Product inspection system and product inspection method

Abstract

A product inspection system is provided with: a spectrum measurement unit for measuring the spectrum of a transported product; and a quality determination unit (44) that determines the quality of the product on the basis of an output obtained by inputting the spectroscopic data of the product measured by the spectroscopic measurement unit into a learned model generated by machine learning. The learned model is generated by machine learning using teacher data including spectral data measured in mutually different postures.

Description

Technical Field

[0001] The present disclosure relates to a product inspection system and a product inspection method. Background Art

[0002] Patent Document 1 discloses a technology for high-speed product quality assessment using non-destructive spectroscopy. According to Patent Document 1, the spectrum of a moving product can be measured, and its quality can be assessed without stopping the product, enabling high-efficiency product quality assessment.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Laid-Open No. 2020-159971 Summary of the Invention

[0006] Technical problem to be solved by the invention

[0007] It would be desirable to be able to determine whether a product is good or bad with higher accuracy. The present inventors have conducted extensive research and have come up with a technology that can achieve higher-accuracy determination of whether a product is good or bad using spectroscopic analysis.

[0008] One aspect of the present disclosure has been made in view of such circumstances, and an object of the present disclosure is to provide a technology capable of determining the quality of a product using spectroscopic analysis with higher accuracy.

[0009] Technical solutions to technical problems

[0010] To address the aforementioned issues, a product inspection system according to one embodiment of the present disclosure includes: a spectrum measuring unit for measuring the spectrum of a conveyed product; and a quality determination unit for determining the quality of the product based on an output obtained by inputting the spectrum data of the product measured by the spectrum measuring unit into a learned model generated by machine learning. The learned model is generated by machine learning using teacher data containing spectrum data measured in different postures.

[0011] Another aspect of the present disclosure is a product inspection method. This method includes the steps of measuring the spectrum of a transported product; and determining whether the product is good or bad based on the output obtained by inputting the measured spectral data of the product into a learned model generated by machine learning. The learned model is generated by machine learning using teacher data containing spectral data measured in different postures.

[0012] Furthermore, any combination of the above-described constituent elements, and any form of expression of the present disclosure in the form of a method, an apparatus, a system, a recording medium, a computer program, etc., may also be practiced as additional aspects of the present disclosure.

[0013] Effects of the Invention

[0014] According to one aspect of the present disclosure, it is possible to more accurately determine whether a product is good or bad using spectroscopic analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a diagram showing the configuration of a product inspection system according to an embodiment.

[0016] Figure 2 Yes Figure 1 A three-dimensional diagram of a conveying mechanism conveying tablets.

[0017] Figure 3 Yes Figure 1 A diagram showing the situation when the clamp is used.

[0018] Figure 4 Yes Figure 1 A diagram illustrating the posture of tablets in a conveying mechanism.

[0019] Figure 5 Yes Figure 1 A diagram illustrating the posture of tablets in a conveying mechanism.

[0020] Figure 6 Yes Figure 1 A diagram illustrating the posture of tablets in a conveying mechanism.

[0021] Figure 7 Yes Figure 1 A diagram illustrating the posture of tablets in a conveying mechanism.

[0022] Figure 8 Yes Figure 1 A diagram illustrating the posture of tablets in a conveying mechanism.

[0023] Figure 9 Yes Figure 1 A diagram illustrating the posture of tablets in a conveying mechanism.

[0024] Figure 10 Yes Figure 1 A diagram illustrating the posture of tablets in a conveying mechanism.

[0025] Figure 11 Yes Figure 1 Block diagram of the control system of the product inspection system.

[0026] Figure 12This is a flowchart illustrating the operations in the preparation phase.

[0027] Figure 13 This is a flowchart explaining the operation of determining whether a product is good or not during the inspection phase.

