Apparatus and method for determining defect type, and recording medium recording instructions

Abstract

A technique for determining a defect type is disclosed. An apparatus according to an aspect of the present disclosure may include: one or more processors; and one or more memories storing instructions and a machine learning model, the instructions, when executed by means of the one or more processors, causing the one or more processors to perform operations; wherein the one or more processors may perform the following operations: obtaining measurement data indicating a state of a solder paste printed on a plurality of printed circuit boards by a screen printer; generating a plurality of input images in one-to-one correspondence with the plurality of printed circuit boards, and indicating whether the measurement data satisfies a predetermined reference, the plurality of input images include a first input image corresponding to a first printed circuit board and a plurality of second input images corresponding to each printed circuit board printed with solder paste in sequence before the first printed circuit board; and determining a defect type occurring on at least one of the screen printer and the first printed circuit board from the first input image and the plurality of second input images based on the machine learning model.

Description

Technical Field

[0001] The present disclosure relates to techniques for determining defect types. Background Art

[0002] The process of printing solder paste on a printed circuit board (PCB) can be performed by a screen printer. In the process of printing solder paste, when at least one of the screen printer and the printed circuit board is defective, the printed circuit board will be defective. However, based on the defect occurring in one printed circuit board, it is difficult to accurately determine whether the corresponding defect is caused by the screen printer or the printed circuit board. Summary of the invention

[0003] Technical issues

[0004] The present disclosure provides techniques for determining defect types.

[0005] Technical Solution

[0006] An apparatus according to one aspect of the present disclosure may include: one or more processors; and one or more memories storing instructions and machine learning models, wherein the instructions, when executed with the aid of the one or more processors, enable the one or more processors to perform operations; wherein the one or more processors may perform the following operations: obtaining measurement data indicating a state of solder paste printed on a plurality of printed circuit boards by a screen printer; generating a plurality of input images, the plurality of input images corresponding one-to-one to the plurality of printed circuit boards and indicating whether the measurement data meets a predetermined benchmark, the plurality of input images including a first input image corresponding to a first printed circuit board and a plurality of second input images corresponding to each printed circuit board on which solder paste is sequentially printed before the first printed circuit board; and determining, based on the machine learning model, from the first input image and the plurality of second input images a type of defect occurring on at least one of the screen printer and the first printed circuit board.

[0007] In one embodiment, the one or more processors may perform the following operations: generating first data analyzing the state of solder paste printed on the multiple printed circuit boards based on the timing changes between the first input image and each pattern in the multiple second input images; generating second data analyzing the position of solder paste printed on the multiple printed circuit boards based on the first input image and the pattern of each of the multiple second input images; and determining the type of defect occurring on at least one of the screen printer and the first printed circuit board based on the first data and the second data.

[0008] In one embodiment, the one or more processors may determine, based on the first data, that a defect has occurred on the screen printing machine when the same pattern appears periodically between the first input image and the multiple second input images, or determine that a defect has occurred on the first printed circuit board when the same pattern does not appear periodically between the first input image and the multiple second input images.

[0009] In one embodiment, the one or more processors may correspond to determining that a defect occurs on the screen printing machine, and determine the type of defect occurring on the screen printing machine from the second data based on the machine learning model.

[0010] In one embodiment, the first data may include data indicating whether the same pattern appears between the first input image and the multiple second input images and the period of appearance of the same pattern, and the second data may include data on the pattern shape for each of the first input image and the multiple second input images.

[0011] In one embodiment, the one or more processors may segment the measurement data into multiple data corresponding to each printed circuit board, generate multiple images indicating whether the solder paste printed on each printed circuit board meets a predetermined benchmark based on the segmented multiple data, and change the size of the multiple images to a predetermined size and generate the multiple input images.

[0012] In one embodiment, the one or more processors may determine a first probability of a first defect type occurring and a second probability of a second defect type occurring on the screen printer or the first printed circuit board based on the pattern of the first input image and each of the plurality of second input images.

[0013] In one embodiment, it may also include: a communication circuit, which is communicatively connected to the substrate inspection device; the communication circuit can obtain the measurement data from the substrate inspection device, and the measurement data indicates the state of the solder paste printed on the multiple printed circuit boards.

[0014] In one embodiment, the defect types may include at least one of scraper defects, scraper fixing defects, template lower end bracket defects, defects caused by poor fixture contact and damage, defects caused by poor solder paste placement and solder paste modulation, defects caused by insufficient solder paste in specific parts, defects caused by poor contact between the template and the substrate, template design defects, and substrate design defects.

[0015] In one embodiment, the one or more processors may train the machine learning model through a plurality of images having at least one pattern, wherein the at least one pattern is determined based on whether the solder paste printed on each of a plurality of printed circuit boards classified according to at least one defect type of the screen printer meets a predetermined benchmark.

[0016] In one embodiment, the machine learning model may be trained based on whether the solder paste printed on each of the learning printed circuit boards having different sizes and shapes satisfies a predetermined benchmark.

[0017] In one embodiment, the machine learning model may be a transformer model.

[0018] A method according to another aspect of the present disclosure may include: a step of obtaining measurement data, the measurement data indicating a state of solder paste printed on multiple circuit boards by a screen printer; a step of generating multiple input images, the multiple input images corresponding one-to-one to the multiple printed circuit boards and indicating whether the measurement data meets a predetermined benchmark, the multiple input images including a first input image corresponding to a first printed circuit board and multiple second input images corresponding to each printed circuit board on which solder paste is printed sequentially before the first printed circuit board; and a step of determining a type of defect occurring on at least one of the screen printer and the first printed circuit board from the first input image and the multiple second input images based on a machine learning model stored in one or more memories.

[0019] In one embodiment, the step of determining the defect type may include: a step of generating first data analyzing the state of the solder paste printed on the multiple printed circuit boards based on the timing changes between the first input image and each pattern in the multiple second input images; a step of generating second data analyzing the position of the solder paste printed on the multiple printed circuit boards based on the first input image and the pattern of each of the multiple second input images; and a step of determining the defect type occurring in at least one of the screen printer and the first printed circuit board based on the first data and the second data.

[0020] In one embodiment, the step of determining the defect type may include: based on the first data, when the same pattern appears periodically between the first input image and the multiple second input images, determining that a defect occurs on the screen printing machine; when the same pattern does not appear periodically between the first input image and the multiple second input images, determining that a defect occurs on the first printed circuit board.

