Artificial intelligence-based cyanide-free process circuit board quality traceability method and system

Abstract

The application discloses a cyanide-free process circuit board quality tracing method and system based on artificial intelligence, and particularly relates to the field of circuit board quality tracing, and comprises the following steps: writing a board part identifier, a hanger identifier and a process serial number into a circuit board entering a cyanide-free process production line, extracting fixed pixel points and fixed electrical measurement points in a side strip area, and corresponding to key processes to output an observation index table; the application continuously collects fixed observation positions of the same circuit board based on edge computing at the outlets of the key processes, forms process records, and then performs continuous comparison, abnormal connection determination and same-hanger merging early warning on the front and rear process records according to the same observation positions, so as to solve the problem that it is difficult to accurately trace the real generation position of defects of the cyanide-free process circuit board.

Description

Technical Field

[0001] This invention relates to the field of circuit board quality traceability technology, and more specifically, to a method and system for tracing the quality of cyanide-free circuit boards based on artificial intelligence. Background Technology

[0002] In the production of cyanide-free circuit boards, existing quality control usually focuses on early detection of defects and quick backtracking of related processes after discovery. Common handling methods include collecting board images, electrical test results, bath parameters, equipment operating status and batch history at the pretreatment, electroplating, cleaning, posttreatment and final inspection locations, and then the edge computing unit on the production line side completes data aggregation and preliminary analysis, and traces the source of the anomaly in combination with the defect detection results. Taking a continuous rack plating production line as an example, under the conditions that multiple plates are carried on the same rack at the same time, the transfer between processes is continuous, the replenishment of the plating solution and the switching of processes are carried out alternately, the graphic load of different plates varies greatly, and the processing time of a single station is limited by a predetermined cycle time, the local anomalies in the preceding process often do not manifest as obvious defects on the spot, but only become apparent after subsequent cleaning, reactivation, redeposition and final inspection. This will cause a situation that occurs repeatedly and can be verified on the spot: the final inspection can identify the abnormal plate, but when checking back, the scope can often only be narrowed down to a few adjacent processes, and it is difficult to further confirm where the defect first started to form. Defects with similar appearance in the same batch may also correspond to different preceding states. The reason is that most of the existing records reflect the test results and final appearance of each station at that time, and lack a formation basis that can be continuously retained with the plate during the process flow. Therefore, the subsequent analysis is more at the level of post-event comparison and related judgment, and it is difficult to distinguish the defect appearance location from the actual defect generation location. Therefore, the technical problem to be solved by this application is: how to establish a quality formation basis for cyanide-free process circuit boards under edge computing conditions that can be continuously retained and traced back for verification across processes of the board, so as to accurately determine the actual location of defect generation. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for quality traceability of cyanide-free process circuit boards based on artificial intelligence. By continuously collecting data at fixed observation locations of the same circuit board at the exit of each key process based on edge computing and forming process records, and then performing continuous comparison, abnormal acceptance judgment, and same fixture merging warning on the records of the preceding and following processes according to the same observation locations, the problems mentioned in the background art can be solved.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for quality traceability of cyanide-free process circuit boards based on artificial intelligence, comprising: S1. Write board identification, fixture identification and process number to the circuit board entering the cyanide-free process production line. Extract fixed pixel points and fixed electrical test points in the edge strip area and correspond them with key processes. Output the observation index table. S2. The edge computing nodes at the exit of each key process collect the plate surface image, local electrical measurement results, tank liquid parameters and equipment parameters according to the observation index table, input the plate surface image into the defect recognition network, obtain the defect boundary, defect category and defect location, and write it into the process record according to the plate identification, process number and observation position and output it. S3. Perform coordinate matching on the process records of adjacent processes under the same board part identification according to the same observation position, calculate the overlapping area of ​​defect boundaries and the difference of local electrical measurement values, and classify new abnormal records, continued abnormal records and aggravated abnormal records according to the defect category comparison results, and output the abnormal sequence. S4. Scan the abnormal sequence by process number and take the first newly generated abnormal record at the same observation position as the candidate generated record. When there are similar continuous abnormal records or aggravated abnormal records in the two key processes after the candidate generated record, determine the candidate generated record as the actual defect generated record and output it. S5. Merge the actual defect generation record with the process records of other circuit boards under the same fixture identifier at the same process number and the same observation position; when there are two or more actual defect generation records with the same defect category and the same observation position, output the batch warning record and the corresponding process record.

[0005] In a preferred embodiment, S1 includes: S1-1. Read the edge strip area image, edge strip area layout data and contact position data of each key process of the circuit board. Establish a unified coordinate system with the starting angle of the edge strip area as the origin and the transmission direction and board width direction. Divide the edge strip area coordinate grid with the same spacing and output the edge strip area coordinate grid set. S1-2. Calculate the minimum distance between the grid center and the line boundary, hole boundary and plate boundary of the coordinate grid set of the edge strip area grid cell by cell, delete the coordinate grid that coincides with the contact position of any key process, determine the grid center of the remaining coordinate grid as a fixed pixel point, and select a unique fixed electrical measurement point in the coordinate grid where each fixed pixel point is located, and output the observation point set. S1-3. Number the fixed pixel points and fixed electrical measurement points in the observation point set according to the same coordinate grid and correspond them to the key processes, and associate them with the board identifier, fixture identifier and process sequence number to generate an observation index table.

[0006] In a preferred embodiment, S2 includes: S2-1. Read the plate identifier, process number, fixed pixel point and fixed electrical test point in the observation index table, collect the whole plate image, the local electrical test results corresponding to the fixed electrical test point, the tank liquid parameters and equipment parameters at the corresponding key process exit, and merge them according to the plate identifier and process number to output the acquisition group; S2-2. Extract local patches from the whole plate image in the acquisition group according to fixed pixel coordinates, write the center coordinates of each local patch and the coordinates of the corresponding fixed electrical measurement point as the same observation position, and match the local patch with the local electrical measurement result according to the same observation position, and output the observation group.

[0007] In a preferred embodiment, S2 further includes: S2-3. Input the local patches in the observation group into the defect recognition network, extract the defect boundaries in the local patches, determine the defect location according to the coordinate difference between the center of the outer rectangle of the defect boundary and the observation location, determine the defect category according to the image features of the area covered by the defect boundary, and output the recognition group. S2-4. Write the defect boundaries, defect categories, and defect locations in the identification group, along with the local electrical test results, tank parameters, and equipment parameters corresponding to the same observation location, into the process record according to the plate identification, process number, and observation location, and then output the record.

[0008] In a preferred embodiment, S3 includes: S3-1. Read two adjacent process records under the same board identifier according to the process sequence number, extract the defect boundary, defect category and local electrical measurement results corresponding to the same observation position in the two process records, and form a position comparison group according to the order of the previous process record before the next process record, and output the position comparison group. S3-2. For the defect boundary of the next process recorded in the position comparison group, perform translation according to the coordinate difference between the defect location and the defect location of the previous process recorded, so that the defect boundary of the next process recorded after translation falls into the same observation position coordinate system as the defect boundary of the previous process recorded. Calculate the area of ​​the overlapping area of ​​the two as the defect boundary overlap area, and subtract the local electrical measurement result of the previous process recorded from the local electrical measurement result of the next process recorded to obtain the difference of local electrical measurement value, and output the comparison result group.

