Modelling abrupt changes: enhanced learning of behaviour models for manufacturing systems

Abstract

Modern manufacturing systems are complex technical systems that exhibit state-based, continuous, timed, and probabilistic behaviour. Modelling such systems is becoming increasingly hard, and yet their behaviour models are today mostly created manually. This paper gives an asset to learning these models automatically from data The HyBUTLA algorithm for learning the hybrid automata models, which can represent manufacturing system’s characteristics, has been recently proposed. However, it could not model the abrupt changes in the continuous part of the system. The contribution of this paper is as follows: the split function that detects and models abrupt changes is presented; both sufficient and necessary conditions for its success are formally proven; the complete HyBUTLA algorithm enhanced with the split function is given; experimental results conducted in a real manufacturing system are presented.