This November, Mr. Ryu Narusawa, a Factory Team Leader within Sony Semiconductor Solutions Corporation talked with PEAKS Media about the technology demands that the Manufacturing industry has, and why Edge and Vision AI are imperative for long term success.
In the following interview, hear from Narusawa-san as he gives a deep dive into the value points of Edge and Vision AI technologies for the Manufacturing industry, and how early adoption today can have long term benefits to customers.
The following interview, translated from Japanese, has been published in association with PEAKS MEDIA. Read the original article in Japanese, here: https://www.peaks-media.com/8947/
The Future of Edge AI Discussed by the Frontrunner of Manufacturing Digital Transformation (DX) | AITRIOS and Manufacturing DX [ Part 1]
Sony Semiconductor Solutions Corporation's edge AI sensing platform, AITRIOS™, aims to contribute to various social issues under the theme of social implementation of AI. Front-runners who are promoting manufacturing Digital Transformation (DX) will discuss the possibilities and ecosystem of AITRIOS.
The speakers were Mr. Ryu Narusawa, Factory Team Leader of Sony Semiconductor Solutions Corporation, and Mr. Atlantic, CEO of Hutzper Corporation, which is promoting solutions by introducing edge AI at the forefront of manufacturing sites. The moderator was Daisuke Hayashi, an expert in DX support. In the first part, we will explore the role and challenges of AITRIOS in manufacturing DX, and in the second part, we will explore the future of IoT system development.
AITRIOS as a manufacturing DX that can be easily tried in the field
Mr. Hayashi: Thank you very much for joining us today. I would like to start by briefly introducing myself. Today's discussion will be moderated by Daisuke Hayashi of FOCIT.
I myself joined a major electronics manufacturing company as an engineer, then moved to the telecommunications-related industry, and entered the consulting industry about 10 years ago. As a consulting center in the manufacturing and telecommunications industries, we are involved in IoT, support for new business creation, and DX accompaniment. In 2023, we became independent and established FOCit.
Daisuke Hayashi, Manufacturing DX Consultant focit Co., Ltd.. Focusing on digital transformation, new business creation, work style reform, regional revitalization projects, etc. against the background of technological trends. We support the formulation of digital strategies and the implementation of transformations.
Mr. Onishi: My name is Hiroshi Onishi from Hutzpa. Founded in 2020, we provide solutions for the manufacturing industry that support AI-based automation in automated inspections and staffing, as well as preventive maintenance services for equipment using vibration sensors. The head office is located in Osaka, and about 70 employees are working on issues such as how to provide edge AI in an environment where the communication environment is not in place
Edge AI development startup Futsuper Co., Ltd. CEO Hiroshi Onishi. We are developing image recognition edge AI services for the manufacturing industry with the motto "Fast, easy, and super skillful AI." By automating visual inspection work with AI, we will work to solve labor shortages, stabilize quality, improve operational efficiency, and solve problems at manufacturing sites with technology.
Mr. Narusawa: I'm Ryu Narusawa from Sony Semiconductor Solutions. Since joining Sony, I have been involved in new business development for 26 years. After developing a 200MB floppy disk and launching a medical business, he moved into the cloud business around 2013. Initially, he was engaged in DX for the medical industry, and then cloud services for the manufacturing industry. In recent years, we have been focusing on data collection and supporting factory DX using AITRIOS and the intelligent vision sensor IMX500.
Ryu Narusawa, Factory Team Leader, Sony Semiconductor Solutions Corporation. Through Sony's edge AI sensing platform "AITRIOS", Sony will work with the manufacturing industry to develop solutions, such as building a system that enables collaboration between the edge and the cloud, and sensing solutions using edge AI by partners.
Mr. Hayashi: I understand that AITRIOS provides cloud services for factories, but could you briefly introduce AITRIOS's strengths from the perspective of DX in the manufacturing industry?
Mr. Narusawa: First of all, I would like to talk about what kind of challenges we faced in utilizing the vision. Prior to being in charge of AITRIOS, I was providing video analysis services for Sony Group factories. However, at that time, the series of tasks to provide services using AI was extremely heavy and difficult.
