Developer Interview

INTERVIEW

The Developer's Story:
The Journey to the Release of ABC Software, and
the Passion Behind the FraD‘s Continuous Ongoing Evolution

KENTA YAMAGIWA National Institute of Occupational Safety and Health

PROFILE

Mainly active in fracture surface analysis and accident investigation, as well as related lectures and academic activities. He has been involved in fracture surface analysis research since he was a graduate student. After graduation, he worked at the National Institute of Occupational Safety and Health.

The First Step of FraD Development, Born from a Sudden Inspiration
Please tell us about the inspiration and process behind the development of FraD.
At first, I was working on creating a database of fracture surfaces in a graduate school research lab, but gradually I began to have the idea that it would be nice if similar fracture surfaces could be searched for. However, implementing this idea was quite difficult. After a while, machine learning became a hot topic. I had a sudden idea of “what if we could teach the features of fracture surfaces?”, and wrote a program for an application example that came to mind on a whim. As it turned out, it produced a pretty good results. So I thought it was interesting and decided to start working on it in earnest.
" In any case, the value is on the ability to reason about fracture surface images."
Please tell us about the value and appeal of FraD as you see it?
In any case, the value is on the ability to reason about fracture surface images. The fracture surface images used for learning are created by scanning the cross sections of destructively tested fractured materials. Of course, the actual fracture surface differs from the surface produced in the experiment, so we do not yet have a complete picture of what results will be obtained when we let FraD make inferences. However, the fracture surface images from destructive tests, where the cause is evident, remain strong and reliable data. We believe that the service based on the quality and explainability*1 of such data is attractive. In fact, our service has been used in a variety of fields, including large companies, universities, and research institutions.
*1: The degree of ability in machine learning to explain the justification for the answer generated
List of Companies, Universities, and Institutions using FraD
The development of FraD mainly required two things: experimentation and software development.
What was the development structure of FraD?
The development of FraD mainly required two things: experimentation and software development. The experimentation was conducted by professional companies, including Kobe Material Testing Laboratory as well as IHI and other companies. On the other hand, the software development was led by me. Since it was my first time writing a machine learning program, I often did not understand about the interpretation of the calculation results. Regarding this issue, I asked for advice from other participants who are familiar with machine learning and made adjustments as we developed the software.
“Is Python easy to read? (wry smile) It took me three years to think that way...”
Did you encounter any difficulties in the development of FraD?
FraD is developed in Python*2, but since I have been using C*3 and C++,*4 it was difficult to get used to the writing style of Python. It seems to be often said that Python code is easy to read, but is it easy to read? (wry smile) It took me three years to think that way... I had quite a bit of difficulty in this regard.
Other than that, I had enough support in the consortium activities and didn’t feel much difficulty. But maybe that’s why other people had a harder time. (wry smile)
*2: A type of programming language commonly used in the development of artificial intelligence and machine learning systems
*3: A type of programming language known for its high versatility and fast processing speed
*4: A programming language that extends the capabilities of the C language
FraD aims for even more accurate analysis.
What are the challenges for FraD in the future?
It is impossible for a machine to learn all the fracture mechanisms of all the materials in the world. However, the accuracy of inference of actual fracture surfaces will depend on how many data points can be gathered. Since the population size is unknown, it cannot be said with certainty how many data points are necessary to produce accurate results. However, I believe that one of the challenges of FraD is to collect as much actual fracture surface data as possible and to determine how accurately it can be inferred.
Expectations for a wide range of FraD applications
How do you see FraD developing in the future?
I believe that this is really just the first step, as we are only able to infer fracture surfaces. On the other hand, the basic part has been completed, so I want to apply it for the future. We have many ideas, for example, creating an environment where the members from companies and schools use their analyzed fracture surfaces as learning data, or creating a system that infers while incorporating other information besides fracture surface. I have many ideas.
“Do you want to give this a try?”- that led me to start researching fracture surface analysis.
You have been involved in fracture surface analysis since you were a graduate student. What led you to start getting involved in this field?
When I was in the graduate school, we could choose the research laboratory we wanted to enter on the basis of our entrance exam grades, but my grade was at the bottom (wry smile). I was assigned to the lab where a person who was going to join had declined admission, and that happened to be the laboratory of Professor Shinsuke Sakai*5. He asked me, “Do you want to give this a try?”—that led me to start researching fracture surface analysis. I was a reckless student who only worked part-time jobs to earn money without doing research, so I still remember being scolded by the professor quite often (laughs). My part-time job was with a company that solved PC-related problems and I worked on various projects. One of my fondest memories was when a fellow part-time worker from my university and I solved a request to convert a program written in N88-BASIC*6 into an Excel macro.
*5: Organizer of consortium activities for the FraD system. Professor emeritus at the University of Tokyo.
*6: A standard programming language developed by NEC corporation used on part of their personal computer series.
My part-time job was with a company that solved PC-related problems and I worked on various projects. One of my fondest memories was when a fellow part-time worker from my university and I solved a request to convert a program written in N88-BASIC*6 into an Excel macro.
*5: Organizer of consortium activities for the FraD system. Professor emeritus at the University of Tokyo.
*6: A standard programming language developed by NEC corporation used on part of their personal computer series.
I think that accident with the bolt was fateful. (laughs)
We would like to hear more interesting episodes related to the fracture analysis!
I boldly bought an Impreza STi*7 with my starting salary, but I enjoyed driving so much that my monthly gas bill exceeded 100,000 yen while gasoline was then 100 yen per liter... I am good at math, but I realized that I am not good at calculating money (laughs). In no time, I had used up all the savings from my student days and started doing maintenance myself to reduce the costs of keeping the car. Before I knew it, I was living every day as a 'hobby Impreza.' However, my Impreza was involved in an accident on the Tomei Expressway where a bolt from a truck hit it. I was very depressed at the time, and my frustration was directed toward that bolt. Then, when I examined the bolt closely, the fracture surface was perfect for my fracture surface analysis! (laughs) The recovered bolt became not only the first step of my fracture surface analysis but also a good source of material for anecdotes. Looking back, I think that accident with the bolt was fateful. (laughs)
*7: The name of a car sold by Subaru
Finally, please give a message to those who are considering using FraD!
I am very happy to have been able to launch such a service with the strong support of the consortium members, including Kobe Material Testing Laboratory. It is the result of the work that I have been focusing on for the past few years. I believe that such machine-learning based services will be used more and more in the future. However, I also think that there are still many aspects users may feel unsatisfactory. Therefore, I hope to be able to respond to requests, “I want something like this,” from users who use this service as their core.

