INTEGRATION WITH PHYSICAL SYSTEMS

Many high-valued expert system applications are interfaced with physical systems. These applications often reside in the feedback path as part of the control logic of a dynamic system. Positioning the expert system in the feedback path greatly simplifies knowledge acquisition owing to the adaptive nature of closed loop systems. The expert system doesn't have to be as precise.

The JTEC panel saw two distinct control situations in Japan: low frequency systems and high frequency systems, where low and high are defined in relation to the execution speed of the expert system.

The blast furnace controller developed at NKK (see Chapter 2) is an excellent example of a low frequency application. The time constant of the blast furnace is six to eight hours. The controller utilizing conventional expert system techniques executes in a few minutes and is cycled every 20 minutes. At this rate, there are no dynamic stability problems.

Other applications, such as the control of a tunneling machine or a group of elevators, had much smaller time constants. In these cases, simple fuzzy logic or a fuzzy expert system was utilized. The fuzzy logic has several advantages:

1. It produces a smoother control function and smoother system performance, and therefore does not have to cycle as often. This obviates the expert system performance problem.
2. Fuzzy rules may perform the same function as five to ten conventional rules. This allows the inferencing to operate faster. It also reduces the size of the knowledge base, which simplifies both original knowledge acquisition and knowledge base maintenance.
3. Depending upon whom you ask, dynamic system stability with a fuzzy expert system controller may or may not be an issue. It appears that there may be a theoretical problem in proving system stability, but stability has not been a practical problem.

Expert systems integrated with physical systems generally have been initiated in Japan by engineers in the operations area of the organization. This has two advantages: user-instigated systems are much more likely to succeed than systems originated by a technology group, such as AI researchers; and users are more likely to identify high-valued problems.

At the same time, it is often more difficult to associate a quantified value with the expert system. The engineers are more concerned with the total system than with any component. Benefits and value are likely to be attributed to the larger system and may not be broken down into individual components.