Automobile circuit board solder paste printing quality detection method and system

Abstract

The invention belongs to the technical field of image analysis and processing, and relates to an automobile circuit board solder paste printing quality detection method and system. The method comprises the steps of constructing a Laplacian pyramid for a target solder paste area image, extracting multi-level geometric and position features, performing weighting based on local phase consistency to obtain a first fusion feature, and comparing the first fusion feature with a first fusion feature of a historical qualified sample to obtain a second fusion feature; calculating a time sequence drift correction characteristic capable of correcting the drift in the production process; constructing a neighborhood feature map by combining the first fusion features of the target region and the spatial neighborhood of the target region, and extracting spatial morphological features by using an orthogonal Gabor filter; and performing vector splicing on the first fusion feature, the time sequence drift correction feature and the spatial form feature to generate a comprehensive combined feature, and inputting the comprehensive combined feature into a predetermined classification model to obtain a printing quality detection result. According to the invention, the detection rate of the solder paste area defect on the automobile circuit board can be improved.

Description

Technical Field

[0001] This invention relates to the field of image analysis and processing technology, specifically to a method and system for inspecting the printing quality of solder paste on automotive circuit boards. Background Technology

[0002] Automotive circuit boards, due to their connection to driving safety, have much higher quality requirements than circuit boards in everyday office equipment. In surface mount technology, solder paste printing is a core process, and its quality directly determines the effectiveness of subsequent component placement and reflow soldering. Therefore, to ensure product quality, the industry typically incorporates an optical inspection (SPI) process after solder paste printing to monitor the solder paste pattern on each pad in real time.

[0003] Existing SPI technology primarily acquires images of various solder paste areas on a circuit board using 2D or 3D vision imaging systems, and then extracts a series of geometric and positional features of the solder paste printing, such as area, volume, height, perimeter, centroid coordinates, and X / Y offset. Subsequently, these extracted feature values ​​are compared with pre-set upper and lower limits of engineering specifications to determine whether the printing quality of a single solder paste is up to standard. However, this method has drawbacks; it is highly sensitive to changes in lighting, image noise, and minor imperfections caused by uneven distribution of metal powder particles on the solder paste surface. This leads to unstable feature extraction and is prone to misjudgment.

[0004] Furthermore, in actual production environments, solder paste printing is a process influenced by a combination of factors, including squeegee pressure, stencil tension, and ambient temperature and humidity. These factors may change slowly and continuously over time, causing a holistic temporal drift in the printing pattern of the entire batch or multiple consecutive batches of solder paste. Traditional inspection methods cannot track this process drift. When the normal printing baseline changes, qualified solder paste is easily misjudged as defective, or early failure trends cannot be detected in time. In addition, the distribution of solder paste defects on circuit boards often exhibits spatial correlation. For example, defects caused by stencil blockage, substrate warping, or equipment calibration deviations usually manifest as abnormal morphology of multiple adjacent solder pastes within a local area. Traditional point-by-point inspection methods cannot extract spatial morphological information representing regional defects from neighborhood relationships, resulting in a low detection rate for such correlated defects. Summary of the Invention

[0005] Therefore, the purpose of this invention is to propose a method for inspecting the solder paste printing quality of automotive circuit boards, in order to solve the problems of existing inspection methods, such as sensitivity to light noise leading to misjudgment, inability to track process timing drift, and low detection rate of regionally related defects due to neglect of spatial correlation of defects.

