Dispensing path intelligent planning method and system based on machine vision profile features

Abstract

The application discloses a point-gluing path intelligent planning method and system based on machine vision profile features, relates to the technical field of process vision detection, and comprises the following steps: acquiring an initial profile of a workpiece to be point-glued and generating an initial point-gluing path to carry out point-gluing operation. In the operation, a formed glue line image is collected, real-time profile data is obtained through sub-pixel level edge extraction, and three types of flow front states, i.e., spreading state, accumulation state and fracture state, are distinguished. Correction weights of subsequent path segments are calculated according to the flow front states, and the path segments are gradually corrected to obtain a first corrected point-gluing path, and a control mechanism is executed to cyclically carry out image collection and path correction. The application solves the technical problem that the existing point-gluing control can only compensate for position deviation after the fact and cannot actively adjust subsequent process parameters according to the real-time flow state of the glue, achieves the technical effect that the execution parameters of subsequent processes are continuously perceived and gradually adjusted according to the spreading state of the glue, and realizes closed-loop adaptive control of the glue line forming quality.

Description

Technical Field

[0001] This invention relates to the field of process vision inspection technology, and in particular to a method and system for intelligent planning of dispensing paths based on machine vision contour features. Background Technology

[0002] Precision dispensing is widely used in electronic packaging and component sealing. The quality of the adhesive line formation directly determines the product yield, making adaptive control of the dispensing path based on visual intelligence a necessity in the industry. Existing vision-based closed-loop dispensing solutions rely solely on rigid coordinate compensation, only correcting adhesive line position deviations, which is insufficient to adapt to spreading, accumulation, and breakage defects caused by the flow fluctuations of transparent adhesives. Traditional vision methods only inspect the formed adhesive line after the fact, failing to determine the state of the adhesive flow front and unable to adaptively adjust subsequent path process parameters based on real-time morphology. Traditional vision compensation methods cannot predict the continuous impact of fluid defects on subsequent adhesive paths, nor can they adjust dispensing speed, nozzle height, and residence time according to the gradual changes in defect distance, easily leading to continuous forming defects. The detection and control data is singular and lagging, making it difficult to achieve precise and dynamic control of the entire dispensing process. Summary of the Invention

[0003] This application provides a method and system for intelligent dispensing path planning based on machine vision contour features, which solves the technical problem that existing dispensing control can only compensate for position deviations after the fact and cannot actively adjust subsequent process parameters according to the real-time flow state of the adhesive.

[0004] The first aspect of this application provides an intelligent dispensing path planning method based on machine vision contour features. The method includes: acquiring an initial contour of the workpiece to be dispensed, generating an initial dispensing path, and performing dispensing based on the initial dispensing path; during the dispensing process, continuously acquiring images of formed adhesive lines through a vision sensing module, and performing sub-pixel-level edge extraction on the formed adhesive line images to obtain real-time contour data of the adhesive lines; determining the current flow front state of the adhesive based on the real-time contour data, wherein the flow front state includes at least a spreading state, an accumulation state, and a broken state; calculating correction weights for subsequent dispensing path segments based on the flow front state, and performing gradual corrections on the dispensing path segments based on the correction weights to generate a first corrected dispensing path; controlling a dispensing mechanism to perform operations according to the first corrected dispensing path, and iteratively cyclically iterating the vision acquisition and path correction processes.

[0005] A second aspect of this application provides an intelligent dispensing path planning system based on machine vision contour features. The system includes: an initial dispensing path acquisition module, used to acquire the initial contour of the workpiece to be dispensed, generate an initial dispensing path, and perform dispensing based on the initial dispensing path; a glue line contour data acquisition module, used to continuously acquire images of formed glue lines through a vision sensing module during the dispensing process, and perform sub-pixel-level edge extraction on the formed glue line images to obtain real-time contour data of the glue line; a flow front state discrimination module, used to discern the current flow front state of the glue based on the real-time contour data, the flow front state including at least a spreading state, an accumulation state, and a broken state; a corrected dispensing path generation module, used to calculate the correction weight of subsequent dispensing path segments based on the flow front state, and perform gradual correction on the dispensing path segments based on the correction weight to generate a first corrected dispensing path; and a corrected dispensing path execution module, used to control the dispensing mechanism to perform operations according to the first corrected dispensing path, and iteratively cycle through the vision acquisition and path correction processes.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application acquires real-time images of the formed adhesive line, extracts its contour features, determines the adhesive flow conditions, calculates the influence weight based on the defect location, and dynamically adjusts the process parameters corresponding to the remaining processing path according to the strength of the influence, continuously updating the parameters in a loop. This adaptively controls the subsequent dispensing process, reduces defects caused by the continuous spread of forming defects, and achieves closed-loop precise control of the entire dispensing process. It achieves the technical effect of continuously sensing the adhesive spreading status and smoothly adjusting the execution parameters of subsequent processes, thus realizing closed-loop adaptive control of the adhesive line forming quality. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating the intelligent dispensing path planning method based on machine vision contour features provided in this application embodiment.

[0009] Figure 2 This is a schematic diagram of the structure of the intelligent dispensing path planning system based on machine vision contour features provided in the embodiments of this application.

[0010] Figure labeling: Initial dispensing path acquisition module 1, glue line contour data acquisition module 2, flow front state discrimination module 3, corrected dispensing path generation module 4, corrected dispensing path execution module 5. Detailed Implementation

[0011] This application provides a method and system for intelligent dispensing path planning based on machine vision contour features, which solves the technical problem that existing dispensing control can only compensate for position deviations after the fact and cannot actively adjust subsequent process parameters according to the real-time flow state of the adhesive.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, an intelligent dispensing path planning method based on machine vision contour features is described, wherein the method includes: Obtain the initial outline of the workpiece to be glued, generate an initial glue dispensing path, and perform glue dispensing based on the initial glue dispensing path.

[0015] In this embodiment of the application, the workpiece to be glued refers to various electronic components and precision structural parts that need to be sealed, bonded or encapsulated through the glue dispensing process.

[0016] Specifically, when acquiring the initial contour of the workpiece to be glued, an industrial area array camera fixedly installed next to the dispensing mechanism is used to capture a complete top surface image of the workpiece. The acquired color image is converted to grayscale, converting the three-channel pixel values ​​into single-channel grayscale values. Then, Gaussian filtering is used to denoise the grayscale image, eliminating noise interference from workpiece surface reflections and ambient stray light. The Canny edge detection algorithm is used to extract the edge pixels of the glue-to-be-glued area on the workpiece surface, obtaining a coarsely extracted set of edge pixels. Then, the Zernike moment sub-pixel edge localization method is used to optimize the sub-pixel accuracy of the coarsely extracted edge pixels. Specifically, a 7×7 pixel Zernike moment template is used to perform convolution operations in the neighborhood of each coarsely extracted edge pixel. The real and imaginary parts of the Zernike moment corresponding to each pixel are calculated. The rotation invariance principle of Zernike moments is used to establish the edge position parameter solution equation, and the sub-pixel level edge position parameters are obtained, thereby improving the edge localization accuracy from the pixel level to the sub-pixel level. Finally, the initial contour coordinate point set of the workpiece to be glued with accurate coordinate information is obtained.

[0017] The segmented preprocessing of the initial dispensing path will be described in detail in subsequent embodiments. Then, based on the obtained initial contour coordinate point set, an initial dispensing path is generated. According to the preset glue line forming width requirement, a fixed-distance offset calculation is performed along the normal direction of each coordinate point of the initial contour to obtain the set of points on the dispensing head's travel trajectory. The specific process of offset calculation is as follows: For each contour point in the initial contour coordinate point set, firstly, using the position coordinates of this point and its two adjacent contour points, the tangent direction vector at this point is calculated using the difference method. Then, the tangent direction vector is rotated 90 degrees around the normal of the plane where the point is located to obtain the normal direction vector at this point. According to the preset glue line forming width requirement, half of the preset width value is used as the offset distance, and the contour point is translated outward by the offset distance along the normal direction vector to obtain the corresponding offset point coordinates. After traversing all contour points and performing the above calculations sequentially, the initial offset point set is obtained. Local self-intersection detection is performed on the initial set of offset points to identify self-intersection regions of the offset curves caused by concave angle offsets. For offset point clusters that have undergone self-intersection, an invalid offset point within the self-intersection loop is removed using a trimming method. The offset points at the self-intersection boundary are then directly connected by straight line segments to form a set of offset trajectory points without self-intersection. This set of offset trajectory points without self-intersection is the set of travel trajectory points for the dispensing head.

