JP2020046270A - Packaging and inspection system - Google Patents

Abstract

To quickly detect an abnormal package.SOLUTION: A packaging and inspection system 100 includes: a packaging machine 20; an inspection section 40; a pre-inspection conveyor 30; a pre-inspection camera 31; and an image determination section 92. The packaging machine 20 forms a package P by packaging an article A with a film F. The inspection section 40 inspects the package P. The pre-inspection conveyor 30 conveys the package P discharged from the packaging machine 20 to the inspection section 40. The pre-inspection camera 31 photographs the package P conveyed by the pre-inspection conveyor 30. The image determination section 92 determines an abnormality of the package P by analyzing a still image or a moving image that is an output of the pre-inspection camera 31.SELECTED DRAWING: Figure 1

Description

  The present disclosure relates to packaging and inspection systems.

  BACKGROUND ART A bag making and packaging machine fills and packs an article such as food into a bag made of a film. Patent Document 1 (Japanese Patent Application Laid-Open No. 2009-161228) describes that an inspection unit including a seal checker, a weight checker, an X-ray foreign substance inspection device, and the like is provided downstream of a bag making and packaging machine.

  If the operation of the bag making and packaging machine such as sealing is inappropriate, the contents of the bag may be scattered on the production line. In this case, dirt due to oil or the like spreads not only to the bag making and packaging machine but also to the subsequent inspection section and the like, forcing the operator to perform cleaning. Therefore, it takes time to restore the production line.

  An object of the present disclosure is to quickly detect an abnormal package in a packaging and inspection system.

  A packaging and inspection system according to a first aspect includes a packaging machine, an inspection unit, a conveyor, a camera, and an image determination unit. The packaging machine creates a package by packaging an article with a packaging material. The inspection unit inspects the package. The conveyor carries the package discharged from the packaging machine to the inspection unit. The camera shoots the packages carried by the conveyor. The image determination unit determines an abnormality of the package by analyzing a still image or a moving image output from the camera.

  According to this configuration, the conveyor and the camera are arranged in front of the inspection unit. Therefore, the conveyor can be stopped when the camera detects that the abnormal package has reached the conveyor, so that the abnormal package is quickly detected and the inspection unit is contaminated with the abnormal package. Can be suppressed.

  A packaging and inspection system according to a second aspect is the packaging and inspection system according to the first aspect, wherein the inspection unit includes at least one of an X-ray inspection machine, a weight inspection machine, and a press seal inspection machine.

  According to this configuration, the inspection unit includes at least one of an X-ray inspection machine, a weight inspection machine, and a press-type seal inspection machine. Therefore, the possibility that these inspection machines need to be cleaned can be reduced.

  A packaging and inspection system according to a third aspect is the packaging and inspection system according to the first aspect or the second aspect, wherein the image determination unit includes a computer having a learned model formed by machine learning.

  According to this configuration, the image determination unit has a learned model formed by machine learning. Therefore, the accuracy of the abnormality determination can be improved by using the teacher data based on the image of the actual package.

  The packaging and inspection system according to a fourth aspect is the packaging and inspection system according to the third aspect, wherein the machine learning is deep learning. The trained model is in the form of a neural network.

  According to this configuration, machine learning is deep learning. Therefore, the accuracy of abnormality determination can be improved by increasing the amount of teacher data.

  The packaging and inspection system according to the fifth aspect is the packaging and inspection system according to any one of the first to fourth aspects, wherein the state of the package that the image determination unit determines to be abnormal is determined by the packaging material. Torn formed bags, failure to seal the packaging material, failure to cut the packaging material, incorrect bag size, scattering of articles out of the bag, incorrect spacing between bags and next bag, packaging Upside down, improper posture of the package, and improper timing at which the package is carried.

  According to this configuration, various aspects of the abnormality of the package can be determined.

  A packaging and inspection system according to a sixth aspect is the packaging and inspection system according to any one of the first to fifth aspects, further including an image storage unit that stores a still image or a moving image.

