JP7143388B2 - Inspection methods - Google Patents

Description

The present invention relates to a technology of an inspection method for inspecting a work.

2. Description of the Related Art Conventionally, techniques of inspection apparatuses for inspecting workpieces are publicly known. For example, it is as described in Patent Document 1.

A nut attachment presence/absence inspection device (inspection device) described in Patent Document 1 includes an arm portion of a general-purpose robot, a CCD camera, and a control portion. The arm of the general-purpose robot grips a press-molded product (workpiece) and holds it in a predetermined posture. A CCD camera images the pierce hole of the press-molded product. The control unit inspects whether or not the nut is attached to the pierce hole based on the imaging result of the CCD camera.

In the inspection apparatus as disclosed in Patent Document 1, there is a possibility that the position of the device may be shifted due to the operation of each part. If the positional deviation of the equipment occurs, it becomes difficult to perform a normal inspection (imaging, etc.), and as a result, the inspection cannot be performed normally. Therefore, a technique for detecting the positional deviation of the device is desired.

JP 2015-175652 A

SUMMARY OF THE INVENTION The present invention has been made in view of the circumstances as described above, and an object thereof is to provide an inspection method capable of detecting the positional deviation of equipment.

The problems to be solved by the present invention are as described above, and the means for solving the problems will now be described.

In claim 1 , a movement step of moving the workpiece to the imaging position by a moving unit configured by a multi-axis robot, and moving the workpiece to the imaging position by the moving unit, the hole formed in the workpiece is The imaging unit is inserted into the hole from one side and the other side of the hole, and in a state of being inserted into the hole, a predetermined imaging portion of the hole is captured from each of the one side and the other side. an inspection step of inspecting the work by determining whether the work is good or bad by a judgment unit based on the imaging result of the imaging unit; and imaging results of the plurality of works imaged by the imaging unit. and a detection step of detecting a positional deviation of the imaging unit and the workpiece based on.

As effects of the present invention, the following effects are obtained.

In claim 1 , it is possible to detect the positional deviation of the device.

BRIEF DESCRIPTION OF THE DRAWINGS The schematic plan view which showed the whole structure of the test|inspection apparatus which concerns on 1st embodiment. Similarly, a schematic front view. Block diagram as well. (a) The front view which showed the workpiece|work. (b) AA sectional view. (a) An enlarged plan view showing a lift section. (b) Enlarged side view showing a lift section. (a) A side view showing a multi-axis robot. (b) An enlarged view showing a grip portion. The front view which shows a misalignment correction|amendment part. The front view which shows an interference check part and an imaging part. Explanatory drawing which showed the deep neural network (henceforth "DNN" is called). 4 is a flow chart showing an inspection procedure; (a) A front view showing a state in which the workpiece has been moved to the first transfer position. (b) Similarly, a side view. (c) A plan view showing the workpiece and the first and second pins at the first transfer position. FIG. 4 is a cross-sectional plan view showing a state in which a workpiece is positioned; (a) A front view showing the work moved to the second transfer position. (b) A front view showing a state in which the deformation portion is moved above and below the workpiece. (c) A front view showing how the gripping portion grips a workpiece. (a) The figure which shows the result of having imaged the 1st hole with the camera of the misalignment correction|amendment part. (b) The figure which shows the result of having imaged the 1st hole after correct|amending a position shift. (a) A front view showing a state before an image of a workpiece is imaged by an imaging unit. (b) A cross-sectional view showing how an imaging unit captures an image of a workpiece. (a) A diagram showing a result of capturing an image of a workpiece before gain correction. (b) A diagram showing a result of capturing an image of a workpiece after gain correction. (a) The figure which shows the result of having imaged the 1st hole with the 1st borescope. (b) The figure which shows the result of having imaged the land part shown to Fig.17 (a) with the 2nd borescope. (a) The figure which shows the result of having imaged the defect. (b) The figure which shows the heat map in Fig.18 (a). The block diagram which showed the inspection apparatus which concerns on 2nd embodiment. (a) Sectional drawing which shows a mode that a 1st hole is imaged with a 1st borescope. (b) Explanatory diagram showing how the pass/fail judgment is performed by the pass/fail judgment DNN. (a) Explanatory diagram showing how a retest necessity DNN is learned. (b) Explanatory diagram showing how re-examination is determined by the re-examination necessity DNN. 4 is a flow chart showing an inspection procedure; (a) A diagram showing a heat map. (b) Sectional drawing which shows a mode that a 1st hole is imaged in reexamination. The block diagram which showed the inspection apparatus which concerns on 3rd embodiment. (a) Sectional drawing which shows a mode that a 1st hole is imaged with a 1st borescope. (b) Explanatory diagram showing a state of learning a positional deviation DNN. 4 is a flow chart showing an inspection procedure; (a) The figure which shows the result of having imaged the 1st hole with the 1st borescope when position shift generate|occur|produced. (b) A diagram showing the result of correcting the positional deviation. The block diagram which showed the inspection apparatus which concerns on 4th embodiment. (a) Sectional drawing which shows a mode that a 1st hole is imaged with a 1st borescope. (b) Explanatory diagram showing how the positional deviation determination DNN and the information estimation DNN are learned. 4 is a flowchart showing determination processing;

Hereinafter, the directions indicated by arrow U, arrow D, arrow F, arrow B, arrow L, and arrow R in the drawings are defined as upward, downward, forward, backward, leftward, and rightward, respectively. will be explained.

First, an example of a workpiece W to be inspected by the inspection apparatus 1 will be described below with reference to FIG. 4 .

The workpiece W is used, for example, as a housing for a valve device (not shown). The workpiece W has a body portion W10 and a flange portion W20. The body portion W10 is formed in a substantially rectangular parallelepiped shape. The body W10 includes a first hole W11, a second hole W12, a third hole W13, and a communication hole W14.

The first hole W11 is a hole in which a spool (not shown) of the valve device is provided. The first hole portion W11 extends through the body portion W10 in the front-rear direction. The first hole portion W11 is formed in a generally circular shape when viewed from the front. The first hole portion W11 is formed in the left-right central portion of the body portion W10. The inner diameters of the front and rear end portions of the first hole portion W11 are formed to be larger than the inner diameter of the front and rear halfway portion. A sliding portion W11a and a land portion W11b are formed in the first hole portion W11. The sliding portion W11a is a portion that can slide with respect to the spool. The sliding portion W11a is formed so as to have a constant diameter in the front-rear middle portion of the first hole portion W11. The land portion W11b is formed to have a larger diameter than the sliding portion W11a in the front-rear midway portion of the first hole portion W11.

The second hole W12 is a hole in which the spool is provided. The second hole portion W12 extends through the body portion W10 in the front-rear direction. The second hole W12 is formed in a generally circular shape in a front view having an inner diameter smaller than that of the first hole W11 (sliding portion W11a). The second hole W12 is formed to the left of the first hole W11. A sliding portion W12a and a land portion W12b are formed in the second hole portion W12, similarly to the first hole portion W11.

The third hole portion W13 is a hole in which the spool is provided. The third hole portion W13 penetrates the main body portion W10 in the front-rear direction. The third hole portion W13 is formed in a generally circular shape in a front view having an inner diameter that is approximately the same as that of the second hole portion W12. The third hole W13 is formed on the upper left side of the first hole W11. A sliding portion and a land portion are formed in the third hole portion W13 (not shown) in the same manner as in the first hole portion W11.

The communication hole W14 is a hole that communicates the first hole portion W11 and the outside of the work W with each other. The communication hole W14 is formed so as to extend in the left-right direction. A plurality (three) of the communication holes W14 are formed at intervals in the front-rear direction.

The flange portion W20 is formed in a substantially disc shape with its plate surface directed in the up-down direction. The flange portion W20 is formed below the body portion W10.

The inspection apparatus 1 detects the sliding portion with the spool, that is, the first hole W11, the second hole W12, and the third hole W13 (hereinafter referred to as "first hole W11, etc.") by the imaging unit 70 described later. ) is imaged to confirm (inspect) that there is no defect in the workpiece W. In this way, the first hole W11 and the like are holes (inspection holes) to be inspected by the inspection apparatus 1 . The inspection device 1 confirms, for example, burrs (excessive protrusions) and remaining foreign matter (whether shot balls or the like used for removing sand remain), etc., as defects occurring in the hole.

Next, the configuration of the inspection apparatus 1 according to the first embodiment will be described with reference to FIGS. 1 to 9. FIG. The inspection apparatus 1 is provided, for example, in a production line of works W or the like. As shown in FIG. 1, the inspection apparatus 1 includes a partitioning member 10, a conveying slider 20, an elevating slider 30, a multi-axis robot 40, a misalignment correction unit 50, an interference check unit 60, an imaging unit 70, a control unit 80, and a PC 90. equip.

The partitioning member 10 is a fence-like member that surrounds the devices (the transport slider 20 and the like) of the inspection apparatus 1 . A loading port 11 (opening) for loading the workpiece W from the outside to the inside of the partitioning member 10 is formed in the front left portion of the partitioning member 10 . The inlet 11 is provided with a table 12 (see FIG. 2) on which the workpiece W can be placed. The space inside the partition member 10 is hereinafter referred to as an "internal space 13".

The conveying slider 20 is for conveying the workpiece W from the input port 11 to an elevation slider 30 which will be described later. The conveying slider 20 is provided adjacent to the inlet 11 in the front left portion of the internal space 13 . As shown in FIGS. 1 and 2 , the transport slider 20 includes a transport conveyor 21 and a mounting section 22 .

The transport conveyor 21 is capable of moving a placement section 22, which will be described later. The transport conveyor 21 is provided so as to span between the inlet 11 of the partitioning member 10 and the elevation slider 30 .

The placing portion 22 is a portion for placing the work W thereon. The mounting section 22 is provided on the transport conveyor 21 and can move in the horizontal direction as the transport conveyor 21 is driven.

The transport slider 20 configured as described above can move the work W (mounting section 22) to the input position P1 and the first transfer position P2 by driving the transport conveyor 21. As shown in FIG. The input position P<b>1 is a position for receiving the work W input from the input port 11 on the placement section 22 . Specifically, the input position P1 is the position where the work W is closest to the input port 11 (moved to the left end of the conveyer 21). The first delivery position P2 is a position where the work W is delivered between the transport slider 20 and the lifting slider 30 . Specifically, the first transfer position P2 is the position where the work W is closest to the lifting slider 30 (moved to the right end of the conveyer 21).

The elevating slider 30 is for elevating the workpiece W. As shown in FIG. The elevating slider 30 is provided in the front right portion of the internal space 13 . As shown in FIGS. 2 and 5, the elevating slider 30 includes an elevating conveyor 31, a lift section 32, a first pin 33, a second pin 34, and a cylinder 35. As shown in FIG.

The elevating conveyor 31 is capable of elevating a lift section 32, which will be described later. The elevator conveyor 31 is fixed to the right end of the transport slider 20 (transport conveyor 21).

The lift portion 32 is a portion that moves up and down as the elevator conveyor 31 is driven. The right part of the lift part 32 is supported by the elevator conveyor 31 so that it can be raised and lowered. As shown in FIG. 5, the lift portion 32 is formed with a concave portion 32a.

The recessed portion 32 a is a portion formed in a recessed shape on the left side of the lift portion 32 . The recessed portion 32 a is formed in the left-right middle portion of the lift portion 32 . The concave portion 32a has a certain width in the left-right direction so that a grasping portion 45 of the multi-axis robot 40, which will be described later, can enter.

The first pin 33 is for determining the position of the workpiece W. As shown in FIG. The first pin 33 is arranged with its axial direction directed in the front-rear direction. A pair of front and rear first pins 33 are provided across the recess 32a. The tip of the first pin 33 is tapered so that it can be inserted into the first hole W11 of the work W (see FIG. 12). The first pin 33 is supported by the lift portion 32 via a cylinder 35 which will be described later. The first pin 33 is driven by the cylinder 35 and configured to move forward and backward with respect to the concave portion 32a in plan view.

The second pin 34 is for restricting the rotation of the work W. As shown in FIG. The second pin 34 is arranged on the right side of the concave portion 32a with its axial direction oriented in the left-right direction. The second pin 34 is formed to be insertable into the communication hole W14 of the work W (see FIG. 4). The second pin 34 is supported by the lift portion 32 via a cylinder (not shown). The second pin 34 is driven by the cylinder and configured to advance and retreat in the horizontal direction with respect to the concave portion 32a in plan view.

The cylinder 35 is for moving the first pin 33 in the left-right direction. The cylinders 35 are fixed to the front and rear portions of the lift portion 32, respectively. The front and rear cylinders 35 are configured to allow the pair of front and rear first pins 33 to move forward and backward.

The lifting conveyor 31 configured as described above can position the workpiece W by inserting the first pin 33 into the first hole W11 (see FIG. 12). Further, the lifting conveyor 31 can restrict the rotation of the workpiece W by inserting the second pin 34 into the communication hole W14. By raising and lowering the lift portion 32 in this state, the elevator conveyor 31 can raise and lower the workpiece W to the first transfer position P2 and the second transfer position P3 shown in FIG.

The second transfer position P3 is a position where the workpiece W is transferred between the lifting slider 30 and the multi-axis robot 40. As shown in FIG. Specifically, the second transfer position P3 is a position where the work W is lifted from the first transfer position P2 to the upper end portion of the elevator conveyor 31 . The space between the second transfer position P3 and the first transfer position P2 is set to be larger than the vertical width (height) of the workpiece W.

The multi-axis robot 40 is for moving the workpiece W and changing its posture. The multi-axis robot 40 is provided substantially in the center of the internal space 13 in plan view (see FIG. 1). As shown in FIGS. 1 and 6 , the multi-axis robot 40 includes a pedestal 41 , a swivel section 42 , an arm 43 , a joint section 44 and a grasping section 45 .

The pedestal portion 41 is a portion that supports a turning portion 42 and the like, which will be described later. The swivel part 42 is mounted on the base part 41, and is provided so that an arm 43, a grip part 45, etc., which will be described later, can swivel (rotate about a rotation axis pointing in the vertical direction). A plurality of arms 43 and joints 44 are provided between the revolving part 42 and the gripping part 45 . The joint part 44 is provided so as to connect the two arms 43, and is configured to be able to rotate the arm 43 and the grip part 45 around a predetermined rotation axis.

The gripping portion 45 is a portion that grips the workpiece W. As shown in FIG. The grip portion 45 includes a fixed member 45a, a movable member 45b and a deformation portion 45c.

The fixing member 45 a is a substantially plate-shaped member that is fixed to the end of the arm 43 . In the state shown in FIG. 6(b), the fixing member 45a is formed in a substantially rectangular shape in a front view with its longitudinal direction oriented vertically.

The movable member 45b is a member that is relatively movable in the vertical direction with respect to the fixed member 45a. A pair of upper and lower movable members 45b are provided. The pair of upper and lower movable members 45b can move toward and away from each other (up and down) by power from a predetermined drive source.

The deformable portion 45c is a deformable member. The deformation portions 45c are provided on the pair of upper and lower movable members 45b. The upper and lower deformation portions 45c are arranged so as to face each other. Two deformation portions 45c on the upper side are provided side by side. A large number of spheres (beads) are accommodated in the deformation portion 45c. The deformable portion 45c can be deformed along the shape of the object by appropriately moving the spherical body when it comes into contact with another member. Further, the deformable portion 45c can be evacuated (internal air is sucked) by a predetermined pump (not shown), thereby restricting the movement of the sphere and fixing its own shape.

The multi-axis robot 40 configured as described above can deform the deformable portion 45c along the shape of the work W by bringing the pair of upper and lower movable members 45b close to each other and pressing the deformable portion 45c against the work W. can. The multi-axis robot 40 can grip the workpiece W by vacuuming the deformable portion 45c in this state (see FIG. 7). In this state, the multi-axis robot 40 can move the workpiece W and change its posture by rotating the turning section 42 and the joint section 44 . In addition, the multi-axis robot 40 can release the workpiece W gripped by the gripping unit 45 (release the grip) by performing an operation opposite to gripping the workpiece W.

The misalignment correction unit 50 is for calculating the positional misalignment between the imaging unit 70 and the work W (the first hole W11, etc.) and correcting the robot coordinates. The misalignment correction unit 50 is provided on the left side of the multi-axis robot 40 . As shown in FIGS. 1 and 7 , the misalignment correction section 50 includes a camera 51 , an upper illumination 52 , a lower illumination 53 and a processing section 54 .

