Quality analysis method, system and storage medium for circuit board

Abstract

This application relates to the field of image processing technology and discloses a method, system, and storage medium for quality analysis of circuit boards. The method includes: extracting feature information from a dataset to generate a defect distribution map; extracting spectral reflectance data based on the map, calculating material uniformity, and obtaining a soldering quality data table; collecting quality feature parameters, establishing a quality grade index, and generating a graded dataset; extracting process data, performing cross-validation and deviation analysis, and forming a process adjustment table; merging the dataset and adjustment table, combining historical comparisons, and outputting a quality assessment report. This application achieves comprehensive detection of surface and internal defects on circuit boards by combining multispectral imaging technology with deep learning algorithms; and realizes automatic diagnosis of quality problems and optimization adjustment of process parameters by establishing a data-driven analysis model.

Description

Technical Field

[0001] This application relates to the field of image processing, and more particularly to a quality analysis method, system, and storage medium for circuit boards. Background Technology

[0002] Existing circuit board quality analysis technologies primarily employ optical inspection, electrical testing, and X-ray inspection. Optical inspection uses AOI (Automated Optical Inspection) equipment to acquire images of the circuit board surface and compares them with standard images to identify defects. Electrical testing uses flying probe testing or ICT (In-Circuit Testing) equipment to test conductivity and insulation. X-ray inspection is used to check for hidden defects such as BGA soldering and inner layer interconnects. In terms of data processing, traditional image processing algorithms are used for defect identification, statistical analysis methods are used to evaluate quality parameters, and process parameters are adjusted and optimized based on expert experience. The quality assessment process establishes a quality control system based on sampling inspection, comprehensively evaluating the product's appearance, performance, and reliability.

[0003] However, existing circuit board quality analysis technologies have the following shortcomings: First, various detection methods operate independently, lacking an effective data fusion mechanism and failing to achieve collaborative analysis of detection data; second, traditional image processing algorithms primarily focus on defect detection, lacking in-depth analysis of the causes of defects, making it difficult to provide effective guidance for process improvement; third, the quality analysis process relies excessively on human experience, lacking intelligent data analysis methods, affecting analysis efficiency and accuracy; finally, there is a lack of clear correlation between quality data and process parameters, making it impossible to achieve precise optimization of process parameters and continuous quality improvement. These problems lead to unsatisfactory quality control during circuit board manufacturing, affecting product yield and reliability. Summary of the Invention

[0004] This application provides a quality analysis method, system, and storage medium for circuit boards, which enables comprehensive detection of surface and internal defects of circuit boards through the combination of multispectral imaging technology and deep learning algorithms; and achieves automatic diagnosis of quality problems and optimization adjustment of process parameters by establishing a data-driven analysis model.

[0005] Firstly, this application provides a quality analysis method for circuit boards, comprising: acquiring original image data of the circuit board using multispectral imaging; performing signal denoising and image registration; fusing band data to form a spatial feature database; and then obtaining a standardized multidimensional image dataset through data standardization; extracting surface feature information of the circuit board based on the spectral band data in the standardized multidimensional image dataset; comparing and analyzing the feature point data with preset thresholds; generating a defect distribution feature map through multidimensional defect feature marking; extracting spectral reflectance data points on the surface of the circuit board based on the defect distribution feature map; calculating material uniformity by combining the light intensity attenuation curves of each band; obtaining a welding quality data table; collecting quality feature parameters based on the defect distribution feature map and the welding quality data table; establishing a circuit board quality grade index by dividing the data into graded threshold intervals; generating a quality graded dataset; extracting key process data points based on the quality graded dataset; performing cross-validation and deviation analysis on process parameters to form a process adjustment parameter table; merging the quality graded dataset and the process adjustment parameter table to construct a quality analysis data matrix; and outputting a circuit board quality assessment report by comparing historical parameters.

[0006] Secondly, this application provides a quality analysis system for circuit boards, the quality analysis system for circuit boards comprising:

[0007] The registration module is used to acquire the original image data of the circuit board using multispectral imaging. After signal denoising and image registration, the spatial feature database is formed by fusion of band data. Then, through data standardization processing, a standardized multidimensional image dataset is obtained.

[0008] The marking module is used to extract surface feature information of the circuit board based on the spectral band data in the standardized multidimensional image dataset, compare and analyze the feature point data with a preset threshold, and generate a defect distribution feature map through multidimensional defect feature marking.

[0009] The calculation module is used to extract spectral reflectance data points on the surface of the circuit board based on the defect distribution feature map, and combine them with the light intensity attenuation curves of each band to calculate the material uniformity and obtain the welding quality data table.

[0010] A module is established to collect quality characteristic parameters based on the defect distribution feature map and the welding quality data table, establish a circuit board quality grade index by dividing the graded threshold interval, and generate a quality graded dataset.

[0011] The analysis module is used to extract key process data points based on the quality grading dataset, perform cross-validation and deviation analysis on process parameters, and form a process adjustment parameter table.

[0012] The merging module is used to merge the quality grading dataset and the process adjustment parameter table to construct a quality analysis data matrix, and output a circuit board quality assessment report by combining historical parameter comparisons.

[0013] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the aforementioned quality analysis method for circuit boards.

[0014] The technical solution provided in this application utilizes multispectral imaging to acquire raw image data of the circuit board, followed by signal denoising and image registration processing to achieve high-precision acquisition of circuit board surface features, improving the accuracy and reliability of data acquisition. It employs band data fusion to form a spatial feature database, effectively integrating multi-band information and enhancing the comprehensiveness of feature extraction. Standardized multidimensional image datasets are obtained through data standardization, providing a standardized data foundation for subsequent analysis. Accurate defect identification is achieved by extracting circuit board surface feature information and comparing it with preset thresholds. A defect distribution feature map is generated using multidimensional defect feature markers, providing intuitive defect distribution information. Material uniformity calculations are performed using light intensity attenuation curves, enabling quantitative assessment of material quality. A scientific quality grading method is provided by establishing a circuit board quality grade index. A process adjustment parameter table is formed using cross-validation and deviation analysis, providing a reliable basis for process optimization. Finally, a comprehensive circuit board quality assessment report is output through comparison of the quality analysis data matrix and historical parameters, achieving systematization and standardization of the quality assessment process. This method organically combines multiple analytical techniques, effectively improving the quality control level and process optimization effect of the circuit board manufacturing process. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of one embodiment of the quality analysis method for circuit boards in this application.

[0017] Figure 2 This is a schematic diagram of one embodiment of the quality analysis system for circuit boards in this application. Detailed Implementation

[0018] This application provides a quality analysis method, system, and storage medium for circuit boards. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings 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 described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the quality analysis method for circuit boards in this application includes:

[0020] Step S101: Obtain the original image data of the circuit board using multispectral imaging. After signal denoising and image registration, a spatial feature database is formed by fusion of band data. Then, a standardized multidimensional image dataset is obtained through data standardization processing.

[0021] Step S102: Based on the spectral band data in the standardized multidimensional image dataset, extract the surface feature information of the circuit board, compare and analyze the feature point data with the preset threshold, and generate a defect distribution feature map through multidimensional defect feature marking.

[0022] Step S103: Based on the defect distribution feature map, extract the spectral reflectance data points on the surface of the circuit board, combine them with the light intensity attenuation curves of each band, calculate the material uniformity, and obtain the welding quality data table.

[0023] Step S104: Based on the defect distribution feature map and welding quality data table, collect quality feature parameters, establish a circuit board quality grade index by dividing the graded threshold interval, and generate a quality graded dataset.

