Beginning of this page.
Displaying present location on the site.
  1. Home
  2. 100 Innovators Series
  3. Kenji Yoshihira
Main content starts here.

A developer of advanced technology for the automatic detection of system faults

Kenji Yoshihira

"My goal is to assist secure operations by detecting signs of system faults."

The SIAT advanced analytics engine uses data collected from sensors to learn the behavior of systems, including computer systems, power plants, factories, and buildings, enabling the system itself to automatically detect faults. Kenji Yoshihira, one of the developers of this unique analytics engine, discusses its inner workings, and shares stories from the development process. He also explains how this technology contributes to safe and secure social infrastructure, and talks about what motivates him in his work.

Enabling systems to automatically detect faults independently

Photo: Kenji Yoshihira

--First up, what sort of technology is SIAT? Can you break it down for us?

Yoshihira:
Put simply, SIAT (System Invariant Analysis Technology) is an analytics engine developed around the concept of understanding system behavior.

The roots of SIAT development can be traced back to a research project involving autonomic computing that was put together by three staff members hired by NEC Laboratories America.

Let me briefly explain what autonomic computing is all about. For example, when a person gets sick or is injured, the body can recover without outside help, and detect abnormalities such as fevers on its own. The basic theme of our research was to enable computers to automatically recover in the same way, and detect and defend against faults independently.

--Can you explain how SIAT learns the behavior of systems?

Yoshihira:
The process of learning system behavior starts with extracting information found in data collected from sensors or other sources. Let's use a water pipe as an example.

We'll assume that sensors have been placed at the inlet and outlet points of a water pipe to measure the flow rate of water inside. If the inlet point has a fast flow rate, the outlet flow rate will also increase, and conversely the outlet flow rate will slow when flow slows at the inlet. This is what is known as invariant relationship.

If there is a significant difference between the flow rate at the inlet and outlet points, there may be a fault somewhere, such as a hole or crack in the pipe. SIAT differs from conventional methods because it focuses on the relationship between inlet and outlet point data, rather than monitoring the flow of water itself. Instead of using the sensor values, information on the relationship between sensors is used.

With SIAT, a model is first created for the relationship between two data points representing normal system behavior collected from sensors. This is then compared with sensor data values collected in real time to detect faults based on any discrepancies found. By collecting more sensor data and generating relationship models from it, an even better overall view of system behavior can be obtained.

The main advantage of SIAT is that is makes it possible to comprehensively identify the many relationships within an entire system, and monitor all of them exhaustively. This enables minor faults that even an expert might overlook to be detected.

Monitoring the safety of power plants, factories, roads, and bridges

--What kinds of systems is SIAT used with, and how does it benefit them?

Yoshihira:
SIAT can learn the behavior of any system designed and built by people. This includes everything from the computer systems in data centers or communication networks systems, to structures such as power generation systems, factories, chemical plants, buildings, bridges, aircraft, and ships.

The normal behavior of these systems is ascertained by reading data measured using a range of sensor technology, covering parameters such as temperature, pressure, flow rate, and vibration. Data is the language that SIAT uses to understand the systems that it interacts with.

Figure: (NEC Site - Big Data Solutions - Operation Enhancement/Optimization Diagram)(NEC Site - Big Data Solutions - Operation Enhancement/Optimization Diagram)

The NEC Group is currently focusing on its social solutions business. SIAT is also effective when applied to monitoring systems that protect communications infrastructure and lifelines such as water and power from failures.

When an earthquake strikes, secondary damage may also occur due to the destruction of buildings, highways, bridges, or tunnels. SIAT can be used to minimize this kind of damage by detecting faults or weaknesses not apparent to the human eye before they become a problem. In this way, I believe SIAT can be applied to a broad range of areas to help create a society that provides a safe and secure environment for people to live in.

--What role did you play in the development of SIAT?

Yoshihira:
As I mentioned earlier, research into an analytics engine for assisting system operations was initiated in 2004 by three staff members at NEC Laboratories America. One of our team worked on proposals for the design concept and framework of SIAT as an analytics engine, as well as development of the fundamental algorithms. Another was an expert on statistics and machine learning, who conducted mathematical verification. I was responsible for the preliminary design and functional evaluation of the software serving as the analytics engine that actually incorporates the data.

I still remember the moment when the concept of SIAT was born like it was yesterday. The three of us were racking our brains, writing down various ideas on the whiteboard of a meeting room, when we hit upon a concept for SIAT that we thought showed promise.

Photo: Kenji Yoshihira

--Do you have any stories about the origins of SIAT to share with us?

Yoshihira:
Development of the analytics engine wasn't exactly smooth sailing from the outset. We develop the underlying technology in the lab, with the ultimate goal of implementing it as an NEC product or service.

A few months after the project started, I brought a prototype of the first analytics engine we had conceived and built to Japan, and made a proposal to the division that handles NEC's software products.

The approach was similar to SIAT, but the technology was based on a fundamentally different premise and concept, and the feedback I received from the software division was very negative. After returning to the U.S., I told the other two lab staffs that it wasn't going to be usable in its current form.

We then went back to the drawing board, and eventually developed SIAT. When I returned to the same division at NEC with a prototype of SIAT, this time they showed a lot of interest. From there we worked toward the practical application of the technology, backed by support and cooperation from everyone in the software division, including advice and evaluation using actual systems.

The underlying SIAT technology was first commercialized in NEC's Master Scope Invariant Analyzer integrated management software, which detects silent failures in computer systems that are normally hard to identify. That was the first time SIAT was introduced to the world.

Relying on data without considering its meaning

--I hear that the SIAT technology is very unique. Can you elaborate on that?

Yoshihira:
The most unique aspect of SIAT is that it doesn't assess the meaning of sensor data such as water pressure, flow rate, or temperature. The concept and approach of detecting abnormal behavior solely from regularities in data is very unique, and during development many skeptical opinions were voiced even within the lab in Japan.

By simply importing data into SIAT, it is possible to determine how a system should be behaving, even if the significance of that data is not known. I'm not a designer of cars, and I have no specialist automotive knowledge, but by reading the regularity of data from sensors installed in a car, through SIAT I can understand how that car should run.

Photo: Kenji Yoshihira

--Why was NEC able to develop this advanced SIAT technology?

Yoshihira:
I think there are two main reasons. NEC has provided a range of products and technologies, including ICT systems, networks, and devices, to customers in a broad range of business sectors both in Japan and overseas. Through offering these products and solutions, and gaining deep insight into our customers' business, NEC has the advantage of being able to comprehensively evaluate what new value we should look at providing.

Additionally, to increase customer value using an analytics engine, a system for collecting and processing data is first required. Another of NEC's advantages is the fact we have our own diverse platforms covering areas such as M2M and sensing technology.

NEC has a number of world-leading analytics engines such as SIAT, but it is the fact we also have the ability to provide platforms and infrastructure that enables us to offer new value to customers through implementing and bringing products to market.

Another thing that works to our benefit is that fact that NEC as a company has great interest and zeal for developing new technologies, with a company culture that fairly assesses everything we develop. Even after a failure or two, if the next technology we come up with is innovative and appealing, and directly connected to the market, NEC will thoroughly evaluate our proposals with a scrutinizing eye.

Menu starts here.
End of menu.

Top of this page