window after the scraper leaves the screen, images of the first type of marker points and the second type of target are acquired simultaneously;

[0014] The center of the mark is located by subpixel edge detection algorithm, and the difference between the actual distance and the nominal distance is calculated. The actual distance is the physical distance between the center point of the screen mark and the center point of the substrate target, and the nominal distance is the theoretical distance preset during the design.

[0015] Step 3: Generation of thermo-mechanical coupling compensation parameters:

[0016] Based on the difference, real-time scraper pressure, and changes in ambient temperature, the local deformation rate of the screen is calculated using a pre-established deformation mapping model.

[0017] Input the deformation rate into the compensation decision model, and output the substrate translation compensation amount and the screen rotation compensation angle;

[0018] Step 4: Multi-degree-of-freedom cooperative localization:

[0019] The motion control commands for the substrate support platform and the screen tensioning mechanism are generated based on the translation compensation amount and the rotation compensation angle.

[0020] Complete the positioning adjustments and verify the marking overlap before the start of the next printing cycle;

[0021] Step 5: Iterative optimization of closed-loop error:

[0022] After printing, the actual outline of the solder paste is collected, and its deviation vector from the design position is extracted;

[0023] When the deviation vector exceeds the set threshold, the weight parameters of the compensation decision model are dynamically updated.

[0024] Preferably, the construction of the collaborative calibration and labeling system in step 1 specifically includes:

[0025] The first type of marker is a circular pit with an infrared high reflectivity metal layer coated on the surface. The pit depth is determined by the set ratio of the screen thickness. The coating adopts a composite structure of first depositing a nickel layer and then depositing a gold layer. The thickness of the nickel layer is controlled within the range that makes the surface roughness of the substrate less than the set threshold. The thickness of the gold layer meets the condition of total reflection of the wavelength of the industrial camera light source.

[0026] The second type of target is a cross-shaped protrusion structure embedded with near-infrared sensitive material, and the height of the protrusion is less than a set ratio of the solder paste thickness;

[0027] The asymmetric distribution must satisfy the following: the interior angle of the triangle formed by any three marker points is greater than the minimum angle threshold calculated by the maximum deformation rate of the screen, and the minimum spacing between marker points is greater than a set multiple of the single grid size, and this multiple is calculated based on the calibrated value of the thermal expansion coefficient of the substrate.

[0028] Preferably, the dynamic acquisition of deformation data in step 2 includes:

[0029] The set time window is determined in the following way: after the scraper is lifted, a first set time is delayed to eliminate mechanical vibration. This time is calculated based on the screen tension value and elastic modulus. The image acquisition window needs to cover the period during which the screen's free oscillation decays to a steady-state threshold, and the total time is less than a set fraction of the period of the screen's free oscillation.

[0030] The specific process of subpixel edge detection is as follows: Gaussian filtering is performed on the marked image, and then Laplacian convolution is performed to detect zero-crossing points as initial edges. The marked image is an image of the first type of marked points and the second type of target acquired simultaneously. The gray-level gradient distribution is scanned along the edge normal direction at subpixel resolution, and the center coordinates are determined by fitting the extreme points of the gradient curve. The fitting process is optimized using the least squares method, and the objective function is the sum of squared errors between the gradient magnitude and the coordinate offset.

[0031] Preferably, the construction of the deformation mapping model in step 3 includes:

[0032] A stepped pressure sequence is applied to the screen at standard temperature. After each pressure step stabilizes, the displacement of the marked point is measured by a laser interferometer. The displacement measurement accuracy reaches the submicron level.

[0033] A polynomial response surface model was established based on the pressure-displacement dataset. The independent variables of the model include pressure value, temperature change and material thermal expansion coefficient, and the dependent variable is deformation rate.

[0034] The order of the polynomial is determined by cross-validation: the order is gradually increased until the error reduction rate of the test set is less than a set threshold, and the final model coefficients are solved using the ridge regression algorithm to avoid overfitting.

[0035] Preferably, the update of the compensation decision model in step 3 includes:

[0036] The weight parameter update adopts a proportional-integral-derivative (PID) control algorithm, and its integral gain coefficient is dynamically adjusted according to the consistency of the historical weight change direction.

[0037] The integral gain coefficient is adjusted by monitoring the convergence trend of the most recent deviation vectors: when the rate of decrease of the standard deviation of the magnitude of the deviation vector for a set number of consecutive cycles exceeds a threshold, the coefficient is increased; otherwise, it is decreased.

[0038] The model input layer adds environmental humidity sensor data as a compensation term, and the humidity compensation coefficient is calibrated through a constant humidity chamber control experiment.

[0039] Preferably, the generation of motion control commands in step 4 includes:

[0040] The calculation of the substrate support stage displacement introduces rotation center offset compensation. The offset is obtained by inverting the measured displacement curve of the laser tracker at a set rotation angle. The offset is the physical displacement deviation of the actual rotation center of the substrate support stage relative to the theoretical center.

[0041] The calculation of the rotation angle of the mesh tensioning mechanism needs to be integrated with real-time tension feedback. The tension value is measured by the micro-strain gauge array installed at the four corners of the mesh frame. The output voltage of the strain gauge is converted into the tension value through a Wheatstone bridge, and then the equivalent elastic modulus is calculated back according to Hooke's law.

