JP2021018064A - Visual appearance inspection method and visual appearance inspection device - Google Patents

Abstract

To provide a visual appearance inspection method for detecting a defect generated in the visual appearance of an inspection object, the visual appearance inspection method which can stably perform highly accurate defect detection in a visual appearance inspection device, and the visual appearance inspection device.SOLUTION: A visual appearance inspection method for inspecting a defect on the visual appearance of an inspection object by using a luminance abnormal part extraction model generated by deep learning comprises: a first step of acquiring a plurality of images by imaging the inspection object while changing a lighting condition; a second step of extracting a luminance abnormal part generated in the visual appearance of the inspection object by inputting the plurality of images to the luminance abnormal part extraction model; a third step of acquiring a shape feature amount by measuring the shape of the luminance abnormal part; and a fourth step of determining whether or not the luminance abnormal part is defective by evaluating the shape feature amount.SELECTED DRAWING: Figure 2

Description

The present invention relates to a visual inspection method and a visual inspection apparatus.

Conventionally, there has been known a technique for inspecting the appearance of an inspected object by using an image of an inspected object such as a product or a product part.
In particular, recently, the feature quantities extracted by machine learning from a plurality of learning images including non-defective products and defective products are classified and compared with the feature quantities of the inspected object included in the image of the inspected object. A technique for automatically determining the quality of an inspected object by evaluation without human intervention has attracted attention and is being developed (see, for example, Patent Document 1).
At this time, a classification model (classifier) is usually used when classifying non-defective products and defective products, but in order to improve the classification accuracy of the classification model, a deep layer based on a plurality of learning images including a defect image is used. A classification model is generated by learning (see, for example, Patent Document 2).

Japanese Unexamined Patent Publication No. 2017-21159 JP-A-2018-005640

However, since it is common to use a two-dimensional image taken from a predetermined direction using one camera and a plurality of lights as a learning image, a blind spot may occur or a defective part may appear. There was a situation where it was not converted. Therefore, the amount of information in the learning image was insufficient, and it was not possible to extract and classify a sufficient amount of features even by using deep learning.
As a result, in the detection of defects generated in the appearance of the object to be inspected, there are many over-detections and oversights, and there is a problem that the quality determination accuracy of the inspected object in the appearance inspection is lowered.

Therefore, in the present invention, the visual inspection method for detecting defects generated in the appearance of the object to be inspected, the visual inspection method capable of stably and highly accurate defect detection in the visual inspection apparatus, and the appearance. The purpose is to provide an inspection device.

In order to achieve the above-mentioned object, the first invention is an appearance inspection method for inspecting an appearance defect of an object to be inspected by using a luminance abnormality part extraction model generated by deep learning, and lighting conditions. The brightness generated in the appearance of the object to be inspected by the first step of acquiring a plurality of images by imaging the object to be inspected while changing the above and by inputting the plurality of images into the luminance abnormality portion extraction model. The second step of extracting the abnormal portion, the third step of acquiring the shape feature amount by measuring the shape of the brightness abnormality portion, and the evaluation of the shape feature amount indicate whether the shape feature amount is defective. It is a visual inspection method characterized by having a fourth step of determining whether or not it is present.

The second invention is an appearance inspection device for inspecting an appearance defect of an object to be inspected by using a luminance abnormality part extraction model generated by deep learning, and includes a lighting device, an imaging device, and the above. A gripping device for gripping an object to be inspected, a transporting device for transporting the object to be inspected gripped by the gripping device, a shape measuring device, the lighting device, the imaging device, the gripping device, the transporting device, and the above. The control device for controlling the shape measuring device and the arithmetic device are provided, and the lighting device, the imaging device, the gripping device, and the transport device are operated by the control device, and the inspection is performed while changing the lighting conditions. An image acquisition means for capturing an object, a brightness abnormality extraction means for extracting a brightness abnormality generated in the appearance of the object to be inspected by using the brightness abnormality extraction model based on the plurality of images, and the brightness abnormality extraction means. The shape measuring device is operated by the control device, the shape of the luminance abnormality portion is measured, the feature amount acquisition means for acquiring the shape feature amount, and the shape feature amount is evaluated, and the shape feature amount is defective. The visual inspection apparatus is provided with a second defect determining means for determining whether or not the defect is present.

