JP7451364B2 - Tablet printing device and tablet printing method - Google Patents

Description

The present invention relates to a tablet printing device and a tablet printing method that print on the surface of a tablet while conveying the tablet.

BACKGROUND ART Tablet printing devices have been known that use an inkjet method to print images such as letters and logos on the surface of tablets such as pharmaceuticals and supplements. A conventional tablet printing device is described in, for example, Patent Document 1.

In a tablet printing device, a plurality of tablets are conveyed in a lined state and each tablet is photographed. Then, the shape, position, rotation angle, etc. of each tablet are determined based on the obtained captured image. Thereafter, printing is performed on the tablets while finely adjusting the position of the ink droplet to be ejected onto each tablet according to the above determination result. For example, in the apparatus of Patent Document 1, the rotation angle of the score line of the tablet is determined based on a photographed image of the tablet, and printing is performed along the score line.

JP 2017-200495 Publication

However, the shapes and surface conditions of the plurality of tablets conveyed in this type of tablet printing apparatus are not strictly constant. For this reason, in a photographed image of a tablet, irregular gloss or roughness due to the roughness of the drug component may appear on the surface of the tablet. Additionally, some tablets may have defects such as chips or black spots. Therefore, the shape, position, or rotation angle of the tablet may not be accurately determined due to the influence of these gloss, roughness, or defects.

The present invention has been made in view of these circumstances, and provides a tablet printing method that can accurately determine the shape, position, or rotation angle of a tablet, and thereby accurately print on the surface of the tablet. An object of the present invention is to provide an apparatus and a tablet printing method.

In order to solve the above problems, a first invention of the present application is a tablet printing device that prints on the surface of a tablet while transporting the tablet, the apparatus comprising a transport mechanism that transports the tablet, and a tablet printer that prints on the surface of the tablet while transporting the tablet. an imaging unit that captures images to obtain captured images; an autoencoder that performs dimension reduction and reconstruction on the captured images to output a reconstructed image; and an autoencoder that outputs a reconstructed image based on the reconstructed image a determination unit that determines at least one of the shape, position, and rotation angle of the tablet; a printing unit that performs printing on the surface of the tablet based on the determination result of the determination unit; and a printing unit that performs printing on the surface of the tablet based on the reconstructed image. The autoencoder includes an inspection area setting section that sets an inspection area, and an inspection section that inspects the tablet for defects in the inspection area set by the inspection area setting section of the photographed image, and the autoencoder is configured to The dimension reduction and the reconstruction are performed using parameters adjusted in advance by machine learning so that the difference between the tablet image and the output tablet image is reduced.

A second invention of the present application is the tablet printing apparatus of the first invention, in which the determination unit determines the rotation angle of the tablet based on the score line of the tablet included in the reconstructed image.

A third invention of the present application is the tablet printing apparatus according to the first invention or the second invention , wherein the inspection area setting section sets the plurality of inspection areas based on the reconstructed image, and the inspection section includes: The tablet is inspected for defects under different conditions for each inspection area.

A fourth invention of the present application is the tablet printing apparatus according to the third invention, wherein the plurality of inspection areas include an area corresponding to an edge of the tablet.

A fifth invention of the present application is a tablet printing method for printing on the surface of a tablet while conveying the tablet, the method comprising: a) inputting an image of the tablet to an autoencoder that performs dimension reduction and reconstruction; a step of adjusting the parameters of the autoencoder so that the difference between the tablet image and the output reconstructed tablet image is small; and b) photographing the tablet while transporting the tablet to obtain a photographed image. c) inputting the photographed image to the autoencoder whose parameters have been adjusted in step a) to obtain a reconstructed image output from the autoencoder, and d) the reconstructed image. e) determining at least one of the shape, position, and rotation angle of the tablet based on the determination result of step d); f) printing on the surface of the tablet based on the determination result of step d); The method includes a step of setting an inspection area of the tablet based on the constituent images, and g) a step of inspecting the tablet for defects in the inspection area of the photographed image set in the step f).

A sixth invention of the present application is the tablet printing method according to the fifth invention, wherein in step d), the rotation angle of the tablet is determined based on the score line of the tablet included in the reconstructed image.

A seventh invention of the present application is the tablet printing method according to the fifth or sixth invention , wherein in the step f), a plurality of the inspection areas are set based on the reconstructed image, and in the step g), The tablet is inspected for defects using different threshold values for each of the inspection areas.

