Dynamic learning of decision trees to acquire knowledge for the diagnosis of dynamic systems

Inductive learning methods became an essential part of knowledge acquisition tools for diagnosis components of expert systems. If dealing with dynamic systems, not only the current state variables are relevant to the classification process of technical system´s states but also their histories. These requirements result in the fact, that the state space seems to be not finite. Up to now, the way out was to base the induction process on extended state vectors built by subsequently concatenating original state vectors. On the one hand, this enlarges the amount of information to be processed enormously. On the other hand, it may be still insufficient, if the depth of history required is not known in advance. This article presents an approach, where the access to the historical values of process data immediately depends on the given learning data set. This dynamic access is called data-driven access to historical values. The use of the presented strategy results in an optimum of efficiency without prior restrictions of the hypothesis space. Consequently, the presented approach is able to generate classifiers in the form of decision trees in a very effective way