Journey beyond the human eye
Industrial Image Processing and Artificial Intelligence
From the phones in our pockets to the reality of self-driving cars, the consumer economy has begun to benefit from the power of artificial intelligence. The same technology, as we might expect, is now rapidly transitioning to use in advanced manufacturing applications based on quality control and other controls.
Leading manufacturers are turning to artificial intelligence and deep learning solutions to solve complex problems. These manufacturers have long been working on systems that provide the reliability, consistency and speed of a computerized system with the flexibility of human visual controls and can precisely solve difficult vision applications that are tedious and often impossible to develop.
In addition, there is a tremendous need for human experience and human senses in assembly and control processes in the manufacturing industry. In recent years, the shortage of skilled operators and rising labor costs have become more critical. Therefore, automating human tasks has become an urgent task for businesses.
One of the processes in which people use their decision-making ability most effectively in the industry is the visual control process. For visual inspection of products with wide variability, it is still very important to have the sensitivities of an experienced operator to be able to recognize defects of various colors and sizes and to distinguish variability between "defective product" and "correct product".
The need that arises at this point is the ability to use traditional camera control tools (such as presence-absence, shape capture, position and measurement, as well as code and OCR reading). out of control that fully meets expectations, which is very difficult to do with existing systems complex (many checkpoint) checks and especially surface error checks In order for it to be implemented, artificial intelligence was supposed to meet this need.
However, for the practical use of artificial intelligence in this role, which can recognize the characteristics of an object and automatically learn decision criteria, it is necessary to prepare it for use with a large amount of image data, as it learns error types. In addition, these projects have problems such as installing field-specific artificial intelligence hardware and keeping software engineers who are experts in artificial intelligence in the field during the long teaching process. All these became roadblocks that slowed down the progress of artificial intelligence in industrial image processing systems.
To solve these problems, OMRON developed an error detection AI algorithm that reproduces the control methods of a skilled operator and the technology to implement it within an existing image processing system.
An "FH Series" image processing system equipped with this AI was first exhibited at the Tokyo Big Sight trade show last year.
With the technology developed by OMRON, you can automatically determine which images to select for optimal learning and achieve high inspection performance even with approximately 10 images.
Additionally, because it is so easily designed, it is possible to quickly get started and make adjustments even without having AI expertise.
Besides this, I would like to talk about the fancy world of artificial intelligence that is very pleasing to the eye, as well as the aspects that may challenge you.
We identify three main limitations here.
First, AI almost always requires large amounts of sample data. You may need to retrieve data for a very long time and revise the system over and over again based on new data.
Second, while AI is very successful in desktop applications, it may fail on real-world images outside the dataset because it is actually operating on comparative datasets. All datasets have biases. In particular, they may correspond to situations where the image processing system may lead to unsuccessful results in case of rare errors.
Third, the generated dataset may be overly sensitive to changes in the image. This means that many of your products with minor changes that are actually OK for you can easily be identified as faulty.
To summarize, competition and efficiency targets in the industry require improving quality every day. Image processing systems are also at the key point of this work. The most important thing at this point is that with the automation of visual control, people are freed from simple and monotonous work.
Omron FH Xpectia AI-Fine Matching and AI-Defect Detection Thanks to artificial intelligence tools, it offers you the opportunity to solve problems that challenge traditional image processing solutions in a robust and reliable way.
