Mounting system

Abstract

To provide a mounting system capable of coping with needs of both of high quality and high-speed production.SOLUTION: A mounting system comprises a production management apparatus and a mounting line having at least one surface mounting machine. The production management apparatus takes an image of an article associated with production of substrates as input and uses a learning model according to machine learning to determine quality of the article. The surface mounting machine comprises a head unit for placing components on the substrates and a controller. The controller can perform switching to a first mounting mode or a second mounting mode. The first mounting mode is a mode in which for quality determination using the learning model, component placing is continued without waiting return of a quality determination result based on the machine learning from the production management apparatus after transmitting the image of the article to the production management apparatus. The second mounting mode is a mode in which for quality determination using the learning model, component placing is performed after waiting return of the quality determination result based on the machine learning from the production management apparatus after transmitting the image of the article to the production management apparatus.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] The present disclosure relates to a mounting system for mounting components on a substrate. [Background technology]

[0002] A surface mounter photographs components to be mounted on a printed circuit board and judges whether the components are good or bad. Patent Document 1 discloses that the quality of components is judged by machine learning. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-110257 A Summary of the Invention [Problem to be solved by the invention]

[0004] Compared to other judgment methods, machine learning has a higher judgment accuracy but takes a long time to make a judgment. Therefore, when machine learning is used to judge the quality of parts, although it is expected to improve quality, it may decrease the production efficiency of boards. For these reasons, it has become an issue to be able to respond to the needs of both high-quality production and high-speed production. Note that this issue is not limited to when judging the quality of parts, but also applies to when judging the quality of marks on a board or a base. [Means for solving the problem]

[0005] (1) A mounting system includes a production management device and a mounting line having at least one surface mounter. The production management device receives an image of an item related to the production of a circuit board and judges the quality of the item using a learning model based on machine learning. The surface mounter includes a head unit that mounts components on a board and a control device, and the control device is capable of switching between a first mounting mode and a second mounting mode. The first mounting mode is a mode in which, after transmitting an image of the item to the production management device for pass / fail judgment using a learning model, mounting is continued without waiting for a return of a pass / fail judgment result based on machine learning from the production management device. The second mounting mode is a mode in which, after transmitting an image of the item to the production management device for pass / fail judgment using a learning model, component mounting is performed after waiting for a return of a pass / fail judgment result based on machine learning from the production management device.

[0006] (2) In the mounting system described in (1), when the surface mounter receives from the production management device, after mounting a component on a board, a result of pass / fail judgment by machine learning that the component mounted on the board is a defective product during the selection of the first mounting mode, the surface mounter may take different error measures according to the working status of the board on which the defective product is mounted. With this configuration, it is possible to take appropriate measures according to the working status of the board on which the defective product is mounted.

[0007] (3) In the mounting system described in (2), when a board carrying a defective product has completed the mounting process and is in the process of being transported to a reflow device, the surface mounter may stop transporting the board and call an operator.

[0008] (4) A mounting system as described in (2), wherein the mounting system is configured to share and execute the operation of mounting components on a single board among a plurality of surface mounting machines connected in series, and if a board carrying a non-returnable product is present in the mounting process, the surface mounting machine may work in cooperation with a transport conveyor to return the board to the surface mounting machine that performed the mounting work of the non-returnable product, and then use a head unit to recover the component determined to be a non-returnable product and remount another component on the board.

[0009] (5) In the mounting system described in (2), the surface mounter may stop transporting the board and call an operator when the board carrying the defective product has completed the reflow process and is about to be removed from the mounting line.

[0010] (6) The mounting system described in (2) may have a plurality of mounting lines each of which shares the work on one board, and a surface mounter on an upstream mounting line may send a skip command to the downstream mounting line after a board carrying a defective product has been discharged to a downstream mounting line, so that the work on the downstream mounting line may be skipped for the board carrying the defective product.

[0011] (7) In the mounting system according to (1) or (2), the surface mounter may have a first RAM, and the production management device may have a second RAM. The surface mounter may store an image to be judged for pass / fail in the first RAM, and the production management device may read the image to be judged for pass / fail directly from the first RAM to the second RAM using RDMA technology. This configuration eliminates the bottleneck in the image transfer speed, and allows image data to be transferred in a short time.