  According to this configuration, the packaging and inspection system includes the image storage unit. Therefore, the still image or the moving image accumulated in the image storage unit can be used for learning of the image determination unit.

  The packaging and inspection system according to the seventh aspect is the packaging and inspection system according to any one of the first to sixth aspects, further comprising a box packing device that packs a package discharged from the inspection unit into a packaging box. Prepare.

  According to this configuration, the packaging and inspection system includes the box packing device. Therefore, by using teacher data with an abnormal label attached to an image corresponding to the orientation of the package that is difficult to handle by the box packing device, it is possible to suppress problems in the box packing device.

  The packaging and inspection system according to an eighth aspect is the packaging and inspection system according to any one of the first to seventh aspects, wherein the camera includes a three-dimensional camera. The still or moving image includes information about the thickness of the package.

  According to this configuration, the camera obtains three-dimensional information of the package. Therefore, information useful for abnormality determination of the package can be obtained.

  The packaging and inspection system according to a ninth aspect is the packaging and inspection system according to any one of the first to eighth aspects, further comprising a sensor that detects that the package has passed a predetermined position on the conveyor. Prepare.

  According to this configuration, the packaging and inspection system includes the sensor that detects the package on the conveyor. Therefore, it is easy to cut out a portion related to the package from the moving image captured by the camera.

  A packaging and inspection system according to a tenth aspect is the packaging and inspection system according to any one of the first to eighth aspects, wherein the image determination unit includes a periodicity extraction unit that extracts data relating to the periodicity of the moving image. And a periodicity determining unit that determines the validity of the periodicity.

  According to this configuration, the image determining unit determines the periodicity of the moving image. Therefore, the abnormality of the package can be determined using the relationship between the two continuously transported packages.

  The inspection method of a package according to the eleventh aspect is as follows: (1) A packaging machine for making a package by packaging an article with a packaging material, a conveyor for carrying the package discharged from the packaging machine, and a package carried by the conveyor. Providing a camera for photographing an object, an image storage unit for storing a still image or a moving image output from the camera, and a function approximator for processing the still image or the moving image; A still image or a moving image generated in the image storage unit by photographing with the camera, a label indicating normality is attached, and a still image or moving image generated in the image storage unit by photographing the abnormal package by the camera. A process in which a function approximator to which teacher data including a video having a label indicating an abnormality attached thereto is input by performing machine learning to form a trained model; Ejected packaging Inputting a still image or a moving image, which is an output of a camera taking a state of being conveyed by a conveyor, to a learned model, and performing abnormality determination of a package based on a result output from the learned model. .

  According to this method, abnormal packages can be quickly detected.

  The packaging and inspection system of the present disclosure can quickly detect an abnormal package.

  The package inspection method according to the present disclosure can quickly detect an abnormal package.

1 is a schematic diagram illustrating a configuration of a packaging and inspection system 100 according to the present disclosure. FIG. 3 is a block diagram illustrating a control unit 80. FIG. 4 is a block diagram showing a camera control unit 84. It is a schematic diagram which shows the example of abnormality of the package P. It is a schematic diagram which shows the example of abnormality of the package P. It is a schematic diagram which shows the example of abnormality of the package P. It is a schematic diagram which shows the example of abnormality of the package P. It is a schematic diagram which shows the example of abnormality of the package P. It is a schematic diagram which shows the structure of the packaging and inspection system 100A which concerns on the modification B. FIG. 18 is a block diagram illustrating a camera control unit 84A according to a modification example D. FIG. 19 is a block diagram showing an abnormality determination engine 84B according to Modification E.

  Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings. The following embodiments are specific examples of the present disclosure, and do not limit the technical scope of the present disclosure.

(1) Overall Configuration FIG. 1 shows a packaging and inspection system 100 according to the present disclosure. The packaging and inspection system 100 inspects a package P, which is a product obtained by packaging an article A having a predetermined weight with a packaging film F, for shipping. The packaging and inspection system 100 includes a weighing machine 10, a packaging machine 20, a pre-inspection conveyor 30, a pre-inspection camera 31, an inspection unit 40, a sorting unit 50, a defective product storage unit 60, a box packing device 70, and a control unit 80. .