[0024] Step S105: Based on the quality grading dataset, extract key process data points, perform cross-validation and deviation analysis on process parameters, and form a process adjustment parameter table.

[0025] Step S106: Merge the quality grading dataset and the process adjustment parameter table to construct a quality analysis data matrix. Combine this with historical parameter comparisons to output a circuit board quality assessment report.

[0026] It is understood that the executing entity of this application can be a quality analysis system for circuit boards, a terminal, or a server; no specific limitation is made here. This application's embodiment uses a server as an example for illustration.

[0027] Specifically, a multispectral camera is used to acquire circuit board images. This camera includes three acquisition units: ultraviolet (200-400nm), visible (400-700nm), and near-infrared (700-1000nm). An adaptive exposure control algorithm is employed during acquisition, adjusting exposure parameters in real-time based on the image grayscale histogram to ensure image quality. The spectral splitting system uses a dichroic mirror group to separate the incident light into different bands, with each band converting the optical signal into an electrical signal via an independent photoelectric converter. The image signal is processed using a wavelet transform denoising algorithm, decomposing the signal into wavelet coefficients at multiple scales. The coefficients are then thresholded, and finally, the image is reconstructed through inverse transform. Image registration employs a deep learning-based method, using an improved U-Net network to extract feature points. Feature matching is used to calculate the image transformation matrix, achieving precise alignment of images across different bands. Circuit board surface features are extracted based on a standardized multidimensional image dataset. An improved SLIC algorithm is used for image segmentation, dividing the image into multiple superpixel regions. Spectral band data is extracted for each region, and statistical characteristics (mean, variance, skewness, kurtosis) are calculated. Feature point data is compared with preset thresholds using an adaptive thresholding method, with the thresholds automatically calculated based on the image grayscale histogram using the OTSU algorithm. Defect feature labeling employs deep learning methods, using an improved ResNet network to extract deep features and combining an attention mechanism to enhance the perception of key features. For different types of defects (cold solder joints, short circuits, open circuits, etc.), a feature template library is established, and feature matching is performed using an improved Siamese network to generate a defect distribution feature map.

[0028] The spectral reflectance of the circuit board surface was measured using an integrating sphere spectrometer. A xenon lamp was used as the incident light source, with a wavelength range of 200-1000 nm and a spectral resolution of 1 nm. The spectral reflectance curve was recorded for each measurement point, and the reflectance variation with wavelength was analyzed. Light intensity attenuation analysis was performed based on the Lambert-Beer law, determining the material's optical properties by analyzing the attenuation relationship of reflected light intensity with propagation distance. Material uniformity analysis employed a spectral angle mapping algorithm, treating the spectral curve of each pixel as a high-dimensional vector and calculating the angle between the spectral vectors of adjacent pixels to evaluate the material surface uniformity. The weld quality data table recorded the spectral characteristics of the welded area, including parameters such as spectral consistency and edge sharpness. A quality feature vector was established by combining the defect distribution feature map and the weld quality data. The feature vector included parameters such as defect density, defect type distribution, and material uniformity index. An improved K-means algorithm was used to cluster the feature vectors, determining the boundaries of different quality levels. A quality level index was established based on the clustering results, classifying the products into multiple levels.

[0029] Key process parameters are extracted from the quality grading dataset. Principal component analysis (PCA) is used for dimensionality reduction to screen major influencing factors. K-fold cross-validation is employed to assess the significance of process parameters' impact on quality. Deviation analysis calculates the deviation of process parameters, identifying parameters requiring adjustment and their direction. A correlation analysis is performed between the quality grading data and process adjustment parameters to establish a quality-process relationship model. Tensor decomposition is used to extract potential correlation patterns between quality characteristics and process parameters. Time series analysis is used to assess quality change trends, predict potential quality risks, and generate a circuit board quality assessment report.

[0030] For example, in the quality analysis of a multilayer printed circuit board: First, image data is acquired using a multispectral camera. After wavelet denoising and image registration, clear multi-band images are obtained. Then, based on the SLIC algorithm, the image is segmented into appropriately sized superpixel regions, and the spectral features of each region are extracted. These features are analyzed using a deep learning network to identify and classify defective regions. For the identified defective regions, detailed spectral reflectance measurements are performed using an integrating sphere spectrometer to analyze material uniformity and soldering quality. Based on the analysis results, the circuit board quality is graded, and adjustment schemes for process parameters are determined through cross-validation and deviation analysis. The quality assessment report includes the analysis results and improvement recommendations.

[0031] In this embodiment, by utilizing multispectral imaging to acquire raw image data of the circuit board, and performing signal denoising and image registration processing, high-precision acquisition of the circuit board surface features is achieved, improving the accuracy and reliability of data acquisition. A spatial feature database is formed by band data fusion, effectively integrating multi-band information and enhancing the comprehensiveness of feature extraction. Standardized multidimensional image datasets are obtained through data standardization, providing a standardized data foundation for subsequent analysis. Accurate defect identification is achieved by extracting circuit board surface feature information and comparing it with preset thresholds. A defect distribution feature map is generated using multidimensional defect feature markers, providing intuitive defect distribution information. Material uniformity calculation is performed using light intensity attenuation curves, enabling quantitative assessment of material quality. A scientific quality grading method is provided by establishing a circuit board quality grade index. A process adjustment parameter table is formed using cross-validation and deviation analysis, providing a reliable basis for process optimization. Finally, a comprehensive circuit board quality assessment report is output by comparing the quality analysis data matrix with historical parameters, achieving systematization and standardization of the quality assessment process. This method organically combines multiple analytical techniques, effectively improving the quality control level and process optimization effect of the circuit board manufacturing process.

[0032] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0033] (1) Simultaneously acquire circuit board images in the ultraviolet, visible and near-infrared bands to form raw image data;

[0034] (2) Using the original image data to identify data interference items, the ambient light fluctuation data and sensor noise data are separated to generate a noise feature spectrum;

[0035] (3) Based on the noise feature spectrum, the interference signal in the original image data is eliminated in a targeted manner, and the image signal is enhanced to obtain the denoised image data;

[0036] (4) Divide the denoised image data into multiple sub-bands according to the wavelength range, and perform mapping transformation on the position coordinates of each sub-band image to obtain position mapping data;

[0037] (5) Convert the location mapping data into a unified coordinate system, and superimpose and fuse image data of different bands according to the spatial distribution pattern to construct a spatial feature database;

[0038] (6) Normalize the data samples in the spatial feature database, calculate the statistical feature parameters, remove abnormal data points, and output a standardized multidimensional image dataset.

[0039] Specifically, raw image data of the circuit board is acquired using a multispectral imaging acquisition device. This device employs three independent acquisition units: an ultraviolet band acquisition unit (wavelength range 200-400nm), a visible light band acquisition unit (wavelength range 400-700nm), and a near-infrared band acquisition unit (wavelength range 700-1000nm). Each acquisition unit includes a spectral splitting system, a photoelectric converter, and an image sensor. The spectral splitting system uses a dichroic mirror group to separate the incident light into different bands; the photoelectric converter converts the optical signal into an electrical signal; and the image sensor uses a back-illuminated CMOS structure with a pixel size of 2.5 micrometers and a resolution of 4096×3072 pixels. High-precision clock synchronization technology is used during the acquisition process to ensure that the three acquisition units are synchronously triggered for sampling. For the acquired raw image data, frequency domain analysis is used to identify data interference. First, the image signal is converted to the frequency domain using a two-dimensional Fourier transform. Ambient light fluctuations mainly manifest as low-frequency interference (0-10Hz), while sensor noise manifests as high-frequency random fluctuations (above 1kHz). A bandpass filter was designed to separate ambient light fluctuation signals from sensor noise signals. The bandwidth parameter of the bandpass filter was dynamically adjusted based on actual test data to achieve optimal separation performance. The separated noise characteristic spectrum recorded the spectral distribution, amplitude statistics, and correlation characteristics of different types of noise.