An eighth invention of the present application is the tablet printing method according to the seventh invention, wherein the plurality of inspection areas include an area corresponding to an edge of the tablet.

According to the first to eighth inventions of the present application, the autoencoder can generate a reconstructed image in which random elements such as gloss, roughness, and defects are reduced from a captured image. Therefore, the shape, position, or rotation angle of the tablet can be determined with high accuracy based on the reconstructed image. Therefore, it is possible to accurately print on the surface of the tablet.

Further , according to the first to eighth inventions of the present application, the inspection area can be set with high accuracy based on the reconstructed image. Therefore, defects in tablets can be inspected with high accuracy.

In particular, according to the third or seventh invention of the present application, a plurality of inspection areas can be set with high accuracy based on the reconstructed image.

In particular, according to the fourth or eighth aspect of the present application, the inspection area corresponding to the edge of the tablet, which is likely to be misaligned, can be set with high accuracy by using the reconstructed image.

1 is a diagram showing the configuration of a tablet printing device. It is a partial perspective view of a conveyance mechanism. FIG. 3 is a bottom view of the head. FIG. 2 is a block diagram showing electrical connections between a computer and various parts of the tablet printing device. FIG. 2 is a block diagram conceptually showing the functions of a computer. FIG. 3 is a diagram conceptually illustrating dimension reduction and reconstruction processing performed by an autoencoder. FIG. 3 is a diagram showing an example of an inspection area. It is a flowchart showing the flow of machine learning. 3 is a flowchart showing the flow of print processing.

Embodiments of the present invention will be described below with reference to the drawings. In addition, in the following description, the direction in which a plurality of tablets are conveyed will be referred to as the "transportation direction", and the direction perpendicular and horizontal to the transportation direction will be referred to as the "width direction".

<1. Overall configuration of printing device>
FIG. 1 is a diagram showing the configuration of a tablet printing apparatus 1 according to an embodiment of the present invention. This tablet printing device 1 is a device that prints images such as a product name, product code, company name, logo mark, etc. on the surface of each tablet 9 while conveying a plurality of tablets 9. As shown in FIG. 1, the tablet printing apparatus 1 of this embodiment includes a transport mechanism 10, a first imaging section 20, a printing section 30, a second imaging section 40, a defective product discharging section 50, a non-defective product discharging section 60, and a computer. It is equipped with 70.

The transport mechanism 10 is a mechanism that transports a plurality of tablets 9 while holding them in an aligned state. The conveyance mechanism 10 includes a pair of pulleys 11 and an annular conveyance belt 12 stretched between the pair of pulleys 11 . A plurality of tablets 9 loaded into the tablet printing apparatus 1 are arranged at equal intervals and supplied onto the surface of the conveyor belt 12 by a carry-in mechanism 80 comprised of a vibrating feeder, a conveyor drum, and the like. One of the pair of pulleys 11 is rotated by power obtained from the transport motor 13. As a result, the conveyor belt 12 rotates in the direction of the arrow in FIG. At this time, the other of the pair of pulleys 11 rotates as the conveyor belt 12 rotates.

FIG. 2 is a partial perspective view of the transport mechanism 10. As shown in FIG. 2, the conveyor belt 12 is provided with a plurality of suction holes 14. The plurality of suction holes 14 are arranged at equal intervals in the transport direction and the width direction. Further, as shown in FIG. 1, the conveyance mechanism 10 includes a suction mechanism 15 that sucks out gas from the space inside the conveyance belt 12. When the suction mechanism 15 is operated, the space inside the conveyor belt 12 becomes a negative pressure lower than atmospheric pressure. The plurality of tablets 9 are suction-held in the suction holes 14 by this negative pressure.

In this way, the plurality of tablets 9 are held on the surface of the conveyor belt 12 in a state where they are aligned in the conveyance direction and the width direction. The transport mechanism 10 then transports the plurality of tablets 9 in the transport direction by rotating the transport belt 12. A plurality of tablets 9 are conveyed in the horizontal direction below four heads 31, which will be described later.

