KR-20260062821-A - METHOD, APPARATUS AND COMPUTER PROGRAM FOR QUALITY INSPECTION OF PRINTED CIRCUIT BOARDS (PCB) BASED ON DEEP LEARNING AND COMPUTER VISION ALGORITHMS

KR20260062821AKR 20260062821 AKR20260062821 AKR 20260062821AKR-20260062821-A

Abstract

A method for inspecting the quality of a printed circuit board (PCB) based on deep learning and computer vision algorithms is performed by a computing device and includes the steps of acquiring an image generated by photographing at least a portion of a printed circuit board (PCB) and inspecting the quality of the printed circuit board by analyzing the acquired image using an artificial neural network model.

Inventors

김준명
이영석
강정운
정근오
탕 칭
김민철
정하일
박정윤

Assignees

주식회사 인터엑스

Dates

Publication Date: 20260507
Application Date: 20250820

Claims (14)

In a quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms performed by a computing device, A step of acquiring an image generated by photographing at least a portion of a printed circuit board (PCB); and A step of inspecting the quality of the printed circuit board by analyzing the above-mentioned acquired image using an artificial neural network model Includes, The step of inspecting the quality of the above printed circuit board is, A step of generating a first segmented image by segmenting a portion corresponding to a defective area as a region of interest from the acquired image as a result of analyzing the acquired image through a first segmentation model; and A step of deriving a first segmentation result including information on the type and size of the defective area included in the first segmentation image and the first segmentation image. A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by including
In paragraph 1, The above first division model is, Characterized by generating training data by marking defective areas on a plurality of images generated by photographing a printed circuit board determined to be defective, and generating training data by using the generated training data according to a machine learning-based learning method. Quality inspection method for printed circuit boards (PCBs) based on deep learning and computer vision algorithms.
In paragraph 1, The above method is, A step of generating a second segmented image by segmenting a portion corresponding to a metal part as a region of interest from the acquired image through a second segmentation model, and deriving a second segmentation result including information regarding the type and size of the metal part included in the generated second segmented image and the generated second segmented image. A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, further comprising
In paragraph 3, The above second division model is, Characterized by generating training data by marking metal parts on a plurality of images generated by photographing a printed circuit board, and generating training based on a machine learning-based learning method using the generated training data. Quality inspection method for printed circuit boards (PCBs) based on deep learning and computer vision algorithms.
In paragraph 1, If the type of the above defective area is any one of contamination, discoloration, lack of plating, or dents in the ball pad area, A step of converting the above-mentioned acquired image into black and white; A step of cropping the ball pad area from the above-mentioned black-and-white converted image; A step of generating a plurality of clusters by clustering pixel values included in the cropped ball pad area; and A step of finally determining whether the printed circuit board is a good product or a defective product by comparing the ratio of the area of the cluster corresponding to the risk area to the area of the cluster corresponding to the normal area among the plurality of clusters generated above with the threshold ratio. A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by further including
In paragraph 5, The above clustering utilizes the K-means clustering method, and is characterized in that the plurality of clusters are generated into three types: a cluster corresponding to a normal area, a cluster corresponding to a defective area, and a cluster corresponding to a background area. Quality inspection method for printed circuit boards (PCBs) based on deep learning and computer vision algorithms.
In paragraph 1, If the type of the above defective part is either surface peeling or foreign matter, A step of analyzing the acquired image to identify defective areas and detecting the maximum pixel value within an area corresponding to the identified defective areas; A step of generating a plurality of clusters by clustering pixel values included within an area corresponding to the identified defective area when the detected maximum pixel value is greater than or equal to a threshold pixel value; and A step of finally determining the printed circuit board as a good product or a defective product based on whether the location of the identified defective part exists within the cluster corresponding to the epoxy region among the generated plurality of clusters. A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by further including
In Paragraph 7, The step of detecting the maximum pixel value above is, A step of converting the above-mentioned acquired image into black and white; A step of histogram analysis of the grayscale converted image and, if the most frequent pixel value is less than a reference value, increasing the brightness of the acquired image; and Step of detecting the maximum pixel value from the image with the brightness increased A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by including
In paragraph 1, If the type of the above defective area is partial exposure of the cheek pad, A step of analyzing the acquired image to identify a ball pad and setting a circle including an area corresponding to the identified ball pad; and A step of finally determining the printed circuit board as good or defective based on comparing the difference between the area of the identified ball pad and the area of the set circle with the threshold difference. A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by further including
In paragraph 3, If the type of the above metal part is metal exposure in the solder resist area, A step of converting the above-mentioned acquired image into black and white; A step of identifying defective areas from the above-mentioned grayscale converted image, and clustering pixel values included in the identified defective areas to generate a plurality of clusters; and If, among the plurality of clusters generated above, there exists at least one remaining cluster that does not correspond to a reference cluster generated by photographing the solder resist area of a printed circuit board determined to be a good product, the step of finally determining the printed circuit board as a good product or a defective product based on whether the average value of the pixels included in the remaining cluster is greater than or equal to a predetermined reference value. A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by further including
In Paragraph 10, The step of finally determining whether the above printed circuit board is a good product or a defective product is, A step of calculating the average value of pixels included in each of the plurality of clusters generated above; A step of selecting the cluster with the highest average value among the plurality of clusters generated above, and detecting the maximum pixel value within the selected cluster; and A step of determining that the defective area is a bump pad when the detected maximum pixel value is less than the threshold pixel value, and finally determining that the printed circuit board is a good product. A quality inspection method for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by further including
In paragraph 1, Prior to the step of generating the first segmented image, The method further includes a step of determining whether the acquired image is diluted or if the defect is spread throughout the image to determine whether the acquired image is segmentable, and is characterized by performing a step of generating the segmented image if the result of determining whether the image is segmentable indicates that the image is segmentable. Quality inspection method for printed circuit boards (PCBs) based on deep learning and computer vision algorithms.
processor; Memory; and It includes a computer program that is loaded into the memory and executed by the processor, The above computer program comprises: an instruction for acquiring an image generated by photographing at least a portion of a printed circuit board (PCB); and It includes an instruction for inspecting the quality of the printed circuit board by analyzing the acquired image using an artificial neural network model, and The instruction for inspecting the quality of the above printed circuit board is, An instruction to generate a first segmented image by segmenting a portion corresponding to a defective area as a region of interest from the acquired image as a result of analyzing the acquired image through a first segmentation model; and Instruction for deriving a first segmentation result including information on the type and size of the defect area included in the first segmentation image and the first segmentation image. A quality inspection device for a printed circuit board (PCB) based on deep learning and computer vision algorithms, characterized by including
A computer program that is combined with a computing device and stored on a recording medium readable by the computing device to execute the method of any one of claims 1 to 12.

