A tin paste printing and mounting offset traceability method based on multi-source detection data fusion

Abstract

This invention discloses a method for tracing solder paste printing and component placement misalignment based on multi-source detection data fusion, belonging to the technical field of automatic inspection systems for printed circuit boards. It includes a data caching module that receives multi-source data, constructs a data pool, and establishes a reference designator-package type mapping; a rule engine module that extracts detection results, corrects thresholds based on package type, converts them into dimensionless risk coefficients, and distributes them to a decision tree for root cause decoupling analysis according to defect type; it categorizes by batch, standardizes root cause expressions, calculates weights and normalized proportions, and outputs dominant and secondary root causes; and generates and outputs a process report. This invention solves the technical problem of how to integrate multi-source data from solder paste printing inspection and component placement inspection, and adaptively corrects judgment thresholds for different package types to achieve accurate tracing of the root causes of placement misalignment defects and batch-level root cause statistics. The invention automatically outputs dominant and secondary root causes, providing a quantitative basis for closed-loop optimization of the production process.

Description

Technical Field

[0001] This invention belongs to the technical field of automatic inspection systems for printed circuit boards, and particularly relates to a method for tracing solder paste printing and mounting offset based on multi-source inspection data fusion. Background Technology

[0002] In surface mount technology (SMT) manufacturing, misalignment between solder paste printing and component placement is a major cause of soldering defects such as tombstones, cold solder joints, solder bridging, and component misalignment. Current technologies typically employ Solder Paste Printing Inspection (SPI) equipment and Automated Optical Inspection (AOI) equipment to inspect the quality of printed pads and components after placement / reflow. However, these inspection systems often operate independently, and the results fail to effectively correlate solder paste printing data with placement data for the same component reference, making it impossible to determine whether misalignment defects originate from solder paste printing deviations or abnormal placement postures. Furthermore, different package types (such as 01005, 0201, BGA, etc.) have significantly different process requirements for solder paste volume and misalignment tolerance, but existing methods often use fixed thresholds for judgment, resulting in high false alarm and false negative rates. Some solutions attempt to introduce decision tree analysis, but lack normalization processing for the dimensionless risk coefficients of SPI and AOI, making it difficult to unify the criteria for determining the root cause of defects across different packages and inspection equipment. Furthermore, existing technologies are mostly limited to isolated analysis of single defects, and do not perform statistical clustering and weight ranking of the root causes of multiple types of defects within the same batch. This makes it impossible to identify systematic process deviations at the batch level, resulting in a lack of clear direction for process improvement. Summary of the Invention

[0003] The purpose of this invention is to provide a method for tracing the source of solder paste printing and component placement misalignment based on multi-source detection data fusion. This method solves the technical problem of how to integrate multi-source data from solder paste printing detection and component placement detection, and adaptively adjust the judgment threshold for different package types, so as to achieve accurate tracing of the root cause of placement misalignment defects and batch-level root cause statistics.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for tracing solder paste printing and mounting offset based on multi-source detection data fusion includes the following steps: S1. The data caching module receives, collects, and manages multi-source data output from the SPI device, AOI device, and MES system, and constructs an associative data pool. The multi-source data includes the detection results output by the SPI device and AOI device, as well as the engineering data output by the MES system. Identify the component package type for each reference number based on engineering data, and establish the association between the reference number and the package type; S2. The rule engine module extracts key feature information from the detection results output by the SPI device and AOI device from the data pool, patches and corrects the preset threshold according to the packaging type, makes a second judgment on the detection results, constructs a defect decision tree, and outputs the decoupling analysis results. S3. The batch decision module retrieves and categorizes the decoupling analysis results by batch. It standardizes the root cause expression in the decoupling analysis results to obtain root cause categories. It calculates the weight of each root cause category and performs normalization. It determines the primary root cause based on the proportion of each root cause category in the total weight and outputs the dominant root cause and secondary root cause. S4. Summarize the results of steps S2 and S3, generate a process report and output it for production closed-loop adjustment.

[0005] Preferably, step S1 includes the following steps: S1-1, The data caching module receives the detection results of solder paste on printed pads from the SPI device, the detection results of components after placement or reflow from the AOI device, and engineering data from the MES system; the engineering data includes PCB design files; based on the pad coordinates and component center coordinates in the PCB design files, the detection results of the SPI device, the pads and the component reference numbers to which the pads belong are bound, and the detection results of the AOI device are bound to the reference numbers; using the reference number as the primary key, the SPI detection data and AOI detection data of the same component are merged to construct a data pool; S1-2 The data cache module retrieves the PCB design file, identifies the component package type corresponding to each reference number, and generates a mapping table between reference numbers and package types.

[0006] Preferably, step S2 includes the following steps: S2-1. The rule engine module extracts the SPI-side process parameters and AOI-side process parameters corresponding to the same bit number from the data pool; it then corrects each process parameter based on the preset packaging patch rules to obtain the corrected packaging allowable threshold. The SPI-side process parameters include solder paste volume, area, height, offset, and offset direction; the AOI-side process parameters include defect type, component offset, and orientation anomaly information. S2-2. Compare and calculate the original process information with the corrected packaging allowable threshold, and convert all process parameters into dimensionless risk coefficients. S2-3. Using the defect type field in the process parameters on the AOI side as the first decision entry, different defect instances are assigned to the corresponding defect decision trees. Based on the dimensionless risk coefficient and using a fixed threshold as a unified judgment benchmark, step-by-step reasoning is performed to obtain the decoupling analysis results.

