Intelligent defect detection system and method applied to PCBA (Printed Circuit Board Assembly)

Abstract

The invention discloses an intelligent defect detection system and method applied to a PCBA (Printed Circuit Board Assembly), and relates to the technical field of big data analysis. By dividing mutually independent sub-detection areas, feature signal interference among different bonding pad areas is eliminated, and a detection range is accurately anchored; the first weight value is determined by counting the proportion of each characteristic exceeding threshold value of the abnormal bonding pad data, the second weight value is calculated by combining with the local time sequence sample data intercepted by the sliding window, and the dynamic weight value is obtained, so that the global historical data universality and the local time sequence data timeliness are balanced; the characteristic reference drift caused by technological parameter fluctuation and consumable batch difference is inhibited; an analysis threshold value is generated by calculating a feature floating value, a fine system of defect judgment and approaching abnormity early warning is constructed, the limitation of single-dimension judgment of a traditional fixed threshold value is broken through, samples which do not meet defect standards but deviate from a normal state are recognized, the problems of detection misjudgment and missed judgment are solved, and the precise quality control requirement of the dynamic production working condition of a production line is met.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to an intelligent defect detection system and method for PCBA boards. Background Technology

[0002] The soldering quality of PCBA directly determines the operational reliability and lifespan of electronic devices. Hidden defects such as cold solder joints and bridging that occur during the production process cannot be identified by the naked eye or conventional optical methods. X-ray imaging technology can penetrate the surface of the board and clearly show the internal structure of the pads and the distribution of solder, enabling accurate detection of hidden defects and preventing defective products from flowing into subsequent processes. It is a key link in the quality control of PCBA production lines.

[0003] In actual PCBA board reflow soldering production lines, slight fluctuations in reflow oven temperature and changes in solder paste batches are difficult to completely avoid. These variables directly cause the local temporal performance of features such as the proportion of low grayscale areas, grayscale dispersion, and edge sharpness in X-ray imaging of solder pads to deviate from the global historical benchmark. Traditional defect detection uses a fixed threshold judgment mode, setting judgment criteria based on only a single-dimensional global historical benchmark. This cannot take into account both the universality of global historical data and the timeliness of local temporal data, and is prone to problems such as false judgment of cold solder joints and missed judgment of bridging due to feature benchmark drift. It is difficult to adapt to the precise quality control requirements of dynamic production line conditions. Therefore, there is an urgent need for an intelligent defect detection system and method for PCBA boards. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent defect detection system and method for PCBA boards to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a defect intelligent detection method applied to PCBA boards, the defect intelligent detection method comprising the following steps:

[0006] Step S1: For PCBA boards of the same model, read the preset pad sample data of PCBA boards of the same model that have been soldered in the past and divide the area, filter out the largest effective area that can completely contain the corresponding preset pad, and obtain the PCBA board divided into several independent sub-detection areas.

[0007] Step S1-1: Read the historical completed soldering preset pad sample data of the same model PCBA board, retrieve the geometric parameters of the preset pads determined in the PCBA board design stage. The geometric parameters represent the shape, center coordinates and outline size parameters of each preset pad in the PCBA board design stage. Based on the geometric parameters, locate the corresponding position of each preset pad in the historical preset pad sample data.

[0008] X-ray imaging data of pre-set solder pads that have been soldered in the past are collected by an X-ray imaging sensor. Based on the geometric parameters of the pre-set solder pads in the PCBA board design stage, the position of each solder pad is located. The largest effective area covering the solder pad and with the largest solder area is extracted. Based on the boundary and size of this area, independent sub-detection areas are divided. The sub-detection area division results with clear boundaries and no cross interference are obtained. The detection range is accurately anchored, the feature signal interference between different solder pad areas is eliminated, and a stable regional foundation is laid for feature extraction and data screening, avoiding feature extraction deviation caused by ambiguous regional boundaries.

[0009] Step S1-2: For each preset pad, extract all candidate areas covering the preset pad from historical preset pad sample data, compare the welding area of ​​each candidate area, select the candidate area with the largest welding area that can completely contain the corresponding preset pad, and determine it as the maximum effective area corresponding to the preset pad. Based on the boundary and size of the maximum effective area corresponding to each preset pad, complete the sub-detection area division of the same model PCBA board; the boundary refers to the outer perimeter outline limit of the maximum effective area, and the size refers to the length, width and area measurement parameters of the maximum effective area. Based on the boundary and size, the range of each sub-detection area and the division boundary of adjacent areas can be clearly defined.

[0010] In the actual production process of PCBA board production lines, small fluctuations in process parameters and batch differences in welding consumables are difficult to completely avoid. These variables directly cause the local temporal performance of features such as the proportion of low grayscale area, grayscale dispersion, and edge sharpness in X-ray imaging of solder pads collected by X-ray imaging sensors to deviate from the global historical benchmark. Traditional defect detection uses a fixed threshold judgment mode, which sets the judgment standard based on only a single dimension of the global historical benchmark. It cannot make a comprehensive judgment from multiple dimensions such as the universality of global historical data and the timeliness of local temporal data. It is easy to cause detection misjudgment or omission due to feature benchmark drift, which is difficult to adapt to the precise quality control needs under the dynamic production conditions of the production line.

[0011] Based on the geometric parameters of the pads preset during the PCBA board design stage, the position of each pad is located, and the largest effective area covering the pad and having the largest welding area is extracted. Based on the boundary and size of this area, independent sub-detection areas are divided. The sub-detection area division results with clear boundaries and no cross interference are obtained. The detection range is accurately anchored, the feature signal interference between different pad areas is eliminated, and a stable regional foundation is laid for feature extraction and data screening, avoiding feature extraction deviation caused by ambiguous regional boundaries.

