Installation surface inspection device

Abstract

Provided is an installation surface inspection device that can more appropriately check for the presence or absence of fallen objects on the installation surface of a device. [Solution] The installation surface inspection device 50 includes a horizontal support part 533 that extends horizontally and at least a portion of which is installed below the PTP packaging machine 10 as an apparatus, and a camera 531 that is attached to the horizontal support part 533 and can capture images of the installation surface. At least the entire surfaces of the horizontal support part 533 and the camera 531 that are visible when viewed from above are sloped parts 533a, 531c that slope downward toward a predetermined outer edge. This makes it difficult for fallen objects Rb to land on the horizontal support part 533, etc., and allows the camera 531 to more reliably capture images of the fallen objects Rb. As a result, it becomes possible to more appropriately check whether or not there are fallen objects Rb on the installation surface.

Description

[Technical Field]

[0001] The present invention relates to an installation surface inspection device for inspecting whether or not there is any fallen object on an installation surface of an apparatus. [Background technology]

[0002] There are widely known devices that have the function of transporting various items, such as manufacturing devices, conveying devices, packaging devices, etc. For example, a packaging device (blister packaging machine) for producing blister sheets transports tablets or capsules as items and packages the tablets or capsules to produce blister sheets.

[0003] In the above-described device, fallen objects (falling objects) may get into the installation surface (surface located below the device). If the objects are pharmaceuticals or the like, the objects must be managed extremely strictly to prevent the mixing of different types of tablets. Therefore, it is necessary to check for the presence or absence of fallen objects on the installation surface.

[0004] However, the task of a worker visually checking the installation surface to see if there are any fallen objects requires effort and time, and there is a risk that the presence or absence of fallen objects cannot be properly confirmed by simply visually checking the installation surface. Therefore, it is conceivable to use an inspection module that includes a horizontal support section that extends parallel to the installation surface, a vertical support section that extends vertically downward from the horizontal support section and whose underside contacts the installation surface, and a camera that is attached to the horizontal support section and can capture images of the installation surface (see, for example, Patent Document 1). By installing this inspection module below the device, it becomes possible to check for the presence or absence of fallen objects based on images obtained by the camera. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2020-94897 Summary of the Invention [Problem to be solved by the invention]

[0006] However, when using the above-described inspection module, there is a risk that a fallen object may land on the horizontal support part or the camera by bouncing off the installation surface, etc. A fallen object that falls on the horizontal support part or the like will be located outside the imaging range of the camera, and as a result, there is a risk that the fallen object will be overlooked or that it will take a long time to find the fallen object.

[0007] The present invention has been made in consideration of the above circumstances, and its purpose is to provide an installation surface inspection device that can more appropriately check for the presence or absence of fallen objects on the installation surface of the device. [Means for solving the problem]

[0008] The following describes each of the means suitable for achieving the above object, with specific effects of the corresponding means added as necessary.

[0009] Means 1. An installation surface inspection device for inspecting the presence or absence of a dropped object, which is a dropped item, on the installation surface of a device having a function of transporting a predetermined item, a horizontal support extending horizontally and at least a portion of which is positioned below the device; a camera attached to the horizontal support portion and capable of capturing an image of the installation surface; An installation surface inspection device characterized in that at least the entire surface of the horizontal support portion and the camera that is visible when viewed from above is an inclined portion that slopes downward toward a specified outer edge.

[0010] According to the above-mentioned method 1, the entire surface of the horizontal support part and the camera that can be seen from above is a sloped surface that slopes downward toward a predetermined outer edge. Therefore, even if a fallen object bounces off the installation surface and ends up on the horizontal support part or the camera, the fallen object is likely to slide down the sloped surface and fall to the installation surface. This makes it difficult for the fallen object to land on the horizontal support part, etc., and ultimately prevents the fallen object from remaining outside the camera's imaging range more reliably. This allows the camera to more reliably capture the fallen object, and ultimately allows for more appropriate confirmation of the presence or absence of the fallen object on the installation surface.

[0011] The camera may include a camera body capable of capturing an image of the installation surface and a cover that surrounds the camera body and has the same shape as the side of a cone or pyramid, or a truncated cone or pyramid, when viewed from above. In this case, the cover has a sloped surface. The gap between the camera body and the cover may be large enough to prevent falling objects from entering, or may be closed.

[0012] Means 2. An installation surface inspection device as described in Means 1, characterized in that the horizontal support part is attached to the device, so that the camera and the entire parts that support the camera are floating above the installation surface.

[0013] According to the above-mentioned method 2, by attaching the horizontal support part to the device, the camera and the parts supporting the camera are all suspended above the installation surface. Therefore, compared to when the legs supporting the horizontal support part are erected on the installation surface, blind spots in the camera's imaging range can be reduced. This makes it possible to more appropriately check for fallen objects on the installation surface.

[0014] Means 3. The horizontal support part is attached to the device only at its base end, The installation surface inspection device described in means 2 is characterized in that the camera is attached to the side of the tip of the horizontal support part.

[0015] According to the above-mentioned means 3, the horizontal support part has a cantilever structure in which only the base end is attached to the device, and the camera is attached to the side of the tip end of the horizontal support part. Therefore, the camera can be placed higher than when the camera is attached to the underside of the horizontal support part. This allows the camera's imaging range to be wider, and ultimately makes it possible to more appropriately check for fallen objects on the installation surface.

[0016] Means 4. A mounting portion is provided between the horizontal support portion and the device, for mounting the horizontal support portion to the device; An installation surface inspection device as described in means 2, characterized in that the entire surface of the mounting portion that can be seen when viewed from above is a mounting portion side slope portion that slopes downward toward a specified outer edge side.

[0017] According to the above-mentioned feature 4, the entire surface of the mounting part that can be seen from above is an inclined mounting part side that slopes downward toward the predetermined outer edge. This makes it less likely that a fallen object will land on the mounting part, making it impossible for the camera to capture the fallen object. This makes it possible to more appropriately check whether or not there is a fallen object on the installation surface.

[0018] Means 5: An installation surface inspection device according to Means 1, characterized in that the acute angle of the angle formed by the inclined surface with respect to the horizontal plane is 50° or more.

[0019] According to the above-mentioned means 5, the acute angle (inclination angle) of the inclined surface with respect to the horizontal plane is set to 50° or more. Therefore, it is possible to more reliably prevent fallen objects from landing on the horizontal support part, etc.

[0020] It is preferable that the inclination angle of the mounting portion side inclined surface described above and the light blocking plate side inclined surface described later is 50° or more.

[0021] Means 6. The horizontal support portion has a shape in which two of the inclined portions, the inclination directions of which are opposite to each other, are connected at the top, The installation surface inspection device according to means 1, wherein the top portion has a curved surface that is convex upward.

[0022] According to the above-mentioned means 6, the horizontal support portion has a shape in which two inclined portions with opposite inclination directions are connected at the top, and this top is curved rather than angular. Therefore, it is possible to more reliably prevent a fallen object from getting caught on the top and landing on the horizontal support portion. This makes it possible to more appropriately check whether or not there is a fallen object on the installation surface of the device. It is also possible to more reliably prevent a fallen object from being damaged by contact with the top.

