Appearance inspection is an inspection service that checks the appearance (appearance, external surface) of products and raw materials to ensure their quality. In the manufacturing industry, visual inspection detects defects such as dirt, foreign matter, scratches, and deformation of products and parts that occur during the manufacturing process to eliminate defective products.

The most basic visual inspection is a visual inspection by human eyes. Automation of visual inspection using inspection equipment is also being promoted at some sites.

AI visual inspection is a method of image inspection utilizing deep learning technology. In recent years, with the development of AI technology, AI has been increasingly used for visual inspection in the manufacturing industry.
AI-based visual inspection is a next-generation visual inspection method that has various advantages, including more accurate and faster inspection than visual inspection, easier setup than conventional inspection equipment, and higher inspection accuracy.

THE MAIN MECHANISM IS TO PROVIDE AI (ARTIFICIAL INTELLIGENCE) WITH IMAGES OF GOOD AND BAD PRODUCTS SO THAT IT CAN MAKE GOOD OR BAD DECISIONS JUST LIKE A HUMAN.

THIS SECTION DESCRIBES THE CHALLENGES OF VISUAL INSPECTIONS CONDUCTED BY VISUAL INSPECTION AND INSPECTION EQUIPMENT, AND WHAT CAN BE EXPECTED FROM AI VISUAL INSPECTIONS IN RESPONSE TO THESE CHALLENGES.

(1-1) Overview of visual inspection and issues

Visual inspection is the most basic visual inspection method. In visual inspection, products and parts are visually compared with limit samples to determine whether they are good or defective. In some cases, a magnifying glass or microscope is used to improve the inspection accuracy of minute parts.

Visual inspection has the advantage of not requiring inspection equipment and being easy to introduce. On the other hand, it has the following problems.

• Labor costs for inspections.
• Difficult to secure personnel to perform visual inspections due to labor shortage
• Quality of inspection depends on the skills of the inspector
• Variation occurs because the judgment criteria depend on the person.
• Takes longer to inspect, especially if detailed confirmation is required

(1-2) ADVANTAGES OF AI VISUAL INSPECTION OVER VISUAL INSPECTION ISSUES

Can reduce costs, solve human resource shortages, increase inspection capacity

AI VISUAL INSPECTION SYSTEMS CAN TAKE PICTURES AND MAKE JUDGMENTS INSTANTLY. IN ADDITION, THE MACHINES ARE "FATIGUE-FREE" AND CAN BE OPERATED 24 HOURS A DAY. ALSO, THE INSPECTION CAPACITY CAN BE DOUBLED OR TRIPLED BY ADDING MORE MACHINES. AS A RESULT, THE SYSTEM CAN SOLVE THE PROBLEMS OF HUMAN RESOURCE SHORTAGE AND HUMAN RESOURCE DEVELOPMENT.

Can improve inspection quality

VISUAL INSPECTIONS BY HUMANS ARE DIFFICULT TO MAINTAIN ACCURACY DUE TO INSPECTORS' ABILITY AND FATIGUE, BUT AI VISUAL INSPECTIONS CAN MAINTAIN A HIGH LEVEL OF INSPECTION ACCURACY. AI VISUAL INSPECTION, HOWEVER, CAN MAINTAIN A HIGH LEVEL OF INSPECTION ACCURACY.

(2-1) Overview and issues of inspection by inspection equipment

In some cases, inspection equipment is installed to automate inspection in order to solve the problems of visual inspection.

The inspection system uses a rule-based method to judge defective products. Rule-based" is a method in which judgment criteria are set in advance for the inspection system, and judgment is made based on whether or not the defect conforms to the rules. The rules can include the length, area, shape, color, etc., of the defective part.

While inspection devices that make rule-based decisions are one effective means of automating visual inspections, they also present the following challenges

• Clear rules must be established, and if there are variations in the inspection target, shape or color of defects, the rules cannot be established and inspection cannot be performed.
• If there are many types of anomalies, the settings become more complex. Also, the more complex the settings, the more over-detection (over-detection)
• Difficult to cope with the loss of know-how when setting up rules becomes a personal task

(2-2) ADVANTAGES OF AI VISUAL INSPECTION OVER INSPECTION EQUIPMENT ISSUES

No need to define clear rules, complex conditions can be handled

The most common method to build a visual inspection AI is to load image data of products and parts sorted into good and defective products into the AI and have it learn. Another method is to train on data of only good products, and then "assign defective products to those with features different from those of good products. AI that has learned classification criteria can automatically classify products and parts captured by a camera.
In the case of AI, there is no need to clearly define good and defective products, as AI can learn the "features" contained in the input data by itself and make a good/defective judgment. In addition, since AI can make complex judgments based on area, color, and other factors, it can handle cases that previously required complex settings.