[0021] In one embodiment, the step of determining the defect type may include: corresponding to determining that a defect occurs in the screen printing machine, determining the defect type occurring in the screen printing machine from the second data based on the machine learning model.

[0022] In one embodiment, the first data may include data indicating whether the same pattern appears between the first input image and the multiple second input images and the period of appearance of the same pattern, and the second data may include data on the pattern shape for each of the first input image and the multiple second input images.

[0023] In one embodiment, the step of determining the defect type may also include: determining a first probability of a first defect type occurring on the screen printer or the first printed circuit board and a second probability of a second defect type occurring based on the first input image and the pattern of each of the multiple second input images.

[0024] In one embodiment, the machine learning model can be trained by multiple images, wherein the multiple images have at least one pattern, and the at least one pattern is determined based on whether the solder paste printed on each of a plurality of printed circuit boards classified according to at least one defect type of the screen printer meets a predetermined benchmark.

[0025] The computer-readable recording medium of another aspect of the present disclosure is a non-temporary computer-readable recording medium having recorded thereon instructions, wherein the instructions, when executed with the aid of one or more processors, cause the one or more processors to perform operations, wherein the instructions can cause the one or more processors to perform the following steps: a step of obtaining measurement data, wherein the measurement data indicates a state of solder paste printed on a plurality of printed circuit boards by a screen printer; a step of generating a plurality of input images, wherein the plurality of input images correspond one-to-one to the plurality of printed circuit boards and indicate whether the measurement data satisfies a predetermined benchmark, wherein the plurality of input images include a first input image corresponding to a first printed circuit board and a plurality of second input images corresponding to each printed circuit board on which solder paste is sequentially printed before the first printed circuit board; and a step of determining a defect type occurring on at least one of the screen printer and the first printed circuit board from the first input image and the plurality of second input images based on a machine learning model stored in one or more memories.

[0026] Effects of the Invention

[0027] According to various embodiments of the present disclosure, the type of defect occurring in at least one of a screen printer and a printed circuit board can be accurately grasped. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1The block diagram briefly shows the structure of an apparatus for determining a defect type according to an embodiment of the present disclosure.

[0029] Figure 2 The process of transforming measurement data into an input image by a device according to an embodiment of the present disclosure is shown.

[0030] Figure 3 1 is a structural diagram briefly showing the structure of a machine learning model according to an embodiment of the present disclosure.

[0031] Figure 4 An embodiment in which a device according to an embodiment of the present disclosure determines a defect type based on a plurality of input images is shown.

[0032] Figure 5 An embodiment in which a device according to an embodiment of the present disclosure determines a defect type based on a plurality of input images is shown.

[0033] Figure 6 An embodiment in which a device according to an embodiment of the present disclosure determines a defect type based on a plurality of input images is shown.

[0034] Figure 7 An embodiment in which a device according to an embodiment of the present disclosure determines various defect types based on a plurality of input images is shown.

[0035] Figure 8 The present invention is a sequence diagram of a method for determining defect types using a plurality of printed circuit boards according to an embodiment of the present invention.

[0036] Fig. 9 The present invention is a sequence diagram of an embodiment in which a device according to an embodiment of the present disclosure determines a defect type based on first data and second data. DETAILED DESCRIPTION

[0037] The various embodiments described in this document are provided for the purpose of clearly describing the technical concept of the present disclosure, and are not intended to be limited to a specific implementation form. The technical concept of the present disclosure includes various modifications, equivalents, alternatives, and embodiments of selectively combining all or part of the embodiments described in this document. In addition, the scope of rights of the technical concept of the present disclosure is not limited to the various embodiments or specific descriptions thereof suggested below.

[0038] Including technical or scientific terms, as long as they are not defined differently, the terms used in this document may have the meaning generally understood by one of ordinary skill in the art to which the present disclosure belongs.

[0039] Expressions such as "including", "may include", "have", "may have", "have", "may have", etc. used in this document mean that the features that become the object (e.g., function, action or component, etc.) exist, and do not exclude the existence of other additional features. In other words, such expressions should be understood as open-ended terms that include the possibility of including other embodiments.

[0040] Singular expressions used in this document may include plural expressions unless the context indicates otherwise. This also applies to singular expressions recorded in the claims.

[0041] As used in this document, the expressions "first", "second", or "the first", "the second", etc., when referring to multiple objects of the same kind, are used to distinguish one object from other objects, unless the context indicates otherwise, and do not limit the order or importance of the corresponding objects.

[0042] As used in this document, expressions such as "A, B, and C", "A, B or C", "A, B and / or C", or "at least one of A, B, and C", "at least one of A, B, or C", or "at least one of A, B, and / or C" may refer to each listed item or all possible combinations of the listed items. For example, "at least one of A or B" may refer to all of (1) at least one A, (2) at least one B, or (3) at least one A and at least one B.

[0043] The expression "unit" as used in this document may refer to software or hardware components such as FPGA (field-programmable gate array) and ASIC (application specific integrated circuit). However, "unit" is not limited to hardware and software. "Unit" can be configured to be stored in an addressable storage medium or to run one or more processors. In one embodiment, "unit" may include components such as software components, object-pointing software components, cluster components, and task components, as well as processors, functions, attributes, programs, subroutines, fragments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, workbenches, displays, and variables.

[0044] The expression "based on..." used in this document is used to describe one or more factors that influence the decision, judgment, behavior or action described in the sentence or article containing the corresponding expression, and the expression does not exclude additional factors that influence the corresponding decision, judgment, behavior or action.

[0045] When used in this document, the expression that a certain component (for example, the first component) is “connected” or “connected to” another component (for example, the second component) means not only that the certain component is directly connected or connected to the other component, but also that the connection or connection is carried out with a new other component (for example, the third component) as the medium.

[0046] The expression "configured to" used in this document may have the meaning of "set to", "having the ability to", "changed to", "formed to", "capable of performing", etc., depending on the context. The corresponding expression is not limited to the meaning of "specially designed in hardware". For example, a processor configured to perform a specific operation may mean a generic-purpose processor that can perform its specific action by running software.

[0047] The following describes various embodiments of the present disclosure with reference to the accompanying drawings. In the drawings and descriptions of the drawings, the same or substantially equivalent components may be given the same reference numerals. In addition, in the following descriptions of various embodiments, repeated descriptions of the same or corresponding components may be omitted, but this does not mean that the corresponding components are not included in the embodiment.