[0023] It should be noted that the radius of curvature of the apex is preferably as small as possible to prevent falling objects from getting caught on the apex. Therefore, it is preferable that the radius of curvature of the apex be 1 mm or less. However, if the radius of curvature of the apex is too small, the shape of the apex will be almost the same as an angular shape. Therefore, it is preferable that the radius of curvature of the apex be 0.5 mm or more. Means 7. An illumination means located below the camera and illuminating a predetermined light at least to the imaging range of the camera on the installation surface; a light shielding plate that covers the illumination means from above and prevents light emitted from the illumination means from directly entering the camera, An installation surface inspection device as described in means 1, characterized in that the entire surface of the shading plate that is visible when viewed from above is a shading plate side slope portion that slopes downward toward a specified outer edge side.

[0024] According to the above-mentioned means 7, the illumination means can illuminate at least the imaging range of the camera on the installation surface. Therefore, the camera can more clearly image fallen objects on the installation surface. Furthermore, when visually checking the installation surface, fallen objects can be more easily found.

[0025] Furthermore, the light shielding plate can prevent the light emitted from the illumination means from directly entering the camera, which can more reliably prevent problems such as halation from occurring in the image data obtained by the camera.

[0026] In addition, the entire surface of the light shielding plate that can be seen from above is a light shielding plate-side slope that slopes downward toward the specified outer edge. This reduces the likelihood of a fallen object landing on the light shielding plate, preventing the camera from capturing an image of the object. This allows for more accurate confirmation of whether or not an object has fallen on the installation surface.

[0027] Means 8. A classification means for generating a neural network having an encoding unit that extracts features from input image data and a decoding unit that reconstructs image data from the features, by training only image data relating to the installation surface without the fallen object as training data; a reconstructed image data acquisition means for inputting the inspection image data obtained by the camera to the identification means and acquiring reconstructed image data as reconstructed image data; a comparison means for comparing the inspection image data and the reconstructed image data; The installation surface inspection device according to Means 1 is characterized in that it comprises a determination means capable of determining the presence or absence of the fallen object on the installation surface based on the comparison result by the comparison means.

[0028] The learning data may be image data (actual image data) obtained by actually capturing an image of an installation surface without any fallen objects, or may be a virtually generated image (virtual image data) of an installation surface without any fallen objects. Furthermore, various types of image data with different factors that may affect the state of the installation surface, such as image data under different ambient light conditions (for example, image data obtained during the day or at night, or image data obtained on a sunny or rainy day), image data when an operator is working, or image data when the device is operating or stopped, may be used as the learning data.

[0029] Furthermore, the "neural network" includes, for example, a convolutional neural network having multiple convolutional layers. The "learning" includes, for example, deep learning. The "identification means (generative model)" includes, for example, an autoencoder and a convolutional autoencoder.

[0030] In addition, the "identification means" is generated by learning only image data relating to the installation surface without any fallen objects. Therefore, when the inspection image data relating to the installation surface with fallen objects is input to the identification means, the reconstructed image data generated will be nearly identical to the inspection image data from which the fallen objects have been removed.

[0031] According to the above-mentioned means 8, the test image data is compared with the reconstructed image data that is reconstructed by inputting the test image data into the identification means, and the presence or absence of a fallen object on the installation surface is determined based on the comparison result. Therefore, the presence or absence of a fallen object can be confirmed more accurately and quickly than when the presence or absence of a fallen object is determined by visually checking the image data.

[0032] Furthermore, the conditions of the installation surface (for example, ambient light) can be matched between the two sets of image data to be compared, which allows for more accurate confirmation of the presence or absence of fallen objects on the installation surface.

[0033] Means 9: The installation surface inspection device according to means 1, characterized in that the inclined surface portion has a buffering means capable of absorbing the impact applied to the falling object when it comes into contact with the falling object.

[0034] According to the above-mentioned means 9, the buffering means of the sloped surface can buffer the impact on the falling object. Therefore, it is possible to more reliably prevent the falling object from moving significantly (e.g., bouncing) due to the impact from the sloped surface. This allows the camera to more reliably capture an image of the falling object, and more appropriately check whether or not there is a falling object on the installation surface.

[0035] Furthermore, by providing a buffer means, it is possible to more reliably prevent the falling object from breaking due to impact, making it easier to find the fallen object in its original shape. Therefore, it is easier to recover the fallen object. Preventing the falling object from breaking is particularly effective when the item (falling object) contains special medicinal ingredients.

[0036] The buffering means may be made of a material capable of absorbing impact, such as soft rubber, felt, urethane, elastic material, etc. The buffering means may also be applied to the above-mentioned mounting portion side slope portion or the light blocking plate side slope portion.

[0037] Furthermore, the technical features relating to the above-mentioned means may be combined as appropriate. For example, the technical feature relating to the above-mentioned means 3 may be combined with at least one of the technical features relating to the above-mentioned means 4 to 9. [Brief explanation of the drawings]

[0038] [Figure 1] FIG. 1 is a perspective view of a PTP sheet. [Figure 2] FIG. 1 is a partially enlarged cross-sectional view of a PTP sheet. [Figure 3] FIG. 1 is a perspective view of a PTP film. [Figure 4] FIG. 1 is a schematic diagram of a PTP packaging machine. [Figure 5] FIG. 2 is a schematic perspective view showing an inspection unit and the like when viewed obliquely from above. [Figure 6] FIG. 2 is a schematic perspective view showing an inspection unit and the like when viewed obliquely from below. [Figure 7] FIG. 6 is a cross-sectional view taken along line JJ in FIG. 5. [Figure 8] FIG. 6 is a cross-sectional view taken along the line KK in FIG. 5. [Figure 9] FIG. [Figure 10] FIG. 6 is a cross-sectional view taken along line LL in FIG. 5. [Figure 11] FIG. 2 is a block diagram showing the functional configuration of the installation surface inspection device. [Figure 12] FIG. 1 is a schematic diagram for explaining the structure of a neural network. [Figure 13] 10 is a flowchart showing the flow of a learning process of a neural network. [Figure 14] 10 is a flowchart showing the flow of a determination process. [Figure 15] 10 is a flowchart showing the flow of a determination process. [Figure 16] 10 is a flowchart showing the flow of a determination process. [Figure 17] FIG. 10 is a cross-sectional schematic view of a horizontal support having a buffer portion in another embodiment. [Figure 18] FIG. 10 is a cross-sectional view showing a horizontal support part, both ends of which are attached to a PTP packaging machine, in another embodiment. [Figure 19] FIG. 10 is a perspective schematic view showing a horizontal support part, both ends of which are attached to a PTP packaging machine, in another embodiment. [Figure 20] FIG. 10 is a schematic perspective view showing a cover according to another embodiment. [Figure 21] FIG. 10 is a schematic plan view showing a cover according to another embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0039] An embodiment will be described below with reference to the drawings. First, a PTP packaging machine will be described as an "apparatus" to which the installation surface inspection device is applied. The PTP packaging machine is a packaging device for producing PTP sheets 1 (see FIGS. 1 and 2) that have a container film 3 with multiple pockets 2 and a cover film 4 attached to the container film 3 so as to cover the pockets 2. The pockets 2 are filled with tablets 5 as "articles." The tablets 5 are circular in plan view and relatively easy to roll.

