JP2022103272A - Vacuum laminating system, molding defect detection method of vacuum laminating system, and molding condition modification method of vacuum laminating system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide: a vacuum laminating system that can perform at least a part of determination of molding defects in a laminated molded product regardless of an operator; a molding defect detection method for a vacuum laminating system; and a molding condition modification method of the vacuum laminating system.

SOLUTION: A vacuum laminating system 11 for performing lamination molding comprises: a detection device that is provided in a subsequent step of a vacuum laminating device, and detects data of a laminated molded product laminated and molded; and a machine learning device unit including a learning unit that learns images of laminated molded products by distinguishing between non-defective and defective based on the detected data and a determination unit for detecting the laminated molded products at least during continuous molding by the detection device and determining whether they are a non-defective product or a defective product, and a type of defect. If a defective rate of the laminated molded products during the continuous molding exceeds a predetermined ratio as a result of the determination of presence or absence of defects and the type of the defects during the continuous molding, a proposed modification of molding conditions corresponding to the type of defects is notified or the molding conditions are modified.

SELECTED DRAWING: Figure 1

COPYRIGHT: (C)2022,JPO&INPIT

Description

The present invention relates to a vacuum laminating system in which a laminated molded product is pressurized by a pressing body in a vacuum chamber formed between an upper plate and a lower plate of a vacuum laminating device to perform laminating molding, a method for detecting molding defects in a vacuum laminating system, and a method for detecting molding defects in a vacuum laminating system. It relates to a method of modifying molding conditions of a vacuum lamination system.

Conventionally, Patent Document 1 to Patent Document 4 describe a vacuum laminating system in which a laminated molded product is pressurized by a pressing body in a vacuum chamber formed between an upper plate and a lower plate of a vacuum laminating device to perform laminating molding. What was done is known. In Patent Document 1, the lower block device is pressed against the upper pressure block by a pressure cylinder in the vacuum chamber of the vacuum press device, and the film-like material is laminated on the substrate. Further, Patent Document 2 pressurizes a laminated molded product using a film body in a vacuum chamber of a vacuum laminating device. Further, Patent Document 2 also describes that the film body floats due to the low degree of vacuum inside the film body, and defects such as air bubbles occur in the laminated molded product.

Further, in Patent Document 3 and Patent Document 4, a CCD camera is attached to a vacuum stacking system, and image information taken by the CCD camera is used for control. Patent Document 3 attaches a CCD camera to the pre-process of the vacuum laminating apparatus and detects whether the laminated molded product is pressurized from the upper surface or the lower surface. Further, Patent Document 4 captures with a CCD camera the position of the laminated molded product sent from the vacuum laminating device on the cutting stage, and moves the cutting mechanism according to the position of the laminated molded product. Is.

JP-A-2005-334902 (Claim 1, (0004), (0016) to (0018), (FIG. 1)) Japanese Unexamined Patent Publication No. 2001-2640154 (Claim 1, (0003), (0009), (0010), (FIG. 1)) JP-A-2006-103177 (Claim 1, (0023)) Japanese Unexamined Patent Publication No. 2013-103346 (Claim 1, (0029), (0040), (FIG. 1))

However, in each of the vacuum laminating systems of Patent Documents 1 to 4, there is no description about how to deal with a molding defect in the laminated molded product. At present, the laminated molded products that have been laminated and molded in the vacuum laminating system are actually checked one by one by an operator for molding defects. Checking for molding defects by these workers has a problem that labor costs are incurred and variations occur depending on the person who makes the judgment. Further, there is a problem that only a skilled worker can correct the molding conditions when there is a molding defect.

Therefore, the first object of the present invention is to provide a vacuum laminating system and a method for detecting molding defects of the vacuum laminating system, which can perform at least a part of determination of molding defects of a laminated molded product without an operator. And. Another object of the second object is to provide a vacuum laminating system capable of performing at least a part of the work of correcting molding conditions when molding defects occur without an operator, and a method for detecting molding defects of the vacuum laminating system. And.

In the vacuum laminating system according to claim 1 of the present invention, a laminated molded product including a circuit board or a wafer is pressed and laminated by a pressing body in a vacuum chamber formed between the upper plate and the lower plate of the vacuum laminating device. In a vacuum laminating system for molding, a detection device provided in a subsequent process of the vacuum laminating device to detect data of a laminated molded product, and an image of a laminated molded product based on the detected data are regarded as non-defective products. A machine learning device unit equipped with a learning unit that distinguishes and learns defective products, and a determination unit that detects at least a laminated molded product with the detection device during continuous molding and determines whether it is a good product or a defective product and the type of defect. , And, the presence or absence of defects and the type during the continuous molding are determined, and if the defect rate of the laminated molded product during the continuous molding exceeds a predetermined ratio, the molding conditions corresponding to the types of defects are corrected. It is characterized by notifying the proposal or modifying the molding conditions.