[0009] In a preferred embodiment, S3 further includes: S3-3. The comparison results group is performed by comparing the defect category of the previous process record with the defect category of the next process record. The next process record with the empty defect boundary of the previous process record and the non-empty defect boundary of the next process record is recorded as a new abnormal record. The next process record with the non-empty defect boundary of the previous process record, the non-empty defect boundary of the next process record, the same defect category, and the non-zero overlap area of ​​the defect boundary is recorded as a continued abnormal record. The classification result group is output. S3-4. For the continuing abnormal records in the classification result group, calculate the defect boundary area of ​​the previous process record and the defect boundary area of ​​the next process record respectively. Rewrite the continuing abnormal records where the defect boundary area of ​​the next process record is greater than the defect boundary area of ​​the previous process record and the difference in local electrical measurement values ​​is not zero as aggravated abnormal records. Arrange the new abnormal records, continuing abnormal records and aggravated abnormal records according to the process sequence number and output the abnormal sequence.

[0010] In a preferred embodiment, S4 includes: S4-1. Read the abnormal records in the abnormal sequence that are arranged by process number under the same board identifier and the same observation position. Extract the newly generated abnormal record with the first process number, write the process number, defect type, defect boundary, defect location and local electrical test results to form a candidate generated record and output it. S4-2. After reading the candidate generated record, read the abnormal records corresponding to the first key process and the second key process. Calculate the defect boundary overlap area between the two abnormal records and the candidate generated record. Compare the defect categories of the two abnormal records and the candidate generated record. Write the abnormal records with the same defect category and non-zero defect boundary overlap area as renewal records. Write the abnormal records with different defect categories or zero defect boundary overlap area as conflict records. Output the first process judgment result and the second process judgment result corresponding to the candidate generated record. S4-3. Write the candidate generated records whose first and second process determination results are both renewal records into the pass-through flag. Write the candidate generated records whose first and second process determination results are both renewal records into the review flag. Write the candidate generated records whose first and second process determination results are both conflict records into the review flag. Write the candidate generated records whose first and second process determination results are both conflict records into the rollback flag. Output the gating result.

[0011] In a preferred embodiment, S4 further includes: S4-4. Read the candidate generated records marked with the verification mark and the process records corresponding to the same process number and the same observation position under the same fixture identifier. Calculate the defect boundary overlap area between the process records and the candidate generated records and compare the defect categories. Write the process records with the same defect category and non-zero defect boundary overlap area as supporting evidence records. When supporting evidence records exist, rewrite the candidate generated records as passing records and output them. When no supporting evidence records exist, rewrite the candidate generated records as rollback records and output them. S4-5. Read the candidate generation record and pass record corresponding to the pass mark, write the candidate generation record and pass record corresponding to the pass mark as the actual defect generation record and output them. Delete the candidate generation record and backtracking record corresponding to the backtracking mark. Then extract the new abnormal record with the same board identifier and the same observation position, whose process number is greater than the deleted candidate generation record and whose process number is the first, to form a new candidate generation record and re-execute the subsequent judgment. When there is no new candidate generation record, output the scan end mark of the current observation position.

[0012] In a preferred embodiment, S5 includes: S5-1. Read the actual defect generation record and the process record of each circuit board under the same fixture identifier, extract the defect category, defect boundary, local electrical test result, tank liquid parameter and equipment parameter corresponding to the same process number and the same observation position, and form a merge group according to the fixture identifier, process number and observation position, and output the merge group; S5-2. Perform defect category comparison and observation location comparison on each defect real generation record in each merge group, write defect real generation records with the same defect category and the same observation location into the same batch record group, count the number of defect real generation records in each batch record group, and output the batch record group. S5-3. Read the corresponding process records for each batch record group, and write the fixture identifier, process number, observation position, defect category, corresponding board identifier and corresponding defect actual generation record into the batch record group with two or more defect actual generation records to form a batch warning record, and output the batch warning record and the corresponding process record.

[0013] In a preferred embodiment, the AI-based cyanide-free process circuit board quality traceability system includes: The index building module writes board identifiers, fixture identifiers and process numbers to the circuit boards entering the cyanide-free process production line, extracts fixed pixel points and fixed electrical test points in the edge strip area, and associates them with key processes to output an observation index table. The data acquisition module collects plate images, local electrical measurement results, tank liquid parameters and equipment parameters through edge computing nodes at the exit of each key process according to the observation index table. The plate images are input into the defect recognition network to obtain the defect boundary, defect category and defect location. The data is then written into the process record and output according to the plate identification, process number and observation location. The anomaly determination module performs coordinate matching on the process records of adjacent processes under the same board part identification according to the same observation position, calculates the overlapping area of ​​the defect boundary and the difference of local electrical measurement values, and classifies new anomaly records, continued anomaly records and aggravated anomaly records according to the defect category comparison results, and outputs the anomaly sequence. The traceability and judgment module scans the abnormal sequence by process number and takes the first newly generated abnormal record at the same observation position as a candidate generated record. When there are similar continuing abnormal records or aggravated abnormal records in the two key processes after the candidate generated record, the candidate generated record is determined to be the actual defect generated record and output. The batch warning module merges the actual defect generation records with the process records of other circuit boards under the same fixture identification at the same process number and the same observation position. When there are two or more actual defect generation records with the same defect category and the same observation position, it outputs the batch warning record and the corresponding process record.

[0014] The technical effects and advantages of this invention are as follows: 1. By establishing an observation index table and continuously generating process records at the exit of each key process, the abnormalities of the same board can be continuously carried over across processes along the same observation position, thereby distinguishing the defect manifestation position from the actual defect generation position, which relatively improves the problem that it is difficult to accurately determine the actual generation position of defects on cyanide-free process circuit boards. 2. By writing the plate image, local electrical test results, tank liquid parameters and equipment parameters into the same process record according to the plate identification, process number and observation position, the image results, electrical test results and process status are closed and corresponded in the same record, reducing secondary splicing errors and record chain breaks in subsequent traceability; 3. By performing coordinate matching, defect boundary overlap area calculation, and local electrical measurement difference calculation on the records of adjacent processes, and classifying new abnormal records, continuing abnormal records, and aggravating abnormal records accordingly, the abnormal evolution process can be continuously expressed, thereby improving the temporal integrity of abnormal source judgment. 4. By using the first newly generated abnormal record as a candidate generated record, and combining the verification results of the next two key processes and the corroborating results of the same fixture process records to perform traceability judgment, short-term abnormalities, isolated abnormalities and continuous abnormalities can be distinguished, which helps to suppress the situation of misjudging occasional abnormalities as the actual generation location. 5. By merging the actual generated records of defects with the same process number and the same observation position under the same fixture identification, and writing the records that meet the quantity conditions into batch warning records and outputting the corresponding process records at the same time, the traceability of single board is further extended to batch verification of the same fixture, thereby relatively improving the timeliness of batch anomaly detection and back-inspection. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method steps of the present invention.

[0016] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0017] 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.