[0042] The motion actuator uses a composite transmission of piezoelectric ceramic driver and servo motor. The piezoelectric ceramic is responsible for micron-level translational compensation, and the servo motor is responsible for milliradian-level rotational compensation.

[0043] Preferably, the specific implementation of real-time tension feedback includes:

[0044] The micro-strain gauge array is arranged along the diagonal of the screen, and each array contains at least three strain gauges distributed in an equilateral triangle.

[0045] When calculating the tension value, temperature drift compensation is first performed on the output values ​​of each strain gauge, and the compensation coefficient is calibrated through temperature rise experiments under no-load conditions.

[0046] The local stress concentration factor of the mesh is calculated based on the strain distribution gradient of the triangular region, and this factor is used to calculate the overall tension by weighted averaging.

[0047] Preferably, the extraction of the deviation vector in step 5 includes:

[0048] The solder paste contour acquisition uses a combination of coaxial and lateral lighting, with the ratio of coaxial light intensity to lateral light intensity adjusted according to the light reflection characteristics of the solder paste components.

[0049] The contour edge detection uses an adaptive threshold segmentation algorithm. The threshold calculation formula includes a weighted term of the mean and standard deviation of the background gray level. The weight coefficients are obtained through training with printing quality acceptance samples.

[0050] The geometric center is located using the gray-scale weighted centroid method, where the pixel gray values ​​are weighted by a Gaussian function to calculate the geometric center of the solder paste outline;

[0051] The calculated actual center coordinates of the solder paste are compared with the preset design coordinates to obtain the deviation vector.

[0052] Preferably, it also includes a pre-printing calibration step:

[0053] After the equipment is preheated, the standard calibration board is aligned with the screen markings by driving the precision displacement platform, and the pixel correction matrix of each industrial camera is recorded.

[0054] Move the calibration plate to the set grid position and collect the mapping relationship between the actual displacement and pixel coordinates;

[0055] Bilinear interpolation is used to generate a full field-of-view correction matrix and store it in the control system.

[0056] Preferably, the verification of the overlap of the markings in step 4 includes:

[0057] Simultaneously capture superimposed images of the first type of markers and the second type of targets using a third industrial camera;

[0058] Calculate the residual between the measured value and the theoretical value of the mark center distance. When the residual exceeds the set tolerance, trigger the recalculation of the compensation amount.

[0059] The recalculation uses an iterative approximation algorithm: the current residual is used as input to generate a compensation increment until the residual is less than the tolerance or the set number of iterations is reached.

[0060] The beneficial effects of this invention are:

[0061] 1. This invention constructs a collaborative calibration marking system, setting spatially corresponding marker points and targets between the screen and the substrate, with the distance between them designed as a set multiple of thermal deformation. Simultaneously, images of the first type of marker points and the second type of target are dynamically acquired, and combined with real-time ambient temperature data, a pre-trained thermo-mechanical coupling deformation model is used to calculate the precise deformation rate and achieve local compensation. This effectively suppresses alignment errors caused by CTE differences and ambient temperature fluctuations, achieving full-cycle, high-precision thermal compensation control.

[0062] 2. This invention employs an asymmetric multi-point marking system, combined with sub-pixel-level image analysis and high-precision coordinate fitting, to capture nonlinear warping or anisotropic expansion of the substrate or screen caused by temperature differences. Simultaneously, the compensation decision model incorporates temperature gradient and material thermal expansion parameters as inputs, ensuring the system can adapt to complex deformations under non-uniform thermal field conditions. This design significantly improves positioning stability and robustness under complex thermal fields.

[0063] 3. This invention acquires images within a specific time window after the scraper detaches, avoiding the influence of mechanical vibration and capturing residual stress and screen deformation caused by the scraper in real time. Through a combination of piezoelectric ceramics and a servo motor, the system can achieve micron-level translational compensation and precise angle adjustment, effectively eliminating the damage to positioning accuracy caused by the scraper's movement.

[0064] 4. This invention employs a closed-loop error iterative optimization mechanism. After each printing, the solder paste outline image is compared with the design drawing to extract the deviation vector for model self-learning and dynamic correction. Simultaneously, by adding multiple input dimensions such as humidity and tension, the adaptability of the compensation model to multi-physical coupling deformation is further enhanced, thereby achieving true online real-time adaptive positioning.

[0065] 5. This invention performs a precise calibration process before printing, using a standard calibration plate combined with a displacement platform and thermal imager monitoring to achieve full-field pixel equivalent calibration and correction under thermal equilibrium conditions, which significantly reduces system initialization errors and image processing errors caused by visual distortion.