The first imaging unit 20 is a part that acquires a photographed image of the tablet 9 before printing. The first imaging unit 20 captures a plurality of tablets 9 transported by the transport belt 12 at a first imaging position P1 downstream of the transport path from the carry-in mechanism 80 and upstream of the transport path from a print position P2, which will be described later. , taken from above. The first imaging unit 20 uses, for example, a line sensor in which imaging elements such as CCD and CMOS are arranged in the width direction. However, other imaging devices such as an area sensor may be used in the first imaging unit 20. The photographed image obtained by photographing is transmitted from the first imaging section 20 to a computer 70, which will be described later. Based on the captured image obtained from the first imaging unit 20, the computer 70 determines the presence or absence of the tablet 9 in each suction hole 14, the position of the tablet 9, the rotation angle of the tablet 9, and whether the tablet 9 has no defects such as chips. Inspect whether or not.

The printing unit 30 prints an image on the surface of the tablet 9 using an inkjet method at a printing position P2 that is downstream of the first imaging position P1 in the transport path and upstream of the second imaging position P3, which will be described later, in the transport path. This is a processing section that performs As shown in FIG. 1, the printing section 30 of this embodiment has four heads 31. The four heads 31 are located above the conveyor belt 12 and arranged in a line along the conveyance direction of the tablets 9. The four heads 31 eject ink droplets of different colors (for example, cyan, magenta, yellow, and black) toward the surface of the tablet 9. Then, a multicolor image is printed on the surface of the tablet 9 by superimposing the monochrome images formed by these colors. Note that the ink ejected from each head 31 is an edible ink manufactured from raw materials approved by the Japanese Pharmacopoeia, the Food Sanitation Law, etc.

FIG. 3 is a bottom view of one head 31. In FIG. 3, the conveyor belt 12 and the plurality of tablets 9 held by the conveyor belt 12 are shown by two-dot chain lines. As shown in an enlarged view in FIG. 3, a plurality of nozzles 311 capable of ejecting ink droplets are provided on the ejection surface 310, which is the lower surface of the head 31. In this embodiment, a plurality of nozzles 311 are two-dimensionally arranged on the lower surface of the head 31 in the transport direction and the width direction. The nozzles 311 are arranged with their positions shifted in the width direction. In this way, by arranging the plurality of nozzles 311 two-dimensionally, the positions of the nozzles 311 in the width direction can be made close to each other. However, the plurality of nozzles 311 may be arranged in a line along the width direction.

As a method for ejecting ink droplets from the nozzle 311, for example, a so-called piezo method is used in which ink in the nozzle 311 is pressurized and ejected by applying a voltage to a piezo element to deform it. However, the ink droplet ejection method may be a so-called thermal method in which the ink droplets are ejected by energizing a heater to heat and expand the ink in the nozzle 311 .

The second imaging unit 40 is a part that acquires a photographed image of the tablet 9 after printing. The second imaging unit 40 captures a plurality of tablets being transported by the transport belt 12 at a second imaging position P3 that is downstream of the transport path from the printing position P2 and upstream of the transport path from a defective product discharge position P4, which will be described later. 9 is photographed from above. The imaging unit 40 uses, for example, a line sensor in which imaging elements such as CCD and CMOS are arranged in the width direction. However, other imaging devices such as an area sensor may be used in the second imaging unit 40. The photographed image acquired by photographing is transmitted from the second imaging section 40 to a computer 70, which will be described later. The computer 70 inspects the quality of the printed image formed on the surface of the tablet 9 based on the second photographed image obtained from the second imaging section 40 .

The defective product discharging unit 50 removes the tablets 9 that have been determined to be defective at a defective product discharging position P4 that is downstream of the second imaging position P3 on the transport path and upstream of the non-defective product discharging position P5, which will be described later. This is the part where it is excreted. The defective product discharging section 50 includes a defective product blowing mechanism 51 disposed inside the conveyor belt 12 and a defective product collecting section 52 disposed below the defective product blowing mechanism 51 with the conveyor belt 12 interposed therebetween. The defective product blowing mechanism 51 blows gas only to the suction hole 14 that holds the tablet 9 determined to be a defective product among the plurality of suction holes 14 of the conveyor belt 12 . Then, the suction hole 14 has a positive pressure higher than atmospheric pressure. As a result, the suction of the tablet 9 in the suction hole 14 is released, and the tablet 9 determined to be a defective product falls from the conveyor belt 12 to the defective product collection section 52.