The molding defect detecting method of the vacuum laminating system according to claim 2 of the present invention is a laminated molded product including a circuit board or a wafer by a pressing body in a vacuum chamber formed between the upper plate and the lower plate of the vacuum laminating device. In the molding defect detection method of the vacuum laminating system that pressurizes and performs laminating, the data of the laminated molded product laminated and molded by the detection device provided in the subsequent process of the vacuum laminating device is detected, and the detected data is used. Based on this, the image of the laminated molded product is classified into a non-defective product and a defective product, and if it is classified as a defective product, the learning unit of the machine learning device is trained to learn the type of the defect, and at least the learning result is used during continuous molding. Then, the laminated molded product is detected by the detection device, and in the case of either a non-defective product or a defective product and a defective product, the type of the defect is determined by the determination unit of the machine learning device.

The molding defect detecting method of the vacuum laminating system according to claim 3 of the present invention is a laminated molded product including a circuit board or a wafer by a pressing body in a vacuum chamber formed between the upper plate and the lower plate of the vacuum laminating device. In the molding condition correction method of the vacuum laminating system that pressurizes and performs laminating, the data of the laminated molded product that was laminated and molded by the detection device provided in the subsequent process of the vacuum laminating device is detected, and the detected data is used. Based on this, the image of the laminated molded product is distinguished into a good product and a defective product, and in the case of the defective product, the learning unit of the machine learning device is made to learn the type of the defective product, and at least during continuous molding, the laminated molding is performed using the learning result. Based on the fact that the product is detected by the detection device and the judgment unit of the machine learning device determines that the product is defective, at least the hot plate temperature, vacuum degree, pressurizing pressure, and molding time correspond to the type of defect. Notify the proposed modification of the molding condition item including one, or modify the molding condition item including at least one of hot plate temperature, vacuum degree, pressure pressure, and molding time according to the type of defect. It is characterized by that.

The vacuum laminating system of the present invention is a vacuum laminating system in which a laminated molded product including a circuit board or a wafer is pressed by a pressing body in a vacuum chamber formed between the upper plate and the lower plate of the vacuum laminating device to perform laminating molding. In the detection device provided in the subsequent process of the vacuum laminating device to detect the data of the laminated molded product, and the image of the laminated molded product is discriminated into good products and defective products based on the detected data. A machine learning device unit provided with a learning unit for detecting at least a laminated molded product by the detection device during continuous molding and a determination unit for determining whether the product is a good product or a defective product and the type of defective product is provided. The presence or absence and type of defects during continuous molding are determined, and if the defect rate of the laminated molded product during continuous molding exceeds a predetermined ratio, a proposal for amending the molding conditions corresponding to the type of defects is notified. Since the molding conditions are modified, at least a part of the determination of molding defects can be made without an operator. Further, the method for detecting molding defects of the vacuum stacking system and the method for modifying the molding conditions of the vacuum stacking system of the present invention also have the same effect in at least a part of them.

It is a schematic diagram of the laminated molding apparatus of this embodiment. It is a block diagram which shows the machine learning apparatus of the laminated molding apparatus of this embodiment. It is a figure which shows the neural network of the machine learning apparatus of the laminated molding apparatus of this embodiment. It is a flowchart at the time of supervised learning whether a laminated molded product is a good product or a defective product by using the machine learning device of the laminated molding apparatus of this embodiment. It is a flowchart at the time of determining whether a laminated molded product is a good product or a defective product at the time of continuous molding by using the machine learning device of the laminated molding apparatus of this embodiment. It is a schematic diagram of the laminated molding apparatus of another embodiment.

The vacuum stacking system 11 according to the embodiment of the present invention will be described with reference to FIG. The vacuum laminating system 11 conveys the laminated molded products A and A1 using the carrier films F1 and F2. The vacuum laminating system 11 includes a laminated molded product mounting portion 12 and a diaphragm pressure type vacuum laminating device 13 (hereinafter, simply vacuum laminating device 13) in order from the front process on the right side in FIG. 1 to the rear process on the left side in FIG. (Abbreviated as), a laminated molded product carry-out unit 14 is provided. Further, the vacuum stacking system 11 is provided with a touch panel type display device 15 that also serves as an operation device, a control device 16 having a control function of the vacuum stacking device 13, and a function of the machine learning device of the present invention.

The laminated molded product mounting portion 12 includes a film feed portion, and the lower carrier film is unwound from the unwinding roll 17 via the driven roll 18 in a horizontal state. A mounting stage SA on which the laminated molded product A is mounted is provided on the portion of the lower carrier film F1 in the horizontal state. A first camera 19, which is a detection device for detecting the state of the laminated molded product A, is arranged above the above-mentioned placement stage SA. A CCD camera is used as the first camera 19, and the first camera 19 is connected to the control device 16 by a signal line (not shown). Then, the first camera 19 detects the shape of the laminated molded product A, the placement position on the carrier film F1, and the like, and the detection data is sent to the control device 16. The first camera 19 is not essential in the present invention.

Above the above-mentioned placement stage SA and the vacuum laminating device 13, the unwinding roll 20 of the upper carrier film F2 of the film feed portion is arranged. The upper carrier film F2 unwound from the unwinding roll 20 is superposed on the upper surface of the laminated molded product A mounted on the mounting stage SA from the portion of the driven roll 21. Then, the laminated molded product A sandwiched between the upper and lower carrier films F1 and F2 is sent to the vacuum laminating device 13 in the next step.