[0066] 6. Through a marker overlap verification mechanism, the system captures an image of overlapping marker points before each cycle and calculates the measured residual. When the residual exceeds the tolerance, iterative compensation calculation is automatically triggered to ensure that the system always maintains the optimal printing alignment state. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0068] Figure 1 This is a flowchart of the steps of the method of the present invention;

[0069] Figure 2 This is a flowchart of step 5 of the method of the present invention for extracting the deviation vector;

[0070] Figure 3 This is a flowchart of the verification steps for marking overlap in step 4 of the method of the present invention. Detailed Implementation

[0071] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0072] Please see Figures 1-3 This invention provides a high-precision positioning method for screen printing based on machine vision. In step 1, at least three first-type marker points (e.g., Φ80μm circles) are formed in the non-printing area of ​​the screen by laser etching or chemical corrosion. The three points are asymmetrically distributed (e.g., right triangles) to ensure uniqueness and avoid positioning ambiguity.

[0073] On the edge of a printed circuit board (such as an Al2O3 ceramic substrate), a second type of target is formed by laser marking or metal deposition, which corresponds to the space of the first type of marker points. The number and distribution position strictly match the projection position of the screen marker points and are easily recognized by machine vision.

[0074] It is understandable that "the center distance between the marker and the target is greater than the set multiple of the maximum expected thermal deformation of the substrate" means that, assuming the maximum thermal deformation is 12μm, the center distance of the marker can be set to more than 10mm, so that the percentage error caused by relative thermal deformation is reduced to 0.12%, ensuring measurement accuracy.

[0075] In step 2, after the squeegee detaches from the screen (e.g., within 100ms–300ms, to avoid vibration interference), a dual industrial camera system (equipped with a 50X lens) simultaneously captures high-resolution images of the first type of marker points and the second type of target. Subpixel edge detection (e.g., Sobel-Canny gradient algorithm + centroid fitting) is used to obtain the actual center coordinates of the marker points and targets. The difference between the actual spacing (projection measurement result) and the designed nominal spacing is calculated and used as the input for real-time thermo-mechanical coupling deformation measurement. The actual spacing is the physical distance between the center point of the screen marker and the center point of the substrate target, while the nominal spacing is the theoretical distance preset during design.

[0076] In step 3, the above difference, combined with the real-time acquired ambient temperature change (±0.1℃ accuracy) and scraper pressure (acquired in real-time by a force sensor, accuracy ±0.01MPa), is input into a pre-established deformation mapping model (generated based on finite element simulation and experimental calibration, stored as a fitting table or neural network regression model). The model outputs the local deformation rate of the screen printing plate (μm / mm), and calculates the overall scaling and rotation trends caused by thermal expansion of the screen printing plate based on the offset direction. The deformation rate is input into the compensation decision model (including PID fine-tuning or gradient descent optimization) to generate the corresponding substrate translation compensation amount (μm level) and screen printing plate rotation compensation angle (0.001° accuracy).

[0077] In step 4, the compensation amount is converted into motion control commands for the substrate carrier stage (XY translation stage) and the screen tensioning mechanism (rotation + tension adjustment), and micron-level translation and angle fine adjustment are executed through the servo control card.

[0078] After all motion localizations are completed and stabilized before the next printing cycle, the overlap between the marker points and the target image is verified again to ensure that the system completes the closed-loop compensation.

[0079] In step 5, after printing is completed, machine vision is used to collect the actual printed pattern outline of the solder paste, compare it with the CAD design file, and extract the key pad position deviation vector.

[0080] If the deviation exceeds the set threshold (e.g., ±3μm), the dynamic update of the weight parameters of the compensation decision model will be automatically triggered (e.g., small-batch incremental training or calibration curve adjustment) to adapt to the effects of long-term thermal drift, screen fatigue or tension decay.

[0081] Through the above implementation steps, the embodiments of the present invention solve the positioning error caused by complex coupling problems such as heat, force, dynamic deformation, and environmental drift in solder paste printing in the field of microelectronic packaging, significantly improve the stability, accuracy and consistency of screen printing in the packaging of micro-components such as 0201 and 01005, and provide highly reliable and mass-producible technical support for high-end packaging production lines.

[0082] In one possible implementation, recesses are drilled at appropriate locations on the screen. A substrate material with excellent plating properties (such as a high-strength alloy or engineering plastic) is selected. Using chemical plating or electroplating techniques, a nickel layer is first deposited on the surface of the recesses, controlling the nickel layer thickness so that the surface roughness of the recesses is below a set threshold (e.g., below nanometer-level roughness). A gold layer is then further deposited on the nickel layer, with a thickness sufficient to ensure total internal reflection for a specified light source wavelength (typically in the near-infrared band, such as 850 nm).

[0083] Near-infrared sensitive materials, such as photosensitive polymers doped with specific metals, are selected and embedded into the cross-shaped raised structure. The height of the raised section needs to be proportional to the solder paste thickness to ensure that the shape of the solder paste is not affected during printing, while also meeting the requirements of machine vision recognition.

[0084] Asymmetrical distribution design requires ensuring that the interior angles of any triangle formed by three marker points are greater than a specific minimum angle to improve resistance to deformation and improve recognition capabilities. The minimum spacing of the marker points is adjusted according to the substrate's coefficient of thermal expansion to ensure a stable distance ratio during printing.