[0012] (8) In the mounting system described in (1) or (2), the article may be configured to include at least one of a component to be mounted on a substrate, a mounting head that adsorbs the component to be mounted on the substrate, a mark affixed to the substrate, and a mark affixed to a base. Effect of the Invention

[0013] According to the present invention, it is possible to meet the needs for both high quality mounting and high speed mounting. [Brief description of the drawings]

[0014] [Figure 1] Overall view of the mounting system [Diagram 2] Plan view of surface mounter [Diagram 3] Side view of surface mounter [Figure 4] Block diagram of the implementation system [Diagram 5] Schematic diagram of deep learning [Figure 6] Diagram showing data exchange in the implementation system [Figure 7A] A chart summarizing methods for determining whether electronic components are good or bad [Figure 7B] Diagram of implementation modes [Figure 8] An illustration of the error processing performed when the first implementation mode is selected [Figure 9] An illustration of the error processing performed when the first implementation mode is selected [Figure 10] An illustration of the error processing performed when the first implementation mode is selected [Figure 11] An illustration of the error processing performed when the first implementation mode is selected [Figure 12] Block diagram of image data transfer function [Figure 13] A chart summarizing the pass / fail judgment results for each part condition [Figure 14] Flowchart showing the process of generating a learning model DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] <Embodiment 1> 1. Mounting System S Fig. 1 is an overall view of a mounting system S. The mounting system S includes a mounting line Lv and a production management device 40. The mounting line Lv includes, in order from the upstream side (the right side of Fig. 1), a printer 11, a surface mounter 12, an inspection machine 13, and a reflow machine 15, and these multiple machines 11, 12, 13, and 15 are connected in series via a transport conveyor.

[0016] The printer 11 is located upstream of the surface mounter 12 and prints solder paste onto the board P to be fed into the surface mounter 12 .

[0017] The surface mounter 12 is located downstream of the printer 11. The surface mounter 12 is a device that mounts electronic components E on the surface of the substrate P. There may be one or more surface mounters 12, and in this embodiment, a line configuration is provided in which three surface mounters 12 are arranged in series. The configuration of the surface mounter 12 will be described later. The electronic components E are an example of a component to be mounted on a substrate.

[0018] The inspection machine 13 is located downstream of the surface mounter 12. The inspection machine 13 is a device that inspects the board P on which the electronic components E are mounted before the board P is input into the reflow device 15.

[0019] The reflow device 15 is located downstream of the inspection machine 13. The reflow device 15 is a device that heats the inspected board P and melts the solder paste applied to the surface of the board P, thereby soldering the board P and the electronic component E together.

[0020] 2. Surface mount machine 2, the surface mounter 12 includes a base 21, a transport conveyor 22, a driving device 23, a head unit 26, a tape feeder F1, and a tray feeder F2. In the following description, the tape feeder F1 and the tray feeder F2 are collectively referred to as the feeder F.

[0021] In the following description, the longitudinal direction of base 21 (left-right direction in Figure 2) will be referred to as the X-direction, the depth direction of base 21 (up-down direction in Figure 2) will be referred to as the Y-direction, and the up-down direction in Figure 3 (the lifting and lowering direction of mounting head 26A) will be referred to as the Z-direction.

[0022] The transport conveyor 22 is disposed in the center of the base 21. The transport conveyor 22 includes a pair of transport belts 22A that are driven to circulate in the X direction, and transports the substrate P on the transport belts 22A in the X direction.

[0023] In this embodiment, the right side shown in Fig. 2 is the entrance, and the substrate P is carried into the machine from the right side shown in Fig. 2 via the transport conveyor 22. The carried-in substrate P is transported by the transport conveyor 22 to a working position in the center of the base.

[0024] On the base 21, tape feeders F1 and tray feeders F2 are installed at four locations around the work position of the substrate P.

[0025] The tape feeder F1 is a device that feeds a component supply tape (not shown) to a supply position to supply a tape component E1. The tape component E1 is, for example, a small electronic component such as a chip capacitor.

[0026] The tray feeder F2 is a device that supplies the tray parts E2 by drawing out a tray that contains the tray parts E2 from a storage cabinet. The tray parts E2 are, for example, medium to large electronic parts such as ICs. In the following description, the tape feeder F1 and the tray feeder F2 are collectively referred to as feeder F. The tape parts E1 and the tray parts E2 are collectively referred to as electronic parts E.

[0027] As shown in FIGS. 2 and 3, the driving device 23 is composed of a pair of Y beams 24, an X beam 25, a Y-axis ball screw 24A, a Y-axis motor 24B, an X-axis ball screw 25A, and an X-axis motor 25B.

[0028] 2, the pair of Y beams 24 are located on both sides in the X direction of the base 21. The pair of Y beams 24 extend parallel to the Y direction, and support the X beam 25 slidably in the Y direction.

