Wafer splitting optimization method based on wafer splitting defective product detection and wafer splitting device

Abstract

The application provides a board splitting optimization method and a board splitting device based on board splitting defective product detection. The board splitting optimization method based on board splitting defective product detection comprises the following steps: after performing the action of splitting the target circuit board, an image of the splitting cut is obtained, and the image of the splitting cut is compared with a preset qualified splitting image template in a visual library; in the case that it is determined that the current splitting result is a defective product, a defective feature is extracted from the cut image of the defective product, the defective feature is compared with existing defective features in the visual library, and a new type of defective feature is determined; the processing parameters of the splitting are updated based on the type of the defective feature, and the splitting action is performed according to the processing parameters. The application aims to improve the subjective difference of manual judgment in the detection of defective products, improve the data reliability, and realize the dynamic self-optimization of parameters.

Description

Technical Field

[0001] This application relates to the field of depaneling technology, specifically to a depaneling optimization method and depaneling device based on the detection of defective products during depaneling. Background Technology

[0002] Traditional PCB depaneling defect handling relies on manual judgment, and defective situations can only be recorded manually. This leads to large fluctuations in defect judgment results, affecting the consistency and reliability of product quality, and also lacks reliable data support for subsequent quality analysis and cause tracing. Summary of the Invention

[0003] The main purpose of this application is to propose a depaneling optimization method and depaneling device based on the detection of defective products during depaneling. This application aims to improve the subjective difference of human judgment during the detection of defective products during depaneling, improve data reliability, and realize dynamic self-optimization of parameters.

[0004] This application proposes a depaneling optimization method based on defective product detection, the method comprising: After performing the action of cutting and separating the target circuit board, an image of the separation cut is obtained, and the image of the separation cut is compared with a preset qualified separation image template in the vision library. If the current board-making result is determined to be a defective product, the defect features are extracted from the cut image of the defective product, the defect features are compared with the existing defect features in the visual library, and new types of defect features are identified. The processing parameters for the separation process are updated based on the type of defective feature, and the separation action is performed according to the processing parameters.

[0005] Optionally, the step of comparing the image of the plate cutting with a preset qualified plate image template in the visual library includes: If the burr length at the depanel cut exceeds a preset length value, the depaneled circuit board is determined to be defective; or, If the offset value of the board separation cut is greater than the preset offset value, the board after separation is determined to be a defective product.

[0006] Optionally, the step of extracting defect features from the cut image of the defective product when the current panelization result is determined to be a defective product, comparing the defect features with existing defect features in the visual library, and determining new types of defect features includes: For cut images identified as defective products, extract the contour features, morphological features, and positional features of the cuts, and generate corresponding defect feature vectors; The similarity is calculated based on the defective feature vector and the known types of defective feature vectors stored in the visual database; If the similarity is less than a preset similarity threshold, the defective feature is determined to be a new type. If the similarity is greater than a preset similarity threshold, the defective feature is determined to be of a known type.

[0007] Optionally, the processing parameters for the plate separation include tool pressure, positioning error threshold for tool entry determination, and cutting speed. The step of updating the processing parameters for the plate separation based on the type of defect features includes: If the type of defect is tool wear, reduce the tool pressure; When the defective feature is of the plate position offset type, the positioning error threshold for determining the cutting edge is reduced from the first error threshold to the second error threshold; If the defective feature is of the burr type, reduce the cutting speed.

[0008] Optionally, the step of updating the processing parameters of the partition plate based on the type of defective feature further includes: If the yield rate of the processed circuit board is greater than or equal to the preset yield rate, maintain the updated board separation parameters; If the yield rate is less than the preset yield rate, restore the board separation parameters to the parameter values ​​before the update.

[0009] Optionally, the step of updating the processing parameters of the partition plate based on the type of defective feature further includes: Add the new types of defective features to the visual library and update the defective product judgment rules; The step of adding the new type of defective features to the visual library and updating the defective product judgment rules includes: If a new feature is detected in a cut image, and the same new type of defective feature is detected in cut images for a consecutive preset number of frames, it is stored in the cache pool. When the preset production batch is reached, the defective features in the cache pool are clustered, denoised, and checked for repetition before being added to the visual library and the defective product judgment rules are updated.

[0010] This application also proposes a slitting device, which includes the slitting optimization method based on the detection of defective slitting products as described above.