The vacuum laminating device 13 itself is known as described in Patent Documents 2 to 4, and the lower board 23 is provided so as to be able to move up and down with respect to the fixedly arranged upper board 22 by an elevating device. Therefore, when the lower plate 23 rises, a vacuum chamber C capable of vacuuming the internal space is formed between the upper plate 22 and the lower plate 23. A hot plate 25 heated by the heater 24 is fixed to the center of the lower surface of the upper plate 22, and an elastic film body 26 such as silicon rubber or fluorine rubber is fixed to the surface of the hot plate 25. A hot plate 28 heated by the heater 27 is fixed to the center of the upper surface of the lower plate 23. The temperatures of these hot plates 25 and 28 are detected by the temperature sensors 29 and 30, respectively, and transmitted to the control device 16.

Above the hot plate 28, a diaphragm 32 made of a film body such as silicon rubber or fluorine rubber that swells from the lower plate 23 when a pressurized gas is introduced from the compressor 31 is arranged. Further, the diaphragm 32 is in close contact with the hot plate 25 by being vacuum-sucked in the space on the back surface side of the vacuum pump 33. In the present invention, the diaphragm 32 corresponds to a pressing body. The atmospheric pressure of the pressurized gas sent from the compressor 31 to the back surface side of the diaphragm 32 is detected by the pressure sensor 34 (atmospheric pressure sensor).

The frame portion 35 formed in the peripheral portion of the lower surface of the upper plate 22 and the frame portion 36 formed in the peripheral portion of the upper surface of the lower plate 23 protrude from the hot plates 25 and 28, respectively, and these frame portions 35 and 36. When they come into contact with each other, a vacuum chamber C isolated from the outside world is formed. The degree of vacuum in the vacuum chamber C is detected by the vacuum sensor 37. Then, one side (right side in FIG. 1) of the frame portion of the vacuum laminating apparatus 13 serves as an inlet for the laminated molded product A before laminating molding, and the other side (left side in FIG. 1) of the frame portion is the laminated molded product A1 after laminating molding. It is the exit of. Further, the vacuum chamber C is connected to the vacuum pump 33 via a pipeline, and the inside of the chamber can be evacuated. In the above, the temperature sensors 29 and 30, the pressure sensor 34, and the vacuum sensor 37 correspond to the sensor 38 that detects the molding state of the vacuum stacking device 13 (see FIG. 2). Further, the heaters 24 and 27, the compressor 31, and the vacuum pump 33 correspond to the operating portion 39 of the vacuum stacking device 13 (see FIG. 2).

A laminated molded product carry-out unit 14 is provided in the post-process of the vacuum laminating device 13. The laminated molded product carry-out section 14 including the film take-up section is provided with a carry-out stage SB in which only the lower carrier film F1 is fed in a horizontal state, and the lower carrier film F1 is folded back via the driven roll 40 to lower the take-up roll. It is taken up by 41. Further, a take-up roll 42 of the upper carrier film F2 is arranged above between the carry-out stage SB and the vacuum laminating device 13, and the upper carrier film F2 is laminated and molded via the driven roll 43 in front of the carry-out stage SB. It is peeled off from the molded product A1 and wound up. In the present embodiment, the laminated molded products A and A1 are conveyed using the upper and lower carrier films F1 and F2, but may be conveyed using only the lower carrier film F1 and the carrier films F1 and F2 are used. It may be transported by a transporting means such as a robot without using it.

A second camera 44, which is a detection device for detecting the state of the laminated molded product A1, is arranged above the carry-out stage SB of the laminated molded product carry-out unit 14. A CCD camera is used as the second camera 44, and the second camera 44 is connected to the control device 16. Then, the second camera 44 detects the shape of the laminated molded product A1, the position on the carrier film F1, and the like, and the detection data is sent to the control device 16 via the signal line. The second camera 44 is not above the carry-out stage SB, but is provided at a robot that conveys the laminated molded product A1, a conveyor at which the laminated molded product A1 is conveyed by the robot, or an inspection site for the laminated molded product A1. But it may be. Therefore, when the conveyor or the second camera 44 of the inspection site is provided, those parts are also included in the vacuum stacking system 11 of the present invention.

Next, the control device 16 of the laminated molded product A1 of the present invention and the machine learning device 51 for determining the presence or absence of molding defects will be described with reference to FIG. As described above, the display device 15, including the operating device of the vacuum stacking system 11, each camera 19.44, each sensor 38, and the operating unit 39 are connected to the control device 16, respectively. Then, the information from the display device 15, the cameras 19, 44, and the sensor 38 is sent as a signal to the input processing unit 52 of the control device 16. In addition to the machine learning device 51 described later, the control device 16 includes a storage unit 53 that stores various molding conditions and molding data of the vacuum stacking system 11 including the vacuum stacking device 13, and an operating unit that processes input values. It includes a molding command unit 54 that generates a command signal to be sent to 39, an image display unit 55 that displays an image of the display device 15, an output processing unit 56, and the like.