[0012] Step S2: Select any sub-detection area of ​​any PCBA board of any model as the research object, and denot it as the target sub-region; based on the PCBA board model to which the target sub-region belongs as the index, obtain the historical grayscale feature benchmark data of X-ray imaging corresponding to the target sub-region under that model, and analyze and extract the pad sample data of abnormal soldering, and denot it as abnormal pad data.

[0013] Step S2-1: Select any sub-detection area from any type of PCBA board as the target sub-region. Using the PCBA board model to which the target sub-region belongs as the retrieval index, retrieve the grayscale feature reference data of the historical X-ray imaging of the corresponding sub-detection area of ​​the PCBA board of that model from the historical soldering sample database. The grayscale feature reference data includes the low grayscale area ratio reference value, grayscale dispersion reference value and edge sharpness reference value of the X-ray imaging.

[0014] The low grayscale area ratio benchmark is used to determine the solder filling status of the pads, defining the quantitative boundary between sufficient solder and insufficient solder, internal voids, and other defects. It is a core indicator for capturing solder filling abnormalities caused by process parameter fluctuations and batch differences in consumables. The grayscale dispersion benchmark is used to measure the uniformity of grayscale distribution in X-ray imaging of the pads, clarifying the distinction between normal soldering and abnormal states such as solder inclusions and uneven distribution. It is a key basis for judging the stability of soldering quality. The edge sharpness benchmark is used to define the clarity of the boundary between the pad and the solder, distinguishing the state threshold between normal soldering and latent defects such as cold solder joints, bridging, and solder overflow. It can identify poor solder wetting problems caused by process fluctuations.

[0015] Step S2-2: Filter all historical pad sample X-ray imaging data of the corresponding sub-detection area of ​​the PCBA board of this model, extract the pad sample data marked as abnormal soldering in history and record it as abnormal pad data, and extract the pad sample data marked as normal soldering in history and record it as normal pad data.

[0016] The target sub-region is selected and PCBA board model is used as the search index. Historical grayscale feature benchmark data of X-ray imaging of the corresponding sub-region is retrieved. The historical pad sample data is filtered to separate abnormal pad data from normal pad data. A special sample dataset with high matching degree with the model of the target sub-region is obtained. The sample data is accurately classified, and the interference of samples from irrelevant models and irrelevant regions is filtered out. The data source for weight calculation and threshold generation is targeted, and reliable data support is provided for balancing global historical data and local time series data.

[0017] Step S3: Extract features from the target sub-region based on the abnormal pad data. The features include the low grayscale area ratio feature, grayscale dispersion feature, and edge sharpness feature. The first weight value corresponding to each feature is obtained through feature analysis. The features are normalized and the defect detection threshold of the target sub-region is calculated by combining the first weight value corresponding to each feature.

[0018] Step S3-1: Extract the low grayscale area ratio feature value, grayscale dispersion feature value and edge sharpness feature value of the normal soldering of each PCBA board in history based on the normal solder pad data. Calculate the mean and three times the standard deviation of each feature value respectively. Determine the normal feature threshold corresponding to each feature based on the mean and three times the standard deviation.

[0019] Step S3-2: Extract the low grayscale area ratio, grayscale dispersion and edge sharpness features of abnormal soldering of each PCBA board in history based on the abnormal solder pad data. Count the number of feature values ​​in the abnormal solder pad data that exceed the corresponding normal feature threshold under each feature, and calculate the sum of the number corresponding to the three features.

[0020] Step S3-3: Calculate the ratio of the number of thresholds exceeding the threshold corresponding to a single feature to the sum of the number of thresholds exceeding the threshold of the three features, and determine each ratio as the first weight value of the corresponding feature.

[0021] Step S3-4: Normalize the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value in the normal pad data; multiply each normalized feature value by the first weight value of the corresponding feature, and perform a summation operation on the multiplication result to obtain the normal weighted value of the target sub-region of each historical PCBA board; calculate the mean and three times the standard deviation of all normal weighted values, and determine the defect detection threshold of the target sub-region based on the mean and three times the standard deviation.

[0022] The specific process of normalization is as follows: determine the maximum and minimum values ​​of the low grayscale area ratio feature, grayscale dispersion feature, and edge sharpness feature in the normal pad data respectively; for each feature's individual feature value, subtract the corresponding feature's minimum value from the value, and then divide by the difference between the feature's maximum and minimum values; after calculation, obtain the normalized result of each feature value, map it to the 0-1 interval, eliminate the dimensional differences between different features, and complete the normalization process;

[0023] Based on normal pad data, the mean and three standard deviations of each feature are calculated to determine the normal feature threshold. The proportion of each feature exceeding the threshold in abnormal pad data is statistically determined to determine the first weight value. After normalizing and weighting the feature values ​​of normal pads, the defect detection threshold is obtained through statistical calculation. The first weight value that quantitatively represents the defect contribution of each feature and the universal defect detection threshold that fits the global historical welding status are obtained. Differentiated weight allocation of defect judgment benchmark is realized, the influence of different features on welding defects is clarified, and the problem of insufficient judgment accuracy caused by the lack of weight differentiation in traditional fixed thresholds is solved.

[0024] Step S4: Retrieve historical normal soldering sample data of PCBA boards of the corresponding model in the target sub-region, extract the normalized values ​​of each feature of the target sub-region of two consecutive boards, calculate the absolute difference of the normalized values ​​of the same feature, and perform weighted calculations based on the first weight value of each feature to obtain the feature floating value; perform analysis and calculations based on the feature floating value to obtain the analysis threshold for screening nearby anomalies.

[0025] Step S4-1: Retrieve the normal feature thresholds corresponding to each feature obtained in step S3-1, and normalize the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value in the normal pad data respectively.