[0007] Preferably, step S3 includes the following steps: S3-1. Based on the decoupling analysis results, the batch decision module extracts data from the same batch and categorizes them by batch. S3-2. Standardize the root cause expressions in the decoupling analysis results and uniformly map them to standard root cause categories; the root cause categories include printing offset, mounting offset, and uneven solder content. S3-3. Based on the frequency of occurrence and average confidence of the root cause category, calculate the weight of each root cause category. The specific formula is as follows: ; ; Among them, W i The weight of the root cause category; N i C represents the number of times the root cause category appears. i This represents the average confidence level of the root cause category.

[0008] S3-4. Normalize the weights of all root cause categories, calculate the proportion of each root cause category in the total weight, and determine the primary cause based on the proportion: ; Among them, P i W is the normalized weight of the i-th root cause; i W represents the original weight of the i-th root cause; j is the original weight of the j-th root cause; j is the index of all root causes traversed; m is the total number of root cause categories.

[0009] Method for determining the main cause: P i ≥80% is a systemic cause; 50% ≤ P i <80% is a mixed primary and secondary cause; P i <50% is considered discrete anomaly.

[0010] S3-5. Sort all root cause categories by weight, select and output the dominant root cause and the secondary root cause.

[0011] Preferably, when performing step S4, the decoupling analysis results output in step S2 and the primary and secondary root causes output in step S3 are summarized, and the detection results of the SPI device and AOI device are combined to form a process report.

[0012] The process report includes standard process report fields such as defect location, defect type, root cause result, and confidence level; based on the root cause type, a preset root cause-suggestion mapping table is queried to generate corresponding process improvement suggestions; the process improvement suggestions are added to the process report; and the process report is output.

[0013] Preferably, when performing step S1-1, for the solder paste inspection results of the printed pads output by the SPI device, the coordinates of each inspection point are matched with the corresponding pad according to the pad coordinates in the PCB design file. After successful matching, the component reference number and pad number to which the pad belongs are bound. For the component inspection results output by the AOI device, the inspection results of each component are matched with the corresponding component reference number according to the component center coordinates in the PCB design file. When a component has multiple pads, a separate record is generated for each pad in the data pool. Each record contains the same AOI inspection data and its own SPI inspection data, and is distinguished by the pad number field.

[0014] Preferably, when performing step S2-2, the dimensionless risk coefficient includes the offset risk coefficient, the insufficient volume risk coefficient, the excessive volume risk coefficient, and the volume imbalance risk coefficient. The calculation methods for each risk coefficient are as follows: The offset risk coefficient includes the SPI offset risk coefficient and the AOI offset risk coefficient: SPI offset risk factor = Offset output by SPI device / Maximum allowable offset for this package; AOI offset risk factor = Offset output by AOI device / Maximum allowable offset for this package; Insufficient volume risk factor = Actual volume ratio / Minimum allowable volume ratio for this package; Excess volume risk factor = Measured volume ratio / Maximum allowable volume ratio of this package; Volume imbalance risk factor = absolute value of volume difference between the two ends / (average value of volume at both ends × allowable volume imbalance ratio of this package); All dimensionless risk coefficients are normalized to a pre-defined risk benchmark threshold.

[0015] Preferably, when performing steps S2-3, the defect decision tree includes a dedicated decision tree for bridging, a dedicated decision tree for cold solder joints, a dedicated decision tree for component offset, and a tombstone defect decision tree; The decision logic of the bridging solder joint decision tree is as follows: when the volume excess risk coefficient is >1.0, the root cause is determined to be excessive solder paste; when the volume excess risk coefficient is <1.0, if the SPI topography is abnormal, the root cause is determined to be printing collapse bridging; otherwise, if the AOI offset risk coefficient is >1.0 and the offset direction points to the adjacent pad, the root cause is determined to be mounting bridging; wherein, the determination that the offset direction points to the adjacent pad is based on the adjacent pad rules preset in the PCB design file. The decision tree for cold solder joints uses the following logic: when the volume deficiency risk coefficient is <1.0, the root cause is determined to be insufficient solder; when the volume deficiency risk coefficient is >1.0, if the reflow temperature obtained from the MES system does not meet the process requirements, the root cause is determined to be insufficient wetting; otherwise, if the placement pressure is abnormal, the root cause is determined to be poor contact. The determination of abnormal placement pressure includes: retrieving the offset output from the AOI equipment, calculating the absolute offset, and if the absolute offset is greater than the preset pressure abnormality offset threshold, it is determined to be abnormal placement pressure. The decision logic of the component offset-specific decision tree is as follows: when the SPI offset risk coefficient is >1.0, the root cause is determined to be the printing tape misalignment; when the SPI offset risk coefficient is <1.0, if the placement pressure is abnormal, the root cause is determined to be the nozzle problem; otherwise, if the MARK recognition is abnormal, the root cause is determined to be the vision calibration problem; whereby the MARK recognition abnormality is achieved by recognizing the error flag in the alignment mark recognition status field output by the AOI equipment vision system. The decision logic of the tombstone defect decision tree is as follows: If the volume imbalance risk coefficient is >1.0, it is determined that the main cause is the imbalance of solder volume at both ends; read the reflow temperature data provided by the MES system. If the reflow temperature does not meet the process requirements and the imbalance of solder volume at both ends has been determined to be the main cause, the root cause is determined to be the imbalance of solder volume plus abnormal reflow temperature; if the reflow temperature does not meet the process requirements and the imbalance of solder volume at both ends has not been determined to be the main cause, the root cause is determined to be insufficient solder paste or component misalignment. At this time: if the volume insufficiency risk coefficient is <1.0, it is determined to be the main cause of insufficient solder paste; if the AOI misalignment risk coefficient is >1.0, it is determined to be the tombstone caused by component misalignment.

