A Comparison of Expert System and Human Operator Performance for Cement Kiln Operation

Abstract

Using an observational study of the operation of a cement kiln and a set of operational rules published by Peray and Waddell, five knowledge-based advisory systems have been developed encapsulating the supervisory expertise of the process operators. The advisory systems were developed in the logic language PROLOG. Each expert adviser was applied to a set of plant data previously logged from the kiln, and the advice offered was compared with the actions taken by the operators. Statistical tests have been applied to the outcome of the comparisons in order to determine which expert adviser emulated the operator performance most successfully. The results demonstrate the value of such tests in quantitatively validating expert systems. In addition, the supervisory techniques that the operators were using in practice were identified.

Keywords:

• control
• expert systems
• knowledge-based modelling
• validation