Method for automatically matching component and bonding pad

Abstract

The invention relates to the technical field of element bonding pad matching, in particular to a method for automatically matching a component and a bonding pad, which comprises the following steps of: directly and dynamically calculating a coverage area according to the size of the component, and detecting a bonding pad layer in real time without pre-storing a mapping relation; the electrical feasibility verification that the number of bonding pads is greater than or equal to the number of pins is increased to solve the risk of insufficient solder or solder skips; the coordinate correction is matched with the mean correction coordinate of the central point of the bonding pad, and the angle correction is realized by identifying a silk-screen polarity identifier through AI to realize a dual-stage correction technology; polarity identification consistency verification is newly added, so that the installation correctness of a high-reliability device is ensured, and polarity direction guarantee is realized; the workload of a client for pre-storing the template is reduced, and the matching precision is improved through dynamic detection and closed-loop correction.

Description

Technical Field

[0001] The present invention relates to the technical field of component pad matching, and in particular to a method for automatically matching components and pads. Background Art

[0002] The role of component pad matching is mainly reflected in three aspects: ensuring welding reliability, improving assembly efficiency and maintaining circuit performance. Mismatch between pad and component pin size will lead to insufficient overlap or overload, causing displacement or short circuit, and there will also be the phenomenon of cold soldering.

[0003] Current methods for matching pads to components pre-store a mapping between pads and components, then find the pre-stored pad information based on the component and compare it with the actual pad information. This requires customers to pre-store the pad-component mapping, which increases their workload. Furthermore, if the pre-stored information is incorrect, matching accuracy will be reduced, increasing the risk of cold or leaky solder joints and compromising installation accuracy.

[0004] In view of this, we propose a method for automatic matching of components and pads to solve the existing problem. Summary of the Invention

[0005] The object of the present invention is to provide a method for automatically matching components and pads to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for automatically matching components and pads, comprising: S1: Obtain the size information of the components and calculate the coverage area of ​​the components on the circuit board based on the size information; S2: Detect all pads in the coverage area in the pad layer to form a candidate pad set; S3: Determine whether the number of pads in the candidate pad set is greater than or equal to the number of pins of the component; if so, determine that the candidate pad set is successfully matched with the component; if not, determine that the match fails; S4: Perform at least one of coordinate correction and angle correction on the successfully matched components.

[0007] Furthermore, in the coordinate correction, the average center coordinate is calculated based on the center point coordinates of each pad in the candidate pad set, and the installation coordinates of the component are corrected based on the average center coordinates.

[0008] Furthermore, during angle correction, AI is used to identify the polarity mark in the silk-screen layer, determine the actual polarity direction of the component, and rotate the installation angle of the component to align with the polarity mark.

[0009] Furthermore, the step of determining whether the number of pads matches includes: if the number of pads is equal to the number of pins, directly binding the candidate pad set to the component; if the number of pads is greater than the number of pins, screening out the pad set with the highest overlap with the pin position for binding.

[0010] Furthermore, the coordinate correction step specifically includes: calculating the coordinate mean of the center points of all pads in the candidate pad set; and offsetting the installation coordinates of the components to the coordinate mean position.

[0011] Furthermore, the angle correction step specifically includes: identifying the polarity identification direction associated with the component in the silk screen layer; calculating the deviation value between the polarity identification direction and the current angle of the component; and rotating the installation angle of the component according to the deviation value.

[0012] Furthermore, for components with polarity, after the angle correction, the process also includes: verifying whether the polarity direction of the component is consistent with the polarity identification direction of the pad; if not, outputting a polarity error alarm.

[0013] Furthermore, the coverage area is a rectangular or polygonal area centered on the component coordinates and covering its physical boundaries.

[0014] Furthermore, after determining that the matching fails, the range of the coverage area is expanded according to a preset step size, and the pads are re-detected until the matching is successful or the maximum expansion threshold is reached.