[0018] Then, the B-spline curve interpolation algorithm is used to smooth the walking trajectory point set, making the trajectory corner positions transition smoothly without abrupt changes, forming a continuous and complete initial dispensing path. Specifically: all coordinate points in the walking trajectory point set are sequentially used as input model points. The chord length parameterization method is used to normalize and parameterize each model point to obtain the node parameter values ​​corresponding to each model point. Based on the node parameter values, the cumulative chord length method is used to determine the node vector of the B-spline curve. The repetition degree of the first and last nodes of the node vector is set to 4 to ensure that the fitted curve passes through the first and last model points. The cubic uniform B-spline basis function is used as the interpolation kernel function. All model points and their corresponding node parameter values ​​are substituted into the B-spline curve equation, and the coordinates of each control point are calculated by solving the tridiagonal matrix equation system. The control point sequence and node vector obtained by the inverse calculation are substituted into the cubic B-spline curve equation, and point-by-point interpolation sampling is performed along the curve arc length direction at equal parameter intervals to generate a smooth trajectory coordinate sequence composed of dense coordinate points. This smooth trajectory coordinate sequence is the smoothed continuous and complete initial dispensing path.

[0019] Finally, the motion coordinate data corresponding to the initial dispensing path is transmitted to the motion control unit of the dispensing mechanism. The motion control unit generates multi-axis linkage motion control commands based on the motion coordinate data, driving the dispensing mechanism to move the glue head along the initial dispensing path. Simultaneously with the glue head's movement, the glue supply system opens the dispensing valve according to the preset dispensing pressure, allowing the glue to flow continuously and adhere to the workpiece surface, completing the dispensing process based on the initial dispensing path.

[0020] During the dispensing process, the visual sensing module continuously acquires images of the formed glue line and performs sub-pixel-level edge extraction on the formed glue line images to obtain real-time contour data of the glue line.

[0021] Optionally, the vision sensing module includes an industrial area scan camera, a diffused ring light source, and an image acquisition unit. The vision sensing module is fixedly mounted on a bracket at the side and rear of the dispensing head, maintaining a fixed relative position and moving synchronously with the dispensing head. The diffused ring light source is arranged around the lens of the industrial area scan camera, illuminating the area of ​​the formed adhesive line on the workpiece surface. It uses an oblique incidence method to supplement the illumination, weakening reflections from the adhesive surface and interference from ambient light. The image acquisition unit is connected to the industrial area scan camera via a data cable, responsible for receiving the image signals output by the camera and transmitting them to the main control processing module. The main control processing module is communicatively connected to the motion control unit and the image processing unit, used to coordinate the timing and data interaction between vision acquisition, image processing, and motion control.

[0022] After the dispensing process begins, the main control processing module sends continuous acquisition commands to the vision sensing module, and the industrial area scan camera triggers shooting according to the preset acquisition interval. The specific acquisition interval is matched with the dispensing travel speed to ensure that the glue line area covered by two adjacent frames partially overlaps, avoiding acquisition blind spots. After each frame of image acquisition is completed, the image acquisition unit transmits the raw image data to the image processing unit of the main control processing module for subsequent edge extraction calculations.

[0023] Next, the image processing unit preprocesses the received pre-formed adhesive line image, converting the color image into a single-channel grayscale image, and then uses Gaussian filtering to smooth and denoise the grayscale image, filtering out random noise points. After preprocessing, the Canny edge detection algorithm is used to coarsely extract pixels-level edges from the adhesive line image. By setting high and low thresholds, edge pixels that meet the gradient conditions are selected, resulting in a set of pixel-level edge points on both sides of the adhesive line. Based on this, the Zernike moment sub-pixel edge localization method is used to precisely locate each pixel-level edge point. By calculating the Zernike moment parameters in the corresponding neighborhood, the sub-pixel coordinate positions of the edge points are obtained, and finally, the real-time contour data of the adhesive line, including the coordinates of the left and right edges, is output. The real-time contour data of the adhesive line is transmitted to the state discrimination unit in the main control processing module for subsequent determination of the flow front state based on the real-time contour data.

[0024] Based on the real-time contour data, the current flow front state of the colloid is determined, and the flow front state includes at least the spreading state, the accumulation state, and the fracture state.

[0025] In this embodiment, the spreading state is when the adhesive flows out and spreads excessively to both sides, resulting in a flattened and widened cross-section of the adhesive line, which easily leads to adhesive overflow. The accumulating state is when the adhesive accumulates and cannot be spread out in time, causing the cross-section of the adhesive line to bulge into a dome shape, resulting in an excessively thick and convex adhesive line. The fractured state is when the adhesive supply is insufficiently continuous, resulting in gaps, breaks, or retraction at the ends of the adhesive line, making it impossible to form a complete and continuous adhesive line.

[0026] In one embodiment of this application, the curvature features of the cross-section edge of the adhesive line and the smoothness of the edges on both sides of the adhesive line are extracted based on real-time contour data. The spread state and the accumulation state are distinguished by combining the contour shape and the corresponding preset conditions. The fracture state is determined based on the interruption of the adhesive line edge and the retraction of the tail. In this way, the current flow front state of the colloid is identified. The specific implementation process of this step will be described in detail later.

[0027] Based on the flow front state, the correction weight of the subsequent dispensing path segment is calculated, and the dispensing path segment is gradually corrected based on the correction weight to generate the first corrected dispensing path.

[0028] Specifically, the first step is to determine the corresponding state influence level based on the flow front state. Then, the basic correction magnitude is obtained by combining the path distance between the current dispensing position and the path segment to be dispensed. The basic correction magnitude is then gradually attenuated according to the path distance to calculate the correction weight corresponding to each path segment to be dispensed. The specific implementation process of this step will be described in detail later.

[0029] Next, the basic operating parameters such as dispensing speed, Z-axis height of the glue head, and start / stop dwell time of each sub-path segment within the dispensing path segment are obtained. The correction weights corresponding to the sub-path segments are coupled with the basic operating parameters to obtain the corrected operating parameters. The corrected operating parameters are then bound to the corresponding sub-path segments. Each sub-path segment is then sequentially concatenated to form a set of corrected path segments, thereby generating the first corrected dispensing path. The specific implementation process of this step will also be described in detail later.

[0030] The dispensing mechanism is controlled to perform the operation according to the first corrected dispensing path, and the visual acquisition and path correction process is iterated in a loop.

[0031] Specifically, the dispensing mechanism performs dispensing operations according to the first corrected dispensing path and the corrected operating parameters, continuously collects images of the formed glue line and processes them to obtain updated real-time contour data, re-determines the state of the glue flow front and performs path weight correction again, and repeats this process until the workpiece to be dispensed is processed. The specific implementation process of this step will also be described in detail later.

[0032] Furthermore, the method provided in this application embodiment includes: Extract the edge curvature features of the cross-section of the adhesive line and the edge smoothness on both sides of the adhesive line from the real-time contour data; when the cross-section of the adhesive line is flat and the edge curvature features meet the first preset condition and the edge smoothness meets the second preset condition, it is determined to be in a spreading state; when the cross-section of the adhesive line is dome-shaped and the edge curvature features meet the third preset condition, it is determined to be in a stacked state; when a continuous interruption or tail retraction is detected at the edge of the adhesive line based on the real-time contour data, it is determined to be in a broken state.

[0033] Optionally, firstly, the cross-sectional edge curvature features and the smoothness of the two sides of the adhesive line are extracted from the acquired real-time contour data of the adhesive line. First, the extension direction of the adhesive line is determined based on the overall coordinate distribution of the adhesive line contour. Multiple cross-sectional positions are selected at equal intervals along the direction perpendicular to the extension direction of the adhesive line. At each cross-sectional position, the corresponding coordinate points of the left and right sides of the adhesive line are extracted, forming a single set of cross-sectional contour points. For each set of cross-sectional contour points, the curvature value of each point is calculated using a numerical differentiation method. The interval extending one-third of the total width to both sides from the highest point of the cross-sectional contour is selected as the top region. The average curvature value within this region is calculated and used as the edge curvature feature value of the cross-section. The tangent direction of continuous edge points is extracted sequentially along the extension direction of the adhesive line. The change in tangent angle between adjacent edge points is calculated, and the average change in tangent angle per unit length is calculated and used as the smoothness value of the corresponding side edge.

[0034] Next, three preset conditions are defined for the discrimination. The first preset condition is that the average curvature of the top of the cross-section is less than the first curvature threshold, and the ratio of the width to the height of the cross-section is greater than the first aspect ratio threshold, corresponding to the curvature feature requirements of the flat profile. The first curvature threshold and the first aspect ratio threshold are obtained as follows: multiple standard glue line samples in normal spreading and forming state are collected in advance. For each standard sample, the average curvature value at the top and the aspect ratio value of the cross-section are calculated according to the aforementioned cross-sectional curvature feature extraction method. The statistical results of all samples are summarized, and the standard deviation of the top average curvature dataset and the standard deviation of the cross-sectional aspect ratio dataset are calculated respectively. The first curvature threshold is set to the average value of the top average curvature dataset plus one standard deviation, and the first aspect ratio threshold is set to the average value of the cross-sectional aspect ratio dataset minus one standard deviation.

