Diagnostic goal-driven reduction of multiscale process models

Fault detection and diagnosis in large-scale process systems is of great practical importance and present various challenging research problems at the same time. One of them is the computational complexity of the algorithms that causes an exponential growth of the computational resources (time and memory) with increasing system sizes [21]. One remedy of this problem is to decompose the system model and effectively focus on its relevant sub-model when doing the fault detection, isolation and loss prevention. Multi-scale modelling is an emerging interdisciplinary field that offers a systematic way of constructing, analyzing and solving dynamic models of large-scale complex systems [22]. The aim of this paper is to propose a model reduction approach based on multi-scale modelling of process systems for diagnostic purposes. Because lumped or concentrated parameter process models are the most important and widespread class of process models for control and diagnostic applications, therefore we also restrict ourselves to this case.