INDUSTRIAL EXPERT-SYSTEM ACQUIRED BY MACHINE LEARNING

This paper describes SERF, an expert system for repairing helicopter b lades. Although very useful, it cannot evolve automatically with advan ces in technology and expert knowledge. To address this problem, we ha ve designed an inductive learning system (MEA) that uses rules or prec lassified examples in order to incrementally build production rules. T hese rules are of the following form: if description of the fault doub le right arrow repair and can be used by KEE (knowledge engineering en vironment), an expert system generator. Both MEA and SERF use a repres entation based on structured objects. In this paper some aspects of ME A will be described, and the way MEA can extend and/or improve the ite rative dichotomizer 3 (ID3) algorithm will be explained. MEA has been tested on a real-life application concerning the repairing of helicopt er blades. Results show that it is efficient, simple to use, and helpf ul to the expert.