[0035] The second preset condition is that the smoothness values ​​of both edges of the adhesive line are less than a preset smoothness threshold, corresponding to the requirement of a smooth edge without obvious undulations. The preset smoothness threshold is obtained by: collecting multiple standard adhesive line samples in a normal spreading and forming state, calculating the smoothness values ​​of both edges of each sample according to the aforementioned edge smoothness extraction method, summarizing the smoothness statistics of all samples, and taking the average of the smoothness dataset plus twice the standard deviation as the preset smoothness threshold.

[0036] The third preset condition is that the average curvature of the top of the cross-section is greater than the second curvature threshold, and the ratio of the width to the height of the cross-section is less than the second width-to-height ratio threshold, corresponding to the curvature feature requirements of the dome-shaped protrusion profile. The second curvature threshold and the second width-to-height ratio threshold are obtained as follows: multiple glue line samples with obvious accumulation and molding states caused by excessive glue supply pressure or slow glue head movement speed are collected in advance. For each accumulation sample, the average curvature value of the top and the width-to-height ratio value of the cross-section are calculated according to the aforementioned cross-sectional curvature feature extraction method. The statistical results of all samples are summarized, and the standard deviation of the top average curvature dataset and the standard deviation of the cross-sectional width-to-height ratio dataset are calculated respectively. The second curvature threshold is set as the average value of the top average curvature dataset minus one standard deviation, and the second width-to-height ratio threshold is set as the average value of the cross-sectional width-to-height ratio dataset plus one standard deviation.

[0037] After feature extraction and threshold comparison, the flow front state is determined sequentially. When the cross-section of the adhesive line is generally flat, and the edge curvature of the corresponding cross-section meets the first preset condition, and the smoothness of the edges on both sides of the adhesive line meets the second preset condition, the adhesive is determined to be in a spreading state. When the cross-section of the adhesive line is generally dome-shaped, and the edge curvature of the corresponding cross-section meets the third preset condition, the adhesive is determined to be in a accumulating state. All edge points are traversed along the extension direction of the adhesive line. If the spatial distance between adjacent edge points exceeds the preset maximum allowable spacing threshold, and the length of this out-of-tolerance state extending continuously along the extension direction of the adhesive line exceeds the preset interruption length threshold, it is determined that there is a continuous edge interruption region in the adhesive line. The appearance of this interruption region indicates that the adhesive is in a fractured state. Alternatively, if the extension direction of the end edge point of the adhesive line is reversed and the distance of the inward contraction of the end point relative to the previous normal edge point after the reverse deflection exceeds the preset retraction threshold, it is determined that the tail of the adhesive has retracted, and the adhesive is also determined to be in a fractured state.

[0038] By quantifying and extracting the curvature and smoothness features of the glue line profile, and combining them with clear threshold conditions, the automatic identification of three types of flow front states is achieved, providing a reliable basis for the adaptive correction of the subsequent dispensing path.

[0039] Furthermore, the method provided in this application embodiment includes: Based on the category of the flow front state, determine the corresponding state influence level; obtain the path distance between the current dispensing position and the path segment to be dispensed; determine the basic correction magnitude based on the state influence level and the path distance; and perform segment-by-segment attenuation processing on the basic correction magnitude in order of path distance from near to far to obtain the correction weight corresponding to each path segment to be dispensed.

[0040] Specifically, after determining the state of the colloid flow front, the corresponding state influence level is matched according to the state category. A mapping table between state categories and influence levels is pre-stored in the main control processing unit. This mapping table is obtained by performing multiple dispensing process experiments with different defect severity levels for the three flow front states: spreading, accumulating, and fractured. The impact range and degree of the current defect on the quality of the glue line forming on subsequent continuous path segments are measured in each experiment. The influence degree is divided into multiple level intervals from mild to severe. The spreading state indicates an excessively large colloid spreading area but undamaged flow continuity, with the lowest impact on subsequent forming quality, corresponding to the lowest level value. The fractured state indicates a physical interruption of the glue line and the flow front has stopped advancing, with the highest degree of damage to subsequent forming quality, corresponding to the highest level value. The accumulating state's influence degree is between the two, mapped to an intermediate level value. The above correspondence between state categories and level values ​​is stored as key-value pairs in the parameter storage area of ​​the main control processing unit. Different flow states correspond to different level values. The level value represents the degree of influence of the state on the subsequent glue line forming. After determining the current state, the corresponding state influence level can be obtained by looking up a table.

[0041] Next, calculate the path distance between the current dispensing position and each dispensing path segment. First, locate the corresponding node of the current dispensing position of the glue head on the initial dispensing path. Along the trajectory extension direction of the initial dispensing path, sequentially accumulate the path arc length between the current node and the starting node of each subsequent dispensing path segment to obtain the cumulative path distance value from the current dispensing position to each dispensing path segment.

[0042] The basic correction magnitude is determined by combining the state impact level and path distance. A maximum correction magnitude parameter is pre-set, and a corresponding level coefficient is pre-configured for each state impact level in the main control processing unit. The level coefficient is a real number between 0 and 1; the higher the state impact level, the closer the level coefficient is to 1, and the lower the state impact level, the closer the level coefficient is to 0. The level coefficient corresponding to the currently determined flow front state is multiplied by the maximum correction magnitude parameter. The product is used as the basic correction magnitude closest to the current dispensing position. This basic correction magnitude represents the maximum adjustment intensity required for the next path segment adjacent to the anomaly occurrence point under the current defect state. The higher the state impact level, the larger the corresponding basic correction magnitude value, and the stronger the effect of parameter adjustment.

[0043] Finally, the base correction amplitude is progressively attenuated according to the path distance from shortest to longest. The attenuation is achieved using a linear attenuation method: the path point corresponding to the current dispensing position is taken as the starting point of the attenuation, and the base correction amplitude at that position is... Pre-set the effective influence distance for the currently determined flow front state. For any subsequent adhesive path segment to be dispensed at a path distance d from the attenuation starting point, its correction weight W(d) is calculated according to the following linear attenuation formula: when hour, Effective influence distance The values ​​are set according to the category of the flow front state. The effective influence distance corresponding to the fractured state is greater than that corresponding to the accumulated state, and the effective influence distance corresponding to the accumulated state is greater than that corresponding to the spreading state. Path segments exceeding the effective influence distance range are too far from the point of anomaly, and the colloidal flow characteristics have naturally recovered to the normal state before reaching the path segment, so the correction weight is set to zero and no parameter adjustment is required. Finally, the correction weight corresponding to each dispensing path segment is obtained. The specific distribution law of weight decay and the multi-scenario adaptation calculation method will be described in detail in the corresponding embodiments later.

[0044] The basic correction range is determined by matching the state level and calculating the path arc length, and the correction weight of each segment is obtained by gradually decreasing along the path direction. This provides a quantitative basis for the differentiated adjustment of subsequent dispensing parameters, taking into account both the defect correction effect and the stability of the path operation.

[0045] Furthermore, the method provided in this application embodiment includes: The mapping range of the correction weight covers N consecutive path segments after the current dispensing position, and the correction weight is distributed exponentially with decreasing along the path extension direction. The closer the path segment is to the point of anomaly, the higher the correction weight.

[0046] Specifically, the location of the point of occurrence corresponding to the flow front anomaly on the initial dispensing path is first determined. Starting from the first dispensing path segment after the anomaly point, consecutive path segments are selected along the path extension direction as the coverage area for weight correction. The standard length of a single path segment is determined jointly by the interpolation cycle of the dispensing motion controller and the rated dispensing travel speed. The standard length equals the rated travel speed multiplied by the duration of a single interpolation cycle, ensuring that each path segment corresponds to a complete control cycle of the controller and ensuring that parameter adjustments can be executed accurately. The total number of covered path segments N is determined according to the category of flow front state. Multiple sets of process calibration tests are conducted in advance to test the shortest path length required for the colloidal flow characteristics to recover from an abnormal state to a normal state under different conditions. The recovery length is divided by the standard length of a single path segment and rounded to obtain the N value for the corresponding state. The fractured state has the longest influence distance on subsequent glue lines, corresponding to the largest N value; the spreading state has the shortest influence distance, corresponding to the smallest N value; the accumulated state has an N value in between. Once the value of N is determined, the weight correction for the current abnormal state only applies to the next N consecutive path segments; path segments outside this range are not subject to any correction or adjustment.