Kenta YAMAGIWA

Hometown: Shimizu City, Shizuoka Prefecture (now Shimizu-ku, Shizuoka City), Japan
Affiliation:
  • Senior Researcher, Mechanical System Safety Research Group, National Institute of Occupational Safety and Health
  • Visiting Professor, Tokyo Denki University (Graduate School)
Membership:
  • Japan Society of Mechanical Engineers
  • The Society of Material Science, Japan (Chairman of Fractography committee)
  • The Japanese Society for Strength and Fracture of Materials
Job Description: Investigation and appraisal commission of industrial accidents
Main Research: Fracture surface analysis and material strength
Background:
June 2003: Completed coursework for Doctoral Program in Mechanical Engineering, Graduate School of Engineering, University of Tokyo
July 2003: Employed at National Institute of Industrial Safety (currently known as National Institute of Occupational Safety and Health)
March 2004: Awarded doctoral degree in Engineering for research on "Characterization of fracture surfaces by introducing local parameters"
September 2007 to February 2008: Visiting Scholar:Department of Mechanical Engineering, University of Utah
April 2013: Concurrently appointed as Visiting Professor (summer semester, equipment safety engineering) at Tokyo Denki University (Graduate School)
Grants Received:
2005-2007: Health and Labor Sciences Research Grants (Quantitative Analysis System Development for Estimating Fracture Load from Fracture Surface)
2009-2011: Grants-in-Aid for Scientific Research (B) "Measurement of 3D Geometry of Creep Voids and New Life Evaluation Method Based on Void Volume Fraction" KAKEN
2013: Research Grant from Japan Boiler Association "Three-Dimensional Quantitative Evaluation of External Pipe Corrosion and Corrosion Simulation Material Based on Fractal Concept"
Awards:

Japan Society for Material Science Paper Award (2002)

ALL Interviewed by

Produce TARO MATSUMOTO (GiRAFFE & Co.)
Photo KOKI MATSUNAGA
Hair Make YUKA KURIHARA
Direction YUMI SHIMOTAHIRA (GiRAFFE & Co.)
Writing YUKI KAWATO (GiRAFFE & Co.)
HARUNA MIKAMI (GiRAFFE & Co.)
Design YUKI KAWATO (GiRAFFE & Co.)
HARUNA MIKAMI (GiRAFFE & Co.)
COMPANY INFORMATION
Company Profile