[0028] Figure 14 It is a diagram showing the configuration of a product inspection system according to a modified example. DETAILED DESCRIPTION

[0029] The product inspection system of the embodiment measures the spectrum of a product through spectroscopic analysis, inputs this data into a learned model (calibration model), estimates the content of a specific component in the product, and determines the product's quality based on the estimated content of the specific component. A "learned model" is a learned model based on machine learning, and has been learned so that, when input with the product's spectral data, it estimates (outputs) the content of the specific component contained in the product. A "specific component" is a specific component contained in the product and is not particularly limited. It can be the component that has the greatest impact on the product's quality or the component with the highest content in the product.

[0030] In a product inspection system, in order to inspect the quality of products at high speed (e.g., less than 10 milliseconds per product), light is irradiated onto products transported and moved by a transport mechanism, and their spectra are measured without stopping the movement of the products.

[0031] However, since the products are arranged at high speed in the conveying mechanism, the posture of the arranged products may differ for each product, although it also depends on the arrangement method.

[0032] Furthermore, the transport mechanism changes over time due to use. For example, if the transport mechanism includes a conveyor belt, the belt may change over time due to use, stretching and bending. Therefore, the position of the product on the transport mechanism may change over time.

[0033] If the posture of the product on the conveying mechanism is different, the posture of the product relative to the measuring unit for measuring the spectrum will also be different, which will affect the measured spectrum and the estimation accuracy of the content of the specific component of the learned model, and further affect the accuracy of the quality judgment.

[0034] Therefore, in this embodiment, a learned model is generated that takes into account that the product posture during spectrum measurement may vary from product to product and over time. This allows for more accurate estimation of the content of specific components in the product, achieving high productivity and more accurate quality determination.

[0035] More specifically, during the "preparation phase," the product inspection system intentionally measures spectra in various postures for products with known specific component content. These spectra are used as teacher (learning) data to generate a learned model. During the "inspection phase," the product inspection system continuously measures spectra of products being transported by a conveyor mechanism. These spectra are input into the learned model generated during the preparation phase to estimate the specific component content in the product and determine its quality.

[0036] The following examples illustrate the case where the product is a tablet. Tablets are typically medicines, particularly pharmaceuticals, but are not limited thereto and may also be, for example, health foods or supplements.

[0037] Furthermore, as long as the product is a solid phase and the content of a specific component affects the quality, it may be a product different from the tablet. In this case, as long as there is no conflict, the technical concept described below can be applied to the other product, so the "tablet" in the following description can be replaced with the other product.

[0038] Figure 1 1 is a diagram showing the configuration of a product inspection system 1 according to an embodiment. The product inspection system 1 includes a spectrometer 10 , a conveyor mechanism 12 , an out-of-system removal device 14 , a jig 16 , and an information processing device 18 .

[0039] The spectrum measuring unit 10 irradiates light to the tablet T and measures the spectrum of the transmitted light. Figure 3 As will be described later, the spectrum of the tablet T held in the holder 16 is measured. The tablet T is a tablet having a known content of a specific component and is also referred to as a sample tablet T. In addition, during the inspection phase, the spectrum measuring unit 10 measures the spectrum of the tablet T held in the holder 16. Figure 1 As shown, the spectrum of the tablets T conveyed sequentially by the conveying mechanism 12 is measured without stopping the movement. These tablets T are tablets to be inspected whose content of a specific component is unknown.

[0040] The spectrum measurement unit 10 includes an irradiation unit 20 and a light receiving unit 22. The irradiation unit 20 irradiates the tablet with measurement light L1. The light receiving unit 22 receives (detects) transmitted light L2 from the tablet T when the irradiation unit 20 irradiates the tablet with measurement light L1.