A large amount of image data is generated from a camera equipped with an image sensor with a resolution of 12 million pixels, and AI processing is performed on NVIDIA's graphics board, and it takes a huge amount of time and effort to save to the hard disk, AI processing, and system version upgrade, etc., and it was very inefficient. While we had the goal of creating a convenient service using APIs, we were plagued by the complexity of the system, as the initial system evolved rapidly, and the old system could not be abandoned, and we continued to support version upgrades.
This was due to the complexity of the data exchange and the need for a lot of preparation to handle it. The intelligent vision sensors IMX500 and AITRIOS can solve these challenges. The IMX500 is equipped with an AI processing function in the image sensor, and the necessary AI processing can be performed in the sensor, eliminating the need for data transfer and writing software for the handling part. Since the AI inference results can be extracted directly from the sensor, the software becomes compact and the entire system is greatly simplified.
Created based on materials from Sony Semiconductor Solutions Corporation
Mr. Narusawa: I also think that the easy price range that is easy to try out will be a big advantage.
Until now, even if you wanted to use AI, it was not easy to start because it required expensive equipment and expertise, and if it was expensive, there were internal procedures. If this is an industrial camera equipped with our IMX500, it can be assembled for about 100,000 yen, and if it is a combination of an inexpensive single board computer such as a Raspberry Pi and a dedicated camera, it can be assembled for tens of thousands of yen, so the hurdle for approval is lowered. In addition, AITRIOS provides everything necessary to apply AI processing to images, such as AI model development tools and cloud-based systems, so you can create AI-based solutions right away.
The concept is that you can easily start developing edge AI.
Mr. Hayashi: The key point is that the hurdles are lowered as much as possible so that it is easy for the manufacturing industry to introduce it.
Mr. Narusawa: When I first introduced AITRIOS, I was told, "You can't use a system that can't be started at the discretion of the section manager at the site." In April 2023, Sony announced the conclusion of a strategic alliance with a small investment in the Raspberry Pi, and in September 2024, the Raspberry Pi AI camera equipped with the IMX500 will go on sale. This AI camera can be purchased for $ 70 (about 10,000 yen), making it easier to judge the introduction to the manufacturing site as an example earlier.
Raspberry Pi AI Camera – 12.3MP with IMX500 (Source) Sony Semiconductor Solutions Group
Mr. Hayashi: It is true that in the manufacturing industry, it is more important than you think to be easy. The cheaper it is, the better... It's impossible if the hurdles aren't low.
By the way, from the standpoint of developing AI systems, what do you think, Mr. Onishi?
Mr. Onishi: The Raspberry Pi that Mr. Narusawa mentioned was exactly that user. Around 2018, just before I founded Hutzpa, edge devices such as NVIDIA's Jetson and Google's Coral USB Accelerator began to appear, and edge AI became popular.
If it was just hardware, it could be bought at the level of tens of thousands of yen, and the code was very complicated, but once implemented, it worked properly. Having experienced such things and getting to where we are now, I can only sympathize with the fact that it is better to have low hurdles in the manufacturing industry. Since we are working on a platform, the lower the hurdle for the hardware to connect to it, the better.
Hardware has become cheaper to the level of tens of thousands of yen, but I think that there are very few people who work in the manufacturing industry who actually touch it, even if it is only hardware. I understand that AITRIOS is trying to make it easier to handle.
Mr. Narusawa: AITRIOS provides a variety of functions, such as development environments, marketplaces, and cloud services, which are necessary to realize efficient development of solutions. For example, the "Local Edition" software kit for evaluation and verification directly in a local network environment provides tools that allow you to create AI models on your home PC without advanced knowledge. It was launched so that you can easily try out the accuracy evaluation and experimentation of AI models.
The "Local Edition" software kit for direct evaluation and verification in the local network environment provided by AITRIOS is available, and AI models for Classification and Object Detection can be easily created and verified without advanced knowledge. Source: Sony Semiconductor Solutions Group
_____________________________________________________________________________________________________________________________
Staying close to OT from the IT side: What are the challenges of manufacturing DX?
Mr. Hayashi: In the context of DX in the manufacturing industry, I think there are companies that are doing well and companies that are not. In general manufacturing DX, from the perspective of "offense and defense", defense such as reduction of man-hours is often featured, but I am very interested in how AITRIOS was born. In general, I feel that the manufacturing industry is not good at creating new businesses that they have, such as creating IoT-like mechanisms and selling them externally, but what kind of corporate culture is behind Sony's ability to embark on "aggressive DX" while having the DNA of a manufacturing company?