[0047] After determining the coverage area, N path segments are numbered sequentially from nearest to farthest along the path extension direction. The first path segment immediately adjacent to the anomaly occurrence point is numbered 1, and subsequent path segments are numbered sequentially. The previously calculated baseline correction magnitude is used as the initial benchmark weight, corresponding to the correction weight of path segment numbered 1. For any path segment numbered i, its correction weight is calculated using an exponential decay rule, calculated as: Correction weight = Initial benchmark weight × e^(-decay coefficient × (i−1)). The decay coefficient is adapted to the value of N, ensuring that the correction weight corresponding to path segment numbered N decays to less than one-tenth of the initial benchmark weight. Specifically, when the decay coefficient needs to be quantified based on the total number of covered path segments N, it is determined according to the inequality... To retrieve values, that is Take the smallest value that satisfies the inequality The value is used as the actual attenuation coefficient to ensure that the correction weight on the end path segment of the coverage area has attenuated to less than one-tenth of the initial baseline weight. This distribution method ensures that the path segment closer to the anomaly point receives a stronger parameter correction force, and the correction force decreases rapidly and gradually with increasing distance, avoiding fluctuations in glue line formation caused by abrupt parameter changes.

[0048] Taking a set of typical process parameters as an example, assuming the current classification is a stacked state, the corresponding basic correction magnitude, i.e., the initial reference weight, is 0.6, and the total number of covered path segments N is 5. Following the attenuation coefficient setting rules mentioned above, we take... The final segment can be accurately attenuated to 0.1 times, but for ease of rounding in engineering implementation, this example uses α=0.4, at which point the final segment... Since the value is less than 0.1 times the baseline value, the adjusted weights for each path segment are calculated sequentially: the weight for the first segment is 0.6, the weight for the second segment is approximately 0.40, the weight for the third segment is approximately 0.27, the weight for the fourth segment is approximately 0.18, and the weight for the fifth segment is approximately 0.12. The adjusted weights for path segments after the fifth segment are set to zero and are not included in the parameter adjustment.

[0049] By limiting the correction coverage of continuous path segments and using an exponential decay rule to allocate the correction weight of each segment, the parameter transition is achieved while ensuring sufficient correction strength near abnormal points, effectively balancing the correction effect of glue line defects with the stability of dispensing operation.

[0050] Furthermore, the method provided in this application embodiment includes: Obtain the basic operating parameters of each sub-path segment in the dispensing path segment. The basic operating parameters include at least the dispensing speed, the Z-axis height of the dispensing head, and the start-stop dwell time. Couple the correction weights corresponding to the sub-path segments with the dispensing speed, the Z-axis height of the dispensing head, and the start-stop dwell time in the basic operating parameters to obtain the corrected operating parameters. Bind the corrected operating parameters to the corresponding sub-path segments to form a set of corrected path segments. Then, concatenate the set of corrected path segments in sequence to generate the first corrected dispensing path.

[0051] Specifically, firstly, based on the sequential numbering of each sub-path segment in the initial dispensing path, the corresponding basic operating parameters are retrieved one by one. The sub-path segments here are of the same granularity as the path segments divided in the previous steps. Each continuous path segment participating in the weight calculation corresponds to one sub-path segment, with completely consistent segment boundaries, ensuring precise matching between the corrected weights and operating parameters. The basic operating parameters are pre-determined through dispensing process calibration experiments and stored in the parameter storage module of the motion control unit, corresponding one-to-one with different path conditions. The calibration process for the basic dispensing speed involves setting multiple sets of different travel speeds for trial application under standard glue head Z-axis height and rated glue flow rate conditions. The actual width of each glue line is visually measured, and the speed value with the smallest deviation from the target glue line width is determined as the basic dispensing speed for the corresponding operating condition. The path is divided into multiple speed levels based on curvature and turning angle. This speed level division is consistent with the high-speed, normal-speed, and low-speed segment division based on curvature extreme points and inflection points in the initial path planning stage of subsequent embodiments. That is, high-speed segments correspond to high-speed levels, low-speed segments correspond to low-speed levels, and normal-speed segments correspond to medium-speed levels. The speed level of each sub-path segment has been marked in the initial path planning stage. During retrieval, the corresponding standard speed value is directly matched and read from the pre-stored parameter table according to the speed level corresponding to the sub-path segment.

[0052] The basic Z-axis height of the dispensing nozzle refers to the vertical distance between the dispensing nozzle and the workpiece surface. This value is calibrated based on the dispensing flow rate, dispensing speed, and target glue line cross-sectional height. The calibration process involves fixing the basic dispensing speed and rated dispensing flow rate for the corresponding speed setting, gradually adjusting the vertical distance between the dispensing nozzle and the workpiece surface, measuring the glue line cross-sectional height and edge smoothness at each distance, and selecting the distance value where the glue line height meets design requirements and has no obvious trailing at the edge as the basic Z-axis height of the dispensing nozzle for the corresponding speed setting. This value is stored in a one-to-one correspondence with the speed settings. When retrieving the value, it is directly read from the parameter table based on the speed setting of the sub-path segment, ensuring that the glue line forming height meets design requirements under basic conditions.

[0053] The basic start-stop dwell time refers to the duration for which the glue head remains stationary at the start and end positions of the path segment, used to compensate for the glue dispensing delay during the start and stop phases. The value is calibrated based on the glue dispensing response speed, needle inner diameter, and joint flatness requirements. The calibration process involves setting multiple different dwell times at path corners at different angles for test coating, observing the continuity of the glue line and the glue accumulation at the corner joint, and selecting the duration without depressions or excessive accumulation at the joint as the basic start-stop dwell time for the corresponding corner condition. These are divided into multiple standard values ​​stored according to the corner size. When retrieving values, the corresponding parameters are matched according to the corner size of the sub-path segment's start and end points; the larger the corner, the longer the basic dwell time, ensuring sufficient glue and no depressions at the path joint. The specific quantification method for the size of the turning angle is to take the angle between the entry direction and the exit direction at the starting point of the sub-path segment as the turning angle of the turning position. The turning angle is divided into three levels: acute angle, right angle and obtuse angle. Each level corresponds to a set of basic start and stop dwell time standard values. The larger the turning angle, the longer the corresponding standard dwell time.

[0054] After obtaining the corrected weights for each sub-path segment in the aforementioned steps, the corrected weights are coupled with the basic dispensing speed, basic glue head Z-axis height, and basic start-stop dwell time of that sub-path segment to obtain the corrected actual operating values ​​for each parameter. The specific calculation rules and parameter adjustment directions for the coupling operation will be detailed in subsequent embodiments. The calculation process is executed sequentially for each sub-path segment, ensuring that each path segment corresponds to an independent set of corrected operating parameters, with a one-to-one correspondence between parameters and path segments, and no misalignment or omissions.

[0055] After correcting the operating parameters of all sub-path segments, each set of corrected operating parameters is bound to the corresponding numbered sub-path segment, forming a set of corrected path segments. The binding operation uses an index matching method, with the sequential number of the sub-path segment as a unique identifier. The corrected dispensing speed, nozzle Z-axis height, and start / stop dwell time are written into the parameter attributes of the path segment to ensure that the motion controller can synchronously obtain the corresponding operating parameters when reading the path. Subsequently, according to the order of the sub-path segments on the initial dispensing path, all corrected path segments are sequentially spliced ​​together, keeping the trajectory coordinates of the original path unchanged, and only updating the operating parameters of each segment, finally generating a complete first corrected dispensing path.

[0056] By clarifying the correspondence between segmented granularities, refining the parameter calibration and retrieval process, and combining segmented parameter binding and path splicing, the operating parameters are gradually adjusted while retaining the original path trajectory, ensuring that the corrected path can be directly executed by the motion controller and that the glue line forming transition is smooth.

[0057] Furthermore, the method provided in this application embodiment includes: If the flow front is in a stacked state, the Z-axis height of the dispensing head is increased and the dispensing speed is decreased, using the correction weight as a scaling factor; if the flow front is in a broken state, the start-stop dwell time is increased, using the correction weight as a scaling factor; if the flow front is in a spreading state, the Z-axis height of the dispensing head is decreased and the dispensing speed is increased, using the correction weight as a scaling factor.