CRITERIA CAN BE LEARNED AND STORED BY AI

SINCE AI VISUAL INSPECTION LEARNS THE CRITERIA FOR JUDGING GOOD OR BAD, IT IS LESS SUSCEPTIBLE TO CHANGES, TRANSFERS, OR RETIREMENTS OF INSPECTION PERSONNEL. IN ADDITION, INSPECTION DATA AND INSPECTION CASES CAN BE ACCUMULATED AND RELEARNED (ADDITIONAL LEARNING) BY THE AI TO ENABLE INSPECTIONS WITH HIGHER ACCURACY.

HERE WE DISCUSS THREE KEY SUCCESS FACTORS IN IMPLEMENTING AI TO AUTOMATE VISUAL INSPECTION.

(1) Shooting environment (lighting and cameras)

IN AI VISUAL INSPECTION, THE TARGET PRODUCT OR PART IS PHOTOGRAPHED AND THE DATA IS FED TO AI FOR LEARNING. THE IMPORTANT POINT WHEN PHOTOGRAPHING IS THAT "ABNORMALITIES ARE PROPERLY CAPTURED AS AN IMAGE. HERE ARE SOME POINTS TO CONSIDER.

Is shooting from one direction sufficient or is shooting from multiple directions necessary?

As the simplest example, if the object to be inspected is a flat surface or only one surface of a three-dimensional object, shooting from only one direction is sufficient.
On the other hand, if the object to be inspected is three-dimensional or has a complex shape, it is necessary to "increase the number of cameras and take pictures from multiple directions simultaneously" or "rotate the object to be inspected so that the surface to be inspected faces the camera". In a similar case, multiple cameras may be used when the inspection target is too large to be inspected with a single camera. For example, in the case of "InspectAI" provided by Araya, multiple cameras can be used to take pictures simultaneously, and if any defect is found anywhere, the product can be determined to be defective. We can also propose conveyors that rotate the inspected product.

Cases that cannot be successfully photographed using common photography methods

Although a camera (color or monochrome) and lighting are used to take photographs for general visual inspections, there are cases in which the simple method of taking photographs does not work well.
For example, in the case of metals and plastics, the shiny surface of the material may reflect the illumination, making it difficult to capture abnormalities (white spots). In such cases, the use of high-performance lighting manufactured by Keyence, for example, can solve the problem. In addition, Arayahas "light experts" who can suggest various ways to capture images with less reflection.
We can also handle other types of photography, such as using near-infrared light to transmit the surface or to highlight irregularities in the surface to be inspected.

Use equipment already installed as is

IF YOU ALREADY HAVE AN EXISTING INSPECTION MACHINE AND HAVE ISSUES WITH ACCURACY, YOU CAN CONSIDER REPLACING THE CAMERA AND LIGHTING OF THE INSPECTION MACHINE WITH AN AI SYSTEM THAT ONLY DETERMINES WHETHER THE MACHINE IS ACCEPTABLE OR NOT. IN THIS CASE, THE COST CAN BE REDUCED BECAUSE THERE IS NO NEED TO INSTALL NEW HARDWARE FOR THE CAMERA AND LIGHTING.

(2) CONSTRUCTION OF AI APPEARANCE INSPECTION MODEL

AI visual inspection requires a preliminary "learning" procedure before it can be used in operation. Traditionally, individual development required a very long preparation period for AI experts to assemble and learn AI models.
In recent years, packaged software packages have emerged that make AI visual inspection easy to implement and use. There are several points to consider in selecting such AI visual inspection software.

CAN YOU USE THE RIGHT TYPE OF AI VISUAL INSPECTION MODEL FOR YOUR INSPECTION?

THERE ARE VARIOUS TYPES OF "AI VISUAL INSPECTIONS," EACH WITH ITS OWN CHARACTERISTICS AND SUITABILITY FOR DIFFERENT INSPECTION TASKS. THE FOLLOWING IS AN INTRODUCTION TO TYPICAL AI MODELS.

• Good product learning model: A model that does not require images of defective products and can learn only with images of good products. It is best suited for inspecting products that are unlikely to have defective products. However, it may not be applicable to products with large variations in the appearance of good products.
• Segmentation model: A model that is good at detecting small defects and foreign objects.
• Object detection model: A well-balanced model that can detect larger defects at high speed.
• Classification model: A model that can handle defects in the entire product (distortion, color defects, etc.).

