TWI792296B - Net mask inspection method and net mask inspection device - Google Patents

Abstract

Provided are a substrate manufacturing method, a mesh mask inspection method, and a mesh mask inspection device capable of suppressing occurrence of defective products by inspecting a cleaned mesh mask.

The metal mask inspection device 30 includes: an illuminating device 32A for irradiating light on the metal mask; a camera 32B for photographing the opening of the metal mask; and a control device 33 for processing the acquired image data. The control device 33 is equipped with an AI model 100, which is constructed by making the neural network use only the image data of the opening of the metal mask with no foreign matter attached thereto as learning data; and acquires and photographs the opening of the metal mask that has been cleaned. The original image data related to the opening part obtained by the department, and the reconstruction image data related to the opening part obtained by inputting the original image data into the AI model 100 to reconstruct (reconstitution), compare the two image data, and determine whether the inside of the opening part Whether there is foreign matter attached to the wall.

Description

Net mask inspection method and net mask inspection device

The present invention relates to a substrate manufacturing method for printing solder on a substrate to mount components, a mesh mask inspection method for inspecting a mesh mask used during solder printing, and a mesh mask inspection device.

Generally, a production line for mounting electronic components on a substrate is first connected to a solder printing machine, and paste solder is printed on the islands of the substrate by screen printing. Then, according to the viscosity of the cream solder, the electronic parts are temporarily fixed on the substrate. Then, soldering is performed by introducing the aforementioned substrate into a reflow furnace.

A solder printer is equipped with a mesh mask that forms many openings corresponding to the islands of the substrate to be pre-printed. In solder printing, with the mesh mask in contact with the surface of the substrate, the paste flux is supplied onto the substrate and the squeegee is slid to push the cream flux into the openings. Then, the screen mask is separated from the substrate, and the cream solder filled in the openings is detached from the screen and printed (transferred) on the islands.

However, when the solder printing process is performed many times, the cream solder adheres to the openings of the mesh mask to cause clogging, and printing failure may occur. Therefore, in the conventional technology, cleaning of the mesh mask was performed every time a predetermined number of solder printing steps were performed.

On the other hand, in a solder printing machine, a camera is used to photograph the opening of the mesh mask, and the opening area of the opening (the amount of creamy solder adhering to the opening) is inspected, and it is notified whether the mesh mask needs to be replaced or cleaned. The technology etc. are known (for example, refer patent document 1). [Prior Art Literature] [Patent Document]

[Patent Document 1] Japanese Patent Application Laid-Open No. 3-154809

[Problem to be solved by the invention]

However, in the prior art, when the flux printing process and the cleaning process are alternately repeated many times, it can be seen that the creamy solder release from the opening of the mesh mask will be transferred to the island part with a worsening ratio. The amount of cream solder tends to decrease.

When the amount of cream solder transferred to the island portion decreases, the soldering operation for mounting parts cannot be performed properly, and the occurrence rate of defective products may increase. Further, if the flux printing process is continued in such a state where the flux dropout is deteriorated, clogging of the opening of the mesh mask will occur in a period shorter than the originally planned number of printings, and printing failures will easily occur.

Therefore, the inventors of the present case have studied in detail the mesh mask with poor solder release, and have found that even after the mesh mask is cleaned, the inner wall of the opening of the mesh mask still has work. Minute residual foreign matters (residues such as cream solder or flux) that cannot be seen by human eyes continue to adhere and remain, which affects the deterioration of solder release during solder printing.

The present invention was developed in view of the above circumstances, and an object thereof is to provide a substrate manufacturing method, a mesh inspection method, and a mesh inspection device capable of suppressing occurrence of defective products by inspecting a cleaned mesh. [Means to solve the problem]

Hereinafter, various means suitable for solving the above-mentioned problems will be described item by item. In addition, the specific function and effect of the corresponding means are also provided as supplementary statements as needed.

Means 1. A method of manufacturing a substrate, which is a method of manufacturing a substrate by printing flux on the substrate to mount components, which comprises: A flux printing process of printing flux on the aforementioned substrate by a predetermined flux printing means; Cleaning process, in the predetermined cleaning means, clean the screen mask used in the aforementioned flux printing process; In the photographing process, photographing the opening of the mesh mask with predetermined photographing means in the subsequent stage after the aforementioned cleaning process and the preceding stage of the aforementioned flux printing process; Cleaning quality judgment process, based on the image data obtained by the above-mentioned shooting process, it is judged whether there is foreign matter attached to the inner wall of the opening of the aforementioned net mask; In the printing preparation process, when it is judged that there is no foreign matter attached to the inner wall of the opening of the mesh mask in the cleaning quality judgment step, the mesh mask is set in the solder printing means; and In the re-transfer process, when it is judged that foreign matter is attached to the inner wall of the opening of the net mask in the cleaning quality judgment process, the net mask is transferred to the cleaning means again.

According to the above-mentioned means 1, after the screen mask used for flux printing is cleaned, inspection is carried out for the presence or absence of foreign matter attached to the inner wall of the opening of the screen mask. Next, in the net mask inspection, if it is determined that foreign matter is attached to the inner wall of the net mask opening, the net mask is cleaned again.

According to this configuration, regardless of whether cleaning of the screen mask is performed or not, the inner wall of the opening of the screen mask will continue to adhere to the fine residual foreign matter (flux or flux residues, etc.), which can prevent the screen from being reused during solder printing. Bad case of masking.

As a result, clogging of the opening of the mesh mask (poor printing) or poor soldering of mounting parts (flux shortage) can be suppressed in advance, which may occur when the flux continues to print in a state where the flux stripping is deteriorated. situation.

In addition, in a general substrate manufacturing method, for example, a flux printing process of printing flux on the substrate, a flux printing inspection process of inspecting the solder printed on the substrate, and mounting predetermined parts on the printed substrate are performed. The component mounting process of the flux, the reflow process of heating and melting the flux printed on the aforementioned substrate to fix the aforementioned components, etc.

Means 2. A method for inspecting a mesh mask, which is used to inspect a cleaned mesh mask before performing flux printing on a substrate, which has: Irradiating process, irradiating predetermined light on the opening of the aforementioned net mask; photographing process, photographing the opening of the aforementioned net mask irradiated by the aforementioned light; The original image data obtaining process is to obtain the original image data related to the aforementioned opening according to the image data obtained by the aforementioned photographing process; A reconstruction (reconstitution) image data acquisition process is to acquire reconstructed image data related to the opening portion reconstructed by inputting the original image data to a discrimination means (generated model), wherein the discrimination means has features extracted from the input image data. The neural network of the encoding part of the quantity and the decoding part of reconstructing the image data from the feature quantity is generated by learning only the image data of the aforementioned opening with no foreign matter attached thereto as the learning data; a comparison process of comparing the aforementioned original image data with the reconstructed image data; and The judging step is to judge whether there is foreign matter attached to the inner wall of the opening of the net mask based on the comparison result in the comparing step.