[0048] Figure 1 A block diagram briefly illustrates the structure of an apparatus 100 for determining a defect type according to an embodiment of the present disclosure. In one embodiment, the apparatus 100 may include a memory 101, a processor 102, and / or a communication circuit 103. In addition, the apparatus 100 may further include an output device 104 for outputting a result processed by the processor 102. In a certain embodiment, at least one of these components of the apparatus 100 may be omitted, or other components may be added to the apparatus 100. In a certain embodiment, a part of the components may be implemented in an integrated manner or in an alternative manner, or may be implemented as a single or multiple individuals.

[0049] At least some of the internal and external components of the device 100 (for example, the memory 101, the processor 102 and the communication circuit 103) can be interconnected and send and receive data and / or signals through a bus, GPIO (General Purpose Input / Output), SPI (Serial Peripheral Interface) or MIPI (Mobile Industry Processor Interface).

[0050] The device 100 may be a device of various forms. For example, the device 100 may be a portable communication device, a computer device, a portable multimedia device, a wearable device, a home appliance device, or a device that is one of the above devices or a combination of the above devices. The device 100 of the present disclosure is not limited to the above devices.

[0051] The processor 102 may drive software to control at least one component of the device 100 connected to the processor 102. In addition, the processor 102 may perform various operations, processes, data generation, processing, etc. related to the present disclosure. In addition, the processor 102 may load data from the memory 101 or store it in the memory 101.

[0052] The memory 101 can store various data. The data stored in the memory 101, as data obtained, processed or used by at least one component of the electronic device 100, may include software. The memory 101 may include volatile and / or nonvolatile memory.

[0053] The communication circuit 103 can perform wireless or wired communication between the device 100 and other external electronic devices. For example, the communication circuit 103 can perform wireless communication based on LTE (long-term evolution), LTE-A (LTE Advance), CDMA (code division multiple access), WCDMA (wideband CDMA), WiBro (Wireless Broadband), WiFi (wireless fidelity), Bluetooth, NFC (near field communication), GPS (Global Positioning System) or GNSS (global navigation satellite system). For example, the communication circuit 103 can perform wired communication based on USB (universal serial bus), HDMI (high definition multimedia interface), RS-232 (recommended standard 232) or POTS (plain old telephone service). The communication circuit 103 may be omitted from the apparatus 100 depending on the embodiment.

[0054] The following is a brief description of the process of surface mount technology (SMT) for printing printed circuit boards using a screen printer 120. The screen printer 120 may include an operating table, a template (or a metal mask), and a squeegee blade. After the printed circuit board enters the screen printer 120 and is fixed on the operating table, it can be configured so that the template is unfolded on the printed circuit board. The template may include a plurality of openings, which allow solder paste to be printed only in predetermined areas of the printed circuit board. After arranging the template on the printed circuit board in such a manner that it is located on the pads of the printed circuit board corresponding to the template openings, the screen printer 120 prints solder paste on the template and uses a squeegee blade to push the solder paste to the openings of the template, thereby enabling soldering on the pads on the printed circuit board. The screen printer then separates the template from the printed circuit board.

[0055] The shape of solder paste printed on a printed circuit board can be inspected by SPI (solder paste inspection) technology. SPI technology is a technology that obtains a two-dimensional or three-dimensional image of solder paste printed on a printed circuit board through optical technology and inspects the shape of solder paste printed on the printed circuit board from the obtained image.

[0056] During the printing process of the printed circuit board as described above, when various types of defects occur in the screen printer 120 or the printed printed circuit board has design defects, the quality of the printed circuit board will be low. In one embodiment, at least one defect type of the screen printer 120 may be at least one defect type that causes abnormal solder paste printed on the printed circuit board among the various defect types that may occur in the screen printer 120. For example, at least one defect type of the screen printer 120 may include at least one of the following defects: a scraper defect of the screen printer 120, a scraper fixing defect (e.g., a screw fixing defect for fixing the scraper, etc.), a support defect of the screen printer 120, a workbench defect of the screen printer 120, a defect caused by poor grid frame setting when a grid frame is used as a support, a defect caused by the state of the solder paste supplied by the screen printer 120, for example, a defect caused by poor placement of the solder paste, poor solder paste modulation, a defect caused by insufficient solder paste supplied by the screen printer 120, and a defect caused by poor template contact. In one embodiment, the design defects of the printed circuit board may include bending and twisting of the printed circuit board, poor impedance, etc.

[0057] However, this is only for description and not limitation, and various defect types of the screen printer 120 that may cause abnormal solder paste printed on the printed circuit board may include at least one defect type of the screen printer 120 .

[0058] The printed circuit board inspection device 110 (solder paste inspector; SPI) (hereinafter referred to as the inspection device) can inspect the printed circuit board and generate various data related to the quality of the printed circuit board. For example, the inspection device 110 can obtain at least one of the following data: whether the solder paste printed on multiple pads of the printed circuit board is excessive volume or insufficient volume, the standard deviation of the printed solder paste, the degree of offset from the alignment position of the printed circuit board and the template, and the process capability index (Capability of Process, Katayori; CPK) of the printed circuit board process. The data generated by the inspection device 110 when inspecting the printed circuit board is referred to as measurement data. The inspection device 110 can inspect multiple printed circuit boards and generate measurement data corresponding to each printed circuit board.

[0059] The processor 102 can communicate with the inspection device 110 using the communication circuit 103 and receive measurement data from the inspection device 110. For example, the processor 102 can receive a plurality of measurement data of a plurality of printed circuit boards from the inspection device 110 and store them in the memory 101. The processor 102 can use the obtained measurement data to determine the type of defect that occurs on the screen printer 120 or the printed circuit board. According to one embodiment, the processor 102 can use the obtained measurement data to determine whether there is a problem with the quality of the printed circuit board. When it is determined that there is a problem with the quality of the printed circuit board, the obtained measurement data can be used to determine the type of defect that occurs on the screen printer 120 or the printed circuit board.

[0060] The following describes a process by which the processor 102 determines the type of defect that occurred on the screen printer 120 or the printed circuit board.

[0061] The processor 102 may generate an image for indicating whether the obtained measurement data meets a predetermined benchmark. According to one embodiment, the input image may be an image indicating whether the solder paste printed on the plurality of pads included in the printed circuit board is over-welded or under-welded. The image indicating whether the solder paste printed on the pad of the printed circuit board is over-welded or under-welded is referred to as the input image below. For example, the processor 102 may generate an input image based on the measurement data, the input image including a pattern for indicating the position of the pad on the printed circuit board where the solder paste is over-welded or under-welded among the plurality of pads of the printed circuit board. When the solder paste is not over-welded or under-welded in the printed circuit board, the pattern may not appear in the input image corresponding to the corresponding printed circuit board. For example, the input image corresponding to the printed circuit board where the solder paste is not over-welded or under-welded may be a black (or white) image. However, the input image is not limited thereto, and may be an image indicating the various states of the solder paste printed on the printed circuit board included in the measurement image.