[0026] Step S4-2: Multiply each normalized feature value by the first weight value of the corresponding feature, and sum the results to obtain the welding defect values ​​of the target sub-regions of each historical PCBA board. The welding defect value is a weighted sum of the normalized values ​​of the low grayscale area ratio feature, grayscale dispersion feature, and edge sharpness feature based on normal pad X-ray imaging, and the first weight value of each feature. It is used to quantify the degree to which the welding state of the pads in the target sub-regions of the PCBA board deviates from the normal welding reference. Sort all welding defect values ​​according to the welding time sequence.

[0027] Step S4-3: Calculate the absolute difference between two adjacent welding defect values ​​after sorting, select the maximum value among the absolute differences, and determine it as the feature floating value. The feature floating value is a quantitative parameter used to screen out pad sample data that are close to abnormal welding conditions but have not yet reached the defect judgment criteria.

[0028] Step S4-4: Subtract the feature floating value from the normal feature threshold corresponding to each feature obtained in step S3-1 to obtain the analysis threshold for nearby anomaly screening corresponding to each feature.

[0029] Normal pad feature values ​​are normalized and weighted by a first weight value to obtain welding defect values. After sorting by welding time, the maximum absolute difference between adjacent defect values ​​is calculated to determine the feature fluctuation value. The normal feature threshold is subtracted from the feature fluctuation value to obtain the analysis threshold. The analysis threshold used to define the near abnormal welding state is obtained. This enables a proactive warning for defect judgment, identifies pad samples that have deviated from the normal state but have not reached the defect detection threshold, and fills the technical gap of traditional fixed thresholds that can only determine whether there is a defect.

[0030] Step S5: Set a sliding window to capture historical normal soldering sample data of the corresponding PCBA board model in the target sub-region, and obtain the second weight value of each feature through feature extraction and analysis; calculate the dynamic weight value by combining the first weight value of each feature, calculate the dynamic defect detection value based on the dynamic weight value, and perform intelligent defect detection on the soldering.

[0031] Step S5-1: Set the window size of the sliding window to the sample data volume corresponding to the preset number of continuously soldered PCBA boards, set the window sliding step size to 1, and use the sliding window to extract historical normal soldering sample data in time sequence to obtain several sets of continuous window sample data.

[0032] Step S5-2: Extract the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value for each group of sample data in the sliding window. Retrieve the normal feature threshold corresponding to each feature obtained in step S3-1. Count the number of each feature value in the sliding window that exceeds the corresponding normal feature threshold. At the same time, calculate the sum of the number of values ​​exceeding the threshold corresponding to the three features in the sliding window. Calculate the ratio of the number of values ​​exceeding the threshold for a single feature to the sum of the number of values ​​exceeding the threshold. Determine each ratio as the second weight value of the corresponding feature.

[0033] Step S5-3: Read the first weight value and the second weight value corresponding to each feature, calculate the average value of the first weight value and the second weight value corresponding to each feature, and determine the average value as the dynamic weight value corresponding to each feature; acquire the X-ray imaging data of the target sub-region of the PCBA board to be detected, extract the low grayscale area ratio feature value, grayscale dispersion feature value and edge sharpness feature value from the data and perform normalization processing, calculate the dynamic defect detection value according to the dynamic weight value corresponding to each feature, when the dynamic defect detection value exceeds the dynamic defect detection threshold, it is determined that there is a welding defect in the target sub-region of the PCBA board to be detected; when the defect detection calculation value does not exceed the dynamic defect detection threshold, it is determined that the welding status of the target sub-region of the PCBA board to be detected is normal.

[0034] The system sets the sliding window size and step size to capture historical normal welding sample data. It extracts the sample feature values ​​within the window and calculates the second weight value. The average of the first and second weight values ​​is calculated to determine the dynamic weight value. Based on the dynamic weight value, the feature values ​​of the sample to be tested are weighted to obtain the dynamic defect detection value and complete the judgment. The system obtains dynamic weight values ​​and dynamic defect detection values ​​that take into account both the universality of global historical data and the timeliness of local time-series data. It achieves adaptive adaptation to feature benchmark drift caused by fluctuations in production line process parameters and batch differences in welding consumables, avoids misjudgment and missed judgment in the detection process, and meets the needs of precise quality control under dynamic production conditions.

[0035] Furthermore, a defect intelligent detection system for PCBA boards includes a region division module, a data filtering module, a weight threshold calculation module, a floating threshold generation module, and a dynamic defect detection module.

[0036] The region division module is used for data processing related to the division of sub-detection regions of PCBA boards of the same model based on preset pads; the data filtering module is used for retrieving the grayscale feature benchmark data of X-ray imaging of the target sub-region corresponding to the PCBA board of the same model and filtering historical pad sample data; the weight threshold calculation module is used for calculating the first weight value of each feature of the target sub-region of the PCBA board and generating the defect detection threshold; the floating threshold generation module is used for calculating the floating value of the feature of the target sub-region of the PCBA board and generating the threshold for screening and analyzing adjacent anomalies; the dynamic defect detection module is used for calculating the dynamic weight value of each feature of the target sub-region of the PCBA board and determining the welding defects of the sample to be detected.

[0037] The region division module includes a pad positioning unit and a region segmentation unit; the pad positioning unit is used to retrieve preset pad geometric parameters, locate positions, and extract the maximum effective area for PCBA boards of the same model; the region segmentation unit is used to divide the detection area of ​​PCBA boards of the same model based on the boundary and size of the maximum effective area.