The camera 51 images the workpiece W from above. The upper lighting 52 irradiates the work W with light from above. A pair of left and right upper illuminations 52 are provided with the camera 51 interposed therebetween. The lower lighting 53 irradiates the work W with light from below. A lower illumination 53 is arranged below the camera 51 .

The processing unit 54 performs calculation processing of positional deviation. The processing unit 54 is connected to the camera 51 . The imaging result B51 of the camera 51 (see FIG. 14A) is input to the processing unit 54 . The processing unit 54 can output the positional deviation to the control unit 80 by performing arithmetic processing on the input imaging result B51. Note that the processing of the processing unit 54 will be described later.

The interference check unit 60 is a portion for confirming that the first hole portion W11 of the workpiece W and the like do not interfere (contact) with the imaging unit 70, which will be described later. The interference check unit 60 is provided on the rear left side of the multi-axis robot 40 . As shown in FIG. 8 , the interference check section 60 includes a mounting member 61 , a shaft member 62 and a sensor section 63 .

The attachment member 61 is a substantially box-shaped member attached to the longitudinal member A<b>1 provided inside the partition member 10 . The attachment member 61 (interference checker 60) is fixed to the longitudinal member A1 and provided immovably.

The shaft member 62 is a member imitating an imaging unit 70 (insertion portions 71b, 72b, and 73b), which will be described later. The shaft member 62 is formed in substantially the same shape (substantially the same outer diameter and length) as the imaging section 70 and is provided so as to assume substantially the same posture as the imaging section 70 . Specifically, the shaft member 62 is formed in a substantially columnar shape with the axial direction directed substantially vertically. The outer diameter of the shaft member 62 is formed to be smaller than the inner diameter of the first hole W11 of the workpiece W and the like. The shaft member 62 is fixed to the mounting member 61 so as to protrude downward from the mounting member 61 .

The sensor unit 63 is a sensor (for example, a proximity switch or the like) that detects that the shaft member 62 is in contact with the first hole W11 or the like of the work W. As shown in FIG. The sensor section 63 is provided inside the mounting member 61 .

The imaging unit 70 is for imaging the first hole W11 of the work W and the like. As shown in FIGS. 1 and 8 , the imaging unit 70 includes a first borescope 71 , a second borescope 72 and a third borescope 73 .

Note that the first borescope 71 to the third borescope 73 are configured in the same manner as each other, except that the arrangement and imaging directions D71 to D73 are different. Therefore, the configuration of the first borescope 71 will be described below as an example, and the differences between the second borescope 72 and the third borescope 73 will be mainly described with respect to the first borescope 71 .

The first borescope 71 is a device capable of imaging the inside of the first hole W11 and the like. The first borescope 71 is provided on the right side of the interference check section 60 . The first borescope 71 includes an attachment portion 71a, an insertion portion 71b, a light source 71c, and a camera 71d.

The attachment portion 71a is a substantially box-shaped member attached to the longitudinal member A1. The mounting portion 71a (the first borescope 71) is fixed to the longitudinal member A1 and provided immovably.

The insertion portion 71b is a substantially cylindrical portion that can be inserted into the first hole portion W11 or the like. The insertion portion 71b is arranged with its axial direction directed substantially vertically. The insertion portion 71b is fixed to the mounting portion 71a so as to protrude downward from the mounting portion 71a. Thus, the insertion portion 71b is provided so as to hang down from above. The outer diameter of the insertion portion 71b is formed so as to be smaller than the inner diameter of the first hole portion W11 of the workpiece W and the like. The outer diameter of the insertion portion 71b is formed to be slightly smaller than the outer diameter of the shaft member 62 of the interference check portion 60. As shown in FIG.

The light source 71c is for irradiating light (illumination) for illuminating the inside of the first hole W11 and the like. The light source 71c is provided inside the mounting portion 71a. The light emitted from the light source 71c is emitted from the lower end portion of the insertion portion 71b to the outside (downward, the same direction as the imaging direction D71 of the camera 71d described later). The light source 71c is configured to change the brightness of light.

The camera 71d is for imaging the inside of the first hole W11 and the like. The camera 71d is provided inside the mounting portion 71a. The camera 71d can capture an image downward (imaging direction D71) from the distal end (lower end) of the insertion portion 71b through the lens in the insertion portion 71b. In this way, the camera 71d captures an image in the imaging direction D71 parallel to the axial direction of the insertion portion 71b, and can obtain an image of the first hole portion W11 and the like viewed straight (directly viewed).

The second borescope 72 includes a mounting portion 72a, an insertion portion 72b, a light source 72c and a camera 72d. The second borescope 72 is arranged to the right of the first borescope 71 . The camera 72d can take an image in an oblique direction (imaging direction D72) from the tip of the insertion portion 72b via the lens in the insertion portion 72b. In this way, the camera 72d captures an image in an imaging direction D72 inclined with respect to the axial direction and the horizontal direction (direction perpendicular to the axial direction) of the insertion portion 72b, and obtains an image of the first hole W11 and the like viewed obliquely. can be done.

The third borescope 73 includes an attachment portion 73a, an insertion portion 73b, a light source 73c and a camera 73d. The third borescope 73 is arranged on the right front side of the first borescope 71 . The camera 73d can take an image in the horizontal direction (imaging direction D73) from the tip of the insertion portion 73b via the lens in the insertion portion 73b. Thus, the camera 73d captures an image in the imaging direction D73 perpendicular to the axial direction of the insertion portion 73b, and can obtain an image of the first hole W11 and the like viewed from the side.

The control unit 80 controls the equipment (the transport slider 20 and the like) of the inspection apparatus 1 . The control unit 80 is provided on the outside (right rear side) of the partition member 10 . As shown in FIG. 3 , the controller 80 is connected to the transport slider 20 , the lift slider 30 , the multi-axis robot 40 , the misalignment corrector 50 and the interference checker 60 .

By transmitting a signal to the transport slider 20, the control unit 80 can control the start and stop of the operation of the transport conveyor 21, the moving direction of the workpiece W (mounting unit 22), and the like.

The control unit 80 sends a signal to the lifting slider 30 to start and stop the operation of the lifting conveyor 31, the lifting direction of the lift unit 32, the operation of the cylinder 35 (the first pin 33 and the second pin 34), and the like. can be controlled.

By transmitting a signal to the multi-axis robot 40, the control unit 80 grips and releases the workpiece W described above, rotates the revolving unit 42 and the joint unit 44, and moves the workpiece W (gripping unit 45). can move and change posture.

By transmitting a signal to the misalignment correction unit 50 , the control unit 80 can perform imaging by the camera 51 and irradiation and stop of light irradiation from the upper illumination 52 and the lower illumination 53 . Further, the control unit 80 can acquire the processing result of the processing unit 54 by receiving a signal from the processing unit 54 .

The control unit 80 can acquire the detection result of the sensor unit 63 by receiving the signal from the interference check unit 60 .

The PC 90 is for determining the quality of the first hole portion W11 and the like based on the imaging results B71 and B72 (see FIG. 17) of the imaging section 70. FIG. The PC 90 is provided outside (left side) of the partition member 10 (see FIG. 1). The PC 90 includes an arithmetic unit such as a CPU/GPU, a storage device such as an HDD, and a display device such as a liquid crystal display. The PC 90 also includes a corrector 91, a pass/fail judgment section 92, and a creation section 93 as functions (programs) relating to pass/fail judgment.

The correction section 91 is for correcting the gain (sensitivity) of the imaging section 70 . The quality determination section 92 is for determining the quality of the work W. As shown in FIG. The processing of the correction unit 91 and the quality determination unit 92 will be described later.

The creating unit 93 is for creating a heat map H (see FIG. 18(b)) indicating the degree of reaction of the DNN 100, which will be described later.

A DNN 100 is constructed in the PC 90 configured as described above, as shown in FIG.

The DNN 100 is an information processing model that mimics the structure of neural circuits in the human brain. More specifically, the DNN 100 connects units 101a, 102a, and 103a that model brain nerve cells to each other, and performs various information processing by exchanging signals with the units 101a and the like. The DNN 100 comprises an input layer 101 , an intermediate layer (hidden layer) 102 and an output layer 103 .

The input layer 101 has a plurality of units 101a. Unit 101 a of input layer 101 is connected to unit 102 a of intermediate layer 102 . The unit 101a of the input layer 101 can perform arithmetic processing on input information and transfer the information to the intermediate layer 102 (unit 102a).

The intermediate layer 102 has a plurality of units 102a. A plurality of intermediate layers 102 are provided between the input layer 101 and the output layer 103 . The unit 102a of the intermediate layer 102 performs arithmetic processing on information passed from the input side (adjacent unit on the input layer 101 side) and outputs the information to the output side (adjacent unit on the output layer 103 side). can pass.

The output layer 103 has a plurality of units 103a. The output layer 103 can perform arithmetic processing on the information passed from the intermediate layer 102 and output the result.

The DNN 100 configured as described above learns in advance the relationship between the result of imaging the first hole W11 and the like of the work W by the imaging unit 70 and the result of quality judgment by supervised learning.

More specifically, in the DNN 100, the result of imaging the workpiece W to be determined as non-defective by the imaging unit 70 and the result of the non-defective determination that the product is non-defective (answer) are input to the input layer 101. 102 and the output layer 103 are passed. In this way, the DNN 100 learns the characteristics of the first hole W11 and the like for determining non-defective products. In addition, the DNN 100 also learns the works W to be determined as defective in the same manner as for the non-defective products.

The PC 90 configured as described above is connected to the imaging section 70 and the control section 80 as shown in FIG.

The PC 90 can acquire the imaging results B71 and B72 of the imaging section 70 by receiving the signal from the imaging section 70 . By inputting the imaging results B71 and B72 into the input layer 101, the DNN 100 generates a score indicating the probability of a good product or a defective product (for example, the probability of a good product is 80% and the probability of a defective product is 20%). ) and the result of pass/fail judgment (for example, the result of OK or NG) can be output.

The PC 90 can communicate with the control unit 80 to exchange control-related information (for example, positional information of the work W and equipment, etc.).

1, 2 and 10, an overview of the inspection of the workpiece W by the inspection apparatus 1 will be described below.

First, the workpiece W is set in the inspection device 1 (input position P1) (step S100).

When the work W is set, the transport slider 20 transports the work W to the first transfer position P2 (step S110).

When the transportation of the work W is completed, the lifting slider 30 positions the work W at the first transfer position P2 (step S120).

When the positioning of the work W is completed, the work W is transferred from the lifting slider 30 to the multi-axis robot 40 at the second transfer position P3 (step S130).

When delivery of the work W is completed, the multi-axis robot 40 moves the work W to a place where the misalignment correction unit 50 is provided. After that, the misalignment correction unit 50 corrects the robot coordinates based on the misalignment between the first hole W11 and the like and the imaging unit 70 (step S140).

When the correction of the robot coordinates is completed, the multi-axis robot 40 moves the workpiece W to a place where the interference check section 60 is provided. After that, the interference check unit 60 confirms in advance that the first hole W11 and the like do not interfere with the imaging unit 70 (step S150).

After confirming the interference, the multi-axis robot 40 moves the workpiece W to the place where the first borescope 71 is provided. After that, the correction unit 91 corrects the gain (sensitivity) of the camera 71d (step S160).

When the gain correction is completed, the first borescope 71 images the first hole W11 and the like with the corrected gain (step S170). A plurality of imaging locations to be imaged by the first borescope 71 are set in advance. The first borescope 71 captures an image of the imaging location with the insertion portion 71b inserted into the first hole W11 or the like (see FIG. 15(b)).

When the imaging by the first borescope 71 is completed, the multi-axis robot 40 moves the workpiece W to a place where the second borescope 72 is provided. The second borescope 72 images the first hole W11 and the like with the corrected gain (step S180). A plurality of imaging locations to be imaged by the second borescope 72 are set in advance. The second borescope 72 captures an image of the imaging location with the insertion portion 72b inserted into the first hole W11 or the like (see FIG. 15(b)).

When the imaging by the second borescope 72 is completed, the multi-axis robot 40, the lifting slider 30 and the transport slider 20 return the workpiece W to the loading position P1 (step S190).

In addition, the pass/fail judgment unit 92 of the PC 90 judges the pass/fail of the work W in parallel with the operation of returning the work W (step S200). Thus, the inspection by the inspection device 1 is completed. It should be noted that the quality determination need not be performed in parallel with the operation of returning the workpiece W, and can be performed at any timing after imaging the first hole portion W11 and the like.

Details of each of the steps S100 to S190 of the inspection described above will be described below.

First, step S100 (setting the workpiece W) will be described with reference to FIGS. 2 and 4. FIG. In step S100, the workpiece W is loaded into the partitioning member 10 through the loading port 11, and placed on the loading section 22 at the loading position P1 (the left end of the conveyor 21). At this time, the workpiece W is placed by a predetermined positioning member provided on the placement portion 22 in such a posture that the communication hole W14 faces right and the first hole portion W11 and the like face the front-rear direction. In this way, the placement section 22 receives the work W loaded from the loading port 11 at the loading position P1.

Next, step S110 (conveying the work W) will be described with reference to FIGS. 1, 2 and 11. FIG. 11 and subsequent drawings, the cross section of the workpiece W is appropriately simplified for convenience of explanation.

In step S110, the transport conveyor 21 is driven by a signal from the control unit 80, and the transport slider 20 moves the placement unit 22 to the right. In this way, as shown in FIGS. 1 and 2, the transport slider 20 transports the workpiece W from the loading position P1 to the first transfer position P2. At this time, the elevator conveyor 31 is driven by a signal from the control section 80, and the lift section 32 is arranged at the same height position as the workpiece W, as shown in FIG. 11(a). Thereby, the first pin 33 is arranged so as to face the first hole W11 of the work W, as shown in FIGS. 11(b) and 11(c). Moreover, the second pin 34 is arranged so as to face the communication hole W14. In this way, at the first transfer position P2, the first pin 33 and the second pin 34 can be inserted into the first hole portion W11 and the communication hole W14.

Next, step S120 (positioning of work W) will be described with reference to FIG. In step S120, the cylinder 35 is driven by a signal from the controller 80, and the tip of the first pin 33 (tapered portion, hereinafter referred to as "tapered portion") is inserted into the first hole W11. More specifically, the tip of the tapered portion is inserted into the sliding portion W11a. Further, the middle portion of the tapered portion abuts on the end portion of the sliding portion W11a during insertion. Thus, the first pin 33 moves the work W so as to follow the shape of the tapered portion, and positions the work W at a predetermined position. Also, the second pin 34 is inserted into the communication hole W14 to restrict the rotation of the work W. As shown in FIG.

Here, as described above, the workpiece W is manufactured by casting. The first hole W11 into which the first pin 33 is inserted is formed by a core during casting. Thus, the first hole portion W11 formed by the core has relatively high processing accuracy in the workpiece W. As shown in FIG. By positioning the work W at such a portion (the first hole portion W11) with high machining accuracy, the elevation slider 30 can position the work W with high accuracy.

Next, step S130 (delivery) will be described with reference to FIGS. 2 and 13. FIG. In step S130, under the control of the control unit 80, the work W is lifted, gripped by the gripping unit 45, and pulled out by the first pin 33, etc., and the work W is transferred from the elevation slider 30 to the multi-axis robot 40. .

More specifically, as shown in FIGS. 2 and 13(a), the elevator conveyor 31 is first driven in step S130, and the elevator slider 30 moves the positioned work W from the first transfer position P2 to the second transfer position P3. up to The amount of elevation at this time is set larger than the vertical width (height) of the workpiece W. As shown in FIG. Thus, the lifting slider 30 lifts the work W to a position higher than the vertical width of the work W (second transfer position P3).

Next, as shown in FIG. 13B, the multi-axis robot 40 causes the gripping portion 45 to enter the concave portion 32a of the lift portion 32 so that the deformation portion 45c and the workpiece W face each other. Thereafter, as shown in FIG. 13(c), the multi-axis robot 40 causes the pair of upper and lower movable members 45b to approach each other, and presses the deformable portion 45c against the workpiece W from above and below. The multi-axis robot 40 grips the workpiece W by vacuuming (fixing the shape of) the deformed portion 45c deformed along the contour of the workpiece W in this way. Further, the position of the pair of upper and lower movable members 45b is changed from the pressed state to the fixed state.

Next, the first pin 33 and the second pin 34 are pulled out from the first hole W11 and the communication hole W14 (see FIG. 11(c)). Thus, the workpiece W is transferred from the lifting slider 30 to the multi-axis robot 40 at the second transfer position P3.

