Site: NKK Kawasaki facility
c/o NKK Corporation
1-1-2 Marunouchi
Tokyo 100, Japan

Date Visited: March 26, 1992

JTEC Attendees:



Isamu Komine

Asst. General Mgr.,
Planning & Coordination Dept.,
R & D Div.

Toshio Okawa

Control Engineering Team,
Electronics Research Ctr.

Taichi Aoki

Asst. Manager,
Process Control Dept.,
Fukuyama Works

Shuichi Yamamoto

Process Control Dept.,
Keihin Works


We visited the Keihin Works located on a man-made island (Ohgishima) in Tokyo Bay.

Most of the time was devoted to NKK's primary application, which is a diagnosis and control system that is used for stabilizing the output of blast furnaces. We saw a video of this application, discussed it with Mr. Komine et al., and then went to a blast furnace control room to see the system in actual operation. The system solves the following problems:

The furnace responds very slowly to changes in input, so one needs a good understanding of the device; one can't rely on immediate feedback from sensors.

The KB is structured into many knowledge sources, each of which has only a few rules.

The system was developed and is maintained by a group of systems engineers, who are domain experts. There are no knowledge engineers at NKK.


NKK has 25 ESs in routine operation, and five more in the field testing stage. Of the 37 systems that have been built or are in some stage of development, 16 combine the functions of advising, process control and diagnosis; 20 combine the functions of advising, planning/scheduling, and management integration aid. The blast furnace expert systems are described in Chapter 2.

All of the fully developed systems are integrated with conventional systems, and also use some high-level language (LISP, FORTRAN, PL-1) in addition to the ES shell (ESHELL for the blast furnace, KT for the planning/scheduling systems).

Rather than emphasize training in knowledge engineering and expert system development, NKK has chosen to train its systems engineers in systems analysis and modelling, which are more important skills for total system development. Expert systems techniques in themselves are of relatively small significance, in NKK's view. On the other hand, the company has developed expert system design tools, used in its Engineering and Construction Division, which embody a methodology for developing ESs. These tools, called NX-7 and NX-8, run on Xerox LISP machines and Sun workstations, and have been applied in developing ESs for operations support of refuse incinerators.

NKK often introduces an ES at the same time as it refurbishes its entire computer system (which itself may be just a part of a larger renewal project), making it difficult to evaluate the impact of the introduction of the ES. However, ESs are introduced only when conventional programming techniques fail to solve the problem at hand.

Regarding future development, our NKK hosts envision more use of in-the-loop control, moving from mainframes to engineering workstations, and providing intelligent assistance on more advanced tasks of engineering and management. They see several problems with current technology: AI tools that are difficult to learn and use; relatively high difficulty in system maintenance; inadequate processing speed; need to obtain knowledge automatically from data and to solve problems (e.g., scheduling) by using previous cases.


The developers of all of NKK's ESs are out in the various departments. They have extensive knowledge of the application domain, but were not skilled in AI or ES when they started the projects. Learning at first proceeded by trial and error; some engineers were sent to training courses. There is now a well established methodology for ES development at NKK. Expectations for this technology may now be too high.

Our hosts expressed the view that the newer techniques of fuzzy control or neural networks were too limited for most of the control applications at NKK. They also stated that the Fifth Generation Project had no influence on their operations. They have a few links with AI departments at Japanese universities. Mr. Okawa, for example, is interested in model-based reasoning and maintains such a link for this purpose.


A blast furnace is a complex, distributed, non-linear process. Conventional mathematical modeling techniques have never been able to predict future states of the furnace with enough accuracy to support automated control. As early as 1986, NKK Steel had developed an expert system to predict abnormal conditions within the furnace. NKK and other Japanese steel companies have since developed blast furnace control systems.

Because the blast furnace feeds all other processes in the steel mill, any instability in the operation of the furnace is compounded by the impact on other processes. Avoiding unstable operation of the furnace requires characterizing the current state of the furnace and projecting the conditions which will occur over the next several hours while there is still time to make adjustments.

An expert system has been developed which successfully models the current state, predicts future trends with sufficient accuracy to make control decisions, and actually makes the control decisions. These decisions can be implemented automatically or the operator can take manual control while still operating through the expert system's interfaces. The system continues to make inferences in manual mode, evaluating the influences of manual operations, then making further inferences.

NKK's Keihin Works system is integrated in the process control computer. It is implemented in LISP with FORTRAN used for data processing. The system has 400 rules, 350 frames, and 20,000 steps.

The benefits of the expert system have not been separately established. It is considered an integral part of a new furnace control system that was installed the last time the blast furnace was shut down for relining. We found that this was a common characteristic of expert systems used as closed loop controllers, viz. benefits are not traced to the component level. This suggests that expert systems have taken their place among the suite of techniques available to the controls engineer and do not require the special attention sometimes afforded new technologies.

Expert systems are not, in general, a complete solution, but must be combined with other technologies to solve almost any real problem. ES is more of a technique and a skill than a solution. We are looking at too small a grain size to speak of a business or of cost/benefit of an application. NKK representatives made this point very explicitly.

Most of NKK's applications were done by the steelmaking division; however, the last five on their list were done by the construction division in cooperation with the R&D lab.

Thus, for the most part, engineers in each department develop ESs. They are already domain experts; they know the factory and the business and only need to learn AI. This they do by trial and error, but with very few errors. Very few of NKK's engineers have attended IBM or Fujitsu training courses.

In designing this application, NKK needed to integrate with the conventional data processing of the steel works, so AI had to couple with high level languages. Development and maintenance of ES was done by the same group of system engineers. They can't quantify the AI development cost since AI is usually used as part of a larger project. They did not see any increase in their software productivity using AI. They also didn't find that ES transferred very well from one steel plant to another. Maintenance by engineers who didn't develop the system is very hard. Hence, they only apply AI where conventional methodology does not work well.

Our hosts stated that they found little methodology for writing ESs. NKK people found little management resistance in employing AI. They initially had great expectations for AI. These are now lower, and they can now select the proper subject area for employing AI.

They feel fuzzy logic is very limited and neural networks are similar to fuzzy in this respect, i.e., neural nets have too limited an input/output relationship. It is good for pattern recognition, which is only part of control technology. Hence, what's done by both neural networks and fuzzy logic can easily be done by other methods. "Sometimes NN [neural network] or Fuzzy techniques are used to make an organization appear 'high tech'." There is no AI group per se at NKK. Applications there have been developed by engineers trained in control theory or industrial engineering or whatever. Although they seem pretty sophisticated about AI (they have interest in model-based reasoning, fuzzy, neural nets, etc.) none of the people have had any formal AI training.

Our NKK hosts said that they have had very few project failures because they determine as part of the feasibility study whether to go ahead. They also said that in their view a good system analysis is the key; anybody experienced in doing a good system analysis can then decide what part of the system should be an expert system, and could probably build it. When they use the term "engineer," they typically mean somebody who can do that kind of analysis.

Blast furnace operators take a long time to become skilled (five to ten years). Furthermore some of the conditions treated, like slip and channeling, happen very rapidly, and thus are difficult to control by hand.

One of our hosts said that they are considering introducing more expert systems "because industry is trying to improve the working environment of laborers." One of our hosts remarked to us that "AI tools are getting easier to use."

Published: May 1993; WTEC Hyper-Librarian