[0018] Refer to the instruction manual appendix Figure 1-2 The present invention provides an artificial intelligence-based method for quality traceability of cyanide-free process circuit boards, comprising: S1. Write board identification, fixture identification and process number to the circuit board entering the cyanide-free process production line. Extract fixed pixel points and fixed electrical test points in the edge strip area and correspond them with key processes. Output the observation index table. Before the circuit board enters the cyanide-free production line, it is necessary to fix the observation positions that will be used continuously across processes. This will ensure that the board surface images and local electrical test results collected at the exits of different key processes on the same circuit board can be placed on the same set of position indexes. This will allow for the continuous formation of process records and the execution of cross-process comparisons and traceability. To this end, the edge strip area is first used as the layout area for observation positions. Then, the edge strip area images, edge strip area layout data, and contact position data of each key process are unified under the same coordinate reference. Coordinate grids are divided within this coordinate reference. Then, positions that will be affected by circuits, holes, board edges, and process contacts are eliminated. Finally, fixed pixel points and fixed electrical test points are determined simultaneously within the remaining coordinate grids, and both are written into the observation index table along with the key processes and basic board identification. The implementation process includes the following steps: First, the edge strip area image, edge strip area layout data, and contact position data of each key process are read. The edge strip area image reflects the actual appearance range of the edge strip area, the edge strip area layout data gives the position of the line boundary, hole boundary, and board boundary, and the contact position data gives the position of the area that has physical or conductive contact with the board in each key process. After reading, the edge strip area image and edge strip area layout data are registered to the same coordinate system with the starting angle of the edge strip area as the origin, the transmission direction of the board in the production line as the first coordinate direction, and the board width direction perpendicular to the transmission direction as the second coordinate direction. Then, the edge strip area is divided along the first and second coordinate directions at the same interval to form the edge strip area coordinate grid set. The same interval here uses the same side length value, which is jointly limited by the minimum line spacing, the minimum hole edge distance, and the fixed electrical test point layout interval in the edge strip area, and remains unchanged in one division to ensure that the subsequent coordinate grid numbering, grid center calculation, and observation position mapping are all performed based on the same scale. After obtaining the coordinate grid set of the edge strip area, the minimum distance from the grid center to the line boundary, the minimum distance to the hole boundary, and the minimum distance to the plate boundary are calculated for each grid. When any minimum distance falls within the line occupation area, hole occupation area, or outside the plate edge, the grid is directly deleted, ensuring that the remaining grids are all within the observable area. Subsequently, the contact position data of each key process are mapped to the same coordinate system, and each grid is checked to see if it coincides with the contact position. When a grid coincides with the contact position of any key process, it is deleted, ensuring that the remaining grids are not affected by contact interference from key processes. For each grid remaining after deletion... The grid is marked, and the center of the grid is directly determined as a fixed pixel. Then, standard electrical measurement point data is read within the grid range. Electrical measurement points falling within the grid are selected as candidate electrical measurement points, and the distance from each candidate electrical measurement point to the grid center is calculated. The candidate electrical measurement point with the highest distance is selected as the fixed electrical measurement point. When two candidate electrical measurement points have the same distance from the grid center, the candidate electrical measurement point with the highest number is selected according to the layout number. When there is no electrical measurement point in the grid, the grid corresponding to the fixed pixel is deleted and not added to the subsequent observation point set. After the above processing, each retained grid corresponds to only one fixed pixel and one fixed electrical measurement point, thus forming the final observation point set. After obtaining the observation point set, each group of fixed pixel points and fixed electrical measurement points in the observation point set is first assigned the same coordinate grid number, and this coordinate grid number is written as the observation position, so that the fixed pixel points and fixed electrical measurement points share the same position index in subsequent processes. Then, each observation position is matched with a key process one by one, so that the same observation position maintains the same number and the same coordinate caliber in each key process. Then, the board identification, fixture identification, and process sequence number are written into the corresponding observation position to form an observation index table composed of board identification, fixture identification, process sequence number, observation position, fixed pixel points, and fixed electrical measurement points. In this way, the board surface image collected at the exit of the key process can be located by fixed pixel points, and the local electrical measurement results can be located by fixed electrical measurement points. The two can be directly matched to the same process record through the same observation position, thereby ensuring that the writing of subsequent process records, the formation of abnormal sequences, and the judgment of the actual generation of defect records are all based on a unified position. Furthermore, by first eliminating the coordinate grids corresponding to the line boundary, hole boundary, board boundary, and contact position, and then determining the fixed pixel point and fixed electrical measurement point, the subsequent board surface image acquisition area and local electrical measurement position can continuously fall within a stable area, reducing the interference of the key process's own actions on the observation results. In practical applications: For a circuit board entering the cyanide-free process production line, first read its edge strip area image and edge strip area layout data, then read the contact position data corresponding to pre-processing, electroplating, and post-processing. After establishing a unified coordinate system based on the starting angle of the edge strip area, divide the edge strip area into coordinate grids. Then delete the coordinate grids that coincide with the circuit area, hole area, outer edge area of ​​the board, and contact position. Extract the grid center from the remaining coordinate grids as fixed pixel points, and select fixed electrical test points in each coordinate grid according to the rule of being closest to the grid center. Finally, write the fixed pixel points, fixed electrical test points, board identification, fixture identification, and process sequence number of each key process into the same observation index table for direct use by the edge computing nodes at the exit of subsequent key processes.

[0019] S2. The edge computing nodes at the exit of each key process collect the plate surface image, local electrical measurement results, tank liquid parameters and equipment parameters according to the observation index table, input the plate surface image into the defect recognition network, obtain the defect boundary, defect category and defect location, and write it into the process record according to the plate identification, process number and observation position and output it. Unified data collection and writing at the exit of key processes is the foundation for subsequent process record generation, anomaly sequence construction, and defect determination. Instead of separately saving the entire board image, local electrical test results, and process parameters, this process uses an observation index table as a reference, mapping fixed pixels, fixed electrical test points, board identifiers, and process numbers to the same key process exit. Then, the collected entire board image, local electrical test results, bath parameters, and equipment parameters are aggregated into a single data set. Subsequently, local patches are extracted from the entire board image, centered on fixed pixels, and the local electrical test results corresponding to the fixed electrical test points are unified to the same observation position. Defect identification is then performed on the local patches to obtain defect boundaries, defect locations, and defect categories. Finally, the identification results, along with the local electrical test results, bath parameters, and equipment parameters, are written into the process record according to the board identifier, process number, and observation position. This process ensures that image information, electrical test information, and process information at the same observation position are all included in the same process record, eliminating the need for secondary stitching during subsequent cross-process comparisons. The implementation process includes the following steps: First, read the board identifier, process number, fixed pixel point, and fixed electrical test point from the observation index table. Then, collect data on the circuit board corresponding to the board identifier at the exit of the corresponding key process. The whole board image is acquired by the visual acquisition unit at the exit, the local electrical test results are acquired by the electrical test unit connected to the fixed electrical test point, the tank liquid parameters are read by the online detection unit of the tank corresponding to the key process, and the equipment parameters are read by the equipment control unit corresponding to the key process. The local electrical test results are directly taken from the measured value of the fixed electrical test point at the current process exit. The tank liquid parameters are read according to the current process exit time, including temperature, conductivity, concentration, and liquid level. The equipment parameters are read according to the current process exit time, including transmission speed, current value, and residence time. After reading, the whole board image, local electrical test results, tank liquid parameters, and equipment parameters under the same board identifier and process number are grouped into the same acquisition group to ensure that subsequent screenshotting, recognition, and writing are all based on data from the same process time. Subsequently, local patches are cropped from the full-board image in the acquisition group according to fixed pixel coordinates. During cropping, the fixed pixel is used as the center, and the same length is taken on both sides along the transmission direction and the board width direction based on the side length of the coordinate grid where the fixed pixel is located, forming a rectangular area covering the current coordinate grid and its adjacent coordinate grids. This rectangular area is written as a local patch. After the local patch is generated, the coordinate grid number where the center of the local patch is located is written as the observation position. At the same time, the fixed electrical measurement point falling into the coordinate grid number is read, and the local electrical measurement result corresponding to the fixed electrical measurement point is written into the same observation position, so that the local patch and the local electrical measurement result are matched one-to-one through the same observation position. After all the matching is completed, the local patches and local electrical measurement results corresponding to each observation position under the same board identifier and the same process number are combined into an observation group. Here, the observation position still uses the coordinate grid number in the observation index table, without changing the previous definition, so as to ensure that the observation position in the subsequent process record is completely consistent with the previous index. After obtaining the observation group, the local patches corresponding to each observation position are input into the defect recognition network. Regions in the local patches that differ from the background area in grayscale distribution, texture direction, or edge closure are segmented, and closed defect boundaries are extracted. The circumscribed rectangle of the defect boundary is then calculated, and the center coordinates of the observation position are subtracted from the center coordinates of the circumscribed rectangle to obtain the defect location. The defect location is recorded using the coordinate difference relative to the center of the observation position, serving as the basis for subsequent cross-process translation and alignment. After determining the defect location, grayscale distribution, texture density, edge direction, and color difference distribution are extracted from the area covered by the defect boundary, and written into the defect category according to the category number output by the defect recognition network. This allows defects of the same type to be directly compared for consistency using the category number during subsequent process record comparisons. After all local patches are processed, an recognition group is formed, consisting of defect boundaries, defect locations, and defect categories. If a local patch does not have a closed region segmented, the defect boundary corresponding to that local patch is written as an empty boundary list, the defect category is written as an empty category marker, and the defect location is written as zero-offset coordinates for subsequent new anomaly record judgment. Furthermore, the defect boundaries, defect categories, and defect locations in the identification group, along with the corresponding local electrical measurement results, bath parameters, and equipment parameters at the same observation location, are written into the process record according to the board identification, process number, and observation location. The process record includes at least the board identification, fixture identification, process number, observation location, defect boundary, defect category, defect location, local electrical measurement results, bath parameters, and equipment parameters. When writing, the board identification, fixture identification, process number, and observation location are written first, followed by the defect boundary, defect category, and defect location, and finally the local electrical measurement results, bath parameters, and equipment parameters. This merges the image results, electrical measurement results, and process parameters under the same process and observation location into a single process record. After all process records are written, they are directly output for subsequent calculation of the overlapping area of ​​defect boundaries between adjacent processes, calculation of local electrical measurement differences, and classification of abnormal records. Through the above processing, the whole plate image, local electrical test results, tank liquid parameters and equipment parameters are simultaneously acquired at the exit of the same key process. Local blocks are uniformly cropped based on fixed pixels. Defect boundaries, defect locations and defect types are uniformly identified under the same observation position. The final process record has complete field sources and consistent location indexes, providing direct input for the subsequent formation of abnormal sequences and the determination of the actual generation of defect records. In practical applications: When a circuit board is transferred to the exit of the critical cyanide-free electroplating process, the edge computing node first reads the fixed pixel points and fixed electrical test points corresponding to the circuit board according to the observation index table. Then, it simultaneously acquires the whole board image of the circuit board, the local electrical test results at the fixed electrical test points, and the temperature, conductivity, concentration, liquid level, equipment conveying speed, current value, and residence time of the cyanide-free electroplating tank. Subsequently, it extracts local patches from the whole board image according to the coordinates of the fixed pixel points, and writes the local patches and the local electrical test results under the same coordinate grid number as the same observation position. Then, it inputs the local patches into the defect recognition network to obtain the defect boundary, defect location, and defect category. Finally, it writes these recognition results, along with the local electrical test results, tank liquid parameters, and equipment parameters, into the process record of the corresponding observation position and outputs them for direct comparison between subsequent adjacent critical processes.