Description

Method, apparatus and computer program for quality inspection of printed circuit boards (PCBs) based on deep learning and computer vision algorithms The present disclosure relates to a method, apparatus, and computer program for quality inspection of a printed circuit board (PCB) based on deep learning and computer vision algorithms. With the growth of the semiconductor industry and increasing demand, the need for automated systems in the inspection of Printed Circuit Boards (PCBs) is emerging. Previously, PCBs were primarily inspected visually by operators; however, due to the complexity of the boards and the diversity of defect types, situations where accurate inspection was difficult frequently occurred. Consequently, interest in inspection automation is growing, and in particular, automated vision inspection systems based on deep learning and computer vision algorithms are being widely researched. Various types of defects exist in the PCB manufacturing process. Representative defects include surface peeling, foreign substances, and metal exposure; detecting these defects is challenging because some of them are visually similar or difficult to distinguish from good products. Such failure to detect defects ultimately results in defective products leaking into the market, which leads to catastrophic losses given the nature of the semiconductor industry. Therefore, it is crucial to reduce the leakage of defective products by automating PCB quality inspections. Semiconductor PCB substrates have a highly complex structure and diverse components. Furthermore, even for the same type of defect, whether a product is classified as good or defective depends on the location where the defect occurs. To address this complexity, existing deep learning-based PCB inspection systems have attempted to improve defect detection performance simply by strengthening inspection criteria. However, this approach leads to the problem of over-detecting products that are actually good as defective, and due to the similarity between defect types, defects are still being released without being detected. In particular, the following four major types of misclassification are typical in PCB board inspection. (1) No defect detected in the ball pad area Ball pads on a PCB board are captured as circular shapes in vision images, and defect types occurring on the ball pads include contamination, discoloration, incomplete plating, and dents. Since all of these defect types are circular in shape and their colors are similar to the background, it is difficult to detect them. As a result, defects are frequently leaked. (2) Surface peeling - Misclassification of foreign substances Surface peeling and foreign substances on the PCB substrate surface appear very similar in vision images as white dots on a black background. As a result, not only is it difficult for inspectors to classify them visually, but deep learning models also fail to accurately distinguish them, leading to a high likelihood of misclassification. (3) Defect in the form of partial exposure of the ball pad The partial ball pad exposure type is a defect where only a portion of the ball pad is exposed on the PCB board. Deep learning-based inspection systems incorrectly recognize this type as an internal structure of the PCB board and fail to classify it as a defect. Consequently, instances of such defects being leaked are occurring. (4) Metal exposure - Bump pad confusion If deep surface delamination occurs in the solder resist area of a PCB board, the metal portion becomes exposed. This metal exposure appears in a form similar to bump pads, which are structures within the PCB board, potentially causing confusion during the inspection process. Such confusion can lead to the release of defective products and is particularly likely to act as a critical defect. To solve these problems, more sophisticated inspection techniques using deep learning and computer vision algorithms are required. This necessitates a new inspection method that can more accurately distinguish various types of defects that may occur within PCB boards and prevent defect leakage. The aforementioned background technology is one that the inventor possessed or acquired in the process of deriving the contents of the disclosure of the present application, and it cannot be considered as prior art disclosed to the general public prior to the filing of this application. The aforementioned background technology is one that the inventor possessed or acquired in the process of deriving the contents of the disclosure of the present application, and it cannot be considered as prior art disclosed to the general public prior to the filing of this application. The following drawings attached to this specification illustrate preferred embodiments of the present disclosure and serve to further enhance understanding of the technical concept of the present disclosure together with the detailed description of the invention; therefore, t