As described above, in step S130, the multi-axis robot 40 grips the workpiece W by deforming the deformation portion 45c. As a result, the deformed portion 45c can be brought into close contact with the work W having a complicated shape or a work W having variations in outer shape, and the work W can be reliably gripped. Further, even if a foreign substance adheres to the work W, the deformation portion 45c can be deformed along the foreign substance to prevent the multi-axis robot 40 and the work W from being out of position.

Next, step S140 (correction of robot coordinates) will be described with reference to FIGS. 7 and 14. FIG.

As described above, the first borescope 71 and the second borescope 72 perform imaging with the insertion portions 71b and 72b inserted into the first hole portion W11 and the like (see FIG. 15(b)). For this reason, the positions of the first bore scope 71 and the second bore scope 72 and the first hole W11 and the like may be misaligned, or the orientation of the first hole W11 and the like may be in the vertical direction (the axial direction of the insertion portion 71b etc.). On the other hand, if it is tilted, there is a possibility that it will be difficult to determine whether it is good or bad in step S190. Correction by the misalignment correction unit 50 (step S140) is performed in order to avoid problems caused by misalignment (center misalignment) between the insertion portions 71b and 72b and the first hole W11 and the like. A specific description will be given below.

In step S140, as shown in FIG. 7, the multi-axis robot 40 changes the posture of the work W so that the first hole W11 and the like face up and down, and the camera 51 of the misalignment correction unit 50 and the lower illumination 53 The workpiece W is moved between In this way, the first hole W11 is arranged vertically below the camera 51 . The upper illumination 52 and the lower illumination 53 emit light toward the first hole W11. The camera 51 takes an image of the illuminated first hole portion W11, and transmits an imaging result B51 shown in FIG. In addition, FIG. 14A shows an example of the imaging result B51 transmitted to the processing unit 54. As shown in FIG.

The processing unit 54 analyzes the imaging result B51, and coordinates the center C11 of the front side (closer to the camera 51) end C10 of the first hole W11 and Side) coordinates of the center C21 of the end C20 are calculated. Then, the processing unit 54 calculates the distance between the center P of the imaging result B51 (image data) and the centers C11 and C21 of the first hole W11 based on the coordinate calculation result. The processing unit 54 also calculates the direction (leftward in FIG. 14A) in which the center C21 of the far side end C20 is shifted from the center C11 of the front side end C10.

Based on the distance and direction calculated in this way, the processing unit 54 adjusts the center P of the imaging result B51 so that the centers C11 and C21 of the first hole W11 coincide (imaging after correction shown in FIG. 14(b) Result B51), the multi-axis robot 40 is controlled. At this time, the processing unit 54 calculates a rotation angle such that the center C11 of the front end C10 and the center C21 of the back end C20 are aligned, and outputs the calculation result to the control unit 80 . The control unit 80 rotates the workpiece W (controls the multi-axis robot 40) based on the calculation result, thereby correcting the inclination (angle) of the first hole W11 with respect to the vertical direction. In addition, the processing unit 54 calculates the direction and amount of movement such that the center C11 of the front end C10 coincides with the center P of the imaging result B51, and outputs the calculation result to the control unit 80 . The control unit 80 moves the workpiece W in the horizontal direction based on the calculation result, thereby correcting the horizontal displacement of the first hole portion W11. Thus, the correction of the robot coordinates by the misalignment correction unit 50 is completed.

In this way, according to the misalignment correction unit 50, the workpiece W is in a determined position (a position where the centers P, C11, and C21 match) and a determined posture (a posture in which the first hole W11 faces the vertical direction). Thus, the multi-axis robot 40 can be corrected. In this way, the misalignment correction unit 50 corrects the position of the work W (corrects the movement of the work W) before moving the work W from the lifting slider 30 to the place where the first borescope 71 is provided. Therefore, positional deviation of the first hole portion W11 and the like with respect to the insertion portions 71b and 72b can be suppressed.

Next, step S150 (interference check) will be described with reference to FIG. The interference check unit 60 uses the shaft member 62 to check in advance that the insertion portions 71b and 72b do not come into contact with the first hole portion W11 and the like in the imaging of the first borescope 71 and the like (see FIG. 15B). Confirm. A specific description will be given below.

In step S150, the multi-axis robot 40 moves the workpiece W, and inserts the shaft member 62 of the interference checker 60 shown in FIG. 8 into the first hole W11. At this time, the multi-axis robot 40 moves the workpiece W in the same manner as the imaging by the first borescope 71 (step S170). Specifically, the multi-axis robot 40 moves upward the workpiece W whose first hole W11 faces in the vertical direction, and inserts the shaft member 62 into the first hole W11.

If the shaft member 62 and the first hole W<b>11 contact each other, the sensor unit 63 detects the contact and transmits the detection result to the control unit 80 . In this case, the control unit 80 suspends the inspection. On the other hand, if the shaft member 62 and the first hole portion W11 do not come into contact with each other, the sensor portion 63 will not detect the contact, so the inspection will not be interrupted. When the sensor portion 63 does not detect contact in this manner, the shaft member 62 is inserted into the second hole portion W12 and the third hole portion W13 in this order. The interference check unit 60 thus confirms that the first hole W11 and the like do not come into contact with the shaft member 62 imitating the insertion portions 71b and 72b.

Thus, according to the interference check unit 60, it is possible to confirm in advance (before imaging by the imaging unit 70) that the first hole W11 and the like do not interfere with the insertion portions 71b and 72b. Thereby, contact between the first hole W11 and the like and the insertion portions 71b and 72b can be reliably prevented. Further, the outer diameter of the shaft member 62 is formed to be larger than the outer diameters of the insertion portions 71b and 72b. With this configuration, it is possible to detect a workpiece W approaching the insertion portions 71b and 72b more than necessary, and more reliably prevent contact between the first hole portion W11 and the like and the insertion portions 71b and 72b.

Next, step S160 (gain correction) will be described with reference to FIGS. 15 and 16. FIG.

As will be described later, the first borescope 71 and the second borescope 72 emit light from the light source 71c when imaging the first hole W11 and the like. The degree of reflection of the light (degree of specular reflection) may differ for each workpiece W depending on the surface (casting surface) condition of the workpiece W and the like. Therefore, for example, as shown in an example in FIG. 16A, when a work W having a high degree of light reflection is imaged by the first borescope 71, the work W (the first hole W11, etc.) is bright in the imaging result B71. It will be photographed. On the other hand, when an image of a workpiece W having a low degree of light reflection is captured, the workpiece W appears dark.

Therefore, the correction unit 91 of the PC 90 corrects the gain of the camera 71d for each work W in advance, so that an image with a constant appearance (an image with a constant brightness) is acquired regardless of which work W is captured. I am making it possible. A specific description will be given below.

In step S160, the first borescope 71 images the first imaging location. That is, as shown in FIG. 15, the multi-axis robot 40 changes the posture of the work W so that the first hole W11 and the like face the vertical direction, and moves the work W from below the first borescope 71 to above. move. In this way, the multi-axis robot 40 inserts the insertion portion 71b of the first borescope 71 into the first hole W11, and moves the workpiece W to a position (imaging position) where the first imaging location can be imaged by the first borescope 71. Let

In this state, the first borescope 71 emits light downward from the light source 71c, and the camera 71d captures an image of the first imaging location. The imaging result B71 (imaging result before gain correction shown in FIG. 16A) imaged in this way is input to the correction unit 91 .

The correction unit 91 analyzes the input imaging result B71, and calculates the brightness in a predetermined range R1 (range between two circles indicated by dashed lines in FIG. 16(a)). Then, the correction unit 91 compares the luminance calculation result with the comparison value (the luminance of the reference image) stored in the storage device, and calculates a correction coefficient based on the comparison result. Using the correction coefficient (by obtaining the product of the gain value and the correction coefficient, etc.), the correction unit 91 adjusts the brightness of the camera 71d so that the first hole W11 and the like can be imaged with a luminance close to the comparison value. Correct the gain.

An imaging result B71 before correction shown in FIG. 16A is an imaging result of the first hole portion W11 having a higher luminance (brighter image) than the comparison value. Taking the imaging result B71 of FIG. 16A as an example, the correction unit 91 determines that the luminance in the range R1 of the imaging result B71 is higher than the comparison value, and reduces the gain of the camera 71d. An imaging result B71 shown in FIG. 16B is the result of imaging the first hole portion W11 with the gain lowered in this way. As shown in FIG. 16B, by lowering the gain, it is possible to image the first hole W11 so as to appear dark.

In this manner, the correction unit 91 adjusts the gain from the imaging result of the first imaging location. Based on the corrected gain, the first borescope 71 and the second borescope 72 capture images of all inspection locations, and capture results used for pass/fail determination are obtained (steps S170 and S180).

In this way, according to the correction unit 91, the gain is corrected for each work W according to the surface (casting surface) state (degree of reflection of light) of the work W, and the brightness of the first hole W11 and the like is uniform. can be improved. This makes it easier to confirm the first hole portion W11 and the like in the imaging result B71, so that quality determination can be performed with high accuracy. In addition, when the DNN 100 performs learning, it is not necessary to prepare a large amount of imaging results (learning data) with different brightnesses, so the burden of learning work can be reduced.

Next, step S170 (imaging with the first borescope 71) will be described with reference to FIGS. 15 and 17(a).

As described above, in step S160, the insertion portion 71b of the first borescope 71 is inserted into the first hole portion W11 by movement of the workpiece W by the multi-axis robot 40, so that the first imaging location can be imaged. Therefore, in step S170, first, the first imaging location is imaged by the first borescope 71 with the corrected gain.

More specifically, the first borescope 71 irradiates light downward from the light source 71c, and images the inside of the first hole W11 (first imaging location) with the camera 71d. At this time, the camera 71d captures an image in the imaging direction D71, that is, in the axial direction of the insertion portion 71b. In this way, the first borescope 71 captures an image of the first hole W21 so as to look straight at the first hole W11 (see directly). The first borescope 71 takes images of the first imaging location multiple times while changing the brightness of the light to be irradiated.

By changing the brightness of the light at the same imaging position, the first borescope 71 has a plurality of different images, such as an image in which the front side of the first hole W11 is clearly captured and an image in which the back side is clearly captured. images can be acquired at the same imaging location. Accordingly, a wide range of the first hole portion W11 can be confirmed by imaging from the same imaging location.

When the first borescope 71 finishes imaging the first inspection location, the multi-axis robot 40 moves the workpiece W vertically, and the first borescope 71 captures the next inspection location. At this time, the image is captured a plurality of times while changing the brightness of the irradiating light. Such movement of the workpiece W and imaging of the first borescope 71 are repeated, and all the imaging locations of the first hole portion W11 are imaged.

In step S170, when imaging of all the imaging locations of the first hole W11 is completed, the multi-axis robot 40 moves the workpiece W downward and extracts the first borescope 71 from the insertion section 71b. After that, the multi-axis robot 40 turns the workpiece W upside down and inserts the insertion portion 71b from the opposite side of the first hole portion W11. In this way, the first borescope 71 changes the direction of insertion into the first hole W11 and images all imaging locations again.

In step S170, when the imaging is completed, the work W is moved by the multi-axis robot 40, and the insertion portion 71b is inserted into the second hole portion W12 and the third hole portion W13 in order. The first borescope 71 sequentially captures images of the second hole W12 and the third hole W13 in the same manner as the first hole W11. In this way, the image pickup results of all inspection locations imaged by the first borescope 71 (see image pickup result B71 shown as an example in FIG. 17A) are transmitted to the PC90.

Next, step S180 (imaging with the second borescope 72) will be described with reference to FIGS. 15 and 17(b).

In step S<b>180 , the multi-axis robot 40 moves the workpiece W upward in the same manner as in step S<b>170 and inserts the first hole portion W<b>11 into the insertion portion 72 b of the second borescope 72 .

In this state, the second borescope 72 irradiates light obliquely from the light source 71c, and images the inside of the first hole W11 (imaging portion) with the camera 72d. In this way, the second borescope 72 images the first hole portion W11 in the imaging direction D72, that is, as viewed obliquely. The second borescope 72 takes multiple images at the same imaging location while changing the brightness of the light irradiating such imaging. After that, the multi-axis robot 40 rotates the workpiece W by a predetermined angle around the axis L72 of the insertion portion 72b (see rotation direction R72 shown in FIG. The portion W11 is moved in the circumferential direction. By rotating the work W about the axis L72 in this way, the multi-axis robot 40 can rotate the work W while the insertion portion 72b is inserted into the first hole W11 or the like, and the second borescope can be rotated. Imaging by 72 can be quickly performed. The second borescope 72 takes multiple images of the thus rotated first hole portion W11 while changing the brightness of the light.

In step S180, the second borescope 72 images the entire periphery of the imaging location of the first hole W11 by such rotation and imaging. After that, the multi-axis robot 40 moves the workpiece W vertically, and the second borescope 72 repeats the operation of capturing an image of the entire circumference of the next inspection location.

In step S180, when the imaging of all imaging locations of the first hole portion W11 is completed, the second borescope 72 changes the direction of insertion into the first hole portion W11 in the same manner as in step S170. All imaging locations of are imaged again.

After completing the imaging, the second borescope 72 images the second hole portion W12 and the third hole portion W13 in the same manner as the first hole portion W11. In this way, the image pickup results of all inspection points imaged by the second borescope 72 (see image pickup result B72 shown as an example in FIG. 17(b)) are transmitted to the PC90.

Next, step S190 (operation of returning the workpiece W) will be described with reference to FIGS. 1 and 2. FIG.

In step S190, the multi-axis robot 40 moves the workpiece W that has been imaged by the second borescope 72 to the second delivery position P3, releases the gripping of the gripping part 45, and delivers the workpiece W to the elevation slider 30. . The lifting slider 30 lowers the work W to the first transfer position P2 and transfers it to the conveying slider 20 . The transport slider 20 transports the work W from the first transfer position P2 to the input position P1. Thus, the work W is returned to the input port 11 .

Next, with reference to FIGS. 9, 17 and 18, step S200 (determining the quality of the work W) will be described. In step S200, the pass/fail determination unit 92 inputs the imaging results B71 and B72 captured in steps S170 and S180 to the input layer 101 of the DNN 100. FIG. The DNN 100 extracts features of the work W from the input imaging results B71 and B72 and determines whether the work W is good or bad.

For example, in step S190, as shown in FIG. 17, if no defects are captured in the imaging results B71 and B72, the DNN 100 receives a score indicating that the workpiece W is likely to be a non-defective product and the result of the quality determination. Output that it is OK.

On the other hand, in step S190, for example, when a burr B71a (defective) is captured in the imaging result B71, the DNN 100 extracts the burr B71a as a feature of the workpiece W, as shown in FIG. Then, the DNN 100 outputs a score indicating that the work W is highly likely to be defective based on the extraction result, and outputs that the quality determination result is NG.

Also, in step S190, the creating unit 93 of the PC 90 creates a heat map H (see FIG. 18B) showing the degree of reaction of the DNN 100. FIG. The heat map H visualizes which part of the imaging results B71 and B72 the DNN 100 focuses on (which part is extracted as a feature of the work W) and how much the focused part affects the determination result. It is something to do. In the heat map H, the part of interest is colored with respect to the imaging results B71 and B72. Also, in the heat map H, different colors are added according to the degree of influence given to the determination result. After the heat map H is created, the quality judgment of the work W is completed.

As described above, in the inspection apparatus 1, the workpiece W positioned by the first pins 33 is gripped by the gripper 45 when the workpiece W is transferred to the multi-axis robot 40 (step S130) (FIGS. See Figure 13). As a result, it is possible to suppress displacement of the gripping position of the workpiece W by the gripping portion 45 . By moving the workpiece W to the imaging section 70, positional deviation between the imaging section 70 (the first borescope 71 and the like) and the first hole portion W11 and the like can be suppressed. As a result, the pass/fail determination section 92 can perform a pass/fail determination with high accuracy.

Further, in the first embodiment, the first hole portion W11 and the like are imaged by moving the workpiece W instead of the imaging unit 70 (steps S170 and S180). According to this, it is possible to prevent the occurrence of troubles (for example, failure of the camera 71d, etc.) associated with the movement of the imaging unit 70 .

Further, in steps S170 and S180, imaging is performed by the first borescope 71 and the second borescope 72 with the first hole W11 and the like directed vertically (see FIG. 15). As a result, the foreign matter adhering to the inside of the first hole portion W11 or the like can be dropped to the outside.