[0011] This application proposes a depaneling optimization method based on defective product detection, the method comprising: After performing the cutting and separating action on the target circuit board, an image of the separating cut is acquired and compared with a preset qualified separating image template in the vision library. If the current separating result is determined to be a defective product, defect features are extracted from the defective product's cut image, compared with existing defect features in the vision library, and new types of defect features are identified. The separating processing parameters are updated based on the type of defect features, and the separating action is executed according to the processing parameters. Through the above settings, the separating cut image is quantitatively compared with the qualified separating image template in the vision library, realizing automated and standardized detection of separating quality. This improves both detection efficiency and the consistency and objectivity of the judgment criteria. Furthermore, by comparing the similarity of defect features with existing defect features in the vision library, the system can identify defect types (such as burrs, offsets, tool wear) or discover new types of defect features, providing an accurate classification basis for subsequent targeted optimization. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0013] Figure 1 This is a schematic flowchart of an embodiment of the depanel optimization method based on the detection of defective depanel products in this application; Figure 2 This is a schematic flowchart of another embodiment of the depanel optimization method based on depanel defect detection in this application; Figure 3 This is a schematic flowchart of another embodiment of the depaneling optimization method based on the detection of defective products in depaneling according to this application; Figure 4 This is a schematic flowchart of another embodiment of the depaneling optimization method based on the detection of defective products in depaneling according to this application; Figure 5 This is a schematic flowchart of another embodiment of the depaneling optimization method based on the detection of defective products in depaneling according to this application; Figure 6 This is a schematic flowchart of another embodiment of the depanel optimization method based on the detection of defective depanel products in this application.

[0014] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0016] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0017] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution that simultaneously satisfies A and B. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0018] Currently, the handling of defective products in traditional PCB depaneling relies on manual judgment. Defective situations can only be recorded manually, making it impossible to form a structured defective product database. This leads to large fluctuations in defective product judgment results, affecting the consistency and reliability of product quality, and also lacking reliable data support for subsequent quality analysis and cause tracing.

[0019] Therefore, this application proposes a depaneling optimization method based on defective product detection. This depaneling method can be applied to a depaneling device, which includes, but is not limited to: an industrial camera (sensing module) for acquiring images, a control module for running vision algorithms, and a depaneling machine (execution module, including cutting tools, transmission modules, etc.) for performing cutting actions.

[0020] The control module can be implemented using a main controller, such as an MCU, PLC, DSP (Digital Signal Processor), SOC (System on Chip), or FPGA (Field Programmable Gate Array). The control module can establish bidirectional communication connections with both the sensing and execution modules via industrial Ethernet, forming a closed-loop control architecture of "sensing-decision-execution".

[0021] refer to Figure 1 In one embodiment of this application, the board separation optimization method based on defective product detection includes: S100. After performing the action of cutting and separating the target circuit board, obtain an image of the separation cut and compare the image of the separation cut with the preset qualified separation image template in the vision library. S200. If the current panelization result is determined to be a defective product, extract the defective features from the cut image of the defective product, compare the defective features with the existing defective features in the visual library, and determine the new type of defective features. S300: Update the processing parameters of the board based on the type of defect characteristics, and perform the board separation action according to the processing parameters.

[0022] In this embodiment, optionally, after completing the positioning calibration and path planning of the target circuit board, the control module controls the execution module (such as a depaneling machine) to perform the depaneling. After the depaneling operation is completed, the system enters the defective product detection stage.

[0023] Optionally, the control module controls the sensing module (e.g., an industrial camera) to acquire images of the cut edges of the target circuit board after depaneling. The images of the cut edges should clearly show the complete shape of the depaneling cut, including details such as the edge contours, burrs, and degree of offset.

[0024] Optionally, after acquiring the slab cutting image, the control module compares the acquired slab cutting image with a preset qualified slab cutting image template in the vision library. The qualified slab cutting image template is a standard cutting image of the slab type after successful slab cutting, representing the quality benchmark for qualified slab cutting. The comparison methods may include: calculating the positional deviation between the cutting edge and the standard template edge, detecting the burr length in the cutting area, and analyzing the continuity and integrity of the cutting.