ArayaInspectAI" of "InspectAI" supports all of the above models, making it possible to handle a wide range of inspections.

How much defective product can be prepared

To build an AI visual inspection model, it is common to prepare a large number of good and defective products in advance; to improve the inspection accuracy of AI visual inspection, it is necessary to learn as many patterns of defective products as possible. However, the frequency of defective products is often low, and preparing a large number of defective products requires a great deal of time and effort.
ArayaInspectAI" of "InspectAI" can perform inspections without defective products by using the aforementioned "Good Product Learning Model". Even in cases where the "Good Product Learning Model" cannot be applied, InspectAI has a mechanism that enables highly accurate inspections with only one-tenth the number of defective images required for training in general AI visual inspections. This system is also recommended for those who have abandoned the introduction of AI visual inspection due to the small number of defective products.

(3) LINKAGE OF AI VISUAL INSPECTION WITH CONVEYORS AND REJECTION EQUIPMENT

Automation of inspections will be further enhanced if the AI visual inspection can automatically stop the line when it detects defective products or eject them to an abnormality lane by an exclusion device (sorting machine). In order to realize these goals, it is necessary to link with a PLC (a device that automatically controls equipment on a production line).
ArayaInspectAI's "InspectAI" can be linked to PLCs as a standard feature. We can also propose new elimination devices (sorting machines).

ArayaThe inspection targets and inspection items that can be handled by InspectAI, the AI visual inspection software provided by

foodstuff

InspectAI is a high-performance AI software that can detect even hairs mixed in food products. In addition to hair, InspectAI can simultaneously inspect the appearance of a variety of food products.

• Foreign matter (e.g., plastic pieces)
• Chips, cracks, peeling, poor shape
• Burnt, over- or under-infested, inconsistency of ingredients
• Defective containers/packages, etc.

metal part

Scratches, dents, cracks, breaks, paint (uneven color), welds

film and sheet

Scratches, dents, dirt

package

Scratches, dents, stains, label presence, label position

Electronic Components and Substrates

soldering

To give you a concrete image of the purposes and applications for which AI visual inspection is used, we will introduce some examples of "InspectAI" introductions and inspections.

Food] Cup Noodles

Food] Bento

Metal parts】Bearings

Cardboard and labels

Various other case studies and inspections are also available. Please contact us for more details.

ArayaThere are five steps to implementing "InspectAI" in the following manner. The introduction process is almost the same as for general AI visual inspections. ArayaWe will tell you about it so that you can imagine what it would be like if you consulted with us.

(1) Hearing

First, we will interview you about the details of the current inspection. The contents of the hearing are as follows

• object to be inspected
• checklist
• Target values (accuracy, speed, etc.)
• Estimated Budget
• Introduction Period

For more specific consideration, please also share the following photos

• Photograph of the object to be inspected
• PHOTOGRAPHS SHOWING TYPICAL NG ITEMS
• Photos of the factory site

The more of the above decision-making factors that are available, the easier it will be for the feasibility study to proceed smoothly.

(2) Initial study

Once the requirements are organized, we will conduct an "initial study" to determine the feasibility of AI visual inspection.
It is necessary for the customer to prepare several images (or workpieces) of good and defective products. We will check how the defective parts appear in the supplied images and determine whether "InspectAI" can perform the inspection.

(3) Accuracy verification

We will actually prepare a certain number of image data and verify how accurate it is with "InspectAI". Specifically, we will proceed with the following steps.

• Customer prepares actual good and defective products
• Taken in our lab
• InspectAI learns to perform inspections and verify accuracy.

(4) Final Proposal

Based on the results of the accuracy verification, we will propose a configuration plan for software and hardware to be installed on the actual line.
One of the strengths of Arayais that we can also make proposals for areas other than AI, as described below.

• Optical equipment (cameras and lighting)
• Conveyors and eliminators (sorting machines)
• PC/WORKSTATION TO RUN AI VISUAL INSPECTION

For hardware, we will make proposals in cooperation with partner companies that have extensive experience in equipment installation. If you want to use hardware from the current line, we will examine whether it can be used without modification.

(5) Introduction

We install the software and hardware at the customer's plant site.
We will perform any necessary setups that Arayacan perform, after having agreed on the details in advance. We will do our best to help you get the most out of InspectAI.