Further, in steps S170 and S180, the first hole portion W11 and the like are imaged by the first borescope 71 and the second borescope 72 having different imaging directions D71 and D72. FIG. 17 shows the first hole W11 imaged by the first borescope 71 or the like (imaging result). More specifically, the imaging result B71 shown in FIG. 17A is the imaging result of the first borescope 71. FIG. Moreover, the imaging result B72 shown in FIG.17(b) is the result of having imaged the 1st hole W11 shown in FIG.17(a) with the 2nd borescope 72. In FIG.

As shown in FIG. 17(a), the first borescope 71 can be used to check the entire circumference of the first hole W11 by looking straight at the first hole W11. Moreover, as shown in FIG.17(b), a part of inner peripheral surface of the 1st hole W11 can be confirmed obliquely by the 2nd borescope 72. As shown in FIG. In this way, by imaging in different imaging directions D71 and D72, the first hole W11 can be inspected from different viewpoints, and the quality can be determined with high accuracy.

In addition, the land W11b may vary in the degree of light reflection depending on the state of the surface (casting surface), and the appearance of the land W11b may change in the imaging results B71 and B72. Therefore, for example, depending on the direction in which the image of the land W11b is taken, the land W11b may appear too bright, making it difficult to grasp the shape of the land W11b. Even in such a case, a plurality of images (imaging results B71 and B72 of the land portion W11b) with different appearances (brightness) can be obtained by imaging in a plurality of imaging directions D71 and D72. As a result, the influence of the degree of reflection can be reduced, and the pass/fail determination can be performed with high accuracy.

Further, for example, when the imaging unit 70 (the first borescope 71 or the like) is moved with respect to the fixed workpiece W and the imaging unit 70 is inserted into the first hole W11 or the like to perform imaging, the imaging unit 70 Since the degree of freedom of movement is restricted, blind spots may occur depending on the surface structure of the work W, making imaging difficult. In contrast, in the present embodiment, by combining a borescope (first borescope 71, etc.) with a different field of view and a multi-axis robot 40 that rotates the workpiece W, the machining surface of the workpiece W (first hole W11, etc.) inner surface) can be imaged without blind spots.

Moreover, the first borescope 71 and the second borescope 72 change the direction of insertion into the first hole W11 and the like to image the first hole W11 and the like. By changing the insertion direction in this way, it is possible to easily check a portion or defect that is difficult to see in one insertion direction by imaging in the other insertion direction. As a result, it is possible to inspect in detail whether or not the first hole portion W11 or the like is defective, and it is possible to accurately determine the quality of the product.

In addition, the pass/fail determination section 92 performs pass/fail determination of the workpiece W using the learned DNN 100 in step S190. With such a configuration, it is possible to extract the features of the workpiece W from the imaging results B71 and B72 based on the learning results, and to accurately determine the quality of the work W. In particular, by using the DNN 100, the number of intermediate layers 102, that is, the number of exchanges of information among the units 101a, 102a, and 103a is increased (see FIG. 9), and the features of the work W can be learned (extracted) in detail. . For this reason, quality determination can be performed more accurately.

In steps S170 and S180, the workpiece W is imaged by the first borescope 71 and the second borescope 72, but the workpiece W may be imaged by the third borescope 73 in addition to the imaging. By using the result of imaging by the third borescope 73 for quality determination, the work W can be confirmed using the result of imaging the first hole W11 and the like sideways (from the imaging direction D73). As a result, the shape of the first hole W11 and the like can be grasped in more detail, so that the accuracy of quality determination can be improved.

In addition, by using the heat map H created in step S190, it is possible to verify the contents of the pass/fail judgment and improve convenience. Hereinafter, description will be made with reference to FIG. 18 as an example.

An imaging result B71 shown in FIG. 18A shows a burr B71a. In this case, as described above, the DNN 100 extracts the burr B71a, and the quality determination result is NG.

FIG. 18(b) shows a heat map H created in the NG determination. In the heat map H, since the DNN 100 extracted the burr B71a (reacted to the burr B71a) in the pass/fail judgment, it is shown that the reaction is relatively high in the range H1 that partially overlaps with the burr B71a. Moreover, since the range H2 within the range H1 had a particularly large effect on the quality judgment, the heat map H shows that the reaction in the range H2 within the range H1 is particularly high.

If the operator or the like confirms the ranges H1 and H2 of the heat map H when the result of the pass/fail judgment is NG, etc., the occurrence of the burr B71a in the first hole portion W11 or the like can be quickly confirmed. can be done. Also, if the heat map H is checked when checking the operation of the inspection apparatus 1, it can be estimated that the DNN 100 has extracted the burr B71a, and it can be confirmed that the pass/fail judgment is performed satisfactorily. As a result, it is possible to verify the contents of the quality judgment by the DNN 100, and the convenience can be improved.

As described above, the inspection apparatus 1 according to the first embodiment grips the positioned work W, and moves the work W to the first position (Fig. 15(b)) by rotating the work W around a plurality of rotation axes. The multi-axis robot 40 (moving unit) that moves to the imaging position shown), and the imaging unit 70 ( acquisition unit).

By configuring in this way, the work W can be accurately moved to the first position, and the image of the work W can be easily imaged in a state in which the imaging unit 70 and the work W are aligned. As a result, it is possible to perform quality determination with high accuracy.

Further, the inspection device 1 has a first pin 33 (insertion portion) that can be inserted into the first hole W11, and by inserting the first pin 33 into the first hole W11, the workpiece can be It further comprises an elevating slider 30 (positioning section) that positions W and delivers it to the multi-axis robot 40 .

By configuring in this way, the workpiece W can be positioned with high precision, and quality determination can be performed with high precision.

Further, the inspection apparatus 1 moves the transfer slider 20 (conveyance section) for transferring the work W to the second position (first transfer position P2) where the first pin 33 is inserted into the first hole W11. It is further equipped.

By configuring in this way, the burden on the operator can be reduced. In particular, the transport slider 20 according to the first embodiment can transport from the vicinity of the input port 11 (input position P1) to the second position. This can effectively reduce the burden on the operator.

Further, when the work W is transferred to the multi-axis robot 40, the elevating slider 30 moves from the second position to a third position (second transfer position P3) higher than the height of the work W. It lifts the workpiece W.

With this configuration, the workpiece W can be lifted to the relatively high third position, and the workpiece W can be easily gripped by the multi-axis robot 40 . Further, it is possible to facilitate fine adjustment of the tightness of the first pin 33 and the like. In addition, by blowing air onto the work W when lifting the work W, it becomes easier to remove the foreign matter adhering to the work W.

In addition, the imaging unit 70 is formed to be insertable into the first hole W11 and the like, and the multi-axis robot 40 moves the work W to move the imaging unit 70 into the first hole W11 and the like. is inserted.

By configuring in this way, it is possible to prevent the occurrence of troubles (failures, etc.) due to the movement of the imaging unit 70 .

Further, the multi-axis robot 40 moves the workpiece W from the bottom to the top, thereby inserting the imaging unit 70 into the first hole W11 or the like.

With this configuration, the axial direction of the imaging section 70 (insertion sections 71b and 72b) faces the vertical direction (see FIG. 8), so deformation of the imaging section 70 (deflection due to its own weight, etc.) is suppressed. be able to.

Further, the multi-axis robot 40 grips the work W with a deformed portion 45c that is deformed along the outer shape of the work W. As shown in FIG.

By configuring in this way, the workpiece W can be easily gripped by the deformation portion 45c.

Further, the inspection apparatus 1 inputs the imaging results B71 and B72 (acquisition results of the acquisition unit) to the learned DNN 100 (learning model), thereby making a quality judgment unit 92 (determination unit 92) that judges the quality of the work W. part).

By configuring in this way, the DNN 100 can perform quality determination with high accuracy.

Further, as described above, the inspection method according to the first embodiment includes moving steps (steps S130, S170, and S180) in which the positioned workpiece W is gripped by the multi-axis robot 40 (moving unit) and moved to a predetermined position. , an acquisition step (steps S170 and S180) of acquiring information about the first hole W11 (inspection hole) formed in the work W at the predetermined position by the imaging unit 70 (acquisition unit).

Such a configuration makes it easier for the imaging unit 70 to acquire information about the first hole W11 and the like.

As described above, the inspection apparatus 1 according to the first embodiment includes the multi-axis robot 40 (moving unit) that moves the work W, and the work W that is immovably provided and moved by the multi-axis robot 40. and an imaging unit 70 (first imaging unit) capable of imaging.

With this configuration, it is possible to move the workpiece W instead of the imaging unit 70 and prevent problems (failures, etc.) caused by the movement of the imaging unit 70 .

Further, the imaging unit 70 is capable of imaging the first hole W11 and the like formed in the workpiece W in a state in which the insertion portions 71b and 72b are inserted into the first hole W11 and the like (holes). , and at least two borescopes (a first borescope 71 and a second borescope 72) whose imaging directions D71 and D72 are different from each other.

By configuring in this way, it is possible to determine the quality of the workpiece using the results B71 and B72 (see FIG. 17) of imaging the first hole portion W11 and the like from a plurality of viewpoints (facing the imaging directions D71 and D72). It is possible to improve the accuracy of the quality judgment of W. Moreover, by using at least two borescopes, even if one borescope (eg, the first borescope 71) fails, the other borescope (eg, the second borescope 72) can be used to restore the first hole W11, etc. can be continuously imaged. As a result, inspection can be performed with the inspection device 1 until the failed borescope is replaced, or the inspection device 1 can be used for purposes other than inspection (for example, for purposes such as checking workpieces W that have been determined to be NG). can improve convenience.

Also, the borescope is provided so as to hang down from above.

With this configuration, the axial direction of the borescope (insertion portions 71b and 72b) faces the vertical direction, so deformation of the borescope (deflection due to its own weight, etc.) can be suppressed. In addition, it is possible to prevent foreign matter from adhering to the borescope.

In addition, the borescope is configured to change the brightness of the illumination (light) that illuminates the first hole W11 and the second hole W12 at the same location, and perform the first hole W11 and the second hole W12 multiple times. is imaged.

By configuring in this way, the shape of the first hole W11 and the like can be confirmed in a wide range, so the accuracy of quality determination of the workpiece W can be improved.

The borescope includes a first borescope 71 for imaging the first hole W11 and the like facing the axial direction (imaging direction D71) of the insertion portion 71b, and and a second borescope 72 that faces the direction (imaging direction D72) that is inclined with respect to the orthogonal direction and images the first hole portion W11 and the like.

By configuring in this way, it is possible to determine the quality of the workpiece using the results B71 and B72 (see FIG. 17) of imaging the first hole portion W11 and the like from a plurality of viewpoints (facing the imaging directions D71 and D72). It is possible to improve the accuracy of the quality judgment of W.

The borescope further includes a third borescope 73 capable of imaging the first hole W11 and the like in a direction (imaging direction D73) orthogonal to the axial direction.

By configuring in this way, the shape of the inner surface of the first hole portion W11 and the like can be grasped in detail, so the accuracy of quality determination of the work W can be improved.

Further, the inspection apparatus 1 further includes a shaft member 62 (test object) imitating the borescope, and the multi-axis robot 40 moves the work W to move the shaft member 62 to the first It is inserted into the hole W11 or the like.

By configuring in this way, it is possible to confirm in advance that the borescope does not come into contact with the first hole W11 or the like (step S140). This can prevent contact between the borescope and the first hole W11 or the like.

Further, the multi-axis robot 40 rotates the workpiece W about the axis L72 of the insertion portion 72b.

By configuring in this way, it is possible to rapidly perform imaging of the first hole portion W11 and the like (imaging while changing the position in the circumferential direction).

In addition, the inspection apparatus 1 includes a camera 51 (second imaging unit) that images the first hole W11 formed in the workpiece W and the like, and the multi-axis robot 40 based on the imaging result B51 of the camera 51. A processing unit 54 (correction unit) for correcting movement of the work W is further provided.

By configuring in this way, positional deviation of the first hole W11 and the like with respect to the imaging unit 70 can be suppressed.

Further, the inspection apparatus 1 further includes a pass/fail judgment section 92 (judgment section) that judges the quality of the work W by inputting the imaging results B71 and B72 of the imaging section 70 to the learned DNN 100 (learning model). It is equipped.

By configuring in this way, the DNN 100 can perform quality determination with high accuracy.

Further, as described above, the inspection method according to the first embodiment includes a moving step of moving the workpiece W by the moving unit (the multi-axis robot 40), and a moving step of moving the workpiece W moved by the multi-axis robot 40. (steps S170 and S180).

By configuring in this way, it is possible to prevent the occurrence of troubles (failures, etc.) due to the movement of the imaging unit 70 .

Further, as described above, the inspection apparatus 1 according to the first embodiment includes the imaging unit 70 capable of imaging the first hole W11 and the like (hole) formed in the workpiece W, and the learned DNN 100 (learning model). and a pass/fail judgment section 92 (judgment section) for judging the quality of the work W by inputting the image pickup results B71 and B72 of the image pickup section 70 to the image pickup section 70 .

By configuring in this way, the DNN 100 can perform quality determination with high accuracy.

In addition, the DNN 100 captures the image of the first hole W11 and the like of the work W to be determined to be non-defective by the imaging unit 70, and the first hole of the work W to be determined to be defective. The result of imaging W11 and the like by the imaging unit 70 is learned by deep learning.

With this configuration, the DNN 100 can finely extract the features of the work W and perform the quality determination more accurately.

Moreover, the inspection apparatus 1 further includes a creation unit 93 that creates a heat map H indicating the degree of reaction of the DNN 100 to the imaging results B71 and B72 of the imaging unit 70 .

By configuring in this way, it becomes possible to verify the contents of the pass/fail judgment, and convenience can be improved.

Further, in the imaging unit 70, a first borescope 71 that is inserted into the first hole W11 and the like and faces in its own axial direction (imaging direction D71) to image the first hole W11 and the like. (first imaging unit), in a state of being inserted into the first hole W11 or the like, facing a direction (imaging direction D72) inclined with respect to its own axial direction and a direction orthogonal to the axial direction and a second borescope 72 (second imaging unit) that images the first hole W11 and the like.

By configuring in this way, it is possible to determine the quality using the results B71 and B72 (see FIG. 17) of imaging the first hole W11 and the like from a plurality of viewpoints (facing the imaging directions D71 and D72). Determination can be performed with higher accuracy.

In addition, in the imaging unit 70, the first hole W11 or the like of the workpiece W is inserted into the first hole W11 or the like, and the first hole W A third borescope 73 (third imaging unit) capable of imaging W11 and the like is further included.

By configuring in this way, the shape of the inner surface of the first hole W11 and the like can be grasped in detail, so that the quality judgment can be performed more accurately.

The inspection apparatus 1 further includes a correction section 91 that corrects the gain of the imaging section 70 for each workpiece W based on the imaging result B71 of the imaging section 70 .

By configuring in this way, it becomes easier to check the first hole W11 and the like in the imaging results B71 and B72 (image data) of the first borescope 71 and the second borescope 72, so that the quality determination is performed more accurately. becomes possible.

Further, as described above, the inspection method according to the first embodiment includes the imaging step (steps S170 and S180) of imaging the first hole W11 formed in the workpiece W and the like by the imaging unit 70, and the learned DNN 100 ( a determination step (step S190) of determining whether the work W is good or bad by inputting the imaging result B71 of the imaging unit 70 into the learning model).

By configuring in this way, the DNN 100 can perform quality determination with high accuracy.

Note that the multi-axis robot 40 according to the first embodiment is an embodiment of the moving unit.
Also, the first hole W11 and the like according to the first embodiment are an embodiment of an inspection hole and a hole.
Also, the imaging unit 70 according to the first embodiment is an embodiment of the acquisition unit and the first imaging unit.
Also, the first pin 33 according to the first embodiment is an embodiment of the insertion portion.
Also, the elevation slider 30 according to the first embodiment is an embodiment of the positioning portion.
Further, the transport slider 20 according to the first embodiment is an embodiment of the transport section.
Also, the DNN 100 according to the first embodiment is an embodiment of a learning model.
Also, the quality determination unit 92 according to the first embodiment is an embodiment of the determination unit.
Further, the shaft member 62 according to the first embodiment is an embodiment of the test body.
Also, the camera 51 according to the first embodiment is an embodiment of the second imaging unit that images the hole for correcting the movement of the work.
Also, the processing unit 54 according to the first embodiment is an embodiment of a correction unit that corrects movement of the work.
Also, the first borescope 71 according to the first embodiment is an embodiment of the first imaging unit that takes an image of the hole in the axial direction.
In addition, the second borescope 72 according to the first embodiment is an embodiment of the second imaging unit that images the hole in the axial direction and in a direction that is inclined with respect to the direction perpendicular to the axial direction. be.
Further, the third borescope 73 according to the first embodiment is an embodiment of the third imaging unit capable of imaging the hole in a direction orthogonal to the axial direction.