[0038] The data filtering module includes a benchmark retrieval unit and a sample sorting unit. The benchmark retrieval unit is used to retrieve and retrieve historical X-ray imaging grayscale feature benchmark data of the corresponding sub-detection area according to the PCBA board model. The sample sorting unit is used to filter historical pad sample data of the corresponding PCBA board model in the target sub-area and separate abnormal pad data from normal pad data.

[0039] The weight threshold calculation module includes a weight calculation unit and a threshold generation unit; the weight calculation unit is used to calculate the proportion of each feature exceeding the normal feature threshold in the abnormal pad data and determine the first weight value of each feature; the threshold generation unit is used to normalize and weight the feature values ​​of the normal pad data and generate the defect detection threshold of the target sub-region by combining the statistical results.

[0040] The floating threshold generation module includes a floating value calculation unit and a floating value calculation unit; the floating value calculation unit is used to normalize and weight the feature values ​​of normal pad data to obtain welding defect values ​​and determine feature floating values; the analysis threshold generation unit is used to combine normal feature thresholds and feature floating values ​​to generate adjacent anomaly screening analysis thresholds corresponding to each feature.

[0041] The dynamic defect detection module includes a dynamic weight calculation unit and a defect determination unit. The dynamic weight calculation unit is used to calculate the average of the first weight value and the second weight value of each feature to determine the dynamic weight value of each feature. The defect determination unit is used to perform normalized weighted calculation of the feature values ​​of the target sub-region of the PCBA board to be inspected, and to complete the determination of the welding defect status by combining the dynamic defect detection value.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] 1. This invention locates the pad position by retrieving preset pad geometric parameters, selects the largest effective area covering the pad and with the largest welding area, and divides independent sub-detection areas based on the boundary and size of this area, eliminating feature signal interference between different pad areas, accurately anchoring the detection range, providing a stable regional basis for feature extraction and data screening, avoiding feature extraction deviations caused by ambiguous regional boundaries, and ensuring the relevance and reliability of the data source.

[0044] 2. This invention determines the first weight value by statistically analyzing the proportion of each feature exceeding the threshold in abnormal pad data, and calculates the second weight value by combining the local time-series sample data captured by the sliding window. The two weight values ​​are analyzed and calculated to obtain a dynamic weight value, which balances the universality of global historical data and the timeliness of local time-series data, suppresses feature benchmark drift caused by process parameter fluctuations and batch differences of consumables, and solves the problem of poor adaptability of traditional fixed weight judgment.

[0045] 3. This invention calculates the maximum absolute difference of soldering defect values ​​in continuous PCBA boards to obtain a feature fluctuation value. This value is then subtracted from the normal feature threshold to generate an analysis threshold. This constructs a refined judgment system that combines defect judgment with early warning of impending anomalies. This system breaks through the limitation of traditional fixed thresholds, which can only determine whether a defect exists. It identifies solder pad samples that do not meet the defect standard but have deviated from the normal state, improves the foresight of production line quality control, and reduces the risk of terminal equipment failure. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating an intelligent defect detection method for PCBA boards according to the present invention.

[0047] Figure 2 This is a schematic diagram of the structure of an intelligent defect detection system for PCBA boards according to the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1: As Figure 1 As shown, the present invention provides a technical solution, an intelligent defect detection method applied to PCBA boards, the intelligent defect detection method comprising the following steps:

[0050] Step S1: For PCBA boards of the same model, read the preset pad sample data of PCBA boards of the same model that have been soldered in the past and divide the area, filter out the largest effective area that can completely contain the corresponding preset pad, and obtain the PCBA board divided into several independent sub-detection areas.

[0051] Step S1-1: Read the historical completed soldering preset pad sample data of the same model PCBA board, retrieve the geometric parameters of the preset pads determined in the PCBA board design stage. The geometric parameters represent the shape, center coordinates and outline size parameters of each preset pad in the PCBA board design stage. Based on the geometric parameters, locate the corresponding position of each preset pad in the historical preset pad sample data.

[0052] Step S1-2: For each preset pad, extract all candidate areas covering the preset pad from historical preset pad sample data, compare the welding area of ​​each candidate area, select the candidate area with the largest welding area that can completely contain the corresponding preset pad, and determine it as the maximum effective area corresponding to the preset pad. Based on the boundary and size of the maximum effective area corresponding to each preset pad, complete the sub-detection area division of the same model PCBA board; the boundary refers to the outer perimeter outline limit of the maximum effective area, and the size refers to the length, width and area measurement parameters of the maximum effective area. Based on the boundary and size, the range of each sub-detection area and the division boundary of adjacent areas can be clearly defined.

[0053] In practice, historical soldering sample data of a certain PCBA board model is selected for operation. The preset pad geometric parameters retained during the design stage of the PCBA board model are retrieved. The precise positioning of each pad in the sample data is completed according to the parameters. For each pad, the entire coverage candidate area is extracted and the soldering area is compared. The largest effective area that meets the requirements is selected. Then, the sub-detection area is divided based on the boundary and size of the area to ensure that each sub-detection area is independent and does not overlap.

[0054] Step S2: Select any sub-detection area of ​​any PCBA board of any model as the research object, and denot it as the target sub-region; based on the PCBA board model to which the target sub-region belongs as the index, obtain the historical grayscale feature benchmark data of X-ray imaging corresponding to the target sub-region under that model, and analyze and extract the pad sample data of abnormal soldering, and denot it as abnormal pad data.

[0055] Step S2-1: Select any sub-detection area from any type of PCBA board as the target sub-region. Using the PCBA board model to which the target sub-region belongs as the retrieval index, retrieve the grayscale feature reference data of the historical X-ray imaging of the corresponding sub-detection area of ​​the PCBA board of that model from the historical soldering sample database. The grayscale feature reference data includes the low grayscale area ratio reference value, grayscale dispersion reference value and edge sharpness reference value of the X-ray imaging.