[0015] Furthermore, the acquisition of component size information includes: identifying the physical boundaries of the component body through computer vision, extracting the length, width and contour features; the calculation of the coverage area includes: taking the actual installation coordinates of the component as the center, combining the length and width dimensions to construct an initial rectangular coverage area, and expanding it into a polygonal coverage area when the contour feature is an irregular shape.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention directly and dynamically calculates the coverage area according to the component size and detects the pad layer in real time without the need for pre-stored mapping relationships; adds electrical feasibility verification that the number of pads is greater than or equal to the number of pins to resolve the risk of cold soldering or leaking solder joints; coordinate correction is performed to match the mean value of the pad center point to correct the coordinates, and angle correction is performed through AI to identify the silk screen polarity mark, realizing a two-stage correction technology; newly added polarity mark consistency verification ensures the correct installation of high-reliability components and realizes polarity direction protection; it not only reduces the workload of customers in pre-stored templates, but also improves matching accuracy through dynamic detection and closed-loop correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the workflow of a method for automatically matching components and pads according to the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0019] The following is a discussion of the method of Example 1: like Figure 1 As shown, a method for automatically matching components and pads includes: S1: Obtain the size information of the component and calculate the coverage area of ​​the component on the circuit board based on the size information; the coverage area is a rectangular or polygonal area centered on the component coordinates and covering its physical boundaries; S2: Detect all pads in the coverage area in the pad layer to form a candidate pad set; S3: Determine whether the number of pads in the candidate pad set is greater than or equal to the number of pins of the component; if so, determine that the candidate pad set is matched with the component successfully; if not, determine that the match fails; after determining that the match fails, expand the coverage area according to a preset step size and re-detect the pads until the match is successful or the maximum expansion threshold is reached; S4: Perform at least one of coordinate correction and angle correction on the successfully matched components; in coordinate correction, calculate the average center coordinate based on the center point coordinates of each pad in the candidate pad set, and use this to correct the installation coordinates of the component; in angle correction, use AI to identify the polarity mark in the silk screen layer to determine the actual polarity direction of the component, and rotate the installation angle of the component to align with the polarity mark.

[0020] The step of determining whether the number of pads matches includes: if the number of pads is equal to the number of pins, directly binding the candidate pad set to the component; if the number of pads is greater than the number of pins, screening out the pad set with the highest overlap with the pin position for binding.

[0021] The coordinate correction steps specifically include: calculating the coordinate mean of all pad center points in the candidate pad set; and offsetting the installation coordinates of the components to the coordinate mean position.

[0022] The angle correction steps specifically include: identifying the polarity identification direction associated with the component in the silk screen layer; calculating the deviation value between the polarity identification direction and the current angle of the component; and rotating the installation angle of the component according to the deviation value.

[0023] For components with polarity, after angle correction, the following steps are also performed: verifying whether the polarity direction of the component is consistent with the polarity identification direction of the pad; if not, outputting a polarity error alarm.

[0024] The working principle of the method for automatically matching components and pads based on the first embodiment is as follows: The acquisition of component size information includes: identifying the physical boundaries of the component body through computer vision, extracting the length, width and contour features; the calculation of the coverage area includes: taking the actual installation coordinates of the component as the center, combining the length and width dimensions to construct an initial rectangular coverage area, and expanding it into a polygonal coverage area when the contour features are irregular shapes.

[0025] The steps of dimension acquisition specifically include: edge detection on the component body image to generate the minimum circumscribed rectangle; when there are convex or concave features on the body, the convex hull polygon is calculated based on the contour point set and the vertex coordinates are recorded.

[0026] The steps of calculating the coverage area specifically include: if the contour feature is a rectangle, generating a coverage area that coincides with the minimum circumscribed rectangle; if the contour feature is a polygon, generating a minimum area convex polygon coverage area that includes all convex hull vertices.

[0027] The formation of a candidate pad set involves coarse pad screening, topological relationship construction, and hierarchical clustering. Coarse pad screening involves identifying all conductive patterns in the pad layer within the component footprint and extracting those that meet area thresholds and shape rules as the initial pad set. Topological relationship construction involves calculating the spatial topological relationships between pads in the initial pad set and generating a topological relationship graph containing a distance matrix and an azimuth matrix. Hierarchical clustering involves clustering pads using the DBSCAN algorithm based on the topological relationship graph, outputting pad clusters that meet the component pin distribution rules as candidate pad sets.