[0025] Optionally, if the current board separation result is determined to be a defective product, the control module analyzes the cut image of the defective product and extracts defect features that can characterize the defect type of the defective product. The extracted feature types include, but are not limited to: contour features (such as geometric attributes such as the shape, edge curvature, and area size of the defective region), morphological features (such as the extension direction of burrs and the shape of chipped edges), and location features (such as the location of the defect on the circuit board, such as the board edge, corner, or rib connection).

[0026] Optionally, the control module compares the extracted defective features with the feature vectors of known defect types already stored in the visual database. This comparison method can perform similarity calculations. The similarity can be measured using methods such as Euclidean distance or cosine similarity; a higher similarity indicates a higher degree of matching between the current defective feature and the defective features of known defect types.

[0027] Optionally, if the similarity is greater than or equal to a preset similarity threshold, it indicates that the current defective feature highly matches a certain defect type already existing in the visual library, belonging to a known type of defective feature. In this case, the system can directly identify the type of the defective product (e.g., "burr defect," "offset defect," "tool wear defect," etc.) without needing to register a new type. If the similarity is less than the preset similarity threshold, it indicates that the current defective feature differs significantly from all known defect types in the visual library, belonging to a new type of defective feature. In this case, the control module identifies the defective feature as a new type and prepares to add it to the visual library to expand the coverage of the system's visual library.

[0028] It is important to understand that regardless of whether the current defect is a new type or a known type, the control module will update and optimize the processing parameters of subsequent board separations based on the type of defect. By adjusting the processing parameters, the probability of the same type of defect occurring is reduced.

[0029] By using the above settings, the panel cutting images are quantitatively compared with qualified panel image templates in the visual database, achieving automated and standardized inspection of panel quality. This improves both inspection efficiency and the consistency and objectivity of the judgment criteria. Furthermore, the system compares the similarity of defective features with existing defective features in the visual database. In this way, the system can identify defect types (such as burrs, offsets, and tool wear) or discover new types of defective features, providing accurate classification criteria for subsequent targeted optimization.

[0030] It is important to understand that traditional manual visual inspection for defective products relies on personal experience, and the judgment criteria may differ between different operators, or even between the same operator at different times.

[0031] Therefore, according to some embodiments of this application, optionally, in one embodiment of this application, reference is made to... Figure 2 The steps of comparing the image of the panel cutting edge with the preset qualified panel image template in the visual library include: S110. If the burr length at the depanel cut exceeds a preset length value, the depaneled circuit board is determined to be defective; or, S120. If the offset value of the board separation cut is greater than the preset offset value, the board after separation is determined to be a defective product.

[0032] In this embodiment, it is important to understand that burrs are excess protrusions or rough parts generated at the cut edge during the board separation process, and are one of the important indicators for measuring the quality of board separation. Excessively long burrs may affect the risk of the circuit board detaching during subsequent use, leading to an electrical short circuit.

[0033] Optionally, the control module performs edge extraction on the image of the board separation cut to detect burr features at the cut edge. For example, the system identifies the edge line of the cut and then measures the maximum length of the burr extending outward from the cut edge. The control module compares the detected burr length with a preset length value (e.g., 0.02 mm): if the burr length detected in the board separation cut image is greater than 0.02 mm, it is determined that the burr length exceeds the standard, and the board after separation is determined to be a defective product; if the burr length is less than or equal to 0.02 mm, the burr index is determined to be qualified, and the judgment of other indicators continues.

[0034] It's important to understand that cut offset refers to the degree of deviation between the actual cutting position and the theoretical board separation path. Excessive cut offset may cause the cutting path to intrude into the component area, the cutting line to deviate from the predetermined position and affect the functional area of ​​the circuit board, or result in excessive deviation in the dimensions of the board edge after separation.

[0035] Optionally, the control module compares the detected cut offset value with a preset offset value (e.g., 0.01mm): if the cut offset value detected in the board separation cut image is greater than 0.01mm, it is determined that the offset exceeds the standard and the board after separation is determined to be a defective product; if the cut offset value is less than or equal to 0.01mm, the offset index is determined to be qualified.

[0036] It should be noted that this preset offset value is set based on the positioning accuracy requirements of the PCB depaneling industry, representing the upper limit of the allowable cut position deviation for qualified depaneling. Cut offset can be determined based on the maximum offset value, the average offset value, or the cumulative length proportion exceeding the preset value. For example, the system continuously monitors the offset distribution along the entire cut length; if any point has an offset value exceeding the preset offset value by 0.01mm, it is judged as a defective product, ensuring the overall consistency of depaneling quality.