[0040] Based on the obtained noise feature spectrum, an adaptive Wiener filtering algorithm is used to selectively eliminate interference signals in the original image data. This algorithm dynamically adjusts filter parameters according to the signal-to-noise ratio (SNR) estimation of local image regions, achieving adaptive noise reduction for different areas. The denoised image is then enhanced using methods such as histogram equalization and contrast enhancement to improve the SNR and detail representation. Each processing step is monitored by quality evaluation metrics to ensure that the processing results meet the requirements of subsequent analysis.

[0041] The denoised image data was subdivided according to its spectral response characteristics. The ultraviolet band was divided into 4 sub-bands (200-250nm, 250-300nm, 300-350nm, 350-400nm), the visible light band into 6 sub-bands (each 100nm), and the near-infrared band into 3 sub-bands (700-800nm, 800-900nm, 900-1000nm). Affine transformation was applied to each sub-band image for geometric correction to eliminate distortions caused by the optical system. A feature point matching algorithm was used during the transformation process to establish the spatial correspondence between different sub-bands.

[0042] Spatial transformation parameters are optimized using the least squares method to convert the positional mapping data of each band to a unified standard coordinate system. Image fusion employs a multi-resolution analysis method, including three steps: pyramid decomposition, coefficient fusion, and image reconstruction. During the fusion process, weight coefficients are dynamically adjusted based on the information content of different bands. The fused data is stored in a spatial feature database, recording information such as spatial coordinates, spectral response values, and feature descriptions. The samples in the spatial feature database are standardized. Statistical parameters (mean, standard deviation, skewness, kurtosis) for each feature dimension are calculated, and then the Z-score standardization method is used to transform the data to a distribution with a mean of 0 and a variance of 1. Outlier identification uses the Local Outlier Factor (LOF) algorithm, which identifies outliers by comparing the local density of data points. After these processing steps, a standardized multidimensional image dataset is finally obtained.

[0043] For example, image data of the PCB board was acquired using a multispectral imaging device, scanning simultaneously in three bands. The raw data revealed low-frequency fluctuations caused by changes in ambient light and high-frequency interference from sensor thermal noise. Frequency domain analysis confirmed that ambient light fluctuations were concentrated in the 0-10Hz band, while sensor noise was distributed above 1kHz. Based on these characteristics, corresponding filters were designed for signal separation, yielding the noise characteristic spectrum. Directional noise cancellation was performed using the noise characteristic spectrum, with an adaptive Wiener filter dynamically adjusting parameters according to the signal-to-noise ratio of local areas. The denoised image data was further subdivided according to a pre-defined band division scheme, with each sub-band undergoing geometric correction. Precise image registration was achieved through feature point matching, followed by data fusion in a unified coordinate system. The fusion process employed a 5-layer pyramid decomposition, using Gaussian and Laplacian operators to extract features at each layer, adjusting weight coefficients based on the information content of different bands. Finally, the fused data was standardized, and outliers were detected and removed using the LOF algorithm, with the outlier threshold set to three times the standard deviation of the local density deviation. The processed standardized multidimensional image dataset contains the complete spectral response characteristics of the PCB board surface.

[0044] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0045] (1) The standardized multidimensional image dataset is segmented according to the surface area of ​​the circuit board, and the spectral band data of each region is extracted to generate a regional spectral data matrix;

[0046] (2) Map the data points in the regional spectral data matrix to spatial coordinates, locate and mark the feature points on the circuit board surface, and output the feature point data;

[0047] (3) Perform density clustering calculation on the feature point data, identify the distribution pattern of the feature points, count the degree of clustering of the feature points, and form a feature distribution statistics table;

[0048] (4) Based on the feature distribution statistics table, the density values ​​of feature points are compared with the preset threshold, and the areas that exceed the threshold range are marked to obtain the abnormal area data table.

[0049] (5) Classify and statistically analyze the data points in the abnormal area data table according to the defect type, calculate the spatial distribution characteristics of each type of defect, and generate a defect type distribution map;

[0050] (6) Perform multi-dimensional feature extraction on the defect type distribution map, perform correlation analysis on parameters such as defect location, area, and type, and output defect distribution feature map.

[0051] Specifically, region segmentation is performed on a standardized multidimensional image dataset. The segmentation employs an improved SLIC (Simple Linear Iterative Clustering) superpixel segmentation algorithm, which iteratively optimizes the clustering of adjacent similar pixels into superpixels. The process includes: initializing uniformly distributed cluster centers; calculating the distance and color similarity between each pixel and its surrounding cluster centers; assigning pixels to the nearest cluster center based on the combined distance; updating the cluster center positions and color features; and repeating this process until convergence. Segmentation parameters include: a preset number of superpixels (determined based on the circuit board size), a compactness factor (controlling the weight ratio of spatial distance to color distance), and a maximum number of iterations. Spectral band data is extracted for each segmented region, including the spectral response values ​​of the three main bands (ultraviolet, visible, and near-infrared) and sub-bands, forming a regional spectral data matrix. The data points in the regional spectral data matrix are mapped to the actual spatial coordinate system through an affine transformation, taking into account factors such as camera distortion and viewing angle differences. Feature point detection uses the Harris corner detection algorithm, which identifies corner features by calculating the gray-level change matrix of local image regions. The core of the Harris algorithm is to calculate the gray-level changes within a window surrounding each pixel and determine whether it is a corner point through feature value analysis. Detected feature points are then classified and labeled, and the location coordinates, spectral response value, and local gradient information of each feature point are recorded.

[0052] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) density clustering calculations are performed on feature point data. The DBSCAN algorithm automatically discovers clusters based on density connectivity, without requiring pre-specified cluster numbers. The core parameters of the algorithm include: neighborhood radius (defining the neighborhood range of a point) and minimum number of points (the threshold for determining core points). By analyzing the spatial distribution patterns of feature points, statistics such as local density, cluster radius, and inter-point distance are calculated to evaluate the degree of clustering of feature points, forming a feature distribution statistics table. Based on the feature distribution statistics table, a dynamic threshold method is used to analyze the feature point density values. The threshold setting is based on historical data statistics and expert experience, taking into account the characteristic differences of different types of circuit boards. Regions exceeding the threshold range are marked, and the location, area, and degree of exceeding the limit of abnormal regions are recorded, generating an abnormal region data table. A multi-level threshold strategy is used for threshold judgment, with different levels of thresholds corresponding to different degrees of anomalies.

[0053] A deep learning-based classification method is used for abnormal regions, employing an improved ResNet network structure that adds an attention mechanism module between traditional convolutional layers. The network training data includes feature samples of various typical defects, such as cold solder joints, short circuits, open circuits, and solder pad detachment. Spatial distribution features, including distribution density, clustering, and directionality, are calculated for each type of defect. Finally, a defect type distribution map is generated, visually displaying the spatial distribution of various defects. Multi-dimensional feature extraction is performed on the defect type distribution map, including: defect location coordinates, area size, shape features, edge features, and texture features. Correlation analysis is used to study the relationships between different feature parameters to identify potential quality problems. The final output defect distribution feature map contains defect feature information.

