Previously, I worked at a major chemical manufacturer dealing with synthetic fibers and resins until 2022, and I joined Daikin in 2023. The reason I came to Daikin was primarily because of the presence of an in-house information technology university. Daikin had a well-established system for large-scale training of young digital talent and was producing excellent personnel, but there was a shortage of instructors, and I believed my experience could be beneficial. Additionally, Daikin's revenue has tripled in less than 25 years, and it has a high operating profit margin and excellent performance, which were also factors in my decision.
NEWS
【Part 1】Can Materials Informatics Become the Savior of the Materials Industry? — Interview with the Head of Digital Materials Design at Daikin TIC on the Challenges of Digital Development of Materials
FEATURE
2024.11.13
Daikin Industries is a rare company that develops almost all of its components for commercial air conditioners in-house. For example, in-house manufacturing accounts for a variety of parts, including the core components of compressors, motors, and inverters along with the power semiconductors that drive them and air filters. Moreover, as the world's only air conditioner manufacturer that develops and manufactures its own refrigerants, Daikin has also formed an identity from its technological prowess as a chemical manufacturer. Viewed from this perspective, one of TIC's most important tasks is new materials development in support of diverse industries ranging from household goods to automobiles and semiconductors. Currently, Daikin is accelerating materials development using digital materials design technology called "Materials Informatics," which leverages information science technology rather than a traditional analog approach. In Part 1, we speak with Executive Engineer Isamu Shigemoto, head of Digital Materials Design, about Daikin's new approach and efforts towards materials design.
Transfer from a Major Chemical Manufacturer to a Leader in Digital Talent Development
— Mr. Shigemoto, we understand that you joined Daikin mid-career. First, could you tell us about your background and why you decided to join Daikin?
Shigemoto: I have been appointed as Executive Engineer at TIC, where I oversee all aspects of digital materials design. I'm involved in everything related to digital technology in the chemical field. For example, I handle a wide range of tasks including the use of generative AI for materials development and computational chemistry simulations to predict and analyze material properties based on physical laws that have been used for a long time.
Shigemoto: I have been appointed as Executive Engineer at TIC, where I oversee all aspects of digital materials design. I'm involved in everything related to digital technology in the chemical field. For example, I handle a wide range of tasks including the use of generative AI for materials development and computational chemistry simulations to predict and analyze material properties based on physical laws that have been used for a long time.
Breaking Through the Limits of Polymer Materials Development with MI: A Chance for Japan
— Could you tell us about the approach and challenges of developing new materials using digital materials design and materials informatics (MI), which utilizes theoretical calculations and machine learning?
Shigemoto: MI is an approach that uses information science, such as statistics and machine learning, to efficiently conduct material development. Specifically, it is expected to be very useful in predicting combinations and manufacturing methods that meet the desired performance for organic, inorganic, and metal materials development. Recent advances in digital technology have made it possible to effectively support materials design by analyzing vast amounts of experimental data and academic papers. However, applying MI to polymer materials currently presents many challenges, and fundamental technological development is needed.
For example, in drug discovery, MI can be used to input chemical structure formulas and predict pharmacological activity as output, which facilitates the development of new drugs. This is because the world of drug discovery allows input parameters to be clearly defined as structure formulas, making it easier to create machine learning models (functions) that predict the relationship between input and output.
In MI, (y) properties are predicted for a given (x) candidate material / design for x that satisfies y.
However, in the development of polymer materials, input parameters cannot be uniquely defined. For example, consider the molecular structure of polyethylene (PE), which is widely used in containers and packaging films. PE is a polymer compound formed by the polymerization of ethylene (C2H4, CH2=CH2), which has a simple structure with many CH2 units linked together. Even though PE is a single molecule, there are many types of PE because the molecular chain length and shape (branching) result in countless structural variations.
In polymer material development, it is difficult to uniquely define input parameters, making the application of MI challenging. For example, PE has a wide range of structures depending on the molecular chain length and shape.
Furthermore, even the same type of PE can exhibit different properties depending on the molding process used. By varying the combinations of molecular structures and molding processes, you can create materials with differing properties, from inexpensive and tearable plastic bags to strong fibers used in bulletproof vests, and even materials with heat resistance or resistance to acids and alkalis. Given the numerous factors to consider, such as molecular structure, chain length, branching, and molding techniques, clearly defining the concept of the materials that we aim to develop is crucial and showcases the expertise of polymer specialists. However, current mainstream MI machine learning models only handle chemical structure levels. Therefore, focusing on areas beyond chemical structure could become a winning strategy for Japanese chemical manufacturers, who have accumulated extensive knowledge in functional materials, and this is where we intend to concentrate our efforts.