[0085] By utilizing a nickel-gold composite coating and near-infrared responsive materials, the marking system exhibits high reflectivity and excellent contrast in machine vision, significantly improving positioning accuracy. Through asymmetrical distribution and a rational marking dot spacing design, it ensures strong resistance to thermal deformation of the screen during use, maintaining printing precision. The system can adapt to changes in light source and temperature in different printing environments, offering excellent application flexibility.

[0086] In one possible implementation, the lifting of the squeegee after completing a printing stroke causes mechanical vibration, which can affect image sharpness and accuracy. Therefore, a delay time is introduced after the squeegee lifts, calculated based on the screen tension and the elastic modulus of its material. These parameters determine the duration of the mechanical vibration, allowing a delay time to be set to ensure vibration attenuation.

[0087] The duration of image acquisition is strictly controlled within a set proportion of the free oscillation period of the screen. This setting ensures that the movement of the screen is as smooth as possible during image acquisition, reducing image deviation caused by oscillation.

[0088] To refine edge detection, Gaussian filtering is applied to the labeled image to reduce noise interference and ensure accuracy. The Laplacian convolution method is used to locate the initial edges of the image. Zero-crossing point detection provides a preliminary framework for further edge analysis.

[0089] Based on the initial edge, a detailed grayscale gradient distribution scan is performed along the edge normal direction. By fitting the extreme points of the gradient curve, the precise center coordinates at the sub-pixel level are calculated.

[0090] The least squares method is used for fitting and optimization. The objective function of this scheme is designed as the sum of squared errors between the gradient magnitude and the coordinate offset. The positioning accuracy is improved by optimizing the sum of squared errors.

[0091] In one possible implementation, different pressure steps are gradually applied to the screen in a temperature-controlled standard environment to ensure that each pressure step reaches stability.

[0092] Displacement of marked points is detected using a laser interferometer. Due to the extremely high measurement accuracy of the laser interferometer, displacement measurements can be made with sub-micron precision. This is crucial for accurately evaluating the deformation response of the screen printing plate.

[0093] Based on the measurement data, a polynomial response surface model was constructed. The independent variables of the model include the applied pressure, the temperature change, and the coefficient of thermal expansion of the material, while the dependent variable is the deformation rate. The combination of these factors can comprehensively reflect the deformation behavior under actual printing conditions.

[0094] To ensure a good fit and predictive ability, cross-validation is used to gradually increase the order of the polynomial. This optimizes the model's complexity, avoiding problems such as reduced predictive ability due to oversimplification or overfitting due to overcomplexity. Furthermore, the criterion for determining the order is that the test set error reduction rate is less than a predetermined threshold, ensuring the model's accuracy and robustness.

[0095] To avoid overfitting, the ridge regression algorithm is used to solve for the coefficients of the multinomial model in the final model. This regularization method can effectively control the complexity of the model, enabling it to exhibit stronger generalization ability when dealing with real data.

[0096] Through precise laser interferometry and complex polynomial modeling, the deformation of the screen printing plate under different pressure and temperature conditions can be predicted with high accuracy. This is crucial for improving printing precision. The multivariate characteristics of the model, along with optimizations through cross-validation and ridge regression, give it greater adaptability and stability under different operating environments and material properties. By compensating for and adjusting the deformation in real time, the precision and product quality of screen printing can be significantly improved, and the defect rate reduced.

[0097] In one possible implementation, to improve model training efficiency and convergence speed, motive gradient descent is used to update the weight parameters of the compensation decision model. Momentum gradient descent can utilize gradient information from previous iterations in each update, thereby accelerating model convergence.

[0098] The momentum factor is dynamically adjusted based on the consistency of the direction of historical weight changes: if the gradient direction is consistent in the past few rounds, the momentum factor will be increased accordingly to promote the convergence process faster; if the direction fluctuates greatly, the momentum factor will be reduced to avoid convergence instability caused by over-adjustment.

[0099] The learning rate adjustment mechanism automatically adjusts the learning rate by monitoring the convergence trend of the bias vector in the most recent iterations. When the rate of decrease of the standard deviation of the bias for a set number of consecutive iterations exceeds a certain preset threshold, it indicates that the optimization trend of the model is clear, and the learning rate can be appropriately increased to accelerate the optimization process.

[0100] If the rate of decrease of the standard deviation of the bias slows down, the learning rate is reduced to avoid the model getting stuck in local optima or oscillations, and to ensure smooth convergence to the global optimum.

[0101] To better adapt to environmental factors, humidity sensor data has been added to the model's input layer. This data reflects humidity changes during the printing process, which directly affect the deformation and positioning accuracy of the printing material.

[0102] To accurately quantify the impact of humidity on screen printing deviations, it is necessary to calibrate the compensation coefficients under different humidity conditions through constant humidity chamber experiments. The relationship between humidity and printing error is obtained experimentally and incorporated into the model as a compensation term.

[0103] The combination of momentum gradient descent and dynamically adjusted momentum factors makes the optimization process more efficient and stable, reducing the oscillation problems that may occur during training in traditional gradient descent, thereby accelerating training convergence. By monitoring the convergence trend of the bias in real time and adjusting the learning rate, the model can dynamically optimize the learning process according to the actual situation, avoiding convergence instability caused by an excessively large learning rate and improving the accuracy of the final model.