The non-defective product discharging section 60 is a part that discharges the tablets 9 determined to be non-defective products at a non-defective product discharging position P5 downstream of the defective product discharging position P4 on the conveying path. The non-defective product discharge section 60 includes a non-defective product blowing mechanism 61 disposed inside the conveyor belt 12 and a non-defective product collecting section 62 disposed below the non-defective product blow mechanism 61 with the transport belt 12 in between. The non-defective blow mechanism 61 blows gas toward the plurality of suction holes 14 of the conveyor belt 12 . Then, the suction hole 14 has a positive pressure higher than atmospheric pressure. As a result, the suction of the tablet 9 in the suction hole 14 is released, and the tablet 9 determined to be a good product falls from the conveyor belt 12 to the good product collection section 62.

The computer 70 is an information processing device that controls the operation of each part in the tablet printing device 1 and performs machine learning described later. FIG. 4 is a block diagram showing electrical connections between the computer 70 and each part of the tablet printing apparatus 1. As conceptually shown in FIG. 4, the computer 70 includes a processor 701 such as a CPU, a memory 702 such as a RAM, and a storage section 703 such as a hard disk drive. In the storage unit 703, a computer program CP for executing print processing and machine learning to be described later is stored. Furthermore, the storage unit 703 also stores a trained model M obtained by machine learning.

The computer 70 can communicate with the above-described transport mechanism 10, first imaging section 20, four heads 31, second imaging section 40, defective product blowing mechanism 51, good product blowing mechanism 61, and carry-in mechanism 80 by wire or wirelessly. It is connected to the. The computer 70 controls the operation of each of these parts according to the computer program CP and the learned model M.

<2. About computers＞
FIG. 5 is a block diagram conceptually showing the functions of the computer 70 described above. As shown in FIG. 5, the computer 70 includes an autoencoder 71, a determination section 72, a print instruction section 73, an inspection area setting section 74, and an inspection section 75. The functions of the autoencoder 71, the determination section 72, the print instruction section 73, the inspection area setting section 74, and the inspection section 75 are realized by the processor 701 of the computer 70 operating according to the computer program CP.

The autoencoder 71 is a processing unit that outputs a reconstructed image D3 by performing dimension reduction (encoding) and reconstruction (decoding) on the input captured image D1 using a multilayer neural network. FIG. 6 is a diagram conceptually showing the dimension reduction and reconstruction processing executed by the autoencoder 71.

As shown in FIG. 6, dimension reduction is a process of compressing an input high-dimensional captured image D1 into low-dimensional data called latent variables D2. Reconstruction is a process of reproducing the image of the tablet 9 from the latent variable D2 to generate a reconstructed image D3. In the dimension reduction process, the autoencoder 71 reduces random information such as gloss, roughness, and defects while leaving essential information such as the shape, color, and direction of the dividing line of the tablet 9. Therefore, the reconstructed image D3 generated by the reconstruction is an image of the tablet 9 in which random elements such as gloss, roughness, and defects are reduced.

The autoencoder 71 can adjust parameters used in each process of dimension reduction and reconstruction by performing machine learning. In machine learning, a large number of photographed images D1 of tablets 9 of good quality are used, and the difference between the photographed image D1 of the tablet 9 input to the autoencoder 71 and the reconstructed image D3 of the tablet 9 output from the autoencoder 71 is calculated. Adjust the dimensionality reduction and reconstruction parameters to make it smaller. As a result, a learned model M including the adjusted parameters is obtained.

The learned model M is stored in the storage unit 703 of the computer 70. When printing the tablet 9 in the tablet printing apparatus 1, the autoencoder 71 executes dimension reduction and reconstruction processing using the parameters indicated by the learned model M obtained through prior machine learning. Therefore, when the photographed image D1 of the tablet 9 acquired by the first imaging unit 20 is input to the autoencoder 71, the reconstructed image D3 of the tablet 9 in which random elements such as gloss, roughness, and defects are reduced is input to the autoencoder 71. It can be output from.

The determination unit 72 is a processing unit that determines the shape, position, and rotation angle of the tablet 9 conveyed by the conveyance mechanism 10. The determination unit 72 determines the shape, position, and rotation angle of the tablet 9 by performing image processing on the reconstructed image D3 output from the autoencoder 71. Specifically, the determining unit 72 extracts a portion corresponding to the edge of the tablet 9 from the reconstructed image D3, and determines the shape and position of the tablet 9 based on the shape and position of the edge. Further, the determination unit 72 extracts a portion corresponding to the score line of the tablet 9 from the reconstructed image D3, and determines the rotation angle of the tablet 9 based on the shape and position of the score line. In addition, in this application, "rotation angle" means the rotation angle of the tablet 9 with respect to the axis which passes through the center of the tablet 9 and is perpendicular to the conveyance direction and the width direction. The determination unit 72 outputs these determination results D4 to the print instruction unit 73.