[0058] In one embodiment, before performing the coupling operation, the flow front status identifier corresponding to the current sub-path segment is first read, and the calculated correction weight value for that sub-path segment is retrieved. Simultaneously, the original values ​​of the basic dispensing speed, basic nozzle Z-axis height, and basic start-stop dwell time corresponding to that sub-path segment are read from the pre-stored parameters. A maximum adjustment ratio for each type of flow state is pre-set. This ratio is determined through process calibration experiments to test the improvement effect of glue line formation and equipment operational stability under different adjustment ranges. The adjustment range with the optimal correction effect and without causing secondary defects is selected as the maximum adjustment ratio for the corresponding parameter and stored in the parameter table of the control unit. For example, in a conventional dispensing process scenario, the maximum upward and downward adjustment ratios of the nozzle Z-axis height are typically set in the range of 5% to 20%, the maximum upward and downward adjustment ratios of the dispensing speed are typically set in the range of 10% to 30%, and the maximum upward adjustment ratio of the start-stop dwell time is typically set in the range of 20% to 50%. The actual values ​​can be calibrated and adjusted within this range according to the glue viscosity, nozzle inner diameter, and target glue line accuracy requirements. Table 1 shows the adjustment parameters, adjustment directions, and typical values ​​for conventional processes corresponding to the three types of flow front states. The typical calibration values ​​in this table are applicable to conventional desktop dispensing scenarios with epoxy adhesives of viscosity 3000~5000cps, needle inner diameter 0.4mm, and target adhesive line width 0.8mm. In actual production, these values ​​can be flexibly adjusted through calibration tests according to the adhesive properties and product precision requirements. The calculation process adopts a linear proportional adjustment method, using the correction weight as a proportional coefficient to control the proportion of the actual adjustment amplitude to the maximum adjustment amplitude. The closer to the anomaly point, the higher the correction weight, and the larger the parameter adjustment amplitude; conversely, the further away from the anomaly point, the smaller the adjustment amplitude, thus achieving gradual correction.

[0059] Table 1: Rules for Adjusting Operating Parameters under Different Flow Front States ; When the flow front is identified as being in an accumulated state, adjustments are made to the Z-axis height of the dispensing head and the dispensing speed. The Z-axis height is calculated upwards; the corrected Z-axis height equals the base Z-axis height plus the product of the base height, the correction weight, and the maximum height increase ratio. The dispensing speed is calculated downwards; the corrected dispensing speed equals the base dispensing speed minus the product of the base dispensing speed, the correction weight, and the maximum speed decrease ratio. In the accumulated state, the glue line cross-section bulges and the glue volume is excessive. Increasing the Z-axis height of the dispensing head elongates the cross-sectional shape of the glue, reducing the glue line accumulation height. Decreasing the dispensing speed matches the flow rhythm of the glue, preventing continuous accumulation and bulging of the glue. These two parameters, adjusted together, improve the accumulation defects.

[0060] When the flow front is identified as being in a spreading state, inverse adjustment calculations are performed on the Z-axis height of the dispensing head and the dispensing speed. The Z-axis height is calculated downwards; the corrected Z-axis height equals the base Z-axis height minus the product of the base height, the correction weight, and the maximum height reduction ratio. The dispensing speed is calculated upwards; the corrected dispensing speed equals the base dispensing speed plus the product of the base dispensing speed, the correction weight, and the maximum speed increase ratio. In the spreading state, the adhesive line is excessively spread and exceeds the width limit. Reducing the Z-axis height of the dispensing head restricts the adhesive's spreading space to both sides, constraining the adhesive line width. Increasing the dispensing speed reduces the total amount of adhesive dispensed per unit path length, preventing continuous diffusion of adhesive to the edges. These two parameters work together to improve spreading defects.

[0061] When a flow front is identified as being in a broken state, only the start-stop dwell time is adjusted upwards. The corrected start-stop dwell time equals the base start-stop dwell time plus the product of the base start-stop dwell time, the correction weight, and the maximum dwell time adjustment ratio. In a broken state, the adhesive line appears interrupted or retracts at the tail, essentially due to insufficient local adhesive delivery. Extending the start-stop dwell time increases the dispensing time at the start and end points of the path, replenishing local adhesive delivery, filling gaps in the adhesive line, and restoring its continuity. For broken positions in the middle of the path, a predetermined length of path area is extended forward and backward from the broken point as a deceleration adjustment section. This length is set according to the adhesive flow recovery distance, typically ranging from 5 to 20 mm. Smaller values ​​are used for low-viscosity adhesives or high-speed dispensing scenarios, while larger values ​​are used for high-viscosity adhesives or low-speed dispensing scenarios. Specific values ​​are determined through flow recovery calibration tests. After extension, several continuous sub-path segments are divided. Simultaneously, the start and stop times of these sub-path segments are adjusted upward according to the same rules to ensure sufficient overall adhesive in the fracture area and fill the continuous gaps.

[0062] After each set of sub-path segments has undergone parameter calculation, the corrected parameter values ​​are compared and verified against preset safety upper and lower limits. If the corrected dispensing speed is lower than the equipment's minimum stable dispensing speed, the minimum stable speed value is used; if it is higher than the equipment's maximum rated operating speed, the maximum rated speed value is used. If the corrected Z-axis height of the dispensing head is lower than the safe distance from the workpiece surface, the safe distance value is used; if it is higher than the upper limit of the effective dispensing height, the upper limit value is used. If the corrected start-stop dwell time exceeds the maximum allowable dwell time, the maximum allowable dwell time is used to prevent excessive glue accumulation and overflow. All sub-path segments are calculated and verified sequentially according to the path extension order to ensure that the parameters gradually return to the base value as the correction weight decreases, achieving a smooth transition.

[0063] By matching differentiated parameter adjustment directions for three types of flow front states, and using the correction weight as a proportional factor combined with the maximum adjustment ratio to achieve linear gradual operation, along with parameter boundary verification and segmented adjustment of fracture areas, the defects in glue line forming are accurately improved while ensuring smooth parameter transition and avoiding step-like abrupt changes in the appearance of the glue line.

[0064] Furthermore, the method provided in this application embodiment includes: The dispensing mechanism is controlled to move along the first corrected dispensing path and perform dispensing operations according to the corrected operating parameters. After completing the current dispensing path segment, the vision sensing module is triggered again to acquire the next frame of the formed glue line image. Subpixel-level edge extraction is performed on the next frame of the formed glue line image to obtain updated real-time contour data. Based on the updated real-time contour data, the flow front state of the glue is re-determined, and based on the updated flow front state, the weight correction of the subsequent dispensing path segment is performed until the workpiece to be dispensed is completed.

[0065] Optionally, the transformation matrix between the imaging coordinate system of the vision sensing module and the motion coordinate system of the dispensing mechanism can be obtained in advance through hand-eye calibration to determine the spatial position offset between the two. The specific implementation of hand-eye calibration is as follows: a checkerboard calibration plate of known size is fixedly placed on the workpiece table. The dispensing mechanism is controlled to move the vision sensing module to multiple different poses to capture images of the calibration plate. The pixel coordinates of the checkerboard corner points in each image are extracted. At the same time, the physical coordinates of the axis end of the motion controller under the corresponding pose are recorded. The correspondence between multiple sets of pixel coordinates and physical coordinates is established. The rotation matrix and translation vector are solved by the least squares method to obtain the transformation matrix between the imaging coordinate system and the motion coordinate system, as well as the spatial position offset between the two.

[0066] The main control unit breaks down the generated first corrected dispensing path into continuous motion interpolation commands, which are then sent segment by segment to the dispensing motion controller. The motion controller drives the dispensing head to move along the planned trajectory coordinates according to the commands. Simultaneously, based on the corrected operating parameters bound to each sub-path segment, it synchronously controls the dispensing valve's dispensing timing, the dispensing head's Z-axis lifting position, and the dwell actions at the start and end points of the path, ensuring precise matching between dispensing volume and movement speed. The motion controller provides real-time feedback on the dispensing head's current physical position via an axis-end encoder. Using a pre-stored transformation matrix, it can calculate the corresponding coordinate position for visual acquisition in real time, ensuring precise correspondence between the visual acquisition area and the just-completed dispensing area. When the dispensing head reaches the end coordinate of the current sub-path segment, the dispensing operation for the current segment is considered complete.

[0067] After the current path segment is completed, the motion controller fine-tunes the spatial position of the glue head based on the calibrated position offset, ensuring that the newly completed glue line segment is completely within the center of the vision of the vision sensing module. It then immediately outputs a hardware trigger signal to the vision sensing module, initiating image acquisition. The vision sensing module, according to pre-calibrated exposure parameters, focal length, and frame rate, captures the formed glue line area directly below, covering the entire completed glue line and the surrounding workpiece reference area, ensuring complete capture of the forming morphology of the glue flow front. The acquired glue line image is transmitted in real-time to the image processing unit via the data bus, awaiting subsequent contour extraction processing.

[0068] The image processing unit performs sub-pixel-level edge extraction processing on the new frame of the formed adhesive line image, and the processing flow is consistent with the initial adhesive line contour acquisition stage. First, the color image is converted into a single-channel grayscale image, then Gaussian filtering is used to filter out surface reflections and noise interference in the image. Subsequently, an edge detection algorithm is used to extract the pixel-level edge contour of the adhesive line. Finally, quadratic polynomial interpolation is used to refine the coordinates of the pixel-level edge points with sub-pixel precision, obtaining updated edge coordinate data on both sides of the adhesive line. Based on this, real-time contour data such as the real-time width, cross-sectional bulge degree, and continuity of the adhesive line are calculated.