(2) Detailed configuration (2-1) Weighing machine 10
The weighing machine 10 is called a combination weighing machine. The weighing machine 10 sorts out the articles A having a predetermined weight and transfers them to the packaging machine 20.

(2-2) Packaging machine 20
The packaging machine 20 is called a bag making and packaging machine. The packaging machine 20 makes a package P by making a bag with the film F pulled out from the film roll FR and packing an article A having a predetermined weight with the bag.

(2-3) Conveyor 30 before inspection
The pre-inspection conveyor 30 carries the package P discharged from the packaging machine 20 to the inspection unit 40.

(2-4) Pre-inspection camera 31
The pre-inspection camera 31 captures a moving image of the package P carried by the pre-inspection conveyor 30.

(2-5) Inspection unit 40
The inspection unit 40 inspects whether the package P is a good product or a defective product. The inspection unit 40 includes, for example, an X-ray inspection machine 41, a weight inspection machine 42, and a press-type seal inspection machine 43. The X-ray inspection machine 41 detects entry of foreign matter by irradiating the product with X-rays. The weight checker 42 determines whether the weight of the package P is appropriate. The press-type seal inspection machine 43 inspects the bag of the package P for holes or poor sealing by pressing the bag.

  The inspection unit 40 is not limited to this mode, and may include at least one of the X-ray inspection machine 41, the weight inspection machine 42, and the pressing seal inspection machine 43.

(2-6) Distributing unit 50
The distribution unit 50 distributes the defective package P found by the inspection unit 40 from the non-defective package P.

(2-7) Defective product storage unit 60
The defective product storage unit 60 stores the defective product package P.

(2-8) Boxing device 70
The box packing device 70 packs the package P into the packing box B for shipping.

(2-9) Control unit 80
The control unit 80 controls various devices (the weighing machine 10, the packaging machine 20, the pre-inspection conveyor 30, the pre-inspection camera 31, the inspection unit 40, the sorting unit 50, the box packing device 70, etc.) included in the packaging and inspection system 100. Perform control. The control unit 80 performs wired communication or wireless communication with various devices. The control unit 80 includes, for example, a computer equipped with a GPU.

  As shown in FIG. 2, the control unit 80 executes a dedicated program to execute a weighing machine control unit 81, a packaging machine control unit 82, a conveyor control unit 83, a camera control unit 84, an inspection unit control unit 85, It functions as the division unit control unit 86 and the boxing device control unit 87. The weighing machine control unit 81 receives the output of the weighing machine 10 and controls the weighing machine 10. The packaging machine control unit 82 receives the output of the packaging machine 20 and controls the packaging machine 20. The conveyor control unit 83 operates or stops the pre-inspection conveyor 30. The camera control unit 84 receives and accumulates the moving image output from the pre-inspection camera 31, and performs various calculations based on the received moving image. The inspection unit control unit 85 receives the output of the inspection unit 40 and controls the inspection unit 40. The distribution unit control unit 86 controls the distribution unit 50. The box packing device control section 87 controls the box packing device 70.

  FIG. 3 shows details of the camera control unit 84. The camera control unit 84 further functions as an image storage unit 91 and an image determination unit 92. The image storage unit 91 stores the moving image output by the pre-inspection camera 31. The image determination unit 92 determines the abnormality of the package P by analyzing the moving image output from the pre-inspection camera 31.

(3) Details of Image Determination Unit 92 The image determination unit 92 has a function approximator X that can learn the relationship between input and output. The function approximator X is typically configured as a multilayer neural network, but is not limited to this mode, and may be realized by another mode such as a support vector machine. The function approximator X has acquired the learned model M as a result of the machine learning in advance. Machine learning is typically performed as deep learning, but is not limited to this mode. The function approximator X constituting the learned model M can infer the abnormality of the package P. Here, the abnormality of the package P is, for example,
・ Tear of the bag formed by the film F of the packaging material ・ Failed to seal the film F of the packaging material (FIG. 4)
-Failure to cut the film F of the packaging material (Fig. 5)
・ Inappropriate bag size (Fig. 6)
・ Scatter of article A out of the bag (FIG. 7)
-The package P is turned upside down.-The posture of the package P is inappropriate (Fig. 8, arrow indicates the transport direction of the pre-inspection conveyor 30).
And so on.