[0016] This invention presents a method for tracing solder paste printing and component placement misalignment based on multi-source detection data fusion. It addresses the technical challenge of integrating multi-source data from solder paste printing and component placement inspections, and adaptively adjusting judgment thresholds for different package types. This enables accurate tracing of the root causes of placement misalignment defects and batch-level root cause statistics. The invention uses a data caching module to correlate and fuse multi-source data from SPI, AOI, and MES, indexed by component reference numbers, constructing a data pool containing pad-level printing data and component-level placement data. Using package type as a constraint variable, preset thresholds are patched and converted into dimensionless risk coefficients. A defect decision tree is constructed using a unified threshold of 1.0 as the judgment benchmark, achieving decoupled analysis of the mechanisms of defects such as tombstone defects, solder bridging, cold solder joints, component misalignment, and side-standing defects. Furthermore, through batch aggregation, root cause standardization mapping, weighted calculation, and normalized proportion analysis, the invention automatically outputs dominant and secondary root causes, providing quantitative basis for closed-loop optimization of the production process. Attached Figure Description

[0017] Figure 1 This is the main flowchart of the present invention; Figure 2This is a flowchart of step S1 of the present invention; Figure 3 This is a flowchart of step S2 of the present invention; Figure 4 This is a flowchart of step S3 of the present invention; Figure 5 This is a schematic diagram of the tin-connecting decision tree of the present invention; Figure 6 This is a schematic diagram of the decision tree for cold solder joints in this invention; Figure 7 This is a schematic diagram of the dedicated decision tree for element offset in this invention; Figure 8 This is a schematic diagram of the side-standing dedicated decision tree of the present invention; Figure 9 This is a schematic diagram of the tombstone defect decision tree of the present invention. Detailed Implementation

[0018] Depend on Figures 1-9 This paper presents a method for tracing solder paste printing and mounting offset based on multi-source detection data fusion, comprising the following steps: S1. The data caching module receives, aggregates, and manages multi-source data output from SPI devices, AOI devices, and the MES system, and constructs an associative data pool. The multi-source data includes the detection results output by the SPI devices and AOI devices, as well as the engineering data output by the MES system. Identify the component package type for each reference number based on engineering data, and establish the association between the reference number and the package type; In this embodiment, the data caching module first receives multi-source data from SPI devices, AOI devices, and MES systems, and then aggregates and manages the multi-source data to build an associative data pool.

[0019] Subsequently, the data caching module retrieves engineering data output by the MES system, such as product model, batch number, BOM part number table, PCB design documents, and reflow process records, and identifies and maps the component reference number, pad position, and package type corresponding to each inspection point, establishing the association between reference number and package type.

[0020] The reflux process record includes parameters such as temperature profile data, timestamps, peak temperature, heating rate, and holding time.

[0021] In this embodiment, the data caching module is mainly responsible for integrating the detection result data with the product design data, so that the system can not only identify the defect location, but also the corresponding packaging category, providing basic data support for subsequent rule-based adaptive analysis.

[0022] The data caching module ultimately outputs a structured data pool and a tag encapsulation type mapping table.

[0023] When executing step S1, the following steps are specifically included: S1-1, the data cache module receives the detection results of the solder paste on the printed pads from the SPI device, the detection results of the components after placement or reflow from the AOI device, and the engineering data from the MES system; the engineering data includes PCB design files; based on the pad coordinates and component center coordinates in the PCB design files, the detection results of the SPI device, the pads and the component reference numbers to which the pads belong are bound, and the detection results of the AOI device are bound to the reference numbers; using the reference number as the primary key, the SPI detection data and AOI detection data of the same component are merged to construct a data pool.

[0024] In this embodiment, the data caching module first receives raw data streams from the SPI device, AOI device, and MES system, i.e., multi-source data, and aggregates and manages them according to production batch number, product model, and PCB version number to build an associative data pool. The multi-source data includes the detection results output by the SPI device and AOI device, as well as the engineering data output by the MES system.

[0025] The raw data stream output by the SPI device is the solder paste inspection result of the printed pads, specifically including information such as defect type, volume, area, height, offset, offset direction, and morphological anomaly data (e.g., collapse, outward expansion). Each inspection result corresponds to a specific pad, and the SPI device will output the coordinates of that pad.

[0026] The raw data stream output by the AOI equipment consists of the component inspection results after placement or reflow, specifically including defect type, offset, offset direction, and orientation anomaly information. Each inspection result corresponds to a component reference number (e.g., C12).

[0027] The raw data stream output by the MES system includes product model, batch number, BOM part number table, and PCB design files (Gerber or CAD data).

[0028] The data caching module uses the current work order information as the entry point to load multi-source data into the data cache, forming a set of data to be associated.

[0029] In this embodiment, for the solder paste detection results of the printed pads output by the SPI device, the coordinates of each detection point output by the SPI device are matched with the corresponding pads (Pad) according to the pad coordinates in the PCB design file. After successful matching, the component reference number and pad number to which the pad belongs are bound (such as Pad1 or Pad2 of C12). For the component inspection results output by the AOI equipment, based on the component center coordinates in the PCB design file, each inspection result output by the AOI is matched with the corresponding component reference number and the reference number is bound. For the current work order information output by the MES system, the BOM part number table and PCB design file are read. Using the tag number as a unified index and the tag number as the primary key, the data of the SPI device (distinguished by pad number) and AOI device of the same component are merged to construct a mapping relationship corresponding to each tag number. After mapping all tag numbers in the PCB design file, all tag numbers and their mapping relationships are collected to construct a data pool.

