Subspace angles for fault detection

摘要

In this paper we present subspace angles between two models and subspace angles within a model. We showrnhow they can be used in fault detection problems. From these angles, model norms and distance measures arernderived that are related to the cepstrum of the models. We show that one of the norms is equivalent to the mutualrninformation of the past and the future output process. Finally, we show how the distance between a model andrnobservations can be defined and computed.