In addition, also in the means described below, the aforementioned "neural network" also includes, for example, a convolutional neural network having a plurality of convolutional layers. In addition, the above-mentioned "learning" includes, for example, deep learning and the like. The aforementioned "discrimination means (generated model)" includes, for example, an autoencoder (self-encoder), a convolutional autoencoder (a convolutional autoencoder), and the like.

According to the above method 2, after the screen mask used for flux printing is cleaned, it is possible to inspect whether there is foreign matter attached to the inner wall of the opening of the screen mask. As a result, the same operation and effect as the above-mentioned means 1 can be achieved.

In particular, in this method, since a discrimination means (generated model) such as a self-encoder constructed by learning a neural network is used to determine whether there is foreign matter attached to the inner wall of the opening of the net mask, it is also possible to detect Small foreign objects that are difficult to detect are formed by simply comparing the area, shape, and size of openings or foreign objects on the image data as in the known technology. For example, in this detection technique, the area of a thin and wide foreign object adhering to the inner wall of the opening is very small when viewed from the normal direction of the mesh mask, that is, the area on the image data, making it difficult to detect.

Furthermore, in this method, because the original image data obtained by photographing is compared with the reconstructed image data reconstructed based on the original image data, in the compared two image data, there will be no damage due to the mesh mask of the object to be inspected. The shooting conditions on the side (for example, the arrangement position or arrangement angle of the net mask, bending, etc.) For more accurate detection of foreign matter.

In addition, as in the case of the above-mentioned patent document 1, etc., when the inspection related to the predetermined opening is carried out at the predetermined position of the net mask, as a quality judgment criterion, the configuration information (position data or dimension data, etc.) of the opening is required. ), the board design information such as Jaber data is memorized in advance, and the configuration information of the opening as the inspection object is appropriately obtained, and the quality of the opening as the inspection object is judged while comparing with the configuration information. , so there is concern that the inspection efficiency will be reduced. Furthermore, it is also necessary to correctly position the inspection position of the net mask.

On the other hand, according to this method, since the inspection of each opening is performed by means of discrimination such as a self-encoder, it is not necessary to memorize the information of each configuration for each of the plurality of openings, so it is not necessary to refer to these when checking. Information, but can attempt to improve the efficiency of inspection.

Furthermore, when a configuration is used to check the clogging of the opening of the screen mask after the solder printing process is completed, the opening of the screen mask without clogging before the start of printing may be photographed and stored in advance as reference data. However, at this time, when the mesh mask is not unused, but has been cleaned to make it unclogged, the remaining foreign matter that has not been washed away by cleaning will also be included in the reference data. In clogging inspection, there is also a concern that residual foreign matter cannot be detected.

In this point, like this means, the structure which inspects the cleaned mesh mask before starting a flux printing is adopted.

Means 3. The mesh mask inspection method as in means 2, wherein in the comparison process, the image related to the inner wall of the opening in the original image data, and the image in the reconstructed image data and the opening in the aforementioned The relevant images of the inner wall are compared.

According to the above-mentioned method 3, it is constituted by comparing the images related to the inner wall of the opening in the two image data (for example, the straightness of the flat part of the inner wall of the opening, or the corner of the inner wall Compared with the configuration that simply compares the area, shape, and size of openings, or foreign objects, for example, it can detect finer foreign objects more accurately.

Means 4. A mesh mask inspection device, which is used to inspect the cleaned mesh mask before performing flux printing on the substrate, having: The irradiating means can irradiate predetermined light to the opening of the aforementioned mesh cover; A photographing means capable of photographing the opening of the aforementioned net mask irradiated by the aforementioned light; and The image processing means can use the image data obtained by the aforementioned shooting means as a basis to perform relevant inspections on the inner wall of the opening of the mesh mask; The aforementioned image processing means have: The discrimination means (generated model) is a neural network that has an encoding unit (encoder) that extracts feature quantities from input image data and a decoding unit (decoder) that reconstructs image data from the feature quantities. The video data of the aforementioned opening with the foreign matter adhered thereto is generated by learning as learning data; The original image data acquisition means can use the aforementioned shooting means to photograph the opening of the aforementioned cleaned net mask, and obtain the original image data related to the aforementioned opening; The reconstructed image data acquisition means can obtain the reconstructed image data related to the aforementioned opening, which can obtain the aforementioned original image data and input it into the aforementioned identification means and perform reconstruction; comparison means, which can compare the aforementioned original image data with the reconstructed image data; and The judging means judges whether there is foreign matter attached to the inner wall of the opening of the mesh cover based on the comparison result obtained by the comparing means.

According to the above-mentioned method 4, after masking the screen used for flux printing, it is checked whether there is foreign matter attached to the inner wall of the opening of the screen mask. As a result, the same effects as those of the above means 1 and 2 can be achieved.

In addition, what is not particularly mentioned in this method is that it is preferable to arrange the irradiation means on the side of the contact surface of the mesh mask that is in contact with the substrate during solder printing, and to arrange the irradiation means on the side of the mesh mask that is not in contact with the substrate. The photographing means is arranged on the non-contact surface side, and the photographing means photographs the light passing through the opening of the net mask among the light irradiated from the illuminating means.

However, it is not excluded that the aforementioned irradiation means and the aforementioned photographing means are uniformly arranged on the contact surface side of the mesh mask that is in contact with the substrate, or on the non-contact surface side of the mesh mask that is not in contact with the substrate. The configuration of light reflected by the mesh mask including the openings among the light irradiated by the irradiation means can also be implemented with this configuration.

Of course, the inspection may be performed in a state where the contact surface side (or the non-contact surface side) of the mesh cover is arranged in either upward or downward direction.

Means 5 is the net mask inspection device as in means 4, wherein the comparison means is to compare the image related to the inner wall of the opening in the original image data with the inner wall of the opening in the reconstructed image data Related images are compared.

According to the above-mentioned means 5, the same effect as that of the above-mentioned means 3 can be achieved.

[Mode for Carrying Out the Invention]

Hereinafter, one embodiment of the present invention will be described. First, the structure of the substrate on which components are mounted by printing solder will be described. FIG. 1 is a partially enlarged plan view enlarging a part of a printed circuit board as a substrate.