[0037] By setting specific preset length values ​​(e.g., 0.02mm) and preset offset values ​​(e.g., 0.01mm), defect determination is transformed into an objective numerical comparison. This quantitative standard improves the consistency and reproducibility of the determination results and eliminates the uncertainty caused by subjective judgment.

[0038] According to some embodiments of this application, optionally, in one embodiment of this application, reference is made to... Figure 3 If the current panelization result is determined to be a defective product, the steps of extracting defect features from the cut image of the defective product, comparing the defect features with existing defect features in the visual database, and identifying new types of defect features include: S210. For the cut image of the product identified as defective, extract the contour features, morphological features and position features of the cut, and generate the corresponding defect feature vector. S220. Calculate the similarity between the defective feature vector and the known types of defective feature vectors stored in the visual database; S221. If the similarity is less than the preset similarity threshold, the defective feature is identified as a new type. S222. If the similarity is greater than the preset similarity threshold, the bad feature is determined to be of a known type.

[0039] In this embodiment, optionally, the control module extracts contour features from the cut image to obtain contour information of the cut area. Contour features include, but are not limited to: the geometric shape of the cut area (such as the sharp shape of burrs, the arc shape of chipped edges, the triangle shape of missing corners, etc.), the curvature change of the contour, the area size and perimeter of the defect area, etc., reflecting the geometric shape of the cut defect.

[0040] Optionally, the control module extracts the morphological features of the cut. The morphological features include, but are not limited to: the extension direction of the burr, the length and width ratio of the burr, the distribution pattern of the offset (e.g., whether the offset is uniform or localized throughout the cut), the depth of the chipped edge and the opening angle, etc., which reflect the generation mechanism of the cut defect. For example, uniform offset may be due to positioning deviation, while local offset may be due to local deformation of the plate or interference from foreign objects.

[0041] Optionally, the control module analyzes the location of the cut defect on the circuit board. Location characteristics include, but are not limited to: the relative coordinates of the cut defect from the circuit board reference point, whether the cut defect is located at the edge of the board or inside the board, whether the cut defect is located on a straight cut segment or a curved / corner cut segment, and whether the cut defect is located at a rib connection. Location characteristics help determine whether the cut defect is related to a specific structure; for example, burrs are more likely to occur at corners, and chipping is more likely to occur at rib connections.

[0042] It is important to understand that after completing the feature extraction described above, the control module integrates the extracted features to generate one or more defective feature vectors. Optionally, the defective feature vector is a numerical representation of a feature, for example, it can be an N-dimensional vector, where each dimension represents a specific feature value (such as burr length, offset distance, curvature value, position coordinates, etc.). This vectorized representation facilitates subsequent similarity calculation and comparison.

[0043] Optionally, the control module calculates the similarity between the generated defective feature vector and the feature vectors of each known defect type stored in the visual library. The known defect types in the visual library are confirmed and classified defective feature templates accumulated during historical segmentation processes, with each defect type corresponding to one or more feature vectors. Optionally, the similarity can be measured using various standard algorithms; for example, Euclidean distance can be used as a similarity metric, where a smaller Euclidean distance indicates greater similarity between the two feature vectors.

[0044] Optionally, if the similarity is greater than or equal to a preset similarity threshold, it indicates that the current defective feature vector highly matches the feature vector of a known defect type in the visual library, and the similarity between the two reaches a level that can be identified as belonging to the same type. In this case, the control module determines that the current defective feature is of a known type and classifies it into the defect type with the highest similarity. If the similarity is less than the preset similarity threshold, it indicates that the current defective feature vector does not meet the identification criteria for similarity with the feature vectors of all known defect types in the visual library, and the difference between the two is significant enough that it cannot be classified into any existing category. In this case, the control module determines that the current defective feature is a new type of defective feature.

[0045] It should be noted that the preset similarity threshold is set based on engineering experience in classifying defect features in the PCB depaneling field. Generally, the higher the preset similarity threshold, the more the system tends to classify defect features as new types; the lower the preset similarity threshold, the more the system tends to classify defect types as known types.