[0069] Finally, the updated real-time contour data is compared with the preset standard glue line contour parameters. These standard glue line contour parameters are a set of reference parameters stored in the main control unit after digital conversion, based on the target glue line width, target cross-sectional height, and edge flatness requirements marked in the process design drawings of the workpiece to be glued. The current flow front state of the glue is re-evaluated according to the previously defined state determination rules. After state determination, using the current glue head position as the correction starting point, the correction weight calculation, running parameter coupling operation, and path parameter update process are re-executed for all remaining unexecuted glue path segments, generating a new corrected path to replace the original subsequent path. The dispensing mechanism continues to perform operations along the updated path, repeating the acquisition, determination, and correction process after each path segment is completed, until the motion controller detects that the glue head has reached the end position of the entire dispensing path, determining that the workpiece to be glued has been processed, and then the cycle terminates.

[0070] By using hand-eye coordinate transformation to achieve precise synchronization of acquisition position, combined with segment-triggered closed-loop visual inspection and iterative path dynamic correction, real-time control and continuous deviation correction of the entire dispensing process are achieved, effectively ensuring the dimensional consistency and molding quality stability of the entire dispensing line.

[0071] Furthermore, the method provided in this application embodiment includes: Along the extension direction of the real-time contour data, several contour micro-segments of equal length are divided; the frequency of abrupt changes in the contour gradient direction within each contour micro-segment is counted to obtain the local burr index; the local burr indices of all contour micro-segments are summarized to obtain the overall edge burr index; the offsets of the left and right edges of the adhesive line in the real-time contour data at corresponding positions are calculated, and the dispersion of the offsets is obtained to obtain the edge symmetry; the overall edge burr index and the edge symmetry are fused to calculate the edge roughness, and the edge roughness is used as an auxiliary discrimination parameter for flow front state discrimination.

[0072] In one embodiment, based on the aforementioned flow front state discrimination based on cross-sectional morphology and edge smoothness, to further improve the reliability and anti-interference capability of state discrimination, this embodiment introduces edge roughness as an auxiliary discrimination parameter. Edge roughness does not independently perform state determination; instead, it intervenes as secondary verification evidence when the confidence level of the state discrimination output by the main discrimination parameter is lower than a preset acceptable threshold. Combined with adhesive line width, cross-sectional bulge degree, and continuity characteristics, it assists in confirming the three types of flow front states, thereby reducing the risk of misjudgment caused by fluctuations in a single parameter.

[0073] First, the left and right edge contour data obtained from sub-pixel extraction are sorted sequentially along the adhesive line extension direction to ensure that the contour points are arranged in order according to the path forward direction. For a single edge contour, the Euclidean distance between adjacent contour points is calculated sequentially from the starting point and accumulated to obtain the cumulative arc length corresponding to each point. The unit length of the contour micro-segment is preset. This length is set according to the adhesive line size and detection accuracy requirements. It can typically be set to the pixel length corresponding to the physical length of 0.3 to 1 mm. The entire edge is evenly divided into several contour micro-segments of equal length according to the cumulative arc length. Each micro-segment contains multiple consecutive contour points, and adjacent micro-segments are connected end-to-end without overlap. The same segmentation rules and segmentation positions are used for the left and right edges to ensure that the micro-segments of the two edges correspond one-to-one.

[0074] Next, for all edge points within each contour micro-segment, the contour tangent direction of each point is calculated sequentially. The calculation method involves fitting a straight line to the two adjacent points before and after the current point, and using the inclination angle of the fitted line as the gradient direction of that point. A pre-set direction abrupt change threshold is used, typically between 12 and 20 degrees in typical examples. The gradient direction difference between two adjacent contour points within the micro-segment is compared sequentially. When the difference exceeds the pre-set threshold, it is considered an edge burr abrupt change. The total number of abrupt changes within the current micro-segment is counted, and the number of abrupt changes is divided by the arc length of the micro-segment to obtain the local burr index corresponding to that micro-segment. A higher value indicates a denser density of burrs and uneven defects at the edge of that segment.

[0075] Then, the local burr indices of all contour segments along the same edge are arithmetically averaged to obtain the overall edge burr index for a single side. The overall edge burr indices of the left and right edges of the adhesive line are then averaged to obtain the overall edge burr index for the entire adhesive line. This index quantifies the density and severity of burrs along the entire edge of the adhesive line; a higher value indicates a rougher overall edge and more irregular defects. After the index calculation, a min-max linear normalization method is used for numerical processing. The lower and upper limits of the overall edge burr index are pre-calibrated using multiple sets of process samples. The calculated index values ​​are then proportionally mapped to the range of 0 to 1 to eliminate dimensional differences and provide a unified numerical benchmark for subsequent feature fusion.

[0076] First, based on the coordinate data of the left and right edges, the center trajectory line of the adhesive line is obtained by fitting using the midpoint method. Several sampling points are selected along the center trajectory line with a fixed step size, consistent with the length of the contour micro-segment. At each sampling point, a normal line to the center trajectory is drawn, and the intersection of this normal line with the left and right edges is calculated. The perpendicular distances from the left and right edges to the center trajectory at these positions are obtained. The difference between the distances of the right and left edges at the same normal position is calculated as the edge offset at that position. After calculating the offsets for all sampling points, the standard deviation of all offsets is calculated. This standard deviation is used as an index of edge symmetry; a larger standard deviation indicates poorer symmetry and consistency between the left and right edges of the adhesive line, and lower edge regularity. The edge symmetry is also normalized using the min-max linear normalization method, with pre-defined upper and lower limits, mapping the calculation results to a numerical range of 0 to 1.

[0077] The weighting coefficients for the overall edge burr index and edge symmetry are pre-determined through process calibration, with the sum of the two weights being 1. Examples show values ​​of 0.6 and 0.4 respectively. The normalized overall edge burr index and edge symmetry are multiplied by their corresponding weights and summed to obtain the final edge roughness value. A higher value indicates a higher overall roughness of the adhesive line edge. Edge roughness is used as an auxiliary discrimination parameter, participating in the flow front state discrimination along with main discrimination parameters such as adhesive line width deviation, cross-sectional bulge degree, and adhesive line continuity index. When the state discrimination confidence obtained from the main parameters is lower than a preset acceptable threshold, edge roughness is introduced for secondary verification: if the edge roughness is high and the overall width of the adhesive line exceeds the standard range, the discrimination confidence for the spread state is increased; if the edge roughness is high and the cross-sectional bulge degree of the adhesive line exceeds the standard, the discrimination confidence for the packed state is increased; if the edge roughness is high and there are continuity breaks in the adhesive line, the discrimination confidence for the broken state is increased. By using auxiliary parameters to further distinguish critical states, the problem of misjudgment caused by fluctuations in a single parameter can be reduced.

[0078] By combining micro-segment gradient mutation statistics with edge symmetry analysis, the roughness of the adhesive line edge is quantitatively calculated. With clear normalization rules and auxiliary discrimination logic, the edge morphology features are transformed into quantifiable auxiliary discrimination indicators, which enriches the discrimination dimensions of the flow front state and effectively improves the accuracy of state identification and anti-interference ability.

[0079] Furthermore, the method provided in this application embodiment includes: Extract the curvature extrema and inflection points from the initial contour. Based on the distribution of the curvature extrema and inflection points, divide the initial dispensing path into a high-speed segment, a normal-speed segment, and a low-speed segment. Configure corresponding basic operating parameters for the high-speed segment, the normal-speed segment, and the low-speed segment respectively. During the iterative process, add the correction weights to the basic operating parameters of the high-speed segment, the normal-speed segment, and the low-speed segment respectively to obtain the final operating parameters of each segment.

[0080] In one embodiment, firstly, the initial contour of the workpiece dispensing path obtained from the initial acquisition is uniformly sampled using discrete points to obtain a sequence of contour coordinate points arranged in the path order. To reduce the interference of contour sampling noise on curvature calculation, a fixed-window moving average filter is used to smooth the contour point coordinates; the window length can be set to 5 to 9 points. Then, the curvature value of each contour point is calculated using the three-point circular arc method: the current point and one adjacent point before and after it are taken sequentially, and the radius of the fitted arc is calculated using the three-point coordinates. The curvature value is the reciprocal of this radius. The sign of the curvature corresponds to the concavity / convexity direction of the contour; positive for convexity towards the outside of the path and negative for concaveness towards the inside of the path. After traversing all contour points, points where the absolute value of curvature exhibits a local maximum along the path extension direction are selected as curvature extrema points, including local maximum and local minimum points. Simultaneously, points where the curvature sign switches between positive and negative are selected as inflection points. These two types of points together serve as feature boundary points for path segmentation.