[0040] 4, a raw roll 11 formed by winding a strip-shaped container film 3 is provided at the most upstream side of the PTP packaging machine 10, and the container film 3 sent out from here is transported along a predetermined path. The PTP packaging machine 10 is equipped with, in order from upstream along the transport path of the container film 3, a heating device 15, a pocket forming device 16, a filling device 21, an inspection device 22, a sealing device 25, a slit forming device 33, a marking device 34, and a sheet punching device 37.

[0041] The heating device 15 preheats the container film 3. The pocket forming device 16 forms a pocket 2 in the container film 3 heated by the heating device 15.

[0042] The filling device 21 fills the tablets 5 into the pockets 2 of the continuously conveyed container film 3. The inspection device 22 inspects, for example, whether the tablets 5 are securely filled in each pocket 2, whether there are any abnormalities in the tablets 5, and whether there is any foreign matter mixed into the pockets 2.

[0043] The sealing device 25 includes a heating roll 25a having a heating function and a film feed roll 25b that continuously transports the container film 3, and attaches the cover film 4 to the container film 3 so as to close the pocket portion 2. More specifically, the strip-shaped cover film 4 is guided toward the heating roll 25a, and the container film 3 and the cover film 4 pass between the two rolls 25a and 25b in a heated and pressure-welded state, thereby attaching the cover film 4 to the container film 3. By attaching the cover film 4 to the container film 3, a strip-shaped PTP film 6 (see FIG. 3) is produced in which tablets 5 are housed in each pocket portion 2.

[0044] The slit forming device 33 forms a separation slit at a predetermined position on the PTP film 6. The marking device 34 marks a mark at a predetermined position (for example, a tag portion) on the PTP film 6. Note that the separation slit and marking are not shown in FIG.

[0045] The sheet punching device 37 has the function of punching the outer edge of the PTP film 6 into PTP sheets 1, that is, separating the PTP sheets 1 from the PTP film 6. The PTP sheets 1 obtained by the sheet punching device 37 are transported by a conveyor 39 and stored in a finished product hopper 40. Meanwhile, scraps 42 remaining in the PTP film 6 after punching are cut to a predetermined size by a cutting device 41 and stored in a scrap hopper 43.

[0046] In this way, in the PTP packaging machine 10, the tablets 5 filled in the pockets 2 are transported together with the container film 3, and the tablets 5 are packaged with both films 3, 4, thereby producing the PTP sheet 1. In other words, the PTP packaging machine 10 has the function of transporting the tablets 5.

[0047] In addition, in the PTP packaging machine 10, the conveying mode of the container film 3 and PTP film 6 is changed appropriately to continuous conveying or intermittent conveying to accommodate the processes performed by the various devices described above. To prevent slack in the container film 3, etc. due to such changes in the conveying mode, tension rolls are provided at predetermined positions in the conveying path of the container film 3, etc., to maintain the tension of the container film 3, etc. within a certain range.

[0048] Next, with reference to Fig. 5 etc., an installation surface inspection device 50 for inspecting the presence or absence of fallen objects Rb, which are dropped tablets 5, on the installation surface 100 of the PTP packaging machine 10 will be described. As shown in Figs. 5 to 11, the installation surface inspection device 50 includes an inspection unit 51 and a control device 52. Note that Fig. 5 etc. shows only the lower part of the PTP packaging machine 10 in a simplified manner. The PTP packaging machine 10 is supported by a plurality of legs 10k.

[0049] First, the inspection unit 51 will be described. The inspection unit 51 is a device for obtaining image data for inspection, and a plurality of inspection units 51 (four in this embodiment) are provided at equal intervals along the longitudinal direction of the installation surface 100 so as to cover almost the entire area of ​​the installation surface 100. Each inspection unit 51 includes one camera unit 53 and a plurality of lighting units 54 (four in this embodiment).

[0050] The camera unit 53 is a unit in which a camera 531, an attachment part 532 which is a part for attaching the camera 531 to the PTP packaging machine 10, and a horizontal support part 533 are integrated.

[0051] Camera 531 is a device for capturing an image from directly above installation surface 100. Camera 531 is attached to the side of the tip of horizontal support part 533, and includes camera body 531a and cover 531b (see FIGS. 6 and 8).

[0052] The camera body 531a has sensitivity to the wavelength region of light irradiated from a lighting device 542 (described later) of the lighting unit 54, and is configured by, for example, a CCD camera or a CMOS camera. The camera 531 captures two-dimensional images of light reflected from the installation surface 100 and the falling object Rb when the light irradiated from the lighting device 542 illuminates the installation surface 100. Image data (e.g., luminance image data) captured by the camera 531 is converted into a digital signal (image signal) inside the camera 531 and then input in the form of a digital signal to the control device 52. In this embodiment, the multiple cameras 531 can capture images of almost the entire installation surface 100.

[0053] The camera 531 may be a 3D camera (for example, a TOF (Time-of-Flight) camera) capable of acquiring distance information to the installation surface 100 and the falling object Rb. The camera 531 may also be one capable of acquiring image data including height information.

[0054] Cover 531b covers the periphery of camera body 531a, and the entire surface visible when viewed from above has the same shape as the side of a cone. As a result, the entire surface of the camera visible when viewed from above is inclined downward toward a predetermined outer edge (the outer edge onto which anything that reaches it will fall), forming camera-side slope 531c configured to prevent falling object Rb from landing on it. The acute angle α1 (see FIG. 8) of camera-side slope 531c relative to the horizontal plane is 50° or greater.

[0055] In this embodiment, the size of gap 531d (see FIG. 6) formed between camera body 531a and cover 531b and opening downward is set to a size that prevents falling object Rb from entering gap 531d. Note that gap 531d may be blocked to prevent falling object Rb from entering gap 531d.

[0056] The mounting portion 532 constitutes the mounting portion of the camera unit 53 relative to the PTP packaging machine 10 (for example, the outer frame of the PTP packaging machine 10), and is shaped like a plate extending in the vertical direction. The entire surface of the mounting portion 532 visible when viewed from above is inclined downward toward a predetermined outer edge (the outer edge onto which an object will fall once it has reached it), and is formed as a mounting portion-side slope 532a configured to prevent a falling object Rb from landing on it. The acute angle α2 (see FIG. 8) of the angle formed by the mounting portion-side slope 532a relative to the horizontal plane is 50° or greater.