  When the image determination unit 92 detects these abnormalities, the conveyor control unit 83 of the control unit 80 stops the pre-inspection conveyor 30. Thereby, for example, it is possible to suppress the scattered articles A from soiling the inspection unit 40. Alternatively, it is possible to prevent the package P having the inappropriate posture reaching the box packing device 70 from exploding in the packing box B.

(4) Package P Inspection Method (4-1) Setting First, the weighing machine 10, the packaging machine 20, the pre-inspection conveyor 30, the pre-inspection camera 31, the inspection unit 40, the sorting unit 50, the defective product storage unit 60, The packaging and inspection system 100 is constructed by preparing the boxing device 70, the control unit 80, and the like.

(4-2) Learning Phase The learning phase for acquiring the learned model M is performed by supervised learning. The actual modes of learning vary. The following is merely an example.

(4-2-1) Imaging of Moving Image First, the packaging and inspection system 100 is operated. However, in the learning phase, devices after the inspection unit 40 are not essential. The pre-inspection camera 31 captures an image of the packaging machine 20 discharging a large number of packages P. The output of the pre-inspection camera 31 is stored in the image storage unit 91 as a moving image. The moving image includes both the normal package P and the abnormal package P.

(4-2-2) Extraction of Still Image From the moving image output from the pre-inspection camera 31 stored in the image storage unit 91, an image including the package P is extracted as a plurality of still images. The timing for cutting out the still image may be specified by a person such as a service person or a user. Alternatively, the timing at which the control unit 80 cuts out a still image may be determined with reference to the sensor output of the weighing machine 10 or the packaging machine 20.

(4-2-3) Creation of Teacher Data Next, teacher data is created from a still image. The teacher data is created by attaching a label “normal” or “abnormal” to the still image. The label of “normal” or “abnormal” may be attached based on the result of a human visually confirming a still image of each package P. Alternatively, when the inspection unit 40 detects an abnormality, the control unit 80 may automatically label the still image of the package P corresponding to the abnormality as “abnormal”.

(4-2-4) Supervised learning The supervised learning is performed by the function approximator X using the supervised data thus obtained. Thereby, the function approximator X acquires the learned model M by, for example, the error backpropagation method.

(4-3) Inference Phase The inference phase is a usage scene by the user of the packaging and inspection system 100. A user operates the packaging and inspection system 100. The pre-inspection camera 31 photographs the package P discharged from the packaging machine 20 and transported by the pre-inspection conveyor 30. A still image portion of the package P in the output moving image of the pre-inspection camera 31 is input to the learned model M of the image determination unit 92. The learned model M outputs, for example, the probability that the package P is “normal” and the probability that it is “abnormal”. If the probability of being “abnormal” is greater than the probability of being “normal”, the image determination unit 92 determines that the package P is abnormal. At this time, the control unit 80 stops the weighing machine 10, the packaging machine 20, and the pre-inspection conveyor 30, and issues an alarm to the user.

(5) Features (5-1)
The pre-inspection conveyor 30 and the pre-inspection camera 31 are arranged in front of the inspection unit 40. Therefore, when the pre-inspection camera 31 and the image determination unit 92 detect that the abnormal package P has reached the pre-inspection conveyor 30, the pre-inspection conveyor 30 can be stopped. Therefore, the abnormal package P is quickly detected, and the inspection unit 40 can be prevented from being contaminated by the abnormal package P.