[0056] Step S2-2: Filter all historical pad sample X-ray imaging data of the corresponding sub-detection area of ​​the PCBA board of this model, extract the pad sample data marked as abnormal soldering in history and record it as abnormal pad data, and extract the pad sample data marked as normal soldering in history and record it as normal pad data.

[0057] In practice, any model of PCBA board with pre-defined sub-inspection areas can be selected, and one of the sub-inspection areas can be chosen as the target sub-area. Using the model of the PCBA board as the search condition, the X-ray imaging grayscale feature benchmark data of the corresponding sub-inspection area can be retrieved from the historical soldering sample database. Then, all historical pad sample X-ray imaging data of the corresponding sub-inspection area of ​​the PCBA board of that model are comprehensively screened, and abnormal pad data and normal pad data are strictly separated according to the soldering status labeling results to ensure that the two types of data are clearly classified and not confused.

[0058] Step S3: Extract features from the target sub-region based on the abnormal pad data. The features include the low grayscale area ratio feature, grayscale dispersion feature, and edge sharpness feature. The first weight value corresponding to each feature is obtained through feature analysis. The features are normalized and the defect detection threshold of the target sub-region is calculated by combining the first weight value corresponding to each feature.

[0059] Step S3-1: Extract the low grayscale area ratio feature value, grayscale dispersion feature value and edge sharpness feature value of the normal soldering of each PCBA board in history based on the normal solder pad data. Calculate the mean and three times the standard deviation of each feature value respectively. Determine the normal feature threshold corresponding to each feature based on the mean and three times the standard deviation.

[0060] Step S3-2: Extract the low grayscale area ratio, grayscale dispersion and edge sharpness features of abnormal soldering of each PCBA board in history based on the abnormal solder pad data. Count the number of feature values ​​in the abnormal solder pad data that exceed the corresponding normal feature threshold under each feature, and calculate the sum of the number corresponding to the three features.

[0061] Step S3-3: Calculate the ratio of the number of thresholds exceeding the threshold corresponding to a single feature to the sum of the number of thresholds exceeding the threshold of the three features, and determine each ratio as the first weight value of the corresponding feature.

[0062] Step S3-4: Normalize the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value in the normal pad data; multiply each normalized feature value by the first weight value of the corresponding feature, and perform a summation operation on the multiplication result to obtain the normal weighted value of the target sub-region of each historical PCBA board; calculate the mean and three times the standard deviation of all normal weighted values, and determine the defect detection threshold of the target sub-region based on the mean and three times the standard deviation.

[0063] In practice, based on the selected normal and abnormal pad data, specific values ​​of the low grayscale area ratio, grayscale dispersion, and edge sharpness features are extracted from the two types of data. The normal feature threshold is determined by calculating the mean and three times the standard deviation of each feature in the normal pad data. The number and total number of feature values ​​exceeding the normal feature threshold in the abnormal pad data are counted. The first weight value corresponding to each feature is obtained by calculating the proportion. Then, the feature values ​​of the normal pad data are normalized and weighted by combining the first weight value. The defect detection threshold of the target sub-region is determined by calculating the mean and three times the standard deviation of all normal weighted values.

[0064] Step S4: Retrieve historical normal soldering sample data of PCBA boards of the corresponding model in the target sub-region, extract the normalized values ​​of each feature of the target sub-region of two consecutive boards, calculate the absolute difference of the normalized values ​​of the same feature, and perform weighted calculations based on the first weight value of each feature to obtain the feature floating value; perform analysis and calculations based on the feature floating value to obtain the analysis threshold for screening nearby anomalies.

[0065] Step S4-1: Retrieve the normal feature thresholds corresponding to each feature obtained in step S3-1, and normalize the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value in the normal pad data respectively.

[0066] Step S4-2: Multiply each normalized feature value by the first weight value of the corresponding feature, and sum the results to obtain the welding defect values ​​of the target sub-regions of each historical PCBA board. The welding defect value is a weighted sum of the normalized values ​​of the low grayscale area ratio feature, grayscale dispersion feature, and edge sharpness feature based on normal pad X-ray imaging, and the first weight value of each feature. It is used to quantify the degree to which the welding state of the pads in the target sub-regions of the PCBA board deviates from the normal welding reference. Sort all welding defect values ​​according to the welding time sequence.

[0067] Step S4-3: Calculate the absolute difference between two adjacent welding defect values ​​after sorting, select the maximum value among the absolute differences, and determine it as the feature floating value. The feature floating value is a quantitative parameter used to screen out pad sample data that are close to abnormal welding conditions but have not yet reached the defect judgment criteria.

[0068] Step S4-4: Subtract the feature floating value from the normal feature threshold corresponding to each feature obtained in step S3-1 to obtain the analysis threshold for nearby anomaly screening corresponding to each feature.

[0069] In practice, historical normal welding sample data of the corresponding PCBA board model in the target sub-region is retrieved. The normal feature threshold determined in step S3-1 is used to normalize the feature values ​​of the sample data. The first weight value is combined to complete the weighted calculation to obtain the welding defect value of the target sub-region of each PCBA board. All welding defect values ​​are arranged in order according to the welding time. The absolute difference between two adjacent welding defect values ​​after arrangement is calculated in turn. The maximum value is selected as the feature floating value. The normal feature threshold of each feature obtained in step S3-1 is subtracted from the feature floating value to obtain the nearest anomaly screening analysis threshold corresponding to each feature.

[0070] Step S5: Set a sliding window to capture historical normal soldering sample data of the corresponding PCBA board model in the target sub-region, and obtain the second weight value of each feature through feature extraction and analysis; calculate the dynamic weight value by combining the first weight value of each feature, calculate the dynamic defect detection value based on the dynamic weight value, and perform intelligent defect detection on the soldering.