[0057] The horizontal support part 533 is attached to the PTP packaging machine 10 via an attachment part 532, and plays a role of supporting the camera 531 below the PTP packaging machine 10 together with the attachment part 532. The horizontal support part 533 has a rod shape extending horizontally, and at least a part of it (in this embodiment, most of it excluding the base end) is installed below the PTP packaging machine 10.

[0058] The horizontal support part 533 has a cantilever structure in which only the base end part is attached to the PTP packaging machine 10 via the attachment part 532. By attaching the horizontal support part 533 to the PTP packaging machine 10 via the attachment part 532, the camera 531 and the attachment part 532 and horizontal support part 533 (i.e., the camera unit 53), which are parts that support the camera 531, are raised above the installation surface 100.

[0059] Furthermore, the entire surface of horizontal support part 533 that can be seen when viewed from above is configured as support part-side slope part 533a that slopes downward toward a predetermined outer edge (the outer edge onto which anything that reaches it will fall) and is configured to prevent falling object Rb from landing on it. As a result, at least the entire surfaces of horizontal support part 533 and camera 531 that can be seen when viewed from above are configured as slope parts 533a, 531c that slope downward toward the predetermined outer edge. In this embodiment, camera side slope part 531c and support part side slope part 533a each correspond to a "slope part."

[0060] It should be noted that the phrase "surfaces of horizontal support part 533 and camera 531 visible when viewed from above" can be interpreted as surfaces visible when horizontal support part 533 and camera 531 are viewed together from above. Therefore, for example, when camera 531 is attached to the lower surface of horizontal support part 533, the surface of camera 531 covered by horizontal support part 533 does not necessarily have to be an inclined surface.

[0061] Moreover, the acute angle α3 (see FIG. 7) of the angle formed by the support-side inclined surface 533a with respect to the horizontal plane is set to 50° or more.

[0062] In addition, the horizontal support portions 533 of the two inspection units 51 that correspond to the extreme ends of the installation surface 100 among the multiple inspection units 51 have a cross-sectional right-angled triangular shape with only one support-side slope portion 533a (see Figure 7).

[0063] On the other hand, the horizontal support portion 533 of the other inspection units 51 has a cross section of an isosceles triangle in which two support-side inclined portions 533a, which are inclined in opposite directions, are connected at an apex 533b (see FIGS. 7 and 9). The apex 533b has an upwardly convex curved surface, and its radius of curvature r is, for example, 0.5 mm or more and 1 mm or less.

[0064] Furthermore, the support part side inclined surface 533a of each inspection unit 51 has an inclined surface shape that gradually descends toward the imaging range of the camera 531 of the inspection unit 51 arranged adjacent to the inspection unit 51. Therefore, the support part side inclined surface 533a of each inspection unit 51 has the function of guiding the tablet 5 that comes into contact with it to the imaging range of a camera 531 other than the camera 531 associated with it.

[0065] The lighting unit 54 is a unit that integrates a lighting attachment portion 541, a lighting device 542, and a light blocking plate 543. In this embodiment, the lighting device 542 constitutes the "illumination means."

[0066] The lighting attachment portion 541 constitutes the attachment portion of the lighting unit 54 to the PTP packaging machine 10 (for example, the outer frame of the PTP packaging machine 10), and is shaped like a plate extending in the vertical direction. The entire surface of the lighting attachment portion 541 seen when viewed from above is inclined downward toward a predetermined outer edge (the outer edge onto which objects that reach it will fall), and is formed as a lighting-side slope portion 541a configured to prevent falling objects Rb from landing on it. The acute angle α4 (see FIG. 10) of the lighting-side slope portion 541a relative to the horizontal plane is 50° or greater.

[0067] Furthermore, by attaching the lighting device 542 and the light blocking plate 543 to the PTP packaging machine 10 via the lighting attachment part 541, the lighting attachment part 541, the lighting device 542, and the light blocking plate 543 (i.e., the lighting unit 54) are raised above the installation surface 100. As described above, in combination with the camera unit 53 being raised above the installation surface 100, the entire inspection unit 51 is raised above the installation surface 100.

[0068] The lighting device 542 irradiates predetermined light (for example, infrared light) from diagonally above at least the imaging range of the camera 531 (camera body 531a) on the installation surface 100. In this embodiment, the lighting device 542 is provided corresponding to one end side in the width direction of the PTP packaging machine 10, and is configured to irradiate predetermined light onto the installation surface 100 from one direction.

[0069] The light shielding plate 543 covers the lighting device 542 from above to prevent light emitted from the lighting device 542 from directly entering the camera 531. The entire surface of the light shielding plate 543 visible from above is inclined downward toward a predetermined outer edge (the outer edge onto which objects will fall once they reach it) to form a light shielding plate-side slope 543a configured to prevent falling objects Rb from landing on it. The acute angle α5 (see FIG. 7) of the light shielding plate-side slope 543a relative to the horizontal plane is 50° or greater. Note that it is preferable that the angles α1, α2, α3, α4, and α5 have larger values. Therefore, it is more preferable that the angles α1, α2, α3, α4, and α5 are each 60° or greater.

[0070] Next, the control device 52 will be described. The control device 52 is configured by a computer system including a CPU that executes predetermined arithmetic processing, a ROM that stores various programs and fixed value data, etc., a RAM that temporarily stores various data when executing various arithmetic processing, and peripheral circuits for these. As shown in FIG. 11 , the control device 52 functions as various functional units such as a main control unit 52a, an illumination control unit 52b, a camera control unit 52c, an image capture unit 52d, a reconstructed image data acquisition unit 52e, a learning unit 52f, and a determination unit 52g, as the CPU operates in accordance with various programs. In this embodiment, the reconstructed image data acquisition unit 52e constitutes a "reconstructed image data acquisition means," and the determination unit 52g constitutes a "comparison means" and a "determination means."

[0071] However, the various functional units are realized by the cooperation of various hardware such as the CPU, ROM, RAM, etc., and there is no need to clearly distinguish between functions realized by hardware and functions realized by software, and some or all of these functions may be realized by hardware circuits such as ICs.

[0072] The control device 52 further includes an input unit 521 including a keyboard, mouse, touch panel, etc., a display unit 522 having a display screen such as a liquid crystal display, a communication unit 526 capable of transmitting and receiving various data to and from the outside, etc. The control device 52 also includes an examination information storage unit 523, an AI storage unit 524, and an image data storage unit 525, each of which is configured using a hard disk drive (HDD), a solid state drive (SSD), etc.

[0073] First, before describing the various functional units that configure the control device 52, the input unit 521, the display unit 522, the storage units 523 to 525, and the communication unit 526 will be described.

[0074] The input unit 521 is an input means for inputting information to the control device 52. The input unit 521 can be used to change various setting information stored in the control device 52.

[0075] Display unit 522 is configured to be able to display, for example, various types of information stored in each of storage units 523 to 525. Therefore, display unit 522 can display image data obtained by camera 531, determination results regarding the presence or absence of fallen object Rb on installation surface 100, and the like.