(5-2)
The inspection unit 40 has at least one of an X-ray inspection machine 41, a weight inspection machine 42, and a pressing seal inspection machine 43. Therefore, since the abnormal package P can be detected before reaching the inspection unit 40, the possibility that the X-ray inspection machine 41, the weight inspection machine 42, or the pressing seal inspection machine 43 needs to be cleaned can be reduced.

(5-3)
The image determination unit 92 has a learned model M formed by machine learning. Therefore, the accuracy of the abnormality determination can be improved by using the teacher data based on the image of the actual package P.

(5-4)
You can use deep learning as machine learning. Therefore, the accuracy of abnormality determination can be improved by increasing the amount of teacher data.

(5-5)
Various modes of abnormality of the package P can be determined by image recognition by machine learning.

(5-6)
The packaging and inspection system 100 includes an image storage unit 91. Therefore, the moving image accumulated in the image storage unit 91 can be used for learning of the image determination unit 92.

(5-7)
The packaging and inspection system 100 includes a boxing device 70. Therefore, at the time of creating the teacher data, by attaching an abnormal label to an image corresponding to the attitude of the package P that is difficult to handle by the box packing device 70, it is possible to suppress a problem in the box packing device 70.

(6) Modifications A plurality of modifications described below may be combined.

(6-1) Modification A
As the pre-inspection camera 31, for example, a three-dimensional camera such as a TOF camera may be adopted. The TOF camera can acquire information on the distance in the light propagation direction. When the three-dimensional camera is adopted as the pre-inspection camera 31, the output moving image includes information on the thickness of the package P.

  According to this configuration, the pre-inspection camera 31 obtains three-dimensional information of the package P. Therefore, useful information for determining the abnormality of the package P can be obtained.

(6-2) Modification B
FIG. 9 shows a packaging and inspection system 100A according to Modification B. The packaging and inspection system 100A is different from the packaging and inspection system 100 according to the above-described embodiment in that the packaging and inspection system 100A includes a sensor 32 that detects that the package P has passed a predetermined position of the pre-inspection conveyor 30. Any sensor can be used as the sensor 32, such as an optical sensor, an infrared sensor, and a capacitance sensor.

  According to this configuration, by using the output of the sensor 32, the packaging and inspection system 100A can easily cut out the portion related to the package P from the moving image captured by the pre-inspection camera 31.

(6-3) Modification C
In the above-described embodiment, the image storage unit 91 accumulates the moving image output from the pre-inspection camera 31. Instead, the image storage unit 91 may sequentially accumulate still images. The sensor output of the weighing machine 10 or the packaging machine 20 may be used as a trigger signal for acquiring a still image. Alternatively, the output of the sensor 32 in the modified example B may be used as a trigger signal.

  Further, the trigger signal may be input to the pre-inspection camera 31, and the pre-inspection camera 31 may output a still image instead of a moving image.

  According to these configurations, the capacity of the image storage unit 91 can be reduced. Alternatively, the image storage unit 91 can accumulate data on a larger number of packages P.

(6-4) Modification D
In the above-described embodiment, the image determination unit 92 performs learning and inference based on individual still images corresponding to each package P. Instead, the image determination unit 92 may perform learning and inference based on the moving image itself having a certain time length.

In this case, the function approximator X configuring the learned model M can infer the following abnormality of the package P.
-Inappropriate spacing between the bag and the next bag.
-Inappropriate timing at which the package P is carried.

  For this purpose, an image determination unit 92A shown in FIG. 10 may be used instead of the image determination unit 92 of the above-described embodiment. The image determining unit 92A includes a periodicity extracting unit 93 and a periodicity determining unit 94. The periodicity extracting unit 93 extracts information related to periodicity from a moving image having a certain length. The periodicity determination unit 94 determines the validity of the periodicity. That is, the periodicity determination unit 94 determines whether the information regarding the periodicity is within the assumed range.

  According to this configuration, the image determination unit 92A determines the periodicity of the moving image. Therefore, the abnormality of the package P can be determined using the relationship between the two continuously transported packages P.