[0104] By incorporating humidity sensor data, the compensation model not only relies on visual information but also senses and compensates for the impact of environmental humidity changes on printing accuracy. This design ensures that the printing process maintains high accuracy under various environmental conditions, especially in environments with significant humidity variations, where the introduction of the compensation coefficient effectively reduces errors caused by humidity changes.

[0105] The calibration of the humidity compensation coefficient ensures that environmental factors accurately reflect the model's predictions and compensation results, improves the accuracy of the printing process, and reduces errors. It has significant practical application value, especially in the field of high-precision screen printing.

[0106] By optimizing the compensation decision model and combining the gradient descent method with momentum, the adaptive learning rate adjustment mechanism, and the humidity compensation strategy, this method significantly improves the positioning accuracy of the screen printing process and has higher stability and adaptability.

[0107] In one possible implementation, the displacement of the substrate carrier stage needs to be calculated with high precision during the screen printing process to ensure accurate printing position. Rotation center offset is a significant factor affecting printing accuracy, especially when the rotation angle is large; rotational deviation of the carrier stage can lead to displacement errors.

[0108] The offset is obtained by measuring the actual displacement curve of the substrate at a set rotation angle using a laser tracker, and then using an inversion method to obtain the offset of the rotation center. This compensation, by calculating the actual displacement curve, accurately corrects the error caused by the offset of the rotation center, ensuring precise positioning.

[0109] In screen printing, the tension of the screen has a significant impact on print quality. The angle calculation of the screen tensioning mechanism requires real-time feedback of the screen tension. To accurately measure the tension, an array of micro-strain gauges installed at the four corners of the screen frame can sense changes in tension.

[0110] The micro-strain gauge array measures the strain value generated by deformation and outputs a voltage signal. This signal is converted into a tension value via a Wheatstone bridge, and then, using Hooke's law, the equivalent elastic modulus of the screen is derived, further calculating the change in screen tension. This data provides real-time tension feedback for the motion control system, avoiding positioning errors caused by uneven tension.

[0111] The motion actuator uses a composite transmission system of piezoelectric ceramic driver and servo motor to precisely control the movement of the substrate and screen during the screen printing process.

[0112] Among them, piezoelectric ceramic actuators are mainly used for micrometer-level translational compensation. Piezoelectric ceramics have extremely high displacement accuracy, enabling precise control of minute displacements, and are suitable for micrometer-level motion compensation.

[0113] Servo motor: Responsible for milliradian-level rotational compensation. Servo motors can provide high torque and precise angle control, ensuring that every change in angle during rotation can be accurately compensated.

[0114] The combination of the two ensures that both translation and rotation during the printing process can be efficiently and accurately compensated at different precision levels, optimizing the accuracy and response speed of motion control.

[0115] By precisely controlling rotational offset, screen tension, and micron-level motion, the positioning accuracy and printing quality in the screen printing process are significantly improved. Its composite drive system combines the advantages of piezoelectric ceramics and servo motors, providing more precise and stable compensation, making this technology of significant application value in the field of high-precision screen printing.

[0116] In one possible implementation, micro-strain gauge arrays are mounted diagonally on the screen, with each array consisting of at least three strain gauges arranged in an equilateral triangle. The advantage of this arrangement is its ability to efficiently sense strain changes at different locations on the screen, particularly the non-uniformity of the screen's tension distribution.

[0117] Strain gauge arrays can monitor minute deformations of a screen in real time, measure local strain, and thus calculate tension changes on the screen. This arrangement method allows for comprehensive strain feedback in key areas of the screen (such as the four corners and the central area), ensuring the accuracy and comprehensiveness of tension monitoring.

[0118] The voltage signal output by a micro-strain gauge is affected by temperature changes, leading to data instability. To eliminate the influence of temperature on the measurement results, temperature drift compensation is required.

[0119] The temperature drift compensation coefficient was obtained through temperature rise experiments under no-load conditions. The compensation coefficient was determined by experimentally measuring the output changes of micro-strain gauges under different temperature conditions. Then, during actual measurements, the output values ​​of each strain gauge were corrected based on this compensation coefficient, thereby eliminating errors caused by temperature changes and ensuring the accuracy of the tension data.

[0120] The strain distribution of a mesh is usually not uniform, especially with stress concentration in localized areas. To accurately calculate the overall tension, it is necessary to consider the stress concentration in these localized areas.

[0121] The local stress concentration factor of the screen can be calculated based on the strain distribution gradient within the triangular region. This factor takes into account the strain distribution and stress concentration effects in different regions, transforming the local stress situation into global tension compensation data.

[0122] The local stress concentration factor is used for weighted averaging to calculate the overall tension. This method, through accurate modeling of local stress, avoids the problem that single tension calculation methods cannot handle local tension fluctuations, thus making the calculation of overall tension more accurate.

[0123] By precisely arranging the micro-strain gauge array, compensating for temperature drift, and calculating the local stress concentration factor, this real-time tension feedback system can effectively improve the accuracy, stability, and adaptability of the screen printing process, providing strong support for high-precision printing.