The machine learning device 51 learns whether or not the laminated molded product A1 is defective by using AI (Artificial Intelligence) technology, which is artificial intelligence, and makes a part or all (at least a part of both) judgments of the operator. It is possible to judge the quality of a laminated molded product without the need for it. The machine learning device 51 is a learning unit 57 that learns the image of the laminated molded product A1 so that it can be distinguished into a non-defective product and a defective product based on the detected data, and a detection device that detects the laminated molded product A1 at least during continuous molding. It is provided with a determination unit 58 that is detected by a second camera 44 or the like and is determined to be either a non-defective product or a defective product. The learning unit 57 includes a learning processing unit 59 and a learning data storage unit 60.

The learning processing unit 59 includes a neural network 71 having a plurality of layers such as an input layer 72, an intermediate layer 73, and an output layer 74 as shown in FIG. Then, in the learning processing unit 59, good product data which is a good product image taken by a camera 44 or the like and defective product data which is a defective product image are input from the input layer 72 of the neural network 71, supervised learning is performed, and stacking is performed. A functional expression such as a regression equation for distinguishing a non-defective product from a defective molded product A1 is generated from the output layer 74. The determination formula generated by the learning processing unit 59 is stored in the learning data storage unit 60 judgment formula storage unit 61. Further, it is desirable that the learning processing unit 59 includes a defective state learning unit 62 that can classify the types of defects such as voids, material outflow, uneven pressure, wrinkles, foreign matter contamination, and misalignment.

The learning data storage unit 60 includes a product data storage unit 63 in addition to the determination formula storage unit 61. Further, the product data storage unit 63 is provided with a non-defective product data storage unit 64 and a defective product data storage unit 65. The non-defective product data storage unit 64 links to the image of the laminated molded product A1 determined to be a non-defective product by the operator, and stores molding data such as molding conditions when the non-defective product is molded and physical quantity data detected by the sensor. Further, the defective product data storage unit 65 links to the image of the laminated molded product A1 determined to be defective by the operator, and forms molding data such as molding conditions when the defective product is molded and physical quantity data detected by the sensor. It will be saved. Further, the defective product data storage unit 65 can classify and store what the operator has determined and input for each type of defect, such as voids, material outflow, uneven pressure, wrinkles, foreign matter contamination, and misalignment. More desirable.

Next, the determination unit 58 of the machine learning device 51 will be described. The determination unit 58 is provided with a quality determination unit 66 for determining the quality of the laminated molded product A1. The quality determination unit 66 determines the quality of the newly sent image data of the laminated molded product A1 based on the function formula stored in the judgment formula storage unit 61 of the learning data storage unit 60. Further, it is more desirable that the quality determination unit 66 includes a defect state determination unit 67 capable of classifying the types of defects such as voids, material outflow, uneven pressure, wrinkles, foreign matter contamination, and misalignment.

Further, when the determination unit 58 determines that the laminated molded product A1 is defective from the sent image, the determination unit 58 corrects the molding conditions of the vacuum lamination device 13 based on the determination that the laminated molded product A1 is defective. It is provided with a molding condition correction unit 68 for determining whether or not to perform the molding condition and how to change the molding condition when the molding condition is changed. The molding condition correction unit 68 formulates better molding conditions by performing reinforcement learning using a neural network 71 or the like. The molding condition modification unit 68 may change the molding condition automatically, or may be displayed on the display device 15 as a molding condition modification proposal and determine whether or not the operator executes the modification proposal. It should be noted that the "modification" of the molding condition in the present invention includes both a completely different molding condition and a slightly corrected molding condition.

The machine learning device 51 is stored not in the control device 16 attached to the vacuum stacking system 11 but in the computer (including the server) of the central control device in the same factory connected to other vacuum stacking systems. It may be processed by being stored in a computer (including a server) in the head office connected by communication, a computer of a manufacturer of a vacuum stacking system, or the like. In that case, those computers and the like are also included in the components for constructing the vacuum stacking system of the present invention. Further, when the data of the vacuum stacking device 13 is transferred to the machine learning device of the computer not connected to the vacuum stacking device 13 by using a portable object such as a flash memory, the computer not connected to the vacuum stacking device 13 is also present. Included in the vacuum stacking system of the invention.

Next, supervised learning by the machine learning device 51 of the vacuum stacking system 11 will be described with reference to the flowchart of FIG. First, the molding conditions of the vacuum laminating system 11 including the vacuum laminating device 13 are set (s1). At this time, if the molding conditions of the laminated molded product A are stored in the storage unit 53 of the control device 16 of the vacuum stacking system 11, the molding conditions are used. If the molding conditions of the same laminated molded product A are not preserved, the operator operates the size, mass, uneven height and spacing, and materials (manufacturer part number, manufacturer recommended molding temperature, material, etc.) of the laminated molded product A. When inputting from the display device 15 that also serves as the above, it is also possible to calculate and display a draft of the molding conditions from the comparison with the molding conditions stored in the past by the machine learning device 51.