[0046] Through the above settings, this application can automatically determine whether the current defective feature is a recurrence of a known type or the first discovery of a new type of defect without manual intervention. For known types, the system can directly apply existing optimization strategies for a quick response; for new types, the system can trigger a supplementary entry process to expand the knowledge base. This distinguishing ability reduces the probability of incorrectly classifying new types of defects into known types, leading to optimization strategy mismatch, and also reduces the probability of visual library redundancy caused by repeatedly registering known types of defects as new types.

[0047] According to some embodiments of this application, optionally, in one embodiment of this application, reference is made to... Figure 4The machining parameters for the panel separation include tool pressure, positioning error threshold for tool entry determination, and cutting speed. The steps for updating the machining parameters for the panel separation based on the type of defect feature include: S310. If the type of defect is tool wear, reduce the tool pressure; S320. When the type of defective feature is plate position offset, reduce the positioning error threshold for tool entry determination from the first error threshold to the second error threshold. S330. If the defective feature is burr-like, reduce the cutting speed.

[0048] In this embodiment, optionally, tool wear defects typically manifest as a gradual deterioration in the quality of the slit as the tool's usage time increases, resulting in increased burrs, rough cuts, and even chipping. This is because as the tool edge dulls, the cutting resistance increases under the same pressure, leading to a decrease in cut quality. When the control module determines that the current defect is of the tool wear type, the parameter update operation performed is to reduce the tool pressure.

[0049] For example, the tool pressure can be reduced by 0.5N to 1N. For instance, if the current tool pressure is set to 18N and tool wear is detected, the control module will reduce the tool pressure by 0.5N to 1N (e.g., adjust it to 17N or 17.5N). The purpose of reducing tool pressure is to reduce the cutting load on the worn tool during the cutting process, reducing the risk of chipping or tool breakage due to excessive pressure, and also improving the cut quality to some extent by reducing the pressure.

[0050] Optionally, board position offset defects typically manifest as a systematic offset of the board separation cut from the theoretical path, with a consistent offset direction and relatively stable offset amount. This is due to insufficient initial positioning accuracy of the circuit board. Under the existing tool entry judgment positioning error threshold, boards with a certain degree of offset are allowed to enter the board separation process, leading to a deviation in the cutting position. When the control module determines that the current defect type is board position offset, the parameter update operation performed is to reduce the tool entry judgment positioning error threshold from the first error threshold to the second error threshold.

[0051] For example, a first error threshold is set to 0.01mm, and a second error threshold is set to 0.008mm. The first error threshold is the initial cutting judgment standard; when the actual position coordinate deviation of the circuit board is less than or equal to 0.01mm, the system allows cutting. After detecting a board position offset defect, the control module reduces the cutting judgment positioning error threshold from 0.01mm to 0.008mm. This means that subsequent circuit boards can only be cut if their actual position coordinate deviation is less than or equal to 0.008mm. By reducing the cutting judgment positioning error threshold, the system will require more stringent standards for the positioning accuracy of subsequent circuit boards, thereby reducing offset defects caused by positioning deviations.

[0052] Optionally, burr defects typically manifest as excess protrusions at the edges of the board separation cut, resulting in an uneven cut. This is caused by excessively high cutting speeds, leading to insufficient cutting action between the tool and the circuit board, causing the material to be torn instead of cleanly cut, forming burrs. If the control module determines that the current defect is of the burr type, the parameter update operation performed is to reduce the cutting speed.

[0053] For example, the cutting speed can be reduced by 10 mm / s to 20 mm / s. If the current cutting speed is set to 70 mm / s and burr-like defects are detected, the control module will reduce the cutting speed by 10 mm / s to 20 mm / s (e.g., adjust it to 50 mm / s or 60 mm / s). The purpose of reducing the cutting speed is to increase the cutting time between the tool and the circuit board, allowing for more thorough cutting and resulting in smoother, cleaner cut edges, thus reducing burr formation.

[0054] Through the above settings, this application designs targeted parameter adjustment strategies for defects such as tool wear (reducing tool pressure), plate position misalignment (reducing the tool entry positioning error threshold), and burrs (reducing cutting speed). This precise optimization method ensures that each parameter adjustment directly addresses the root cause of the problem, improving the effectiveness of the optimization measures. Furthermore, the quantification of parameters provides the control module with clear execution instructions, eliminating the need for manual judgment of the adjustment amount and automating parameter optimization.