[0028] The steps of constructing the topological relationship specifically include: establishing a polar coordinate system with the center of the coverage area as the origin; calculating the polar radius ρ and polar angle θ of the center point of each pad, and generating an azimuth matrix: M angle [i][j]=|θ i -θ j |, i and j are pad indices, and ρ i <ρ j ; Generate distance matrix: M dist [i][j]= , x i with y i is the horizontal and vertical coordinate value corresponding to pad index i, x j with y j are the horizontal and vertical coordinate values ​​corresponding to pad index j.

[0029] The hierarchical clustering process includes first-level clustering and second-level filtering. The first-level clustering uses a distance matrix as input, setting a neighborhood radius of 1.5 times the pin spacing. Pads with spacing smaller than the neighborhood radius are grouped together. The second-level filtering calculates the standard deviation of the orientation matrix within each cluster. If the standard deviation exceeds the angle tolerance threshold, the cluster is split until each sub-cluster satisfies the standard deviation less than or equal to the threshold.

[0030] The hierarchical clustering steps specifically include pad type verification: extracting the contour features of each pad in the candidate pad set; if the component is a polar device, identifying the pad containing the polarity mark and marking it as the reference pad; verifying whether the relative position of the reference pad and the remaining pads conforms to the pin distribution template of the component.

[0031] The dynamic generation method of the pin distribution template includes: counting the center coordinates of all pads in the candidate pad set; fitting the minimum circumscribed polygon of the pad center point; comparing the number of polygon vertices with the number of component pins, and generating a distribution template based on the vertices if they are consistent.

[0032] Before the pad coarse screening step, grayscale difference enhancement processing is also included. The operations on the pad layer image include: extracting the channel with the maximum contrast between the pad and the substrate in the RGB channel; performing histogram equalization on the channel to enhance the contrast of the pad edge; and using adaptive threshold segmentation to obtain a binary pad mask.

[0033] Coordinate correction includes pad weight assignment, weighted centroid calculation, outlier robustness handling, and coordinate shift execution. Pad weight assignment calculates weight factors based on the geometric properties of each pad in the candidate pad set. Geometric properties include pad area, shape regularity, and position confidence. Weighted centroid calculation uses a weighted average algorithm to calculate the corrected reference point coordinates based on the center coordinates of each pad and its weight factor. Outlier robustness handling uses an iterative reweighted least squares method to eliminate pads whose reference point offset exceeds a threshold and recalculate the weighted centroid until convergence. Coordinate shift execution corrects the component's installation coordinates to the final reference point position.

[0034] The pad weight allocation steps specifically include: calculating the area weight factor W area =(A i / A max ) α , where A i is the current pad area, A max is the maximum pad area in the candidate set, α=0.5-1.0; calculate the shape regularity factor W shape =1-|1-[Pi 2 / (4πAi)]|, where P i Is the pad perimeter, circular pad W shape =1; comprehensive weight W i =β·W area +γ·W shape , where β+γ=1, and β>γ.

[0035] The steps of robust outlier processing include: initializing the weights of all pads to W i ; Calculate the weighted centroid coordinates (μ x ,μy ): μ x =Σ(W i ·x i ) / ΣW i , μ y =Σ(W i ·y i ) / ΣW i ; Calculate the Euclidean distance di from each pad to the center of mass; if di>k·σ d (σ d is the distance standard deviation, k=2.0-3.0), then reset the pad weight to zero; repeat the above process until no pad is removed or the maximum number of iterations is reached.

[0036] The position confidence is obtained by performing sub-pixel edge detection on the pad image and fitting the standard deviation of the edge points as the position fluctuation σ edge ; Position confidence: C pos =1 / (1+σ edge / δ), where δ is the maximum allowed edge fluctuation threshold; add the confidence term W to the weight factor i '=W i ·C pos .