[0054] Taking defect detection on a four-layer PCB board as an example: The board surface is segmented using SLIC (Segmentation-Limited Image Processing), with segmentation parameters set so that each superpixel region is approximately 1 square millimeter. This granularity ensures sufficient detail for analysis while avoiding excessive computation. Full-band spectral data is extracted from each segmented region and recorded in a spectral data matrix. Spatial mapping is performed on the matrix data, converting image coordinates to actual physical coordinates. Harris corner detection is used to extract feature points, with a response threshold set to ensure that detected feature points have significant corner characteristics. DBSCAN clustering analysis is performed on the feature points, with a neighborhood radius set to the actual physical distance of 0.5 millimeters and a minimum number of points set to 5. These parameter settings effectively identify potential defect clusters. Based on the clustering results, the feature point density of each region is calculated, generating a density distribution heatmap. The heatmap is compared with a preset density threshold to mark regions with abnormal density. A deep learning model is used to classify defects in these abnormal regions. The model's input includes the multispectral image data and local feature descriptors of the region. The classification results reveal the spatial distribution patterns of different types of defects. For example, welding defects tend to be locally concentrated, while substrate defects are more widely dispersed. Finally, multi-dimensional feature extraction and correlation analysis were performed on each type of defect to study the interrelationships between defect features. By analyzing the location distribution and area size of welding defects, it was found that the problem of cold solder joints in certain areas is related to the pad design.

[0055] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0056] (1) Perform spectral reflectance sampling on the location information in the defect distribution feature map, record the reflection intensity of each sampling point in different bands, and form a spectral reflectance data set;

[0057] (2) Arrange the spectral reflectance data groups in wavelength order, calculate the difference in reflectance intensity between adjacent bands, and generate a light intensity variation gradient table;

[0058] (3) Normalize the data in the light intensity change gradient table, plot the light intensity attenuation curve for each sampling point, record the key inflection point positions, and output the attenuation characteristic data.

[0059] (4) Perform regional statistics on the attenuation characteristic data, calculate the variance of the attenuation curve in the local area, and form a material uniformity distribution map;

[0060] (5) Based on the material uniformity distribution map, the welding point is located, the slope of the attenuation curve of the welding area is extracted, the welding quality parameters are statistically analyzed, and the welding point quality matrix is ​​established.

[0061] (6) Compare and analyze the parameters in the welding point quality matrix with the standard parameters, mark abnormal welding points, and output the welding quality data table.

[0062] Specifically, spectral reflectance sampling was performed at the locations marked in the defect distribution feature map. A high-precision integrating sphere spectrometer was used, with a wavelength range covering 200-1000 nm and a spectral resolution of 0.5 nm. A xenon lamp was used as the incident light source, guided to the measurement point via an optical fiber guide system. Spectral reflectance data was recorded at each sampling point, including reflectance intensity values ​​in the ultraviolet (200-400 nm), visible (400-700 nm), and near-infrared (700-1000 nm) bands. An autofocus system was used during sampling to ensure measurement accuracy, and the data acquisition time for each measurement point was 100 milliseconds to ensure signal stability. The acquired spectral reflectance data sets were sorted from shortest to longest wavelength to form a continuous spectral curve. The central difference method was used to calculate the rate of change of reflectance intensity between adjacent wavelength points, obtaining the light intensity gradient. A five-point difference format was used for gradient calculation to improve calculation accuracy. The light intensity gradient table records the reflectance variation characteristics of each sampling point in different wavelength ranges, reflecting the spectral response characteristics of the material. The data in the light intensity variation gradient table were normalized to their maximum and minimum values, mapping the data to the [0, 1] interval. A third-order spline interpolation method was used to fit the light intensity attenuation curve, and the key inflection point locations were identified by analyzing the second derivative of the curve. Inflection point features include location coordinates and curvature values, which are directly related to the optical properties of the material. The attenuation characteristic data for each sampling point contains spectral response information. Regional statistical analysis was performed on the attenuation characteristic data. A sliding window method was used to calculate the variance of the attenuation curve within a local region, with the window size dynamically adjusted according to the circuit board's feature dimensions. The variance calculation reflects the uniformity characteristics of the material surface, and the generated material uniformity distribution map visually displays the spatial distribution of surface quality.

[0063] For the problem of welding point location, the location formula based on spectral features is as follows:

[0064] ,

[0065] in: Indicates the index of the welding point location; is the weighting factor for the i-th spectral feature; This is a weighting factor for material uniformity; Let be the spectral characteristic value of the i-th region; The region-averaged spectral characteristics; For local regional variance; Here, M represents the gradient weight coefficients; M is the material homogeneity function. These are the gradients of uniformity in the x and y directions, respectively.

[0066] Based on the location results, the spectral characteristics of the welding area are extracted, and the changes in the slope of the attenuation curve are analyzed. A welding point quality matrix is ​​established, recording parameters such as welding strength, uniformity, and interface characteristics. By comparing with a standard parameter database, abnormal welding points are identified, and a welding quality data table is output.

[0067] Taking the analysis of an actual circuit board as an example: The circuit board surface is sampled using a grid, and spectral reflectance data is recorded at each sampling point. Spectral data shows that normal soldering areas exhibit a smooth reflectance curve in the 500-600nm wavelength range, while areas with poor soldering show significant discontinuities. Analysis of the light intensity gradient reveals that the gradient value of poor soldering points is significantly higher than that of normal soldering points within a specific wavelength range. After normalization, the light intensity attenuation curve characteristics of each sampling point are extracted. The attenuation curve of normal soldering points shows characteristic inflection points near 350nm and 750nm, and the inflection point positions are stable. Regional statistical analysis shows that the attenuation curve variance of high-quality soldering areas is small, indicating uniform material bonding. Solder point location is based on the material uniformity distribution map, and the location of the solder point is accurately identified by calculating the spectral feature score of the local area. Solder quality analysis shows that the slope of the attenuation curve of good soldering points remains stable within a specific wavelength range, with a smooth transition to adjacent areas. By comparing with a standard parameter library, a solder quality evaluation standard is established to achieve accurate identification and classification of abnormal soldering points.

[0068] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0069] (1) Extract the defect density parameter from the defect distribution feature map, divide the defect location distribution data into regions, and output the regional defect statistics table;

[0070] (2) Classify the welding point data in the welding quality data table according to the quality level, perform numerical statistics on the welding quality parameters, and generate a welding quality distribution map;

[0071] (3) Integrate the data from the regional defect statistics table and the welding quality distribution map, calculate the comprehensive quality score for each region, and form a quality characteristic parameter matrix;

[0072] (4) Perform numerical analysis on the quality characteristic parameter matrix, statistically analyze the parameter distribution pattern, calculate the key interval nodes, and obtain the quality threshold distribution table;

[0073] (5) Divide the quality threshold distribution table into grade intervals according to the parameter value range, assign quality grade labels to different intervals, and establish a circuit board quality grade index;

[0074] (6) Classify the quality feature parameter matrix based on the circuit board quality grade index, organize and statistically analyze the classification results, and output the quality grade dataset.

[0075] Specifically, a grid method is used to divide the circuit board surface into uniform rectangular regions, with the grid size determined based on the actual dimensions and feature dimensions of the circuit board. The number of defects per unit area and the area ratio are calculated for each grid region to generate a defect density distribution map. An adaptive grid subdivision strategy is employed for region division, using smaller grid sizes in defect-dense areas to improve analysis accuracy. A regional defect statistics table records information such as the type, quantity, distribution location, and area ratio of defects within each grid region. Welding point data in the welding quality data table is categorized and graded according to welding quality parameters (including welding strength, uniformity, and interface characteristics). The quality grade classification uses a multi-parameter comprehensive evaluation method, comprehensively considering the geometric characteristics, spectral characteristics, and material properties of the welding points. Numerical statistics employ a weighted scoring method, with the weight of different parameters determined based on their influence on welding quality. The welding quality distribution map visually displays the spatial distribution characteristics of welding points of different quality grades.