As shown in FIG. 1 , a printed circuit board 1 has a wiring pattern (not shown) made of copper foil or a plurality of islands 3 formed on the surface of a flat base substrate 2 made of glass epoxy resin or the like. In addition, on the surface of the base substrate 2 , a resist film 4 is applied to a portion other than the island portion 3 . Next, as shown in FIGS. 1 and 3 , a viscous cream solder 5 is printed on the island portion 3 . In addition, in Figs. 1 and 3 , expediently, a scatter pattern is added to the portion showing the cream solder 5 .

Next, a production line (manufacturing process) for manufacturing the printed circuit board 1 will be described with reference to FIG. 2 . FIG. 2 is a block diagram showing the composition of the production line 10 of the printed substrate 1 . In the production line 10 of this embodiment, viewed from the front side, it is set so that the printed circuit board 1 is transferred from left to right.

As shown in FIG. 2 , in the production line 10 , a solder printing machine 12 , a solder printing inspection device 13 , a component mounting machine 14 , and a reflow device 15 are installed sequentially from the upstream side (left side in FIG. 2 ).

The solder printing machine 12 is used to perform a solder printing process of printing cream solder 5 on the island portion 3 of the printed circuit board 1, and constitutes a solder printing means in this embodiment.

As shown in FIG. 3 , the solder printing machine 12 includes: a flat metal mask 20 formed with a plurality of openings 21 corresponding to each island 3 on the printed circuit board 1; The top of the mask 20 slides.

In addition, the metal mask 20 (refer to FIG. 4 ) in this embodiment is composed of a rectangular thin-plate-shaped metal mask body part 20a formed with the aforementioned plurality of openings 21, and a peripheral edge part supporting the mask body part 20a. It is composed of a rectangular frame-shaped metal frame portion 20b with four sides. The deflection of the mask main body 20a can be suppressed by supporting the thin plate-shaped mask main body 20a in a tensioned state on the highly rigid frame part 20b.

Under the above configuration, in the solder printing process, first, the metal mask 20 is aligned and brought into contact with the surface side of the printed circuit board 1 . Next, cream solder 5 is supplied onto the upper surface of metal mask 20 . Next, the squeegee 22 is slid along the upper surface of the metal mask 20 to push the cream flux 5 into the opening 21 .

Then, the metal mask 20 is separated from the printed circuit board 1, and the cream solder 5 filled in the opening 21 is printed (transferred) on the island 3, and the solder printing is completed.

The solder printing inspection device 13 is a mechanism that performs an inspection process on the state (for example, printing position, height, printing amount, etc.) of the cream solder 5 printed as described above.

The component mounting machine 14 executes a component mounting process of mounting an electronic component 25 (see FIG. 1 ) on the island portion 3 on which the cream solder 5 is printed. The electronic component 25 includes a plurality of electrodes or lead wires, and each of the electrodes or lead wires is temporarily fixed to predetermined respective cream solders 5 .

The reflow device 15 is used to heat and melt the cream solder 5 , and perform a reflow process of soldering (soldering) the island portion 3 and the electrodes or wires of the electronic component 25 .

In addition, what is omitted in the illustration is that the production line 10 is provided with a conveyor etc. between the solder printing machine 12 and the solder printing inspection device 13 to transfer the printed circuit board 1 between the above-mentioned devices. Furthermore, a branching device is provided between the solder print inspection device 13 and the component mounting machine 14 . Next, the printed circuit board 1 judged to be a good product by the solder printing inspection device 13 is guided to the component mounting machine 14 on the downstream side, and the printed circuit board 1 judged to be a defective product is stored in a defective product by a branching device. Department discharge.

Furthermore, a cleaning device 29 as a cleaning means and a metal mask inspection device 30 as a mesh mask inspection device are provided in the vicinity of the solder printing machine 12 .

The cleaning device 29 is a cleaning process for cleaning the metal mask 20 serving as a mesh mask to which dirt and the like adhered in the solder printer 12 .

The metal mask inspection device 30 checks whether the metal mask 20 that has been cleaned in the cleaning device 29 is properly cleaned.

The configuration of the metal mask inspection device 30 will now be described in detail with reference to FIGS. 4 and 5 . FIG. 4 is a schematic configuration diagram schematically showing a metal mask inspection device 30 . FIG. 5 is a block diagram of the functional configuration of the metal mask inspection device 30 .

The metal mask inspection device 30 is equipped with: a transfer mechanism 31, which performs the transfer or positioning of the metal mask 20; an inspection unit 32, which is used to perform inspection of the metal mask 20; and a control device 33, which uses the transfer mechanism 31 or the inspection unit 32 Firstly, various controls, image processing, and calculation processing in the metal mask inspection device 30 are executed. The control device 33 constitutes video processing means related to this embodiment.

The transfer mechanism 31 includes: a pair of transfer rails 31a arranged along the loading and unloading direction of the metal mask 20; an endless conveyor belt 31b rotatably arranged on each transfer rail 31a; and a driving means such as a motor for driving the conveyor belt 31b. (illustration omitted); and a chuck mechanism (illustration omitted), in order to locate the metal mask 20 at a predetermined position;

With the above configuration, the metal mask 20 carried into the metal mask inspection device 30 and the frame portions 20b on both side edges in the width direction perpendicular to the carrying-in and carrying-out direction are respectively inserted into the transfer rail 31a and placed on the conveyor belt. 31b on. Next, the conveyor belt 31b starts to move, and moves the metal mask 20 to a predetermined inspection position. When the metal mask 20 reaches the inspection position, the conveyer belt 31b stops, and at the same time, the chuck mechanism operates. Due to the action of the chuck mechanism, the conveyor belt 31b is pushed up, and the frame portions 20b at both side edges of the metal mask 20 are clamped by the conveyor belt 31b and the upper edge of the transfer rail 31a. Through this mechanism, the metal mask 20 is positioned and fixed at the inspection position. When the inspection is finished, the fixing action by the chuck mechanism is released, and at the same time, the conveyor belt 31b starts to move. Thereby, the metal mask 20 is carried out by the metal mask inspection apparatus 30. As shown in FIG. Of course, the configuration of the transfer mechanism 31 is not limited to the above-mentioned configuration, and other configurations may be employed.