Current efforts at Daikin in MI-based development of organic polymers and composite materials.
— Even though MI is used in both fields of drug discovery and polymer material development, the nature of the input parameters handled is completely different, and predicting physical properties based solely on chemical structure makes creating learning models very challenging.
Shigemoto: Yes, whether one finds this complexity intriguing or not is a matter of personal preference, but I find it fascinating. If I believed that AI alone could complete polymer material development, I would have switched jobs to an IT vendor. However, in reality, no matter how advanced the AI technology is, it cannot design materials on its own. AI can only be effectively utilized by a manufacturer with research and development and production capabilities. This is also one of the reasons I joined Daikin.
What Is Daikin's Vision and Approach for the New Digital Materials Design?
— So, what is Daikin's goal for the future of materials design using MI, as it overcomes these hurdles?
Shigemoto: This is not unique to Daikin but a general observation. Traditional material development relied heavily on the expertise and skills of veterans, often summarized as "intuition, proficiency, and experience," the so-called 3Ks expression in Japanese. A major direction is to shift from this experience-based approach to a logic-based one using data and theory. Of course, intuition and experience are valuable, and consulting a veteran can resolve issues quickly. However, if this expertise becomes too personalized, it becomes problematic when the veteran retires, and the technical skills are not passed on. There is a sense of crisis that if we do not preserve the collective know-how of veterans in the form of data and pass it on to younger engineers, the company itself may not survive.
The second point is that I believe AI is unlikely to propose innovative materials including processing methods. AI is a technology for optimizing input parameters, and it is the developer's job to define the fundamental issue of what inputs (molecular structure or processing methods) should be optimized to control a certain output (properties). In other words, developers must first decide the broad concept of material development. However, once the concept is well-defined, solving optimization problems using AI becomes straightforward. Therefore, verifying ideas proposed by developers becomes incredibly fast with AI. Previously, the viability of an idea could only be assessed by creating samples and measuring their properties, but having your own AI means that you can continuously enhance it with development data and quickly evaluate the effectiveness of ideas. Companies that can rapidly generate and test ideas and quickly respond to customer needs will be the winners in the market.
The third point is the ability to use MI to uncover customer needs. In situations with no initial ideas, conducting a zero-based interview with people often yields vague requirements. However, if you propose, "We have several new material candidates with MI, can they be useful for your business?" then you will receive some reaction. Repeated interactions of this kind will help extract customer needs, quickly establish development policies, and solidify development strategies shared with customers.
MI as a tool to accelerate idea verification will speed up our business with customers. Conversely, if we do not create such an environment, we will lose to companies using similar approaches. Simply proposing materials after developing them to our satisfaction using AI or simulations will not be sufficient. Material development will need to shift to an agile approach in which it will be important to continuously generate ideas, present predictions to customers, even those in preliminary stages, and adjust development strategies based on their feedback.
I believe that material developers will need to transition from working as experts on basic research in research labs and actively engage customers directly. I used to be secluded in a central research institute, but now I actively participate in important meetings with customers.
Shigemoto: This is not unique to Daikin but a general observation. Traditional material development relied heavily on the expertise and skills of veterans, often summarized as "intuition, proficiency, and experience," the so-called 3Ks expression in Japanese. A major direction is to shift from this experience-based approach to a logic-based one using data and theory. Of course, intuition and experience are valuable, and consulting a veteran can resolve issues quickly. However, if this expertise becomes too personalized, it becomes problematic when the veteran retires, and the technical skills are not passed on. There is a sense of crisis that if we do not preserve the collective know-how of veterans in the form of data and pass it on to younger engineers, the company itself may not survive.
The second point is that I believe AI is unlikely to propose innovative materials including processing methods. AI is a technology for optimizing input parameters, and it is the developer's job to define the fundamental issue of what inputs (molecular structure or processing methods) should be optimized to control a certain output (properties). In other words, developers must first decide the broad concept of material development. However, once the concept is well-defined, solving optimization problems using AI becomes straightforward. Therefore, verifying ideas proposed by developers becomes incredibly fast with AI. Previously, the viability of an idea could only be assessed by creating samples and measuring their properties, but having your own AI means that you can continuously enhance it with development data and quickly evaluate the effectiveness of ideas. Companies that can rapidly generate and test ideas and quickly respond to customer needs will be the winners in the market.
