Designing models for online use with Modelica and FMI

Abstract

Model-based online applications such as soft-sensing, fault detection or model predictive control require rep- resentative models. Basing models on physics has the advantage of naturally describing nonlinear pro- cesses and potentially describing a wide range of op- erating conditions. Implementing adaptivity is essen- tial for online use to avoid model performance degra- dation over time and to compensate for model imper- fection. Requirements for identifiability and observ- ability, numerical robustness and computational speed place an upper limit on model complexity. These con- siderations motivate that models for online use should be balanced-complexity, physically based with online adaption possible. Despite potential benefits, the effort required to im- plement balanced-complexity models, particularly at large scales, may deter their use. This paper presents techniques used in the design of balanced-complexity models. A Modelica-based approach is chosen to reduce implementation effort by interfacing exported Modelica models with application code by means of the generic interface FMI. The suggested approach is demonstrated by parameter estimation for a process of offshore oil production: a subsea well-manifold- pipeline production system.