[0071] Step S5-1: Set the window size of the sliding window to the sample data volume corresponding to the preset number of continuously soldered PCBA boards, set the window sliding step size to 1, and use the sliding window to extract historical normal soldering sample data in time sequence to obtain several sets of continuous window sample data.

[0072] Step S5-2: Extract the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value for each group of sample data in the sliding window. Retrieve the normal feature threshold corresponding to each feature obtained in step S3-1. Count the number of each feature value in the sliding window that exceeds the corresponding normal feature threshold. At the same time, calculate the sum of the number of values ​​exceeding the threshold corresponding to the three features in the sliding window. Calculate the ratio of the number of values ​​exceeding the threshold for a single feature to the sum of the number of values ​​exceeding the threshold. Determine each ratio as the second weight value of the corresponding feature.

[0073] Step S5-3: Read the first weight value and the second weight value corresponding to each feature, calculate the average value of the first weight value and the second weight value corresponding to each feature, and determine the average value as the dynamic weight value corresponding to each feature; acquire the X-ray imaging data of the target sub-region of the PCBA board to be detected, extract the low grayscale area ratio feature value, grayscale dispersion feature value and edge sharpness feature value from the data and perform normalization processing, calculate the dynamic defect detection value according to the dynamic weight value corresponding to each feature, when the dynamic defect detection value exceeds the dynamic defect detection threshold, it is determined that there is a welding defect in the target sub-region of the PCBA board to be detected; when the defect detection calculation value does not exceed the dynamic defect detection threshold, it is determined that the welding status of the target sub-region of the PCBA board to be detected is normal.

[0074] In specific implementation, a sliding window is set up to perform time-series interception operation on the historical normal soldering sample data of the corresponding PCBA board model in the target sub-region. The specific values ​​of three types of features are extracted from each group of window sample data. Combined with the normal feature threshold determined in step S3-1, the number and sum of each feature exceeding the threshold are counted, and the proportion is calculated to obtain the second weight value corresponding to each feature. The first weight value obtained in step S3-3 is read and the arithmetic mean of the two is calculated to obtain the dynamic weight value. The X-ray imaging data of the target sub-region of the PCBA board to be detected is obtained and the three types of feature values ​​are extracted and normalized. The dynamic defect detection value is calculated by combining the dynamic weight value. The soldering defect status is determined by comparing the value with the dynamic defect detection threshold.

[0075] Example 2, as Figure 2 As shown, the present invention provides a defect intelligent detection system for PCBA boards. The defect intelligent detection system includes a region division module, a data filtering module, a weight threshold calculation module, a floating threshold generation module, and a dynamic defect detection module.

[0076] The region division module is used for data processing related to the division of sub-detection regions of PCBA boards of the same model based on preset pads; the data filtering module is used for retrieving the grayscale feature benchmark data of X-ray imaging of the target sub-region corresponding to the PCBA board of the same model and filtering historical pad sample data; the weight threshold calculation module is used for calculating the first weight value of each feature of the target sub-region of the PCBA board and generating the defect detection threshold; the floating threshold generation module is used for calculating the floating value of the feature of the target sub-region of the PCBA board and generating the threshold for screening and analyzing adjacent anomalies; the dynamic defect detection module is used for calculating the dynamic weight value of each feature of the target sub-region of the PCBA board and determining the welding defects of the sample to be detected.

[0077] The output of the region division module is electrically connected to the input of the data filtering module; the output of the data filtering module is electrically connected to the input of the weight threshold calculation module; the output of the weight threshold calculation module is electrically connected to the input of the floating threshold generation module; and the output of the floating threshold generation module is electrically connected to the input of the dynamic defect detection module.

[0078] The region division module includes a pad positioning unit and a region segmentation unit; the pad positioning unit is used to retrieve preset pad geometric parameters, locate positions, and extract the maximum effective area for PCBA boards of the same model; the region segmentation unit is used to divide the detection area of ​​PCBA boards of the same model based on the boundary and size of the maximum effective area.

[0079] The data filtering module includes a benchmark retrieval unit and a sample sorting unit. The benchmark retrieval unit is used to retrieve and retrieve historical X-ray imaging grayscale feature benchmark data of the corresponding sub-detection area according to the PCBA board model. The sample sorting unit is used to filter historical pad sample data of the corresponding PCBA board model in the target sub-area and separate abnormal pad data from normal pad data.

[0080] The weight threshold calculation module includes a weight calculation unit and a threshold generation unit; the weight calculation unit is used to calculate the proportion of each feature exceeding the normal feature threshold in the abnormal pad data and determine the first weight value of each feature; the threshold generation unit is used to normalize and weight the feature values ​​of the normal pad data and generate the defect detection threshold of the target sub-region by combining the statistical results.

[0081] The floating threshold generation module includes a floating value calculation unit and a floating value calculation unit; the floating value calculation unit is used to normalize and weight the feature values ​​of normal pad data to obtain welding defect values ​​and determine feature floating values; the analysis threshold generation unit is used to combine normal feature thresholds and feature floating values ​​to generate adjacent anomaly screening analysis thresholds corresponding to each feature.

[0082] The dynamic defect detection module includes a dynamic weight calculation unit and a defect determination unit. The dynamic weight calculation unit is used to calculate the average of the first weight value and the second weight value of each feature to determine the dynamic weight value of each feature. The defect determination unit is used to perform normalized weighted calculation of the feature values ​​of the target sub-region of the PCBA board to be inspected, and to complete the determination of the welding defect status by combining the dynamic defect detection value.