[0062] The processor 102 may generate an input image of a predetermined size while maintaining the aspect ratio of the printed circuit board. When the input image has the same size as the actual size of the printed circuit board, the size of the data processed by the processor 102 may be too large, and thus the processor 102 may generate an input image of a size smaller than the actual size of the printed circuit board. According to one embodiment, the processor 102 may generate an input image corresponding to a printed circuit board having a variety of sizes and patterns. For example, the processor 102 may generate a first input image corresponding to a first printed circuit board having a first area and a first set of pads, and may generate a second input image corresponding to a second printed circuit board having a second area different from the first area and a second set of pads different from the first set of pads. The areas of the first printed circuit board and the second printed circuit board may be different, but the areas of the first input image and the second input image may be the same.

[0063] The memory 101 can store a machine learning model. The machine learning model can be trained to derive at least one value representing the correlation between at least one defect type of the screen printer 120 or the printed circuit board and the input image. The machine learning model can be, for example, a transformer model. In addition, the machine learning model can also be stored in the memory 101 of an external device (e.g., an external server, etc.) that is linked to the processor 102 in a wired or wireless manner. At this time, the processor 102 can send and receive information with the external device that is linked in a wired or wireless manner, and the information is used to determine the defect type associated with at least one solder paste that is sensed to be abnormal.

[0064] The at least one value generated by the processor 102 indicating the correlation between at least one defect type and the input image may be a probability value. For example, the at least one value may indicate that the correlation between the input image and the first defect type of the screen printer 120 or the printed circuit board is a%, and the correlation between the input image and the second defect type of the screen printer 120 or the printed circuit board is b%. That is, the processor 102 may determine, based on the input image, a plurality of defect types that have the possibility of having occurred on the screen printer 120 or the printed circuit board. However, this is for description only and not limitation, and a variety of values ​​that can represent correlation may be used.

[0065] The machine learning model stored in the memory 101 may be trained by a plurality of input images classified according to at least one defect type of the screen printer 120 or the printed circuit board. For example, the machine learning model may be trained by a plurality of input images classified as a first defect type of the screen printer 120 or the printed circuit board, and a plurality of input images classified as a second defect type of the screen printer 120. If the machine learning model trained as described above is input with a plurality of input images, a value indicating the correlation between the input image and at least one of the first defect type and the second defect type may be output.

[0066] The processor 102 may obtain a plurality of measurement data of a plurality of printed circuit boards printed sequentially, and generate input images corresponding to the plurality of printed circuit boards respectively based on the plurality of measurement data. For example, the processor 102 may generate a first input image of a first printed circuit board that is printed last, and generate a plurality of second input images of a plurality of second printed circuit boards that are printed before the first printed circuit board. According to one embodiment, the plurality of second input images may be used to sense whether the solder paste printed on the first printed circuit board is abnormal. For example, the processor 102 may determine the type of defect that occurs on the screen printer 120 or the first printed circuit board during the printing process of the first printed circuit board based on the timing changes of the plurality of second input images and the first input image.

[0067] The processor 102 can generate first data representing the time-series change of the input image pattern and second data representing the location where over-welding or under-welding of solder paste occurs in multiple printed circuit boards from the first input image and multiple second input images. The processor 102 can generate the first data by analyzing the pattern change of the input images of multiple printed circuit boards sequentially input to the screen printer 120 in time sequence. The processor 102 can generate the second data based on each pattern appearing in the multiple input images, and the second data represents the location where over-welding or under-welding of solder paste occurs on the printed circuit board.

[0068] The processor 102 may use a machine learning model to determine the type of defect that occurs on the screen printer 120 or the printed circuit board based on the first input image, the plurality of second input images, the first data, and the second data. After the machine learning model obtains the first input image and the plurality of second input images, it may derive at least one value representing the correlation between at least one type of defect of the screen printer 120 or the printed circuit board and the input image. According to one embodiment, when the same pattern periodically appears in the first input image and the plurality of second input images, the processor 102 may determine that the screen printer 120 has a defect. According to one embodiment, when the same pattern appears in both the first input image and the plurality of second input images, it may be determined that the screen printer 120 has a defect and the same problem continues to occur. For example, when the same pattern appears in both the first input image and the plurality of second input images, the processor 102 may determine that the screen printer 120 has a bracket defect. For example, when the same pattern appears in the first input image and the odd-numbered (or even-numbered) input images in the plurality of second input images, the processor 102 may determine that the screen printer 120 has a scraper defect. According to another embodiment, when the same pattern does not appear periodically in the first input image and the plurality of second input images, the processor 102 may determine that a temporary defect has occurred on the first printed circuit board or the screen printer. For example, when the pattern appears only in the first input image but not in the plurality of second input images, the processor 102 may determine that a design defect has occurred on the first printed circuit board. A temporary defect means a defect that occurs only during the printing of a predetermined number of printed circuit boards, and may occur in the first printed circuit board or the screen printer for various reasons. For example, a temporary defect may occur when the first printed circuit board has a design defect or when the screen printer prints in a state where the solder is not sufficiently modulated. When the solder is not sufficiently modulated in the screen printer, after several printings, the temporary defect may not occur after the solder is modulated.

[0069] According to one embodiment, the processor 102 may train the machine learning model using a plurality of input images and data mapping defect types corresponding to each input image. According to one embodiment, the training of the machine learning model may be performed offline, and the plurality of input images and data mapping defect types corresponding to each input image may be obtained from a previously performed circuit board printing process.

[0070] Figure 2The process of converting measurement data into an input image by an apparatus of an embodiment of the present disclosure is shown. The measurement data can be generated by combining information of multiple printed circuit boards in a time sequence according to the order of inspection by the inspection device. The measurement data may include data measured for each of multiple solder pastes printed on each printed circuit board. For example, the measurement data may include data on the printing amount of multiple solder pastes on multiple pads included in each printed circuit board and whether there is over-soldering / under-soldering.

