MPRI Ship Analytics


Implementing Expert Systems within the WISE Environment

Knowledge can be defined as a set of known conditions and the ability to be able to link those conditions in such a way that a future result can be determined or 'inferred'

Within an 'expert system' the known conditions are defined as a set of rules within the system. The conditions upon which the expert system is monitoring are then checked against the defined values within the rules. If certain rules then become true in a particular sequence or point in time the system can link these results together and by using an inference engine, which runs through all the rules and checks all the various combinations, determine a predicted outcome in a similar way to an expert.

Most expert systems are developed for a single application, such as providing assistance with a particular area of medical diagnosis. Even then they have been restricted by the two problems of obtaining sufficient data and the subsequent processing required.

Within the WISE Systems these problems have been overcome in the following ways:

Knowledge capture

Traditionally the knowledge of an expert is captured by another person who then has the ability to be able to define that knowledge in the correct way. Unfortunately, the way the person capturing the knowledge perceives the information he is being given may be different from what the expert meant. Consequently to avoid this the expert has to be involved in extensive testing of the final system to ensure the correct results are obtained.

By using a new programming environment 'Clarity' we have developed a system which allows an expert to input his knowledge directly using rules of a very simple construction. This enables him to see the results immediately avoiding any misconceptions.

For information on the Clarity programming environment click here or follow the link on the page menu.

Processing capability

Apart from the actual performance of the machine hosting the system the processing capability of such systems is a function of the number of rules and their structure. By re-designing the structure of the rules so that it is optimized considerable performance increases have been achieved to the extent that such a system can be used satisfactory in a real time environment. Current systems are able to check in excess of 1300 rules, which in turn check in excess of 5000 parameters within a time period of 2secs.

Portability

Being developed for specific applications means that even if another application may be similar traditional methods have meant that a complete new set of rules has to be defined.

This has been overcome by using a 'Rule Generator' which enables the creation of multiple rule sets for different applications such as two oil tankers to be developed automatically.






© 2008 MPRI