[0006] To solve the above problems, the present invention provides a technical solution for a method of inspecting the solder paste printing quality of automotive circuit boards: A method for inspecting the printing quality of solder paste on automotive circuit boards includes the following steps: S1, acquire two-dimensional images of each solder paste area on the automotive circuit board to be inspected, and perform the following operations for each target solder paste area: S11, construct a Laplacian pyramid containing multiple levels based on the image of the target solder paste area, and extract the geometric and positional features of the target solder paste area as level features at each level of the Laplacian pyramid; S12, calculate the local phase consistency of each layer image as the confidence weight of the layer, and perform a weighted summation of the layer features of the multiple layers based on the confidence weight to obtain the first fusion feature; S2, extract the first fusion features of a predetermined number of the most recent qualified samples with the same label as the target solder paste area from the historical database; based on the extracted first fusion features of the most recent qualified samples, calculate their mean vector and the inverse matrix of their covariance matrix; perform a difference operation between the first fusion features of the current target solder paste area and the mean vector, and multiply the difference result by the inverse matrix to obtain the time drift correction features; S3. Based on the geometric center coordinates of the target solder paste area in the circuit board design file, determine a predetermined number of spatial nearest neighbor solder paste areas of the target solder paste area, and construct a neighborhood feature map composed of the first fusion feature of the target solder paste area and the predetermined number of nearest neighbor solder paste areas; use a set of orthogonal Gabor filters with preset parameters to filter the neighborhood feature map, and take the maximum value or mean value of all filtered responses as spatial morphological features. S4, the first fusion feature, spatial morphology feature and temporal drift correction feature are vector-concatenated to generate a combined feature; the combined feature is input into a predetermined classification model to obtain the printing quality detection result of the target solder paste area.

[0007] Furthermore, the method for constructing the Laplace pyramid includes: Gaussian smoothing and interlaced downsampling are applied to the image of the target solder paste region to construct a Gaussian pyramid. The Laplacian pyramid is generated by subtracting each layer of the Gaussian pyramid from the upsampled and blurred image of the next layer.

[0008] Furthermore, the number of levels in the Laplace pyramid is set to 4; the geometric and positional features include parameters in 7 dimensions: area, perimeter, aspect ratio, compactness, offset, X-coordinate of the centroid, and Y-coordinate of the centroid. These 7-dimensional parameters are combined to form the feature vector of the corresponding level.

[0009] Furthermore, the calculation method for the first fusion feature is as follows: For all levels of the Laplace pyramid, multiply the feature vector of each level by the confidence weight of that level. The first fusion feature is calculated by summing the product results of all levels and then dividing by the sum of the confidence weights of all levels.

[0010] Furthermore, the method for extracting the geometric and positional features of the target solder paste region at each level of the Laplace pyramid is as follows: For each level of the Laplacian pyramid image, binarization is performed using the Otsu method to segment the solder paste region, and the parameters of the seven dimensions are calculated based on the segmented solder paste region.

[0011] Furthermore, the method for calculating the local phase consistency of each layer image is as follows: apply a set of preset multi-scale multi-directional logarithmic Gabor filters to each layer image, calculate the sum of the energy of the filter response, and obtain the local phase consistency value of the layer image.

[0012] Furthermore, the method for constructing the neighborhood feature map includes: The number of solder paste regions that are closest to the geometric center of the target solder paste region in terms of Euclidean distance are designated as the nearest neighbor solder paste regions. A neighborhood feature map is constructed, and the first fused feature vector of the target solder paste region is placed at the center of the neighborhood feature map. The first fused feature vectors of the predetermined number of nearest neighbor solder paste regions are then filled into the remaining positions of the neighborhood feature map according to their orientation.

[0013] Furthermore, the method for obtaining the spatial morphological features includes: The neighborhood feature map is filtered using eight orthogonal Gabor filters. The orthogonal Gabor filters have four directional parameters and two different scale parameters. Calculate the filtering response of each orthogonal Gabor filter across all feature dimensions of the neighborhood feature map, and take the maximum value of the response as the response value of the orthogonal Gabor filter; The spatial morphological features are obtained by concatenating the response values ​​of eight orthogonal Gabor filters into vectors.

[0014] Furthermore, the predetermined classification model is a support vector machine model, and the kernel function of the support vector machine model is a radial basis function.