[0041] The following describes an example of the irradiation unit 20 and the light receiving unit 22. The irradiation unit 20 and the light receiving unit 22 are, of course, not limited to the following configuration. The irradiation unit 20 may also include a broadband pulse light source and a pulse stretching element that stretches the pulse width of the emitted pulse light so that the wavelength and elapsed time in a pulse have a one-to-one relationship. The pulse stretching element may also include an arrayed waveguide diffraction grating that spatially divides the pulse light emitted from the pulse light source according to wavelength, and a number of optical fibers corresponding to the number of wavelengths divided by the arrayed waveguide diffraction grating.

[0042] return Figure 1 The measuring light L1 is irradiated onto the first surface of the tablet T ( Figure 1 The light L1 is transmitted through the tablet T and radiated from the second surface (upper surface) as transmitted light (hereinafter also referred to as object light) L2. L1 (λ), the transmittance X(λ) of the object light L2 has a wavelength dependence, the spectrum I of the object light L2 L2 (λ) is represented by the following formula (1).

[0043] I L2 (λ)=X(λ)×I L1 (λ)…(1)

[0044] The light receiving unit 22 is provided on the opposite side of the irradiation unit 20 across the conveying mechanism 12, and detects the diffused transmitted light emitted from the second surface of the tablet T. The light receiving unit 22 includes a photodetector that detects the diffused transmitted light of the tablet T as the object light L2. In addition to the photodetector, the light receiving unit 22 may also include an A / D converter, a focusing optical system, etc., but in the embodiment of FIG. Figure 1 A photodetector is a photoelectric conversion element that converts an optical signal into an electrical signal, and examples thereof include a photodiode, an avalanche photodiode, a phototransistor, a photomultiplier using the photoelectric effect, and a photoconductive element using a change in resistance due to light irradiation.

[0045] The spectrum measuring unit 10 generates a spectrum I of the object light L2 based on the output signal of the light receiving unit 22. L2 (λ). Then, based on the spectrum I of the measurement light L1 L1 (λ) and the spectrum I of the object light L2 L2 (λ), and the transmittance X(λ) of the tablet T was calculated by the following formula (2).

[0046] X(λ)=I L2 (λ) / I L1 (λ)…(2)

[0047] As a modified example, the spectrum measuring unit 10 may also measure the spectrum of light reflected from the tablet T. In this case, the light receiving unit 22 is disposed, for example, on the same side of the conveying surface 12a of the conveying mechanism 12 as the irradiating unit 20, and receives light reflected from the tablet T when the irradiating unit 20 irradiates the tablet T with the measuring light L1.

[0048] As a further modification, the spectrum measuring unit 10 may measure the spectra of both the transmitted light L2 and the reflected light from the tablet T.

[0049] The measurement light L1 is light capable of estimating the content of a specific component in the tablet T. The measurement light L1 is not particularly limited, and may be, for example, light within a wavelength range encompassing the near-infrared region, specifically light within a wavelength range of 1000 nm to 1300 nm, more preferably light within a wavelength range of 1100 nm to 1200 nm. The measurement light L1 may be pulsed light.

[0050] The conveying mechanism 12 is a mechanism for conveying the tablets T in the direction in which the tablets T are arranged during the inspection phase. Figure 1 The tablet T is conveyed from left to right in the middle. The tablet T may be sucked by a suction mechanism (not shown) and adsorbed on the conveying surface 12a. In this case, the tablet T can be conveyed stably.

[0051] Hereinafter, the direction in which the conveying mechanism 12 conveys the tablets T is referred to as a conveying direction x, the direction perpendicular to the conveying direction x is referred to as a width direction y, and the direction perpendicular to both the conveying direction x and the width direction is referred to as a height direction z.

[0052] Figure 2 1 is a perspective view showing a situation where the conveying mechanism 12 conveys the tablet T. As an embodiment, Figure 2 As shown, the conveying mechanism 12 includes a pair of conveying belts 52. The pair of conveying belts 52 are arranged at a predetermined interval in the width direction y. The pair of conveying belts 52 are endless belts wound around a plurality of rollers (not shown) whose rotation axes extend in the width direction y. The tablet T is placed across the pair of conveying belts 52. When a drive unit (not shown), such as a motor, drives the rollers, the pair of conveying belts 52 rotates accordingly, conveying the tablet T. The measurement light L1 from the irradiation unit 20 passes between the pair of conveying belts 52 and irradiates the tablet T.