[0081] Then, based on the acceleration / deceleration performance of the equipment's motion axis and the requirements for glue line formation quality, two curvature thresholds are pre-calibrated: a low-speed threshold and a high-speed threshold. The calibration process involves setting test paths with different curvature radii on a standard workpiece, gradually increasing the operating speed, and recording the highest speed at which the equipment runs smoothly without significant vibration and the glue line formation is qualified. Correspondingly, the curvature boundaries of the three operating ranges—high-speed, normal-speed, and low-speed—are defined, and the boundary curvature values ​​are determined as the high-speed threshold and the low-speed threshold. In typical desktop dispensing equipment scenarios, the high-speed threshold can be taken as 0.03 to... The low-speed threshold can be set to 0.15. The smaller the absolute value of curvature, the smoother the path. Traverse all contour points sequentially along the path extension direction, and use the feature boundary points as interval boundaries to count the absolute value of curvature within each interval.

[0082] If the absolute value of the maximum curvature within a given interval is less than the high-speed threshold, it indicates that the path segment is a large-radius curve or a straight line segment with high motion stability, and is classified as a high-speed segment. If the absolute value of the maximum curvature within a given interval is between the high-speed threshold and the low-speed threshold, it indicates that the segment is a smooth transition curve, and is classified as a normal-speed segment. If the absolute value of the maximum curvature within a given interval is greater than the low-speed threshold, it indicates that the segment is a small-radius turn or a sharp bend, with high motion inertia, and is classified as a low-speed segment. After classification, the corresponding speed segment attribute is marked for each sub-path segment to ensure that the speed level is consistent within the same continuous path segment.

[0083] Independent basic operating parameters are configured for high-speed, normal-speed, and low-speed sections. These parameters include three categories: basic dispensing speed, basic glue head Z-axis height, and basic start-stop dwell time. These parameters are determined through process calibration tests: under standard glue dispensing volume and viscosity conditions, test coatings are performed on the corresponding path patterns for each of the three speed sections. The values ​​of each parameter are adjusted to ensure that the glue line width, height, and edge quality meet the acceptable standards. The final parameter values ​​are then solidified as the basic operating parameters for the corresponding speed section. The high-speed section has the highest basic dispensing speed and the shortest basic start-stop dwell time; the low-speed section has the lowest basic dispensing speed and the longest basic start-stop dwell time. The glue head Z-axis height is fine-tuned as the speed decreases to ensure stable glue dispensing and molding. Independent maximum adjustment ratios can be set for each of the three speed sections. Due to the high motion precision requirements of the low-speed section, the maximum adjustment ratio is set relatively small, while it can be appropriately increased for the high-speed section to avoid exceeding the stable operating range of the equipment after correction.

[0084] In the iterative correction process based on visual feedback, the correction weight for each sub-path segment is first calculated based on the currently determined flow front state and path distance. Each sub-path segment retrieves its corresponding level of basic operating parameters according to its own velocity segment attributes. The coupling operation uses a linear proportional adjustment method, adjusting the basic operating parameters according to the parameter adjustment direction corresponding to the current flow front state, using the correction weight as a proportional coefficient, and matching the maximum adjustment ratio of the corresponding velocity segment to perform increase / decrease calculations. During the adjustment process, the basic operating parameters of all sub-path segments are adjusted in the same direction and proportionally with the same correction weight; that is, if the basic value of the high-speed segment is high, it remains high after correction, and if the basic value of the low-speed segment is low, it remains low after correction, ensuring that the velocity hierarchy between the high-speed and low-speed segments is not reversed due to correction. Simultaneously, a smaller maximum adjustment ratio is used for the low-speed segment to further ensure operational stability under low-speed sharp curves. The correction magnitude decreases with path distance, achieving a gradual transition of all path parameters and compatibility and unity of velocity classification.

[0085] By dividing the speed into multiple speed segments based on the contour curvature features and matching them with corresponding basic parameters, and combining a clear threshold calibration method with a unified weight coupling rule to adjust the speed by segment, dynamic parameter correction under the graded speed system is achieved while ensuring the smoothness of path motion and the quality of basic forming. This balances dispensing efficiency with the smoothness of defect correction.

[0086] In summary, the intelligent dispensing path planning method based on machine vision contour features provided in this application has the following technical effects: This application acquires images of formed adhesive lines, obtains real-time contour data through sub-pixel edge extraction, determines the state of the adhesive flow front, calculates exponential decay correction weights for continuous path segments, couples and adjusts dispensing operation parameters, and combines iterative correction with edge roughness-assisted discrimination to achieve intelligent dispensing path planning. This effectively improves the accuracy and quality consistency of adhesive line formation, and achieves the technical effect of continuously sensing the adhesive spreading status and smoothly adjusting the execution parameters of subsequent processes segment by segment, thus realizing closed-loop adaptive control of adhesive line formation quality.

[0087] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides an intelligent dispensing path planning system based on machine vision contour features, the system comprising: Initial dispensing path acquisition module 1 is used to acquire the initial contour of the workpiece to be dispensed, generate an initial dispensing path, and perform dispensing based on the initial dispensing path.

[0088] The adhesive line contour data acquisition module 2 is used to continuously acquire images of the formed adhesive line through a vision sensing module during the dispensing process, and to perform sub-pixel level edge extraction on the formed adhesive line images to obtain real-time contour data of the adhesive line.

[0089] The flow front state discrimination module 3 is used to discriminate the current flow front state of the colloid based on the real-time contour data. The flow front state includes at least the spreading state, the accumulation state, and the fracture state.

[0090] The corrected dispensing path generation module 4 calculates the correction weight of the subsequent dispensing path segment based on the flow front state, and performs a gradual correction on the dispensing path segment based on the correction weight to generate a first corrected dispensing path.

[0091] The dispensing path correction execution module 5 is used to control the dispensing mechanism to perform operations according to the first corrected dispensing path and to iterate the visual acquisition and path correction process.

[0092] Furthermore, the flow front state determination module 3 is used to perform the following steps: Extract the edge curvature features of the cross-section of the adhesive line and the edge smoothness on both sides of the adhesive line from the real-time contour data; when the cross-section of the adhesive line is flat and the edge curvature features meet the first preset condition and the edge smoothness meets the second preset condition, it is determined to be in a spreading state; when the cross-section of the adhesive line is dome-shaped and the edge curvature features meet the third preset condition, it is determined to be in a stacked state; when a continuous interruption or tail retraction is detected at the edge of the adhesive line based on the real-time contour data, it is determined to be in a broken state.

[0093] Furthermore, the modified dispensing path generation module 4 is used to perform the following steps: Based on the category of the flow front state, determine the corresponding state influence level; obtain the path distance between the current dispensing position and the path segment to be dispensed; determine the basic correction magnitude based on the state influence level and the path distance; and perform segment-by-segment attenuation processing on the basic correction magnitude in order of path distance from near to far to obtain the correction weight corresponding to each path segment to be dispensed.

[0094] Furthermore, the modified dispensing path generation module 4 is used to perform the following steps: The mapping range of the correction weight covers N consecutive path segments after the current dispensing position, and the correction weight is distributed exponentially with decreasing along the path extension direction. The closer the path segment is to the point of anomaly, the higher the correction weight.

[0095] Furthermore, the modified dispensing path generation module 4 is used to perform the following steps: Obtain the basic operating parameters of each sub-path segment in the dispensing path segment. The basic operating parameters include at least the dispensing speed, the Z-axis height of the dispensing head, and the start-stop dwell time. Couple the correction weights corresponding to the sub-path segments with the dispensing speed, the Z-axis height of the dispensing head, and the start-stop dwell time in the basic operating parameters to obtain the corrected operating parameters. Bind the corrected operating parameters to the corresponding sub-path segments to form a set of corrected path segments. Then, concatenate the set of corrected path segments in sequence to generate the first corrected dispensing path.

[0096] Furthermore, the modified dispensing path generation module 4 is used to perform the following steps: If the flow front is in a stacked state, the Z-axis height of the dispensing head is increased and the dispensing speed is decreased, using the correction weight as a scaling factor; if the flow front is in a broken state, the start-stop dwell time is increased, using the correction weight as a scaling factor; if the flow front is in a spreading state, the Z-axis height of the dispensing head is decreased and the dispensing speed is increased, using the correction weight as a scaling factor.