[0124] In one possible implementation, embodiments of the present invention employ a combined coaxial and lateral lighting scheme to acquire solder paste contours. The combination of coaxial and lateral lighting enhances image contrast under different reflectivity characteristics, thereby obtaining clearer and more accurate solder paste contours.

[0125] The light reflection properties of solder paste components affect the intensity of reflected light. Therefore, the ratio of coaxial light intensity to lateral light intensity needs to be adjusted according to the solder paste composition to ensure optimal illumination. Different solder paste components (such as metal powder, resin, etc.) may have different effects on light reflection. Adjusting the illumination ratio can effectively improve image quality and the accuracy of contour extraction.

[0126] The contour edge detection employs an adaptive threshold segmentation algorithm, a method that automatically adjusts the threshold based on the specific characteristics of the image. This method can flexibly adjust the threshold according to different lighting conditions, solder paste thickness, and other factors to ensure accurate detection of solder paste edges.

[0127] The threshold calculation formula includes a weighted term based on the mean and standard deviation of the background grayscale. This formula makes edge detection more accurate and adaptable to different background lighting variations. The weighting coefficients are obtained through training on print quality acceptance samples. This process ensures the algorithm's adaptability and accuracy under various printing conditions. For example, through training, the algorithm can be adjusted in real-world printing environments to recognize different types of solder paste or printing materials, reducing errors caused by environmental factors.

[0128] After contour detection, the geometric center was located using a gray-scale weighted centroid method. This method not only considers the positional information of each pixel in the image, but also weights the pixels according to their gray-scale values, enabling a more accurate determination of the geometric center of the solder paste contour.

[0129] The calculated actual center coordinates of the solder paste are compared with the preset design coordinates to obtain the deviation vector.

[0130] When calculating the geometric center of the solder paste outline, using Gaussian weighting can reduce errors caused by image noise or local irregularities. The Gaussian weighting effect means that pixels with higher grayscale values ​​have a greater impact on centroid calculation within the central region of the solder paste outline, thus improving the accuracy of centroid positioning, especially when the solder paste distribution is irregular, ensuring accurate positioning even in such cases.

[0131] The deviation vector extraction method, by optimizing the process of solder paste contour acquisition, edge detection, and geometric center positioning, significantly improves the positioning accuracy and quality control capability in the screen printing process while ensuring accuracy and robustness, and has broad application prospects.

[0132] In one possible implementation, during calibration, a precision displacement platform is first used to align the standard calibration plate with the markings on the screen. The displacement platform allows for precise movement of the calibration plate, ensuring accurate alignment with the markings on the screen. This step is crucial for ensuring the calibration plate is correctly positioned, providing a reliable foundation for subsequent calibration processes.

[0133] The calibration plate has known markers or grids with known physical locations, so the relationship between pixels and actual physical coordinates can be established by recording the pixel coordinates when the industrial camera captures these markers.

[0134] After aligning the calibration plate with the screen markings, the pixel correction matrix of each industrial camera is recorded. The correction matrix describes the mapping relationship between pixel coordinates within the camera's field of view and actual physical coordinates. The pixel equivalent of each camera is independent because different cameras have different viewing angles, focal lengths, and other factors, which may lead to differences in the correspondence between pixels and actual positions.

[0135] The calibration matrix is ​​constructed by moving the calibration board to a preset grid position within the field of view. In this way, the actual displacement of the calibration board at different positions and the corresponding pixel coordinates are recorded.

[0136] After recording the mapping relationship between actual displacement and pixel coordinates, a bilinear interpolation algorithm is used to generate a full field-of-view correction table. This interpolation method can calculate the pixel value at any position within the field of view based on the known coordinates and pixel values ​​of points, thus providing an accurate pixel-physical coordinate mapping for positions in all fields of view.

[0137] The calibration process should be performed after the equipment has preheated. This is because thermal expansion may occur during heating, leading to slight changes in the mechanical structure that could affect calibration accuracy. The preheating time should be determined based on the uniformity of the temperature field monitored by the thermal imager to ensure that the equipment reaches a stable operating temperature during calibration, eliminating errors caused by temperature variations.

[0138] A precise pre-printing calibration process can not only improve the positioning accuracy of machine vision systems, but also ensure the stability and consistency of equipment under different working conditions, thereby significantly improving the quality and efficiency of screen printing.

[0139] In one possible implementation, a third industrial camera is used to capture images related to the printing process. This camera is capable of simultaneously acquiring superimposed images of a first type of marker points and a second type of target. The first type of marker points typically refer to specific patterns or markings used for positioning, while the second type of target may be the printing target or other important reference points. Through this superimposed image, the overlap between the marker points and the target can be compared and analyzed to determine whether it meets predetermined accuracy requirements.

[0140] After capturing and overlaying images, the system calculates the center-to-center distance of the marker points (i.e., the distance between marker points) based on the image information and compares it with the theoretical value. The theoretical value is a calibration value derived from the printing template or target design position, while the measured value is the distance between the center points of the markers captured by an industrial camera in actual applications.

[0141] The difference between the measured value and the theoretical value is called the residual. When the residual exceeds the set tolerance, it indicates that there is a deviation in the positioning accuracy, and subsequent compensation calculations must be triggered.