Although the first embodiment of the present invention has been described above, the present invention is not limited to the above configuration, and various modifications are possible within the scope of the invention described in the claims.

For example, the inspection apparatus 1 inspects the workpiece W used for the housing of the valve device, but the type of inspection object of the inspection apparatus 1 is not particularly limited. Also, the object to be inspected does not necessarily have to be an object manufactured by casting, and may be an object manufactured by other techniques.

Also, in the first embodiment, the PC 90 (correction unit 91) adjusts the gain, but the invention is not limited to this, and image processing may be performed. Specifically, the PC 90 analyzes the result B71 (see FIG. 16A) of imaging the first hole W11 first, and calculates a correction value for correcting the luminance. Then, the PC 90 performs image processing (processing for changing brightness and processing for adjusting contrast) on the imaging result B71 using the correction value. The PC 90 also performs image processing using the correction values for the results of the second and subsequent imaging. Thus, the PC 90 can cope with the difference in surface conditions by image processing.

As described above, the inspection apparatus 1 further includes the PC 90 (processing section) that performs image processing on the imaging results B71 and B72 of the imaging section 70 .

By configuring in this way, it becomes easier to confirm the first hole W11 and the like in the imaging results B71 and B72, so it is possible to perform quality determination with higher accuracy.

Note that the PC 90 that performs the image processing described above is an embodiment of the processing unit.

It should be noted that the configuration of the PC 90 is not particularly limited as long as it can determine whether the work W is good or bad. Therefore, the PC 90 does not necessarily need to perform gain adjustment or image processing. Also, the PC 90 does not have to create the heat map H.

In addition, although the inspection apparatus 1 moves the workpiece W during inspection, it is not limited to this. For example, the image pickup unit 70 may be separately moved to pick up an image of the workpiece W.

In addition, although the multi-axis robot 40 grips the work W with the deformable portion 45c, the present invention is not limited to this, and the work W may be gripped by pressing an undeformable member against the work W. .

In addition, the multi-axis robot 40 rotates the workpiece W while the insertion portion 72b is inserted into the first hole portion W11 or the like at the time of imaging by the second borescope 72, but the present invention is not limited to this. For example, the multi-axis robot 40 may rotate the workpiece W by extracting the insertion portion 72b from the first hole W11 or the like. In this case, contact between the first hole W11 and the like and the insertion portion 72b can be prevented even if the workpiece W is rotated about a direction different from the axis of the insertion portion 72b.

Further, although the lifting slider 30 positions the work W by inserting the first pin 33 into the first hole W11 (see FIG. 12), the method of positioning the work W is not limited to this. . For example, the lifting slider 30 may position the workpiece W by inserting the first pin 33 into the second hole W12 or the third hole W13. Further, the lifting slider 30 does not necessarily have to be positioned by inserting a pin into the hole. For example, the work W may be positioned by pressing a predetermined member against the side surface of the work W.

Also, the lift slider 30 lifted the work W from the first transfer position P2 to the second transfer position P3, which is higher than the height of the work W (see FIG. 2), but the height for lifting the work W is particularly limited. You can lift it to any height instead.

Further, the outer diameter of the shaft member 62 is formed to be larger than the outer diameter of the imaging section 70 (insertion section 71b, etc.) (see FIG. 8). Any shape or posture similar to that of the imaging unit 70 may be used to the extent that contact can be confirmed. Therefore, the shaft member 62 may have, for example, the same outer diameter as the insertion portion 71b.

In addition, the insertion portions 71b, 72b, and 73b of the first borescope 71 to the third borescope 73 are provided so as to hang downward (see FIG. 8), but the posture of the insertion portion 71b and the like is limited to this. It can be in any position, rather than being For example, the insertion portion 71b and the like may be in a posture in which the axial direction is oriented horizontally. In this case, the multi-axis robot 40 moves the work W along the horizontal direction and inserts the insertion portion 71b and the like into the first hole portion W11.

In the inspection, the imaging unit 70 images the first hole W11 and the like with two borescopes (the first borescope 71 and the second borescope 72) (steps S170 and S180). The number is not particularly limited. For example, the imaging unit 70 may image the first hole W11 and the like only with the first borescope 71 .

In addition, although the first borescope 71 and the second borescope 72 performed imaging multiple times at the same location while changing the brightness of the light, the present invention is not limited to this. Imaging may be performed only once.

In addition, although the PC 90 performs pass/fail determination using the DNN 100, the method for performing the pass/fail determination is not particularly limited. For example, the PC 90 may use a neural network with one intermediate layer 102 to determine the quality. Also, the PC 90 may perform pass/fail judgment using a pre-captured reference image instead of the neural network. In this case, the PC 90 can, for example, extract the difference between the reference image and the imaging results B71 and B72 of the imaging section 70, and perform pass/fail determination based on the extraction result.

Also, although the DNN 100 is learned by supervised learning, the learning method is not particularly limited, and may be learned by, for example, unsupervised learning.

Other configurations of the inspection apparatus 1 are not particularly limited as long as they can image the workpiece W and determine whether it is good or bad. Therefore, the inspection apparatus 1 does not necessarily need to include the transport slider 20 (see FIG. 2), the misalignment corrector 50 (see FIG. 7), and the interference checker 60 (see FIG. 8).

Also, although the PC 90 adjusts the gain for each workpiece W, the gain is not limited to this, and may be adjusted for each production lot of the workpiece W, for example. A specific description will be given below.

When the work W is manufactured by casting, for example, a different production lot is assigned to each ladle. When the PC 90 inspects the work W of the given manufacturing lot for the first time, the PC 90 determines the characteristics of the work W according to the manufacturing lot (for example, the degree of light reflection and the work W color, etc.). In addition, when a defect is extracted in the quality judgment of the work W (step S190), the PC 90 extracts the type of the defect (type of burr, remaining foreign matter, etc.) and the extraction location as the feature.

The PC 90 adjusts the gain for each manufacturing lot based on the characteristics thus extracted. For example, when the PC 90 extracts a feature that the work W inspected first in a certain production lot has a relatively bright color, the PC 90 detects that the work W in the same production lot is subsequently inspected. Lower the gain.

Also, the PC 90 may change the determination conditions (parameters for determination) of the work W instead of the gain for each manufacturing lot (according to characteristics). The determination conditions for the work W are the values and the contents of the arithmetic processing used in the quality determination (step S190), and more specifically, the contents of the calculation processing by the DNN 100. FIG. A plurality of DNNs 100 are constructed in the PC 90 so that the determination conditions can be changed for each manufacturing lot. A plurality of DNNs 100 are trained using workpieces W with different characteristics (degree of light reflection, color of workpiece W, etc.) as described above, and can execute pass/fail judgments according to the characteristics (and thus the manufacturing lot). configured to

When changing the determination conditions, the PC 90 determines which DNN 100 to use among the plurality of DNNs 100 based on the features (color, defect extraction points, etc.) of the work W to be re-inspected first for each manufacturing lot. to decide. The PC 90 thus changes the determination conditions (DNN 100) for each manufacturing lot, and determines whether the work W is good or bad.

As described above, the work W is manufactured by casting, and the inspection device 1 includes the PC 90 (extracting unit) for extracting the characteristics of the work W for each production lot of the work W, and based on the extraction results, It further comprises a PC 90 (adjustment unit) that adjusts at least one of parameters relating to imaging by the imaging unit 70 and parameters relating to determination by the quality determination unit 92 for each manufacturing lot.

By configuring in this way, it is possible to optimize the parameter (gain) related to imaging or the parameter (DNN 100) related to determination according to the manufacturing lot, and it is possible to perform quality determination with higher accuracy.

Note that the PC 90 that adjusts the imaging conditions and the like for each manufacturing lot described above is an embodiment of the extractor and the adjuster.

Next, an inspection apparatus 201 and an inspection method according to a second embodiment will be described with reference to FIGS. 19 to 23. FIG.

The inspection device 201 according to the second embodiment is greatly different from the inspection device 1 according to the first embodiment in that it is determined whether or not to re-inspect the workpiece W. First, the re-examination will be explained.

Depending on how the first hole portion W11 and the like captured by the imaging unit 70 are captured, it is assumed that the quality determination (step S200) of the work W described in the first embodiment may not be performed accurately. In such a case, it is desirable to re-inspect and inspect in detail the work W that is difficult to determine whether it is good or bad, or the work W that should be reconfirmed at a predetermined location even if it is determined to be a non-defective product. .

The inspection apparatus 201 according to the second embodiment is configured to be able to extract such works W that require re-inspection (work W that is difficult to determine whether it is good or bad, etc.) and perform re-inspection. A specific description will be given below. In addition, about the member comprised similarly to the inspection apparatus 1 which concerns on 1st embodiment, the same code|symbol as 1st embodiment is attached|subjected, and the description is abbreviate|omitted.

In the inspection apparatus 201 according to the second embodiment shown in FIG. 19, the configuration of the PC 290 is different from that of the PC 90 (see FIG. 3) according to the first embodiment. The PC 290 includes a correction section 291 , a pass/fail determination section 292 , a creation section 293 and a re-inspection determination section 294 .

The correction unit 291 is for correcting the gain (sensitivity) of the imaging unit 70, and is configured similarly to the correction unit 91 according to the first embodiment. The quality determining section 292 is for determining the quality of the work W, and is configured in the same manner as the quality determining section 92 according to the first embodiment. The creation unit 293 is for creating a heat map H293 (see FIG. 23(a)) indicating the degree of reaction of the retest necessity DNN 297, which will be described later.

The reexamination determination unit 294 is for determining whether or not to reexamine. The processing of the reexamination determination unit 294 will be described later.

A plurality of DNNs (see FIG. 9) having an input layer, a plurality of intermediate layers and an output layer are constructed in the PC 290 configured as described above. Specifically, in the PC 290, a pass/fail determination DNN 296, a reexamination necessity DNN 297, and an imaging condition DNN 298 are constructed.

The pass/fail judgment DNN 296 is for judging the pass/fail of the workpiece W, and is configured in the same manner as the DNN 100 according to the first embodiment. That is, the pass/fail judgment DNN 296 learns in advance the relationship between the result B271 (see FIG. 20) of imaging the first hole W11 and the like (image pickup location) by the imaging unit 70 and the pass/fail judgment result. As shown in FIG. 20(b), when the imaging result B271 from the imaging unit 70 is input, the quality judgment DNN 296 receives a quality judgment score S296 indicating the probability of a good product or a defective product, and the judgment result of a good product or a defective product. R296 can be output. The quality judgment score S296 outputs the probability of a non-defective product (determination of quality is OK) and the probability of a defective product (determination of quality is NG) with 100% as the parameter. For example, the probability of a non-defective product is 80% and the probability of a defective product is 20%. As the determination result R296, a determination result of OK or NG is output based on the probability of a non-defective product and the probability of a defective product.

Here, for example, in the imaging result B271 of the workpiece W that needs to be re-inspected, there are parts that should be confirmed in detail in the re-inspection, that is, parts that are defective or difficult to determine the quality, etc. Therefore, the quality determination score S296 (eg, 30% chance of being good, 70% chance of being bad). On the other hand, the imaging result B271 of a work W that does not require re-inspection, for example, a work W that is clearly good, does not show any defects, so the pass/fail judgment score S296 is a high value (for example, the probability of a good product is 100%). , the probability of defective products is 0%). In this way, the imaging result B271 and the pass/fail judgment score S296 input/output to/from the pass/fail judgment DNN 296 are information closely related to the necessity of the above-described reexamination.

The re-inspection necessity DNN 297 is for determining whether or not the workpiece W should be re-inspected. As described above, the imaging result B271 and the pass/fail judgment score S296 are related to the necessity of reexamination. Therefore, the re-inspection necessity DNN 297 can determine whether or not to re-inspect the work W by learning in advance the relationship between the imaging result B 271, the pass/fail judgment score S296, and the re-inspection judgment result through supervised learning. is configured to

Specifically, as shown in FIG. 21(a), learning data A200 including imaging result B271 and pass/fail judgment score S296 of workpiece W requiring reinspection, and the fact that reinspection is necessary (answer). The necessity of retesting is input to the DNN 297 (input layer), and the information is transferred to the output side. This allows the re-inspection necessity DNN 297 to learn the characteristics of the work W that requires re-inspection. As shown in FIG. 21B, when the imaging result B271 and the pass/fail judgment score S296 are input, the re-examination necessity DNN 297 uses the imaging result B271 to determine the characteristics such as defects that have affected the pass/fail judgment score S296. It is possible to extract and determine that re-examination is necessary.

In addition, the re-inspection necessity DNN 297 performs learning based on the learning data A 200 for works W that do not require re-inspection in the same way as for works W that require re-inspection. Here, the work W that does not require re-inspection is the work W for which the inspection result (defective judgment) is clear. (a work having a somewhat low pass/fail judgment score S296). The learned re-examination necessity DNN 297 is capable of extracting features such as the presence or absence of defects from the input imaging result B 271 and determining that re-inspection is unnecessary.

An imaging condition DNN 298 shown in FIG. 19 is used to determine imaging conditions (parameters relating to imaging) of the imaging unit 70 at the time of reexamination. The imaging condition is a condition for determining how to image the workpiece W (image location). Specifically, the gain, the imaging direction, the brightness of the light irradiated to the first hole W11, etc. is.

Here, at the time of reinspection, it is desirable to image the first hole portion W11 and the like under optimum imaging conditions (gain, etc.) that can properly image the location to be checked for defects and the like. Such imaging conditions differ for each imaging result B271. Therefore, the imaging condition DNN 298 pre-supervised learning of the relationship between the imaging result B271 and the optimum imaging condition so that the optimum imaging condition for reexamination can be determined for each imaging result B271 (Fig. 21(a) reference).

Specifically, the imaging result B271 of the workpiece W requiring re-inspection and the imaging conditions (answers) appropriate for the imaging result B271 are input to the imaging condition DNN 298 (input layer), and the information is sent to the output side. hand over. This allows the imaging condition DNN 298 to learn the optimum imaging condition corresponding to the imaging result B271. When the imaging result B271 is input, the learned imaging condition DNN 298 acquires information necessary for determining the imaging condition from the imaging result B271, specifically, the type of defect, the color of the workpiece W, the imaging direction, and the like. It is possible to extract the features and output the optimum imaging conditions.

The flow of inspection of the work W by the inspection device 201 will be described below.

As shown in FIG. 22, the inspection device 201 (control unit 80) performs steps S100 to S180 described in the first embodiment. That is, the inspection apparatus 201 causes the multi-axis robot 40 to move the set work W to a position (imaging position) where the imaging unit 70 can image, and the imaging unit 70 causes the first hole W11 of the work W (imaging position). point) is imaged.

After that, the pass/fail determination section 292 performs pass/fail determination of the workpiece W in step S200 described in the first embodiment. At this time, as shown in FIG. 20(b), the imaging result B271 obtained in steps S170 and S180 is input to the pass/fail determination DNN 296. FIG. The good/bad judgment DNN 296 outputs a good/bad product judgment result R296 and a good/bad judgment score S296 based on the imaging result B271.

As shown in FIG. 22, after the quality determination of the work W is performed, the reinspection determination unit 294 determines whether the quality determination score S296 is equal to or greater than a predetermined threshold value (for example, whether the probability of a good product is 80%). It is determined whether or not (step S210). If the pass/fail determination score S296 is equal to or greater than the predetermined threshold (step S210: Yes), the workpiece W is returned to the input position P1 (see FIG. 1) in step S190 described in the first embodiment and inspected by the inspection device 201. ends. In this way, the inspection apparatus 201 does not reinspect a work W with a high pass/fail determination score S296, that is, a work W that is clearly good and has a low need for reinspection.

On the other hand, when the pass/fail determination score S296 is less than the predetermined threshold (step S210: No), the reinspection determination unit 294 uses the reinspection necessity DNN 297 to determine whether or not to reinspect the workpiece W (step S220). In this way, the re-inspection determination unit 294 determines the work W having a low quality determination score S296, that is, the work W whose quality determination is difficult, or the work W that should be checked at a predetermined location even if it is determined as a non-defective product (re-inspection can be extracted. At this time, the creating unit 293 creates a heat map H293 indicating the degree of reaction of the retest necessity DNN 297 .