[0076] When integrating data from the regional defect statistics table and welding quality distribution map, a multi-dimensional evaluation system is adopted. The comprehensive quality score for each region is calculated, considering factors including defect density, defect type weight, welding quality grade, and regional importance coefficient. Each element in the quality characteristic parameter matrix contains multi-dimensional quality characteristic information, reflecting the overall quality status of the region. Numerical analysis of the quality characteristic parameter matrix employs various statistical methods, including mean, standard deviation, skewness, and kurtosis analysis of the parameters. Cluster analysis is used to determine the key interval nodes of the quality parameters, and an improved K-means algorithm is used to segment the parameter space. The quality threshold distribution table records the threshold distribution of each quality parameter.

[0077] The parameter space is divided into multiple grade intervals, taking into account the distribution characteristics of the parameters and actual quality requirements. Quality grade labels are assigned to different intervals, establishing a standardized circuit board quality grade index. The index system covers multiple grades from excellent to unqualified, with each grade having a clearly defined parameter range. A dynamic threshold strategy is used for grade division, adjusted according to product type and quality requirements. Based on the established quality grade index, the data in the quality feature parameter matrix is ​​classified. The classification results are systematically organized, recording the data distribution of each grade. The quality grading dataset contains complete information such as quality grade distribution, anomaly region identification, and improvement suggestions.

[0078] For example, a 10mm × 10mm grid was used to divide the circuit board surface, with a further subdivision of 5mm × 5mm grid used in areas with dense component density. Defect analysis within each grid revealed defect clustering in certain areas, primarily located at the circuit board edges and in areas with high-density component mounting. Based on the spectral characteristics and material uniformity parameters of the solder joints, they were classified into three levels: A, B, and C. Level A indicates excellent solder quality with stable characteristic parameters; Level B indicates acceptable solder quality with parameter fluctuations within acceptable limits; and Level C indicates potential solder quality issues. Spatial distribution analysis revealed a positive correlation between Level C solder joints and the defect density of the surrounding area.

[0079] Integrated analysis revealed that soldering quality was generally lower in areas with high defect density, indicating potential process issues. The comprehensive quality score calculation considered defect density, solder joint quality level distribution, and regional importance coefficients within the region; areas with lower scores often exhibited a combination of quality problems. Based on parameter distribution patterns, the quality characteristic space was divided into five grade intervals: Excellent, Good, Acceptable, Needs Improvement, and Unacceptable. Each grade interval has a clearly defined parameter range; for example, the Excellent grade requires defect density below a threshold and all solder joints meeting Grade A standards. This grading method established a quality assessment system. Based on the established quality grade index, the entire circuit board was graded for quality. The grading results not only displayed the overall quality level of the circuit board but also identified areas requiring focused attention.

[0080] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0081] (1) Extract key process data from the quality grading dataset, classify the data according to process flow nodes, and form a process data classification table;

[0082] (2) Perform correlation analysis on the data in the process data classification table, calculate the correlation coefficient between process parameters, and generate a parameter correlation matrix;

[0083] (3) Sort the data in the parameter correlation matrix according to the correlation coefficient, filter the highly correlated process parameter combinations, and output the key parameter group;

[0084] (4) Cross-compare and verify the key parameter groups, statistically analyze the parameter fluctuation range, calculate the parameter deviation value, and obtain the process parameter deviation table;

[0085] (5) Based on the process parameter deviation table, perform data analysis, calculate the adjustment direction and adjustment range of each parameter, and form a parameter adjustment plan;

[0086] (6) Verify the feasibility of the adjustment values ​​in the parameter adjustment scheme, integrate the verification results, and output the process adjustment parameter table.

[0087] Specifically, key process data are extracted from the quality grading dataset. Principal component analysis is used to identify process parameters that significantly impact quality. These parameters are categorized according to the manufacturing process, including: pretreatment parameters (cleanliness, surface treatment, etc.), board fabrication parameters (lamination temperature, pressure, etc.), drilling parameters (rotation speed, feed rate, etc.), electroplating parameters (current density, time, etc.), solder resist parameters (exposure energy, development time, etc.), and surface treatment parameters (tin immersion temperature, time, etc.). A process data classification table records the key process parameters and their value ranges for each stage.

[0088] The correlation between process parameters is calculated using the following formula:

[0089] ,

[0090] in: The correlation coefficient between parameters a and b; Let a and b be the values ​​of parameters a and b in the k-th sample; Let a be the mean of parameters a and b; These are the sample weight coefficients; This is the parameter variance adjustment factor; This refers to the process distance between parameters; is the scale parameter; n is the number of samples.

[0091] Correlation analysis employed a combination of Pearson correlation coefficient and Spearman rank correlation coefficient to assess the linear and nonlinear relationships between parameters. Each element in the parameter correlation matrix represents the strength of the correlation between two process parameters. By setting a correlation coefficient threshold, highly correlated parameter combinations were selected to form key parameter groups. These key parameter groups included information such as parameter name, correlation coefficient, and direction of influence. The k-fold cross-validation method was used to validate the key parameter groups. The variations of each parameter under different process conditions were recorded, and the fluctuation range and standard deviation of the parameters were calculated. By comparing with standard process parameters, the deviation of the actual parameters was calculated, generating a process parameter deviation table. Deviation analysis considered the historical trends of parameter changes and the degree of their impact on quality.

[0092] Based on the deviation table data, the direction and magnitude of adjustment required for each parameter are calculated. The parameter adjustment plan takes into account the coupling relationships between parameters, and an optimization algorithm is used to find the optimal adjustment combination. The feasibility of the adjustment plan is evaluated through historical data analysis and a process expert knowledge base.

[0093] Taking the optimization of process parameters for a batch of multilayer PCBs as an example: Key process parameters, including lamination temperature, pressure, drilling speed, and electroplating time, were extracted from quality grading data. These parameters were then categorized and organized according to the manufacturing process to form a process data classification table. Correlation analysis showed a significant correlation between lamination temperature and interlayer bonding strength, and a high correlation between drilling speed and the uniformity of copper thickness on the hole wall. Through parameter screening, several highly correlated parameter combinations were identified. Cross-validation results showed that the correlation of these parameter combinations was consistent across different batches of products. Parameter fluctuation analysis revealed deviations between the actual and standard values ​​of some process parameters, such as a faster heating rate in the lamination temperature affecting the resin curing effect. Based on these analysis results, specific adjustment plans were developed: optimizing the lamination temperature curve, adjusting drilling parameters, and improving the electroplating process. Feasibility verification confirmed that these adjustment plans met process requirements and were practically operable. The generated process adjustment parameter table provided clear guidance for parameter optimization on the production line. This systematic process parameter optimization method achieved precise control and quality improvement in the PCB manufacturing process.

[0094] In one specific embodiment, the process of performing step S106 may specifically include the following steps:

[0095] (1) Match the quality parameters in the quality grading dataset with the process parameters in the process adjustment parameter table to generate a parameter association table;

[0096] (2) Organize the data items in the parameter association table into a structured format, group the data according to quality grade and process type, and form a data grouping matrix;

[0097] (3) Perform statistical analysis on each group of data in the data grouping matrix, calculate the correspondence between quality parameters and process parameters, and output a comprehensive analysis data table;

[0098] (4) Compare the data in the comprehensive analysis data table with the historical quality data, mark the points of abnormal parameter changes, and generate a quality trend analysis chart;

[0099] (5) Based on the quality trend analysis chart, data mining is performed to extract the pattern of quality change, key influencing factors are statistically analyzed, and a quality assessment conclusion is formed;

[0100] (6) Integrate the quality assessment conclusions with specific parameter data, classify and organize them according to the assessment dimensions, and output the circuit board quality assessment report.