[0142] If the residual between the measured and theoretical values ​​of the mark center distance exceeds the set tolerance, the compensation amount will be recalculated. The purpose of the compensation amount calculation is to correct errors in the machine vision system and ensure printing accuracy. The system will automatically compensate based on the current residual to correct positioning errors caused by various factors.

[0143] The compensation amount is calculated using an iterative approximation algorithm. This algorithm is implemented through the following steps:

[0144] Using the current residual as input parameter, calculate the initial compensation increment.

[0145] The calculated compensation increment is corrected, and a new compensation value is generated.

[0146] Repeat this process until the residual is less than the set tolerance or the preset maximum number of iterations is reached.

[0147] The iterative approximation algorithm can continuously adjust the compensation amount, gradually approach the optimal value, and ensure that the residual eventually reaches the allowable range, thereby achieving high-precision positioning correction.

[0148] By verifying the overlap between the marker points and the target, errors between the actual marker points and the theoretical target positions can be detected in a timely manner and corrected through compensation recalculation. This maintains high positioning accuracy throughout the printing process, avoiding the cumulative effect of initial errors and thus improving print quality.

[0149] By employing an iterative approximation algorithm to recalculate the compensation amount, the compensation can be adaptively adjusted based on the actual residual situation, avoiding accuracy deviations caused by excessive or insufficient compensation in a single step. The iterative process makes the compensation correction more accurate, gradually reducing errors and ultimately achieving the ideal accuracy standard.

[0150] By automatically triggering the recalculation of compensation quantities, the entire process requires no manual intervention and can monitor and correct errors in real time. The system automates error detection and correction, greatly improving production efficiency and reducing the complexity and error rate of manual operations.

[0151] By combining the difference between actual residuals and theoretical values ​​for precise compensation calculations, the system can effectively address errors caused by factors such as equipment variations, temperature fluctuations, or material inconsistencies. Regardless of operating conditions, the system maintains high stability and robustness, ensuring efficient production over extended periods.

[0152] Regularly verifying and compensating for mark overlap can effectively prevent substandard product quality or defective products caused by insufficient positioning accuracy. This method ensures accurate execution of every print job, thereby reducing scrap rates and rework costs, and improving overall production efficiency.

[0153] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0154] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A high-precision positioning method for screen printing based on machine vision, characterized in that, include: Step 1: Construct a collaborative calibration and labeling system: At least three asymmetrically distributed first-type markers are etched in the non-printing area of ​​the screen printing plate, and second-type targets corresponding to the first-type markers are processed at the edge of the printed substrate. The design value of the center distance between the first type of marker point and the target is greater than the set multiple of the maximum expected thermal deformation of the substrate; Step 2: Dynamic acquisition of deformation data: Within a set time window after the scraper leaves the screen, images of the first type of marker points and the second type of target are acquired simultaneously; The center of the mark is located by subpixel edge detection algorithm, and the difference between the actual distance and the nominal distance is calculated. The actual distance is the physical distance between the center point of the screen mark and the center point of the substrate target, and the nominal distance is the theoretical distance preset during the design. Step 3: Generation of thermo-mechanical coupling compensation parameters: Based on the difference, real-time scraper pressure, and changes in ambient temperature, the local deformation rate of the screen is calculated using a pre-established deformation mapping model. Input the deformation rate into the compensation decision model, and output the substrate translation compensation amount and the screen rotation compensation angle; Step 4: Multi-degree-of-freedom cooperative localization: The motion control commands for the substrate support platform and the screen tensioning mechanism are generated based on the translation compensation amount and the rotation compensation angle. Complete the positioning adjustments and verify the marking overlap before the start of the next printing cycle; Step 5: Iterative optimization of closed-loop error: After printing, the actual outline of the solder paste is collected, and its deviation vector from the design position is extracted; When the deviation vector exceeds the set threshold, the weight parameters of the compensation decision model are dynamically updated.

2. The high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, Step 1, which involves building a collaborative calibration and labeling system, specifically includes: The first type of marker is a circular pit with an infrared high reflectivity metal layer coated on the surface. The pit depth is determined by the set ratio of the screen thickness. The coating adopts a composite structure of first depositing a nickel layer and then depositing a gold layer. The thickness of the nickel layer is controlled within the range that makes the surface roughness of the substrate less than the set threshold. The thickness of the gold layer meets the condition of total reflection of the wavelength of the industrial camera light source. The second type of target is a cross-shaped protrusion structure embedded with near-infrared sensitive material, and the height of the protrusion is less than a set ratio of the solder paste thickness; The asymmetric distribution must satisfy the following: the interior angle of the triangle formed by any three marker points is greater than the minimum angle threshold calculated by the maximum deformation rate of the screen, and the minimum spacing between marker points is greater than a set multiple of the single grid size, and this multiple is calculated based on the calibrated value of the thermal expansion coefficient of the substrate.

