| Process models based on first principles, herein referred to as physical models, often are available or it is feasible to develop them. These physical models usually are capable of capturing the dominant static and dynamic behavior of a process. Nevertheless, significant additional dynamics might be needed to represent the process which are not accounted for by the physical model. In this dissertation it is suggested to identify these additional dynamics, thereby reducing model uncertainties significantly and possibly increasing insight into the physical process. Thus, methods are devised to identify a combined physical/empirical model of a process, assuming that measured input/output data from the process are available.; One way to identify additional dynamics is to remove that part of the data which is accounted for by the physical model and to use the remaining altered data to identify the additional dynamics. This approach is referred to as data extraction in this dissertation.; When both the physical and the additional dynamics are linear, the above concept can be developed in a rather straightforward manner. However, in the event a nonlinear physical model is involved, some alteration in the identification scheme has to be considered.; The most challenging case of identifying unknown dynamics occurs when they operate in series with the known dynamics (because of potential identifiability problems). Therefore, this dissertation focuses on that case. It is shown theoretically for a simple dynamic system that the data extraction method provides advantages over the case when the complete process model must be identified. Using several simulated processes it is shown that the theoretical results carry over to the simulations.; For an experimental evaluation of the data extraction method, a stirred-tank heating process is used. The stirred-tank heater also exhibits decidedly nonlinear characteristics when the inlet flow rate is changed. This feature is used to demonstrate the nonlinear capabilities of the data extraction method.; The data extraction method is also applied to data obtained from an industrial process, the recovery boiler. A rather simple though nonlinear physical model of the process, based on an energy balance, is shown to predict quite well one of the most important relations in the recovery boiler, namely the relation between steam generation rate and black liquor firing rate. |