[0029] A Y-axis ball screw 24A extending in the Y direction and a Y-axis motor 24B at its shaft end are disposed on the upper surface of the Y beam 24. The Y-axis ball screw 24A is a drive shaft in the Y-axis direction, and a ball nut (not shown) fixed to the lower surface of the X beam 25 is screwed into the Y-axis ball screw 24A.

[0030] By driving the Y-axis motor 24B to rotate the Y-axis ball screw 24A about its axis, the X-beam 25 and the head unit 26 can be moved in the Y direction (Y-axis servo mechanism).

[0031] X-beam 25 is elongated in the X-direction, and supports head unit 26 so that it can slide along the X-direction. An X-axis ball screw 25A extending in the X-direction and an X-axis motor 25B at the shaft end are arranged on the side of X-beam 25. X-axis ball screw 25A is a drive shaft in the X-axis direction, and a ball nut (not shown) fixed to head unit 26 is screwed into the X-direction.

[0032] By driving the X-axis motor 25B to rotate the X-axis ball screw 25A about its axis, the head unit 26 can be moved in the X direction relative to the X-beam 25 (X-axis servo mechanism).

[0033] From the above, it is possible to move head unit 26 to any position on base 21 by controlling X-axis motor 25B and Y-axis motor 24B in a composite manner.

[0034] The head unit 26 has a plurality of mounting heads 26A. The mounting heads 26A are arranged in a row in the X direction with respect to the head unit 26. Each mounting head 26A can move in the vertical direction by being driven by a Z-axis motor 26B (Z-axis servo mechanism).

[0035] The mounting head 26A is a shaft that is long in the Z direction, and has a suction nozzle at its tip (lower end). By switching the internal pressure of the suction nozzle between negative and positive pressure, the tape component E1 and the tray component E2 can be held or released.

[0036] With this configuration, by operating the X-axis motor 25B, the Y-axis motor 24B, and the Z-axis motor 26B at a predetermined timing, the tape components E1 supplied by the tape feeder F1 and the tray components E2 supplied by the tray feeder F2 can be picked up by the mounting head 26A and mounted on the substrate P.

[0037] Note that reference numeral 27 in Fig. 2 denotes a component recognition camera, and reference numeral 28 in Fig. 3 denotes a board recognition camera. Component recognition camera 27 is fixed on base 21 with its imaging surface facing upward, and captures images (bottom surface images) of tape components E1 and tray components E2 picked up by mounting head 26A.

[0038] The board recognition camera 28 is attached to the head unit 26 with its imaging surface facing downward, and can move together with the head unit 26 on the base 21. The board recognition camera 28 images each mark provided on the board P, the correction marks on the base 21, and the like.

[0039] Next, the electrical configuration of the surface mounter 12 will be described with reference to Fig. 4. The surface mounter 12 includes an arithmetic processing unit 30, a communication unit 31, a data storage unit 32, a program storage unit 33, a motor control unit 34, a recognition unit 35, a display unit 37, an operation unit 38, and the like.

[0040] The arithmetic processing unit 30 includes a CPU, a RAM, etc. The arithmetic processing unit 30 mounts the electronic component E on the substrate P by controlling the motor control unit 34 in accordance with a control program. The arithmetic processing unit 30 corresponds to the "control device" of the present invention.

[0041] The communication unit 31 is an interface for communicating information with the production management device 40. The communication unit 31 is communicably connected to the production management device 40 via a communication cable.

[0042] The recognition unit 35 is connected to the component recognition camera 27 and the board recognition camera 28. The recognition unit 35 performs image recognition of the object using the images output from the cameras 27 and 28, and executes pass / fail judgment of the object. The pass / fail judgment will be described in detail later.

[0043] The data storage unit 32 temporarily stores images output from the cameras 27 and 28. The captured images are used for image recognition by the recognition unit 35, and are also transmitted to the production management device 40 via a communication cable, where all images are saved. The program storage unit 33 stores execution programs for the first mounting mode and the second mounting mode, and a program for switching between the two mounting modes. It also stores a recognition algorithm for image recognition.

[0044] The display unit 37 is, for example, a display. The operation unit 38 is a user interface that allows the user to perform input operations.

[0045] 3. Configuration of the production management device 40 The production management device 40 is a device that manages the mounting line Lv, and is, for example, a personal computer. As shown in FIG. 4, the production management device 40 includes a communication unit 41, a control unit 42, a storage unit 43, a display unit 44, and an operation unit 45.

[0046] The communication unit 41 is connected to the communication unit 31 of the surface mounter 12 via a communication cable so that they can communicate with each other. Commands (instructions) and data can be transmitted and received between the surface mounter 12 and the production management device 40.