The print instruction unit 73 is a processing unit that causes the printing unit 30 to perform ink ejection processing. The determination result D4 of the determination unit 72 and image data D5 indicating an image to be printed on the tablet 9 are input to the print instruction unit 73. The print instruction unit 73 adjusts the image data D5 according to the determination result D4 input from the determination unit 72. Specifically, the print instruction unit 73 shifts the coordinates of the image data D5 according to the position of the tablet 9 indicated by the determination result D4. Further, the print instruction unit 73 rotates the image data D5 according to the rotation angle of the tablet 9 indicated by the determination result D4. Then, the print instruction section 73 outputs a print instruction D6 to the plurality of heads 31 of the printing section 30 based on the adjusted image data D5.

The plurality of heads 31 eject ink droplets onto the upper surface of the tablet 9 being transported based on the print instruction D6 output from the print instruction section 73. As a result, the image is printed at an appropriate position on the surface of the tablet 9 at an appropriate rotation angle.

Note that a plurality of image data D5 having different rotation angles may be stored in advance in the storage unit 703 of the computer 70. In that case, the print instruction section 73 may select and read image data D5 having an appropriate rotation angle from the storage section 703 according to the rotation angle of the tablet 9 indicated by the determination result D4. In this way, the processing load on the computer 70 for rotating the image data D5 can be reduced.

The inspection area setting section 74 is a processing section for setting an area to be inspected (hereinafter referred to as "inspection area") of the captured image D1. FIG. 7 is a diagram showing an example of the inspection area of the photographed image D1. In the example of FIG. 7, for the photographed image D1, an inspection area A1 corresponding to the edge of the tablet 9, two semicircular inspection areas A2 and A3 surrounded by the score line and the edge of the tablet 9, and a tablet An inspection area A4 corresponding to the side surface of No. 9 is set. The inspection area setting unit 74 specifies the coordinates of a portion corresponding to the edge of the tablet 9 and a portion corresponding to the dividing line in the reconstructed image D3 output from the autoencoder 71 through image processing. Then, a plurality of inspection areas A1 to A4 in the photographed image D1 are set based on the specified coordinates. That is, although the image to be inspected is the captured image D1, the inspection area setting section 74 sets the inspection areas A1 to A4 in the captured image D1 using the reconstructed image D3.

The inspection unit 75 is a processing unit that inspects the tablet 9 for defects based on the photographed image D1. The photographed image D1 to be inspected may be an image of the tablet 9 before printing acquired by the first imaging unit 20, or an image of the tablet 9 after printing acquired by the second imaging unit 40. Good too. The inspection unit 75 inspects the tablet 9 for defects in each of the plurality of inspection areas A1 to A4 set by the above-described inspection area setting unit 74 in the photographed image D1.

The inspection method, for example, compares the photographed image D1 with a reference image, which is an image of a normal tablet 9, and detects a portion where the difference in pixel values between the two images is larger than a preset threshold as a defect. do. Alternatively, the photographed image D1 and the reconstructed image D3 output from the autoencoder 71 may be compared, and a portion where the difference in pixel values between the two images is larger than a preset threshold may be detected as a defect. .

In this embodiment, a plurality of inspection areas A1 to A4 are set for one captured image D1. Therefore, defects in the tablet 9 can be inspected under different conditions for each of the inspection areas A1 to A4. For example, the threshold value can be relaxed for inspection areas where false detections are likely to occur. Furthermore, the threshold value can be made stricter for inspection areas where false detection is less likely to occur. This allows highly accurate inspection to be performed while suppressing erroneous detection of defects. However, the number of inspection areas set by the inspection area setting section 74 for one captured image D1 may be one.

<3. About machine learning＞
Next, machine learning performed in the autoencoder 71 will be explained. FIG. 8 is a flowchart showing the flow of machine learning. This machine learning is executed in advance before the printing process for the plurality of tablets 9 is executed in the tablet printing apparatus 1.