[0071] The measurement data of the first printed circuit board is referred to as the first measurement data 211, and the measurement data of the plurality of second printed circuit boards is referred to as the second measurement data 212. For example, the first measurement data 211 and the second measurement data 212 may be data in which a plurality of solder pastes printed on a printed circuit board are arranged according to the area between the printed pads. For example, the first measurement data 211 may be data arranged in sequence from the data of the solder paste printed on the pad with the largest area to the data of the solder paste printed on the pad with the smallest area in the first printed circuit board. However, this is only for description and not limitation. The information for each measurement of the plurality of solder pastes included in the first measurement data 211 and the second measurement data 212 may be arranged according to the identification information of the plurality of pads printed with solder pastes, may be arranged in time sequence according to the inspection order of each of the plurality of solder pastes, or may be arranged randomly without other references.

[0072] The processor may compare the first measurement data 211 and the second measurement data 212. The processor may determine the difference between the first measurement data 211 and the second measurement data 212 by comparing the first measurement data 211 and the second measurement data 212. In addition, the processor may determine the portion 213 that exceeds the set range from the determined difference. The processor may determine that the solder paste printed on at least one pad corresponding to the portion 213 that exceeds the set range determined has over-soldering or under-soldering.

[0073] After the inspection of the first printed circuit board is completed, the processor may also add the first measurement data 211 to the second measurement data 212, thereby updating the second measurement data 212. For example, the processor may update the second measurement data 212 only when the multiple solder pastes printed on the first printed circuit board are not over-welded or under-welded. As another example, the processor may update the second measurement data 212 even when at least one of the multiple solder pastes printed on the first printed circuit board is over-welded or under-welded.

[0074] The processor may transform the first measurement data 211 and the second measurement data 212 into an input image 220 for inputting into a machine learning model. According to one embodiment, the processor may generate an input image 220 including a pattern 221 based on the measurement data, wherein the pattern 221 indicates the position of a pad on the printed circuit board where over-soldering or under-soldering of solder paste occurs among a plurality of pads of the printed circuit board. For example, the input image 220 may include a pattern 221 having the topography of a corresponding printed circuit board and indicating the state of solder paste printed on a pad included in the printed circuit board.

[0075] According to one embodiment, the processor may determine the pattern 221 of the input image 220 according to whether the solder paste is over-welded or under-welded. For example, in the case where all pads of the printed circuit board are not over-welded or under-welded with solder paste, the processor may determine the input image 220 as an image with one color (e.g., black) and no other pattern 221. On the contrary, in the case where a portion of the pads of the printed circuit board are over-welded or under-welded with solder paste, the processor may determine to display the pattern 221 of the input image 220 to distinguish the pad positions where the solder paste is over-welded or under-welded. According to one embodiment, when determining the pattern 221 of the input image 220, the processor may display the over-welding of the solder paste and the under-welding of the solder paste as different colors respectively. For example, in the case where the solder paste is over-welded in the first pad of the printed circuit board and the solder paste is under-welded in the second pad, the processor may display the position of the first pad as a first color in the input image 220, and display the position of the second pad as a second color distinguished from the first color.

[0076] Figure 3 The following is a diagram briefly showing the structure of the machine learning model of an embodiment of the present disclosure. Figure 2 The generated plurality of input images 300 are input to the machine learning model 314. The processor may utilize the machine learning model 314 to determine the type of defect occurring on at least one of the screen printer and the printed circuit board based on the plurality of input images 300.

[0077] In order to determine the defect type occurring on at least one of the screen printer and the printed circuit board, the processor may utilize a DETR (Detection with Transformer) structure 310 including a machine learning model 314. For example, the machine learning model 314 may be a transformer model. The processor may input a plurality of input images 300 to the machine learning model 314, and obtain information 320 of the defect type occurring on at least one of the screen printer and the printed circuit board as a result of the plurality of input images 300 being processed in the machine learning model 314.

[0078] The processor may determine a plurality of input images 300 for inputting into the machine learning model 314. In order to accurately determine the type of defect occurring on at least one of the screen printer and the printed circuit board, the processor may analyze the changes in the plurality of input images 300 over time. According to one embodiment, the processor may input a predetermined number (e.g., 9) of input images 300 into the machine learning model 314. For example, in order to analyze the defect of at least one of the screen printer and the printed circuit board that has an impact during the printing of the first printed circuit board, the processor may use the input images 300 of a plurality of second printed circuit boards printed by the screen printer before printing the first printed circuit board. In order to improve the accuracy of determining the type of defect occurring on at least one of the screen printer and the printed circuit board, the processor may further increase the number of second printed circuit boards. According to one embodiment, the processor may determine the number of second printed circuit boards based on external input.

[0079] The processor may transform the plurality of input images 300 into a feature map form 311 that can be processed by the machine learning model 314. The processor may generate first data 312 of temporal changes of the plurality of input images 300. According to an embodiment, the first data 312 may be data obtained by analyzing temporal changes of patterns of the plurality of input images 300. For example, the first data 312 may include data of whether the same pattern appears between the plurality of input images 300 and the period of occurrence of the same pattern.

[0080] The processor may generate second data 313 for pattern shapes of the plurality of input images 300. According to an embodiment, the second data 313 may be data for pad positions where over-soldering or under-soldering of solder paste occurs in the printed circuit board based on the pattern of each of the plurality of input images 300.

[0081] The processor may input the input image transformed into the feature map form 311 to the machine learning model 314 to determine the defect type occurring in at least one of the screen printer and the printed circuit board. For example, the machine learning model 314 may determine the defect type of at least one of the screen printer and the printed circuit board that has an impact on the printing process of the first printed circuit board based on the first data 312 and the second data 313.

[0082] The machine learning model 314 can refer to the first data 312 to determine where the defect occurs in the screen printer and the printed circuit board, and output information 320 of the defect type. According to one embodiment, when the same pattern appears repeatedly in multiple input images 300, it can be determined that a defect occurs on the screen printer. For example, when the same pattern appears in all input images 300, or the same pattern appears in the odd (or even) input images in the input images 300, it can be determined that a defect occurs on the screen printer. For example, when the same pattern appears in the odd (or even) input images in the input images 300, the machine learning model 314 can determine that a scraper defect or a bracket defect has occurred. According to another embodiment, when the same pattern does not appear repeatedly in multiple input images 300, it can be determined that a temporary defect has occurred on the printed circuit board or the screen printer. For example, when a specific pattern appears in only one input image among the multiple input images 300, the machine learning model 314 can determine that a design defect of the printed circuit board has occurred.