[0101] Specifically, parameter matching is performed. The quality grading dataset contains quality parameters such as pass rate and defect rate, while the process adjustment parameter table contains process parameters such as temperature, pressure, and time. A multidimensional mapping algorithm is used to establish the correspondence between parameters, considering the physical meaning and influence mechanism of the parameters during the mapping process. The parameter association table records the matching relationships, including information such as parameter type, numerical range, and influence weight. The parameter association table is structured, and a hierarchical clustering method is used to group the data. Quality grade grouping is based on product quality standards, dividing the data into four levels: excellent, good, average, and poor. Process type grouping is based on the manufacturing process flow, including material parameters (substrate properties, copper foil thickness, etc.), process parameters (pressing temperature, drilling speed, etc.), and environmental parameters (temperature, humidity, cleanliness, etc.). The rows of the data grouping matrix represent quality grades, and the columns represent process types.

[0102] Statistical analysis employed multiple methods to process the data grouping matrix, including analysis of variance (ANOVA), regression analysis, and principal component analysis (PCA). ANOVA assessed the contribution of different process parameters to quality fluctuations, regression analysis established quantitative relationships between parameters, and PCA reduced data dimensionality and extracted key features. A comprehensive analysis table recorded the mapping relationship between quality parameters and process parameters. Historical data comparison used time series analysis to identify parameter trends and outliers. The comparison process employed a sliding window technique to calculate the moving average and standard deviation of the parameters. Outlier identification used the 3σ criterion, marking significantly deviating data points and recording the time and specific parameters of the anomalies. A quality trend analysis graph displayed the trajectory of parameter changes over time. Data mining employed an association rule mining algorithm to discover patterns in quality changes. The algorithm included three steps: data preprocessing, frequent itemset generation, and rule evaluation. Effective rules were screened using support and confidence indices to identify key factors affecting quality. Based on the mining results, the evaluation conclusions systematically described the causes and impacts of quality problems.

[0103] The quality assessment conclusions are integrated with parameter data and categorized according to dimensions such as product performance, process stability, and quality reliability. The assessment report includes quality status analysis, problem diagnosis, and improvement suggestions, providing guidance for production optimization. For example, quality parameters (such as weld quality and circuit accuracy) are matched with process parameters (such as weld temperature and exposure energy). Analysis revealed a significant correlation between the weld quality and weld temperature control of a certain batch of products. The parameter correlation table shows that when the weld temperature deviation exceeds the set range, the weld defect rate increases significantly. Data grouping analysis indicates that the process parameter stability of the high-quality product group is significantly higher than that of other groups. Statistical analysis results show a negative correlation between the fluctuation range of process parameters and the product quality level. Comparison with historical data reveals a recent decrease in the accuracy of weld temperature control, which is consistent with the fluctuation trend of quality levels. Quality trend analysis identifies several key influencing factors: first, the stability of process parameters, especially the accuracy of temperature and pressure control; second, the consistency of material properties, including the quality of the substrate and solder; and finally, the influence of environmental conditions, such as fluctuations in humidity and cleanliness. The combined effect of these factors determines the final product quality. Data mining results revealed significant patterns in quality variation: small fluctuations in process parameters do not significantly affect product quality within a certain range, but quality drops sharply when fluctuations exceed critical values. Furthermore, certain process parameters interact, requiring synergistic optimization to achieve optimal results. Based on these analyses, the evaluation report proposes specific improvement recommendations: optimize the process parameter control system to improve the precision of temperature and pressure control; strengthen raw material inspection to ensure consistent material properties; and improve production environment control to reduce environmental interference. These recommendations are highly targeted and actionable, providing clear guidance for improving circuit board manufacturing quality.

[0104] In one specific embodiment, the process of integrating the quality assessment conclusions with specific parameter data may specifically include the following steps:

[0105] (1) Extract the key assessment points from the quality assessment conclusion, classify them according to the defect type, and generate an assessment focus statistical table;

[0106] (2) Quantitatively analyze the data items in the key evaluation statistics table, calculate the deviation value of each evaluation item, and form a deviation distribution matrix;

[0107] (3) Perform data clustering based on the deviation distribution matrix, classify risk levels according to the degree of deviation, and output a risk assessment list;

[0108] (4) Conduct a retrospective analysis of each indicator in the risk assessment list, extract the root causes of quality problems, and generate a problem attribution table;

[0109] (5) Conduct correlation analysis between the problem attribution table and process parameter data, statistically analyze the quality fluctuation pattern, and generate an improvement direction report;

[0110] (6) Systematically organize the improvement direction reports, sort them according to the severity of the problems, and output a circuit board quality assessment report.

[0111] Specifically, key evaluation points were extracted from the quality assessment conclusions. Text mining algorithms were used to identify important quality feature words and evaluation indicators in the conclusions, with feature word selection based on a professional terminology database for the circuit board manufacturing field. Defect type classification adopted a pre-defined classification standard, including: process defects (such as cold solder joints, short circuits, open circuits), material defects (such as delamination, blistering, copper peeling), and structural defects (such as line width deviation, aperture deviation). A key evaluation statistics table recorded the distribution characteristics, frequency of occurrence, and severity of various defects. Quantitative analysis was performed on the data in the key evaluation statistics table, converting qualitative descriptions into quantitative indicators based on a multi-dimensional evaluation system. Multiple evaluation dimensions were set for each evaluation item, including defect area, depth, location, and impact range. The deviation of each evaluation item from the standard value was calculated using the standard score method, forming a deviation distribution matrix. Rows in the matrix represent different evaluation items, columns represent different evaluation dimensions, and each element represents a specific deviation value. An improved K-means clustering algorithm was used to analyze the deviation distribution matrix. The clustering process considered the magnitude and distribution characteristics of the deviation values, grouping similar deviation patterns into one category. Clustering results are used to determine risk levels, classifying the degree of bias into three risk levels: high, medium, and low. The risk assessment checklist details the risk level, impact, and potential consequences of each assessment item.

[0112] The traceability analysis employs a causal chain analysis method to trace the mechanisms of quality problems layer by layer. By constructing a problem tree, the root causes of quality deviations are identified. The problem attribution table system records the problem type, scope of impact, causal analysis, and solutions. The attribution analysis focuses on the influence of process parameters, material properties, and environmental factors. Correlation analysis is performed between the problem attribution table and process parameter data, using association rule mining algorithms to discover patterns in quality fluctuations. The analysis includes trends in process parameter changes, interactions between parameters, and their degree of impact on quality. Based on the analysis results, the improvement direction report proposes specific optimization suggestions and improvement measures.

[0113] The improvement direction reports were systematically compiled, and a multi-criteria decision-making method was used to prioritize the issues. The prioritization criteria comprehensively considered the severity, frequency of occurrence, difficulty of resolution, and economic impact of the issues. The quality assessment report includes the analysis results and improvement recommendations.

[0114] For example, key issues are extracted from the evaluation conclusions, such as insufficient bonding strength between the inner copper foil and the substrate, and incomplete via filling. These issues are then categorized into material defects and process defects using classification tags. Each defect item is recorded with specific characteristic descriptions and statistical data. In the quantitative analysis phase, each defect item is evaluated from multiple dimensions. For instance, for the issue of insufficient bonding strength, the degree of deviation is assessed through peel strength testing; for the via filling issue, the deviation in filling rate is assessed through microscopic sectioning. These deviation data constitute a multi-dimensional evaluation matrix.