[0015] The technical solution of the automotive circuit board solder paste printing quality inspection system provided by this invention is as follows: An automotive circuit board solder paste printing quality inspection system includes a processor and a memory. The memory stores a computer program. When the computer program is run by the processor, it implements the automotive circuit board solder paste printing quality inspection method described in any of the above technical solutions.

[0016] The beneficial effects of this invention are as follows: Compared with existing technologies, this invention constructs a multi-level image pyramid and uses local phase consistency as a confidence level for feature weighted fusion, generating a more stable first fusion feature that is insensitive to interference factors such as illumination changes and image noise. Through time-drift correction features, a reference benchmark is established using historical data from recent qualified samples, compensating for overall printing offsets caused by slow changes in process parameters, ensuring that the detection standard aligns with the actual conditions of the current production batch. By analyzing the correlation between the target solder paste and other solder pastes in its spatial neighborhood, spatial morphological features representing the consistency of local morphology are extracted, thereby enhancing the ability to identify clustered and correlated defects caused by stencil blockage, substrate warping, etc., and improving the accuracy of defect detection. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the automotive circuit board solder paste printing quality inspection method of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] Specific embodiments of the automotive circuit board solder paste printing quality inspection method of the present invention: like Figure 1 As shown, the method for inspecting the solder paste printing quality of automotive circuit boards includes the following steps: S1, acquire two-dimensional images of each solder paste area on the automotive circuit board to be inspected, and perform the following operations for each target solder paste area: S11, construct a Laplacian pyramid containing multiple levels based on the image of the target solder paste area, and extract the geometric and positional features of the target solder paste area as level features at each level of the Laplacian pyramid; S12, calculate the local phase consistency of each layer image as the confidence weight of the layer, and perform weighted summation of the layer features of the multiple layers based on the confidence weight to obtain the first fusion feature.

[0020] The two-dimensional image of the solder paste area is obtained through a machine vision system. The inspection equipment uses its built-in industrial camera or optical imaging module as a vision acquisition device to take high-speed pictures of the automotive circuit board placed at the inspection station.

[0021] In step S11, the method for constructing the Laplace pyramid includes: Gaussian smoothing and interlaced downsampling are applied to the image of the target solder paste area to construct a Gaussian pyramid. The Laplacian pyramid is generated by subtracting each layer of the Gaussian pyramid from the upsampled and blurred image of the next layer.

[0022] More specifically, the Laplacian pyramid has four levels, each corresponding to an image representation at different resolutions. The original image is level 0, which is then downsampled and Gaussian blurred to obtain level 1, and so on up to level 3. The geometric and positional features include seven dimensions: area, perimeter, aspect ratio, compactness, offset, and the X and Y coordinates of the centroid. For each level of the Laplacian pyramid image, the solder paste region outline is segmented using Otsu's method for binarization. The total number of pixels in the solder paste region outline is calculated as the area, the number of pixels at each edge is calculated as the perimeter, and the ratio of the number of pixels between two adjacent edges is calculated as the aspect ratio. The compactness of the solder paste region outline is calculated based on the perimeter and area. The X and Y coordinates of the centroid, as well as the offset in the X and Y directions, are calculated by comparing the solder paste region outline with the ideal center position in the design file. These seven dimensions are then combined to form the feature vector for that level. The above process is repeated at all four levels to generate four independent hierarchical feature vectors for each target solder paste region.

[0023] In step S12, a set of multi-scale, multi-directional logarithmic Gabor filters is applied to each layer of the image, and the sum of the energy of the filtered responses is calculated to obtain the local phase consistency value of the image at that layer. The local phase consistency value is then used as the confidence weight for that layer. The higher the local phase consistency value, the clearer the image features and the higher the confidence level.

[0024] For all levels of the pyramid, multiply the feature vector of each level by the confidence weight of that level, sum the product of all levels, and then divide by the sum of the confidence weights of all levels to calculate the first fused feature.