The inspection unit 32 is provided with: an illuminating device 32A as an illuminating means arranged below the transfer rail 31a (transfer path of the metal mask 20); a camera 32B as a photographing means arranged above the transfer rail 31a; The mechanism 32D is set so that the camera 32B can move in the X-axis direction (the left-right direction in FIG. 4 ) (see FIG. 5 ); and the Y-axis moving mechanism 32E is set so that the camera 32B can move in the Y-axis direction (the front-back direction in FIG. 4 ). (Refer to FIG. 5 ), and can be driven and controlled by the control device 33 (moving mechanism control section 76 described later).

In addition, the "inspection range" of the metal mask 20 is one of the plurality of areas of the metal mask 20 set in advance with the size of the imaging angle of view (shooting range) of the camera 32B as a unit.

The control device 33 (moving mechanism control part 76) can drive and control the X-axis moving mechanism 32D and the Y-axis moving mechanism 32E to move the camera 32B to any inspection range of the metal mask 20 positioned and fixed at the inspection position. Furthermore, by sequentially moving the camera 32B to a plurality of inspection ranges set on the metal mask 20 , inspections related to the inspection ranges are executed, and the inspection of the entire curved area of the metal mask 20 is performed simultaneously.

The illuminating device 32A includes: an illuminating panel 32Aa that emits predetermined light; and a diffuser plate (including a film, etc.) 32Ab that diffuses and uniformizes the light emitted from the illuminating panel 32Aa; Section 72) performs drive control.

The lighting panel 32Aa is composed of an LED substrate on which a plurality of LEDs (light emitting diodes) are mounted in a matrix, and is arranged so that the LED mounting surface faces upward in the Z-axis direction (upward in the vertical direction in FIG. 4 ). ). In addition, the light emitted from the lighting panel 32Aa passes through the diffuser plate 32Ab to be diffused, and is irradiated substantially uniformly on the entire lower surface of the metal cover 20 .

In addition, in this embodiment, the metal mask 20 is set in the metal mask inspection device 30 so that the contact surface side of the metal mask 20 in contact with the printed circuit board 1 faces downward in the metal mask inspection device 30 when printing flux. Mask 20.

The camera 32B is arranged on the opposite side of the lighting device 32A with the metal cover 20 interposed therebetween. In this embodiment, it is arranged above the metal mask 20, and the predetermined inspection range of the metal mask 20 is photographed from directly above.

The camera 32B has: a CCD (Charge Coupled Device, Charge Coupled Device) type image sensor or a CMOS (Complementary Metal Oxide Semiconductor, Complementary Metal Oxide Semiconductor) type image sensor and other imaging elements; The image of 21 is formed on the optical system (lens unit or aperture) of the photographing element. Of course, the imaging elements are not limited to these, and other imaging elements may also be used.

The camera 32B is driven and controlled by the control device 33 (camera control unit 73 described later). More specifically, the control device 33 synchronizes the illumination processing with the lighting device 32A, and executes the photographing processing by the camera 32B. In this way, among the lights irradiated from the lighting device 32A, the light passing through the opening 21 of the metal mask 20 is photographed by the camera 32B, and image data is generated.

In this way, the image data captured by the camera 32B is converted into a digital signal inside the camera 32B, and then sent to the control device 33 (the image acquisition unit 74 described later) in the form of a digital signal and stored therein. Then, the control device 33 (video processing unit 75 and the like to be described later) executes various video processing and calculation processing described later based on the video data.

The control device 33 is composed of a CPU (Central Processing Unit, Central Processing Unit) that performs predetermined calculation processing, and a ROM (Read Only Memory, Read Only Memory) that stores various programs or fixed value data, etc., and executes various calculation processes. It consists of RAM (Random Access Memory, Random Access Memory) which temporarily stores various data, and a computer including these peripheral circuits.

Moreover, the control device 33 is operated according to various programs through the CPU, so that the main control unit 71, the lighting control unit 72, the camera control unit 73, the image acquisition unit 74, the image processing unit 75, the moving mechanism control unit 76, the learning unit, etc. 77. Various functional parts such as the inspection part 78 and the transfer mechanism control part 79 function.

However, if the above-mentioned various functional parts are realized through the coordinated operation of various hardware such as the above-mentioned CPU, ROM, and RAM, it is not necessary to clearly distinguish between the functions realized by hardware or software. Some or all of these functions can also be realized by IC, etc. hardware circuit to achieve.

Furthermore, the control device 33 is provided with: an input unit 55 composed of a keyboard, a mouse, a touch panel, etc.; a display unit 56 having a display screen such as a liquid crystal display; a memory unit that can store various data or programs, calculation results, etc. 57; and a communication unit 58 that can send and receive various data to and from the outside.

Hereinafter, the various functional units constituting the control device 33 will be described in detail. The main control unit 7 is a functional unit responsible for overall control of the metal mask inspection device 30 , and is configured to transmit and receive various signals with other functional units such as the lighting control unit 72 and the camera control unit 73 .

The lighting control unit 72 is a functional unit that drives and controls the lighting device 32A, and constitutes the irradiation control means of the present embodiment.

The camera control unit 73 is a functional unit that drives and controls the camera 32B, and controls shooting timing and the like based on command signals from the main control unit 71 .

The video acquisition unit 74 is a functional unit for taking in video data captured by the camera 32B.

The video processing unit 75 is a functional unit that executes predetermined video processing on the video data captured by the video acquisition unit 74 . For example, in the learning process described later, video data for learning is generated as learning data used for learning of the deep neural network 90 (hereinafter, simply referred to as "neural network 90", see FIG. 6 ). In addition, inspection video data used when performing inspection processing described later is generated. In addition, it is preferable that the imaging conditions of the image data for learning used in the learning process and the imaging conditions of the image data for inspection used in the inspection process match as much as possible.

The moving mechanism control unit 76 is a functional unit that drives and controls the X-axis moving mechanism 32D and the Y-axis moving mechanism 32E, and controls the position of the camera 32B based on command signals from the main control unit 71 .

The learning unit 77 is a functional unit that performs learning of the neural network 90 using learning data or the like, and constructs an AI (Artificial Intelligence) model 100 as a discrimination means.

In addition, the AI model 100 in this embodiment, as will be described later, uses only the image data related to the opening 21 of the unused metal mask 20 to which no foreign matter is attached as learning data (image data for learning), and makes the neuron The network 90 is a generative model built for deep learning, and has a so-called autoencoder (self-encoder) structure.

The structure of the neural network 90 will now be described with reference to FIG. 6 . FIG. 6 is a schematic diagram conceptually illustrating the structure of the neural network 90 . As shown in FIG. 6 , the neural network 90 has: an encoder unit 91 as an encoding unit, which extracts feature quantities (latent variables) TA from input image data GA; and a decoder unit 92 as a decoding unit, which extracts feature quantities (latent variables) TA from The amount TA reconstructs (reconstitution) the image data GB to form a convolutional auto-encoder (CAE: Convolutional Auto-Encoder) structure.