3. The high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, Step 2, dynamic acquisition of deformation data, includes: The set time window is determined in the following way: after the scraper is lifted, a first set time is delayed to eliminate mechanical vibration. This time is calculated based on the screen tension value and elastic modulus. The image acquisition window needs to cover the period during which the screen's free oscillation decays to a steady-state threshold, and the total time is less than a set fraction of the period of the screen's free oscillation. The specific process of subpixel edge detection is as follows: Gaussian filtering is performed on the marked image, and then Laplacian convolution is performed to detect zero-crossing points as initial edges. The marked image is an image of the first type of marked points and the second type of target acquired simultaneously. The gray-level gradient distribution is scanned along the edge normal direction at subpixel resolution, and the center coordinates are determined by fitting the extreme points of the gradient curve. The fitting process is optimized using the least squares method, and the objective function is the sum of squared errors between the gradient magnitude and the coordinate offset.

4. The high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, Step 3, the construction of the deformation mapping model, includes: A stepped pressure sequence is applied to the screen at standard temperature. After each pressure step stabilizes, the displacement of the marked point is measured by a laser interferometer. The displacement measurement accuracy reaches the submicron level. A polynomial response surface model was established based on the pressure-displacement dataset. The independent variables of the model include pressure value, temperature change and material thermal expansion coefficient, and the dependent variable is deformation rate. The order of the polynomial is determined by cross-validation: the order is gradually increased until the error reduction rate of the test set is less than a set threshold, and the final model coefficients are solved using the ridge regression algorithm to avoid overfitting.

5. The high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, The update of the compensation decision model in step 3 includes: The weight parameter update adopts a proportional-integral-derivative (PID) control algorithm, and its integral gain coefficient is dynamically adjusted according to the consistency of the historical weight change direction. The integral gain coefficient is adjusted by monitoring the convergence trend of the most recent deviation vectors: when the rate of decrease of the standard deviation of the magnitude of the deviation vector for a set number of consecutive cycles exceeds a threshold, the coefficient is increased; otherwise, it is decreased. The model input layer adds environmental humidity sensor data as a compensation term, and the humidity compensation coefficient is calibrated through a constant humidity chamber control experiment.

6. The high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, Step 4, generating motion control commands, includes: The calculation of the substrate support stage displacement introduces rotation center offset compensation. The offset is obtained by inverting the measured displacement curve of the laser tracker at a set rotation angle. The offset is the physical displacement deviation of the actual rotation center of the substrate support stage relative to the theoretical center. The calculation of the rotation angle of the mesh tensioning mechanism needs to be integrated with real-time tension feedback. The tension value is measured by the micro-strain gauge array installed at the four corners of the mesh frame. The output voltage of the strain gauge is converted into the tension value through a Wheatstone bridge, and then the equivalent elastic modulus is calculated back according to Hooke's law. The motion actuator uses a composite transmission of piezoelectric ceramic driver and servo motor. The piezoelectric ceramic is responsible for micron-level translational compensation, and the servo motor is responsible for milliradian-level rotational compensation.

7. The high-precision positioning method for screen printing based on machine vision according to claim 6, characterized in that, The specific implementation of real-time tension feedback includes: The micro-strain gauge array is arranged along the diagonal of the screen, and each array contains at least three strain gauges distributed in an equilateral triangle. When calculating the tension value, temperature drift compensation is first performed on the output values ​​of each strain gauge, and the compensation coefficient is calibrated through temperature rise experiments under no-load conditions. The local stress concentration factor of the mesh is calculated based on the strain distribution gradient of the triangular region, and this factor is used to calculate the overall tension by weighted averaging.

8. The high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, Step 5, the extraction of the deviation vector, includes: The solder paste contour acquisition uses a combination of coaxial and lateral lighting, with the ratio of coaxial light intensity to lateral light intensity adjusted according to the light reflection characteristics of the solder paste components. The contour edge detection uses an adaptive threshold segmentation algorithm. The threshold calculation formula includes a weighted term of the mean and standard deviation of the background gray level. The weight coefficients are obtained through training with printing quality acceptance samples. The geometric center is located using the gray-scale weighted centroid method, where the pixel gray values ​​are weighted by a Gaussian function to calculate the geometric center of the solder paste outline; The calculated actual center coordinates of the solder paste are compared with the preset design coordinates to obtain the deviation vector.

9. The high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, It also includes a pre-printing calibration step: After the equipment is preheated, the standard calibration board is aligned with the screen markings by driving the precision displacement platform, and the pixel correction matrix of each industrial camera is recorded. Move the calibration plate to the set grid position and collect the mapping relationship between the actual displacement and pixel coordinates; Bilinear interpolation is used to generate a full field-of-view correction matrix and store it in the control system. The preheating time is determined based on the uniformity of the equipment temperature field monitored by the thermal imager.

10. A high-precision positioning method for screen printing based on machine vision according to claim 1, characterized in that, Step 4, the verification of the overlap of the markings, includes: Simultaneously capture superimposed images of the first type of markers and the second type of targets using a third industrial camera; Calculate the residual between the measured value and the theoretical value of the mark center distance. When the residual exceeds the set tolerance, trigger the recalculation of the compensation amount. The recalculation uses an iterative approximation algorithm: the current residual is used as input to generate a compensation increment until the residual is less than the tolerance or the set number of iterations is reached.

Images