According to the present invention, in the visual inspection method and the visual inspection apparatus, ambiguous judgments due to insufficient deep learning are eliminated, so that there are extremely few over-detections and oversights in the defect detection of the inspected object. Therefore, it is possible to suppress a decrease in the quality judgment accuracy in the visual inspection.
Therefore, the present invention is extremely useful for stably and accurately inspecting the appearance of various products and product parts.

The whole schematic view of the visual inspection apparatus 1. The flowchart of the appearance inspection by the appearance inspection apparatus 1.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is an overall schematic view of the visual inspection device 1. The visual inspection device 1 includes an imaging device 2, a lighting device 3, a shape measuring device 4, a transport device 5, a gripping device 6, a pedestal 7, a support column 8, and a control device 9 for controlling these. The wiring and the like are not shown.

Here, in the visual inspection device 1, the support column 8 is installed on the pedestal 7, the image pickup device 2 is attached to the uppermost portion of the support column 8, and the lighting device 3 is arranged under the image pickup device 2. An image can be taken by illuminating 10 with light from above, and a shape measuring device 4 and a transport device 5 provided with a gripping device 6 are installed on the support column 8.

The visual inspection device 1 is a device that detects defects on the surface of a complicated shape such as a cast joint as the object 10 to be inspected.

The image pickup device 2 is, for example, a digital camera, and is capable of taking an image of the object 10 to be inspected through a lens (not shown).

The lighting device 3 is, for example, a white LED. The plurality of lighting devices 3 have different positions and irradiation directions from each other so that light can be irradiated toward the object 10 to be inspected. That is, the object 10 to be inspected can be irradiated with light from all directions. In the illustrated example, the lighting device 3 uses four lighting devices 3 to image the object 10 to be inspected.

The shape measuring device 4 is, for example, an optical cutting type three-dimensional measuring device. By scanning the object to be inspected 10, the surface shape of the object to be inspected 10 can be measured.

The transport device 5 is, for example, an articulated robot. The object 10 to be inspected is gripped by the gripping device 6 attached to the tip of the transfer device 5, so that the object 10 can be conveyed to the imaging position or the shape measurement position in a posture capable of imaging and measuring the shape of the entire surface of the object 7 to be inspected. ing.

The gripping device 6 is, for example, a two-claw hand. The surface of the object to be inspected 10 is gripped at an arbitrary position so that the entire surface of the object to be inspected 10 can be in a posture capable of imaging and shape measurement.

The image pickup device 2, the lighting device 3, the shape measuring device 4, the transport device 5, and the gripping device 6 are connected to the control device 9 by wiring (not shown). The control device 9 is, for example, a computer, and includes input units such as a keyboard and a touch panel for inputting instructions such as setting various conditions and starting processing, and a memory, HDD, and HDD in which the operations and pass / fail judgment criteria of each unit are stored in advance. It has a storage unit such as an SSD, and a calculation / processing unit such as a CPU that performs various calculations and processes based on a program input in advance. In addition, the control device 9 can acquire an image from the image pickup device 2 by the communication control unit, and can also store the acquired image and the inspection result in the storage unit. Further, the control device 9 can output the inspection result or the like to an output unit such as a display or a printer.

More specifically, the calculation / processing unit of the control device 9 grips a predetermined position on the surface of the object to be inspected 10 by the gripping device 6, and conveys the object 10 to be inspected in a predetermined posture by the conveying device 5. The lighting device 3 is stopped and the lighting devices 3 are sequentially and individually emitted. Then, the calculation / processing unit of the control device 9 images the object 10 to be inspected by the image pickup device 2 in synchronization with the light emission of each lighting device 3. Further, the control device 9 can transport the object 10 to be inspected to a predetermined posture by the transport device 5 when the light emission of the lighting device 3 and the imaging of the image pickup device 2 are completed. By repeating this on the entire surface of the object 10 to be inspected, it is possible to capture a plurality of images under various imaging conditions.

Next, a method of visual inspection of the object to be inspected will be described.
FIG. 2 is a flowchart of a visual inspection method for determining the quality of defects generated in the appearance of an object to be inspected by using the visual inspection device 1.
First, the object to be inspected 10 is gripped by the gripping device 6, and the object to be inspected 10 is transported in a predetermined posture by the transfer device 5 and the gripping device 6.

After that, in step S01, the surface of the object to be inspected 10 is imaged from a plurality of directions under a plurality of imaging conditions combining the illumination direction and the posture, and a plurality of images are acquired.

In particular, the control device 9 sequentially causes the lighting devices 3 to emit light individually, and irradiates the object 10 to be inspected with light from each direction. At the same time, the control device 9 images the object 10 to be inspected by the image pickup device 2. For example, by using four lighting devices 3, it is possible to acquire four images having different imaging conditions in one posture.