[0030] For example, the data structure of capacitor C12 in a certain batch in the data pool includes basic information data, SPI data, and AOI data; Basic information data includes batch number, product model, tag number, and package type, such as [batch number: B24001, product model: PCA-01, tag number: C12, pad number: 1]; SPI data includes defect type, volume, area, height, offset, and offset direction, such as [defect type: insufficient solder, volume: 78, area: 81, height: 92, offset: 0.018, offset direction: X-]; AOI data includes defect type, offset, and offset direction, such as [defect type: tombstone, offset: 0.025, offset direction: X-].

[0031] Since a component may have multiple pads (e.g., a capacitor has two pads), a separate record is generated for each pad in the data pool. Each record contains the same AOI data and its own SPI data. The "Pad Number" field can be used to distinguish different pads of the same component, facilitating use in S2.

[0032] S1-2 The data cache module retrieves the PCB design file, identifies the component package type corresponding to each reference number, and generates a mapping table between reference numbers and package types.

[0033] After completing the data pool construction, the data caching module retrieves the PCB design file from the MES system, combines it with the current work order information, identifies the component package type corresponding to each reference designator, and generates a mapping table between reference designators and package types. The PCB design file includes pad adjacency topology data, used to define the spatial adjacency relationships between pads.

[0034] In this embodiment, since the SPI device and AOI device usually only output detection results and do not directly contain packaging information, this embodiment needs to perform joint identification through the BOM part number table or PCB design file. Generally, the packaging information is first obtained from the BOM part number table. If there is no packaging information in the BOM part number table, it is then obtained from the PCB design file. If there is no packaging information, an error is reported and an error log is generated.

[0035] When extracting packaging information, the BOM part number table is extracted with the highest priority by default. First, the BOM part number table is read, and its position number, part number and description fields are extracted. Then, the packaging identifier is parsed from the description field.

[0036] If the package type is not explicitly recorded in the BOM, the package pin library name (Footprint) in the PCB design file will be read to extract the package category.

[0037] After the data caching module completes the encapsulation information according to the preset extraction priority, it generates a mapping table between the tag number and the encapsulation type, such as C12: 01005; R15: 0201.

[0038] S2. The rule engine module extracts key feature information from the detection results output by the SPI device and AOI device from the data pool, patches and corrects the preset threshold according to the packaging type, makes a second judgment on the detection results, constructs a defect decision tree, and outputs the decoupling analysis results.

[0039] The rules engine module extracts the detection results of SPI devices and AOI devices from the data pool and extracts key feature information from them; Key feature information includes parameters such as solder paste volume, area, height, offset, defect type, and package type; The rules engine performs different threshold patch corrections based on the packaging type, performs secondary judgments on the detection results of SPI devices and AOI devices, and constructs a defect decision tree.

[0040] When performing step S2, the specific steps include the following: S2-1. The rule engine module extracts the SPI-side process parameters and AOI-side process parameters corresponding to the same bit number from the data pool; it then corrects each process parameter based on the preset packaging patch rules to obtain the corrected packaging allowable threshold. The SPI-side process parameters include solder paste volume, area, height, offset, and offset direction; the AOI-side process parameters include defect type, component offset, and orientation anomaly information.

[0041] In this embodiment, the rule engine module reads the detection results output by the SPI device and AOI device corresponding to the same bit number from the data pool constructed in step 1. It extracts SPI-side process parameters such as solder paste volume, area, height, offset, and offset direction, and AOI-side process parameters such as defect type, component offset, and posture anomaly information. The rule engine module uses the package type as a process constraint variable and corrects each process parameter based on preset package patching rules. The package patching rules adjust the solder paste volume threshold, offset tolerance, and unbalance allowable range according to different package types (e.g., 1005, 201, 402, BGA). Finally, the rule engine obtains the corrected package allowable threshold through the patching rules corresponding to the package type.

[0042] As shown in Table 1, the packaging patch rules include the maximum allowable offset, the minimum allowable volume ratio, the maximum allowable volume ratio, and the allowable volume imbalance ratio. The packaging allowable thresholds are different for each packaging type.

[0043] Table 1. Patch Packaging Rules:

[0044]

[0045] S2-2. Compare and calculate the original process information with the corrected packaging allowable threshold, and convert all process parameters into dimensionless risk coefficients.

[0046] This embodiment uses the SPI-side process parameters and AOI-side process parameters as the original process information. The original process information is compared and calculated with the corresponding corrected packaging allowable thresholds. All process parameters are then converted into dimensionless risk coefficients to unify the scale differences between different physical quantities. These dimensionless risk coefficients include offset risk coefficients, insufficient volume risk coefficients, excessive volume risk coefficients, and volume imbalance risk coefficients.

[0047] The dimensionless risk coefficient is calculated as follows: SPI offset risk factor = Offset output by SPI device / Maximum allowable offset for this package; AOI offset risk factor = Offset output by AOI device / Maximum allowable offset for this package; Insufficient volume risk factor = Actual volume ratio / Minimum allowable volume ratio for this package; Excess volume risk factor = Measured volume ratio / Maximum allowable volume ratio of this package; Volume imbalance risk factor = absolute value of volume difference between the two ends / (average volume of the two ends × allowable volume imbalance ratio of the package).

[0048] In this embodiment, if a total offset risk coefficient is required, the SPI offset risk coefficient and the AOI offset risk coefficient can be combined using methods such as weighted averaging.

[0049] In this embodiment, the preset risk benchmark threshold is 1.0. Through this standardization process using a dimensionless risk coefficient, all process features are uniformly mapped to a risk space with 1.0 as the critical judgment benchmark, so that subsequent decision tree nodes can make judgments with a unified threshold of 1.0. This avoids the rule complexity caused by the dispersion of thresholds under different packaging conditions. In this embodiment, the 1.0 threshold is a normalized risk benchmark value.

