Diagnostic Goal-Driven Reduction of Multiscale Process Models

Abstract

A systematic model reduction method is proposed in this paper that is applicable for multiscale process models with steady-state operating points for prediction-based fault detection, isolation and loss prevention in the initial phase of a fault. Based on the time-scale separation of the dynamics a refined scale-map, a model structure map is constructed for each root cause and symptom and the relevant target time scale is determined. Then the scale-reduced nonlinear model of the system is constructed that only contains the dynamic balance equations on the target time scale with all the other dynamic equations reduced by steady-state assumptions. Thereafter local linearization is applied to obtain the final reduced model.The resulted set of reduced models preserves the engineering meaning of the remaining state variables, while the number of state variables and parameters is decreased significantly.