If the re-inspection determining unit 294 determines that re-inspection is unnecessary (step S220: No), the workpiece W is returned to the input position P1 and the inspection by the inspection device 201 is completed (step S190). In this way, the inspection apparatus 201 does not reinspect works W that do not require reinspection, for example, works W that have obvious defects that do not require reinspection.

On the other hand, if reexamination is necessary (step S220: Yes), the reexamination determination unit 294 uses the imaging condition DNN 298 to determine imaging conditions for reexamination (step S230).

The inspection apparatus 201 re-inspects the workpiece W based on the imaging conditions thus determined (step S240). In other words, the multi-axis robot 40 moves the workpiece W so that the image capturing section 70 can capture an image of the imaged location determined to require re-inspection. Further, the imaging unit 70 images the workpiece W under the imaging conditions determined in step S240. At this time, the image capturing unit 70 captures the image with a borescope corresponding to the image capturing direction included in the image capturing conditions. For example, when the imaging direction is downward, the first borescope 71 (imaging direction D71) is used, and when the imaging direction is oblique, the second borescope 72 (imaging direction D72) is used, and the imaging direction is the horizontal direction. In this case, an image is taken with the third borescope 73 (imaging direction D73) (see FIG. 8). The pass/fail determination unit 292 (pass/fail determination DNN 296) performs a pass/fail determination (re-inspection) based on the result of the imaging.

A specific example of the inspection (steps S100 to S240) by the inspection apparatus 201 described above will be described below. In the following, a specific example will be described by taking as an example a case where reinspection is performed due to the burr B271a shown in FIG. 20(a).

The imaging unit 70 images the first hole W11 and the like of the work W (see FIG. 20(a), steps S100 to S180). The imaging result B271 of the first borescope 71 imaged in this manner includes an image of a burr B271a, as shown in FIG. 20(b).

The reexamination determination unit 294 inputs the imaging result B271 to the pass/fail determination DNN 296 to perform pass/fail determination (step S200). The pass/fail judgment DNN 296 extracts a burr B271a (feature) from the imaging result B271 and outputs a pass/fail judgment score S296 and the like. The quality determination score S296 is X%, which is lower than the threshold value with the probability of being good (step S210: No).

In this case, as shown in FIG. 21B, the reexamination determination unit 294 inputs the imaging result B271 (imaging result used in the pass/fail judgment) and the pass/fail judgment score S296 to the reexamination necessity DNN 297, (step S220). The re-examination necessity DNN 297 extracts features such as a burr B271a from the imaging result B271, and outputs a determination result R297 indicating that re-inspection is necessary. In addition, as shown in FIG. 23(a), the creating unit 293 creates a heat map H293 indicating that the burr B271a has a great influence on the determination result R297 (retest necessity DNN297 shows a high response). do.

Then, the reexamination determination unit 294 inputs the imaging result B271 (imaging result used in pass/fail determination) to the imaging condition DNN 298, and determines the imaging condition (step S230). At this time, the imaging condition DNN 298 extracts information (features) necessary for determining imaging conditions from the imaging result B 271 . Specifically, information such as the burr B271a, the color of the workpiece W, and the imaging direction D71 is extracted. The imaging condition DNN 298 outputs appropriate imaging conditions for the imaging result B 271 based on the extraction result. In this way, the imaging condition DNN 298 outputs an imaging condition such that the imaging direction is an oblique direction (imaging direction D72), for example.

As shown in FIG. 23B, the inspection apparatus 201 performs re-inspection based on the imaging conditions determined in step S230 (step S240). That is, the inspection apparatus 201 controls the multi-axis robot 40 by the control unit 80 to insert the first hole portion W11 into the insertion portion 72b of the second borescope 72, and uses the second borescope 72 to obliquely ( A predetermined imaging location including the burr B271a is imaged (facing the imaging direction D72). The pass/fail judgment unit 292 judges the pass/fail of the workpiece W on the basis of the imaged result.

In this way, in the above-described reinspection, the burr B271a can be inspected intensively. In the reinspection, the burr B271a can be imaged under an imaging condition (appropriate imaging direction D72) that facilitates confirmation of the burr B271a.

In addition, the re-inspection determination unit 294 determines whether or not to perform re-inspection using the re-inspection necessity DNN 297 (step S220). By extracting B271a, etc., it is possible to accurately extract the work W that needs to be re-inspected. By reinspecting the works W extracted in this way, it is possible to improve the quality of the product.

In addition, the re-inspection determination unit 294 compares the pass/fail determination score S296 with a threshold value, and does not perform re-inspection determination for the work W having a high pass/fail determination score S296 (step S210: Yes, step S190). As a result, it is possible to reduce the load on the PC 290 by not performing the re-inspection necessity DNN 297 for the work W for which the need for re-inspection is clearly low.

As described above, the inspection apparatus 201 according to the second embodiment captures an image of the workpiece W using the imaging unit 70, and determines the quality of the workpiece W based on the imaging result B271 of the imaging unit 70. and the learned re-inspection necessity DNN 297 (first learning model), the control unit 80 and the pass/fail determination unit 292 inspect the work W again. and a reexamination determination unit 294 (determination unit) that determines whether or not to perform an inspection.

By configuring in this way, it is possible to accurately extract the workpiece W that needs to be re-inspected.

In addition, the quality determination unit 292 outputs a score (quality determination score S296) indicating the possibility of whether or not the work W is a non-defective product in the quality determination of the work W, and the re-inspection necessity DNN 297 is , the relationship between the imaging result B271 of the imaging unit 70, the score, and whether or not to perform the reexamination is learned.

By configuring in this way, it is possible to extract features from the imaged result B271 and the like, and to more accurately extract workpieces W that need to be re-inspected.

Further, the PC 290 determines parameters (imaging conditions) related to imaging by the imaging unit 70 at the time of the reexamination based on the imaging results B271 (imaging results at steps S170 and S180) before the reexamination. (step S230).

By configuring in this way, it is possible to optimize the parameters related to imaging at the time of reexamination. For example, it becomes possible to image the work W from a direction that facilitates quality determination at the time of reinspection (see FIG. 23(b)), so that quality determination of the work W can be performed more accurately.

Further, the reexamination determination unit 294 determines the parameters using the imaging condition DNN 298 (second learning model) obtained by learning the relationship between the imaging results B271 of the imaging unit 70 and the parameters (step S230 ).

By configuring in this way, features can be extracted from the imaging result B271, and the parameters relating to imaging can be optimized according to the features.

In addition, in the imaging unit 70, while being inserted into the first hole W11 or the like (hole) formed in the work W, the first hole is directed in its own axial direction (imaging direction D71). A first borescope 71 (first imaging unit) for imaging W11 and the like, and in a state of being inserted into the first hole W11 and the like, inclined with respect to its own axial direction and a direction orthogonal to the axial direction and a second borescope 72 (second imaging unit) for imaging the first hole W11 and the like while facing the direction (imaging direction D72).

By configuring in this way, it is possible to determine the quality of the workpiece W in reinspection using the result B271 of imaging the first hole W11 and the like from a plurality of viewpoints (imaging directions D71 and D72). can do well.

Further, the imaging unit 70 is inserted into the first hole W11 or the like in the reexamination and images the first hole W11 or the like in a direction perpendicular to the axial direction (imaging direction D73). It further includes a third borescope 73 (third imaging section).

By configuring in this way, in the reinspection, it is possible to grasp the shape of the inner surface of the first hole W11 and the like in more detail from the imaging result of the third borescope 73, so that the quality of the work W in the reinspection Determination can be performed with higher accuracy.

Further, as described above, in the inspection method according to the second embodiment, the imaging unit 70 images the work W, and the quality determination unit 292 determines the quality of the work W based on the imaging result B271 of the imaging unit 70. By using the inspection process (steps S170, S180, S200) for inspecting the workpiece W and the learned re-inspection necessity DNN 297 (learning model), it is determined whether or not to perform the inspection process again. and a step (step S220).

By configuring in this way, it is possible to accurately extract the workpiece W that needs to be re-inspected.

Note that the control unit 80 and the pass/fail determination unit 292 according to the second embodiment are one embodiment of the inspection unit.
Further, the reexamination determination unit 294 according to the second embodiment is an embodiment of the determination unit.
Further, the retest necessity DNN 297 according to the second embodiment is an embodiment of the first learning model and the learning model.
Also, the imaging condition DNN 298 according to the second embodiment is an embodiment of the second learning model.
Also, the first hole W11 and the like according to the second embodiment are one embodiment of the hole.
Moreover, the first borescope 71 according to the second embodiment is an embodiment of the first imaging unit.
Moreover, the second borescope 72 according to the second embodiment is an embodiment of the second imaging unit.
Also, the third borescope 73 according to the second embodiment is an embodiment of the third imaging section.

Although the second embodiment of the present invention has been described above, the present invention is not limited to the above configuration, and various modifications are possible within the scope of the invention described in the claims.

For example, the re-inspection determination unit 294 does not perform re-inspection determination for works W that have a high probability of being non-defective (step S210: Yes, step S190). W may be determined to be reexamined.

Also, the quality determination DNN 296 outputs the probability of a good product and the probability of a defective product (good/bad decision score S296) and the determination result R296, but the information output by the quality determination DNN 296 is not particularly limited. For example, the good/bad decision DNN 296 may output only the probability of a good product. In this case, the pass/fail determination unit 292, the re-inspection determination unit 294, etc. can appropriately determine the pass/fail of the work W and the need for re-inspection based on the output result of the probability of a non-defective product.

In addition, although the reexamination necessity DNN 297 and the imaging condition DNN 298 are learned by supervised learning (see FIG. 21(a)), the learning method is not particularly limited. It may be learned by, for example.

In addition, the reexamination necessity DNN 297 learns the relationship between the imaging result B 271, the pass/fail judgment score S 296, and the reexamination judgment result. The content is not particularly limited. Also, the learning model for retesting determination need not be a DNN with a plurality of intermediate layers, and may be a neural network with one intermediate layer, for example.

Further, although the reexamination determination unit 294 determines the imaging conditions of the imaging unit 70 for reexamination using the imaging conditions DNN 298 (step S230), the method of determining the imaging conditions is not particularly limited. For example, it is not always necessary to use the DNN, and the imaging conditions may be determined based on the color information (for example, luminance) of the imaging result B271 imaged before the reexamination.

Further, the position of the workpiece W to be imaged in the re-inspection is not limited to a predetermined imaging location, and can be arbitrarily changed. A specific description will be given below with reference to FIG.

FIG. 23(a) shows a heat map H293 when it is determined that reinspection is necessary due to the burr B271a. In the heat map H293, the reaction is high in the ranges H293a and H293b including the burr B271a. The re-inspection determination unit 294 analyzes the heat map H293, and images the burr B271a from the highly responsive ranges H293a and H293b, that is, the position where the burr B271a can be clearly imaged. For the image obtained in this manner, the quality determination can be performed by the quality determination DNN 296 as in the case described above, or the quality can be determined visually by the operator.

As described above, the inspection apparatus 201 further includes a creation unit 293 that creates a heat map H293 indicating the degree of reaction of the re-examination necessity DNN 297 to the imaging result B271 of the imaging unit 70. The inspection determination section 294 determines the imaging portion of the work W at the time of the re-inspection based on the heat map H293.

With such a configuration, it is possible to intensively inspect portions that are likely to be defective (for example, burrs B271a, etc.).

Further, although the reexamination determination unit 294 determines the imaging conditions based on the imaging results B271 in steps S170 and S180 (step S230), the information used for determining the imaging conditions is not particularly limited. For example, the reinspection determination unit 294 may determine the imaging conditions using the manufacturing lot of the work W. FIG. A specific description will be given below.

When the work W is manufactured by casting, for example, a different production lot is assigned to each ladle. When the PC 290 inspects the work W of the given manufacturing lot for the first time, the PC 290 determines the characteristics of the work W according to the manufacturing lot (for example, the degree of light reflection and the work W color, etc.). In addition, when a defect is extracted in the quality judgment of the work W (step S200), the PC 290 extracts the type of the defect (type of burr, remaining foreign matter, etc.) and the extraction location as the feature.

The PC 290 determines imaging conditions for each production lot based on the features thus extracted. For example, when the PC 290 extracts a feature that the work W inspected first in a certain production lot has a relatively bright color, in subsequent reinspections of the works W in the same production lot, the imaging unit 70 Dim the light emitted from

In addition, the PC 290 may change the determination conditions (parameters regarding determination) of the work W at the time of reinspection (according to characteristics) for each manufacturing lot. The determination conditions for the workpiece W are the values and the contents of the arithmetic processing used in the quality determination (step S200), more specifically, the contents of the calculation processing by the quality determination DNN 296. FIG. A plurality of good/bad judgment DNNs 296 are constructed in the PC 290 so that the judgment conditions can be changed for each production lot. The plurality of pass/fail judgment DNNs 296 are trained using works W having different characteristics (the degree of light reflection, the color of the work W, etc.) as described above, and make pass/fail judgments according to the characteristics (and thus the manufacturing lot). executable.

When changing the judgment conditions, the PC 290 selects which of the plurality of pass/fail judgment DNNs 296 based on the features (such as color and defect extraction points) of the work W to be re-inspected first for each manufacturing lot. Determine if DNN 296 is to be used for retesting. The PC 290 thus changes the judgment conditions (good/bad judgment DNN 296) for each production lot, and inspects the workpiece W again.

As described above, the work W is manufactured by casting, and the inspection device 201 includes the PC 290 (extraction unit) for extracting the characteristics of the work W for each production lot of the work W, and based on the extraction results, A PC 290 (adjustment unit) that adjusts at least one of parameters related to imaging by the imaging unit 70 in the reinspection and parameters related to determination by the reinspection determination unit 294 for each production lot. be.

By configuring in this way, it is possible to optimize the parameters in the re-inspection according to the manufacturing lot, and it is possible to determine the quality of the work W more accurately.

Note that the PC 290 that adjusts the parameters for each production lot described above is an embodiment of the extractor and the adjuster.

Next, an inspection apparatus 301 and an inspection method according to the third embodiment will be described with reference to FIGS. 24 to 27. FIG.

The inspection apparatus 301 according to the third embodiment is significantly different from the inspection apparatus 1 according to the first embodiment in that it detects positional deviation of equipment (the multi-axis robot 40, the imaging unit 70, etc.) that occurs during inspection. First, the positional deviation of the equipment that occurs during inspection will be described.

When the workpiece W inspection (steps S100 to S200) described in the first embodiment is performed for a long period of time, the relative positions of the devices may gradually shift. FIG. 27(a) shows an example of a result B371 (imaging result of the first borescope 71) of imaging the first hole W11 in a state where the device is displaced. When the positional deviation of the device becomes large, the position of the center C311 of the first hole W11 or the like is greatly deviated from the center P371 of the imaging result B371 (image data) shown in FIG. Accuracy may deteriorate.

Therefore, the inspection apparatus 301 according to the third embodiment is configured so as to inspect the workpiece W and also to detect the positional deviation of the equipment described above. A specific description will be given below. In addition, about the member comprised similarly to the inspection apparatus 1 which concerns on 1st embodiment, the same code|symbol as 1st embodiment is attached|subjected, and the description is abbreviate|omitted.

An inspection apparatus 301 according to the third embodiment shown in FIG. 24 differs from the PC 90 (see FIG. 3) according to the first embodiment in the configuration of a control unit 380 and a PC 390. FIG. The control section 380 has an adjustment section 381 .

The adjuster 381 is for adjusting the positional deviation of the device. Processing of the adjustment unit 381 will be described later.

The PC 390 includes a correcting section 391 , a good/bad judging section 392 , a creating section 393 and a detecting section 394 .

The correction unit 391 is for correcting the gain (sensitivity) of the imaging unit 70, and is configured similarly to the correction unit 91 according to the first embodiment. The quality determination section 392 is for determining the quality of the work W, and is configured in the same manner as the quality determination section 92 according to the first embodiment. The creation unit 393 is for creating a heat map H (see FIG. 18(b)) indicating the degree of reaction of the quality determination DNN 396, which will be described later, and is configured in the same manner as the creation unit 93 according to the first embodiment. be.