First, the laminated molded product A (including its mounting position) on which the mounting stage SA of the laminated molded product mounting portion 12 is mounted is photographed by the first camera 19, and the data of the photographed laminated molded product A is It is sent to the storage unit 53 of the control device 16. At the start of the next molding cycle, the laminated molded product A is fed to the vacuum laminating apparatus 13 by being driven by the take-up rolls 42 and 41 of the upper and lower carrier films F1 and F2. Next, a vacuum chamber C is formed between the lower plate 23 to which the heated hot plate 28 is attached and the upper plate 22 to which the heated hot plate 25 is attached, and the inside of the vacuum chamber C is further formed. Vacuum is sucked by the vacuum pump 33. At this time, the laminated molded product A between the upper and lower carrier films F1 and F2 is liable to be misaligned. When the vacuum chamber C reaches a predetermined degree of vacuum, pressurized air is supplied from the compressor 31 to the space below the diaphragm 32 which is a pressing body, and the diaphragm 32 bulges toward the upper laminated molded product. Then, the laminated molded product A is pressurized and laminated and molded between the diaphragm 32 and the elastic film body 26 attached to the hot plate 25 of the upper plate 22 (s2).

Laminate molding by the diaphragm 32 is performed under the molding conditions such as the hot plate temperature, the degree of vacuum, and the air pressure set in advance, the hot plate temperature measured by the temperature sensors 29 and 30, and the vacuum measured by the vacuum sensor 37. Molding data of measured values such as air pressure measured by the pressure sensor 34 is also sent to the control device 16 and stored in the storage unit 53 (s3). When the predetermined molding time is completed, the laminated molding is completed, the supply of pressurized air to the back surface side of the diaphragm 32 is stopped, and the pressure inside the vacuum chamber C is returned to normal pressure. Then, when the lower plate 23 is lowered and the vacuum chamber C is opened, the laminated molded product A1 formed by feeding the upper and lower carrier films F1 and F2 is sent to the laminated molded product carry-out unit 14. At that time, the upper carrier film F2 is peeled off before the laminated molded product A1 reaches the carry-out stage SB of the laminated molded product carry-out unit 14. Then, in the carry-out stage SB, the laminated molded product A1 is photographed by the second camera 44 provided above, and the photographed image information is sent to the learning processing unit 59 of the machine learning device 51 of the control device 16 (s4). ..

Further, before and after the shooting by the second camera 44, the laminated molded product A1 on the carry-out stage SB is visually checked by an operator located in the vicinity of the carry-out stage SB, and whether it is a good product or a defective product is good or bad. Judgment is made. Then, when the operator inputs either a good product button or a defective product button on the touch panel of the display device 15 (operation device), the result of the quality determination is a learning process of the control device 16 via the display device 15. It is sent to the department 59. At this time, although it is not essential, the operator may input the type of defect such as void, material outflow, uneven pressure, wrinkle generation, foreign matter contamination, misalignment, etc. when the laminated molded product A1 is defective. More preferred.

The image of the laminated molded product A1 taken by the second camera 44, the result of the quality judgment of the laminated molded product A1 performed by the operator, or the type of defect are linked and sent as one data, and the learning processing unit 59 is sent. It is sent to the product data storage unit 63 of the learning data storage unit 60 via. When the operator determines that the laminated molded product A1 has no defects (s5 = Y), it is saved as an image in a good product state in the non-defective product data storage unit 64 (s6), and the molding conditions and molding data (atmosphere temperature) at that time are also stored. , Atmospheric humidity, hot plate temperature, vacuum degree of vacuum chamber C, number of shots of diaphragm, material and tendency of laminated molded product A) are also stored as molding data of non-defective product molding (s7). If the operator determines that the laminated molded product A1 is defective (s5 = N), it is stored as an image of the defective product state in the defective product data storage unit 65 (s8). At that time, the molding conditions and the molding data are also stored as the molding data of the defective product molding as in the case of the non-defective product (s9). Alternatively, an image of the laminated molded product A1 taken by the second camera 44 may be saved first, and a good product or a defective product may be associated with the saved image.

Then, when the frequency of occurrence of defective products is above a certain level (s10 = Y), the molding conditions are corrected (s11). Further, regardless of whether or not the molding conditions are modified, the quality of the laminated molded product A1 sampled until the set number of times is reached (s12 = N) is determined, and data is collected.

When the number of pass / fail judgment inputs by the operator reaches the set number (s12 = Y), then the learning processing unit 59 of the learning unit 57 includes a large number of images and defective products that are determined to be good products stored in the product data storage unit 63. Using the learning data consisting of a large number of determined images as a teacher, a functional expression used for determining the quality of the laminated molded product A1 is generated using the neural network 71 (s13). In supervised learning using these neural networks 71, a good quality judgment cannot be made unless a certain number of samples are accumulated, so that a certain amount of samples are accumulated first. After accumulating the samples, the learning processing unit 59 inputs the sample in which the image and the pass / fail judgment result are linked to the input layer 72 of the neural network 71, and the machine learning device 51 performs regression analysis for making the pass / fail judgment and performs a function. An equation (regression equation) is generated and output from the output layer 74. The output function expression is stored in the determination expression storage unit 61 of the learning unit 57 (s14).

