next up previous
Next: Experimental Results and Up:

Deep Interoperation between Previous: Communication between Diagnostic


Implementing Deep Interoperation of Distributed Diagnostic Expert Systems

 

As the methods of modeling, cooperating and communicating (wrapping) distributed diagnostic expert systems come up, we put them together into an interoperation environment for distributed diagnostic expert systems, as shown in Figure 5. In order for the interoperation environment to be enacted at each development site, each expert system should be manually modeled there, using IPST as a common task ontology. However, just one kind of SS-based cooperation facilities and communication facilities with conversion function manually developed are available without modification in the interoperation environment. So it is only task to model expert systems in order to get into the interoperation environment.

  
Figure 5: The Interoperation Environment between Distributed Diagnostic Expert Systems

When one expert system finds a fault in itself(for example, when one output was wrong or rejected by a user), it asks the interoperation environment to support it in changing for a better performance. First, the expert system (originator) sends its model description message to the other expert system (recipient) through communication facilities. After getting the message, the recipient tries to make correspondence between two models from the originator and the recipient, based on SS-based cooperation facilities. It finds differences in the context of good correspondence and makes them up into a reply message. Then the recipient sends it to the originator through communication facilities. The reply message is converted into another reply message using a common domain ontology so that the originator can use it to change itself. Using the converted message, the inference engine's structure is modified automatically. However, expert system developers manually to change the implementations of the originator based on the specification level modification. The modified originator's performance is manually tested. If the same fault still exists, or another fault comes up, the above-mentioned interoperation process is repeated and another reply message is given to the originator until the originator's performance improves or the recipient sends no reply message.

The interoperation environment has been implemented by SICStus-Prolog. The size of communication and cooperation facilities is only 24 KB. The size of models of two diagnosis expert systems is only 10 KB. So comparing the size of the interoperation environment with the size of the troubleshooting expert system code (540KB) and the enterprise diagnosis expert system code (300KB), the size of model is so small that it is easy for other expert systems to use the interoperation environment.



next up previous
Next: Experimental Results and Up:

Deep Interoperation between Previous: Communication between Diagnostic




Daiki Kishimoto s0011
Sat Sep 28 20:16:55 JST 1996