Since the structure of a convolutional autoencoder is well known, its detailed description will be omitted, but the encoder unit 91 has a plurality of convolutional layers (Convolution Layer) 93, and in each convolutional layer 93, a plurality of filters are used for the input data. The result of the convolution calculation performed by the device (convolution kernel, kernel) 94 is output as the input data of the next layer. Similarly, the decoder unit 92 has a plurality of deconvolution layers (Deconvolution Layer) 95, and in each deconvolution layer 95, the result of performing deconvolution calculation on the input data using a plurality of filters (convolution kernels) 96 is output as the input data of the next layer. In addition, in the learning process mentioned later, the weight (parameter) of each filter 94,96 is updated.

The inspection unit 78 is a functional unit that inspects whether foreign matter is attached to the opening 21 of the metal mask 20 after cleaning.

The transfer mechanism control unit 79 is a functional unit that drives and controls the transfer mechanism 31 , and controls the position of the metal mask 20 according to the instruction signal from the main control unit 71 .

The memory unit 57 is composed of HDD (hard disk drive, Hard Disk Drive) or SSD (solid state drive, Solid State Drive), etc., for example, has a memory AI model 100 (the learning information obtained by the neural network 90 and its learning) The predetermined memory area. Such a memory area constitutes the model storage means of this embodiment.

The communication unit 58 includes, for example, a wireless communication interface conforming to communication standards such as wired LAN (Local Area Network) or wireless LAN, and is configured to transmit and receive various data to and from the outside. For example, the results of the inspection performed by the inspection unit 78 are transmitted to the cleaning device 29 and the like via the communication unit 58 . According to the inspection result information, the cleaning device 29 may also take the opening portion 21 of the metal mask 20 that is determined to be defective as the focus of cleaning.

Next, the learning process of the neural network 90 executed by the metal mask inspection device 30 will be described with reference to the flowchart of FIG. 7 .

First, the worker prepares the unused metal mask 20 . The metal mask 20 that has not been used here can also be replaced with a metal mask 20 that has been confirmed to be cleaned without remaining foreign matter.

In addition, the metal mask 20 prepared here preferably has an opening 21 having the same shape as the metal mask 20 to be inspected. However, the thickness and material of the metal mask 20 , the size and layout of the opening 21 do not have to be identical, but it is preferable in terms of versatility for learning from various learning data.

Next, the operator arranges the unused metal mask 20 at a predetermined inspection position of the metal mask inspection device 30 , and then instructs the main control unit 71 to execute a predetermined learning program.

According to the execution of the predetermined learning program, when the learning process starts, the main control unit 71 first executes the pre-processing for performing the learning of the neural network 90 in step S101.

Specifically, in accordance with an instruction from the main control unit 71 , first, the lighting control unit 72 turns on the lighting device 32A (lighting panel 32Aa). Next, according to the command from the main control unit 71 , the camera control unit 73 drives the camera 32B to take pictures of the predetermined inspection range of the metal mask 20 . With this configuration, the image data related to the predetermined inspection range of the metal mask 20 can be obtained. Here, the video data obtained by the camera 32B is input by the video acquisition unit 74 and subjected to predetermined video processing (for example, shading compensation or tilt correction) by the video processing unit 75 , and stored in the storage unit 57 .

In addition, the above-mentioned series of processes are repeatedly executed while moving within the inspection range on the metal mask 20 until the image data related to the necessary number of openings 21 is acquired as learning data.

In step S101, when the image data related to learning the required number of openings 21 is obtained, in the next step S102, the learning part 77 prepares the unlearned neural network 90 according to the instruction from the main control part 71. . For example, the neural network 90 previously stored in the memory unit 57 or the like is read. Alternatively, the neural network 90 is constructed based on network configuration information (for example, the number of layers of the neural network, the number of nodes in each layer, etc.) stored in the memory unit 57 or the like.

In step S103, video data for learning as learning data is acquired. Specifically, according to an instruction from the main control unit 71, the image processing unit 75 extracts one opening from among the plurality of openings 21 included in the image data based on the image data stored in the memory unit 57 in step S101. part 21, and the image data related to the opening part 21 is acquired as one image data for learning. Then, the image data for learning is output to the learning unit 77 . That is, only the image data related to the opening 21 of the unused metal mask 20 to which no foreign matter has adhered is used as learning data (image data for learning).

In step S104, reconstructed image data is obtained. Specifically, according to the instruction from the main control unit 71, the learning unit 77 will send the learning image data obtained in step S103 to the input layer of the neural network 90 as input data, so as to obtain the output layer of the neural network 90. The output reconstructed image data.

Next, in step S105, the learning unit 77 compares the learning image data obtained in step S103 with the reconstructed image data output by the neural network 90 in step S104, and determines whether the error is small enough (whether it is within below a predetermined threshold).

Here, when the error is sufficiently small, the neural network 90 and its learning information (updated parameters, etc. described later) are stored in the memory unit 57 as the AI model 100, and this learning process is terminated.

On the other hand, if the error is not small enough, the network update process (learning of the neural network 90 ) is performed in step S106 and then returns to step S103 to repeat the series of processes described above.

Specifically, in the network update process of step S105, the weights (parameters) of the above-mentioned filters 94, 96 in the neural network 90 are set using known learning algorithms such as Backpropagation. It is updated to make the loss function representing the difference between the image data for learning and the reconstructed image data as small as possible, whichever is more appropriate. In addition, as a loss function, BCE (Binary Cross-entropy) etc. can be used, for example.

By repeating these processes many times, in the neural network 90, the error between the image data for learning and the reconstructed image data becomes extremely small, and more accurate reconstructed image data can be output.

Next, the flow of various processes related to the cleaning process of the metal mask 20 will be described. In this embodiment, in the manufacturing process (line 10) of the above-mentioned printed circuit board 1, it is configured such that after the solder printing is performed a predetermined number of times by the solder printing machine 12, even if the solder printing is temporarily stopped, the cleaning device 29 is used for the solder printing. The metal mask 20 is cleaned. Hereinafter, with reference to the flowchart of FIG. 8 , a series of procedures from when the solder printing is stopped to when the cleaned metal mask 20 is returned to the solder printer 12 will be described in detail.