[0020] S3. Perform coordinate matching on the process records of adjacent processes under the same board part identification according to the same observation position, calculate the overlapping area of ​​defect boundaries and the difference of local electrical measurement values, and classify new abnormal records, continued abnormal records and aggravated abnormal records according to the defect category comparison results, and output the abnormal sequence. When forming an anomaly sequence, it is necessary to continuously compare the process records of the same board under adjacent key processes within the same location caliber, so that the anomalies appearing in the later process can be clearly distinguished as new, continued, or aggravated. The focus here is not on looking at the identification results of a single process, but on matching the previous process record with the next process record one by one under the same observation position, and then performing coherent calculations on the defect boundary, defect category, and local electrical measurement results, so as to obtain an anomaly sequence that can be directly entered into subsequent traceability judgment. To this end, first read the two process records that are adjacent to each other under the same board identifier according to the process sequence number to form a location comparison group; then translate the defect boundary in the later process record to the observation position coordinate system of the previous process record, and calculate the defect boundary overlap area and the difference of local electrical measurement values; then classify the later process record according to whether the defect boundary is empty, whether the defect category is consistent, and whether the defect boundary overlap area is zero; finally, further determine whether the continued anomaly record has formed an aggravated anomaly record, and write the anomaly sequence according to the process sequence number. The implementation process includes the following steps: First, read the two adjacent process records under the same board identifier, where adjacent means two consecutive records in the key process sequence that have both generated process records. If a key process has not generated a process record, it is not considered an adjacent process for comparison. Instead, the next generated process record is read, ensuring that the position comparison is always based on two existing process records. After reading, extract the defect boundary, defect category, and local electrical measurement results corresponding to the same observation position from the two process records, maintaining the writing order of the previous process record first and the next process record last. These two are then grouped into a position comparison group. Here, the same observation position is determined by the same coordinate grid number written in the process record. As long as the observation position number of the previous process record and the next process record are consistent, they are considered to be the same observation position. Different observation positions under the same board identifier are established in their own position comparison groups, without cross-mixing. After obtaining the position comparison group, the defect boundary of the next process record in the position comparison group is translated. Specifically, the defect positions in the previous process record and the next process record are read first. A translation vector is formed by subtracting the defect position of the previous process record from the defect position of the next process record. Then, this translation vector is subtracted from all boundary points in the defect boundary of the next process record point by point, so that the translated defect boundary of the next process record and the defect boundary of the previous process record fall into the same observation position coordinate system. Then, the area of ​​the overlapping region enclosed by the two defect boundaries is calculated, and this overlapping area is written as the defect boundary overlap area. The difference in local electrical measurement values ​​is obtained by subtracting the local electrical measurement result of the previous process record from the local electrical measurement result of the next process record. If the local electrical measurement result is a single value, it is directly subtracted. If the local electrical measurement result consists of multiple measurement fields, it is subtracted according to the main measurement field pre-written in the process record, and the resulting difference is written as the local electrical measurement difference. After all calculations are completed, a comparison result group containing the defect boundary overlap area and the local electrical measurement difference is obtained. Next, a comparison is performed between the defect categories of the previous and subsequent process records in the result group. The subsequent process record is then categorized based on the null values ​​of the defect boundaries of the previous and subsequent process records. Here, null defect boundaries are represented by an empty boundary list, and null defect categories are represented by an empty category marker. When the previous process record's defect boundary is in the null boundary list and the subsequent process record's defect boundary is not in the null boundary list, the subsequent process record is recorded as a newly created abnormal record. If the defect boundary of a process record is not an empty boundary list, the defect category of the previous process record is consistent with the defect category of the subsequent process record, and the overlap area of ​​the defect boundaries is not zero, the subsequent process record is recorded as a continuing abnormal record. For subsequent process records that do not meet the above two writing conditions, no new abnormal record or continuing abnormal record is written. After completing the classification of all position comparison groups, the classification result group is output. Each classification result retains the corresponding board identifier, process number, observation position, defect boundary, defect category, defect location, and local electrical test result for subsequent further judgment. Furthermore, for each continuing abnormal record in the classification result group, the defect boundary area of ​​the previous process record and the defect boundary area of ​​the subsequent process record are calculated. The defect boundary area is directly obtained from the area of ​​the region enclosed by the respective defect boundaries. When the defect boundary area of ​​the subsequent process record is greater than that of the previous process record, and the difference in local electrical measurement values ​​is not zero, the continuing abnormal record is rewritten as an aggravated abnormal record. When the defect boundary area of ​​the subsequent process record is not greater than that of the previous process record, or the difference in local electrical measurement values ​​is zero, the record is kept as a continuing abnormal record and no rewriting is performed. After all rewriting is completed, the newly generated abnormal records, continuing abnormal records, and aggravated abnormal records are arranged in ascending order of process number. When the process number is the same, they are arranged in ascending order of observation position number to form an abnormal sequence and output it. After this processing, the abnormal changes of each observation position under the same board identification in each key process can directly enter the subsequent defect real generation record judgment in a unified order. Furthermore, by first establishing a position comparison group with the same observation position, and then performing translation, overlap and difference calculations, the continuous changes of defects in the preceding and following processes can be compressed into the same coordinate caliber, thereby ensuring that the abnormal sequence has a basis for repeatable calculation. In practical applications: After process records are generated for both the pretreatment and cyanide-free electroplating critical processes on a circuit board, the defect boundaries, defect categories, and local electrical measurement results with the same observation location number in these two process records are first read. Then, the defect boundaries corresponding to the cyanide-free electroplating critical process are translated according to the coordinate difference of the defect locations in the two process records, so that the two defect boundaries fall within the same observation location coordinate system, and the overlap area of ​​the defect boundaries and the difference in local electrical measurement values ​​are calculated. Subsequently, if the defect boundary at the observation location is empty in the pretreatment critical process but not empty in the cyanide-free electroplating critical process, then the cyanide-free electroplating... The process record corresponding to the critical process is recorded as a new abnormal record. If the defect categories of the two process records are the same and the overlap area of ​​the defect boundaries is not zero, the process record corresponding to the cyanide-free electroplating critical process is recorded as a continuing abnormal record. On this basis, if the defect boundary area of ​​the cyanide-free electroplating critical process is greater than the defect boundary area of ​​the pretreatment critical process, and the difference in local electrical measurement values ​​is not zero, the continuing abnormal record is rewritten as an aggravated abnormal record. Finally, the new abnormal records, continuing abnormal records, and aggravated abnormal records corresponding to all observation positions of the circuit board are arranged according to the process sequence number to form an abnormal sequence for subsequent traceability and judgment.