[0047] The control unit 42 includes a CPU, a RAM, etc. The storage unit 43 stores information on the board P produced by the surface mounter 12 and information on the electronic components E to be mounted on the board P. The storage unit 43 saves all images received from the surface mounter 12.

[0048] The display unit 44 is, for example, a display. The operation unit 45 is a user interface that allows the user to perform input operations.

[0049] 4. Judging the quality of electronic component E The surface mounter 12 picks up the electronic components E supplied by the feeder F using the mounting head 26A, then photographs them using a component recognition camera 27 and judges whether the electronic components E to be mounted on the board P are good or bad based on the photographed image.

[0050] There are two methods for determining the quality of electronic components E. The first method is a method using a recognition algorithm, and the second method is a method using a learning model M based on machine learning.

[0051] The recognition algorithm performs rule-based judgment based on pre-prepared rules. For example, it judges whether electronic component E is good or bad by comparing feature values ​​extracted from an image (such as the shape, position, and angle of the target object) with threshold values.

[0052] 5, the learning model M is a supervised learning model that uses images of electronic components E and pass / fail results as training data, and judges the pass / fail of the electronic components E by inputting images of the electronic components E. Specifically, it outputs the probability that the electronic components E are pass and the probability that they are fail.

[0053] In addition, an example of such a learning model M is a learning model based on deep learning using a neural network.

[0054] The generation and evaluation of the learning model M imposes a large computational load, and the computational capacity of the surface mounter 12 may be insufficient. Therefore, in this embodiment, as shown in FIG. 6, teacher data is sent from the surface mounter 12, and the learning model M is generated by the production management device 40.

[0055] Furthermore, when determining whether the electronic component E is good or bad (secondary determination) during the production of the board P, image data is sent from the surface mounter 12 to the production management device 40, and the production management device 40 determines whether the electronic component E is good or bad based on the learning model M.

[0056] 5. Implementation Mode The learning model M has higher judgment accuracy than the recognition algorithm, but the judgment time is long. Therefore, when the learning model M is used to judge the quality of the electronic components E, the production efficiency of the board P may decrease, although the quality of the board P can be improved.

[0057] In this embodiment, after the adsorbed electronic component E is photographed by the component recognition camera 27, an image of the electronic component E is sent to the production management device 40 in parallel with the first judgment using the recognition algorithm, and a second judgment is performed using the learning model M.

[0058] Then, when mounting the electronic component E to be judged on the board P, two mounting modes are set depending on whether or not to wait for the result of the second judgment from the production management device 40. The user can select the two mounting modes by operating the operation unit 39 etc. before starting production.

[0059] The first mounting mode is a mode in which, after transmitting an image of the electronic component E to the production management device 40 for pass / fail judgment, component mounting is continued without waiting for a return of the pass / fail judgment result by deep learning from the production management device 40. Note that component mounting is intended to take out the component E from the feeder F and mount it on the board P.

[0060] In the first mounting mode, the quality of the electronic components E is judged and then mounted based only on the result of the first judgment, without waiting for the result of the second judgment. Therefore, compared to the second mounting mode, the judgment time is shorter, and the production of the board P can be performed at high speed.

[0061] The second mounting mode is a mode in which an image of electronic component E is sent to the production management device 40 for pass / fail determination, and then component mounting is performed after waiting for a return of the pass / fail determination result based on deep learning from the production management device 40.

[0062] In the second mounting mode, since the first and second judgments can be used together, the accuracy of the pass / fail judgment of the electronic components E is high. Therefore, defective products can be detected with high accuracy before mounting on the board, so the occurrence frequency of defective boards is low and the production quality of the boards P is high.

[0063] When the first and second judgments are used in combination, if both judgments are OK (good), the electronic component E is judged as a good product, and in other cases it is judged as a defective product.

[0064] Furthermore, in this embodiment, when an image is sent from the surface mounter 12 to the production management device 40 for pass / fail judgment, the surface mounter 12 also transmits data (1) and (3) linked to the image to the production management device 40. In this manner, when the learning model M gives an NG judgment result, the production management device 40 can determine which surface mounter 12 and which board P the electronic component E mounted on is defective.

[0065] <Data transmitted from the surface mounter 12 to the production management device 40> (1) Board number (2) Images of electronic components to be mounted on the board (3) Information to identify the machine (if there are multiple surface mount machines)

[0066] 7.Effects As described above, the surface mounter 12 is set to two mounting modes, the first mounting mode and the second mounting mode, and the user can arbitrarily select and switch between these two mounting modes by inputting operations via the operation unit 38. Therefore, it is possible to meet the needs of both high-quality mounting and high-speed mounting.