[0076] The inspection information storage unit 523 stores the determination result regarding the presence or absence of a fallen object Rb on the installation surface 100, information indicating the position of the fallen object Rb, and the like.

[0077] The AI ​​storage unit 524 stores an AI (Artificial Intelligence) model 200 as an "identification means."

[0078] The image data storage unit 525 stores various image data such as image data obtained by the camera 531 .

[0079] The communication unit 526 is equipped with a wireless communication interface conforming to communication standards such as a wired LAN (Local Area Network) or a wireless LAN, and is configured to be able to transmit and receive various data to and from the outside. For example, the determination result performed by the determination unit 52g can be output to the outside (such as the PTP packaging machine 10) via the communication unit 526.

[0080] Next, the above-mentioned various functional units that constitute the control device 52 will be described in detail.

[0081] The main control unit 52a is a functional unit that controls the entire installation surface inspection device 50, and is configured to be able to send and receive various signals to and from other functional units such as the lighting control unit 52b and the camera control unit 52c.

[0082] The lighting control unit 52b is a functional unit that controls the driving of the lighting device 542, and controls the lighting timing and the like based on a command signal from the main control unit 52a.

[0083] The camera control unit 52c is a functional unit that controls the driving of the camera 531, and controls the timing of image capture and the like based on a command signal from the main control unit 52a.

[0084] In this embodiment, the illumination control unit 52b and the camera control unit 52c execute a process of irradiating the installation surface 100 with light from the illumination device 542 and capturing an image of the light reflected from the installation surface 100 and the like by the camera 531 every time a certain period of time has elapsed. Therefore, a plurality of sets of test image data, which are image data obtained by the camera 531, are acquired within a predetermined period of time. These sets of test image data are sent to the image acquisition unit 52d.

[0085] The image capturing unit 52d is a functional unit for capturing image data captured and acquired by the camera 51. The captured image data is stored in the image data storage unit 525.

[0086] The learning unit 52f is a functional unit that uses learning data to learn a deep neural network 190 (hereinafter simply referred to as "neural network 190"; see FIG. 12) and constructs an AI model 200.

[0087] As will be described later, the AI ​​model 200 in this embodiment is a generative model constructed by deep learning the neural network 190 using only image data relating to the installation surface 100 without any falling objects Rb as training data, and has the structure of a so-called autoencoder.

[0088] Here, the structure of the neural network 190 will be described with reference to Fig. 12. Fig. 12 is a schematic diagram conceptually showing the structure of the neural network 190. As shown in Fig. 12, the neural network 190 has a convolutional auto-encoder (CAE) structure that includes an encoder unit 191 as an "encoding unit" that extracts a feature (latent variable) TA from input image data GA, and a decoder unit 192 as a "decoding unit" that reconstructs image data GB from the feature TA.

[0089] The structure of a convolutional autoencoder is well known, and therefore a detailed description will be omitted. However, the encoder unit 191 has a plurality of convolution layers 193, and in each convolution layer 193, a result of a convolution operation performed on input data using a plurality of filters (kernels) 194 is output as input data for the next layer. Similarly, the decoder unit 192 has a plurality of deconvolution layers 195, and in each deconvolution layer 195, a result of a deconvolution operation performed on input data using a plurality of filters (kernels) 196 is output as input data for the next layer. Then, in the learning process described below, the weights (parameters) of each filter 194, 196 are updated.

[0090] Here, the learning process performed by the learning unit 52f when generating the AI ​​model 200 will be described. First, prior to the learning process, a large number of image data relating to the installation surface 100 without any fallen objects Rb are prepared as learning data. In this embodiment, various types of image data are prepared as learning data, which are image data actually obtained by the camera 531 capturing an image of the installation surface 100, and which vary in factors that may affect the state of the installation surface 100, such as image data under different ambient light conditions (for example, image data obtained during the day or at night, or image data obtained on a sunny or rainy day), image data when an operator is working, and image data when the PTP packaging machine 10 is operating or stopped.

[0091] Note that the learning data may be a virtually generated image (virtual image data) of the installation surface 100 without any fallen object, rather than image data obtained by actually capturing an image of the installation surface 100. The learning data may also be data with a reduced amount of data by masking structural parts of the PTP packaging machine 10 (for example, the legs 10k) and parts that do not need to be checked for the presence or absence of a fallen object Rb.

[0092] 13, in the learning process, first, in step S201, an untrained neural network 190 is prepared. For example, the neural network 190 is read out from a predetermined storage device or the like. Alternatively, the neural network 190 is constructed based on network configuration information (e.g., the number of layers of the neural network, the number of nodes in each layer, etc.) stored in the storage device or the like.

[0093] Next, in step S202, reconstructed image data is acquired. That is, previously prepared learning data is provided as input data to the input layer of the neural network 190. Then, reconstructed image data output from the output layer of the neural network 190 is acquired.

[0094] In the following step S203, the learning data is compared with the reconstructed image data output by the neural network 190 in step S202, and it is determined whether the error is sufficiently small (whether it is equal to or smaller than a predetermined threshold).

[0095] If the error is sufficiently small, it is determined in step S205 whether or not the termination condition for the learning process is met. For example, if a positive determination is made in step S203 a predetermined number of times in succession without going through the process of step S204 (described later), or if learning using all of the prepared learning data is repeated a predetermined number of times, it is determined that the termination condition is met. If the termination condition is met, the neural network 190 and its learning information (such as updated parameters (described later)) are stored in the AI ​​storage unit 524 as the AI ​​model 200, and the learning process is terminated.

[0096] On the other hand, if the termination condition is not met in step S205, the process returns to step S202, and the neural network 190 is trained again.

[0097] If the error is not sufficiently small in step S203, the network update process (learning of the neural network 190) is performed in step S204, and then the process returns to step S202 to repeat the above series of processes.

[0098] In the network update process of step S204, a known learning algorithm such as backpropagation is used to update the weights (parameters) of the filters 194, 196 in the neural network 190 to more appropriate ones so that a loss function representing the difference between the training data and the reconstructed image data is minimized. Note that, for example, BCE (Binary Cross-entropy) can be used as the loss function.

[0099] By repeating the processes of steps S202 to S204 many times, the error between the training data and the reconstructed image data in the neural network 190 is minimized, and more accurate reconstructed image data is output.

[0100] The finally obtained AI model 200 is image data (inspection image data) obtained by the camera 531, and when image data relating to the installation surface 100 without a falling object Rb is input, the AI ​​model 200 generates reconstructed image data that substantially matches the input image data. On the other hand, when image data (inspection image data) obtained by the camera 531 relating to the installation surface 100 with a falling object Rb is input, the AI ​​model 200 generates reconstructed image data that substantially matches the input image data, with noise portions (portions corresponding to the falling object Rb) removed. In other words, when a falling object Rb is present on the installation surface 100, virtual image data relating to the installation surface 100 assuming that there is no falling object Rb is generated as reconstructed image data relating to the installation surface 100.