The method of supervised learning may be a method using a decision tree in addition to regression analysis. The function calculation formula may be used as it is or as an approximate expression. Further, the neural network 71 of FIG. 3 is provided with only one intermediate layer 73 (hidden layer) between the input layer 72 and the output layer 74, but a plurality of intermediate layers are provided between the input layer 72 and the output layer 74. It does not exclude those having a layer 73 (hidden layer) and performing deep learning.

Next, laminating molding is actually performed, good / bad judgment is made by the good / bad judgment unit 66 of the machine learning device 51 using the above-mentioned function formula, and the good / bad judgment result is displayed on the display device 15. On the other hand, the operator checks whether the result of the quality determination performed by the machine learning device 51 (AI) is correct (s15). When there is an error in the quality determination result of the machine learning device 51 (s15 = N), the set number of times (s12) is increased, and supervised learning is further performed. If the worker confirms that the judgment result of the machine learning device 51 is correct (s15 = Y), the supervised learning is completed, and from the next time, the quality of the machine learning device 51 alone is used without the worker's check. Judgment becomes possible.

Next, the quality determination and the modification of the molding conditions at the time of continuous molding of the vacuum lamination system 11 will be described with reference to the flowchart of FIG. In the vacuum laminating device 13 of the vacuum laminating system 11, the laminated molded product A1 is continuously laminated and molded in each molding cycle (s101). The laminated molded product A1 is photographed on the carry-out stage SB of the laminated molded product carry-out unit 14 (s102). The captured image is sent to the quality determination unit 66 of the determination unit 58 as image information. Then, it is determined whether the image information is good image information or defective image information by using the function expression generated by the above supervised learning and stored in the determination expression storage unit 61 (s103). When the quality determination unit 66 determines that there is no defect in the image information of the laminated molded product A1 (s103 = Y), a command (s104) without modification of the molding conditions is issued and the next laminated molded product A1 is molded. Will be. Even when the machine learning device 51 based on AI determines whether the machine learning device 51 is good or bad, the operator may repeat the inspection such as sampling inspection. Therefore, the present invention may be such that at least a part of the machine learning device 51 can be performed without an operator.

When the quality determination unit 66 determines that the image information of the laminated molded product A1 is defective (s103 = N), the laminated molded product A1 of the image is removed from the molding line as a defective product (s105). At this time, the operator may re-inspect only the laminated molded product A1 determined by the machine learning device 51 to be defective, or all the laminated molded products A1. Therefore, the present invention may be such that at least a part of the machine learning device 51 can be performed without an operator. Then, when the machine learning device 51 determines that the product is defective but the operator confirms that the product is non-defective, or when the machine learning device 51 determines that the product is defective but the operator confirms that the product is defective. If there is, it may be subject to supervised learning again in the learning unit 57.

Next, based on the fact that the laminated molded product A1 is determined to be defective by the quality determination unit 66, is it necessary for the molding condition correction unit 68 to correct the molding conditions based on the molding data sent from each sensor? Is made (s106). However, there are various factors that cause defects, and there may be problems with the laminated molded product A itself to be charged or dust may be mixed in. Therefore, even if a defect is detected once, the molding conditions are not changed. .. Therefore, when it is determined that it is not necessary to change the molding conditions even if the laminated molded product A1 is defective (s106 = N), a command (s104) without modification of the molding conditions is issued and the next laminated molded product is issued. Molding is done

When it is determined that the molding conditions need to be corrected (s106 = Y) when the defect rate exceeds a predetermined ratio, the molding conditions are corrected by the molding condition correction unit 68 of the machine learning device 51. In the present embodiment, by performing reinforcement learning in the molding condition correction unit 68 of the determination unit 58, the molding condition can be corrected by the machine learning device 51 with no involvement of the operator or at least a part of the involvement. desirable. Regarding reinforcement learning, when a preset defect rate is reached, etc., the molding condition correction unit 68 of the machine learning device 51 refers to the molding conditions and molding data when a good product is molded, and provisionally Generate molding conditions. Then, continuous molding is performed under the provisional molding conditions, and if the defect rate falls within a predetermined range or improves, a reward is given. In addition, if continuous molding is performed under temporary molding conditions and the defect rate worsens, a negative reward is given. Then, the molding condition correction unit 68 of the machine learning device 51 corrects the molding conditions in the direction in which the reward increases.

The items of the molding conditions to be modified at this time include the hot plate temperature, the degree of vacuum, the pressurizing pressure, the molding time and the like. However, when molding a non-defective product, the items of the molding conditions are intricately intertwined, and if the set value of one item is corrected, the number of defective products may increase unless the set value of another item is also corrected. In adjusting these molding conditions, the machine learning device 51 may approach good molding conditions while balancing each item by performing reinforcement learning using a mathematical programming method or the like.