[0050] S2-3. Using the defect type field in the process parameters on the AOI side as the first decision entry, different defect instances are assigned to the corresponding defect decision trees. Based on the dimensionless risk coefficient and using a fixed threshold as a unified judgment benchmark, step-by-step reasoning is performed to obtain the decoupling analysis results.

[0051] In this embodiment, process parameters from the AOI side are retrieved, and the defect type field is used as the first decision entry point. Different defect instances are automatically assigned to the corresponding defect decision trees, and the defect formation mechanism is decoupled and analyzed by using entry diversion and decision tree reasoning to obtain the decoupling analysis results.

[0052] In all defect decision-making processes, the dimensionless risk coefficient is used as the basis, and a fixed threshold of 1.0 is used as the unified judgment benchmark for step-by-step reasoning.

[0053] The entry point triage plus decision tree reasoning method specifically includes the following steps: S2-3-1, Main entrance diversion rules: If the defect type in the process parameters on the AOI side is tombstone, then the process is routed to the tombstone defect decision tree, i.e., step S2-3-6 is executed. If the defect type in the process parameters on the AOI side is bridging, then the process is diverted to the bridging-specific decision tree, i.e., step S2-3-2 is executed. If the defect type in the process parameters on the AOI side is cold solder joint, then the process is diverted to the cold solder joint-specific decision tree, i.e., step S2-3-3 is executed. If the defect type in the process parameters on the AOI side is equal to component offset, then the process is redirected to the component offset-specific decision tree, i.e., step S2-3-4 is executed. If the defect type in the AOI side process parameters is equal to "side-standing", then the process is redirected to the side-standing dedicated decision tree, i.e., step S2-3-5 is executed. S2-3-2, the Lianxi-specific decision tree includes: When the risk factor for excessive volume is greater than 1.0, the root cause is determined to be excessive solder paste. When the volume excess risk factor is <1.0, if the SPI topography is abnormal (collapse / expansion / stringing), the root cause is determined to be printing collapse bridging; otherwise, if the AOI offset risk factor is >1.0 and the offset direction points to adjacent pads, the root cause is determined to be mounting bridging. The determination of the offset direction pointing to adjacent pads is based on the adjacent pad rules preset in the PCB design file.

[0054] In this embodiment, SPI topography anomalies are determined by topography anomaly data output by the SPI device. The topography anomaly data output by the SPI device (such as collapse markers, outward expansion markers, edge diffusion index, etc.) includes specific detection fields. The specific data field definitions include collapse detection data field, outward expansion detection data field, and anomaly severity field.

[0055] The determination of whether the offset direction points to an adjacent pad involves using the pad coordinates and offset direction information from the AOI device. Based on the pre-defined adjacent pad rules in the PCB design file (e.g., through pad coordinate relationships or grid calculations), it is determined whether there is an anomaly in the adjacent pads caused by directional offset. If the offset direction points to an adjacent pad, it is added to the decision tree to determine whether it is a "mount bridging" or "mount offset".

[0056] The dedicated decision tree for tandem soldering outputs the root causes, confidence levels, and process recommendations for tandem soldering. The confidence level is the pre-set confidence level for each decision in the dedicated decision tree for tandem soldering, and the process recommendations are process recommendations provided by the professional knowledge base. Both the confidence level and the process recommendations are preset by experts in the decision tree.

[0057] S2-3-3, the decision tree for cold solder joints includes: When the risk coefficient for insufficient volume is less than 1.0, the root cause is determined to be tin deficiency. When the volume deficiency risk factor is greater than 1.0, if the reflow temperature does not meet the process requirements, the root cause is determined to be insufficient wetting; otherwise, if the placement pressure is abnormal, the root cause is determined to be poor contact. The determination of abnormal placement pressure includes: retrieving the offset output from the AOI equipment, calculating the absolute offset, and if the absolute offset is greater than a preset pressure abnormality offset threshold, then the placement pressure is determined to be abnormal.

[0058] The output of the decision tree for solder joint defects includes the root cause of the defect, confidence level, and process recommendations.

[0059] In this embodiment, the reflux temperature is obtained from the reflux process record in the MES system. The reflux temperature record data comes from the reflux process record in the MES system, which includes multiple sampling points of the temperature curve and the temperature timestamp.

[0060] If the reflow process record is empty or incomplete, an alarm can be triggered and an error log can be recorded. The decision tree for cold solder joints ignores the judgment of reflow temperature and outputs the error log. At this time, the judgment is only based on the insufficient volume risk factor and abnormal placement pressure.

[0061] Determining abnormal patch pressure involves: retrieving the offset and offset direction output from the AOI device, and calculating the absolute offset. If Δxy > the preset pressure abnormality offset threshold, it is determined that the patch pressure is abnormal. Here, Δxy is the absolute offset, Δx is the offset on the x-axis output by the AOI device, and Δy is the offset on the y-axis output by the AOI device.

[0062] S2-3-4, The dedicated decision tree for component offset includes: When the SPI offset risk coefficient is greater than 1.0, the root cause is determined to be printing tape deviation; When the SPI offset risk coefficient is less than 1.0, if the patch pressure is abnormal, the root cause is determined to be a nozzle problem; otherwise, if the MARK recognition is abnormal, the root cause is determined to be a vision calibration problem.

[0063] The dedicated decision tree output for component offset includes the root cause of the component offset, confidence level, and process recommendations.

[0064] The determination of MARK anomalies is achieved by identifying the error flags in the alignment mark recognition status field (such as the 'MarkError' error flag) output by the vision system of the AOI device.