[0101] Furthermore, when image data (inspection image data) obtained by the camera 531 relating to the installation surface 100 containing foreign matter other than the fallen object Rb (for example, film fragments or dust) is input, the AI ​​model 200 generates reconstructed image data that is approximately identical to the input image data, with noise portions (portions corresponding to the foreign matter) removed.

[0102] It is not necessary to perform the above learning process each time the control device 52 is manufactured. The neural network 190 and its learning information (updated parameters, etc.) may be acquired in advance and stored as the AI ​​model 200 in the AI ​​memory unit 524 of the control device 52.

[0103] The reconstructed image data acquisition unit 52e inputs the image data acquired by the camera 531 into the AI ​​model 200 and acquires the reconstructed image data as the reconstructed image data. More specifically, the reconstructed image data acquisition unit 52e provides the inspection image data acquired by the camera 531 as input data to the input layer of the neural network 190. Then, the reconstructed image data acquisition unit 52e acquires the image data output from the output layer of the neural network 190 as the reconstructed image data.

[0104] The determination unit 52g is a functional unit that performs a determination process as to whether or not a fallen object Rb exists on the installation surface 100. The determination unit 52g compares the inspection image data obtained by the camera 531 with reconstructed image data that is reconstructed by inputting the inspection image data into the AI ​​model 200, and determines whether or not a fallen object Rb exists based on the comparison result.

[0105] The determination process performed by the determination unit 52g will be described in more detail with reference to Figures 14 to 16. The determination process is performed when a predetermined number or more of test image data are newly acquired.

[0106] In the determination process, first, in step S301, the reconstructed image data acquisition unit 52e inputs the inspection image data obtained by the camera 531 into the AI ​​model 200, thereby acquiring reconstructed image data.

[0107] Next, in step S302, the test image data obtained by the camera 531 is compared with the reconstructed image data related to the test image data obtained by the reconstructed image data acquisition unit 52e, and the difference in luminance for each pixel between the two data is calculated. Subsequently, pixels for which the difference is not within a predetermined allowable range are identified as defective pixels.

[0108] Next, in step S303, the area of ​​the connected component of the defective pixel (defect area) is calculated.

[0109] Next, in step S304, it is determined whether the maximum area among the calculated areas is equal to or greater than a predetermined threshold value. In other words, it is determined whether the defective area is outside the allowable range. If the maximum area is equal to or greater than the threshold value, it is determined that there is a fallen object Rb or foreign matter on the installation surface 100, and in step S305, it is determined that an "abnormal portion" exists in the test image data. Then, the portion of the test image data corresponding to the connected component of the defective pixel is identified as the "abnormal portion."

[0110] On the other hand, if the maximum area is below the threshold, it is determined in step S306 that there is no "abnormal part" in the test image data. Note that the above-mentioned determination method is only an example, and it may also be determined that there is an "abnormal part" if there is even one defective pixel, for example.

[0111] Then, after determining whether or not there is an "abnormal portion," in step S307, it is determined whether or not the processes of steps S301 to S306 (i.e., the process of determining whether or not there is an abnormal portion) have been performed on all of the newly acquired test image data. If the processes have been performed on all of the new test image data, the process proceeds to step S308. On the other hand, if there is new test image data that has not yet been processed, the process returns to step S301, and the processes are performed on the test image data that has not yet been processed.

[0112] After performing the above processing on all newly acquired test image data, in step S308, it is determined whether there are multiple test image data in which the presence of "abnormal parts" has been found (i.e., whether the processing of step S305 has been performed multiple times).

[0113] If there are multiple test image data in which the presence of an "abnormal portion" is recognized (step S308: Yes), in step S309, it is determined whether or not there has been a change in the position of the "abnormal portion" between the multiple test image data. If the position of the "abnormal portion" has not changed, it is presumed that the "abnormal portion" is a fallen object Rb, and in step S310 it is determined that a fallen object Rb is present on the installation surface 100. Furthermore, in step S311, information indicating the position of the fallen object Rb (for example, information for identifying the camera 531 among the multiple cameras 531 that captured the falling object Rb, the coordinates of the "abnormal portion" in the test image data, etc.) is acquired.

[0114] Then, in step S312, the determination result that there is a fallen object Rb on the installation surface 100 and information indicating the position of the fallen object Rb are stored in the inspection information storage unit 523. Note that as information indicating the position of the fallen object Rb, inspection image data in which the "abnormal portion" is emphasized or inspection image data in which the "abnormal portion" is replaced with a predetermined mark may be stored in the inspection information storage unit 523.

[0115] On the other hand, if the position of the "abnormal part" has changed (step S308; No), it is assumed that the "abnormal part" is a foreign object (for example, a piece of film or dust) that moves relatively easily, and in step S313 it is determined that there was a foreign object on the installation surface 100 but that there was no fallen object Rb. Then, in step S312, the determination result is stored in the inspection information storage unit 523. At this time, information indicating that a foreign object was present on the installation surface 100 and information indicating the position of the foreign object may be stored in the inspection information storage unit 523 together with the determination result.

[0116] Furthermore, if a negative determination is made in step S308, it is determined in step S314 whether there is only one piece of test image data in which the presence of an "abnormal portion" is recognized. If there is only one piece of test image data in which the presence of an "abnormal portion" is recognized (step S314; Yes), it is presumed that the "abnormal portion" is either a fallen object Rb or a foreign object, and it is determined in step S315 that a fallen object Rb or a foreign object is present on the installation surface 100. Furthermore, in step S316, the determination result is stored in the inspection information storage unit 523. At this time, information indicating the position of the fallen object Rb or foreign object may be stored in the inspection information storage unit 523 together with the determination result.

[0117] On the other hand, if the determination in step S314 is negative, that is, if the presence of an "abnormal portion" is not found in any of the test image data, it is determined in step S317 that there is no fallen object Rb on the installation surface 100. Then, in step S316, the determination result that there is no fallen object Rb on the installation surface 100 is stored in the inspection information storage unit 523. At this time, information indicating that no foreign object was found to be present on the installation surface 100 may be stored in the inspection information storage unit 523 together with the determination result.

[0118] As described above in detail, according to this embodiment, the entire surfaces of the horizontal support member 533 and the camera 531 that are visible when viewed from above are formed as inclined surfaces 533a, 531c that slope downward toward a predetermined outer edge. Therefore, even if a falling object Rb bounces off the installation surface 100 and ends up on the horizontal support member 533 or the camera 531, the falling object Rb is likely to slide down the inclined surfaces 533a, 531c and fall onto the installation surface 100. This makes it difficult for the falling object Rb to land on the horizontal support member 533, etc., and thus more reliably prevents the falling object Rb from remaining outside the imaging range of the camera 531. This allows the camera 531 to more reliably image the falling object Rb, thereby more appropriately confirming whether or not the falling object Rb is present on the installation surface 100. In particular, in this embodiment, the acute angles α3 and α1 of the angles formed by the inclined portions 533a and 531c with respect to the horizontal plane are set to 50° or more, which more reliably prevents the falling object Rb from landing on the horizontal support portion 533, etc.