Alternatively, the results of supervised learning may be incorporated when modifying the molding conditions. Since molding data including molding conditions at the time of good product molding and defective product molding are linked and stored in the non-defective product data storage unit 64 and the defective product data storage unit 65, they are different from the molding conditions and molding data at the time of good product molding. If molding defects occur due to the use of the molding conditions and the molding data associated therewith, corrections such as returning to the molding conditions and molding data at the time of past non-defective product molding may be performed. Further, when the number of samples of molding data including molding conditions at the time of molding is insufficient, deep learning may be incorporated.

The modification of the molding conditions at this time is performed by notifying the modification plan of the molding conditions or by modifying the molding conditions. The notification of the modification plan of the molding condition is usually performed by displaying it on the display device 15, but it may be notified by voice or the like, or when the operator is not in the vicinity of the vacuum stacking device 13, the notification device may be notified to the mobile terminal. good. Then, the worker decides to adopt or rejects the modification plan of the molding condition presented by the machine learning device 51, and makes a modification setting. At this time, if it is inferred that the heater is broken, the vacuum chamber C is not sealed, the diaphragm is deteriorated, or the pressurizing mechanism or vacuum mechanism including the valve is defective, the machine learning device 51 analogizes the message, diaphragm, etc. It is preferable to display a message instructing the replacement of parts on the display device 15. Further, the modification of the molding conditions may be automatically performed by the machine learning device 51. When the machine learning device 51 automatically modifies the molding conditions, the modified molding conditions are stored and can be displayed on the display device. Then, it is determined whether or not the planned number of molded pieces has been reached (s108), and continuous molding is performed until the planned number of molded pieces is reached (s108 = Y).

Regarding the above, the case where the machine learning device 51 only determines the quality of the laminated molded product A1 has been described. However, at the time of initial supervised learning, the worker classifies and inputs at least one defect such as void, material outflow, pressure unevenness, wrinkle generation, foreign matter contamination, and misalignment, and the defect is defective for each type of defect. When a function expression for determining the presence / absence and the type is generated and saved, the determination unit 58 can also display the type of defect by using the neural network 71 similar to the case where only the pass / fail determination is performed. Then, when the molding conditions are corrected, the correction corresponding to the type of defect can be performed. For example, when a void is generated in the laminated molded product, correction measures are taken such as increasing the degree of vacuum to a high vacuum or extending the time until pressurization by the diaphragm 32 after forming the vacuum chamber C. Further, when the material outflow occurs, correction measures such as lowering the hot plate temperature and lowering the pressurizing pressure by the diaphragm 32 are taken. Further, when the laminated molded product A1 is displaced, the time for increasing the degree of vacuum after forming the vacuum chamber C is slowed down, the feeding speeds of the carrier films F1 and F2 are slowed down, and the carrier films F1 and F1 are slowed down. Correction measures such as adjusting the tension of F2 are performed.

Regarding the determination of the type of defect, in the initial supervised learning, the worker inputs only the pass / fail judgment together with the image of the laminated molded product A1, and the machine learning device 51 generates only the function expression of the pass / fail judgment. However, by incorporating unsupervised learning such as clustering by AI during continuous molding, which defects such as voids, material outflow, uneven pressure, wrinkles, foreign matter contamination, misalignment, etc. correspond to defects from the captured image. Classification can be performed by the state determination unit 67. Then, it is possible to correct the molding conditions corresponding to the defect and notify the message.

Further, regarding the misalignment of the laminated molded product A1 and the outflow of materials, a machine learning device can be used by using a predetermined calculation formula or numerical value when comparing the image of the first camera 19 and the image of the second camera 44. In some cases, it is possible to make a pass / fail judgment by a conventional method without using 51. However, if the laminated molded product A1 is slightly misaligned as a whole but does not overlap and good laminated molding is possible, or if it is fuzzy in relation to other good / bad conditions, pass / fail is determined. In that case, it is desirable to incorporate the judgment by the machine learning device 51. Note that the misalignment is detected whether or not molding is performed at a good molding position of the vacuum stacking device 13, even if it is installed only in the second camera 44 on the outlet side of the vacuum stacking device 13. It is possible.

The supervised learning of the machine learning device 51 of the vacuum lamination system 11 is initially performed for each molding of the laminated molded product A1, but as the learning results are accumulated, the molded product that is the same as or similar to the laminated molded product A1. In the case of molding, there are increasing cases where it becomes possible to make a pass / fail judgment without supervised learning. In addition, when this machine learning device 51 is additionally introduced to the newly shipped vacuum stacking system 11 or the old vacuum stacking system, the results of conventional supervised learning can be installed from the beginning, which is new. In many cases, it is possible to reduce the number of times of supervised learning.

Next, the vacuum stacking system 101 of another embodiment of the present invention will be described with reference to FIG. 6, focusing on the differences. In the vacuum laminating system 101, a flattening press device 104 without a vacuum chamber is provided after a laminated molded product mounting portion 102 and a diaphragm pressure type vacuum laminating device 103 (hereinafter, simply abbreviated as the vacuum laminating device 103). A laminated molded product carry-out unit 105 is provided in a post-process of the flattening press device 104. The camera, which is a detection device for the laminated molded product, has a first camera 106 above the laminated molded product mounting portion 102, and a second camera 107 above between the vacuum laminating device 103 and the flattening press device 104. A third camera 108 is provided above the molded product carry-out portion 105.