[0053] return Figure 1 The information processing device 18 centrally controls the product inspection system 1. During the preparation phase, the information processing device 18 uses teacher data, which will be described in detail later, to generate a learned model for determining the quality of tablets T. Furthermore, during the inspection phase, the information processing device 18 uses the learned model to determine the quality of tablets T, as will be described in detail later.

[0054] During the inspection phase, the system removal device 14 removes only the tablets T that are judged as defective during the quality assessment. The configuration of the system removal device 14 is not particularly limited and may be configured using known or future-available technologies.

[0055] Figure 3 Yes Figure 1 FIG. 1 is a diagram showing a situation when the clamp 16 is used. Figure 3This figure shows the spectrometer 10 and its surroundings as viewed along the transport direction x. The clamp 16 is positioned between the irradiating section 20 and the light receiving section 22. In this example, the transport mechanism 12 is retracted from a position where it intersects the space between the irradiating section 20 and the light receiving section 22. Alternatively, the spectrometer 10 is retracted from a position where the transport mechanism 12 intersects the space between the irradiating section 20 and the light receiving section 22.

[0056] The clamp 16 is used to hold the sample tablet T in various postures. Figures 4 to 8 The difference in the posture of each tablet T during the spectrum measurement described in . In the preparation stage, the spectrum measuring unit 10 measures a spectrum used as teacher data while the sample tablet T is held in a desired posture by the jig 16 .

[0057] For example, the fixture 16 includes a circular plate 16a that rotates about a rotation axis R. By rotating the circular plate 16a holding the tablet T, a state of being transported at the transport speed of the transport mechanism 12 is reproduced. In this state, the spectrum of the tablet T held on the circular plate 16a, i.e., the spectrum used as training data, is measured by the spectrum measurement unit 10.

[0058] Furthermore, in the measurement of the spectrum in the preparation stage, i.e., the measurement of the spectrum used as the teaching data, the spectrum may be measured by a spectrum measuring unit (not shown) other than the spectrum measuring unit 10 and having the same configuration as the spectrum measuring unit 10. This other spectrum measuring unit may be provided at a position where the transport mechanism 12 does not cross between the irradiating unit and the light receiving unit.

[0059] Next, refer to Figures 4 to 8 , the different postures of each tablet T in the conveying mechanism 12 will be described. In addition, the "different postures" here also include the case where the positions in the width direction y and the height direction z are different. In addition, here, the case where the conveying mechanism 12 is configured to include a pair of conveying belts 52 will be described.

[0060] Figure 4 This figure shows the conveying mechanism 12 as viewed from a direction perpendicular to the conveying surface 12a. In this example, the positions of the tablets T in the width direction y vary. Although this depends on how the tablets T are arranged, the tablets T are arranged on the conveying mechanism 12 at a high speed (e.g., 10 milliseconds or less per tablet), which may cause positional deviations in the width direction.

[0061] Figure 5 (a) to (c) are views of the conveying mechanism 12 along the width direction y. Figure 5In (a) to (c), at the irradiation position of the measuring light L1 on the tablet T, the position of the tablet T in the height direction z relative to the predetermined reference conveying plane S is different. The reference conveying plane S is not particularly limited and may be, for example, a plane that corresponds to the conveying plane before the conveying mechanism 12 undergoes aging changes. Figure 5 In (a), the conveying surface is consistent with the reference conveying surface S. Figure 5 In (b) and (c), for example, the conveying belt is stretched and bent due to age-related changes, so that the conveying surface 12a is lower than the reference conveying surface S. Figure 5 In (a) to (c), since the heights of the conveying surface 12a at the irradiation position are different from each other, the heights relative to the reference conveying surface S are different from each other.

