Advanced Pattern Recognition method for the Monitoring of Dynamic Systems

The behavior of a dynamic system assumes different states over the time. In Pattern Recognition methods, each state is represented by a set of similar patterns forming restricted regions in the feature space, called classes. Recognizing the state, or class, of a new incoming pattern is performed using a membership function. In this paper, we propose to develop the supervised classification method Fuzzy Pattern Matching to be in addition a non supervised one. The goal is to monitor dynamic systems with a limited prior knowledge about their functioning. The detection of the occurrence of new states as well as the reinforcement of the estimation of their membership functions are performed online thanks to the combination of supervised and non supervised classification modes. No information in advance about the shape of classes or their number is required to achieve this detection and estimation reinforcement.