(6-5) Modification E
In the above embodiment, an image such as a moving image or a still image is used as an input for the abnormality determination. Instead of this, not only the image but also the change over time of the weight output output from the weight checker 42 may be used as an input for abnormality determination. FIG. 11 shows a block diagram of an abnormality determination engine 84B according to this modification. The anomaly determination engine 84B receives a moving image, which is a temporal change in the image, from the pre-inspection camera 31 and a temporal change in the weight from the weight checker 42. Weight is input to one of the input perceptrons of the learned model configured in the function approximator X.

  According to this configuration, when an abnormal package P is found in the weight inspection machine 42, the packaging and inspection system 100 can be stopped. Therefore, the press-type seal inspection machine 43, the sorting unit 50, the box packing device 70, and the like are prevented from being contaminated by the abnormal package P.

10: weighing machine 20: packaging machine 30: pre-inspection conveyor 31: pre-inspection camera 32: sensor 40: inspection section 41: X-ray inspection machine 42: weight inspection machine 43: press-type seal inspection machine 70: box packing device 80: Control unit 91: Image storage unit 92: Image determination unit 92A: Image determination unit 93: Periodicity extraction unit 94: Periodicity determination unit 100: Packaging and inspection system 100A: Packaging and inspection system A: Article F: Film (packaging material) )
M: Learned model P: Package X: Function approximator

JP 2009-161228 A

Claims (11)

A packaging machine for making a package by packaging an article with a packaging material;
An inspection unit that inspects the package,
A conveyor that carries the package discharged from the packaging machine to the inspection unit,
A camera for photographing the package carried by the conveyor,
An image determination unit that determines the abnormality of the package by analyzing a still image or a moving image output from the camera,
Comprising,
Packaging and inspection system.   The packaging and inspection system according to claim 1, wherein the inspection unit includes at least one of an X-ray inspection machine, a weight inspection machine, and a press-type seal inspection machine. The image determination unit includes a computer having a learned model formed by machine learning,
The packaging and inspection system according to claim 1. The machine learning is deep learning,
The trained model is in the form of a neural network,
The packaging and inspection system according to claim 3. The state of the package that the image determination unit determines to be abnormal includes tearing of the bag formed by the packaging material, failure of sealing the packaging material, failure of cutting the packaging material, and size of the bag. Improper, scattering of the article out of the bag, improper spacing between the bag and the next bag, upside down of the package, improper posture of the package, timing when the package is carried Including at least one of:
The packaging and inspection system according to claim 1. An image storage unit that stores the still image or the moving image,
Further comprising,
The packaging and inspection system according to claim 1. A box packing device that packs the package discharged from the inspection unit into a packing box,
Further comprising,
The packaging and inspection system according to any one of claims 1 to 6. The camera includes a three-dimensional camera,
The still image or the moving image includes information on a thickness of the package,
The packaging and inspection system according to claim 1. A sensor for detecting that the package has passed a predetermined position of the conveyor,
Further comprising,
The packaging and inspection system according to claim 1. The image determination unit,
A periodicity extracting unit that extracts data related to the periodicity of the moving image;
A periodicity determining unit that determines the validity of the periodicity,
Having,
The packaging and inspection system according to claim 1. A packaging machine for making a package by packaging an article with a packaging material;
A conveyor for carrying the package discharged from the packaging machine,
A camera for photographing the package carried by the conveyor,
An image storage unit that stores a still image or a moving image output from the camera,
A function approximator for processing the still image or the moving image,
The step of preparing
The still image or the moving image generated in the image storage unit when the normal package is photographed by the camera is labeled with a label indicating normal, and the abnormal package is photographed by the camera. The function approximator to which the still image or the moving image generated in the image storage unit is attached with a label signifying an abnormality, and the teacher data including the teacher data performs machine learning. Forming a trained model,
The still image or the moving image, which is the output of the camera that captures the state in which the package discharged from the packaging machine is conveyed by the conveyor, is input to the learned model, and the result is output from the learned model. Based on, the step of performing an abnormality determination of the package,
A method for inspecting a package, comprising:

Images