The solder printing machine 12 temporarily stops the solder printing operation when the number of printed circuit boards 1 on which the solder has been printed exceeds a predetermined number, and lights unillustrated lamps and the like in order to clean the metal mask 20, and expresses its meaning. Notify the operator (step S201). The operator who has received this notice unloads the used metal mask 20 from the solder printer 12 and transfers it to the cleaning device 29 .

Next, the operator carries the used metal mask 20 into the cleaning device 29 for cleaning (step S202 ). When the cleaning process is finished, the metal mask 20 that is about to be cleaned by the operator is carried out from the cleaning device 29 and transferred to the metal mask inspection device 30 . The cleaning process in this embodiment is constituted by performing a cleaning process.

Then, the worker carries the cleaned metal mask 20 into the metal mask inspection device 30, and performs a metal mask inspection process (step S203). Details about metal mask inspection processing will be described later. The execution procedure of this metal mask inspection process constitutes the mesh mask inspection method in this embodiment.

When the metal mask inspection device 30 is judged as cleaning failure (judgment failure) by the metal mask inspection process (step S203), in order to clean the cleaned metal mask 20 again, the display unit 56 or the communication unit 58 Inform the operator of the intention.

The operator who has received the notification unloads the metal mask 20 with poor cleaning from the metal mask inspection device 30, and then transfers it to the cleaning device 29 (step S204: Yes). Such a process corresponds to the retransfer process in this embodiment. Next, the operator will carry the poorly cleaned metal mask 20 into the cleaning device 29, and perform the cleaning process again (step S202).

On the other hand, when the metal mask inspection device 30 determines that the cleaning process has been properly executed (passed) by the metal mask inspection process (step S203), in order to return the cleaned metal mask 20 to the solder printer 12, This fact is notified to the operator via the display unit 56, the communication unit 58, or the like.

The operator who has received this will take out the cleaned metal mask 20 from the metal mask inspection device 30, and transfer it to the solder printer 12 (step S204: NO). Then, the worker installs the cleaned metal mask 20 on the solder printing machine 12, and prepares for resumption of the solder printing process (step S205). Such a process corresponds to the printing preparation process in this embodiment.

Here, the metal mask inspection process executed by the metal mask inspection device 30 will be described with reference to the flowchart of FIG. 9 . However, the inspection process shown in FIG. 9 is performed for each predetermined inspection range of the metal mask 20 .

When the metal mask 20 is loaded into the metal mask inspection device 30 and positioned at a predetermined inspection position, the inspection process is started according to a predetermined inspection program.

Once the inspection process is started, the lighting process is first performed in step S301. Specifically, the main control unit 71 outputs a predetermined signal to the lighting control unit 72 when it determines that the metal shield 20 is positioned at the inspection position based on a signal from a sensor not shown. Accordingly. The lighting control unit 72 turns on the lighting device 32A. Accordingly, the light irradiates the entire lower area of the metal shield 20 .

In the next step S302, a photographing process is performed. The process of executing this imaging process corresponds to the imaging process in this embodiment. Specifically, the camera control unit 73 drives the camera 32B according to an instruction from the main control unit 71 . Thereby, a predetermined inspection range on the metal mask 20 can be photographed, and image data of an area including a plurality of openings 21 can be obtained. Then, the video data of the area is input to the video acquisition unit 74 .

In step S303, image processing is performed. Specifically, according to an instruction from the main control unit 71 , the image processing unit 75 is made to generate inspection image data related to each opening 21 based on the area image data captured by the image acquisition unit 74 in step S302 .

When generating the image data for inspection, first, all of the plurality of openings 21 included in the area image data are identified and captured. Next, these data are numbered and registered as original image data related to one opening 21 to be inspected. The original image data acquisition means in this embodiment is constituted by the function of executing such processing.

In this process, the original image data related to the opening 21 not attached with foreign matter as shown in FIG. 10( a ) or the opening 21 with foreign matter W1 and W2 attached as shown in FIG. Related original image data, etc.

In step S304, reconstruction processing is performed. The reconstruction image data acquisition means in this embodiment is constituted by executing the function of this processing.

Specifically, according to the instruction from the main control unit 71, the inspection unit 78 will input the original image data related to the opening 21 of the predetermined number (for example, No. 001) acquired in step S303 into the input of the AI model 100. layer. Next, obtain the image data reconstructed by the AI model 100 and output from the output layer and the reconstructed image data related to the opening 21 of the aforementioned predetermined number (for example, No. 001), and the reconstructed image data and the original image of the same number Image data is associated and memorized. In this manner, in this process, reconstructed image data can be obtained for all the openings 21 numbered and registered in step S303 .

Here, in the AI model 100, it is natural to input the original image data related to the opening 21 with no foreign matter attached as shown in FIG. The original image data related to the openings 21 with foreign objects W1 and W2 attached is also learned as described above, and the image data related to the openings 21 without foreign objects attached as shown in FIG. 10(c) It is output as reconstructed image data.

In step S305, a good-failure judgment process is executed. Specifically, according to the instruction from the main control unit 71 , the inspection unit 78 firstly compares the original image data with the same number with the reconstructed image data, and extracts the difference between the two image data. The comparison means in this embodiment can be constituted by the function of executing such processing.

Next, the inspection unit 78 calculates the number of pixels corresponding to the difference between the two image data, and determines whether it is greater than a predetermined threshold. Here, when the difference between the two image data is greater than a predetermined threshold, it is determined that "there is a foreign object".

On the other hand, if the number of pixels corresponding to the difference between the two image data is smaller than the predetermined threshold value, it is judged as "no foreign matter". The judging means of this embodiment can be constituted by performing these judging functions.

Here, when the inspection unit 78 judges that all the openings 21 included in the area image data (predetermined inspection range of the metal mask 20) are “no foreign matter”, the inspection range related to the area image data is judged as “good”. ", and the result is stored in the storage unit 57, and this process ends.

On the other hand, among the plurality of openings 21 included in the area image data (predetermined inspection range of the metal mask 20), even if there is even one opening 21 that is judged to have "a foreign object", it will be compared with the area. The inspection range related to the image data is judged as "defective", and the result is stored in the storage unit 57, and this process ends.

Next, when the metal mask inspection device 30 executes the above-mentioned inspection processing results on the entire inspection range on the metal mask 20, and judges that the entire inspection range is "good", it judges that the cleaning has been properly performed (judgment is passed), and then In the above method, in order to return the cleaned metal mask 20 to the solder printer 12, the operator is notified of the intention through the display unit 56, the communication unit 58, and the like.