[0037] Angle correction includes multimodal feature extraction, feature fusion, and progressive angle optimization. The multimodal feature extraction step includes: synchronously collecting image data of the silk screen layer, solder mask layer, and pad layer, extracting the contour direction and azimuth of the polarity identification symbol of the silk screen layer, the asymmetric geometric features of the window area of ​​the solder mask layer, and the size difference ratio between the polar pad and the non-polar pad of the pad layer. The feature fusion step includes: fusing three types of features through an attention weighted network to output the theoretical polarity direction vector of the component. The progressive angle optimization step includes: starting from the current installation angle, iteratively rotating and searching within the range of ±180° according to the golden section ratio, calculating the cosine similarity between the actual polarity direction and the theoretical vector at each angle, and taking the angle with the largest similarity as the correction output.

[0038] The feature fusion step includes building a dual-channel attention mechanism and outputting a theoretical polarity direction vector. In building a dual-channel attention mechanism, spatial attention generates a heat map for the polarity symbol outline of the silk screen layer, focusing on high gradient areas; channel attention assigns weights to the three types of features, and the weight of the silk screen layer W silkscreen =0.5, the weight of the solder mask layer W soldermask =0.3, pad layer weight Wpad=0.2. Output theoretical polarity direction vector V theory =Σ(W i ·V i ) / ||Σ(W i ·V i )||, where Vi Normalize the direction vector of each layer feature.

[0039] The progressive angle optimization steps include: initialization, golden section iteration, and output of the global optimal angle θ opt In initialization, the angle search interval [θ min ,θ max ]=[current angle -180°, current angle +180°]. The steps of golden section iteration include: calculating the segmentation point θ a =θ max -(θ max -θ min ) / φ,θ b =θ min +(θ max -θ min ) / φ, φ is the golden ratio 1.618; calculate θ a ,θ b The cosine similarity S(θ)=cos(V actual (θ),V theory ); retain the subintervals corresponding to higher values ​​of S(θ); and repeat the iterations until the interval length is < 0.1° or 10 iterations are reached.

[0040] The progressive angle optimization also includes a pad position verification step: opt Calculate the average distance D between the component pin projection point and the candidate pad center at the angle avg If D avg >50% of the pin diameter, the conflict handling is activated. Activating the conflict handling includes: opt -15°,θ opt +15°] interval is scanned in 1° steps; the angles that meet the conditions of S(θ)>0.95 and D avg (θ)≤30% of the pin diameter.

[0041] The extraction of asymmetric geometric features includes: identifying the centroid of the window area of ​​the solder mask layer; calculating the major axis direction of the second-order moment ellipse of the window area; when the major axis eccentricity is greater than 0.7 and the centroid deviates from the geometric center, it is determined that a polar feature exists.

[0042] Angle correction also includes a deep learning enhancement step: using a pre-trained polarity recognition model to process the blurred silk screen image, inputting a multi-scale pyramid image of the silk screen layer ROI area, and outputting the polarity symbol category (triangle / bar / dot) and direction confidence; when the direction confidence is >0.9, the model output direction is directly used as V silkscreen .

[0043] Polarity consistency verification methods include time-space synchronous acquisition, dual-stream feature extraction, and adaptive decision-making. Time-space synchronous acquisition continuously captures multiple frames of relative position images of components and pads through a high-speed camera during the movement of the placement head. In dual-stream feature extraction, the spatial stream extracts the polarity identification direction angle θ of the component in the current frame. component and pad polarity marking direction angle θ pad , time flow analysis of θ in N consecutive frames (N≥3) component and θ pad The adaptive decision is based on the spatial deviation Δθ=|θ component -θ pad | and the time flow stability coefficient K t , perform graded response: if Δθ≤threshold θ th And K t ≥0.8, the polarity is determined to be consistent; if Δθ>θ th But K t >0.9, it is determined to be instantaneous occlusion and re-verification is started; if Δθ>θ th And K t <0.6, triggering a real-time emergency stop alarm.

[0044] Time flow stability coefficient K t The calculation includes: calculating the mean value μ of the angle difference of consecutive frames Δθ and standard deviation σ Δθ ; Stability coefficient K t =exp(-σ Δθ / μ Δθ ).

