Rule evolution in order based diagnostic systems

The authors present a novel system designed to evolve sets of rule bases used to optimise the order of lists of data arrays. Based upon induction learning techniques, an algorithm is described which is able to learn the rules most appropriate to ordering data in an attempt to promote a particular trait. A classifier system is employed as the main sorting engine, with a genetic algorithm in place to evolve newer, more proficient rules. As a test-bench for the sorting technique, the algorithm was trained to optimise lists of suspect components derived from PCB test/repair stations, endeavouring to promote the true fault to the top of the list. The paper initially describes the environment into which the evolvable rule base has been integrated. It then proceeds to disclose the algorithmic workings of a proposed solution using a genetic algorithm based classifier system which has the ability to identify the true fault on average 80% of the time.