The detection unit 394 is for detecting that the equipment (the multi-axis robot 40, the imaging unit 70, etc.) of the inspection apparatus 301 is misaligned. Processing of the detection unit 394 will be described later.

A plurality of DNNs (see FIG. 9) having an input layer, a plurality of intermediate layers and an output layer are constructed in the PC 390 configured as described above. Specifically, in the PC 390, a quality judgment DNN 396 and a position deviation detection DNN 397 are constructed.

The pass/fail judgment DNN 396 is for judging the pass/fail of the work W, and is configured in the same manner as the DNN 100 according to the first embodiment. That is, the pass/fail judgment DNN 396 learns in advance the relationship between the result B371 (see FIG. 27) of imaging the first hole W11 and the like (image pickup location) by the imaging unit 70 and the pass/fail judgment result. When the imaging result B371 from the imaging unit 70 is input, the quality determination DNN 396 can output a score indicating the probability of a non-defective product and a defective product and the determination result of the non-defective product and the defective product (FIG. 20(b) reference).

The positional deviation detection DNN 397 is for detecting the positional deviation of the device. The positional deviation detection DNN 397 is configured to be able to detect positional deviation based on the trend of positional deviation of the device.

Specifically, when the inspection is continuously performed and the positional deviation of the equipment gradually increases, the distance between the center P371 of the imaging result B371 shown in FIG. The distance L311 gradually increases. Therefore, the positional deviation detection DNN 397 performs predetermined learning in advance so as to be able to extract such tendency of positional deviation of the device (variation of the distance L311). For example, as shown in FIG. 25, a result B371 of imaging predetermined imaging locations for a plurality of workpieces W (for example, learning data A300 shown in FIG. ) and the presence or absence of positional deviation (answer) are input to the positional deviation detection DNN 397 (input layer), and the information is transferred to the output side. As a result, the positional deviation detection DNN 397 can extract the trend of positional deviation (variation of the distance L311) from the imaging results B371 captured in the last several tens of inspections, and determine the presence or absence of positional deviation. In this embodiment, the presence or absence of positional deviation is distinguished by whether or not the distance L311 (see FIG. 27A) exceeds a predetermined threshold. In other words, if the distance L311 is small enough to not hinder the inspection (below a predetermined threshold value), it is considered that no positional deviation has occurred. On the other hand, if the distance L311 is large (exceeds a predetermined threshold), it is assumed that a positional deviation has occurred.

The flow of inspection of the work W by the inspection device 301 will be described below.

As shown in FIG. 26, the inspection device 301 (control unit 380) performs steps S100 to S180 described in the first embodiment. That is, the inspection apparatus 301 moves the set work W to a position (imaging position) where the imaging unit 70 can image the set work W by the multi-axis robot 40, and the imaging unit 70 moves the first hole W11 of the work W (inspection location) is imaged (see FIG. 25(a)).

When the imaging unit 70 completes imaging, the detection unit 394 detects the presence or absence of positional displacement of the device using the positional displacement detection DNN 397 (step S310). If no positional deviation is detected (step S310: No), the work W is returned to the input position P1 (see FIG. 1) by steps S190 and S200 described in the first embodiment, and the quality determination unit 392 The quality of the workpiece W is determined.

On the other hand, if positional deviation is detected in step S310 (step S310: Yes), the detection unit 394 calculates the direction and amount of positional deviation based on the imaging result B371 (step S320). The detection unit 394 outputs the calculation result to the display device (liquid crystal display, etc.) of the PC 390 and also transmits a signal to the adjustment unit 381 to notify the calculation result.

After calculating the direction of the positional deviation and the like, the adjustment unit 381 adjusts the positional deviation (robot coordinates) based on the calculation result of the detection unit 394 (step S330).

When the adjustment of the positional deviation is completed, the imaging unit 70 images the workpiece W in the same manner as in steps S170 and S180 (step S340).

In this way, the control unit 380 and the quality determination unit 392 determine the quality of the work W based on the imaging result B371 after the positional deviation adjustment, and return the work W to the input position P1 (step S310: No, steps S190 and S200). .

A specific example of detecting and correcting the positional deviation (steps S310 to S330) will be described below with reference to FIG.

As described above, FIG. 27(a) is the result B371 (imaging result of the first borescope 71) of imaging the first hole W11 in a state where the device is displaced. As shown in FIG. 27(a), when the positional deviation occurs, the center C311 of the first hole W11 shifts from the center P371 of the imaging result B371. Further, as the positional deviation of the device gradually increases, the distance L311 between the centers P371 and C311 gradually increases.

In step S310 (detection of positional deviation), the detection unit 394 inputs the result B371 of imaging the first hole W11 in the last several tens of inspections to the positional deviation detection DNN 397 to increase the distance L311, that is, Check for misalignment trends. The misalignment detection DNN 397 extracts the tendency of misalignment and determines the presence or absence of misalignment. For example, if the positional deviation detection DNN 397 determines from the positional deviation tendency that the amount of deviation is large (the influence on the quality determination is large to some extent), it outputs the determination result that there is a positional deviation (step S310: Yes ).

In this case, in step S320, the detection unit 394 analyzes the imaging result B371 imaged in the current inspection (the most recent steps S170 and S180), and calculates the coordinates of the center C311 of the first hole W11. Then, the detection unit 394 calculates the distance L311 between the center P371 of the imaging result B371 and the center C311 of the first hole W11 based on the coordinate calculation result. The detection unit 394 also calculates the direction in which the center C311 of the first hole W11 is shifted from the center P371 of the imaging result B371 (direction of shift, leftward in FIG. 27A). The detection unit 394 outputs the direction of positional deviation and the distance L311 thus calculated to the display device. In this way, the detection unit 394 notifies the operator or the like of the calculation result such as the direction of the positional deviation.

After that, the adjustment unit 381 controls the multi-axis robot 40 so that the center P371 of the imaging result B371 and the center C311 of the first hole W11 are aligned based on the calculation result of the direction of the positional deviation and the amount of deviation (Fig. 27(b), step S330). In this way, the adjusting unit 381 adjusts (corrects) the positional deviation of the device by correcting the robot coordinates. In this way, the processing of detection and adjustment of positional deviation by the detection unit 394 and the adjustment unit 381 is completed.

In this way, the detection unit 394 can detect the positional deviation of the device using the imaging result B371 used to determine whether the work W is good or bad. As a result, positional deviation can be detected without interrupting the inspection, and workability can be improved. In particular, in this embodiment, the presence or absence of positional deviation is determined based on the trend of change in positional deviation rather than simply the amount of positional deviation (distance L311 (see FIG. 27A)). A misalignment adjustment (step S330) can be performed.

Further, the detection unit 394 detects the positional deviation using the imaging result B371 after gain correction (after step S160). As a result, the detection unit 394 can accurately detect the positional deviation using the imaging result B371 (after gain correction) in which the first hole W11 can be easily confirmed. In addition, since the detection unit 394 can accurately calculate the center C311 of the first hole W11, it can accurately calculate the direction and amount of positional deviation.

Also, the pass/fail judgment unit 392 makes a pass/fail judgment based on the result B371 of the image of the work W after the adjustment of the positional deviation (steps S330, S340, and S200). As a result, the pass/fail determination unit 392 can improve the accuracy of the pass/fail determination.

As described above, the inspection apparatus 301 according to the third embodiment includes the multi-axis robot 40 (moving unit) that moves the workpiece W to the imaging position, and the imaging unit 70 that images the workpiece W at the imaging position. A control unit 380 and a pass/fail judgment unit 392 (inspection unit) for inspecting the work W by making a pass/fail judgment of the work W based on the result B371; and a detection unit 394 for detecting positional deviation of the imaging unit 70 and the work W based on B371.

By configuring in this way, it is possible to detect the positional deviation of the device. In addition, positional deviation can be detected without interrupting the inspection, and workability can be improved.

Further, the detection unit 394 detects a positional deviation using a positional deviation detection DNN 397 (learning model) that learns the relationship between the imaging results B371 of the plurality of workpieces W and the positional deviation.

By configuring in this way, positional deviation can be detected with high accuracy.

Further, when the positional deviation is detected (step S310: Yes), the detection unit 394 calculates the direction and amount of the positional deviation based on the imaging result B371, and notifies the calculation result. Yes (step S320).

With this configuration, the operator or the like can grasp the direction and amount of misalignment based on the notification result from the detection unit 394, so that the work to restore the misalignment (maintenance work) can be performed in a short time. be able to.

Moreover, the inspection apparatus 301 further includes an adjustment unit 381 that adjusts the positional deviation based on the calculation result of the direction of the positional deviation and the amount of deviation.

By configuring in this way, it becomes possible to image the workpiece W by aligning the positions of the workpiece W and the imaging unit 70 (steps S330 and S340), and improve the accuracy of quality determination.

Further, a correction unit 391 (second correction unit) that corrects the gain of the imaging unit 70 for each workpiece W based on the imaging result B371 of the imaging unit 70 is further provided, and the detection unit 394 performs the correction The positional deviation is detected based on the result of imaging the work W with the gain corrected by the section 391 .

By configuring in this way, the workpiece W can be easily confirmed in the imaged result B371, so that the positional deviation can be detected with high accuracy.

Further, as described above, the inspection method according to the third embodiment is such that the image of the work W is imaged by the imaging unit 70, and the quality of the work W is determined by the quality determination unit 392 (determination unit) based on the imaged result B371. and an inspection process (steps S170 to S200) for inspecting the work W, and the positional deviation of the imaging unit 70 and the work W based on the imaging result B371 of the plurality of works W imaged by the imaging unit 70. and a detection step of detecting (step S310).

By configuring in this way, it is possible to detect the positional deviation of the device.

It should be noted that the multi-axis robot 40 according to the third embodiment is one embodiment of the moving unit.
Also, the control unit 380 and the pass/fail determination unit 392 according to the third embodiment are an embodiment of the inspection unit.
Also, the positional deviation detection DNN 397 according to the third embodiment is an embodiment of a learning model.
Also, the correction unit 391 according to the third embodiment is an embodiment of the second correction unit.
Also, the pass/fail determination unit 392 according to the third embodiment is an embodiment of the determination unit.

Although the third embodiment of the present invention has been described above, the present invention is not limited to the above configuration, and various modifications are possible within the scope of the invention described in the claims.

For example, the positional deviation detection DNN 397 is learned by supervised learning (see FIG. 25(b)), but the learning method is not particularly limited. can be anything.

Moreover, although the detection unit 394 detects the positional deviation using the positional deviation detection DNN 397, the method of detecting the positional deviation is not particularly limited. For example, the detection unit 394 may detect misalignment using a neural network with one intermediate layer. Further, the detection unit 394 may detect the positional deviation using a pre-captured reference image instead of the neural network. In this case, the detection unit 394 can, for example, extract the difference between the reference image and the imaging result B371, and detect the positional deviation based on the extraction result.

Also, the PC 390 (the detection unit 394) calculated the direction and amount of misalignment based on the imaging result B371 (see FIG. 27A), but the method of calculating the direction and the like of the misalignment is not particularly limited. not something. For example, the PC 390 may calculate the direction of positional deviation using the production lot in addition to the imaging result B371. A specific description will be given below.

When the work W is manufactured by casting, for example, a different production lot is assigned to each ladle. When the PC 390 inspects the work W of the given manufacturing lot for the first time, the PC 390 determines the characteristics of the work W corresponding to the manufacturing lot (for example, the center P371 and the positional relationship of the center C311 of the first hole W11, etc.) are extracted.

The PC 390 corrects the direction of positional deviation and the like for each production lot based on the features thus extracted. For example, the PC 390 extracts the feature that the center C311 of the first hole portion W11 is deviated from the center P371 of the imaging result B371 in a predetermined direction by a predetermined distance from the work W inspected first in a certain manufacturing lot. In this case, the direction and amount of positional deviation calculated in step S320 are corrected based on the feature in the subsequent inspection of the workpieces W of the same production lot.

As described above, the work W is manufactured by casting, and the inspection device 301 includes the PC 390 (extraction unit) for extracting the characteristics of the work W for each manufacturing lot of the work W, and based on the extraction results, A PC 390 (first correction unit) that corrects the calculation result of the direction of the positional deviation and the amount of deviation for each manufacturing lot.

With this configuration, it is possible to accurately calculate the direction and amount of positional deviation.

It should be noted that the PC 390 that adjusts the direction of positional deviation and the like for each manufacturing lot described above is an embodiment of the extraction unit and the first correction unit.

Note that the PC 390 does not necessarily need to calculate the direction and amount of positional deviation. For example, it is possible to detect only the presence or absence of positional deviation (step S310) and notify the operator accordingly. A worker who has received the notification can check the device and confirm the direction of the positional deviation. Also, the PC 390 does not necessarily need to include the correction section 391 for correcting the gain. Also, the controller 380 does not necessarily have to include the adjuster 381 . For example, it is also possible for an operator to manually correct (adjust) gains and positional deviations.

Next, an inspection apparatus 401 and an inspection method according to a fourth embodiment will be described with reference to FIGS. 28 to 30. FIG.

The inspection apparatus 401 according to the fourth embodiment is similar to the inspection apparatus according to the first embodiment in that it determines whether a positional deviation of the equipment such as the multi-axis robot 40 and the imaging unit 70 has occurred when the equipment shakes. Very different from 1. First, the displacement of the equipment caused by shaking will be described.

For example, when a relatively large tremor such as an earthquake occurs, there is a possibility that the position of each device such as the multi-axis robot 40 may be shifted. When the workpiece W is inspected in this state, the position of the center of the first hole W11 and the like is greatly deviated from the center of the imaging result B471 of the imaging unit 70 (see FIG. 27(a)), and the accuracy of the quality judgment is reduced. It can get worse.

Therefore, the inspection apparatus 401 according to the fourth embodiment is configured so as to inspect the workpiece W and also to detect the displacement of the equipment caused by the shaking described above. A specific description will be given below. In addition, about the member comprised similarly to the inspection apparatus 1 which concerns on 1st embodiment, the same code|symbol as 1st embodiment is attached|subjected, and the description is abbreviate|omitted.

An inspection apparatus 401 according to the fourth embodiment shown in FIG. 28 differs from the inspection apparatus 1 (see FIG. 3) according to the first embodiment in that it includes an acceleration sensor 500 and that the configuration of the PC 490 is different.

An acceleration sensor 500 shown in FIG. 28 is for detecting shaking of equipment. The acceleration sensor 500 is provided at a predetermined location or the like of the partition member 10 (see FIG. 1). The acceleration sensor 500 can detect the magnitude and direction of shaking. The acceleration sensor 500 is connected to the control unit 80 and can transmit a shake detection result B500 to the control unit 80 . Further, the acceleration sensor 500 can transmit the shake detection result B500 to the PC 490 via the control unit 80 .

The PC 490 includes a correcting section 491 , a good/bad determining section 492 , a creating section 493 , a positional deviation determining section 494 and an estimating section 495 .

The correction unit 491 is for correcting the gain (sensitivity) of the imaging unit 70, and is configured similarly to the correction unit 91 according to the first embodiment. The quality determining section 492 is for determining the quality of the work W, and is configured in the same manner as the quality determining section 92 according to the first embodiment. The creation unit 493 is for creating a heat map H (see FIG. 23(a)) indicating the degree of reaction of the quality determination DNN 497, which will be described later, and is configured in the same manner as the creation unit 93 according to the first embodiment. be.

The positional deviation determination section 494 is for determining whether or not there is a positional deviation of the device. The processing of the positional deviation determination unit 494 will be described later.

The estimator 495 is for estimating information about positional deviation. The information about the positional deviation is information for indicating how the positional deviation occurs, specifically, the device in which the positional deviation occurred, the direction of the deviation, the amount of deviation, and the like. Processing of the estimation unit 495 will be described later.

A plurality of DNNs (see FIG. 9) having an input layer, a plurality of intermediate layers and an output layer are constructed in the PC 490 configured as described above. Specifically, in the PC 490, a quality determination DNN 497, a position deviation determination DNN 498, and an information estimation DNN 499 are constructed.

The pass/fail judgment DNN 497 is for judging the pass/fail of the work W, and is configured in the same manner as the DNN 100 according to the first embodiment. That is, the pass/fail judgment DNN 497 learns in advance the relationship between the result B471 (see FIG. 29(a)) of imaging the first hole W11 and the like (image pickup location) by the imaging unit 70 and the pass/fail judgment result. In addition, when the imaging result B471 of the imaging unit 70 is input, the good/bad judgment DNN 497 can output a score indicating the probability of a non-defective product or a defective product and the judgment result of a non-defective product or a defective product (FIG. 20 ( b) see).