The vacuum stacking system 101 of another embodiment is also provided with the same machine learning device 51 as the vacuum stacking system of the embodiments shown in FIGS. 1 and 2. In another embodiment, an image taken by the second camera 107 and laminated only by the vacuum laminating device 103 and an image taken by both the vacuum laminating device 103 and the flattening press device 104 by the third camera 108 are taken. .. Therefore, it is possible to quickly determine whether the vacuum laminating device 103 of the vacuum laminating system 101 is the cause of the molding defect or the flattening press device 104 is the cause of the molding defect.

In this embodiment and another embodiment, a camera such as a CCD camera is mainly used as a detection device for detecting data of a laminated molded product. However, as for the detection device, at least one of the following detection devices may be used, or may be used in combination with the camera. As a detection device, the presence or absence of warpage or material outflow of the laminated molded product may be detected by measuring the height of each measurement point of the laminated molded product with an ultrasonic sensor or the like. Further, the presence or absence of voids and the presence or absence of material outflow may be detected by a photoelectric sensor or a laser sensor that detects the color and brightness by receiving the projected light by the light receiving unit. Alternatively, a thermography that detects the temperature of the laminated molded product may be used to detect the heating unevenness of the laminated molded product.

The present invention is not listed one by one, but is not limited to the above-described embodiment of the present invention. Needless to say, it also applies to.

For example, when the pressing body of the vacuum laminating devices 13 and 103 is a diaphragm, the board on which the diaphragm is provided may be at least one of the upper board and the lower board. Further, the pressing body may be a flat pressure plate provided with an elastic body or a rotatable pressure roll instead of the diaphragm. Further, the number of the vacuum stacking devices 13 and 103 may be one, but two or more may be arranged in series. Further, the laminated molded product to be laminated and molded by the vacuum laminating systems 11 and 101 may be a wafer or another plate-like body in addition to the circuit board, and is not limited thereto.

11,101 Vacuum stacking system 13,103 Vacuum stacking device 19,106 First camera 44,107 Second camera (detection device)
51 Machine learning device 57 Learning unit 58 Judgment unit 59 Learning processing unit 66 Good / bad judgment unit 68 Molding condition correction unit

Claims (3)

In a vacuum laminating system in which a laminated molded product including a circuit board or a wafer is pressed by a pressing body in a vacuum chamber formed between the upper plate and the lower plate of the vacuum laminating device to perform laminating molding.
A detection device provided in a subsequent process of the vacuum laminating device and detecting data of a laminated molded product that has been laminated and molded.
A learning unit that learns by distinguishing images of laminated molded products into non-defective products and defective products based on the detected data, and at least during continuous molding, the laminated molded products are detected by the detection device and are defective as either good products or defective products. A machine learning device unit equipped with a judgment unit that determines the type of
During the continuous molding, the presence or absence of defects and the type are determined, and
When the defect rate of the laminated molded product during continuous molding exceeds a predetermined ratio, the vacuum lamination system is characterized by notifying a proposal for modifying the molding condition corresponding to the type of defect or modifying the molding condition. .. In a molding defect detection method of a vacuum laminating system in which a laminated molded product including a circuit board or a wafer is pressed by a pressing body in a vacuum chamber formed between the upper plate and the lower plate of the vacuum laminating device to perform laminating molding.
The data of the laminated molded product that was laminated and molded by the detection device provided in the subsequent process of the vacuum laminating device was detected, and the data was detected.
Based on the detected data, the image of the laminated molded product is classified into a non-defective product and a defective product, and if it is classified as a defective product, the learning department of the machine learning device is supervised to learn the type of the defect.
At least during continuous molding, the learning result is used to detect the laminated molded product with the detection device, and if it is either a good product or a defective product and a defective product, the judgment unit of the machine learning device determines the type of defect. A method for detecting molding defects in a vacuum lamination system. In the method of modifying the molding conditions of a vacuum laminating system in which a laminated molded product including a circuit board or a wafer is pressed by a pressing body in a vacuum chamber formed between the upper plate and the lower plate of the vacuum laminating device to perform laminating molding.
The data of the laminated molded product that was laminated and molded by the detection device provided in the subsequent process of the vacuum laminating device was detected, and the data was detected.
Based on the detected data, the image of the laminated molded product is classified into a non-defective product and a defective product, and in the case of the defective product, the learning unit of the machine learning device is made to learn the type of the defect.
At least during continuous molding, the learning result is used to detect the laminated molded product with the detection device.
Based on the judgment of the machine learning device that it is a defective product
Notify the correction proposal of the item of the molding condition including at least one of the hot plate temperature, the degree of vacuum, the pressurizing pressure, and the molding time corresponding to the type of the defect, or notify the hot plate temperature corresponding to the type of the defect. , A method for modifying molding conditions of a vacuum lamination system, comprising modifying items of molding conditions including at least one of vacuum degree, pressure pressure, and molding time.

Images