[0021] S4. Scan the abnormal sequence by process number and take the first newly generated abnormal record at the same observation position as the candidate generated record. When there are similar continuous abnormal records or aggravated abnormal records in the two key processes after the candidate generated record, determine the candidate generated record as the actual defect generated record and output it. After an anomaly sequence is formed at the same observation location, the earliest newly emerging anomaly record cannot be directly identified as the actual defect generation record. This is because whether this newly emerging anomaly record is subsequently inherited by similar anomalies determines whether it is a preceding actual generation point or merely a short-term disturbance or a local misjudgment. Therefore, within the anomaly sequence of the same board identifier and the same observation location, candidate generation records need to be extracted first. Then, subsequent anomaly records are read sequentially along the key processes. The overlap of defect boundaries and the consistency of defect categories between subsequent anomaly records and candidate generation records are continuously judged. Based on the judgment results, candidate generation records are classified into three states: direct pass, verification, or rollback. For candidate generated records that have entered the review stage, process records of the same process and observation position under the same fixture identifier are introduced as supporting evidence to supplement the verification of whether the candidate generated records have the basis for continued retention. Finally, candidate generated records that meet the conditions for continued verification or supporting evidence are written as actual defect generated records, and candidate generated records that do not meet the conditions are deleted. New new abnormal records are extracted and extracted until there are no new candidate generated records at the current observation position. Through this process, the actual defect generated records are no longer directly determined by a single abnormal record, but are jointly confirmed by the continuous abnormal inheritance relationship under the same plate and the same observation position and the supporting evidence relationship of the same fixture. The implementation process includes the following steps: First, read the abnormal records in the abnormal sequence that are arranged by process number under the same board identifier and observation position, and retain only the record set corresponding to the board identifier and observation position. Then, sort them by process number from smallest to largest. After sorting, search for the abnormal records marked as new abnormal records one by one, extract the new abnormal record with the first process number, and write the process number, defect type, defect boundary, defect location, and local electrical test result of the record into the candidate generated record. If there are more than two new abnormal records under the same process number, first calculate the defect boundary area of ​​each new abnormal record, and take the new abnormal record with the first defect boundary area as the candidate generated record. When the defect boundary areas are the same, calculate the absolute value of the local electrical test result corresponding to each new abnormal record, and take the new abnormal record with the first absolute value as the candidate generated record. When the absolute values ​​of the local electrical test results are still the same, take the first record according to the writing order of the abnormal records in the abnormal sequence. After completing this process, output the candidate generated record as the starting point for subsequent judgment of the current observation position. After obtaining the candidate generation record, read the exception records corresponding to the first and second critical processes following the candidate generation record. The first and second critical processes are determined according to the order of critical processes. If an exception record has been generated for the critical process immediately following the candidate generation record, that exception record is used as the exception record corresponding to the first critical process. If no exception record has been generated for the adjacent critical process, that critical process is recorded as an empty record, and the process continues to read the exception record corresponding to the next critical process as the exception record corresponding to the second critical process. If there are no exception records at the corresponding positions of either the first or second critical process, the corresponding position is directly used as the conflict resolution object. After reading, calculate the relationship between the two exception records and the candidate generation record. The calculation of the overlapping area of ​​the defect boundaries is as follows: First, the defect boundary of the subsequent abnormal record is translated to the coordinate system of the observation position of the candidate generated record according to the coordinate difference between the defect position of the candidate generated record and the defect position of the subsequent abnormal record. Then, the area of ​​the overlapping area enclosed by the two defect boundaries is calculated. Subsequently, the defect categories of the two abnormal records and the candidate generated record are compared respectively. Abnormal records with the same defect category and a non-zero overlapping area of ​​defect boundaries are written as continuation records. Abnormal records with different defect categories or zero overlapping area of ​​defect boundaries are written as conflict records. If there are no abnormal records in the subsequent first or second key process, the judgment result of that position is directly written as a conflict record. After processing, the judgment results of the first and second processes corresponding to the candidate generated record are output. After obtaining the results of the first and second process determinations, cross-determination is performed on the same candidate generated record: when both the first and second process determinations result in a renewal record, the candidate generated record is marked as "pass-through"; when both the first and second process determinations result in a conflict record, the candidate generated record is marked as "review"; when both the first and second process determinations result in a conflict record, the candidate generated record is marked as "review"; when both the first and second process determinations result in a conflict record, the candidate generated record is marked as "return"; after writing, a gating result is formed, where the "pass-through" mark indicates that the candidate generated record has been accepted by the same type of anomaly in two consecutive key processes, the "review" mark indicates that the candidate generated record has been accepted only on one side and needs to be further verified by the same fixture, and the "return" mark indicates that the candidate generated record has not been accepted in either of the two key processes and is no longer retained as the current candidate generated record; For candidate generated records marked with a verification mark, continue reading the process records corresponding to the same process number and observation position under the same fixture identifier, and compare these process records one by one with the candidate generated records. During the comparison, first read the defect boundary and defect category in the process record, then translate the defect boundary in the process record to the coordinate system of the observation position where the candidate generated record is located, then calculate the overlap area of ​​the defect boundary between the process record and the candidate generated record, and compare the defect categories of the process record and the candidate generated record. When the defect categories of the process record and the candidate generated record are the same and the overlap area of ​​the defect boundary is not zero, write the process record as a supporting record. After completing all supporting comparisons for the same candidate generated record, count whether there is a supporting record. When there is at least one supporting record, rewrite the candidate generated record as a pass record and output it. When there is no supporting record, rewrite the candidate generated record as a rollback record and output it. The pass record here still retains the process number, defect category, defect boundary, defect position and local electrical measurement results in the candidate generated record, which are used to directly write the actual defect generated record later. After the gating results and verification results are formed, the candidate generation records and pass records corresponding to the pass markers are read. The candidate generation records and pass records corresponding to the pass markers are uniformly written as the actual defect generation records and output. For candidate generation records written to the rollback markers and candidate generation records rewritten as rollback records, they are directly deleted from the anomaly sequence corresponding to the current observation position. After deletion, the new anomaly records with the same board identifier and the same observation position, whose process number is greater than the deleted candidate generation record and whose process number is the first, are reread and formed into new candidate generation records. Then, the subsequent anomaly record reading, renewal record and conflict record writing, gating result writing, and corresponding processing of circumstantial verification are executed again in sequence. When there are no new anomaly records under the same board identifier and the same observation position, the current board identifier and the current observation position are written to the scan end marker and output to end the subsequent extraction and judgment of the observation position. After this processing, multiple new anomaly records under the same observation position can enter the judgment process in sequence until the actual defect generation record is retained or all scans are completed. Through the above processing, the selection rules for new anomaly records under the same board identification and observation position, the reading rules for the first and second key processes, the rules for determining the continued verification of the overlapping area of ​​the defect boundary, the writing rules for the through mark, the verification mark and the backtrack mark, the supplementary verification rules for the supporting evidence records, and the rescanning rules after backtracking are all fixed. In the future, there will be no problems such as the inability to uniquely determine the candidate generated record, the inability to process the missing subsequent anomaly records, the lack of supporting evidence sources for the verification record, the inability to continue after backtracking, or the inability to terminate the scan at the current observation position. Furthermore, by first using the succession relationship of the last two key processes to solidify the candidate generated record, and then using the same fixture process record to supplement the verification, it is possible to distinguish between short-term anomalies, isolated anomalies and continuous anomalies, so that the final written defect real generated record has a continuous succession basis and a supporting evidence basis from the same fixture. In practical applications: When three abnormal records are generated sequentially at the same observation position for a circuit board, the first newly generated abnormal record is extracted as a candidate generated record according to the process sequence number. Then, the abnormal records corresponding to the first and second key processes are read, and the defect boundary overlap area between these two abnormal records and the candidate generated record is calculated and compared with the defect category. If the two subsequent abnormal records are consistent with the defect category of the candidate generated record and the defect boundary overlap area is not zero, the candidate generated record is directly written as the actual generated record of the defect. If only one subsequent abnormal record meets the conditions, the process records corresponding to the same process and observation position under the same fixture identifier are further read to check whether there is a corroborating record that is consistent with the defect category of the candidate generated record and the defect boundary overlap area is not zero. If there is, the candidate generated record is rewritten as a pass record and written as the actual generated record of the defect. If there is no, the candidate generated record is deleted, and the next newly generated abnormal record is extracted and the same judgment is performed. When there are no new newly generated abnormal records at the current observation position, the scan end mark of the observation position is output.