When performing machine learning, a large number (for example, 100 to 1000) of photographed images D1 of good tablets 9 with no defects are prepared. Then, as shown in FIG. 8, the photographed image D1 of the good tablet 9 is input to the autoencoder 71 (step S1). The autoencoder 71 generates a reconstructed image D3 by performing dimension reduction and reconstruction processing on the input captured image D1 (step S2). When the reconstructed image D3 is output from the autoencoder 71, the computer 70 performs dimension reduction and reconstruction processing based on a machine learning algorithm so that the difference between the photographed image D1 and the reconstructed image D3 becomes small. (step S3).

The computer 70 repeats the processing of steps S1 to S3 described above until a predetermined termination condition is satisfied (step S4: No). The termination condition may be, for example, that the difference between the photographed image D1 and the reconstructed image D3 converges within a preset tolerance range. Further, the termination condition may be that the number of times the learning process in steps S1 to S3 is repeated reaches a preset number of times.

Eventually, when the termination condition is satisfied, the computer 70 terminates the machine learning (step S4: Yes). As a result, a learned model M including the adjusted parameters is obtained. The autoencoder 71 that has obtained the trained model M can convert the photographed image D1 of the tablet 9 into a reconstructed image D3 of the tablet 9 in which random elements such as gloss, roughness, and defects are reduced and output it. becomes.

<4. About printing process>
Next, the printing process executed in the tablet printing apparatus 1 after the above-described machine learning is completed will be described. FIG. 9 is a flowchart showing the flow of print processing. Note that this tablet printing apparatus 1 processes a plurality of tablets 9 while sequentially conveying them, and FIG. 9 shows the flow of processing when focusing on one tablet 9.

When executing the printing process, first, the tablets 9 are carried onto the upper surface of the conveyor belt 12 via the carrying mechanism 80 (step S5). Then, by the rotation of the conveyor belt 12, the tablet 9 is conveyed toward the downstream side in the conveyance direction. When the tablet 9 reaches the first imaging position P1 on the transport path, the first imaging section 20 photographs the tablet 9. Thereby, the first imaging unit 20 obtains a photographed image D1 of the tablet 9 before printing (step S6). The first imaging unit 20 transmits the obtained photographed image D1 to the computer 70.

The computer 70 inputs the photographed image D1 obtained from the first imaging section 20 to the autoencoder 71. The autoencoder 71 performs dimension reduction and reconstruction processing on the input captured image D1 using parameters indicated by the learned model M. As a result, a reconstructed image D3 of the tablet 9 in which random elements such as gloss, roughness, and defects are reduced is generated (step S7). The autoencoder 71 outputs the generated reconstructed image D3 to the determination section 72 and the inspection area setting section 74.

Subsequently, the determining unit 72 determines the shape, position, and rotation angle of the tablet 9 by performing image processing on the reconstructed image D3 (step S8). Specifically, the determining unit 72 extracts a portion corresponding to the edge of the tablet 9 from the reconstructed image D3, and determines the shape and position of the tablet 9 based on the shape and position of the edge. Further, the determination unit 72 extracts a portion corresponding to the score line of the tablet 9 from the reconstructed image D3, and determines the rotation angle of the tablet 9 based on the shape and position of the score line. The determination unit 72 then outputs these determination results D4 to the print instruction unit 73.

In this way, the determination unit 72 determines the shape, position, and rotation angle of the tablet 9 based on the reconstructed image D3 instead of the photographed image D1. Therefore, the shape, position, and rotation angle of the tablet 9 can be determined with high accuracy while suppressing the effects of random factors such as gloss, roughness, and defects.

Subsequently, the print instruction unit 73 adjusts the image data D5 according to the determination result D4 input from the determination unit 72 (step S9). Specifically, the print instruction unit 73 shifts the coordinates of the image data D5 according to the position of the tablet 9 indicated by the determination result D4. Further, the print instruction unit 73 rotates the image data D5 according to the rotation angle of the tablet 9 indicated by the determination result D4. Then, the print instruction section 73 outputs a print instruction D6 to the plurality of heads 31 of the printing section 30 based on the adjusted image data D5.

When the tablet 9 reaches the printing position P2 on the transport path, the plurality of heads 31 discharge ink droplets onto the upper surface of the tablet 9 according to the printing instruction D6 output from the printing instruction section 73. As a result, the image indicated by the image data D5 is printed at an appropriate position on the surface of the tablet 9 at an appropriate rotation angle (step S10).