On the other hand, even if there is even one inspection area judged as "defective" among all the inspection areas on the metal mask 20 by the metal mask inspection device 30, it is judged as poor cleaning (judgment failed), and if In the above method, in order to clean the cleaned metal mask 20 again, the operator is notified of the intention through the display unit 56 or the communication unit 58 or the like. Therefore, by executing the above-mentioned series of processing steps, the cleaning quality judgment step of the present embodiment can be constituted.

As described in detail above, according to the present embodiment, the metal mask inspection device 30 is configured to inspect the inner wall of the opening 21 of the metal mask 20 after cleaning the metal mask 20 used for solder printing. Check whether there are foreign objects attached to the inside. In this metal mask inspection, the metal mask 20 is cleaned again when it is determined that foreign matter is attached to the inner wall of the opening 21 of the metal mask 20 .

With this configuration, regardless of whether or not the metal mask 20 has been cleaned, tiny foreign matter (residues such as cream solder 5 or flux) remains in a state of adhering to the inner wall of the opening 21 of the metal mask 20, and this occurs. Inappropriate reuse of the metal mask 20 in solder printing can be suppressed.

As a result, clogging of the opening 21 of the metal mask 20 (poor printing), improper soldering of the electronic parts 25 (flux shortage), and various inappropriate situations such as continuous flux printing in a deteriorated state of flux stripping can be suppressed in advance. .

In particular, in the present embodiment, the AI model 100 of the neural network 90 constructed by learning is used to determine whether foreign matter is attached to the opening 21 of the metal mask 20, so the opening 21 on the image data or the Microscopic foreign matter that is difficult to detect in the configuration where the area, shape, and size of the foreign matter are compared can also be detected.

Furthermore, in this embodiment, since the original image data captured by the opening 21 is compared with the reconstructed image data reconstructed based on the original image data, there will be no difference between the two compared image data. The imaging conditions of the metal mask 20 side of the inspection object (for example, the arrangement position or arrangement angle of the metal mask 20, bending, etc.), or the imaging conditions of the metal mask inspection device 30 side (for example, the lighting state or the position of the camera 32B). Due to the influence of differences in the picture angle, etc.), finer foreign matter can be detected more accurately.

In addition, when inspecting the predetermined opening 21 located at a predetermined position on the metal mask 20 , in terms of quality judgment criteria, the configuration information (positional data, dimension data, etc.) of the opening 21 is required. In this method, the board design information such as Jaber data is stored in advance, the configuration information of the opening 21 to be inspected is appropriately obtained, and the quality judgment of the opening 21 to be inspected is performed while comparing with the configuration information. There are concerns about low inspection efficiency. Furthermore, the positioning of the inspection position of the metal mask 20 must be performed correctly.

On the other hand, according to the present embodiment, since the AI model 100 learned for the openings 21 is used to inspect the openings 21, it is not necessary to memorize the respective configuration information of a plurality of openings 21 in advance, and to perform the inspection. There is no need for reference, so the inspection efficiency can be improved.

In addition, it is not limited to the description content of the said embodiment, For example, it can implement as follows. Of course, other application examples and modification examples not illustrated below are also possible.

(a) The configuration of the mesh mask is not limited to the metal mask 20 related to the above-mentioned embodiment, and other configurations may be employed.

(a-1) For example, the metal mask 20 related to the above embodiment is composed of the mask body portion 20a and the frame portion 20b, but the present invention is not limited thereto, and the frame portion 20b may be omitted.

(a-2) The mask body portion 20a related to the above-mentioned embodiment is formed of a metal material. It is not limited thereto, and as the mask main body portion 20a, for example, a “yarn (mesh)” made of yarns made of nylon, polyester, or stainless steel may be used to coat a photosensitive emulsion or the like to form a covering portion. Only the part corresponding to the opening 21 is exposed only to the product of the "yarn" type, and a composite product in which the opening 21 is formed on the "yarn (screen)".

(b) The configuration related to the production of the printed circuit board 1 such as the production line 10 is not limited to the above-mentioned embodiment, and other configurations can be employed.

(b-1) In the above embodiment, the cleaning device 29 and the metal mask inspection device 30 are provided in the vicinity of the solder printer 12 in combination. However, it is not limited thereto, and a structure in which the functions of the cleaning device and/or the metal mask inspection device and the solder printing machine 12 or the solder printing inspection device 13 are integrated may be adopted.

(b-2) In the above-mentioned embodiment, when performing the cleaning process, an operator performs cleaning between the flux printing machine 12 and the cleaning device 29, between the cleaning device 29 and the metal mask inspection device 30, And the operation configuration of transferring the metal mask 20 between the metal mask inspection device 30 and the solder printing machine 12 . Without being limited thereto, transfer means such as a conveyor for transferring the metal mask 20 may be arranged between each of these devices to automate the transfer operation. That is, it is also possible to construct a manufacturing system that combines the production line 10, the cleaning device 29, and the metal mask inspection device 30.

(c) The configuration related to the metal mask inspection device 30 is not limited to the above-mentioned embodiment, and other configurations may be employed.

(c-1) In the above-described embodiment, the metal mask 20 is set in the metal mask inspection device 30 so that the substrate contact surface side of the metal mask 20 that contacts the printed circuit board 1 faces downward during solder printing. constitute. However, the present invention is not limited thereto, and the metal mask 20 may be set in the metal mask inspection device 30 so that the substrate contact surface side of the metal mask 20 faces upward.

(c-2) In the above-mentioned embodiment, during solder printing, the illuminating device 32A is arranged on the substrate contact surface side of the metal mask 20 which is in contact with the printed circuit board 1, and the camera 32B is arranged on the solder printing surface. The substrate non-contact surface side of the metal mask 20 that does not contact the printed circuit board 1 during printing. However, it is not limited thereto, and the camera 32B may be disposed on the side of the metal mask 20 in contact with the substrate, and the illuminating device 32A may be disposed on the side of the metal mask 20 not in contact with the substrate.

(c-3) In the above embodiment, the illuminating device 32A is arranged below the metal mask 20, the camera 32B is arranged above the metal mask 20, and the inspection related to the opening 21 is carried out by the transmitted light inspection. composition. However, it is not limited to this, but the illuminating device 32A and the camera 32B are collectively arranged above or below the metal shield 20, and the inspection related to the opening 21 is performed by reflected light inspection.

(d) The configuration of the AI model 100 (neural network 90 ) as a discrimination means and its learning method are not limited to the above-mentioned embodiments.