[0115] Data clustering analysis revealed that certain defects shared similar deviation characteristics; for example, issues with inner layer bonding strength and via filling were both related to lamination process parameters. These defects were marked as high-risk items through risk level classification. A risk assessment checklist detailed the risk characteristics and impact of various defects. Tracing analysis revealed the root causes of the problems. For instance, insufficient inner layer bonding strength stemmed from improper control of the lamination temperature profile. Analysis of historical data on lamination temperature, pressure, and time parameters revealed that an excessively rapid temperature rise rate led to incomplete resin curing, thus affecting bonding strength.

[0116] Correlation analysis revealed interactions among multiple process parameters. The relationship between pressing temperature and pressure directly affected bond strength, while venting time affected via filling quality. These findings provided clear directions for process improvement. Improvement recommendations included optimizing the pressing temperature profile, adjusting pressure parameters, and improving the venting process. The quality assessment report prioritized issues based on severity. Inner layer bond strength issues, affecting product reliability, were listed as the primary improvement targets, followed by via filling issues, which impacted yield. The report detailed the specific manifestations, causal analysis, and improvement measures for each issue.

[0117] The quality analysis method for circuit boards in the embodiments of this application has been described above. The quality analysis system for circuit boards in the embodiments of this application is described below. Please refer to [link / reference]. Figure 2 One embodiment of the circuit board quality analysis system in this application includes:

[0118] The registration module 201 is used to acquire the original image data of the circuit board using multispectral imaging. After signal denoising and image registration, the spatial feature database is formed by fusion of band data. Then, through data standardization processing, a standardized multidimensional image dataset is obtained.

[0119] The marking module 202 is used to extract surface feature information of the circuit board according to the spectral band data in the standardized multidimensional image dataset, compare and analyze the feature point data with a preset threshold, and generate a defect distribution feature map through multidimensional defect feature marking.

[0120] The calculation module 203 is used to extract spectral reflectance data points on the surface of the circuit board based on the defect distribution feature map, and to calculate the material uniformity by combining the light intensity attenuation curves of each band and obtaining a welding quality data table.

[0121] Module 204 is used to collect quality characteristic parameters based on the defect distribution feature map and the welding quality data table, establish a circuit board quality grade index by dividing the graded threshold interval, and generate a quality graded dataset.

[0122] Analysis module 205 is used to extract key process data points based on the quality grading dataset, perform cross-validation and deviation analysis on process parameters, and form a process adjustment parameter table.

[0123] The merging module 206 is used to merge the quality grading dataset and the process adjustment parameter table to construct a quality analysis data matrix, and output a circuit board quality assessment report by combining historical parameter comparisons.

[0124] Through the collaborative efforts of the aforementioned components, this method utilizes multispectral imaging to acquire raw image data of the circuit board. After signal denoising and image registration, high-precision acquisition of circuit board surface features is achieved, improving the accuracy and reliability of data acquisition. A spatial feature database is formed by fusion of band data, effectively integrating multi-band information and enhancing the comprehensiveness of feature extraction. Standardized multidimensional image datasets are obtained through data standardization, providing a standardized data foundation for subsequent analysis. Accurate defect identification is achieved by extracting circuit board surface feature information and comparing it with preset thresholds. A defect distribution feature map is generated using multidimensional defect feature markers, providing intuitive defect distribution information. Material uniformity calculations are performed using light intensity attenuation curves, enabling quantitative assessment of material quality. A scientific quality grading method is provided by establishing a circuit board quality grade index. A process adjustment parameter table is formed using cross-validation and deviation analysis, providing a reliable basis for process optimization. Finally, a comprehensive circuit board quality assessment report is output by comparing the quality analysis data matrix with historical parameters, achieving systematization and standardization of the quality assessment process. This method organically combines multiple analytical techniques, effectively improving the quality control level and process optimization effect of the circuit board manufacturing process.

[0125] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the quality analysis method for the circuit board.

[0126] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A quality analysis method for circuit boards, characterized in that, The quality analysis method for the circuit board includes: The raw image data of the circuit board is obtained by multispectral imaging. After signal denoising and image registration, the spatial feature database is formed by fusion of band data. Then, through data standardization processing, a standardized multidimensional image dataset is obtained. Based on the spectral band data in the standardized multidimensional image dataset, surface feature information of the circuit board is extracted, the feature point data is compared and analyzed with a preset threshold, and a defect distribution feature map is generated after multidimensional defect feature marking. Based on the defect distribution feature map, spectral reflectance data points on the circuit board surface are extracted. Combined with the light intensity attenuation curves for each wavelength band, material uniformity is calculated to obtain a welding quality data table. This includes: sampling spectral reflectance at location information based on the defect distribution feature map, recording the reflectance intensity of each sampling point in different wavelength bands to form a spectral reflectance data set; arranging the spectral reflectance data set in wavelength order, calculating the reflectance intensity difference between adjacent wavelength bands using the five-point difference method to generate a light intensity variation gradient table; and normalizing the data in the light intensity variation gradient table using third-order spline interpolation. The light intensity attenuation curve of each sampling point is fitted, and the key inflection point position is identified by the second derivative, and the attenuation feature data is output. The spectral angle mapping algorithm is used to treat the spectral curve of each pixel as a high-dimensional vector, calculate the angle between the spectral vectors of adjacent pixels, evaluate the uniformity of the material surface, and perform material uniformity calculation. Based on the material uniformity distribution map, the welding point is located, the slope of the attenuation curve of the welding area is extracted, the welding quality parameters are statistically analyzed, the welding point quality matrix is ​​established, the parameters in the welding point quality matrix are compared and analyzed with the standard parameters, abnormal welding points are marked, and the welding quality data table is output. Based on the defect distribution feature map and the welding quality data table, quality feature parameters are collected, and a circuit board quality grade index is established by dividing the graded threshold interval to generate a quality graded dataset. Based on the quality grading dataset, key process data points are extracted, and cross-validation and deviation analysis are performed on process parameters to form a process adjustment parameter table. This includes: extracting key process data from the quality grading dataset; classifying the data according to process flow nodes to form a process data classification table; performing correlation analysis on the data in the process data classification table to calculate the correlation coefficient between process parameters and generate a parameter correlation matrix; sorting the data in the parameter correlation matrix according to the correlation coefficient, selecting highly correlated process parameter combinations, and outputting key parameter groups; performing cross-validation on the key parameter groups, statistically analyzing parameter fluctuation ranges, calculating parameter deviation values, and obtaining a process parameter deviation table; performing data analysis based on the process parameter deviation table to calculate the adjustment direction and magnitude of each parameter, forming a parameter adjustment plan; verifying the feasibility of the adjustment values ​​in the parameter adjustment plan, integrating the verification results, and outputting the process adjustment parameter table. Key process data were extracted from the quality grading dataset. Principal component analysis was used to identify process parameters that significantly affect quality. The process parameters were classified according to the manufacturing process, including: pretreatment parameters, board making parameters, drilling parameters, electroplating parameters, solder resist parameters, and surface treatment parameters. The process data classification table records the key process parameters and their value ranges for each step. The correlation between process parameters is calculated using the following formula: ； in: The correlation coefficient between parameters a and b; , Let a and b be the values ​​of parameters a and b in the k-th sample; , These are the mean values ​​of parameters a and b, respectively. These are the sample weight coefficients; , This is the parameter variance adjustment factor; This refers to the process distance between parameters; is the scale parameter; n is the number of samples; Correlation analysis employs a combination of Pearson correlation coefficient and Spearman rank correlation coefficient to assess the linear and nonlinear relationships between parameters. Each element in the parameter correlation matrix represents the correlation strength between two process parameters. By setting a correlation coefficient threshold, highly correlated parameter combinations are selected to form key parameter groups. Each key parameter group includes the parameter name, correlation coefficient, and direction of influence. The k-fold cross-validation method is used to validate the key parameter groups. The changes in each parameter under different process conditions are recorded, and the fluctuation range and standard deviation of the parameters are calculated. By comparing with standard process parameters, the deviation of the actual parameters is calculated, and a process parameter deviation table is generated. Deviation analysis considers the historical trend of parameter changes and the degree of quality impact. The quality grading dataset and the process adjustment parameter table are merged to construct a quality analysis data matrix. By comparing historical parameters, a circuit board quality assessment report is output.