[0022] S5. Merge the actual defect generation record with the process records of other circuit boards under the same fixture identifier at the same process number and the same observation position; when there are two or more actual defect generation records with the same defect category and the same observation position, output the batch warning record and the corresponding process record. After determining the actual defect generation record, it is necessary to further determine whether multiple boards with the same generation characteristics have co-occurred under the same fixture identifier, thereby extending single-board traceability to batch warnings. The focus here is no longer on the individual actual defect generation record itself, but on grouping the process records and actual defect generation records corresponding to the same fixture identifier, process number, and observation position into the same merging range. Then, perform similar grouping and quantity statistics on the actual defect generation records within the merging range. Finally, write the groups that meet the quantity conditions into batch warning records, and output the batch warning record together with the corresponding process records that constitute the group. After this processing, the batch warning record retains both the concentrated characteristics of batch anomalies and the traceability chain that can be traced back to the original process records. The implementation process includes the following steps: First, read the actual defect generation record and the process records of each circuit board under the same fixture identifier. The actual defect generation record serves as the master record for subsequent grouping, while the process records of each circuit board serve as the base record for grouping, providing corresponding process information. During reading, first filter out all records under the same fixture by fixture identifier, then extract the defect category, defect boundary, local electrical test results, tank liquid parameters, and equipment parameters corresponding to the same process number and observation position. Write these fields, along with the corresponding board identifier, into the same grouping unit. Each grouping unit corresponds to only one fixture identifier, one process number, and one observation position. If there are multiple process numbers or multiple observation positions under the same fixture identifier, they are formed into different grouping units and are not mixed. After all grouping units are written, they are grouped into a group and output. After obtaining the merge groups, each defect-generated record in each merge group undergoes defect category comparison and observation location comparison. Here, the observation location comparison is still based on the observation location number written in the process record, and the defect category comparison is still based on the category number output by the defect identification network. When two defect-generated records have the same defect category and the same observation location, these two defect-generated records are written into the same batch record group. If there are three or more defect-generated records in a merge group, the first defect-generated record is used as the benchmark, and each subsequent record is compared with the defect category and observation location. Defect-generated records that meet the writing conditions are continuously merged into the same batch record group, while defect-generated records that do not meet the writing conditions are formed into a new batch record group. After all merge groups have been processed, the number of defect-generated records in each batch record group is counted. During the count, different board identifiers are used as the counting unit, and defect-generated records with the same board identifier that are written repeatedly are counted only once. After the count is completed, the batch record groups are output. After forming batch record groups, the corresponding process records for each batch record group are read, and batch record groups with two or more actual defect generation records are selected. For batch record groups that meet this quantity condition, the fixture identifier, process number, observation position, defect category, corresponding board identifier, and corresponding actual defect generation record are written into the same batch warning record. The corresponding board identifier is written as the set of all board identifiers participating in the counting within the batch record group, and the corresponding actual defect generation record is written as the set of all actual defect generation records participating in the counting within the batch record group. After the batch warning record is formed, the process records corresponding to the batch record group are extracted, and the batch warning record and the corresponding process record are output together. The corresponding process record here refers to all process records that constitute the basis for merging the batch record group. It is used to subsequently check the original observation situation of the same process number and the same observation position under the same fixture, so as to ensure that the batch warning record can be directly returned to the original record chain. Through the above processing, the hierarchical relationship between defect real generation records, merging groups, batch record groups, and batch warning records is fixed. Batch anomalies under the same fixture identifier no longer rely on a general judgment of the number of anomalies, but are based on the continuous merging of the same process number, the same observation position, and the same defect category. At the same time, batch warning records are output synchronously with the corresponding process records, so that the original process records that constitute the batch warning can be directly reviewed during subsequent verification. In practical applications: When multiple circuit boards under the same fixture identifier complete the defect generation record judgment, first read these defect generation records and the process records of all circuit boards under the fixture, then group the records with the same process number and the same observation position into the same group; then compare the defect category and observation position of the defect generation records in each group, and write the defect generation records with the same defect category and the same observation position into the same batch record group, and count the number of defect generation records in each batch record group according to different board identifiers; finally, write the fixture identifier, process number, observation position, defect category, corresponding board identifier set and corresponding defect generation record set into the batch record group with a quantity of two, forming a batch warning record, and output the batch warning record together with the process record corresponding to the group, so that the production line can perform subsequent batch verification and handling.