[0067] <Embodiment 2> 8 and 9 are line configuration diagrams of the mounting line Lv. The mounting line Lv of the second embodiment has a plurality of surface mounters 12A to 12C connected in series, and includes a loader 10, a printer 11, three surface mounters 12A to 12C, a conveying device 14, and a reflow device 15.

[0068] The substrate P is sent out from the loader 10 and passes through the printer 11, the surface mounters 12A, 12B, 12C, the conveying device 14, and the reflow device 15 in that order, and undergoes the printing process (printer 11), the mounting process (surface mounters 12A to 12C), and the reflow process (reflow device 15).

[0069] As shown in FIG. 10, a conveying device 16 and an unloader 17 are installed downstream of the reflow device 15, and the board P is discharged from the unloader 17 to outside the mounting line after the reflow process is completed.

[0070] As described in the first embodiment, the surface mounter 12 can select between the first mounting mode and the second mounting mode. In the first mounting mode, components are mounted without waiting for the determination result of the learning model M.

[0071] Therefore, after the mounting work is completed in the mounting machine 12 and the board P is sent to the downstream machine, the surface mounting machine 12 may be notified by the production management device 40 that the electronic component E mounted on the board P is an NG product as a result of pass / fail judgment by the learning model M.

[0072] In this embodiment, when the first mounting mode is selected and a message is received from the production management device 40 indicating that the electronic component E mounted on the board P is a defective product as a result of pass / fail determination by deep learning after the electronic component E has been mounted on the board P, the arithmetic processing unit 30 of the surface mounter 12 takes different error measures depending on the work status of the board P on which the defective product is mounted. In this way, it is possible to take an appropriate error measure depending on the work status of the board P on which the defective product is mounted.

[0073] <Case 1> Figure 8 shows that during mounting work on board No. 2 P, surface mounter 12A is notified by production management device 40 that, as a result of pass / fail judgment by learning model M, "electronic component E linked to board No. 5 P is an NG product."

[0074] In this case, the fifth board P carrying the defective product is located in the downstream machine of three boards (fifth board - second board) from the surface mounter 12A as shown in FIG. 8, that is, the conveying device 14, and it can be determined that the board P carrying the defective product has already completed the mounting process and is in the process of being conveyed to the reflow device 15. In this way, the surface mounter 12A can grasp the working status of the board P carrying the defective product from the number of the board P carrying the defective product, the number of the board P being worked on, and the line configuration of the mounting line Lv. The surface mounter 12 may record the production history of the board P in the data storage unit 32 in association with the board number. By recording the production history, the production status of the board P can be easily grasped. Furthermore, the data of the line configuration of the mounting line Lv may be stored in the data storage unit 32 in advance and read out as necessary.

[0075] 8, the arithmetic processing unit 30 of the surface mounter 12A transmits a stop command to the transport equipment 14 to stop the transport of the fifth board P. Also, an operator call is made to notify that the fifth board P on which the defective product is mounted is stopped on the transport equipment 14.

[0076] This allows the operator to take measures such as retrieving the fifth board P from the transport device 14. The retrieved fifth board P may be discarded, or, since it has not yet been reflowed, it may be returned to the mounting line Lv after manual correction (removal of defective products and remounting normal components).

[0077] <Case 2> FIG. 9 shows that during mounting work on board No. 2 P, surface mounter 12A is notified by production management device 40 that, as a result of pass / fail judgment by learning model M, "component E linked to board No. 3 P is an NG product."

[0078] In this case, board No. 3 P, which is carrying the defective product, is located in the machine one board downstream (board No. 3 - board No. 2) from surface mounting machine 12A, that is, surface mounting machine 12B, and it can be determined that mounting work is currently underway.

[0079] In this case, the arithmetic processing unit 30 of the surface mounter 12A transmits a reverse command to each machine upstream of the surface mounter 12B that is mounting the third board P, that is, the surface mounter 12B, the surface mounter 12A, the printer 11, and the rotor 10. This causes the transport conveyors 22 of each of the machines 12B, 12A, 11, and 10 to rotate in the reverse direction, and sends the board P on each machine upstream in the reverse direction.

[0080] 9, when the third board P with the defective component mounted thereon is returned to the surface mounter 12A, the surface mounter 12A moves the head unit 26 onto the third board, recovers the defective component from the third board P, and remounts a normal component E. In this manner, the mounting of the defective component can be recovered without the occurrence of a defective board.

[0081] <Case 3> Figure 10 shows that during mounting work on board No. 2 P, surface mounter 12A is notified by production management device 40 that, as a result of a pass / fail judgment using learning model M, "component E linked to board No. 9 P is an NG product."