2. The quality analysis method for circuit boards according to claim 1, characterized in that, The process involves acquiring raw image data of the circuit board using multispectral imaging, performing signal denoising and image registration, fusing band data to form a spatial feature database, and then performing data standardization to obtain a standardized multidimensional image dataset, including: Circuit board images in the ultraviolet, visible, and near-infrared bands are acquired simultaneously to form raw image data; The original image data is used to identify data interference items, and the ambient light fluctuation data and sensor noise data are separated to generate a noise feature spectrum; Based on the noise feature spectrum, the interference signals in the original image data are targeted to be eliminated, and the image signal is enhanced to obtain denoised image data; The denoised image data is divided into multiple sub-bands according to the wavelength range, and the position coordinates of each sub-band image are mapped and transformed to obtain position mapping data. The location mapping data is converted into a unified coordinate system, and image data from different bands are superimposed and fused according to spatial distribution rules to construct the spatial feature database; The data samples in the spatial feature database are normalized, statistical feature parameters are calculated, outlier data points are removed, and the standardized multidimensional image dataset is output.

3. The quality analysis method for circuit boards according to claim 1, characterized in that, The step of extracting surface feature information of the circuit board based on the spectral band data in the standardized multidimensional image dataset, comparing and analyzing the feature point data with a preset threshold, and generating a defect distribution feature map through multidimensional defect feature labeling includes: The standardized multidimensional image dataset is segmented according to the surface region of the circuit board, and the spectral band data of each region is extracted to generate a regional spectral data matrix. The data points in the regional spectral data matrix are spatially mapped to the feature points on the circuit board surface, and the feature point data is output. Density clustering calculations are performed on the feature point data to identify the distribution patterns of the feature points, the degree of clustering of the feature points is statistically analyzed, and a feature distribution statistics table is formed. Based on the feature distribution statistics table, the density values ​​of feature points are compared with preset thresholds to mark areas that exceed the threshold range, thus obtaining an abnormal area data table. The data points in the abnormal area data table are classified and statistically analyzed according to defect type, and the spatial distribution characteristics of each type of defect are calculated to generate a defect type distribution map. Multi-dimensional feature extraction is performed on the defect type distribution map, and parameters such as defect location, area, and type are correlated and analyzed to output the defect distribution feature map.

4. The quality analysis method for circuit boards according to claim 1, characterized in that, Based on the defect distribution feature map and the welding quality data table, quality feature parameters are collected, and a circuit board quality grade index is established by dividing the data into graded threshold intervals, generating a quality graded dataset, including: Extract the defect density parameter from the defect distribution feature map, divide the defect location distribution data into regions, and output a regional defect statistics table. The welding point data in the welding quality data table are categorized according to quality level, the welding quality parameters are statistically analyzed, and a welding quality distribution map is generated. By integrating the data from the regional defect statistics table and the welding quality distribution map, the comprehensive quality score of each region is calculated to form a quality characteristic parameter matrix; Numerical analysis is performed on the quality characteristic parameter matrix to statistically analyze the parameter distribution pattern, calculate the key interval nodes, and obtain the quality threshold distribution table. The quality threshold distribution table is divided into grade intervals according to the parameter value range, and different intervals are assigned quality grade labels to establish the circuit board quality grade index. The quality feature parameter matrix is ​​classified based on the circuit board quality grade index, the classification results are organized and statistically analyzed, and the quality grade dataset is output.

5. The quality analysis method for circuit boards according to claim 1, characterized in that, The process involves merging the quality grading dataset and the process adjustment parameter table to construct a quality analysis data matrix. By comparing historical parameters, a circuit board quality assessment report is output, including: The quality parameters in the quality grading dataset are matched with the process parameters in the process adjustment parameter table to generate a parameter association table; The data items in the parameter association table are structured and grouped according to quality level and process type to form a data grouping matrix; Perform statistical analysis on each group of data in the data grouping matrix, calculate the correspondence between quality parameters and process parameters, and output a comprehensive analysis data table; The data in the comprehensive analysis data table is compared with historical quality data, abnormal parameter changes are marked, and a quality trend analysis chart is generated. Based on the aforementioned quality trend analysis chart, data mining is performed to extract patterns of quality changes, identify key influencing factors, and form quality assessment conclusions. The quality assessment conclusions and specific parameter data are integrated, categorized and organized according to the assessment dimensions, and the circuit board quality assessment report is output.

6. The quality analysis method for circuit boards according to claim 5, characterized in that, The process involves integrating the quality assessment conclusions with specific parameter data, categorizing and organizing them according to assessment dimensions, and outputting the circuit board quality assessment report, including: The key assessment points in the quality assessment conclusions are extracted, categorized according to defect type, and an assessment focus statistical table is generated. The data items in the evaluation focus statistics table are quantitatively analyzed, the deviation value of each evaluation item is calculated, and a deviation distribution matrix is ​​formed; Based on the aforementioned deviation distribution matrix, data clustering is performed, risk levels are classified according to the degree of deviation, and a risk assessment list is output. A retrospective analysis was conducted on each indicator in the risk assessment list to extract the root causes of quality problems and generate a problem attribution table. The problem attribution table is correlated with the process parameter data to analyze the quality fluctuation pattern and generate an improvement direction report. The improvement direction reports are systematically organized, sorted according to the severity of the problems, and the circuit board quality assessment report is output.

7. A quality analysis system for circuit boards, used to implement the quality analysis method for circuit boards as described in any one of claims 1 to 6, characterized in that, The quality analysis system for the circuit board includes: The registration module is used to acquire the original image data of the circuit board using multispectral imaging. After signal denoising and image registration, the spatial feature database is formed by fusion of band data. Then, through data standardization processing, a standardized multidimensional image dataset is obtained. The marking module is used to extract surface feature information of the circuit board based on the spectral band data in the standardized multidimensional image dataset, compare and analyze the feature point data with a preset threshold, and generate a defect distribution feature map through multidimensional defect feature marking. The calculation module is used to extract spectral reflectance data points on the surface of the circuit board based on the defect distribution feature map, and combine them with the light intensity attenuation curves of each band to calculate the material uniformity and obtain the welding quality data table. A module is established to collect quality characteristic parameters based on the defect distribution feature map and the welding quality data table, establish a circuit board quality grade index by dividing the graded threshold interval, and generate a quality graded dataset. The analysis module is used to extract key process data points based on the quality grading dataset, perform cross-validation and deviation analysis on process parameters, and form a process adjustment parameter table. The merging module is used to merge the quality grading dataset and the process adjustment parameter table to construct a quality analysis data matrix, and output a circuit board quality assessment report by combining historical parameter comparisons.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the quality analysis method for circuit boards as described in any one of claims 1 to 6.

Images