(d-1) In the above-mentioned embodiment, it is also possible to perform reconstruction processing on various data as necessary when the reconstruction processing is performed in the learning processing of the neural network 90 or the inspection processing of the metal mask. Formation of processing such as normalization.

(d-2) The structure of the neural network 90 is not limited to that shown in FIG. 6 , and for example, a structure in which a pooling layer (pooling layer) is provided after the convolutional layer 93 may also be adopted. Of course, different configurations such as the number of layers of the neural network 90, the number of nodes in each layer, and the connection structure of each node may be used.

(d-3) In the above-mentioned embodiment, the AI model 100 (neural network 90) is a generative model having a convolutional autoencoder (CAE) structure, but it is not limited thereto. A generative model constructed by different types of autoencoders such as VAE (Variational Autoencoder).

(d-4) In the above-mentioned embodiment, the neural network 90 is learned by the error back-propagation method, but it is not limited to this, and other various learning algorithms may be used for learning.

(d-5) The neural network 90 may also be formed by using a special circuit for AI processing such as a so-called AI chip. In this case, only learning information such as parameters can be stored in the storage unit 57 , read out by an AI processing dedicated circuit, and then set in the neural network 90 to form the AI model 100 .

(d-6) In the above-mentioned embodiment, the learning unit 77 is provided, and the learning of the neural network 90 is performed in the control device 33, but it is not limited to this, as long as at least the AI model 100 (that has completed the learning) The neural network 90) can be memorized in the memory unit 57, and the configuration of the learning unit 77 is omitted. Therefore, learning of the neural network 90 may be performed outside the control device 33 and stored in the memory unit 57 .

(e) The inspection method of the opening 21 in the metal mask inspection process is not limited to the above embodiment, and other configurations may be employed.

For example, in the above-mentioned embodiment, the original image data is compared with the reconstructed image data, and the difference between the two image data is extracted. By calculating the number of pixels corresponding to the difference, and determining whether the number of pixels is greater than a predetermined threshold value, and The presence or absence of foreign matter attached to the opening 21 is determined.

It is not limited to this, and it can also be made, for example, by comparing the images related to the inner wall of the opening 21 in the two image data (for example, it is also possible to take the straight line degree of the flat part of the inner wall of the opening 21, or the inner wall of the opening 21). The degree of bending of the corner of the wall, etc.) to detect the composition of foreign objects.

1: Printed substrate 3: Island Department 5: Paste flux 10: Production line 12: Solder printing machine 20: metal mask 21: opening 29: Cleaning device 30: Metal mask inspection device 32A: Lighting device 32B: Camera 33: Control device 77: Learning Department 78: Inspection Department 90: Neural Networks 100: AI model W: foreign matter

FIG. 1 is a partially enlarged plan view of a part of a printed circuit board enlarged. FIG. 2 is a block diagram showing the configuration of a production line for printed circuit boards. Fig. 3 is a schematic diagram for explaining the printing operation of the solder printer. FIG. 4 is a schematic configuration diagram schematically showing a metal mask inspection device. Fig. 5 is a block diagram showing the functional configuration of the metal mask inspection device. FIG. 6 is a schematic diagram illustrating the structure of a neural network. FIG. 7 is a flowchart of the flow of learning processing of the neural network. FIG. 8 is a flow chart of the flow of various processes related to the cleaning process. FIG. 9 is a flowchart of the flow of inspection processing. In Figure 10, (a) is the original image data map related to the opening without foreign matter attached; (b) is the original image data map related to the opening with foreign matter attached; (c) is the original image data map related to the opening with foreign matter attached A reconstructed image data map obtained by reconstructing the original image data related to the opening.

30: Metal mask inspection device

31: transfer mechanism

32: Check unit

32A: Lighting device

32B: Camera

32D: X-axis moving mechanism

32E: Y-axis moving mechanism

33: Control device

55: Input part

56: Display part

57: memory department

58: Department of Communications

71: Main Control Department

72: Lighting Control Department

73: Camera Control Department

74: Image Acquisition Department

75: Image processing department

76: Mobile mechanism control department

77: Learning Department

78: Inspection Department

79: Transfer mechanism control department

100: AI model

Claims (4)

A mesh mask inspection method for inspecting a cleaned mesh mask before performing flux printing on a substrate, comprising: an irradiation step of irradiating predetermined light to an opening of the mesh mask The photographing process is to photograph the opening of the aforementioned net mask irradiated by the aforementioned light; the original image data acquisition process is to obtain the original image data related to the aforementioned opening according to the image data obtained by the aforementioned photographing process; reconstruct the image data The obtaining step is to obtain the reconstructed image data related to the opening part reconstructed by inputting the original image data to the identification means, wherein the identification means has an encoding part which extracts a feature quantity from the input image data and reconstructs the opening from the feature quantity. The neural network of the decoding part of the image data is generated by learning only the image data of the opening part without foreign matter attached thereto as learning data; the comparison process compares the original image data with the reconstructed image data; and the determination process, Based on the comparison result in the comparison step, it is judged whether or not foreign matter is attached to the inner wall of the opening of the mesh cover. The method for inspecting a mesh mask according to claim 1, wherein, in the comparison process, the image related to the inner wall of the opening in the original image data is compared with the inner wall of the opening in the reconstructed image data related images for comparison. A mesh mask inspection device, which is used to perform soldering on a substrate Before the agent printing, it is used to check the cleaned net mask, and it has: the irradiation means, which can irradiate the predetermined light to the opening of the aforementioned net mask; the photographing means, which can photograph the opening of the aforementioned net mask after the aforementioned light The image processing means can perform inspections related to the inner wall of the opening of the aforementioned net mask according to the image data obtained by the aforementioned shooting means, wherein the aforementioned image processing means has: a distinguishing means, which is to enable The neural network of the encoding part that extracts the feature value from the image data and the decoding part that reconstructs the image data from the feature value is generated by learning only the image data of the aforementioned opening with no foreign matter attached thereto; the original image data The obtaining means can obtain the original image data related to the aforementioned opening, and the aforementioned opening is the opening of the aforementioned cleaned net mask photographed by the aforementioned shooting means; the reconstructed image data obtaining means can obtain and convert the aforementioned original image data The reconstructed image data related to the aforementioned opening is input and reconstructed by the aforementioned identifying means; the comparing means can compare the aforementioned original image data with the reconstructed image data; Check whether there is any foreign matter attached to the inner wall of the opening of the mask. Such as the net mask inspection device of claim 3, wherein, the former The comparison means compares the image related to the inner wall of the opening in the original image data with the image related to the inner wall of the opening in the reconstructed image data.

Images