The positional deviation determination DNN 498 is for determining whether or not the positional deviation of the equipment occurs when shaking occurs. As described above, when the device is misaligned, the center of the first hole W11 and the like shifts from the center of the imaging result B471. At this time, it is considered that there is a correlation between the characteristics of the shaking (direction, magnitude, etc.) and the displacement of the device (direction, magnitude, etc.). Therefore, the positional deviation determination DNN 498 is appropriately trained so as to be able to detect the positional deviation of the device from such shaking characteristics.

More specifically, as shown in FIG. 29, it includes a detection result B500 (the direction and magnitude of shaking) of the acceleration sensor 500 and an imaging result B471 (for example, the imaging result of the first imaging location of the first hole W11). The learning data A400 and the presence/absence of positional deviation (answer) are input to the positional deviation determination DNN 498, and the information is transferred to the output side. As a result, the positional deviation determination DNN 498 extracts the trend of positional deviation (such as deviation of the center according to the direction of shaking) from the shake detection result B500 and the imaging result B471 (imaging result after the shaking has stopped), and determines the positional deviation. It is possible to determine whether or not has occurred.

The information estimation DNN 499 is for estimating the above-described information regarding the positional deviation (equipment in which the positional deviation has occurred, etc.). The information estimation DNN 499 is configured to be capable of estimating information related to positional deviation by focusing on the way of shaking. A specific description will be given below.

It is considered that there is a correlation between the characteristics of the shaking (direction, magnitude, etc.) and the equipment that causes the displacement (which equipment is likely to cause the displacement in response to the shaking). For example, since the first borescope 71 and the like are provided so as to hang down from above (see FIG. 29(a)), they are likely to be displaced when shaken sideways. Therefore, the information estimating DNN 499 is appropriately trained so as to be able to extract the trend of device position deviation (features of devices that are likely to be displaced) according to the way of shaking. Specifically, the learning data A400 and the information (answer) on the positional deviation described above are input to the information estimation DNN 499 (input layer), and the information is transferred to the output side (see FIG. 29(b)). As a result, the information estimation DNN 499 is capable of extracting the tendency of equipment positional deviation according to the way of shaking (equipment that is prone to shaking, etc.) and estimating information about the positional deviation.

The inspection device 401 configured as described above performs determination processing as shown in FIG. 30 when the acceleration sensor 500 detects shaking. The determination process will be described below.

The determination processing is processing for determining whether or not there is a positional deviation of the device. The determination process is performed when the acceleration sensor 500 detects a shake having a magnitude exceeding a predetermined threshold. Further, the determination process is performed by interrupting the inspection of the workpiece W (steps S100 to S200 described in the first embodiment).

When the acceleration sensor 500 detects the shaking, the control unit 80 controls the transport slider 20, the lifting slider 30, the multi-axis robot 40, the misalignment correction unit 50 until the shaking stops (until the acceleration sensor 500 stops detecting the shaking). , the operations of the interference check unit 60 and the imaging unit 70 are temporarily stopped (step S410).

When the shaking stops, the control unit 80 restarts imaging of the workpiece W by the imaging unit 70 (step S420). At this time, the control unit 80 resumes the processes other than the pass/fail determination (step S200) among the above-described steps S100 to S200. That is, the control unit 80 carries out the conveyance of the workpiece W (step S110), the correction of the gain by the correction unit 491 (step S160), and the like described in the first embodiment. Then, the control unit 80 moves to a position (imaging position, see FIG. 29A) where the work W can be imaged by the multi-axis robot 40 or the like, images the work W with the imaging unit 70, and places the work W at the input position. Return to P1 (see FIG. 1) (step S190).

Thus, in step S<b>420 , the quality determination unit 492 suspends the quality determination of the work W based on the instruction from the control unit 80 . In addition, the positional deviation determination unit 494 counts the number of works W that have been held based on instructions from the control unit 80 (hereinafter referred to as “holding number”).

After that, the positional deviation determination unit 494 determines the presence or absence of positional deviation using the positional deviation determination DNN 498 (step S430). At this time, the positional deviation determination unit 494 inputs the shaking detection result B500 and the imaging result B471 imaged in step S420 to the positional deviation determination DNN 498 . The positional deviation determination DNN 498 extracts the aforementioned positional deviation tendency (center deviation according to the direction of shaking, etc.) from the input imaging result B471, and outputs a determination result as to whether or not positional deviation has occurred. .

In step S430, the estimating unit 495 uses the information estimation DNN 499 to estimate information about the positional deviation, that is, the device in which the positional deviation occurred, the amount of deviation, the direction of the deviation, and the like. At this time, the estimation unit 495 inputs the shake detection result B500 and the imaging result B471 imaged in step S420 to the information estimation DNN 499, and estimates information about the positional deviation.

After determining the positional deviation, the PC 490 outputs the positional deviation determination result to the display device and notifies the operator or the like (step S440). In addition, the PC 490 also outputs to the display device that the estimated result of the information on the positional deviation and the quality determination are pending.

Upon receiving the notification, the operator can check the positional deviation determination result and the like displayed on the display device, and perform inspection and adjustment of each device. The operator checks each device, etc., and if it is judged that the quality judgment can be resumed, the operator appropriately operates the input device provided in the control unit 80 to issue an instruction to restart the quality judgment (step S450). : Yes). As a result, the quality determination unit 492 determines the quality of the work W whose quality determination has been suspended, and then continues the inspection of the work W (steps S100 to S200).

On the other hand, if there is no instruction from the operator for a predetermined period of time (step S450: No), the positional deviation determination unit 494 checks whether the number of pending positions exceeds a predetermined threshold value (eg, 10) (step S470). If the pending number is equal to or less than the predetermined threshold (step S470: No), the image of the workpiece W is continuously captured by the imaging unit 70, and the pending number increases. Then, the positional deviation determination unit 494 waits for an instruction from the operator (steps S420 to S440).

On the other hand, when the number of reservations exceeds the predetermined threshold (step S470: Yes), the positional deviation determining unit 494 checks whether the amount of deviation of each work W estimated in step S430 is less than the predetermined threshold. (step S480). If the deviation amount of each workpiece W is less than the predetermined threshold (step S480: Yes), the control section 80 restarts the quality determination by the quality determination section 492 (step S460). In this way, when the amount of deviation is small even if shaking occurs, that is, when it is considered that the accuracy of quality judgment will not deteriorate, the inspection of the workpiece W is resumed without receiving an instruction from the operator.

On the other hand, if the amount of deviation is greater than or equal to the predetermined threshold value (step S480: No), the controller 80 stops the movement of the transport slider 20 and the like (step S490). In addition, the control unit 80 causes a predetermined display unit to display a message indicating that the operation has been stopped, information regarding positional deviation, and the like, and outputs a warning sound from the speaker to notify the operator that the operation has been stopped. Thus, the determination processing ends.

As described above, in the determination process, it is possible to confirm whether or not a positional deviation has occurred when the device or the like shakes (step S430). As a result, when a positional deviation occurs, it is possible to quickly deal with the positional deviation (steps S450 to S490).

Further, the positional deviation determination unit 494 determines the positional deviation of the device using the imaging result B471 after the gain correction (after step S160). Accordingly, the positional deviation determination unit 494 can accurately determine the positional deviation using the imaging result B471 (after gain correction) that makes it easy to confirm the first hole W11. In addition, the estimating unit 495 can easily estimate the information about the positional deviation by using the imaging result B471 after the gain correction.

As described above, the inspection apparatus 401 according to the fourth embodiment includes the multi-axis robot 40 (moving unit) that moves the workpiece W to a predetermined position, and the imaging unit 70 that captures an image of the workpiece W that has been moved to the predetermined position. , a pass/fail judgment unit 492 (first judgment unit) that judges the quality of the work W based on the image pickup result B471 of the image pickup unit 70, and a shake of equipment including the multi-axis robot 40 and the image pickup unit 70 is detected. It is determined whether or not the device is displaced based on the acceleration sensor 500 (detection unit) and the result of imaging the work W by the imaging unit 70 when the acceleration sensor 500 detects shaking. and a positional deviation determination unit 494 (second determination unit).

With this configuration, it is possible to check the positional deviation of the device when the acceleration sensor 500 detects a shake.

Further, the positional deviation determination unit 494 determines whether or not a positional deviation has occurred using a positional deviation determination DNN 498 (first learning model) that has learned the relationship between the imaging result B 471 and the positional deviation of the device. It is.

By configuring in this way, it is possible to extract the feature of the positional deviation from the imaging result B471 and determine the positional deviation with high accuracy.

Further, the positional deviation determination unit 494 determines whether the positional deviation of the equipment has occurred based on the result B471 of the imaging unit 70 imaging the first hole W11 (hole) formed in the work W. It is determined whether or not (steps S420 to S440).

By configuring in this way, it is possible to determine the positional deviation using the work W to be subjected to quality determination, thereby improving convenience.

Further, a control unit 80 is provided for controlling the operations of the multi-axis robot 40, the imaging unit 70, the quality determination unit 492, and the positional deviation determination unit 494. The control unit 80 controls the motion of the acceleration sensor 500. is detected, the quality determination unit 492 suspends the quality determination of the workpiece W, and the positional deviation determination unit 494 determines whether or not the device is misaligned (step S420-S440).

By configuring in this way, it is possible to prevent the quality determination result of the work W from being output in a state in which positional deviation has occurred, and to suppress deterioration in the accuracy of quality determination.

Further, when the number of works W for which quality judgment is suspended reaches a predetermined number (step S470: Yes), the control unit 80 controls the multi-axis robot 40, the imaging unit 70, and the quality judgment unit 492. is stopped (step S490).

By configuring in this way, it is possible to prevent the number of works W for which quality judgment is suspended from becoming excessively large.

Further, the positional deviation determination unit 494 calculates the amount of deviation of the equipment based on the imaging result B471 of the imaging unit 70 (step S430), Even if the number reaches the predetermined number, if the calculation result of the deviation amount of the device is less than the predetermined threshold value (step S470: Yes, step S480: Yes), the quality determination unit 492 determines the quality of the work W. It is to be restarted (step S460).

By configuring in this way, it is possible to smoothly restart the quality determination of the workpiece W when the positional deviation is relatively small (less than the predetermined threshold value).

Further, the positional deviation determination section 494 determines whether or not the positional deviation of the device has occurred based on the imaging result B471 of the imaging section 70 and the detection result B500 of the acceleration sensor 500 .

By configuring in this way, by using the shake detection result B500 in addition to the imaging result B471, positional deviation can be detected with high accuracy.

Further, an estimating unit 495 is provided for estimating information about the positional deviation including the device in which the positional deviation occurred, the direction of the deviation, and the amount of the deviation based on the detection result B500 of the acceleration sensor 500 .

With this configuration, it is possible to provide the operator with information for determining which device is displaced and to what degree (information on positional displacement), so that the work of restoring the positional displacement can be performed in a short time. can be done.

Also, the estimation unit 495 estimates the information on the positional deviation using the information estimation DNN 499 (second learning model) that learns the relationship between the detection result B500 of the acceleration sensor 500 and the information on the positional deviation. be.

By configuring in this way, it is possible to extract features from the detection result B500 and accurately estimate information related to positional deviation.

Further, a correction unit 491 for correcting the gain of the imaging unit 70 when the workpiece W is imaged based on the imaging result B471 of the imaging unit 70 is further provided, Based on the result of imaging with the gain corrected in 491, it is determined whether or not positional deviation has occurred (steps S420 to S440).

With this configuration, it becomes easier to check the workpiece W in the imaging result B471, so it is possible to accurately determine whether or not there is any positional deviation.

Further, as described above, the inspection method according to the fourth embodiment includes the moving step (steps S100 to S180) of moving the work W to a predetermined position by the multi-axis robot 40, and moving the work W moved to the predetermined position. Imaging step (steps S170 and S180) of imaging by the imaging unit 70, and first judgment for performing quality judgment of the work W by the quality judgment unit 492 (first judgment unit) based on the imaging result B471 of the imaging unit 70 a detection step of detecting shaking of equipment including the multi-axis robot 40 and the imaging unit 70 with an acceleration sensor 500; and a second determination step (step S430) in which a positional deviation determination unit 494 (second determination unit) determines whether or not a positional deviation of the device has occurred based on the result of imaging by 70 .

With this configuration, it is possible to check the positional deviation of the device when the acceleration sensor 500 detects a shake.

It should be noted that the multi-axis robot 40 according to the fourth embodiment is one embodiment of the moving unit.
Also, the quality determination unit 492 according to the fourth embodiment is an embodiment of the first determination unit.
Also, the acceleration sensor 500 according to the fourth embodiment is an embodiment of the detection unit.
Also, the positional deviation determination unit 494 according to the fourth embodiment is an embodiment of the second determination unit.
Also, the positional deviation determination DNN 498 according to the fourth embodiment is an embodiment of the first learning model.
Also, the information estimation DNN 499 according to the fourth embodiment is an embodiment of the second learning model.

Although the fourth embodiment of the present invention has been described above, the present invention is not limited to the above configuration, and various modifications are possible within the scope of the invention described in the claims.

For example, the PC 494 uses two individually prepared DNNs (the positional deviation determination DNN 498 and the information estimation DNN 499) to individually determine the positional deviation and estimate the information on the positional deviation, but the present invention is limited to this. A common DNN may also be used to determine the positional deviation and estimate the information on the positional deviation.

Further, although the positional deviation determination unit 494 determines the positional deviation using the positional deviation determination DNN 498 (step S430), the method of determining the positional deviation is not particularly limited. For example, the misalignment determining unit 494 may determine misalignment using a neural network with one intermediate layer. Further, the positional deviation determination unit 494 may detect the positional deviation using a pre-captured reference image instead of the neural network. In this case, the positional deviation determination unit 494 can, for example, extract the difference between the reference image and the imaging result B471, and detect the positional deviation based on the extraction result.

Moreover, although the estimation unit 495 estimated the information about the positional deviation using the information estimation DNN 499 (step S430), the method of estimating the information about the positional deviation is not particularly limited. For example, the estimating unit 495 may estimate information about the positional deviation using a neural network with one hidden layer. Alternatively, the estimating unit 495 may estimate the information about the positional deviation using a pre-captured reference image instead of the neural network. In this case, the estimation unit 495 can, for example, extract the difference between the reference image and the imaging result B471, and estimate the information regarding the positional deviation based on the extraction result.

In addition, other configurations are not particularly limited as long as the PC 494 can determine the positional deviation. Therefore, the PC 494 does not necessarily have the corrector 491 for correcting the gain and the estimator 495 for estimating information about the positional deviation.

Further, although the positional deviation determination unit 494 determines the positional deviation using the work W to be judged as good or bad, the object to be imaged when judging the positional deviation does not necessarily have to be the work W to be judged as good or bad. For example, the positional deviation determination unit 494 may image the workpiece W (reference object) that serves as a reference when adjusting (teaching) the multi-axis robot 40 to determine the positional deviation.

As described above, the positional deviation determination section 494 determines whether or not the positional deviation of the device has occurred, based on the result of the imaging of the reference object, which serves as a reference for determining the positional deviation of the device, by the imaging section 70 . is determined.

By configuring in this way, positional deviation can be determined with high accuracy.

Also, in the determination process, when the number of reservations exceeds a predetermined threshold, the amount of deviation is confirmed (step S470: Yes, step S480), but the processing when the number of reservations exceeds the predetermined threshold It is not particularly limited. For example, when the number of reservations exceeds a predetermined threshold (step S470: Yes), the operation may be stopped regardless of the amount of deviation.

40 multi-axis robot 70 imaging unit 80 control unit (inspection unit)
301 inspection device 392 pass/fail determination unit (inspection unit)
394 Detector B371 Imaging result W Work

Claims (1)

a moving step of moving the workpiece to an imaging position by a moving unit configured by a multi-axis robot;
By moving the work to the imaging position by the moving part, the imaging part is inserted into the hole formed in the work from one side and the other side of the hole, and the imaging part is inserted into the hole. , an imaging step of imaging a predetermined imaging portion of the hole from each of the one side and the other side by the imaging unit;
an inspection step of inspecting the work by determining whether the work is good or bad by a determination unit based on the imaging result of the imaging unit;
a detection step of detecting a positional deviation between the imaging unit and the workpiece based on imaging results of the plurality of workpieces captured by the imaging unit;
including,
Inspection methods.

Images