Then, the same imaging operation is repeated each time the transport device 5 and the gripping device 6 are transported in a predetermined posture. For example, by changing the posture eight times, 32 images are acquired with 8 postures x 4 images.

In the next step S02, a plurality of images acquired in step S01 are input to the luminance abnormality part extraction model generated in advance by deep learning and stored in the storage unit of the control device 9, and the luminance of the object 10 to be inspected 10 is input. Extract the abnormal part.

In the next step S03, based on the detection result of step S02, it is determined whether or not the luminance abnormality portion is detected in all the images of the object 10 to be inspected. If the luminance abnormality part is not extracted from all the images of the object 10 to be inspected, the process proceeds to step S07, and it is determined that there is no defect, and the object 10 to be inspected is passed.

On the other hand, when a brightness abnormality portion is detected from an image of the object 10 to be inspected, the position information of the brightness abnormality portion in the image is stored in the storage unit of the control device 9, and the process proceeds to step S04.

In step S04, the shape feature amount of the luminance abnormality portion is measured by using the shape measuring device 4. The shape feature amount is, for example, a diameter, a depth, a height, or the like.

At this time, based on the position information of the brightness abnormality portion stored in the storage unit of the control device 9, the transfer device 5 and the gripping device 6 are used to convey the brightness abnormality portion to the shape measuring device 4 in a posture in which the brightness abnormality portion can be measured.

In the next step S05, the measured value related to the shape feature amount is compared with the reference value of the shape feature amount stored in advance in the storage unit of the control device 9, and at least one of the elements in the shape feature amount exceeds the reference value. If this is the case, the process proceeds to step S06, the abnormal brightness portion is determined to be defective, and the object to be inspected 10 is rejected as having a defect.

On the other hand, if the measured value is less than the reference value in all the elements in the shape feature amount, the process proceeds to step S07, and it is determined that there is no defect and the inspected object 10 is passed.

Although the embodiments of the invention have been described so far, the present invention is not limited to the above embodiments.
For example, in the visual inspection device 1 of the present embodiment, the shape measuring device 4 and the transport device 5 are installed on the support column 8, but they may be installed on the pedestal 7 separately from the support column 8.
Further, for example, although the visual inspection device 1 of the present embodiment includes a set of transport devices 5 and a gripping device 6, a plurality of sets of transport devices 5 and gripping devices 6 are installed to provide a plurality of objects 7 to be inspected. It may be handed over by a set of transporting device 5 and gripping device 6 to grip and transport.

1: Visual inspection device 2: Imaging device 3: Lighting device 4: Shape measuring device 5: Conveyor device 6: Gripping device 7: Pedestal 8: Support 9: Control device 10: Object to be inspected

Claims (2)

It is a visual inspection method that inspects the appearance defects of the object to be inspected by using the luminance abnormality extraction model generated by deep learning.
The first step of capturing an image of the object to be inspected while changing the lighting conditions to acquire a plurality of images, and
A second step of extracting the luminance abnormality portion generated in the appearance of the object to be inspected by inputting the plurality of images into the luminance abnormality portion extraction model, and
The third step of acquiring the shape feature amount by measuring the shape of the luminance abnormality portion, and
A visual inspection method comprising a fourth step of determining whether or not the luminance abnormality portion is a defect by evaluating the shape feature amount.
It is a visual inspection device that inspects defects in the appearance of the object to be inspected using the luminance abnormality extraction model generated by deep learning.
The lighting device, the image pickup device, the gripping device for gripping the object to be inspected, the transport device for transporting the object to be inspected gripped by the gripping device, the shape measuring device, the lighting device, the imaging device, and the above. A control device for controlling the gripping device, the transport device, and the shape measuring device, and a calculation device are provided.
An image acquisition means for imaging the object to be inspected while changing the lighting conditions by operating the lighting device, the imaging device, the gripping device, and the transporting device by the control device.
Based on the plurality of images, the luminance abnormality extraction model is used to extract the luminance abnormality portion generated in the appearance of the object to be inspected, and the luminance abnormality extraction means.
A feature amount acquisition means for operating the shape measuring device by the control device, measuring the shape of the luminance abnormality portion, and acquiring the shape feature amount.
An visual inspection apparatus comprising a second defect determining means for evaluating the shape feature amount and determining whether or not the luminance abnormality portion is a defect.