[0082] In this case, as shown in FIG. 10, the 9th board P carrying the defective product is located in a machine 7 boards downstream (9th board - 2nd board) from the surface mounting machine 12A, that is, in the conveying device 16 located downstream of the reflow device 15, and it can be determined that the reflow process has already been completed and that it is about to be removed from the mounting line Lv.

[0083] In this case, the arithmetic processing unit 30 of the surface mounter 12A transmits a stop command to the conveying device 16 and the machines located downstream thereof, that is, the conveying device 16 and the unloader 17, and stops the conveying of the ninth board P.

[0084] Thereafter, the surface mounting machine 12 makes a call to the operator to inform that the 9th board P on which the defective product is mounted is stopped at the transport device 16. The operator call may be, for example, an abnormality light or a warning sound attached to the surface mounting machine 12. In addition, an error message may be displayed on the display unit 37 in combination with these.

[0085] By doing so, it is possible to prevent the board P carrying the defective product from being discharged outside the mounting line. Also, the operator can take measures such as retrieving the ninth board P from the transport device 16 and discarding it.

[0086] <Case 4> Case 4 discloses a method of recovering boards between multiple lines. As shown in Fig. 11, the mounting line Lv is made up of two lines, a first mounting line Lv1 and a second mounting line Lv2, and after the mounting work (e.g., mounting work on the first surface of the board P) is performed on the first mounting line Lv1, the board P is sent to the second mounting line Lv2 to perform the remaining mounting work (e.g., mounting work on the second surface of the board P).

[0087] FIG. 11 shows that the surface mounter 12A on the first mounting line Lv1, during mounting work on the second board P, is notified by the production management device 40 that, as a result of the pass / fail judgment by the learning model M, "the electronic component E linked to the eleventh board P is an NG product."

[0088] In this case, the 11th board P is located in a machine that is 9 boards downstream (the 11th board−the 2nd board) from the surface mounter 12A. In other words, it is located in the printer 11 of the second mounting line Lv2.

[0089] In this case, the calculation processing unit 30 of the surface mounter 12A sends a skip command to each machine that makes up the second mounting line Lv2, namely, the printing machine 11, the surface mounters 12A and 12B, and the AOI (inspection machine), causing them to skip work (printing, mounting, inspection) on the 11th board P containing the defective product.

[0090] In this way, after the board P with the defective products mounted thereon has been discharged onto the downstream mounting line Lv2, the surface mounter 12A sends a skip command to each machine on the downstream mounting line Lv2, and the board with the defective products mounted thereon is skipped from the downstream mounting line Lv2. This allows the 11th board P with the defective products mounted thereon to be discharged off the line in an unprocessed state, without having to be processed on the second mounting line Lv2.

[0091] <Embodiment 3> The mounting system of the third embodiment uses RDMA (Remote Direct Memory Access) technology for transferring image data between the surface mounter 12 and the production management device 40. RDMA is a technology for directly transferring data between memories of different computers.

[0092] FIG. 12A is a block diagram relating to a data transfer function between the surface mounter 12 and the production management device 40. As shown in FIG.

[0093] The arithmetic processing unit 30 of the surface mounter 12 has a first RAM 30A, and the control unit 42 of the production management device 40 has a second RAM 42A.

[0094] The first RAM 30A and the second RAM 42A are connected via a bus 50 so as to be capable of transferring data therebetween, and are remotely accessible to each other.

[0095] The image data captured by the surface mounter 12 is temporarily stored in the first RAM 30A. Then, the image data temporarily stored in the first RAM 30A is directly transferred from the first RAM 30A to the second RAM 42A via the bus 50 and stored therein.

[0096] 12(B) shows a comparative example, in which image data captured by the surface mounter 12 is temporarily stored in RAM and then stored in a hard disk. The image data is then transferred from the hard disk of the surface mounter 12 to the hard disk of the production management device 40 via Ethernet.

[0097] Generally, the read / write speed of a RAM is faster than that of a hard disk. Therefore, by applying the RDMA technology, image data can be transferred between the surface mounter 12 and the production management device 40 in a short time.

[0098] Therefore, even when a large amount of image data is transmitted, communication congestion can be suppressed, and the time required for data transfer between devices can be prevented from becoming a bottleneck in the mounting work.

[0099] An example of the bus 50 connecting the first RAM 30A and the second RAM 42A is InfiniBand (registered trademark), which has a high transfer speed and a short delay time.

[0100] <Embodiment 4> In the first embodiment, an image of the electronic component E and a pass / fail result are used as training data to generate a learning model M for determining the pass / fail quality of the electronic component E (see FIG. 6).