[0097] Furthermore, the corrected dispensing path execution module 5 is used to perform the following steps: The dispensing mechanism is controlled to move along the first corrected dispensing path and perform dispensing operations according to the corrected operating parameters. After completing the current dispensing path segment, the vision sensing module is triggered again to acquire the next frame of the formed glue line image. Subpixel-level edge extraction is performed on the next frame of the formed glue line image to obtain updated real-time contour data. Based on the updated real-time contour data, the flow front state of the glue is re-determined, and based on the updated flow front state, the weight correction of the subsequent dispensing path segment is performed until the workpiece to be dispensed is completed.

[0098] Furthermore, the flow front state determination module 3 is used to perform the following steps: Along the extension direction of the real-time contour data, several contour micro-segments of equal length are divided; the frequency of abrupt changes in the contour gradient direction within each contour micro-segment is counted to obtain the local burr index; the local burr indices of all contour micro-segments are summarized to obtain the overall edge burr index; the offsets of the left and right edges of the adhesive line in the real-time contour data at corresponding positions are calculated, and the dispersion of the offsets is obtained to obtain the edge symmetry; the overall edge burr index and the edge symmetry are fused to calculate the edge roughness, and the edge roughness is used as an auxiliary discrimination parameter for flow front state discrimination.

[0099] Furthermore, obtaining the initial contour of the workpiece to be glued is used to perform the following steps: Extract the curvature extrema and inflection points from the initial contour. Based on the distribution of the curvature extrema and inflection points, divide the initial dispensing path into a high-speed segment, a normal-speed segment, and a low-speed segment. Configure corresponding basic operating parameters for the high-speed segment, the normal-speed segment, and the low-speed segment respectively. During the iterative process, add the correction weights to the basic operating parameters of the high-speed segment, the normal-speed segment, and the low-speed segment respectively to obtain the final operating parameters of each segment.

[0100] The intelligent dispensing path planning system based on machine vision contour features provided in this embodiment of the invention can execute the intelligent dispensing path planning method based on machine vision contour features provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0101] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0102] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for intelligent dispensing path planning based on machine vision contour features, characterized in that, The method includes: Obtain the initial outline of the workpiece to be glued, generate an initial glue dispensing path, and perform glue dispensing based on the initial glue dispensing path; During the dispensing process, the visual sensing module continuously acquires images of the formed glue line and performs sub-pixel level edge extraction on the images of the formed glue line to obtain real-time contour data of the glue line. Based on the real-time contour data, the current flow front state of the colloid is determined, and the flow front state includes at least the spreading state, the accumulation state, and the fractured state. Based on the flow front state, the correction weight of the subsequent dispensing path segment is calculated, and the dispensing path segment is gradually corrected based on the correction weight to generate the first corrected dispensing path. The dispensing mechanism is controlled to perform the operation according to the first corrected dispensing path, and the visual acquisition and path correction process is iterated in a loop.

2. The intelligent dispensing path planning method based on machine vision contour features as described in claim 1, characterized in that, Based on the real-time contour data, the current flow front state of the colloid is determined, including: Extract the edge curvature features of the cross-section of the adhesive line and the edge smoothness on both sides of the adhesive line from the real-time contour data; When the cross-section of the adhesive line is flat and the edge curvature features meet the first preset condition and the edge smoothness meets the second preset condition, it is determined to be in the spreading state; When the cross-section of the adhesive line is dome-shaped and the edge curvature characteristics meet the third preset condition, it is determined to be in a stacked state; When the real-time contour data detects a continuous interruption at the edge of the adhesive line or a retraction at the tail, it is determined to be in a broken state.

3. The intelligent dispensing path planning method based on machine vision contour features as described in claim 1, characterized in that, Based on the flow front state, the correction weights for subsequent dispensing path segments are calculated, including: Based on the category of the flow front state, determine the corresponding state impact level; Obtain the path distance between the current dispensing position and the dispensing path segment; The basic correction range is determined based on the state impact level and the path distance; The basic correction magnitude is gradually attenuated according to the path distance from near to far to obtain the correction weight corresponding to each dispensing path segment.

4. The intelligent dispensing path planning method based on machine vision contour features as described in claim 3, characterized in that, The mapping range of the correction weight covers N consecutive path segments after the current dispensing position, and the correction weight is distributed exponentially with decreasing along the path extension direction. The closer the path segment is to the point of anomaly, the higher the correction weight.

5. The intelligent dispensing path planning method based on machine vision contour features as described in claim 4, characterized in that, Based on the correction weight, the dispensing path segment to be dispensed is gradually corrected to generate a first corrected dispensing path, including: Obtain the basic operating parameters of each sub-path segment in the dispensing path segment. The basic operating parameters include at least the dispensing speed, the Z-axis height of the dispensing head, and the start and stop dwell time. The correction weights corresponding to the sub-path segments are coupled with the dispensing speed, the Z-axis height of the dispensing head, and the start-stop dwell time in the basic operating parameters to obtain the corrected operating parameters. The modified operating parameters are bound to the corresponding sub-path segments to form a modified path segment set, and the modified path segment set is spliced ​​in sequence to generate the first modified dispensing path.

6. The intelligent dispensing path planning method based on machine vision contour features as described in claim 5, characterized in that, The correction weights corresponding to the sub-path segments are coupled with the dispensing speed, dispensing head Z-axis height, and start / stop dwell time in the basic operating parameters for calculation, including: If the flow front state is a stacked state, then the value of the Z-axis height of the dispensing head is increased and the value of the dispensing speed is decreased, using the correction weight as a proportional factor. If the flow front state is a fractured state, then the value of the start-stop dwell time is increased by using the correction weight as a proportional factor. If the flow front is in a spreading state, then the Z-axis height of the dispensing head is reduced and the dispensing speed is increased, using the correction weight as a scaling factor.

7. The intelligent dispensing path planning method based on machine vision contour features as described in claim 6, characterized in that, The dispensing mechanism is controlled to perform the operation according to the first corrected dispensing path, and the vision acquisition and path correction process is iterated cyclically, including: The dispensing mechanism is controlled to move along the first corrected dispensing path and to perform dispensing operations according to the corrected operating parameters; After completing the current dispensing path segment, the visual sensing module is triggered again to acquire the next frame of the formed glue line image. Subpixel-level edge extraction is performed on the next frame of the formed adhesive line image to obtain updated real-time contour data; Based on the updated real-time contour data, the flow front state of the colloid is re-determined, and based on the updated flow front state, the weight correction of the subsequent dispensing path segment is performed until the workpiece to be dispensed is completed.

8. The intelligent dispensing path planning method based on machine vision contour features as described in claim 1, characterized in that, The method also includes the calculation of edge roughness, including: Divide the real-time contour data into several equal-length contour segments along the extension direction of the real-time contour data; The frequency of abrupt changes in the gradient direction within each contour micro-segment is counted to obtain the local burr index. The overall edge burr index is obtained by summing the local burr indices of all contour micro-segments. The offsets of the left and right edges of the adhesive line in the real-time contour data at corresponding positions are calculated, and the dispersion of the offsets is obtained to obtain the edge symmetry. The edge roughness is calculated by combining the overall edge burr index and the edge symmetry, and the edge roughness is used as an auxiliary discrimination parameter to determine the state of the flow front.

9. The intelligent dispensing path planning method based on machine vision contour features as described in claim 1, characterized in that, The generation of the initial dispensing path also includes: Extract the curvature extrema and inflection points in the initial contour, and divide the initial dispensing path into high-speed segment, normal-speed segment and low-speed segment based on the distribution of the curvature extrema and inflection points. Configure corresponding basic operating parameters for the high-speed segment, the constant-speed segment, and the low-speed segment respectively; During the iterative process, the corrected weights are superimposed on the basic operating parameters of the high-speed segment, the normal-speed segment, and the low-speed segment respectively to obtain the final operating parameters of each segment.

10. A machine vision contour feature-based intelligent dispensing path planning system, characterized in that, The system is used to implement the intelligent dispensing path planning method based on machine vision contour features as described in any one of claims 1-9, the system comprising: The initial dispensing path acquisition module is used to acquire the initial contour of the workpiece to be dispensed, generate the initial dispensing path, and perform dispensing based on the initial dispensing path. The glue line contour data acquisition module is used to continuously acquire images of the formed glue line through a vision sensing module during the glue dispensing process, and to perform sub-pixel level edge extraction on the formed glue line images to obtain real-time contour data of the glue line. The flow front state discrimination module is used to determine the current flow front state of the colloid based on the real-time contour data. The flow front state includes at least the spreading state, the accumulation state, and the fracture state. The corrected dispensing path generation module calculates the correction weight of the subsequent dispensing path segment based on the flow front state, and performs a gradual correction on the dispensing path segment based on the correction weight to generate the first corrected dispensing path. The dispensing path correction execution module is used to control the dispensing mechanism to perform the operation according to the first corrected dispensing path, and to iteratively cycle through the visual acquisition and path correction process.

Images