[0045] Threshold θ th The dynamic calculation includes: basic threshold θ base =arctan(pin diameter / pad center distance); speed compensation term θ vel =k·v·Δt, where v is the placement head speed, Δt is the image sampling interval, and k is the coefficient; the final threshold θ th =θ base +θ vel .

[0046] The re-verification operation includes: controlling the placement head to rise to a safe height; starting multi-angle scanning, rotating the component around the Z axis in 10° steps, and collecting images in 12 orientations; constructing a polarity direction probability distribution map, and taking the direction with the highest probability as the final polarity direction.

[0047] The revalidation operation also includes a deep learning arbitration step: when the graded response result is in doubt, multiple frames of images are input into a pre-trained polarity arbitration network; the network outputs a three-dimensional arbitration vector [probability of agreement, probability of disagreement, uncertainty]; if the probability of agreement is greater than 0.7 and the uncertainty is less than 0.2, the original judgment result is overwritten.

[0048] The polarity arbitration network adopts a spatiotemporal fusion architecture: in the spatial branch, 3D-ResNet extracts single-frame spatial features; in the temporal branch, LSTM encodes the temporal relationship between consecutive frames; in the fusion layer, the spatial-temporal features are weighted through the attention mechanism.

[0049] The above specific embodiments are only several preferred embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for automatically matching components and pads, characterized in that: include: S1: Obtain the size information of the components and calculate the coverage area of ​​the components on the circuit board based on the size information; S2: Detect all pads in the coverage area in the pad layer to form a candidate pad set; S3: Determine whether the number of pads in the candidate pad set is greater than or equal to the number of pins of the component; if so, determine that the candidate pad set is successfully matched with the component; if not, determine that the match fails; S4: Perform at least one of coordinate correction and angle correction on the successfully matched components.

2. The method for automatically matching components and pads according to claim 1, wherein: In coordinate correction, the average center coordinates are calculated based on the center point coordinates of each pad in the candidate pad set, and the installation coordinates of the components are corrected based on this average center coordinates.

3. The method for automatically matching components and pads according to claim 1, wherein: During angle correction, AI is used to identify the polarity mark in the silk screen layer, determine the actual polarity direction of the component, and rotate the installation angle of the component to align with the polarity mark.

4. The method for automatically matching components and pads according to claim 1, wherein: The step of determining whether the number of pads matches includes: if the number of pads is equal to the number of pins, directly binding the candidate pad set to the component; if the number of pads is greater than the number of pins, screening out the pad set with the highest overlap with the pin position for binding.

5. The method for automatically matching components and pads according to claim 2, wherein: The coordinate correction steps specifically include: calculating the coordinate mean of all pad center points in the candidate pad set; and offsetting the installation coordinates of the components to the coordinate mean position.

6. The method for automatically matching components and pads according to claim 3, characterized in that: The angle correction steps specifically include: identifying the polarity identification direction associated with the component in the silk screen layer; calculating the deviation value between the polarity identification direction and the current angle of the component; and rotating the installation angle of the component according to the deviation value.

7. The method for automatically matching components and pads according to claim 6, characterized in that: For components with polarity, after angle correction, the following steps are also performed: verifying whether the polarity direction of the component is consistent with the polarity identification direction of the pad; if not, outputting a polarity error alarm.

8. The method for automatically matching components and pads according to claim 1, wherein: The coverage area is a rectangular or polygonal area centered on the component coordinates and covering its physical boundaries.

9. The method for automatically matching components and pads according to claim 1, wherein: After determining that the matching fails, the coverage area is expanded according to the preset step size and the pads are re-detected until the matching is successful or the maximum expansion threshold is reached.

10. The method for automatically matching components and pads according to claim 1, characterized in that: The acquisition of component size information includes: identifying the physical boundaries of the component body through computer vision, extracting the length, width and contour features; the calculation of the coverage area includes: taking the actual installation coordinates of the component as the center, combining the length and width dimensions to construct an initial rectangular coverage area, and expanding it into a polygonal coverage area when the contour features are irregular shapes.

Images