[0065] S2-3-5, Side-standing Dedicated Decision Tree includes: When the AOI offset risk coefficient is greater than 1.0, the root cause is determined to be abnormal mounting posture. When the volume imbalance risk factor is <1.0, if the package is 01005 or 0201, the root cause is determined to be sensitive to the small package; otherwise, if the reflow temperature jumps abnormally, the root cause is determined to be the reflow disturbance. When the volume imbalance risk coefficient is greater than 1.0, the root cause is determined to be solder paste imbalance inducing lateral tilting. The side-mounted dedicated decision tree output includes the root cause of component side-mounted defects, confidence level, and process recommendations.

[0066] In this embodiment, abnormal temperature jumps in reflow refers to drastic fluctuations or unexpected abrupt changes in the temperature curve within a preset time window during the reflow soldering process, which exceed the allowable range of the process. This can be determined based on whether the first derivative of the temperature curve exceeds a threshold.

[0067] S2-3-6, Tombstone Defect Decision Tree includes: If the volume imbalance risk coefficient > 1.0, it is determined that the main cause is the imbalance of solder quantity at both ends; Read the reflow temperature data provided by the MES system. If the reflow temperature does not meet the process requirements and the imbalance of solder quantity at both ends has been determined to be the main cause, the root cause is the imbalance of solder quantity plus abnormal reflow temperature; If the reflow temperature does not meet the process requirements and the imbalance of solder quantity at both ends has not been determined to be the main cause, the root cause is insufficient solder paste or component misalignment. At this time: If the volume insufficiency risk coefficient < 1.0, it is determined that the main cause is insufficient solder paste; if the AOI misalignment risk coefficient > 1.0, it is determined that the tombstone is induced by component misalignment.

[0068] In this embodiment, the specific steps of the tombstone defect decision tree are as follows: S2-3-6-1. If the volume imbalance risk coefficient is >1.0, it is determined that the imbalance of tin quantity at both ends is the main factor, and S2-3-6-2 is executed; otherwise, the imbalance judgment is skipped and S2-3-6-2 is executed directly. S2-3-6-2. Read the reflow temperature data and determine whether it meets the process requirements (such as whether the peak temperature, holding time, and heating rate are within the preset window): If the reflow temperature does not meet the process requirements, and it has been determined in S2-3-6-1 that the imbalance of solder quantity at both ends is the main cause, then the root cause is determined to be the imbalance of solder quantity + abnormal reflow temperature; If the reflow temperature does not meet the process requirements, and it has not been determined in S2-3-6-1 that the imbalance of solder quantity at both ends is the main cause, then the root cause is determined to be insufficient solder paste or component misalignment, and proceed to S2-3-6-3; S2-3-6-3. When the root cause is insufficient solder paste or component misalignment, if the volume deficiency risk coefficient is <1.0, the root cause is determined to be insufficient solder paste; if the AOI misalignment risk coefficient is >1.0, the root cause is determined to be component misalignment-induced tombstoning; if the volume deficiency risk coefficient is >1.0 and the AOI misalignment risk coefficient is <1.0, the root cause is determined to be other reasons. In this embodiment, this situation is an unidentifiable abnormality, and the system will issue an alarm and suggest manual review.

[0069] In this embodiment, if the volume deficiency risk coefficient is <1.0 and the AOI offset risk coefficient is >1.0, then the one with higher confidence is selected according to priority; generally, the AOI offset risk coefficient has a higher priority than the volume deficiency risk coefficient.

[0070] The output of the tombstone defect decision tree includes the root cause of the tombstone defect, the confidence level, and process recommendations.

[0071] S3. The batch decision module retrieves and categorizes the decoupling analysis results by batch. It standardizes the root cause expressions in the decoupling analysis results to obtain root cause categories. It calculates the weight of each root cause category and performs normalization. Based on the proportion of each root cause category in the total weight, it determines the primary root cause and outputs the dominant root cause and secondary root cause.

[0072] In this embodiment, based on the decoupling analysis results obtained in steps S2-3, data from the same batch are extracted, categorized by batch, and the root cause expression in the decoupling analysis results is standardized and uniformly mapped to standard root cause categories to eliminate expression differences. Based on the occurrence frequency and average confidence of root cause categories, the weight of each root cause category is calculated, the weights of all root cause categories are normalized, the proportion of each root cause category in the total weight is calculated, and the main cause is determined according to the proportion. All root cause categories are sorted by weight, the dominant root cause and secondary root cause are selected, and the batch stability is judged.

[0073] When performing step S3, the specific steps include the following: S3-1. The batch decision module extracts data from the same batch based on the decoupling analysis results and categorizes them by batch.

[0074] In this embodiment, based on the decoupling analysis results obtained in steps S2-3, data from the same batch are extracted and categorized by batch. S3-2. Standardize the root cause expressions in the decoupling analysis results and uniformly map them to standard root cause categories; the root cause categories include printing offset, mounting offset, and uneven solder content.

[0075] This embodiment standardizes the root cause expression in the decoupling analysis results, mapping them uniformly to standard root cause categories to eliminate expression differences.

[0076] In this embodiment, a dictionary mapping is used to standardize the root cause expression. For example, "printing deviation" is mapped to "printing offset"; "mount bridging" is mapped to "mount offset"; and "solder paste imbalance / dominant solder amount imbalance at both ends" is mapped to "uneven solder amount", etc.

[0077] S3-3. Based on the frequency of occurrence and average confidence of the root cause category, calculate the weight of each root cause category. The specific formula is as follows: ; ; Among them, W i The weight of the root cause category; N i C represents the number of times the root cause category appears. i This represents the average confidence level of the root cause category.

[0078] S3-4. Normalize the weights of all root cause categories, calculate the proportion of each root cause category in the total weight, and determine the primary cause based on the proportion: ; Among them, P i W is the normalized weight of the i-th root cause;i W represents the original weight of the i-th root cause; j is the original weight of the j-th root cause; j is the index of all root causes traversed; m is the total number of root cause categories.