[0055] It is important to understand that simply detecting defective features and adjusting the depaneling parameters may lead to a local optimization trap. Therefore, to improve the accuracy of depaneling parameter optimization and contribute to improving and stabilizing the yield of the depaneled circuit board, alternatively, according to some embodiments of this application, refer to... Figure 5 The step of updating the processing parameters of the depanel based on the type of defect features also includes: S340. If the yield rate of the processed circuit board is greater than or equal to the preset yield rate, maintain the updated board separation parameters. S350. If the yield rate is less than the preset yield rate, restore the board separation parameters to the parameter values ​​before the update.

[0056] In this embodiment, optionally, after the control module updates the processing parameters (such as tool pressure, tool entry positioning error threshold, cutting speed, etc.) according to the type of defect characteristics, the device enters an observation period to continuously monitor the yield rate of subsequent batches of circuit boards. The yield rate is calculated as the proportion of circuit boards judged as good within a certain production batch or quantity to the total number of boards separated. For example, if 100 circuit boards are produced continuously, and 95 are judged as good, the yield rate is 95%. Optionally, the separation device can determine the yield rate of the separated circuit boards based on the content of some of the above embodiments. Alternatively, the yield rate can be entered periodically by R&D personnel or testing personnel.

[0057] Optionally, the preset yield rate can be preset in the control device by the R&D personnel according to the requirements, such as 95%.

[0058] Optionally, if the yield rate of the processed circuit boards is greater than or equal to the preset yield rate, the control module determines that the current parameter adjustment is effective and the board separation quality of the device has reached a stable state. In this case, the control module maintains the updated board separation parameters and continues to use them as the benchmark parameters for subsequent board separation operations. Meanwhile, the device can continue to monitor the yield rate, but will no longer actively adjust the maintained parameters unless defective products are detected again.

[0059] Optionally, if the yield rate is lower than the preset yield rate, the control module determines that the current parameter adjustment may be excessive or have side effects, leading to a deterioration in the board separation quality. The criteria for judging a decline in yield rate may include: the yield rate falling below a preset target value (e.g., dropping from above 95% to below 90%), a continuous downward trend in yield rate (e.g., the yield rate decreasing successively for three consecutive batches), or the detection of frequent recurrence of defect types related to previous optimization targets.

[0060] If the yield rate drops, the control module will restore the parameters of the sub-board to their pre-optimization values. The aim is to minimize the negative impact of this parameter adjustment, returning the equipment to a relatively stable state and reducing the likelihood of widespread quality problems caused by a single unsuccessful parameter optimization.

[0061] The above settings ensure that only parameter adjustments that have been verified in production and can reliably improve quality are retained, reducing the quality risks associated with blind adjustments. Furthermore, if the yield rate decreases after adjustment, the device can automatically revert to the parameter values ​​before optimization. This reversible design allows the device to boldly attempt parameter optimization without worrying about long-term quality problems caused by a single failed adjustment, thus reducing the risk and cost of optimization exploration.

[0062] It's important to understand that in industrial environments, factors such as momentary light flickering, brief dust obstruction, and occasional image acquisition anomalies can lead to false defect features in a single image. If every newly detected feature is immediately added to the database, the visual database may be contaminated with a large amount of invalid, noisy data, causing subsequent defect judgment rules to become ineffective.

[0063] Therefore, according to some embodiments of this application, optionally, in one embodiment of this application, reference is made to... Figure 6 The step of updating the processing parameters of the depanel based on the type of defect features also includes: S360. Add new types of defective features to the visual library and update the defective product judgment rules; The steps for adding new types of defective features to the visual database and updating the defective product judgment rules include: S361. If a new feature is detected in a cut image, and the same new type of defective feature is detected in cut images for a consecutive preset number of frames, store it in the cache pool. S362. When the preset production batch is reached, the features in the cache pool are clustered, denoised, and checked for repetition before being added to the visual library and the defective product judgment rules are updated.

[0064] In this embodiment, it is important to understand that when the control module determines that a certain defective feature is a new type, the system will not immediately add it to the visual library. Instead, it first enters a controlled caching mechanism to prevent false defective features caused by a single accidental event (such as a momentary flash of light, a brief obstruction by dust, or an abnormal image acquisition) from polluting the visual library.