[0023] Furthermore, an AI-based quality traceability system for cyanide-free process circuit boards includes: The index building module writes board identifiers, fixture identifiers and process numbers to the circuit boards entering the cyanide-free process production line, extracts fixed pixel points and fixed electrical test points in the edge strip area, and associates them with key processes to output an observation index table. The data acquisition module collects plate images, local electrical measurement results, tank liquid parameters and equipment parameters through edge computing nodes at the exit of each key process according to the observation index table. The plate images are input into the defect recognition network to obtain the defect boundary, defect category and defect location. The data is then written into the process record and output according to the plate identification, process number and observation location. The anomaly determination module performs coordinate matching on the process records of adjacent processes under the same board part identification according to the same observation position, calculates the overlapping area of ​​the defect boundary and the difference of local electrical measurement values, and classifies new anomaly records, continued anomaly records and aggravated anomaly records according to the defect category comparison results, and outputs the anomaly sequence. The traceability and judgment module scans the abnormal sequence by process number and takes the first newly generated abnormal record at the same observation position as a candidate generated record. When there are similar continuing abnormal records or aggravated abnormal records in the two key processes after the candidate generated record, the candidate generated record is determined to be the actual defect generated record and output. The batch warning module merges actual defect generation records with process records of other circuit boards under the same fixture identifier at the same process number and observation location. When there are two or more actual defect generation records with the same defect category and observation location, it outputs a batch warning record and the corresponding process record. Working Principle: This solution first pre-determines a set of reusable observation positions in the edge strip area of ​​the circuit board, and writes the fixed pixel point, fixed electrical test point, board identification, fixture identification, and process sequence number corresponding to each observation position into an observation index table. Subsequently, at the exit of each critical process, edge computing nodes synchronously collect board surface images, local electrical test results, bath parameters, and equipment parameters according to this index table, and write them into process records. Next, the process records of the same board in adjacent critical processes are compared according to the same observation position, the overlap area of ​​defect boundaries and the difference of local electrical test values ​​are calculated, and new abnormal records, continued abnormal records, and aggravated abnormal records are identified in sequence to form an abnormal sequence. Then, candidate generated records are extracted one by one from the abnormal sequence. Combined with the follow-up certification in subsequent key processes and the corroborating evidence in the same fixture records, the actual defect generated records that can be continuously accepted are screened out. Finally, the actual defect generated records under the same fixture, with the same process number and the same observation position are merged to form batch warning records and corresponding process records. The previous indexing ensures the consistency of position, the subsequent continuous comparison ensures that the abnormal evolution is traceable, and the subsequent follow-up certification and corroborating evidence judgment ensures that the actual generated position can be confirmed. Therefore, the whole process does not only look at the final inspection results, but also connects the entire chain of defects from their appearance, continuation to batch spread along the process. For example, on a continuous production line for cyanide-free circuit boards, the same rack carries multiple circuit boards through key processes such as pretreatment, cyanide-free electroplating, and post-treatment. Edge computing nodes are deployed at the exit of each key process. These edge computing nodes collect local images and electrical measurement data for each board at fixed observation positions according to a pre-written observation index table, and simultaneously read the bath parameters and equipment parameters for that process. If a board shows no obvious abnormalities after pretreatment, but a defect boundary first appears at a certain observation position after cyanide-free electroplating, and the same defect persists at that position after post-treatment with the boundary continuing to expand, then the system will identify the corresponding cyanide-free electroplating line. The abnormal record serves as a candidate generation record. Combined with the verification results of subsequent processes and corroborating records from other boards in the same location on the same rack, it is confirmed that the defect was first formed in the cyanide-free electroplating process. If two or more circuit boards under the same rack form actual generation records of the same type of defect at the same process number and observation location, the system will output corresponding batch warning records and process records, reminding the production line to promptly check the bath solution status and equipment status corresponding to that process. In this way, in practical applications, on-site personnel will no longer only see which board is faulty, but will be able to directly see in which key process and observation location the defect started to form, and whether it has already spread in the same rack.

[0024] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for quality traceability of cyanide-free process circuit boards based on artificial intelligence, characterized in that, include: S1. Write board identification, fixture identification and process number to the circuit board entering the cyanide-free process production line. Extract fixed pixel points and fixed electrical test points in the edge strip area and correspond them with key processes. Output the observation index table. Read the edge strip area image, edge strip area layout data, and contact position data of each key process of the circuit board. Establish a unified coordinate system with the starting angle of the edge strip area as the origin and the transmission direction and board width direction. Divide the edge strip area coordinate grid with the same spacing and output the edge strip area coordinate grid set. Calculate the minimum distance between the grid center and the line boundary, hole boundary, and board boundary of the edge strip area coordinate grid set grid by grid. Delete the coordinate grid that coincides with the contact position of any key process. Determine the grid center of the remaining coordinate grid as a fixed pixel point. Select a unique fixed electrical test point in the coordinate grid of each fixed pixel point and output the observation point set. The fixed pixel points and fixed electrical measurement points in the observation point set are numbered according to the same coordinate grid and corresponded to the key processes, and associated with the board identifier, fixture identifier and process sequence number to generate an observation index table; S2. The edge computing nodes at the exit of each key process collect the plate surface image, local electrical measurement results, tank liquid parameters and equipment parameters according to the observation index table, input the plate surface image into the defect recognition network, obtain the defect boundary, defect category and defect location, and write it into the process record according to the plate identification, process number and observation position and output it. Read the plate identifier, process number, fixed pixel point, and fixed electrical test point from the observation index table. At the exit of the corresponding key process, collect the whole plate image, the local electrical test results corresponding to the fixed electrical test point, the tank liquid parameters, and the equipment parameters. Merge them according to the plate identifier and process number and output the collection group. For the whole plate image in the collection group, extract local blocks according to the coordinates of the fixed pixel point. Write the center coordinates of each local block and the coordinates of the corresponding fixed electrical test point as the same observation position. Correspond the local blocks with the local electrical test results according to the same observation position and output the observation group. Input the local patches in the observation group into the defect recognition network, extract the defect boundaries in the local patches, determine the defect location according to the coordinate difference between the center of the outer rectangle of the defect boundary and the observation location, and determine the defect category according to the image features of the area covered by the defect boundary, and output the recognition group; write the defect boundaries, defect categories and defect locations in the recognition group, along with the local electrical test results, tank liquid parameters and equipment parameters corresponding to the same observation location, into the process record according to the plate identification, process number and observation location and output it; S3. Perform coordinate matching on the process records of adjacent processes under the same board part identification according to the same observation position, calculate the overlapping area of ​​defect boundaries and the difference of local electrical measurement values, and classify new abnormal records, continued abnormal records and aggravated abnormal records according to the defect category comparison results, and output the abnormal sequence. S4. Scan the abnormal sequence by process number and take the first newly generated abnormal record at the same observation position as the candidate generated record. When there are similar continuous abnormal records or aggravated abnormal records in the two key processes after the candidate generated record, determine the candidate generated record as the actual defect generated record and output it. S5. Merge the actual defect generation record with the process records of other circuit boards under the same fixture identifier at the same process number and the same observation position; when there are two or more actual defect generation records with the same defect category and the same observation position, output the batch warning record and the corresponding process record.

2. The method for quality traceability of cyanide-free process circuit boards based on artificial intelligence according to claim 1, characterized in that: S3 includes: S3-1. Read two adjacent process records under the same board identifier according to the process sequence number, extract the defect boundary, defect category and local electrical measurement results corresponding to the same observation position in the two process records, and form a position comparison group according to the order of the previous process record before the next process record, and output the position comparison group. S3-2. For the defect boundary of the next process recorded in the position comparison group, perform translation according to the coordinate difference between the defect location and the defect location of the previous process recorded, so that the defect boundary of the next process recorded after translation falls into the same observation position coordinate system as the defect boundary of the previous process recorded. Calculate the area of ​​the overlapping area of ​​the two as the defect boundary overlap area, and subtract the local electrical measurement result of the previous process recorded from the local electrical measurement result of the next process recorded to obtain the difference of local electrical measurement value, and output the comparison result group.