[0079] Method for determining the main cause: P i ≥80% is a systemic cause; 50% ≤ P i <80% is a mixed primary and secondary cause; P i <50% is considered discrete anomaly.

[0080] S3-5. Sort all root cause categories by weight, select and output the dominant root cause and the secondary root cause.

[0081] S4. Summarize the results of steps 2 and 3, combine them with the original SPI / AOI data to form a process report, generate standard process report fields based on the summarized data, including defect location, defect type, root cause result, confidence level, etc., query the preset root cause-suggestion mapping table based on the root cause type to generate corresponding process improvement suggestions; add the process improvement suggestions to the process report; output the process report.

[0082] This embodiment summarizes the decoupling analysis results output in step S2 and the primary and secondary root causes output in step S3, and combines them with the detection results of the SPI device and AOI device to form a process report. The process report includes standard process report fields such as defect location, defect type, root cause result, and confidence level; based on the root cause type, a preset root cause-suggestion mapping table is queried to generate corresponding process improvement suggestions; the process improvement suggestions are added to the process report; and the process report is output.

[0083] This invention presents a method for tracing solder paste printing and component placement misalignment based on multi-source detection data fusion. It addresses the technical challenge of integrating multi-source data from solder paste printing and component placement inspections, and adaptively adjusting judgment thresholds for different package types. This enables accurate tracing of the root causes of placement misalignment defects and batch-level root cause statistics. The invention uses a data caching module to correlate and fuse multi-source data from SPI, AOI, and MES, indexed by component reference numbers, constructing a data pool containing pad-level printing data and component-level placement data. Using package type as a constraint variable, preset thresholds are patched and converted into dimensionless risk coefficients. A defect decision tree is constructed using a unified threshold of 1.0 as the judgment benchmark, achieving decoupled analysis of the mechanisms of defects such as tombstone defects, solder bridging, cold solder joints, component misalignment, and side-standing defects. Furthermore, through batch aggregation, root cause standardization mapping, weighted calculation, and normalized proportion analysis, the invention automatically outputs dominant and secondary root causes, providing quantitative basis for closed-loop optimization of the production process.

Claims

1. A method for tracing solder paste printing and mounting offset based on multi-source detection data fusion, characterized in that: Includes the following steps: S1. The data caching module receives, aggregates, and manages multi-source data output from SPI devices, AOI devices, and the MES system, and constructs an associative data pool. The multi-source data includes the detection results output by the SPI devices and AOI devices, as well as the engineering data output by the MES system. Identify the component package type for each reference number based on engineering data, and establish the association between the reference number and the package type; S2. The rule engine module extracts key feature information from the detection results output by the SPI device and AOI device from the data pool, patches and corrects the preset threshold according to the packaging type, makes a second judgment on the detection results, constructs a defect decision tree, and outputs the decoupling analysis results. S3. The batch decision module retrieves and categorizes the decoupling analysis results by batch. It standardizes the root cause expression in the decoupling analysis results to obtain root cause categories. It calculates the weight of each root cause category and performs normalization. It determines the primary root cause based on the proportion of each root cause category in the total weight and outputs the dominant root cause and secondary root cause. S4. Summarize the results of steps S2 and S3, generate a process report and output it for production closed-loop adjustment.

2. The solder paste printing and mounting offset tracing method based on multi-source detection data fusion as described in claim 1, characterized in that: When performing step S1, the specific steps include the following: S1-1, The data cache module receives the inspection results of the solder paste on the printed pads from the SPI device, the inspection results of the components after placement or reflow from the AOI device, and the engineering data from the MES system; the engineering data includes PCB design files; Based on the pad coordinates and component center coordinates in the PCB design file, bind the detection results of the SPI device, the pad and the component reference number to which the pad belongs, and bind the detection results of the AOI device to the reference number; Using the tag number as the primary key, SPI detection data and AOI detection data of the same component are merged to construct a data pool; S1-2 The data cache module retrieves the PCB design file, identifies the component package type corresponding to each reference number, and generates a mapping table between reference numbers and package types.

3. The solder paste printing and mounting offset tracing method based on multi-source detection data fusion as described in claim 1, characterized in that: When performing step S2, the specific steps include the following: S2-1. The rule engine module extracts the SPI-side process parameters and AOI-side process parameters corresponding to the same bit number from the data pool; it then corrects each process parameter based on the preset packaging patch rules to obtain the corrected packaging allowable threshold. The SPI-side process parameters include solder paste volume, area, height, offset, and offset direction; the AOI-side process parameters include defect type, component offset, and orientation anomaly information. S2-2. Compare and calculate the original process information with the corrected packaging allowable threshold, and convert all process parameters into dimensionless risk coefficients. S2-3. Using the defect type field in the process parameters on the AOI side as the first decision entry, different defect instances are assigned to the corresponding defect decision trees. Based on the dimensionless risk coefficient and using a fixed threshold as a unified judgment benchmark, step-by-step reasoning is performed to obtain the decoupling analysis results.