Mr. Narusawa: AITRIOS itself was not started with factories as a target, but one of the reasons for this was the fact that there are multiple factories within the Sony Group.
Factory workers have clear challenges and objectives, such as wanting to improve productivity and reduce costs through automation with AI and data linkage through the cloud. However, there were voices that it was difficult to start with a single development team at the factory, so we decided to provide cloud services cross-sectionally through AITRIOS. Since it is within the group, we can discuss with the people in the field, make it a project, and have them set a budget, and we can do it together, and confidentiality and other matters go smoothly. I think the main reason was that there was an environment within the company where it started to turn as a business.
Mr. Hayashi: I'm sure there are many manufacturing companies that are talking about the same context, but there are still many companies that are not doing well, and I was wondering what was different from the Sony Group.
As I listened to your story, I felt that AITRIOS is more IT-oriented, but the people in the field are working in the world of OT (Operational Technology), which operates facilities and equipment, and I think there is a big gap between IT and OT. I feel that this gap is also the reason why manufacturing DX is not working, but can you tell us what kind of things you have done to break down the barriers and lower the hurdles when promoting new businesses?
Mr. Narusawa: I think it's important to find out where the field is in serious trouble.
For example, when we receive a consultation that says, "I want to pull data from the PLC of a production facility and manage it, and it is really hard to type handwritten notes into a PC every day." Instead of proposing a large system all of a sudden, start with something simple that can solve the problems in the field. "Let's do it with a Raspberry Pi and a sensor, and we can easily manage the data." Then, the number of Raspberry Pis will increase and it will be difficult to manage them, so I said, "Let's do it in the cloud."
The site is really in trouble, so they will persuade the upper management to proceed.
Mr. Hayashi: I would say that we are very close to the OT side from the IT side.
I believe that AITRIOS's strength itself is that it can do more than just provide a platform because it knows various contexts in various industries. Is there a kind of culture that Sony can relate to? Do you have any tips or advice for other manufacturers to make it work?
Mr. Narusawa: I think it depends on how close you can get to the site (OT side) from the cloud side and the IT side. It depends on how well the IT side can understand the site. If you don't understand the real issues of the other party, their pockets, and the decision-making process, how they differ from factory to factory, you won't be able to proceed. There are cultural differences between industries and segments, so I think it's important to understand that.
Mr. Hayashi: Mr. Onishi, you are nodding quite a bit, but do you think that is still the case?
Mr. Onishi: In the manufacturing industry, I think it is especially necessary for the IT side to be close. I didn't expect the other party (OT side) to come by.
When I was going around the site for about half a year after the company's founding, I was told, "I don't understand what you're talking about, so I'm not going to listen to you anymore, so please put on your work clothes first" (laughs), and I've been wearing work clothes ever since. If you open an IT MacBook and go to explain it with a scrunchy feeling, it will be closed by the other party from the start. Small and medium-sized enterprises in particular are stronger in the field, so I think it is important to remove barriers from here.
Mr. Onishi: I would like to ask Mr. Narusawa one question about the development of AITRIOS, which is closer, is AITRIOS developed in an agile manner with a voice from the field, or is it a long-term roadmap for future business plans, image sensors for autonomous vehicles, etc., and is it built according to it?
Mr. Narusawa: AITRIOS has both hardware and software aspects, and hardware cannot be agile. Since the operation of the factory has been decided, it will take five years for each voice. In that respect, AI processing chips that can change the AI model installed are grateful and interesting.
In the past, what could be done with hardware specifications was determined by the specifications at the moment of development.
Therefore, four or five years ago, when AI chips appeared, I actually wondered what could be done with about 8MB of memory. However, due to the tremendous progress in the world in the method of data compression of AI models, it is now possible to do quite amazing things with 8 MB. But the hardware hasn't changed. I think this is a device that has never existed in the world before.
So, we invest in software to increase what we can do with a small size. Requests such as "I want to do this because there was such a case" or "It works with a large AI model, but how can I make it smaller?" are conveyed to the development team and compressed. The AI model is from the perspective of the field, and it seems that it will change to something new and agile.
Sensing solutions that expand the possibilities of "seeing"
Mr. Hayashi: So our strength is that we have continued to develop products that meet the needs of the field.