The third point is the ability to use MI to uncover customer needs. In situations with no initial ideas, conducting a zero-based interview with people often yields vague requirements. However, if you propose, "We have several new material candidates with MI, can they be useful for your business?" then you will receive some reaction. Repeated interactions of this kind will help extract customer needs, quickly establish development policies, and solidify development strategies shared with customers.
MI as a tool to accelerate idea verification will speed up our business with customers. Conversely, if we do not create such an environment, we will lose to companies using similar approaches. Simply proposing materials after developing them to our satisfaction using AI or simulations will not be sufficient. Material development will need to shift to an agile approach in which it will be important to continuously generate ideas, present predictions to customers, even those in preliminary stages, and adjust development strategies based on their feedback.
I believe that material developers will need to transition from working as experts on basic research in research labs and actively engage customers directly. I used to be secluded in a central research institute, but now I actively participate in important meetings with customers.
The Challenges of Setting Themes Revealed by MI Implementation and Future Outlook
—
What have you learned while developing digital material using MI?
Shigemoto: There are two major issues. The first is the difficulty in aligning business requirements with what can be achieved with current digital technology. For example, if there is a material that could be sold if developed within a year, the digital technology to design it may not always be available. The challenge is to match "what can be done" with "what should be done," which means setting themes for digital efforts is difficult. In this sense, it is also important to cultivate personnel with excellent judgment rather than just blindly enhancing digital technology.
The second issue, related to judgment, is to improve the density and level of communication between digital specialists and experimental researchers, or domain experts. We want to increase communication so that we can not only propose solutions with digital technology based on our understanding of business needs but also encourage business sides to present challenges that digital technology might solve. Instead of focusing on technical issues, there seems to be a lack of a development culture that effectively utilizes available technologies across the organization. Improving this will be crucial for creating a sustainable foundation.
— What are your thoughts on Daikin's future in material design and research and development?
Shigemoto: Instead of gaining an advantage from these technologies, in the future, the intention may be to avoid falling behind, whether it's AI, machine learning, or simulations. Therefore, it is essential to steadily advance initiatives like MI and turn them into competitive business opportunities. At the same time, by introducing advanced digital technologies, researchers will be able to achieve results more quickly, and gather ideas more efficiently than through traditional methods of collecting information from technical papers. AI will allow for quicker identification of promising leads and easier execution of various tasks.
I hope that this technology can be used to save time for more creative work or personal free time, leading to a more balanced and satisfying professional life. Rather than being driven by technology, we should aim to benefit from it. In the future, I think that the world will shift from valuing numerical specs to valuing personal preferences and sensitivities. Maintaining a sense of real-life experiences as a consumer is crucial for generating genuinely valuable product ideas. For this reason, I would say, "Enjoy life to generate great ideas."
— Do you have a message to our readers who want to join Daikin?
Shigemoto: We are always welcoming new talent! If you are an exceptional individual, please come and join us. This reflects Daikin's management's attitude of valuing people and wanting them to excel. The company is growing steadily, with expanding business and profits creating a virtuous cycle of further investments. I have felt this momentum since joining the company mid-career. Additionally, the close distance between executives and the on-site workplaces and the exchange of opinions with staff are positive aspects of our culture. Daikin may not be suitable for those who only wait for instructions from supervisors, but it is an exciting company for those who can proactively seek and pursue interesting opportunities.
(Continued in the next part)
※The information and profiles are based on the time of the interview.
Isamu Shigemoto
Executive Engineer, Technology and Innovation Center
Joined in January 2023. Originally from Osaka Prefecture.
Responsible for digital materials design (Materials Informatics, Computational Chemistry).
"I aim to realize overwhelming development capabilities through a new style of human-AI collaboration and to continually deliver groundbreaking new products that will astonish the world."
Executive Engineer, Technology and Innovation Center
Joined in January 2023. Originally from Osaka Prefecture.
Responsible for digital materials design (Materials Informatics, Computational Chemistry).
"I aim to realize overwhelming development capabilities through a new style of human-AI collaboration and to continually deliver groundbreaking new products that will astonish the world."
Related article