Thereafter, when the tablet 9 reaches the second imaging position P3 on the transport route, the second imaging section 40 photographs the tablet 9. Thereby, the second imaging unit 40 obtains a photographed image D1 of the printed tablet 9 (step S11). The second imaging unit 40 transmits the obtained photographed image D1 to the computer 70.

Subsequently, the inspection area setting unit 74 of the computer 70 sets inspection areas A1 to A4 for the captured image D1 (step S12). Here, the photographed images D1 to be inspected are both an image of the tablet 9 before printing obtained from the first imaging section 20 and an image of the tablet 9 after printing obtained from the second imaging section 40. The inspection area setting unit 74 sets a plurality of inspection areas A1 to A4 based on the reconstructed image D3 output from the autoencoder 71. Therefore, the influence of random factors such as gloss, roughness, and defects can be suppressed, and a plurality of inspection areas A1 to A4 can be set with high precision. In particular, if the inspection area A1 corresponding to the edge of the tablet 9 is set based on the photographed image D1 itself, it is likely to be misaligned due to the influence of gloss, roughness, chipping, etc. of the tablet 9. However, if the reconstructed image D3 is used, the influence of gloss, roughness, or chipping of the tablet 9 can be suppressed, and the inspection area A1 can be set with high accuracy.

Thereafter, the inspection unit 75 of the computer 70 inspects the tablet 9 for defects in each of the plurality of inspection areas A1 to A4 set in the above-described step S12 in the photographed image D1 (step S13). This determines whether the tablet 9 is a good product with no defects or a defective product with defects. As described above, the plurality of inspection areas A1 to A4 are set with high accuracy based on the reconstructed image D3. Therefore, the inspection unit 75 can accurately inspect the tablet 9 for defects in each region. The inspection unit 75 may inspect the tablet 9 for defects under different conditions for each of the inspection areas A1 to A4.

If the tablet 9 is determined to be a defective product, the computer 70 operates the defective product blow mechanism 51 when the tablet 9 reaches the defective product discharge position P4 on the conveyance path. The defective product blowing mechanism 51 blows gas to the suction hole 14 that holds the tablet 9 determined to be a defective product among the plurality of suction holes 14 of the conveyor belt 12 . Thereby, the tablet 9 determined to be a defective product is discharged from the conveyor belt 12 to the defective product collection section 52 (step S14).

If the tablet 9 is determined to be a non-defective product, the computer 70 operates the non-defective product blowing mechanism 61 when the tablet 9 reaches the non-defective product discharge position P5 on the transport path. The non-defective blow mechanism 61 blows gas toward the plurality of suction holes 14 of the conveyor belt 12 . As a result, the tablets 9 determined to be non-defective are discharged from the conveyor belt 12 to the non-defective product collection unit 62 (step S15).

As described above, in this tablet printing apparatus 1, the autoencoder 71 can generate a reconstructed image D3 in which random elements such as gloss, roughness, and defects are reduced from the photographed image D1. The generated reconstructed image D3 is then used to determine the shape, position, and rotation angle of the tablet 9. Therefore, the shape, position, and rotation angle of the tablet 9 can be determined with high accuracy while suppressing the influence of the random factors described above. Therefore, printing can be performed on the surface of the tablet 9 with high accuracy.

Furthermore, in this tablet printing apparatus 1, the reconstructed image D3 generated by the autoencoder 71 is used to set the inspection areas A1 to A4. Therefore, the influence of the random factors mentioned above can be suppressed, and the inspection areas A1 to A4 can be set with high accuracy. Therefore, defects in the tablet 9 can be inspected with high accuracy.

<5. Modified example>
Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments.

In the above embodiment, the determination unit 72 of the computer 70 determined the shape, position, and rotation angle of the tablet 9. However, the determination unit 72 does not necessarily need to determine all of the shape, position, and rotation angle of the tablet 9. The determining unit 72 may be any unit that determines at least one of the shape, position, and rotation angle of the tablet 9 for printing processing by the printing unit 30.

Further, in the above embodiment, the inspection area setting section 74 of the computer 70 sets a plurality of inspection areas A1 to A4 for the photographed image D1. Then, the inspection section 75 inspected each of these inspection areas A1 to A4. However, among the plurality of inspection areas A1 to A4, there may be an area where no inspection is performed.

Furthermore, the inspection method by the inspection unit 75 is not limited to the comparative inspection exemplified in the above embodiment.