[0065] Optionally, the system acquires subsequent cut images and checks whether the same new type of defect feature can be detected within a preset number of consecutive frames. For example, the preset number of frames is set to 3 frames. That is, the system needs to detect the same new type of defect feature (e.g., the same morphology of burrs or the same type of offset pattern) in 3 consecutive different cut images to consider the defect feature to be stable rather than accidental. Only then will the control module store the new type of defect feature in the cache pool.

[0066] Optionally, the cache pool is a temporary storage area used to temporarily store new types of defect features that have passed preliminary verification but have not yet been formally added to the database. The data stored in the cache pool includes defect feature vectors, corresponding cut images, detection time, board type information, etc.

[0067] Optionally, defective features stored in the cache pool are not immediately incorporated into the visual library, but require further batch verification and screening. The system sets a batch threshold, triggering a batch ingestion operation only after production reaches a preset batch size. For example, if the preset production batch size is 100 times, the system performs centralized processing on all candidate new types of defective features accumulated in the cache pool every 100 board splitting processes (or every 100 circuit boards processed). This centralized processing includes clustering noise reduction and repeatability verification.

[0068] Optionally, the clustering denoising steps include: the system performs clustering analysis on the bad feature vectors in the cache pool, and classifies candidate features with high feature similarity into the same potential category. During the clustering denoising process, isolated features that cannot be classified into any category are considered as possible noise or occasional anomalies and are filtered out in the denoising stage, not entering the subsequent database entry process.

[0069] Optionally, the duplication verification steps include: for each potential new category formed after clustering, the system further verifies the consistency of its internal features. If there are significant differences between the feature vectors contained in the same category (for example, features with excessively different morphologies are mixed into the same "new type"), the system will split the category or mark it for manual review. At the same time, the system will also perform a secondary comparison between the potential new category and the known bad types already in the main visual library to confirm that it is indeed significantly different from all existing types (similarity is below a preset threshold), reducing the probability of duplicate entries into the library.

[0070] Optionally, after clustering noise reduction and repeatability verification, new defect types that are confirmed to be real and stable will be batch-integrated into the main visual library. The integration operation includes: adding the defect feature vector of the new type to the visual library, and updating the defect judgment rules so that subsequent board separation quality inspection can identify this new type of defect.

[0071] By setting the above, only consistently occurring undesirable features can enter the cache pool, ensuring that the vision library remains effective during long-term operation and will not gradually become invalid due to various interference factors in the industrial field, thus enhancing the long-term robustness of the system.

[0072] It is important to understand that the depaneling optimization method based on defect detection includes loading a visual library corresponding to the circuit board to be processed. This visual library includes the reference point features of the circuit board, images of successful depaneling, and known defective features. The step of loading the visual library corresponding to the target circuit board further includes: Read the circuit board's preset parameters; If the circuit board is a new type, import a preset number of standard images, automatically extract features, and generate a unique visual library template.

[0073] Optionally, before loading the vision library, the control module first reads the preset parameters of the circuit board to be processed. These preset parameters can be obtained in various ways: automatically identifying the board type by reading the identification code (such as a QR code or barcode) on the circuit board, obtaining the board type information of the current batch of products from the production management device, or being manually entered by the operator. The preset parameters of the circuit board include, but are not limited to: board type number, board thickness, material, number and theoretical coordinates of reference points, component layout information, preset cutting speed range, preset tool pressure range, and positioning accuracy requirements. These parameters form the basis for subsequent vision library generation and board separation parameter settings. For example, the preset parameters are stored in the control module in the form of configuration files, with different configuration files corresponding to different board types. The control module automatically loads the corresponding configuration file based on the read board type identifier.

[0074] With the above setup, this device can accurately compare the acquired circuit board image with a pre-stored standard template to calculate the actual position coordinate deviation of the circuit board. Without this standardized reference benchmark, visual positioning would lack a reference point and would be unable to achieve high-precision deviation calculation.

[0075] Optionally, when the control module determines that the board type of the circuit board to be processed does not have a corresponding template in the vision library (i.e., the circuit board is a new board type), the device enters the initialization mode and automatically generates a dedicated vision library template for the new board type, without the need to retrain the algorithm model or perform complex manual annotation. After the initialization is completed, the newly generated dedicated vision library template is stored in the vision library for direct loading and use during subsequent board separation operations of the same board type. This process does not require retraining the algorithm model or manual feature annotation, achieving rapid adaptation to the new board type.