[0101] In the fourth embodiment, a method for extracting teacher data and a flow for generating a learning model M will be described. Fig. 13 is a diagram summarizing pass / fail results of an electronic component E according to a recognition algorithm for each component state.

[0102] In this embodiment, position and angle determination and direction determination are performed as recognition algorithm determinations. The position and angle determination determines whether the position and angle of the electronic component E sucked and held by the mounting head 26A are correct, and the direction determination determines whether the direction of the electronic component E sucked and held by the mounting head 26A is correct.

[0103] The learning default judgment result shown in FIG. 13 is a judgment of the quality of electronic component E based on the position angle judgment and the direction judgment. If both the position angle judgment and the direction judgment are OK, the electronic component E is judged as a good product, and if either one is NG, the electronic component E is judged as a defective product.

[0104] In this way, the result of judging whether the recognition algorithm is good or bad can be used as training data for the learning model M.

[0105] In addition, in the recognition algorithm judgment, the events that can be judged as good or bad depend on the algorithm, and if the number of algorithm types is small, there are events that cannot be judged (for example, the presence or absence of cracks or the attachment of foreign matter). Therefore, to supplement the algorithm judgment, an operator judgment may be used in combination.

[0106] By using two judgment methods (algorithm judgment and operator judgment) in combination, it is possible to judge the pass / fail of electronic component E even for phenomena that cannot be judged by algorithm judgment alone (for example, the presence or absence of cracks or the attachment of foreign matter).

[0107] 14 is a flowchart showing the procedure for generating a learning model. The generation flow of the learning model M is composed of S10 to S70. In the generation flow of the learning model M, when registering the teacher data, a quality judgment of the electronic component E is performed by using a recognition algorithm judgment and an operator judgment in combination (S30, S40).

[0108] Furthermore, the production management device 40 has a full image storage function, and during the production of the board P, all images (including images of things other than components) taken by the surface mounters 12A to 12C are transmitted from the surface mounter 12 to the production management device 40 and stored in the memory unit 43 of the production management device 40.

[0109] Furthermore, for learning targets such as electronic components E, not only images but also pass / fail judgment results based on a recognition algorithm are transmitted from the surface mounter 12 to the production management device in association with the images. When a predetermined number of images are accumulated in the production management device 40, the production management device 40 starts machine learning.

[0110] After the machine learning starts, in S10, the production management device 40 reads and acquires an image from the storage unit 43. Then, the process proceeds to S20, where the production management device 40 determines whether the image read from the storage unit 43 is a learning object.

[0111] When the image is a learning object (in this example, an image of an electronic component E sucked and held by mounting head 26A), production management device 40 acquires the recognition algorithm judgment result for the learning object image in S30. Then, the process proceeds to S40, where the production management device 40 determines the pass / fail judgment result for electronic component E in consideration of the operator's judgment result for the recognition algorithm judgment result acquired in S30.

[0112] Then, the process proceeds to S50, where the image of the electronic component acquired in S10 and the pass / fail judgment result determined in S40 are registered as training data.

[0113] After registering the training data, the production management device 40 performs deep learning and generates a learning model M that uses an image of the electronic component E as input to determine whether the electronic component E is good or bad.

[0114] Then, the process proceeds to S70. If there are remaining images (S70: YES), the process returns to S10 and the next image is read out. In this manner, the teacher data is registered in sequence, and the deep learning is repeated. Then, when there are no remaining images, the deep learning ends (S70: NO). After the deep learning, the learning model M can be used to determine pass / fail.

[0115] <Other embodiments> The present invention is not limited to the embodiments described above and illustrated in the drawings, and the following embodiments, for example, are also included within the technical scope of the present invention.

[0116] (1) In the above embodiment, the quality of the electronic components E sucked and held by the mounting head 26A is judged using a recognition algorithm or a learning model. The object of judgment may be an item other than the components to be mounted on the substrate P as long as it is related to the production of the substrate P. For example, the object of judgment may be a substrate mark 81 or a base mark 82 (see FIG. 2). In addition to the above, the mounting head 26A may also be included in the judgment object. The substrate mark 81 is a fiducial mark for position confirmation or a butt mark for identifying the quality of the substrate. The base mark 82 is a correction mark for correcting the position of the head unit 26 by detecting, for example, thermal elongation of the shaft.