4. The solder paste printing and mounting offset tracing method based on multi-source detection data fusion as described in claim 1, characterized in that: When performing step S3, the specific steps include the following: S3-1. Based on the decoupling analysis results, the batch decision module extracts data from the same batch and categorizes them by batch. S3-2. Standardize the root cause expressions in the decoupling analysis results and uniformly map them to standard root cause categories; the root cause categories include printing offset, mounting offset, and uneven solder content. S3-3. Based on the frequency of occurrence and average confidence of the root cause category, calculate the weight of each root cause category. The specific formula is as follows: ； ； Among them, W i The weight of the root cause category; N i C represents the number of times the root cause category appears. i The average confidence level of the root cause category; k is the summation index, specifically representing the k-th sample; C ik This represents the confidence level of the k-th sample in the i-th root cause; S3-4. Normalize the weights of all root cause categories, calculate the proportion of each root cause category in the total weight, and determine the primary cause based on the proportion: ； Among them, P i W is the normalized weight of the i-th root cause; i W represents the original weight of the i-th root cause; j is the original weight of the j-th root cause; j is the index of all root causes traversed; m is the total number of root cause categories; Method for determining the main cause: P i ≥80% is a systemic cause; 50% ≤ P i <80% is a mixed primary and secondary cause; P i <50% is considered discrete anomaly; S3-5. Sort all root cause categories by weight, select and output the dominant root cause and the secondary root cause.

5. The solder paste printing and mounting offset tracing method based on multi-source detection data fusion as described in claim 1, characterized in that: When performing step S4, the decoupling analysis results output from step S2 and the primary and secondary root causes output from step S3 are summarized, and the detection results of the SPI device and AOI device are combined to form a process report. The process report includes standard process report fields such as defect location, defect type, root cause results, and confidence level; Based on the root cause type, a pre-defined root cause-suggestion mapping table is queried to generate corresponding process improvement suggestions; Add the proposed process improvement suggestions to the process report; Output the process report.

6. The solder paste printing and mounting offset tracing method based on multi-source detection data fusion as described in claim 2, characterized in that: When performing step S1-1, for the solder paste inspection results of the printed pads output by the SPI device, the coordinates of each inspection point are matched with the corresponding pad according to the pad coordinates in the PCB design file. After successful matching, the component reference number and pad number to which the pad belongs are bound. For the component inspection results output by the AOI device, the inspection results of each component are matched with the corresponding component reference number according to the component center coordinates in the PCB design file. When a component has multiple pads, a separate record is generated for each pad in the data pool. Each record contains the same AOI inspection data and its own SPI inspection data, and is distinguished by the pad number field.

7. The solder paste printing and mounting offset tracing method based on multi-source detection data fusion as described in claim 3, characterized in that: When performing step S2-2, the dimensionless risk coefficients include offset risk coefficient, insufficient volume risk coefficient, excessive volume risk coefficient, and volume imbalance risk coefficient. The calculation methods for each risk coefficient are as follows: The offset risk coefficient includes the SPI offset risk coefficient and the AOI offset risk coefficient: SPI offset risk factor = Offset output by SPI device / Maximum allowable offset for this package; AOI offset risk factor = Offset output by AOI device / Maximum allowable offset for this package; Insufficient volume risk factor = Actual volume ratio / Minimum allowable volume ratio for this package; Excess volume risk factor = Measured volume ratio / Maximum allowable volume ratio of this package; Volume imbalance risk factor = absolute value of volume difference between the two ends / (average value of volume at both ends × allowable volume imbalance ratio of this package); All dimensionless risk coefficients are normalized to a pre-defined risk benchmark threshold.

8. The solder paste printing and mounting offset tracing method based on multi-source detection data fusion as described in claim 7, characterized in that: When performing steps S2-3, the defect decision tree includes a dedicated decision tree for bridging, a dedicated decision tree for cold solder joints, a dedicated decision tree for component offset, and a tombstone defect decision tree. The decision logic of the PCB decision tree for PCBs is as follows: when the risk coefficient of excessive volume is greater than 1.0, the root cause is determined to be excessive solder paste. When the risk factor for excessive volume is less than 1.0, if the SPI morphology is abnormal, the root cause is determined to be printing collapse bridging. Otherwise, if the AOI offset risk coefficient is greater than 1.0 and the offset direction points to the adjacent pad, the root cause is determined to be a mounting bridge; wherein, the determination that the offset direction points to the adjacent pad is based on the adjacent pad rules preset in the PCB design file. The decision tree for cold solder joints uses the following logic: when the volume deficiency risk coefficient is <1.0, the root cause is determined to be insufficient solder; when the volume deficiency risk coefficient is >1.0, if the reflow temperature obtained from the MES system does not meet the process requirements, the root cause is determined to be insufficient wetting; otherwise, if the placement pressure is abnormal, the root cause is determined to be poor contact. The determination of abnormal placement pressure includes: retrieving the offset output from the AOI equipment, calculating the absolute offset, and if the absolute offset is greater than the preset pressure abnormality offset threshold, it is determined to be abnormal placement pressure. The decision logic of the component offset-specific decision tree is as follows: when the SPI offset risk coefficient is >1.0, the root cause is determined to be the printing tape misalignment; when the SPI offset risk coefficient is <1.0, if the placement pressure is abnormal, the root cause is determined to be the nozzle problem; otherwise, if the MARK recognition is abnormal, the root cause is determined to be the vision calibration problem; whereby the MARK recognition abnormality is achieved by recognizing the error flag in the alignment mark recognition status field output by the AOI equipment vision system. The decision logic of the tombstone defect decision tree is as follows: If the volume imbalance risk coefficient is >1.0, it is determined that the main cause is the imbalance of solder volume at both ends; read the reflow temperature data provided by the MES system. If the reflow temperature does not meet the process requirements and the imbalance of solder volume at both ends has been determined to be the main cause, the root cause is determined to be the imbalance of solder volume plus abnormal reflow temperature; if the reflow temperature does not meet the process requirements and the imbalance of solder volume at both ends has not been determined to be the main cause, the root cause is determined to be insufficient solder paste or component misalignment. At this time: if the volume insufficiency risk coefficient is <1.0, it is determined to be the main cause of insufficient solder paste; if the AOI misalignment risk coefficient is >1.0, it is determined to be the tombstone caused by component misalignment.

Images