3. The method for quality traceability of cyanide-free process circuit boards based on artificial intelligence according to claim 2, characterized in that: S3 further includes: S3-3. The comparison results group is performed by comparing the defect category of the previous process record with the defect category of the next process record. The next process record with the empty defect boundary of the previous process record and the non-empty defect boundary of the next process record is recorded as a new abnormal record. The next process record with the non-empty defect boundary of the previous process record, the non-empty defect boundary of the next process record, the same defect category, and the non-zero overlap area of ​​the defect boundary is recorded as a continued abnormal record. The classification result group is output. S3-4. For the continuing abnormal records in the classification result group, calculate the defect boundary area of ​​the previous process record and the defect boundary area of ​​the next process record respectively. Rewrite the continuing abnormal records where the defect boundary area of ​​the next process record is greater than the defect boundary area of ​​the previous process record and the difference in local electrical measurement values ​​is not zero as aggravated abnormal records. Arrange the new abnormal records, continuing abnormal records and aggravated abnormal records according to the process sequence number and output the abnormal sequence.

4. The method for quality traceability of cyanide-free process circuit boards based on artificial intelligence according to claim 3, characterized in that: S4 includes: S4-1. Read the abnormal records in the abnormal sequence that are arranged by process number under the same board identifier and the same observation position. Extract the newly generated abnormal record with the first process number, write the process number, defect type, defect boundary, defect location and local electrical test results to form a candidate generated record and output it. S4-2. After reading the candidate generated record, read the abnormal records corresponding to the first key process and the second key process. Calculate the defect boundary overlap area between the two abnormal records and the candidate generated record. Compare the defect categories of the two abnormal records and the candidate generated record. Write the abnormal records with the same defect category and non-zero defect boundary overlap area as renewal records. Write the abnormal records with different defect categories or zero defect boundary overlap area as conflict records. Output the first process judgment result and the second process judgment result corresponding to the candidate generated record. S4-3. Write the candidate generated records whose first and second process determination results are both renewal records into the pass-through flag. Write the candidate generated records whose first and second process determination results are both renewal records into the review flag. Write the candidate generated records whose first and second process determination results are both conflict records into the review flag. Write the candidate generated records whose first and second process determination results are both conflict records into the rollback flag. Output the gating result.

5. The method for quality traceability of cyanide-free process circuit boards based on artificial intelligence according to claim 4, characterized in that: S4 further includes: S4-4. Read the candidate generated records marked with the verification mark and the process records corresponding to the same process number and the same observation position under the same fixture identifier. Calculate the defect boundary overlap area between the process records and the candidate generated records and compare the defect categories. Write the process records with the same defect category and non-zero defect boundary overlap area as supporting evidence records. When supporting evidence records exist, rewrite the candidate generated records as passing records and output them. When no supporting evidence records exist, rewrite the candidate generated records as rollback records and output them. S4-5. Read the candidate generation record and pass record corresponding to the pass mark, write the candidate generation record and pass record corresponding to the pass mark as the actual defect generation record and output them. Delete the candidate generation record and backtracking record corresponding to the backtracking mark. Then extract the new abnormal record with the same board identifier and the same observation position, whose process number is greater than the deleted candidate generation record and whose process number is the first, to form a new candidate generation record and re-execute the subsequent judgment. When there is no new candidate generation record, output the scan end mark of the current observation position.

6. The method for quality traceability of cyanide-free process circuit boards based on artificial intelligence according to claim 5, characterized in that: S5 includes: S5-1. Read the actual defect generation record and the process record of each circuit board under the same fixture identifier, extract the defect category, defect boundary, local electrical test result, tank liquid parameter and equipment parameter corresponding to the same process number and the same observation position, and form a merge group according to the fixture identifier, process number and observation position, and output the merge group; S5-2. Perform defect category comparison and observation location comparison on each defect real generation record in each merge group, write defect real generation records with the same defect category and the same observation location into the same batch record group, count the number of defect real generation records in each batch record group, and output the batch record group. S5-3. Read the corresponding process records for each batch record group, and write the fixture identifier, process number, observation position, defect category, corresponding board identifier and corresponding defect actual generation record into the batch record group with two or more defect actual generation records to form a batch warning record, and output the batch warning record and the corresponding process record.

7. A quality traceability system for cyanide-free process circuit boards based on artificial intelligence, characterized in that, include: The index building module writes board identifiers, fixture identifiers and process numbers to the circuit boards entering the cyanide-free process production line, extracts fixed pixel points and fixed electrical test points in the edge strip area, and associates them with key processes to output an observation index table. Read the edge strip area image, edge strip area layout data, and contact position data of each key process of the circuit board. Establish a unified coordinate system with the starting angle of the edge strip area as the origin and the transmission direction and board width direction. Divide the edge strip area coordinate grid with the same spacing and output the edge strip area coordinate grid set. Calculate the minimum distance between the grid center and the line boundary, hole boundary, and board boundary of the edge strip area coordinate grid set grid by grid. Delete the coordinate grid that coincides with the contact position of any key process. Determine the grid center of the remaining coordinate grid as a fixed pixel point. Select a unique fixed electrical test point in the coordinate grid of each fixed pixel point and output the observation point set. The fixed pixel points and fixed electrical measurement points in the observation point set are numbered according to the same coordinate grid and corresponded to the key processes, and associated with the board identifier, fixture identifier and process sequence number to generate an observation index table; The data acquisition module collects plate images, local electrical measurement results, tank liquid parameters and equipment parameters through edge computing nodes at the exit of each key process according to the observation index table. The plate images are input into the defect recognition network to obtain the defect boundary, defect category and defect location. The data is then written into the process record and output according to the plate identification, process number and observation location. Read the plate identifier, process number, fixed pixel point, and fixed electrical test point from the observation index table. At the exit of the corresponding key process, collect the whole plate image, the local electrical test results corresponding to the fixed electrical test point, the tank liquid parameters, and the equipment parameters. Merge them according to the plate identifier and process number and output the collection group. For the whole plate image in the collection group, extract local blocks according to the coordinates of the fixed pixel point. Write the center coordinates of each local block and the coordinates of the corresponding fixed electrical test point as the same observation position. Correspond the local blocks with the local electrical test results according to the same observation position and output the observation group. Input the local patches in the observation group into the defect recognition network, extract the defect boundaries in the local patches, determine the defect location according to the coordinate difference between the center of the outer rectangle of the defect boundary and the observation location, and determine the defect category according to the image features of the area covered by the defect boundary, and output the recognition group; write the defect boundaries, defect categories and defect locations in the recognition group, along with the local electrical test results, tank liquid parameters and equipment parameters corresponding to the same observation location, into the process record according to the plate identification, process number and observation location and output it; The anomaly determination module performs coordinate matching on the process records of adjacent processes under the same board part identification according to the same observation position, calculates the overlapping area of ​​the defect boundary and the difference of local electrical measurement values, and classifies new anomaly records, continued anomaly records and aggravated anomaly records according to the defect category comparison results, and outputs the anomaly sequence. The traceability and judgment module scans the abnormal sequence by process number and takes the first newly generated abnormal record at the same observation position as a candidate generated record. When there are similar continuing abnormal records or aggravated abnormal records in the two key processes after the candidate generated record, the candidate generated record is determined to be the actual defect generated record and output. The batch warning module merges the actual defect generation records with the process records of other circuit boards under the same fixture identification at the same process number and the same observation position. When there are two or more actual defect generation records with the same defect category and the same observation position, it outputs the batch warning record and the corresponding process record.

Images