[0117] (2) In the above embodiment, in the first mounting mode, the electronic component E is mounted on the board P after the pass / fail judgment by the first judgment (recognition algorithm). In the first mounting mode, since the pass / fail judgment is performed by the learning model, the first judgment may or may not be performed as long as the mounting of the component is continued after transmitting an image of the electronic component E to the production management device 40 and without waiting for a return of the pass / fail judgment result by deep learning from the production management device 40. The same is true for the second mounting mode, since the pass / fail judgment is performed by the learning model M, since the mounting of the component is performed after transmitting an image of the electronic component E to the production management device 40 and without waiting for a return of the pass / fail judgment result by deep learning from the production management device 40, the first judgment may or may not be performed.

[0118] (3) In the above embodiment, an example has been shown in which the mounting line Lv is configured to include the loader 10, the printer 11, the surface mounter 12, the conveying device 14, the reflow device 15, the conveying device 16, and the unloader 17. The configuration of the mounting line Lv is not limited to the example of the embodiment, and may be a line configuration different from that of the embodiment as long as it includes at least one surface mounter 12.

[0119] (4) In the above-described second embodiment, if it is found after mounting the electronic component E on the board P that the electronic component E mounted on the board P is a defective product as a result of pass / fail determination by deep learning, the surface mounter 12 takes different error measures depending on the work status of the board P on which the defective product is mounted. The error measures may be performed by the production management device 40 instead of the surface mounter 12. For example, the production management device 40 may identify the work status of the board P on which the defective product is mounted from monitoring data of the production status of the board P (data monitoring the number of the board to be worked on and the work status for each machine), and take different error measures depending on the work status.

[0120] (5) In the above embodiment 1, deep learning is exemplified as an example of machine learning. Machine learning is not limited to deep learning, and a learning method other than deep learning can be used as long as it is used to determine the quality of electronic components. The same applies to the case of determining the quality of other items. [Explanation of symbols]

[0121] 12: Surface mount machine 26: Head unit 26A: Mounting head 27: Parts recognition camera 28: Circuit board recognition camera 30: Calculation processing unit (control device) 40: Production control device Lv: Implementation line P: Substrate E:Electronic parts S: Implementation system

Claims

1. 1. An implementation system comprising: A production management device; A mounting line having at least one surface mount machine, The production management device receives an image of an item related to the production of a substrate, and judges the quality of the item using a learning model based on machine learning; The surface mount machine includes: A head unit having components mounted on a board; A control device; The control device is capable of switching between a first mounting mode and a second mounting mode, the first mounting mode is a mode in which, after transmitting an image of the article to the production management device for determining whether the article is good or bad using a learning model, component mounting is continued without waiting for a return of a result of the quality determination by machine learning from the production management device; The second mounting mode is a mode in which an image of the item is sent to the production management device for pass / fail determination using a learning model, and then the component mounting is performed after waiting for a return of the pass / fail determination result based on machine learning from the production management device.

2. The mounting system according to claim 1 , When the surface mounter receives, during the selection of the first mounting mode, from the production management device, a message indicating that the component mounted on the board is an NG product as a result of the pass / fail judgment by machine learning after mounting the component on the board, A mounting system that handles different errors depending on the work situation when a board is equipped with a defective product.

3. The mounting system according to claim 2, In this mounting system, if the substrate carrying the defective product has completed the mounting process and is in the process of being transported to a reflow device, the surface mounting machine stops transport of the substrate and calls an operator.

4. The mounting system according to claim 2, The mounting system is configured to share and execute the operation of mounting components on a single board among a plurality of surface mounters connected in series, If a board carrying a defective product is in the mounting process, the surface mounter returns the board to the surface mounter that performed the mounting operation of the defective product in cooperation with the transport conveyor. This mounting system then uses a head unit to collect the components determined to be defective and remounts them on the board with different components.

5. The mounting system according to claim 2, The surface mounting machine stops conveying the board and calls an operator when the board on which the defective product is mounted has completed the reflow process and is not yet to be discharged from the mounting line.

6. The mounting system according to claim 2, We have multiple mounting lines that share the work for one board, In this mounting system, a surface mounting machine on an upstream mounting line transmits a skip command to the downstream mounting line after a board carrying a defective product has been discharged to the downstream mounting line, and skips work on the downstream mounting line for the board carrying the defective product.

7. The mounting system according to claim 1 or 2, the surface mount machine has a first RAM; the production management device has a second RAM, the surface mounter stores an image of a defect determination target in the first RAM; The production management device reads out an image to be subjected to quality determination directly from the first RAM to the second RAM using RDMA technology.

8. The mounting system according to claim 1 or 2, The article relating to the production of substrates is Components to be mounted on the board, A mounting head that picks up the components to be mounted on the board, Marks on the board, A mounting system including at least one of a mark on a base.

Images