[0119] Furthermore, by attaching the horizontal support part 533 to the PTP packaging machine 10, the camera 531 and the parts that support the camera 531 (the horizontal support part 533 and the attachment part 532) are entirely suspended above the installation surface 100. Therefore, it is possible to further reduce blind spots in the imaging range of the camera 531 compared to when legs for supporting the horizontal support part 533 are erected on the installation surface 100. This makes it possible to more appropriately check whether or not there is a fallen object Rb on the installation surface 100.

[0120] In addition, the horizontal support part 533 has a cantilever structure in which only the base end is attached to the PTP packaging machine 10, and the camera 531 is attached to the side of the tip end of the horizontal support part 533. Therefore, the camera 531 can be positioned higher than when the camera 531 is attached to the underside of the horizontal support part 533. This allows the imaging range of the camera 531 to be wider, and ultimately makes it possible to more appropriately check whether or not there is a fallen object Rb on the installation surface 100.

[0121] Furthermore, the entire surface of the mounting portion 532 that is visible when viewed from above is configured as a mounting portion-side slope 532a that slopes downward toward a predetermined outer edge. This makes it less likely that a fallen object Rb will land on the mounting portion 532, preventing the camera 531 from capturing an image of the fallen object Rb. This makes it possible to more appropriately check whether or not a fallen object Rb is present on the installation surface 100. In particular, in this embodiment, the acute angle α2 of the mounting portion-side slope 532a with respect to the horizontal plane is set to 50° or greater, which more reliably prevents a fallen object Rb from landing on the mounting portion 532.

[0122] Furthermore, the lighting mounting portion 541 has the lighting-side slope portion 541a, which can effectively prevent the fallen object Rb from landing on the lighting mounting portion 541. In particular, in this embodiment, the acute angle α4 of the lighting-side slope portion 541a with respect to the horizontal plane is set to be 50° or more, which can more reliably prevent the fallen object Rb from landing on the lighting mounting portion 541.

[0123] Additionally, the top 533b of the horizontal support portion 533 is not angular but curved. Therefore, the falling object Rb can be caught on the top 533b, which more reliably prevents the falling object Rb from landing on the horizontal support portion 533. This makes it possible to more appropriately check whether or not a falling object Rb is present on the installation surface 100. Furthermore, it is possible to more reliably prevent the falling object Rb from being damaged by contacting the top 533b.

[0124] Additionally, in this embodiment, the lighting device 542 can illuminate at least the imaging range of the camera 531 on the installation surface 100. Therefore, the camera 531 can more clearly image the fallen object Rb on the installation surface 100. Furthermore, when visually checking the installation surface 100, the fallen object Rb can be found more easily.

[0125] Furthermore, light shielding plate 543 can prevent light emitted from lighting device 542 from directly entering camera 531. Therefore, it is possible to more reliably prevent problems such as halation from occurring in image data obtained by camera 531.

[0126] In addition, the entire surface of the light-shielding plate 543 that is visible when viewed from above is configured as a light-shielding plate-side inclined surface 543a that slopes downward toward a predetermined outer edge. This makes it less likely that a falling object Rb will land on the light-shielding plate 543, preventing the camera 531 from capturing an image of the falling object Rb. This makes it possible to more appropriately check whether or not a falling object Rb is present on the installation surface 100. In particular, in this embodiment, the acute angle α5 of the light-shielding plate-side inclined surface 543a with respect to the horizontal plane is set to 50° or more, which more reliably prevents a falling object Rb from landing on the light-shielding plate 543.

[0127] Furthermore, the determination unit 52g compares the test image data obtained by the camera 531 with reconstructed image data that is reconstructed by inputting the test image data into the AI ​​model 200, and determines the presence or absence of a fallen object Rb on the installation surface 100 based on the comparison result. Therefore, the presence or absence of a fallen object Rb can be confirmed more accurately and quickly compared to when the presence or absence of a fallen object Rb is determined by visually checking the image data.

[0128] Furthermore, the conditions (for example, ambient light, etc.) of the installation surface 100 can be matched between the two sets of image data to be compared, so that the presence or absence of a fallen object Rb on the installation surface 100 can be confirmed with higher accuracy.

[0129] Additionally, in this embodiment, the determination unit 52g determines the presence or absence of a fallen object Rb based not simply on whether or not there is an "abnormal portion" in the test image data, but on whether or not there is a change in the position of the "abnormal portion." Therefore, the presence or absence of a fallen object Rb on the installation surface 100 can be determined with even greater accuracy.

[0130] The present invention is not limited to the above-described embodiment, and may be implemented as follows: Of course, other applications and modifications not exemplified below are also possible.

[0131] (a) In the above embodiment, the lighting device 542 is provided corresponding to one widthwise end side of the PTP packaging machine 10, and is configured to irradiate the installation surface 100 with light from one direction. However, the lighting device 542 may be provided corresponding to both widthwise end sides of the PTP packaging machine 10, so that predetermined light is irradiated onto the installation surface 100 from two different directions. Alternatively, predetermined light may be irradiated onto the installation surface 100 from all four sides of the installation surface 100.

[0132] (b) As shown in Fig. 17, the support-side inclined surface 533a may include a buffer 535 capable of buffering the impact applied to a falling object Rb when the support-side inclined surface 533a comes into contact with the falling object Rb. The buffer 535 corresponds to a "buffering means." The buffer 535 may be made of a material capable of absorbing impact, such as soft rubber, felt, urethane, or an elastic material.

[0133] By providing the buffer portion 535, it is possible to buffer the impact applied to the falling object Rb from the support-side slope portion 533a. Therefore, it is possible to more reliably prevent the falling object Rb from moving significantly (for example, bouncing) due to the impact applied from the support-side slope portion 533a. This allows the falling object Rb to be more reliably imaged by the camera 531, and it is possible to more appropriately check whether the falling object Rb is present on the installation surface 100.

[0134] Furthermore, by providing the buffer portion 535, damage (such as cracking) of the fallen object Rb due to impact can be more reliably prevented, and the fallen object Rb can be more easily found in its original shape. Therefore, collection of the fallen object Rb can be more easily performed. Preventing cracking of the fallen object Rb is particularly effective when the tablet 5 (falling object Rb) contains a special pharmaceutical ingredient.

[0135] The buffer portion 535 may be applied to the above-mentioned camera-side slope portion 531c, mounting portion-side slope portion 532a, lighting-side slope portion 541a, and light-shielding plate-side slope portion 543a.

[0136] (c) In the above embodiment, the horizontal support section 533 has a cantilever structure in which only the base end is attached to the PTP packaging machine 10. In contrast to this, as shown in Figures 18 and 19, mounting sections 532 may be provided at both ends of the horizontal support section 533, and the horizontal support section 533 may be attached to the PTP packaging machine 10 via the mounting sections 532, thereby giving the horizontal support section 533 a double-supported structure. In this case, the camera 531 may be attached to the underside of the horizontal support section 533.