[0083] According to an embodiment, the machine learning model 314 can accurately determine the defect type that occurs in at least one of the screen printer and the printed circuit board based on the second data 313. For example, when it is determined that a defect occurs in the screen printer, the machine learning model 314 can refer to the second data 313 and determine that the screen printer has a first defect type when the pattern is a first shape, and determine that the screen printer has a second defect type when the pattern is a second shape.

[0084] Figure 4 The present invention shows an embodiment in which a device according to an embodiment of the present invention determines a defect type based on multiple input images. The processor can determine the defect type based on multiple input images. The processor can determine the defect type based on whether the same pattern appears repeatedly in the multiple input images. Figure 4 In FIG. 1 , an embodiment of determining the defect type using 8 input images is shown, but the embodiments of the present invention are not limited thereto. Figure 4 Among the plurality of input images 400 and 402, the 8th printed circuit board corresponding to the input image No. 8 is described as the most recently printed printed circuit board. According to one embodiment, the processor may obtain information about defect types from the machine learning model 314, and use the obtained information to generate charts 401 and 403 representing defect types that occur on the screen printer or the printed circuit board when printing each printed circuit board.

[0085] The machine learning model 314 can determine the type of defect that occurs on at least one of the screen printer and the printed circuit board based on the first data and the second data generated based on the input multiple input images. For example, the multiple input images 400 input to the machine learning model 314 can all have the same pattern. The machine learning model 314 receives the input of the multiple input images 400 and determines that a defect occurs on the screen printer based on the first data, and determines that a support defect of the screen printer occurs based on the second data.

[0086] According to another embodiment, the same pattern may not appear in the multiple input images 402 input to the machine learning model 314. For example, among the multiple input images 402 input to the machine learning model 314, only input image No. 8 has a specific pattern, and the remaining input images do not have the pattern, or other patterns may appear. The machine learning model 314 may determine that a temporary defect has occurred based on the first data and the second data. For example, it may be determined that a design defect has occurred in the printed circuit board.

[0087] Figure 5 The present invention shows an embodiment in which a device according to an embodiment of the present invention determines a defect type based on a plurality of input images. Figure 5 In the embodiment of determining the defect type using 8 input images, the embodiment of the present invention is not limited thereto. In addition, among the multiple input images 500 and 510, the 8th printed circuit board corresponding to the 8th input image is described as the most recently printed printed circuit board.

[0088] The machine learning model 314 may determine the type of defect that occurs on at least one of the screen printer and the printed circuit board based on the plurality of input images 500, 510 that are input. Figure 5 , the same pattern may appear repeatedly in multiple input images 500. For example, the same pattern may appear repeatedly in each odd-numbered (or even-numbered) input image. The processor may generate first data and second data based on multiple input images 500. The first data may include information indicating that the same pattern appears repeatedly in the odd-numbered (or even-numbered) input images in the multiple input images 500. The second data may include information that the pattern that appears repeatedly in the multiple input images 500 has a shape that is arranged vertically. The machine learning model 314 may determine that a defect has occurred on the screen printing machine based on the first data, and may determine the type of defect that has occurred on the screen printing machine based on the second data. For example, the machine learning model 314 may determine that the type of defect that has occurred on the screen printing machine is a scraper defect. The machine learning model 314 may output information indicating that a scraper defect has occurred on the screen printing machine, and the processor may obtain corresponding information to determine that a scraper defect has occurred.

[0089] For the case where the pattern occurs in only one of the input images 510 input to the machine learning model 314, Figure 4 Therefore, the repeated contents will not be repeated here.

[0090] Figure 6 The present invention shows an embodiment in which a device according to an embodiment of the present invention determines a defect type based on multiple input images. The processor may generate first data based on temporal changes of patterns appearing in multiple input images 600 and second data related to the shape of the pattern. The processor may use a machine learning model 314 to determine the defect type based on the first data and the second data. In the past, since the defect type was determined based on one input image, the first data could not be considered to determine the defect type. However, by referring to the temporal changes of the patterns included in the multiple input images 600, the defect type may be determined more accurately. For example, referring to Figure 6 , when the same pattern appears in the even-numbered input images 600a, 600b, 600c, and 600d of the multiple input images 600, if the temporal change of the pattern is taken into account, the defect type can be accurately determined. Without considering the first data 610, it is wrongly determined that a defect caused by poor solder paste placement and solder paste modulation has occurred, but when considering the first data 620, the defect type can be accurately determined as a scraper defect. That is, if the temporal change of the pattern of each input image 600 is taken into account, the processor can more accurately determine the defect type of the same input image 600.

[0091] Figure 7 The present invention shows an embodiment in which a device according to an embodiment of the present invention determines various defect types based on multiple input images. The processor can use multiple input images to determine various types of defects that occur on the screen printer and the printed circuit board. For example, when the first pattern appears in all of the multiple input images 700 in the same manner, the processor can use the machine learning model 314 to determine that a scraper defect occurs in the screen printer.

[0092] In another embodiment, when a second pattern appears in an even-numbered (or odd-numbered) input image among the plurality of input images 701 , the processor may determine, using the machine learning model 314 , that a scraper defect has occurred on the screen printing machine.

[0093] In another embodiment, when the third pattern appears in only one input image among the plurality of input images 702 , the processor may use the machine learning model 314 to determine that the screen printer is not defective but a design defect occurs on the printed circuit board.

[0094] In another embodiment, when the fourth pattern appears identically in all of the plurality of input images 703 , the processor may utilize the machine learning model 314 to determine that a defect caused by poor placement and modulation of the solder paste has occurred on the screen printer.

[0095] In another embodiment, when the fifth pattern appears identically in all of the multiple input images 704 , the processor may use the machine learning model 314 to determine that a stent defect has occurred on the screen printing machine.

[0096] The embodiments of the present invention are not limited to the examples mentioned above, and the processor may determine that at least one of various defect types occurs based on a pattern appearing in an input image corresponding to a printed circuit board and a temporal change in patterns of a plurality of input images.