[0062] Figure 6 (a) and (b) are views of the transport mechanism 12 along the transport direction x. Figure 6 In (a) and (b), at the position where the measurement light L1 irradiates the tablet T, the angle θy of the tablet T relative to the predetermined conveyance reference plane S, specifically the angle θy about an axis parallel to the conveyance direction x, varies from tablet to tablet. The reference conveyance plane S is not particularly limited; for example, it may be the same conveyance plane as that of the conveyance mechanism 12 before aging. For example, if the curved outer surface of the tablet T contacts the conveyance surface 12a, the angle θy of the tablet T relative to the reference conveyance plane S may vary from tablet to tablet.

[0063] Figure 7 (a) and (b) are views of the transport mechanism 12 along the transport direction x. Figure 7 In (a) and (b), Figure 6 Similarly to (a) and (b), the angle θy of the tablet T at the irradiation position is different. For example, the angle θy may be different depending on the degree of progress of the aging of the pair of conveyor belts 52.

[0064] Figure 8 (a) and (b) are views of the transport mechanism 12 along the width direction y. Figure 8 In (a) and (b), at the irradiation position of the measurement light L1 on the tablet T, the angle θx of the tablet T relative to the predetermined conveyance reference plane S, and this angle θx about an axis parallel to the width direction y, is different for each tablet T. The reference conveyance plane S is not particularly limited; for example, it may be the same conveyance plane as that of the conveyance mechanism 12 before aging. For example, if the curved outer surface of the tablet T contacts the conveyance surface 12a, the angle θx of the tablet T relative to the reference conveyance plane S may differ for each tablet T.

[0065] Figure 9 (a) and (b) are views of the transport mechanism 12 along the width direction y. Figure 9 In (a) and (b), Figure 8 Similarly to (a) and (b), the angle θx of the tablet T at the irradiation position is different. For example, the angle θx may be different due to the change of the conveyor belt 52 over time.

[0066] Figure 10 12a is a diagram showing the conveying mechanism 12 as viewed from a direction perpendicular to the conveying surface 12a. Figure 8 In the embodiment, the tablets T have a non-rotationally symmetrical shape, with orientations differing from one another about an axis perpendicular to the conveying surface 12a (in other words, about an axis parallel to the height direction z). The tablets T have a dividing line L, and the extending directions of the dividing line L differ from one another when viewed perpendicular to the conveying surface 12a.

[0067] In addition, here, the case where the conveying mechanism 12 is configured to include a pair of conveying belts 52 is described, but even if the conveying mechanism 12 is configured otherwise, it is possible that Figures 4-10 At least one of the postures is different.

[0068] Figure 11 This is a block diagram showing the control system of product inspection system 1. While each module shown here can be implemented in hardware using components such as a computer's CPU (central processing unit) and mechanical devices, and in software using computer programs, the functional modules described here are implemented through the collaboration of these components. Therefore, those skilled in the art who have access to this specification should understand that these functional blocks can be implemented in various forms through a combination of hardware and software.

[0069] The information processing device 18 includes a spectrum data receiving unit 40, a model generating unit 42, a quality determination unit 44, and a model storage unit 46. The components of the information processing device 18 are shown only as components that are of interest in this embodiment.

[0070] There is no limit to the number of physical devices (frames) that comprise information processing device 18. Information processing device 18 can be implemented as a single device or through the collaboration of multiple devices. For example, information processing device 18 can be implemented through the collaboration of a first device and a second device, where the first device includes spectral data receiving unit 40, model generating unit 42, and model storage unit 46, and the second device includes quality determination unit 44.

[0071] The model storage unit 46 stores a learned model (calibration model) generated by the model generation unit 42 described later.