[0083] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A defect intelligent detection method applied to PCBA boards, characterized in that: The intelligent defect detection method includes the following steps: Step S1: For PCBA boards of the same model, read the preset pad sample data of PCBA boards of the same model that have been soldered in the past and divide the area, filter out the largest effective area that can completely contain the corresponding preset pad, and obtain the PCBA board divided into several independent sub-detection areas. Step S2: Select any sub-detection area of ​​any PCBA board of any model as the research object, and denot it as the target sub-region; based on the PCBA board model to which the target sub-region belongs as the index, obtain the historical grayscale feature benchmark data of X-ray imaging corresponding to the target sub-region under that model, and analyze and extract the pad sample data of abnormal soldering, and denot it as abnormal pad data. Step S3: Extract features from the target sub-region based on the abnormal pad data. The features include the low grayscale area ratio feature, grayscale dispersion feature, and edge sharpness feature. The first weight value corresponding to each feature is obtained through feature analysis. The features are normalized and the defect detection threshold of the target sub-region is calculated by combining the first weight value corresponding to each feature. Step S4: Retrieve historical normal soldering sample data of PCBA boards of the corresponding model in the target sub-region, extract the normalized values ​​of each feature of the target sub-region of two consecutive boards, calculate the absolute difference of the normalized values ​​of the same feature, and perform weighted calculations based on the first weight value of each feature to obtain the feature floating value; perform analysis and calculations based on the feature floating value to obtain the analysis threshold for screening nearby anomalies. Step S5: Set a sliding window to capture historical normal soldering sample data of the corresponding PCBA board model in the target sub-region, and obtain the second weight value of each feature through feature extraction and analysis; calculate the dynamic weight value by combining the first weight value of each feature, calculate the dynamic defect detection value based on the dynamic weight value, and perform intelligent defect detection on the soldering.

2. The intelligent defect detection method for PCBA boards according to claim 1, characterized in that: The specific steps of step S1 are as follows: Step S1-1: Read the historical completed soldering preset pad sample data of the same model PCBA board, retrieve the geometric parameters of the preset pads determined in the PCBA board design stage. The geometric parameters represent the shape, center coordinates and outline size parameters of each preset pad in the PCBA board design stage. Based on the geometric parameters, locate the corresponding position of each preset pad in the historical preset pad sample data. Step S1-2: For each preset pad, extract all candidate areas covering the preset pad from the historical preset pad sample data, compare the welding area of ​​each candidate area, select the candidate area with the largest welding area that can completely contain the corresponding preset pad, and determine it as the maximum effective area corresponding to the preset pad. Based on the boundary and size of the maximum effective area corresponding to each preset pad, complete the sub-detection area division of the same model PCBA board; the boundary refers to the outer perimeter outline limit of the maximum effective area, and the size refers to the length, width and area measurement parameters of the maximum effective area. Based on the boundary and size, the range of each sub-detection area and the division boundary of adjacent areas can be clearly defined.

3. The intelligent defect detection method for PCBA boards according to claim 2, characterized in that: The specific steps of step S2 are as follows: Step S2-1: Select any sub-detection area from any type of PCBA board as the target sub-region. Using the PCBA board model to which the target sub-region belongs as the retrieval index, retrieve the grayscale feature reference data of the historical X-ray imaging of the corresponding sub-detection area of ​​the PCBA board of that model from the historical soldering sample database. The grayscale feature reference data includes the low grayscale area ratio reference value, grayscale dispersion reference value and edge sharpness reference value of the X-ray imaging. Step S2-2: Filter all historical pad sample X-ray imaging data of the corresponding sub-detection area of ​​the PCBA board of this model, extract the pad sample data marked as abnormal soldering in history and record it as abnormal pad data, and extract the pad sample data marked as normal soldering in history and record it as normal pad data.

4. The intelligent defect detection method for PCBA boards according to claim 3, characterized in that: The specific steps of step S3 are as follows: Step S3-1: Extract the low grayscale area ratio feature value, grayscale dispersion feature value and edge sharpness feature value of each PCBA board with normal soldering based on the normal pad data. Calculate the mean and three times the standard deviation of each feature value. Determine the normal feature threshold corresponding to each feature based on the mean and three times the standard deviation. Step S3-2: Extract the low grayscale area ratio, grayscale dispersion and edge sharpness features of abnormal soldering of each PCBA board in history based on the abnormal solder pad data. Count the number of feature values ​​in the abnormal solder pad data that exceed the corresponding normal feature threshold under each feature, and calculate the sum of the number corresponding to the three features. Step S3-3: Calculate the ratio of the number of thresholds exceeding the threshold corresponding to a single feature to the sum of the number of thresholds exceeding the threshold of the three features, and determine each ratio as the first weight value of the corresponding feature. Step S3-4: Normalize the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value in the normal pad data respectively; multiply each normalized feature value by the first weight value of the corresponding feature, and perform a summation operation on the multiplication result to obtain the normal weighted value of the target sub-region of each historical PCBA board. The mean and three standard deviations of all normally weighted values ​​are statistically analyzed, and the defect detection threshold of the target sub-region is determined based on the mean and three standard deviations.

5. The intelligent defect detection method for PCBA boards according to claim 4, characterized in that: The specific steps of step S4 are as follows: Step S4-1: Retrieve the normal feature thresholds corresponding to each feature obtained in step S3-1, and normalize the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value in the normal pad data respectively. Step S4-2: Multiply each normalized feature value by the first weight value of the corresponding feature, and sum the results to obtain the welding defect values ​​of the target sub-regions of each historical PCBA board. The welding defect value is a weighted sum of the normalized values ​​of the low grayscale area ratio feature, grayscale dispersion feature, and edge sharpness feature based on normal pad X-ray imaging, and the first weight value of each feature. It is used to quantify the degree to which the welding state of the pads in the target sub-regions of the PCBA board deviates from the normal welding reference. Sort all welding defect values ​​according to the welding time sequence.