[0025] S2, extract the first fusion features of a predetermined number of the most recent qualified samples with the same label as the target solder paste area from the historical database; based on the first fusion features of the extracted most recent qualified samples, calculate their mean vector and the inverse matrix of their covariance matrix; perform a difference operation between the first fusion features of the current target solder paste area and the mean vector, and multiply the difference result by the inverse matrix to obtain the time drift correction features.

[0026] In actual testing, the production history database records the characteristic data and results of each pad's previous testing. When a new target solder paste area is detected, the unique identifier of the solder paste area in the circuit board design file is identified, for example, the capacitor pad with identifier C101. The database storing all historical testing data is accessed, and all solder paste sample records with the same identifier C101 and whose testing results are deemed qualified are queried. Qualified records are sorted in descending order according to the testing time, and selection starts from the latest record until 500 qualified samples are selected. The first fusion feature of these 500 qualified samples is extracted, and the average value of each first fusion feature is calculated to form a mean vector. At the same time, the covariance matrix between the first fusion features of these 500 qualified samples is calculated, and the inverse of the covariance matrix is ​​obtained. The first fusion feature vector of the current C101 pad under test is subtracted from the previously calculated mean vector to obtain a difference vector. This difference vector is multiplied by the inverse of the covariance matrix, and the result is the timing drift correction feature. The time-series drift correction feature represents the degree of deviation of the current sample from its recent stable production state.

[0027] Of course, in other embodiments, the number of samples is not limited to 500 in this embodiment, and can be reasonably selected according to actual detection needs.

[0028] S3. Based on the geometric center coordinates of the target solder paste area in the circuit board design file, determine a predetermined number of spatial nearest neighbor solder paste areas of the target solder paste area, and construct a neighborhood feature map composed of the first fusion feature of the target solder paste area and the predetermined number of nearest neighbor solder paste areas; use a set of orthogonal Gabor filters with preset parameters to filter the neighborhood feature map, and take the maximum value or mean value of all filtered responses as spatial morphological features.

[0029] In this step, the methods for constructing the neighborhood feature map include: A set number of solder paste regions with the closest Euclidean distance to the geometric center of the target solder paste region are designated as nearest neighbor solder paste regions. A neighborhood feature map is constructed, and the first fused feature vector of the target solder paste region is placed at the center of the neighborhood feature map. The first fused feature vectors of the set number of nearest neighbor solder paste regions are then filled into the remaining positions of the neighborhood feature map according to their orientation.

[0030] Specifically, the geometric center coordinates of the current target solder paste area and the center coordinates of all other solder paste areas on the same circuit board are obtained. The Euclidean distance between the target solder paste area and the geometric centers of all other solder paste areas is calculated one by one, and the eight nearest solder paste areas are selected as the nearest neighbor solder paste areas according to the distance from smallest to largest, creating a 3×3 neighborhood feature map. The first fused feature vector of the target solder paste area itself is placed at the center of the neighborhood feature map. For the selected eight nearest neighbor solder paste areas, based on their actual orientation relative to the target solder paste area (e.g., top left, top right, etc.) or by dividing the angle into eight sectors, with each sector corresponding to a position on the grid, their respective first fused feature vectors are filled into the corresponding eight positions in the 3×3 neighborhood feature map, forming a spatial structure where each position stores a 7-dimensional feature vector.

[0031] In step S3, the methods for obtaining spatial morphological features include: The neighborhood feature map is filtered using eight orthogonal Gabor filters. The orthogonal Gabor filters have four directional parameters and two different scale parameters. Calculate the filtering response of each orthogonal Gabor filter across all feature dimensions of the neighborhood feature map, and take the maximum value of the response as the response value of the orthogonal Gabor filter; The spatial morphological features are obtained by concatenating the response values ​​of eight orthogonal Gabor filters into vectors.