[0072] The spectrum data receiving unit 40 previously receives spectrum data of the sample tablet T measured by the spectrum measuring unit 10. Alternatively, as described above, the spectrum of the sample tablet T may be measured by a spectrum measuring unit different from the spectrum measuring unit 10. In this case, the spectrum data receiving unit 40 receives the spectrum data of the sample tablet T measured by the different spectrum measuring unit.

[0073] Furthermore, the spectrum data receiving unit 40 receives spectrum data of the tablet T sequentially measured by the spectrum measuring unit 10 during the inspection phase.

[0074] The model generation unit 42 performs machine learning using the teacher data in the preparatory stage to generate a learned model that, when input with spectral data of a tablet T, outputs (estimates) the content of a specific component in the tablet T. The learned model can be constructed using any machine learning method that is known or will be available in the future. The model generation unit 42 stores the generated learned model in the model storage unit 46. Teacher data will be described later.

[0075] During the inspection phase, the quality determination unit 44 uses the learned model stored in the model storage unit 46 to determine the quality of the tablet T. Specifically, the quality determination unit 44 inputs the spectral data received by the spectral data receiving unit 40 into the learned model, thereby causing the learned model to estimate (output) the content of a specific component in the tablet T.

[0076] Next, the quality determination unit 44 determines the quality of the tablet T based on the content of the specific component estimated by the learned model. The quality determination unit 44 determines the quality of the tablet T based on the content of the specific component estimated by the learned model. If the ratio of the content of the estimated characteristic component (in other words, the content of the specific component actually contained in the tablet T) to the content of the specific component that should be contained in the tablet T (hereinafter referred to as the specified content ratio) is within a specified threshold range, for example, if the specified content ratio is greater than 98% and less than 102%, the tablet T is determined to be a qualified product. If the specified content ratio is outside the threshold range, the tablet T is determined to be a defective product. The quality determination unit 44 may also determine the quality of the tablet T based on at least one other factor in addition to the content of the specific component. The other factor may also be the appearance of the tablet T, for example. In this case, even if the specified content ratio is within the specified threshold range, the quality determination unit 44 will determine the tablet T as a defective product as long as the appearance of the tablet T is poor. Whether the appearance of the tablet T is defective can also be determined by analyzing an image of the tablet T captured by an imaging unit (not shown). For example, if the tablet T is deformed such as chipped or cracked, the tablet T can be determined to have a defective appearance.

[0077] The system-external rejection device 14 includes a rejection control unit 48. The rejection control unit 48 centrally controls the system-external rejection device 14. The rejection control unit 48 controls the system-external rejection device 14 and rejects tablets T determined as defective by the quality determination unit 44 from the system. Tablets T determined as defective are automatically rejected, thereby preventing accidents in which defective products are shipped.

[0078] Next, the “teacher data” will be described. The teacher data is a plurality of data sets regarding a plurality of sample tablets T whose contents of specific components are known, and each of the data sets includes the contents of the specific components and spectral data.

[0079] The plurality of sample tablets T include sample tablets T having various contents, that is, sample tablets T having different contents of a specific component. Preferably, the plurality of sample tablets T include a sample tablet T having a specific component content of 100%.

[0080] The plurality of sample tablets T preferably include sample tablets T that cover the assumed content range of the specific component of the tablets T. That is, the plurality of sample tablets T preferably include sample tablets T having a specific component content within the assumed range, sample tablets T having a specific component content below the assumed range, and sample tablets T having a specific component content above the assumed range.

[0081] Variations in the content of a specific component in multiple sample tablets T may affect the accuracy of the generated model. Therefore, for example, the variation in the content of a specific component in multiple sample tablets T can be determined by trial and error. For example, if the expected content of a specific component in multiple sample tablets T is between 90% and 110%, sample tablets T with a content of 2% within the range of 88% to 112% can be prepared, that is, sample tablets T with a specific component content of 88%, 90%, 92%, ..., and 112%.