I would like to know the specific use cases of AITRIOS, such as how you want it to be used for such sites and how it can be used, but what kind of places do you expect it to be used from the perspective of AITRIOS?
Mr. Narusawa: The easiest way to use it is to use visual inspection. It's a place where you can take the place of people's eyes. This means that the machine takes the place of the person and checks the product for scratches and dirt, whether the parts are installed correctly, etc.
When I joined the company in '98, I continued to visit factories around the country on business trips, and at that time, there were many young people and the organization was pyramid-shaped. In today's workplace, the number of young people is decreasing, and if the so-called veterans leave, we are approaching a situation where we will not be able to go around anymore. In fact, one of our customers once told us, "This is not ROI (return on investment), but BCP related to business survival." The fact that the number of people is decreasing is really serious, so the first expectation is to have people use it from the perspective of a place where it can be automated and make it a little easier in that area.
In addition, we would like to use sensors as new "eyes".
Going back in time, the advent of digital cameras has greatly changed the way we shoot and share. Compared to conventional Polaroids and sketches, digital cameras enable real-time sharing of on-site conditions and achieve visualization. This "being visible" is very important. Image sensors mounted on digital cameras have been able to visualize the invisible by capturing light and converting it into electrical signals.
Sony's image sensors are used in various places in our lives. High-quality smartphone cameras, factory automation, safe driving support systems for cars, etc. Source: Sony Semiconductor Solutions Group
Mr. Narusawa: An easy-to-understand example is the litmus test paper.
Acidity/alkalinity, which is difficult to judge visually, can be identified by using litmus test paper, such as red for acidity and blue for alkaline.
However, this "color" is difficult to handle, and "light red" and "dark red" may be judged differently by different people. By using AI for these things, even the same "red" can output the pH value (pH value) corresponding to the intensity of the color as a result. We believe that being able to make the same decisions without relying on people is important not only in terms of quality assurance, but also in terms of passing on technology.
With our AITRIOS and IMX500, we will be able to convert "things that are difficult to quantify" into numerical values with the power of AI and cameras. By using edge AI technology, the IMX500 successfully compresses the data required for image processing and retrieves the necessary data. In other words, it will be possible to convert "things that can be seen with the eye but are difficult to quantify" into numerical values with the power of AI and cameras.
We would like to further develop this technology and think together with you about what new possibilities will open up.
Mr. Hayashi: Do you digitize things that are ambiguous about what they are, such as the "fox color" that often appears in cooking, and the sensor tells you that it is a fox color right now, for example?
Mr. Narusawa: Exactly. We exhibited at the Hannover Messe held in Germany in April 2024, and at that time, we demonstrated that the "cookie doneness" can be processed by AI and displayed in three stages to be checked.
To put it more concretely, it can replace what you see with your eyes, such as the color and turbidity of factories and plants.
Demonstration of cookie doneness (Source: Sony Semiconductor Solutions Group)
Mr. Hayashi: That's exactly how it's used. Mr. Onishi, do you have any ideas about how to use it in the field?
Onishi: In terms of replacing the human eye, I think it is important to create value that is unique to digitalization.
Until now, I think that we have only looked at the last result of the litmus test paper, but by using a sensor to look at it all the time, we can see when it has changed and the granularity of the information can be seen.
In the plant you mentioned as an example, there is a place that conducts a test called SV30 (activated sludge sedimentation rate) twice a day to check the quality by scooping up the sludge in the sedimentation layer and letting it stand for 30 minutes. It is judged by the sedimentation rate of the sludge after 30 minutes, but the quality should be different if it sinks in 3 minutes and if it sinks in 10 minutes. By placing a sensor and constantly monitoring it, you will be able to see changes other than simple OK/NG, such as when separation is slower than usual. I felt that it would be possible to use it in places where it is difficult to do it by humans.
Mr. Hayashi: We don't replace people, we go beyond where people can't. Is that the zone where DX succeeds? Thank you.
AITRIOS and its logo are registered trademarks or trademarks of Sony Group Corporation or its affiliates.
PEAKSMEDIA EDITORIAL TEAM
PEAKS MEDIA is a website that disseminates news articles and useful information that supports the transformation of the manufacturing industry and open innovation.
We will introduce a variety of "PEAKS" such as outstanding technologies, materials, people, and companies in each project, and deliver information that will support innovators in companies who are promoting reform.