Mechanistic Non-linear Models

Non-linear physical models tend to be relatively complex and nsnally leqniie considerable development effort to arrive at an adeqnate form. In addition, they reqnire intimate process knowledge and understanding. Their primary nse has been in process design and simnlation. It is usually the steady-state version of these models that is used to size new equipment or to simulate the process at different conditions such as feed rate, temperature and pressure, etc. These steady-state models are also useful for supervisory control where they can be used in an on-line or offline mode to optimize the process conditions at various time intervals and readjust the setpoints of the controllers in a basic control structure. 

Dynamic forms of these mechanistic models may be useful in either simulation, prediction or real time control. They can, among others, be used to develop start-up procedures for continuous processes. They could also be used for optimal control of batch processes, where a process variable has to follow a pre-defined trajectory. 

Unless one is interested in operating the process over a wide range of conditions, the simple (linear) empirical models discussed in the next section are easier to develop. 

One of the major drawbacks of empirical linear models is the limited range of applicability. Especially extrapolation capabilities of the models beyond the region for which they were developed can be poor owing to process non-linearities. One way to avoid this is to adapt the process model parameters to the changing process conditions. 

---