[0032] Specifically, this step is used to identify deep spatial texture and structural information from the neighborhood feature map. A predefined set of eight orthogonal Gabor filters is used, each composed of four directions (0°, 45°, 90°, and 135°) paired with two different scale parameters. For each orthogonal Gabor filter, such as the one with a 45° direction and a scale of 2, the filter is applied to each of the seven feature dimensions of the neighborhood feature map. This means the filter will filter a 3×3 matrix composed of the area values ​​of nine regions to obtain a response value. Similarly, the other six dimensions (perimeter, aspect ratio, etc.) are also filtered independently to obtain six more response values. The largest response value among the seven obtained from the orthogonal Gabor filters is selected as the response of that filter. This process is repeated for all eight filters, resulting in eight response values. These eight response values ​​are then concatenated sequentially to form an eight-dimensional vector, representing the spatial morphological features of local region consistency.

[0033] S4, the first fusion feature, spatial morphology feature and temporal drift correction feature are vector-concatenated to generate a combined feature; the combined feature is input into a predetermined classification model to obtain the printing quality detection result of the target solder paste area.

[0034] The first fusion feature represents the stable geometric and positional information of the solder paste itself, which is fused across multiple scales; the temporal drift correction feature represents the comparison between the current solder paste and the historical qualified sample benchmark, reflecting its deviation during the production process; the spatial morphology feature represents the correlation between the current solder paste and other solder pastes in its neighborhood, reflecting whether there are regional defects.

[0035] In step S4, the predetermined classification model is a Support Vector Machine (SVM) model, and the kernel function of the SVM model is a radial basis function (RBF). Before quality judgment, a combined feature vector needs to be constructed. This combined feature vector is formed by sequentially concatenating the previously calculated 7-dimensional first fusion feature and the 8-dimensional spatial morphological feature, resulting in a 15-dimensional comprehensive feature vector. This 15-dimensional combined feature vector is then input into a pre-trained SVM classification model. The SVM classification model uses the RBF as its kernel function, enabling it to handle complex classification problems. Based on the input combined feature vector, the SVM classification model calculates the position of the solder paste region in the feature space and determines whether that position is within a region deemed acceptable or unacceptable. A classification result is output; for example, output label 1 indicates a printing defect in the target solder paste region, while output label 0 indicates acceptable printing quality.

[0036] This invention achieves more stable, adaptive, and accurate solder paste printing quality inspection by integrating a first fusion feature that is insensitive to noise, a timing drift correction feature that can compensate for process drift, and a spatial morphology feature that can identify clustered defects.

[0037] Specific embodiments of the automotive circuit board solder paste printing quality inspection system of the present invention: The automotive circuit board solder paste printing quality inspection system includes a processor and a memory. The memory stores a computer program. When the computer program is run by the processor, it implements the automotive circuit board solder paste printing quality inspection method in any of the above embodiments.

[0038] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0039] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0040] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for inspecting the printing quality of solder paste on automotive circuit boards, characterized in that, Includes the following steps: S1, acquire two-dimensional images of each solder paste area on the automotive circuit board to be inspected, and perform the following operations for each target solder paste area: S11, construct a Laplacian pyramid containing multiple levels based on the image of the target solder paste area, and extract the geometric and positional features of the target solder paste area as level features at each level of the Laplacian pyramid; S12, calculate the local phase consistency of each layer image as the confidence weight of the layer, and perform a weighted summation of the layer features of the multiple layers based on the confidence weight to obtain the first fusion feature; S2, extract the first fusion feature of a predetermined number of the most recent qualified samples with the same label as the target solder paste area from the historical database; Based on the first fusion features of the most recently qualified samples, calculate their mean vector and the inverse of their covariance matrix; perform a difference operation between the first fusion features of the current target solder paste region and the mean vector, and multiply the difference result by the inverse matrix to obtain the time drift correction features; S3, based on the geometric center coordinates of the target solder paste area in the circuit board design file, determine a predetermined number of spatial nearest neighbor solder paste areas of the target solder paste area, and construct a neighborhood feature map composed of the first fusion feature of the target solder paste area and the predetermined number of nearest neighbor solder paste areas; The neighborhood feature map is filtered using a set of orthogonal Gabor filters with preset parameters, and the maximum or mean value of all filtered responses is taken as the spatial morphological feature. S4, the first fusion feature, spatial morphology feature and temporal drift correction feature are vector-concatenated to generate a combined feature; the combined feature is input into a predetermined classification model to obtain the printing quality detection result of the target solder paste area.