[0082] The plurality of data sets include data sets of spectral data measured in different postures. Figures 4-10 At least one of the position in the width direction y, the position in the height direction z, the angle θy of the tablet T relative to the specified conveying reference plane (i.e., the angle θy around the axis parallel to the conveying direction x), the angle θx of the tablet T relative to the specified conveying reference plane (i.e., the angle θx around the axis parallel to the width direction y), and the orientation around the axis perpendicular to the conveying surface 12a (i.e., around the axis parallel to the height direction z) described in the figure is different from each other.

[0083] The plurality of data sets may include only one data set obtained by measuring each sample tablet T in one posture, or may include a plurality of data sets obtained by measuring each sample tablet T in a plurality of postures.

[0084] The above is the basic structure of the product inspection system 1. Next, its operation will be described.

[0085] Figure 12 This is a flowchart illustrating the operations in the preparation phase. First, spectra are measured for multiple sample tablets T with known specific component contents in various postures (S10). Next, using training data containing the measured spectral data of the sample tablets T, a learned model for determining the quality of the tablets T is generated (S12).

[0086] Figure 13 This is a flowchart explaining the operation of determining whether a product is good or not during the inspection phase. Figure 13 The processing is repeated at a predetermined cycle (e.g., a predetermined cycle of 10 milliseconds or less), in other words, for each tablet being conveyed sequentially. First, the spectrum of the tablet T being conveyed by the conveying mechanism 12 is measured without stopping (S20). Next, the measured spectrum data is input into the learned model to estimate the content of a specific component in the tablet T (S22). Next, based on the estimated content of the specific component, the quality of the tablet T is determined (S24).

[0087] According to this embodiment, in the preparation stage, a learned model is created using teacher data containing spectral data of multiple sample tablets T with known specific component content, i.e., spectral data intentionally measured in various postures. In the inspection stage, this learned model is used to estimate the specific component content of the tablets and determine the tablet's quality. Specifically, the specific component content is estimated based on the learned model, which takes into account that the posture of the tablets T during spectral measurement may vary from tablet to tablet and over time. This allows for more accurate estimation of the specific component content and more accurate determination of the tablet's quality.

[0088] The present disclosure has been described above based on the embodiments. These embodiments are merely illustrative, and those skilled in the art will appreciate that various modifications are possible in the combination of these components and processing steps, and that such modifications are also within the scope of the present disclosure.

[0089] (Variation 1)

[0090] In the embodiment, a case is described in which the sample tablet T is held in various postures using the clamp 16 and the spectra of the sample tablet T in various postures, i.e., the spectra of the sample tablet T used as teacher data, are measured. However, as long as the spectra of the sample tablet T in various postures can be measured, the measurement of the spectra using the clamp 16 is not limited.

[0091] Figure 141 is a diagram showing the configuration of a product inspection system 101 according to a modified example.

[0092] The product inspection system 101 includes a spectrometer 10, a conveyor mechanism 12, an external removal device 14, an information processing device 18, and a posture detection unit 108. That is, the product inspection system 101 of this modification does not include the jig 16.

[0093] In the preparation stage of this modification, similarly to the inspection stage, the transport mechanism 12 sequentially transports the sample tablets T, and the spectrum measurement unit 10 measures the spectra of the sample tablets T sequentially transported by the transport mechanism 12 without stopping its movement.

[0094] During the preparation phase, the posture detection unit 108 detects the posture of the sample tablet T being conveyed by the conveying mechanism 12. The posture detection unit 108 may detect the posture of the sample tablet T at a position upstream of the spectrum measurement unit 10 as shown in the figure, or at a position downstream of the spectrum measurement unit 10, or at a position where the spectrum is measured by the spectrum measurement unit 10 (e.g., a position where the sample tablet T is irradiated with the measurement light L1 from the irradiation unit 20).