[0097] Various embodiments of the present disclosure may be implemented in a machine-readable storage medium as software. The software may be software for implementing various embodiments of the present disclosure. The software may be inferred from various embodiments of the present disclosure by a programmer skilled in the art of the present disclosure. For example, the software may be a program including machine-readable instructions (e.g., code or code fragments). A machine is a device that can operate according to instructions extracted from a storage medium, such as a computer. In one embodiment, the machine may be the device 100 of an embodiment of the present disclosure. In one embodiment, the processor of the machine may run the extracted instructions, and the constituent elements of the machine may perform functions corresponding to the corresponding instructions. In one embodiment, the processor may be the processor 102 of an embodiment of the present disclosure. The storage medium may refer to all kinds of recording media (recording medium) that store data readable by a machine. The storage medium may include, for example, ROM (read-only memory), RAM (random access memory), CD-ROM (read-only optical disk drive), magnetic tape, floppy disk, optical data storage device, etc. In one embodiment, the storage medium may be a memory 101. In one embodiment, the storage medium may also be implemented in a distributed form in a computer system connected via a network. The software may be distributed, stored, and run in a computer system, etc. The storage medium may be a non-transitory storage medium. A non-transitory storage medium refers to a tangible medium that exists independently of semi-permanent or temporary storage of data, and does not include a signal that is transmitted transiently.

[0098] Figure 8This is a sequence diagram of a method for determining defect types using multiple printed circuit boards according to an embodiment of the present disclosure. In step 800, the device 100 may receive measurement data indicating the state of the solder paste from the inspection device 110. The device 100 may obtain measurement data from the inspection device 110, and the measurement data may include information about the printed circuit board printed by the screen printer 120. The measurement data may include information about the amount of solder paste printed on the printed circuit board and whether there is over-soldering or under-soldering. According to one embodiment, the measurement data may be data of multiple solder pastes printed on a printed circuit board arranged according to the area size of the printed pads.

[0099] In step 802, the device 100 may generate a plurality of input images for indicating whether the measurement data meets a predetermined benchmark. The input image may be an image indicating whether the solder pastes respectively printed on a plurality of pads included in the printed circuit board are over-welded or under-welded. The device 100 may compare the first measurement data with the second measurement data, and determine the difference between the first measurement data and the second measurement data. The device 100 may determine the portion exceeding the set range from the determined difference. The device 100 may determine that the solder paste printed on at least one pad corresponding to the portion exceeding the set range has over-welded or under-welded. For example, the device 100 may generate an input image including a pattern based on the measurement data, the pattern indicating the position of the pads on the printed circuit board where the solder paste has been over-welded or under-welded among the plurality of pads of the printed circuit board. When the solder paste has not been over-welded or under-welded in the printed circuit board, the pattern may not appear in the input image corresponding to the corresponding printed circuit board.

[0100] According to an embodiment, the apparatus 100 may generate an input image having a predetermined size while maintaining the same aspect ratio as the aspect ratio of the printed circuit board. For efficient data processing, the apparatus 100 may generate the size of the input image to be smaller than the size of the actual printed circuit board. According to an embodiment, the apparatus 100 may generate an input image of a printed circuit board having different sizes and pad sets.

[0101] In step 804, the apparatus 100 may input the first input image and the plurality of second input images to the machine learning model 314. The memory of the apparatus 100 may store the machine learning model 314. The machine learning model 314 may be trained to derive at least one value representing the correlation between at least one defect type of the screen printer 120 or the printed circuit board and the input image. The machine learning model 314 may be trained by a plurality of input images classified according to at least one defect type of the screen printer 120 or the printed circuit board. The machine learning model 314 may be, for example, a transformer model.

[0102] In step 806, the device 100 may use the machine learning model 314 to determine the defect type of at least one of the screen printer 120 and the printed circuit board. The device 100 may generate first data representing the temporal change of the input image pattern and second data representing the positions where over-welding or under-welding of solder paste occurs in the plurality of printed circuit boards from the first input image and the plurality of second input images. The device 100 may generate the first data by analyzing the pattern changes of the input images of the plurality of printed circuit boards sequentially input to the screen printer 120 in a temporal manner. The device 100 may generate the second data based on the patterns respectively appearing in the plurality of input images, the second data representing the positions where over-welding or under-welding of solder paste occurs on the printed circuit board.

[0103] After obtaining the first input image and the plurality of second input images, the machine learning model 314 may derive at least one value representing the correlation between at least one defect type of the screen printer 120 or the printed circuit board and the input image. According to one embodiment, when the same pattern periodically appears in the first input image and the plurality of second input images, the device 100 may determine that a defect has occurred on the screen printer 120. According to one embodiment, when the same pattern appears in both the first input image and the plurality of second input images, it may be determined that a defect has occurred on the screen printer 120 and the same problem continues to occur. According to another embodiment, when the same pattern does not periodically appear in the first input image and the plurality of second input images, the device 100 may determine that a temporary defect has occurred on the first printed circuit board or the screen printer.

[0104] Fig. 9 The present invention is a sequential diagram of an embodiment of an apparatus according to an embodiment of the present invention determining a defect type based on first data and second data. In step 900, the apparatus 100 may generate first data based on a temporal change between a pattern of each of a first input image and a plurality of second input images. For example, the first data may include information on whether the same pattern appears repeatedly between the plurality of input images and the period of the repeated occurrence.

[0105] In step 902, the apparatus 100 may generate second data analyzing solder paste positions of a plurality of printed circuit boards based on the first input image and the pattern of each of the plurality of second input images. The second data may include information on the shape of the pattern included in the plurality of input images.

[0106] In step 904, the apparatus 100 may determine a defect type of at least one of the screen printer 120 and the first printed circuit board based on the first data and the second data. The apparatus 100 may determine the defect type based on the first data and the second data using the machine learning model 314.

[0107] exist Figure 8 and Fig. 9In the sequence diagram shown, each step of the method or algorithm of the present disclosure is described in sequence, but in addition to being performed in sequence, each step can also be performed in an order that can be arbitrarily combined in the present disclosure. The description based on this sequence diagram does not exclude the situation where changes or revisions are imposed on the method or algorithm, and does not mean that any step is necessary or recommended. In one embodiment, at least a portion of the steps can be performed in parallel, repeatedly or exploratory. In one embodiment, at least a portion of the steps can be omitted, or other steps can be added.

[0108] The above describes the technical concept of the present disclosure based on various embodiments, but the technical concept of the present disclosure includes various substitutions, deformations and changes that can be realized within the scope understandable to those skilled in the art of the present disclosure. In addition, such substitutions, deformations and changes should be understood to be included in the attached claims.