Furthermore, the autoencoder 71 is not limited to a simple autoencoder, but may be an advanced autoencoder such as VAE (Variational Auto Encoder) or UR (UnRegularized Anomaly Score)-VAE.

Further, in addition to the configuration of the above embodiment, the tablet printing device 1 may include a drying mechanism that dries the ink attached to the tablets 9. The drying mechanism is preferably provided downstream of the printing unit 30 in the transport direction and upstream of the defective product discharge unit 50 in the transport direction. The drying mechanism may be, for example, a mechanism that blows hot air toward the tablets 9 conveyed by the conveyor belt 12.

Further, in the above embodiment, the printing section 30 was provided with four heads 31. However, the number of heads 31 included in the printing section 30 may be one to three, or five or more.

Moreover, the "tablet" in the present invention is not limited to a tablet as a medicine, but may also be a tablet as a supplement or a tablet confectionery such as Ramune.

Other details of the tablet printing device and the tablet printing method may be modified as appropriate without departing from the spirit of the present invention. Furthermore, the elements appearing in the above-described embodiments and modifications may be combined as appropriate to the extent that no contradiction occurs.

1 Tablet printing device 9 Tablet 10 Transport mechanism 20 First imaging section 30 Printing section 31 Head 40 Second imaging section 50 Defective product discharge section 60 Good product discharge section 70 Computer 71 Autoencoder 72 Judgment section 73 Print instruction section 74 Inspection area setting section 75 Inspection section 80 Carrying mechanism A1 to A4 Inspection area D1 Photographed image D2 Latent variable D3 Reconstructed image D4 Judgment result D5 Image data D6 Print instruction M Learned model

Claims (8)

A tablet printing device that prints on the surface of a tablet while conveying the tablet,
a conveyance mechanism that conveys the tablet;
an imaging unit that photographs the tablets conveyed by the conveyance mechanism to obtain a photographed image;
an autoencoder that outputs a reconstructed image by performing dimension reduction and reconstruction on the captured image;
a determination unit that determines at least one of the shape, position, and rotation angle of the tablet based on the reconstructed image;
a printing unit that prints on the surface of the tablet based on the determination result of the determination unit;
an inspection area setting unit that sets an inspection area of the tablet based on the reconstructed image;
an inspection unit that inspects the tablet for defects in the inspection area set by the inspection area setting unit in the photographed image;
Equipped with
The autoencoder performs the dimension reduction and the reconstruction using parameters adjusted in advance by machine learning so that the difference between the input tablet image and the output tablet image is small. Tablet printing equipment. The tablet printing device according to claim 1,
In the tablet printing apparatus, the determination unit determines a rotation angle of the tablet based on a score line of the tablet included in the reconstructed image. The tablet printing device according to claim 1 or 2 ,
The inspection area setting unit sets the plurality of inspection areas based on the reconstructed image,
In the tablet printing apparatus, the inspection unit inspects the tablet for defects using a different threshold value for each inspection area. The tablet printing device according to claim 3 ,
A tablet printing device, wherein the plurality of inspection areas include an area corresponding to an edge of a tablet. A tablet printing method that prints on the surface of a tablet while conveying the tablet,
a) Input a tablet image into an autoencoder that performs dimension reduction and reconstruction, and set the parameters of the autoencoder so that the difference between the input tablet image and the output reconstructed tablet image is small. a process of adjusting the
b) a step of photographing the tablet while conveying the tablet to obtain a photographed image;
c) inputting the captured image to the autoencoder whose parameters have been adjusted in step a) to obtain a reconstructed image output from the autoencoder;
d) determining at least one of the shape, position, and rotation angle of the tablet based on the reconstructed image;
e) a step of printing on the surface of the tablet based on the determination result of step d);
f) setting an inspection area of the tablet based on the reconstructed image;
g) inspecting the tablet for defects in the inspection area set in step f) of the photographed image;
A tablet printing method comprising: 6. The tablet printing method according to claim 5 ,
In the tablet printing method, in step d), the rotation angle of the tablet is determined based on the score line of the tablet included in the reconstructed image. The tablet printing method according to claim 5 or 6 ,
In the step f), setting a plurality of the inspection areas based on the reconstructed image,
In the step g), a tablet printing method includes inspecting the tablet for defects using a different threshold value for each inspection area. The tablet printing method according to claim 7 ,
The tablet printing method, wherein the plurality of inspection areas include areas corresponding to edges of the tablet.

Images