[0095] The configuration of the posture detection unit 108 is not particularly limited. For example, the posture detection unit 108 may include at least one camera. In this case, the posture detection unit 108 can also capture images of the tablet T from three directions: the transport direction x, the width direction y, and the height direction z. Furthermore, for example, the posture detection unit 108 may include an ultrasonic sensor. In short, the posture detection unit 108 may be configured to detect the posture of the sample tablet T.

[0096] The multiple datasets of training data in this variation include datasets of spectral data of the sample tablet T measured while being conveyed by the conveying mechanism 12, as described above. The multiple datasets of training data are not particularly limited and may be datasets selected by the user from the spectral data of the sample tablet T measured. The user may select multiple datasets of training data that include datasets of spectral data measured in various postures based on the posture detection results of the sample tablet T by the posture detection unit 108, for example, by referring to images of the sample tablet T captured by at least one camera serving as the posture detection unit 108. For example, the user may select multiple datasets of training data that include datasets of spectral data measured in at least one of the following postures: position in the width direction y, angle θy of the tablet T relative to a predetermined conveying reference plane, angle θx of the tablet T relative to a predetermined conveying reference plane, and orientation around an axis perpendicular to the conveying surface 12a.

[0097] According to this modification, the same functions and effects as those of the embodiment can be achieved.

[0098] (Variation 2)

[0099] In the embodiment, the spectrometer 10 is communicably connected to the information processing device 18. However, the spectrometer 10 may include the information processing device 18 therein.

[0100] (Variation 3)

[0101] In the embodiment, the learned model used in the good / bad judgment is described as being learned in a manner that outputs (estimates) the content of a specific component of the tablet when the spectral data of the tablet is input, but this is not limited to this. The learned model can also be learned in a manner that outputs (estimates) the good / badness of the tablet from the perspective of the content of the specific component.

[0102] Industrial applicability

[0103] The present invention relates to a product inspection system and a product inspection method.

[0104] Description of Reference Numerals

[0105] 1 product inspection system, 10 spectrum measurement unit, 44 quality judgment unit.

Claims

1. A product inspection system, characterized in that: have: a spectrometer section for measuring a spectrum of a product being transported; and a quality determination unit that determines the quality of the product based on an output obtained by inputting the spectrum data of the product measured by the spectrum measurement unit into a learned model generated by machine learning, The learned model is generated by machine learning using teacher data including spectral data measured in different postures.

2. The product inspection system according to claim 1, characterized in that: The learned model is configured to output the content of a specific component.

3. The product inspection system according to claim 1, characterized in that: The spectral data measured in the different postures include spectral data measured at different positions in a direction perpendicular to the conveying direction.

4. The product inspection system according to claim 1, characterized in that: The spectral data measured in different postures include spectral data measured in postures at different heights relative to a reference conveying surface.

5. The product inspection system according to claim 1, characterized in that: The spectral data measured in different postures include spectral data measured in postures having different angles with respect to a reference conveying plane when viewed in the conveying direction.

6. The product inspection system according to claim 1, characterized in that: The spectral data measured in different postures include spectral data measured in postures having different angles with respect to a reference conveying plane when viewed in a direction perpendicular to the conveying direction.

7. The product inspection system according to claim 1, characterized in that: When the product has a non-rotationally symmetrical shape, the spectral data measured in different postures include spectral data measured in postures having different orientations around an axis perpendicular to the conveying surface.

8. The product inspection system according to claim 1, characterized in that: When the product is a tablet having a dividing line, the spectral data measured in different postures include spectral data measured in postures in which the extending directions of the dividing lines are different when viewed in a direction perpendicular to the conveying surface.

9. The product inspection system according to claim 1, characterized in that: The quality determination unit further determines the quality of the product based on the appearance of the product.

10. A product inspection method, characterized in that: The following steps are involved: Determine the spectrum of the product being handled; and The quality of the product is determined based on the output obtained by inputting the measured spectral data of the product into a learned model generated by machine learning. The learned model is generated by machine learning using teacher data including spectral data measured in different postures.

Images