6. The intelligent defect detection method for PCBA boards according to claim 5, characterized in that: Step S4 also includes: Step S4-3: Calculate the absolute difference between two adjacent welding defect values ​​after sorting, select the maximum value among the absolute differences, and determine it as the feature floating value. The feature floating value is a quantitative parameter used to screen out pad sample data that are close to abnormal welding conditions but have not yet reached the defect judgment criteria. Step S4-4: Subtract the feature floating value from the normal feature threshold corresponding to each feature obtained in step S3-1 to obtain the analysis threshold for nearby anomaly screening corresponding to each feature.

7. The intelligent defect detection method for PCBA boards according to claim 6, characterized in that: The specific steps of step S5 are as follows: Step S5-1: Set the window size of the sliding window to the sample data volume corresponding to the preset number of continuously soldered PCBA boards, set the window sliding step size to 1, and use the sliding window to extract historical normal soldering sample data in time sequence to obtain several sets of continuous window sample data. Step S5-2: Extract the low grayscale area ratio feature value, grayscale dispersion feature value, and edge sharpness feature value for each group of sample data in the sliding window. Retrieve the normal feature threshold corresponding to each feature obtained in step S3-1. Count the number of each feature value in the sliding window that exceeds the corresponding normal feature threshold. At the same time, calculate the sum of the number of values ​​exceeding the threshold corresponding to the three features in the sliding window. Calculate the ratio of the number of values ​​exceeding the threshold for a single feature to the sum of the number of values ​​exceeding the threshold. Determine each ratio as the second weight value of the corresponding feature. Step S5-3: Read the first weight value and the second weight value corresponding to each feature, calculate the average value of the first weight value and the second weight value corresponding to each feature, and determine the average value as the dynamic weight value corresponding to each feature; acquire the X-ray imaging data of the target sub-region of the PCBA board to be detected, extract the low gray area area ratio feature value, gray level dispersion feature value and edge sharpness feature value from the data and normalize them, calculate the dynamic defect detection value according to the dynamic weight value corresponding to each feature, and determine that there is a welding defect in the target sub-region of the PCBA board to be detected when the dynamic defect detection value exceeds the dynamic defect detection threshold; When the calculated defect detection value does not exceed the dynamic defect detection threshold, the soldering status of the target sub-area of ​​the PCBA board to be inspected is determined to be normal.

8. A defect intelligent detection system for PCBA boards, which applies the defect intelligent detection method for PCBA boards described in any one of claims 1-7, characterized in that: The intelligent defect detection system includes a region division module, a data filtering module, a weighted threshold calculation module, a floating threshold generation module, and a dynamic defect detection module. The region division module is used for data processing related to the division of sub-detection regions of PCBA boards of the same model based on preset pads; the data filtering module is used for retrieving the grayscale feature benchmark data of X-ray imaging of the target sub-region corresponding to the PCBA board of the same model and filtering historical pad sample data; the weight threshold calculation module is used for calculating the first weight value of each feature of the target sub-region of the PCBA board and generating the defect detection threshold; the floating threshold generation module is used for calculating the floating value of the feature of the target sub-region of the PCBA board and generating the threshold for screening and analyzing adjacent anomalies; the dynamic defect detection module is used for calculating the dynamic weight value of each feature of the target sub-region of the PCBA board and determining the welding defects of the sample to be detected. The output of the region division module is electrically connected to the input of the data filtering module; the output of the data filtering module is electrically connected to the input of the weight threshold calculation module; the output of the weight threshold calculation module is electrically connected to the input of the floating threshold generation module; and the output of the floating threshold generation module is electrically connected to the input of the dynamic defect detection module.

9. The intelligent defect detection system for PCBA boards according to claim 8, characterized in that: The region division module includes a pad positioning unit and a region segmentation unit; the pad positioning unit is used to retrieve preset pad geometric parameters, locate positions, and extract the maximum effective area for PCBA boards of the same model; the region segmentation unit is used to divide the detection area of ​​PCBA boards of the same model based on the boundary and size of the maximum effective area. The data filtering module includes a benchmark retrieval unit and a sample sorting unit; the benchmark retrieval unit is used to retrieve and retrieve historical X-ray imaging grayscale feature benchmark data of the corresponding sub-detection area according to the PCBA board model; The sample sorting unit is used to filter historical pad sample data of PCBA boards of the corresponding model in the target sub-region and separate abnormal pad data from normal pad data. The weight threshold calculation module includes a weight calculation unit and a threshold generation unit. The weight calculation unit is used to calculate the proportion of each feature exceeding the normal feature threshold in the abnormal pad data and determine the first weight value of each feature. The threshold generation unit is used to perform normalized weighted calculation on the feature values ​​of normal pad data and generate the defect detection threshold of the target sub-region based on the statistical results.

10. The intelligent defect detection system for PCBA boards according to claim 8, characterized in that: The floating threshold generation module includes a floating value calculation unit and a floating value calculation unit; the floating value calculation unit is used to normalize and weight the feature values ​​of normal pad data to obtain welding defect values ​​and determine feature floating values; the analysis threshold generation unit is used to combine normal feature thresholds and feature floating values ​​to generate adjacent anomaly screening analysis thresholds corresponding to each feature. The dynamic defect detection module includes a dynamic weight calculation unit and a defect determination unit. The dynamic weight calculation unit is used to calculate the average of the first weight value and the second weight value of each feature to determine the dynamic weight value of each feature. The defect determination unit is used to perform normalized weighted calculation of the feature values ​​of the target sub-region of the PCBA board to be inspected, and to complete the determination of the welding defect status by combining the dynamic defect detection value.

Images