Claims

1. A device for determining a defect type, comprising: More than one processor; and One or more memories storing instructions and machine learning models, wherein the instructions, when executed by means of the one or more processors, cause the one or more processors to perform operations; The one or more processors perform the following operations: obtaining measurement data indicating a condition of solder paste printed on a plurality of printed circuit boards by a screen printer; generating a plurality of input images, the plurality of input images corresponding one-to-one to the plurality of printed circuit boards and indicating whether the measurement data satisfies a predetermined benchmark, the plurality of input images comprising a first input image corresponding to a first printed circuit board and a plurality of second input images corresponding to respective printed circuit boards on which solder paste is sequentially printed before the first printed circuit board; and Based on the machine learning model, a type of defect occurring on at least one of the screen printer and the first printed circuit board is determined from the first input image and the plurality of second input images.

2. The device according to claim 1, wherein: The one or more processors perform the following operations: generating first data analyzing a state of solder paste printed on the plurality of printed circuit boards based on a time-series change between each pattern in the first input image and the plurality of second input images; generating second data analyzing positions of solder paste printed on the plurality of printed circuit boards based on the first input image and the pattern of each of the plurality of second input images; and Based on the first data and the second data, a type of defect occurring in at least one of the screen printer and the first printed circuit board is determined.

3. The device according to claim 2, wherein: The one or more processors determine a type of defect occurring on the screen printer or the first printed circuit board based on the first data and based on whether the same pattern periodically appears between the first input image and the plurality of second input images.

4. The device according to claim 3, wherein: The one or more processors correspond to determining that a defect occurs on the screen printing machine, and determining the type of defect occurring on the screen printing machine from the second data based on the machine learning model.

5. The device according to claim 2, wherein: The first data includes data for indicating whether the same pattern appears between the first input image and the plurality of second input images and a period during which the same pattern appears. The second data includes data of a pattern shape for each of the first input image and the plurality of second input images.

6. The device according to claim 1, wherein: The one or more processors divide the measurement data into a plurality of data respectively corresponding to one printed circuit board, generating a plurality of images indicating whether the solder paste printed on each printed circuit board satisfies a predetermined reference based on the divided plurality of data, The sizes of the plurality of images are changed to a predetermined size to generate the plurality of input images.

7. The device according to claim 1, wherein: The one or more processors determine a first probability of a first defect type occurring and a second probability of a second defect type occurring on the screen printer or the first printed circuit board based on the first input image and the pattern of each of the plurality of second input images.

8. The device according to claim 1, wherein: Also included: a communication circuit, the communication circuit being communicatively connected to the substrate inspection device; The communication circuit obtains the measurement data from the substrate inspection device, and the measurement data indicates the state of the solder paste printed on the plurality of printed circuit boards.

9. The device according to claim 1, wherein: The defect types include at least one of scraper defects, scraper fixing defects, template lower end bracket defects, defects caused by poor fixture contact and damage, defects caused by poor solder paste placement and solder paste preparation, defects caused by insufficient solder paste in specific parts, defects caused by poor contact between the template and the substrate, template design defects, and substrate design defects.

10. The device according to claim 1, wherein: The one or more processors train the machine learning model using a plurality of images, The plurality of images have at least one pattern determined based on whether solder paste printed on each of a plurality of printed circuit boards classified according to at least one defect type of the screen printer satisfies a predetermined reference.

11. The device according to claim 10, wherein: The machine learning model is a model trained based on whether the solder paste printed on each of the learning printed circuit boards having different sizes and shapes satisfies a predetermined benchmark.

12. The device according to claim 10, wherein: The machine learning model is a transformer model.

13. A method for determining a defect type, comprising: the step of obtaining measurement data indicating a state of solder paste printed on a plurality of printed circuit boards by a screen printer; a step of generating a plurality of input images, the plurality of input images corresponding one-to-one to the plurality of printed circuit boards and indicating whether the measurement data satisfies a predetermined benchmark, the plurality of input images comprising a first input image corresponding to a first printed circuit board and a plurality of second input images corresponding to respective printed circuit boards on which solder paste is sequentially printed before the first printed circuit board; and The step of determining a defect type occurring on at least one of the screen printer and the first printed circuit board from the first input image and the plurality of second input images based on a machine learning model stored in one or more memories.

14. The method according to claim 13, wherein: The step of determining the defect type comprises: a step of generating first data analyzing a state of solder paste printed on the plurality of printed circuit boards based on a time-series change between each pattern in the first input image and the plurality of second input images; generating second data analyzing positions of solder pastes printed on the plurality of printed circuit boards based on the first input image and the pattern of each of the plurality of second input images; and The step of determining a type of defect occurring in at least one of the screen printer and the first printed circuit board based on the first data and the second data.

15. The method according to claim 14, wherein: The step of determining the defect type comprises: A step of determining a defect type occurring on the screen printer or the first printed circuit board based on the first data and based on whether the same pattern periodically appears between the first input image and the plurality of second input images.

16. The method according to claim 15, wherein: The step of determining the defect type comprises: Corresponding to determining that a defect has occurred in the screen printing machine, a step is performed to determine the type of defect that has occurred in the screen printing machine from the second data based on the machine learning model.

17. The method according to claim 14, wherein: The first data includes data for indicating whether the same pattern appears between the first input image and the plurality of second input images and a period during which the same pattern appears. The second data includes data of a pattern shape for each of the first input image and the plurality of second input images.

18. The method according to claim 13, wherein: The step of determining the defect type further comprises: A first probability of a first defect type occurring on the screen printer or the first printed circuit board and a second probability of a second defect type occurring on the screen printer or the first printed circuit board are determined based on the first input image and the pattern of each of the plurality of second input images.

19. The method according to claim 13, wherein: Training the machine learning model with multiple images, The plurality of images have at least one pattern determined based on whether solder paste printed on each of a plurality of printed circuit boards classified according to at least one defect type of the screen printer satisfies a predetermined reference.

20. A computer-readable recording medium, the computer-readable recording medium being a non-transitory computer-readable recording medium having recorded thereon instructions, the instructions, when executed by means of one or more processors, causing the one or more processors to perform operations, wherein: The instructions cause the one or more processors to perform the following steps: the step of obtaining measurement data indicating a state of solder paste printed on a plurality of printed circuit boards by a screen printer; a step of generating a plurality of input images, the plurality of input images corresponding one-to-one to the plurality of printed circuit boards and indicating whether the measurement data satisfies a predetermined benchmark, the plurality of input images comprising a first input image corresponding to a first printed circuit board and a plurality of second input images corresponding to respective printed circuit boards on which solder paste is sequentially printed before the first printed circuit board; and The step of determining a defect type occurring on at least one of the screen printer and the first printed circuit board from the first input image and the plurality of second input images based on a machine learning model stored in one or more memories.

Images