[0076] By importing a preset number (e.g., 3-5) of standard images and automatically extracting features, a custom visual library template for the new pattern type can be quickly generated. This process requires no retraining of the algorithm model or manual annotation, significantly shortening the preparation time for launching the new pattern type and improving the production line's ability to respond quickly to multiple product varieties.

[0077] This application also proposes a board separation device, including a control module; the control module is used to execute the board separation method as described in any of the preceding claims.

[0078] Optionally, the control module can be implemented using a main controller, such as an MCU, PLC, DSP (Digital Signal Processor), SOC (System on Chip), or FPGA (Field Programmable Gate Array). The control module can establish bidirectional communication connections with both the sensing and execution modules via industrial Ethernet, forming a closed-loop control architecture of "sensing-decision-execution".

[0079] Optionally, since the technical solution and its effects of the plate separating device have been described in detail in the description of the plate separating method, those skilled in the art can understand the specific working principle of the device, so it will not be repeated here.

[0080] The above description is merely an exemplary embodiment of this application and does not limit the patent scope of this application. Any equivalent structural transformations made based on the technical concept of this application and the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the patent protection scope of this application.

Claims

1. A depaneling optimization method based on defective product detection, characterized in that, The method includes: After performing the action of cutting and separating the target circuit board, an image of the separation cut is obtained, and the image of the separation cut is compared with a preset qualified separation image template in the vision library. If the current board-making result is determined to be a defective product, the defect features are extracted from the cut image of the defective product, the defect features are compared with the existing defect features in the visual library, and new types of defect features are identified. The processing parameters for the separation process are updated based on the type of defective feature, and the separation action is performed according to the processing parameters.

2. The board separation optimization method based on defective product detection as described in claim 1, characterized in that, The step of comparing the image of the plate cutting edge with a preset qualified plate image template in the visual library includes: If the burr length at the depanel cut exceeds a preset length value, the depaneled circuit board is determined to be defective; or, If the offset value of the board separation cut is greater than the preset offset value, the board after separation is determined to be a defective product.

3. The board separation optimization method based on defective product detection as described in claim 1, characterized in that, The step of extracting defect features from the cut image of the defective product when the current board-making result is determined to be a defective product, comparing the defect features with existing defect features in the visual library, and determining new types of defect features includes: For cut images identified as defective products, extract the contour features, morphological features, and positional features of the cuts, and generate corresponding defect feature vectors; The similarity is calculated based on the defective feature vector and the known types of defective feature vectors stored in the visual database; If the similarity is less than a preset similarity threshold, the defective feature is determined to be a new type. If the similarity is greater than a preset similarity threshold, the defective feature is determined to be of a known type.

4. The board separation optimization method based on defective product detection as described in claim 1, characterized in that, The processing parameters for the plate separation include tool pressure, positioning error threshold for tool entry determination, and cutting speed. The step of updating the processing parameters for the plate separation based on the type of defect features includes: If the type of defect is tool wear, reduce the tool pressure; When the defective feature is of the plate position offset type, the positioning error threshold for determining the cutting edge is reduced from the first error threshold to the second error threshold; If the defective feature is of the burr type, reduce the cutting speed.

5. The board separation optimization method based on defective product detection as described in claim 4, characterized in that, After the steps of updating the processing parameters of the board and performing the board separation action according to the processing parameters, the method further includes: If the yield rate of the processed circuit board is greater than or equal to the preset yield rate, maintain the updated board separation parameters; If the yield rate is less than the preset yield rate, restore the board separation parameters to the parameter values ​​before the update.

6. The board separation optimization method based on defective product detection as described in claim 1, characterized in that, The step of updating the processing parameters of the partition plate based on the type of defect features further includes: Add the new types of defective features to the visual library and update the defective product judgment rules; The step of adding the new type of defective features to the visual library and updating the defective product judgment rules includes: If a new feature is detected in a cut image, and the same new type of defective feature is detected in cut images for a consecutive preset number of frames, it is stored in the cache pool. When the preset production batch is reached, the defective features in the cache pool are clustered, denoised, and checked for repetition before being added to the visual library and the defective product judgment rules are updated.

7. A plate separating device, characterized in that, The slitting device includes a control module; the control module is used to execute the slitting optimization method based on defective product detection as described in any one of claims 1-6.

Images