[0137] (d) In the above embodiment, the top 533b of the horizontal support portion 533 has an upwardly convex curved shape, but the top 533b may have an angular shape.

[0138] (e) In the above embodiment, the entire surface of cover 531b of camera 531 that is visible when viewed from above has the same shape as the side of a cone, but the shape of cover 531b is not limited to this. Therefore, for example, as shown in FIGS. 20 and 21 , cover 531b may be configured so that the entire surface that is visible when viewed from above has the same shape as the side of a pyramid. Furthermore, in cases such as when camera 531 is attached to the underside of horizontal support part 533, cover 531b may be configured so that the entire surface that is visible when viewed from above has the same shape as the side of a truncated cone or pyramid.

[0139] (f) In the above embodiment, the judgment unit 52g estimates whether the “abnormal part” is a falling object Rb or a foreign object based on the change in the position of the “abnormal part.” However, without making such an estimation, the “abnormal part” may simply be uniformly estimated to be a falling object Rb.

[0140] (g) In the above embodiment, the presence or absence of a falling object Rb is determined by the determination unit 52g, but the presence or absence of a falling object Rb may also be determined by an operator or the like visually checking the image data obtained by the camera 531. Furthermore, the presence or absence of a falling object Rb may also be determined without using the AI ​​model 200.

[0141] (h) The learning unit 52f may have a function of performing additional learning of the AI ​​model 200 using image data obtained by the camera 531.

[0142] (i) The configuration of the AI ​​model 200 (neural network 190) as the "identification means" and its learning method are not limited to those described in the above embodiment. For example, the neural network 190 may be configured to perform normalization or other processing on various data as needed when performing the learning process or the acquisition process of reconstructed image data. Furthermore, the structure of the neural network 190 is not limited to that shown in FIG. 12 , and may be configured, for example, with a pooling layer provided after the convolution layer 193. Of course, the number of layers of the neural network 190, the number of nodes in each layer, and the connection structure of each node may be different.

[0143] Furthermore, in the above embodiment, the AI ​​model 200 (neural network 190) is a generative model having the structure of a convolutional autoencoder (CAE), but it is not limited to this and may be a generative model having the structure of a different type of autoencoder, such as a variational autoencoder (VAE).

[0144] Furthermore, in the above embodiment, the neural network 190 is configured to learn using the error backpropagation method, but this is not limiting, and the neural network 190 may be configured to learn using various other learning algorithms.

[0145] Additionally, neural network 190 may be configured by a dedicated AI processing circuit such as an AI chip. In this case, only learning information such as parameters may be stored in AI storage unit 524, which may be read by the dedicated AI processing circuit and set in neural network 190 to configure AI model 200.

[0146] (j) In the above embodiment, the tablet 5 is given as an example of the article, but the article is not limited to the tablet 5. Therefore, the article may be, for example, an electronic component.

[0147] Furthermore, in the above embodiment, the tablet 5 as the article has a circular shape in plan view, but the shape of the article may be other than circular in plan view. However, articles with a circular cross section, such as a tablet with a circular shape in plan view, a cylindrical capsule, or a spherical pill, tend to roll on a slope (such as the slope portions 533a, 531c), and the effect of providing the slope portions 533c, 531, etc., is more likely to be exhibited. Therefore, the installation surface inspection device 50 is particularly suitable for use with a device having a function for transporting articles with a circular cross section.

[0148] (k) In the above embodiment, the installation surface inspection device 50 is applied to the PTP packaging machine 10, but the device to which the installation surface inspection device 50 is applied may be any device that has the function of transporting a specified item (i.e., any device in which an item may fall during transportation), and is not limited to the PTP packaging machine 10. [Explanation of symbols]

[0149] 5...tablet (item), 10...PTP packaging machine (device), 50...installation surface inspection device, 52e...reconstructed image data acquisition unit (reconstructed image data acquisition means), 52g...determination unit (comparison means, determination means), 191...encoder unit (encoding unit), 192...decoder unit (decoding unit), 200...AI model (identification means), 531...camera, 531c...camera side slope portion (slope portion), 532...mounting unit, 532a...mounting unit side slope portion, 533...horizontal support portion, 533a...support unit side slope portion (slope portion), 533b...top, 535...buffer portion (buffer means), 542...lighting device (irradiation means), 543...shade plate, 543a...shade plate side slope portion, Rb...falling object.

Claims

1. An installation surface inspection device for inspecting the presence or absence of a dropped object, which is a dropped item, on an installation surface of a device having a function of transporting a predetermined item, a horizontal support extending horizontally and at least a portion of which is positioned below the device; a camera attached to the horizontal support portion and capable of capturing an image of the installation surface; An installation surface inspection device characterized in that at least the entire surface of the horizontal support portion and the camera that is visible when viewed from above is an inclined portion that slopes downward toward a specified outer edge.

2. 2. The installation surface inspection device according to claim 1, wherein the horizontal support portion is attached to the device, so that the camera and all of the components supporting the camera are suspended above the installation surface.

3. the horizontal support is attached to the device only at its proximal end; 3. The installation surface inspection device according to claim 2, wherein the camera is attached to a side surface of a tip end of the horizontal support part.

4. a mounting portion interposed between the horizontal support portion and the device for mounting the horizontal support portion to the device; The installation surface inspection device according to claim 2, characterized in that the entire surface of the mounting portion that can be seen when viewed from above is a mounting portion side slope portion that slopes downward toward a predetermined outer edge side.

5. 2. The installation surface inspection device according to claim 1, wherein an acute angle of the inclined surface relative to a horizontal plane is 50 degrees or more.

6. The horizontal support portion has a shape in which two of the inclined portions, the inclination directions of which are opposite to each other, are connected at their tops, 2. The installation surface inspection device according to claim 1, wherein the top portion has an upwardly convex curved surface shape.

7. an illumination means positioned below the camera and configured to illuminate at least an imaging range of the camera on the installation surface with predetermined light; a light shielding plate that covers the illumination means from above and prevents light emitted from the illumination means from directly entering the camera, The installation surface inspection device described in claim 1, characterized in that the entire surface of the light-shielding plate that is visible when viewed from above is a light-shielding plate side slope portion that slopes downward toward a predetermined outer edge side.

8. a recognition means for generating the image data by training a neural network having an encoding unit that extracts features from the input image data and a decoding unit that reconstructs the image data from the features, using only image data relating to the installation surface without the fallen object as training data; a reconstructed image data acquisition means for inputting the inspection image data obtained by the camera to the identification means and acquiring reconstructed image data as reconstructed image data; a comparison means for comparing the inspection image data and the reconstructed image data; 2. The installation surface inspection device according to claim 1, further comprising a determination means for determining the presence or absence of the fallen object on the installation surface based on the comparison result by the comparison means.

9. 2. The installation surface inspection device according to claim 1, wherein the inclined surface portion has a buffering means capable of absorbing an impact applied to the falling object when the inclined surface portion comes into contact with the falling object.

Images