2. The method for inspecting the solder paste printing quality of automotive circuit boards according to claim 1, characterized in that, The method for constructing the Laplace pyramid includes: Gaussian smoothing and interlaced downsampling are applied to the image of the target solder paste region to construct a Gaussian pyramid. The Laplacian pyramid is generated by subtracting each layer of the Gaussian pyramid from the upsampled and blurred image of the next layer.

3. The method for inspecting the printing quality of solder paste on automotive circuit boards according to claim 2, characterized in that, The Laplace pyramid has 4 levels. The geometric and positional features include 7 parameters: area, perimeter, aspect ratio, compactness, offset, X-coordinate of the centroid, and Y-coordinate of the centroid. These 7 parameters are combined to form the feature vector of the corresponding level.

4. The method for inspecting the printing quality of solder paste on automotive circuit boards according to claim 3, characterized in that, The calculation method for the first fusion feature is as follows: For all levels of the Laplace pyramid, multiply the feature vector of each level by the confidence weight of that level. The first fusion feature is calculated by summing the product results of all levels and then dividing by the sum of the confidence weights of all levels.

5. The method for inspecting the printing quality of solder paste on automotive circuit boards according to claim 3, characterized in that, The method for extracting the geometric and positional features of the target solder paste region at each level of the Laplace pyramid is as follows: For each level of the Laplacian pyramid image, binarization is performed using the Otsu method to segment the solder paste region, and the parameters of the seven dimensions are calculated based on the segmented solder paste region.

6. The method for inspecting the solder paste printing quality of automotive circuit boards according to claim 1, characterized in that, The method for calculating the local phase consistency of each layer image is as follows: apply a set of preset multi-scale multi-directional logarithmic Gabor filters to each layer image, calculate the sum of the energy of the filter response, and obtain the local phase consistency value of the layer image.

7. The method for inspecting the printing quality of solder paste on automotive circuit boards according to claim 1, characterized in that, The method for constructing the neighborhood feature map includes: The number of solder paste regions that are closest to the geometric center of the target solder paste region in terms of Euclidean distance are designated as the nearest neighbor solder paste regions. A neighborhood feature map is constructed, and the first fused feature vector of the target solder paste region is placed at the center of the neighborhood feature map. The first fused feature vectors of the predetermined number of nearest neighbor solder paste regions are then filled into the remaining positions of the neighborhood feature map according to their orientation.

8. The method for inspecting the printing quality of solder paste on automotive circuit boards according to claim 7, characterized in that, The method for obtaining the spatial morphological features includes: The neighborhood feature map is filtered using eight orthogonal Gabor filters. The orthogonal Gabor filters have four directional parameters and two different scale parameters. Calculate the filtering response of each orthogonal Gabor filter across all feature dimensions of the neighborhood feature map, and take the maximum value of the response as the response value of the orthogonal Gabor filter; The spatial morphological features are obtained by concatenating the response values ​​of eight orthogonal Gabor filters into vectors.

9. The method for inspecting the printing quality of solder paste on automotive circuit boards according to claim 1, characterized in that, The predetermined classification model is a support vector machine model, and the kernel function of the support vector machine model is a radial basis function.

10. A quality inspection system for solder paste printing on automotive circuit boards, characterized in that, It includes a processor and a memory, the memory storing a computer program, which, when run by the processor, implements the method for detecting the solder paste printing quality of automotive circuit boards as described in any one of claims 1-9.

Images