Process Simulation

Process simulation or "flowsheeting" has found widespread use in the chemical industries for the evaluation and optimisation of process designs and the investigation o f plant operation. A process simulation is carried out using computer models which mathematically describe the actions of major items o f equipment or unit operations in a process thereby enabling mass and energy balances to be performed. The models may be performance based, i.e. feed stream information and equipment parameters are supplied and product stream information calculated, or design based, i.e. a combination o f feed and product stream information relating to equipment function are supplied so that design calculations such as sizing and costing can be carried out (Jackson and De Silva, 1985).

The simulation problem itself is defined by:

• providing physical property data or calculation procedures

• supplying the required feed stream information

• specifying equipment parameters or product stream information depending on the type o f models used.

A process simulator is a computer program which sets up the simulation problem with the input o f the user and then carries out its solution.

There are three different methods o f solving the model equations within a process simulator:

• the sequential modular approach

• the equation-orientated approach

• a combination o f the sequential modular and equation-orientated approaches.

Sequential modular simulators were first developed in the late 1950's for use in the petrochemical industry (Biegler, 1989). In the sequential modular approach the output o f each unit is calculated as a function o f the input variables. The unit models are solved one after the other so that the calculation sequence parallels the flow o f material in the actual process. Since each unit model is a self contained module, a large flowsheet can be constructed very quickly simply by specifying unit connections. The model equations within each unit module can also be altered, from simple to more rigorous, with minimum changes to the flowsheet topology. The main problem with this method is that awkward iteration loops are required when there are recycle streams and when design or optimisation calculations are performed (Biegler, 1989). The simplified structure o f a sequential modulai simulator is shown in Figure 1.7 overleaf.

In the academic community, most early research in process simulation was devoted to the equation-orientated approach (Biegler, 1989). In this approach, all o f the unit equations are collected together and solved simultaneously. Flowsheets containing recycle streams

or design constraints can therefore be solved more easily with an equation-orientated approach than a sequential modular approach. The large amount o f derivative and function information available in an equation-orientated simulator also allows more sophisticated optimisation strategies to be applied. However, this approach is limited by the capabilities o f the equation solver, typically a Newton-Raphson method, and convergence failures may occur. The simplified structure o f an equation-orientated simulator is shown in Figure 1.7.

Equation-Orientated Sequential Modular

Unit Operation Model Library Simultaneous Equation Creation and Solution Executive Program (Set-up flowsheet and unit equations)

Executive Program (Set-up flowsheet, sequence and control

unit calculations)

Physical Properties Library

Physical Properties Library

Figure 1.7: Structure o f process simulators, from Biegler (1989).

By the late 1970's and early 1980's, combinations o f the sequential modular and equation- orientated approaches were developed (Biegler, 1989). In simultaneous modular simulators, the sequential modular approach was modified to incorporate simultaneous solution o f the streams connecting the unit modules, although the unit modules themselves remained intact. Equation-orientated approaches were also modified to include procedural sub-modules for the calculation o f physical properties which linked in with the equation solver. The development o f these new approaches was aided by the application o f quasi-

Newton convergence methods, which require less derivative information than previously used Newton-Raphson methods but have a much faster convergence rate than sequential modular iteration loops (Biegler, 1989).

It is also possible to carry out two different types o f process simulation, known as steady- state and dynamic. In steady-state simulations the equations which make up each unit model are algebraic and therefore can only describe continuous processes at steady-state conditions. On the other hand, in dynamic simulations the equations used are both ordinary differential equations and algebraic equations. This means processes which display transient behaviour, such as batch operations, can be described. Due to the complexity o f dynamic systems, as much as ten times more computer power is required to solve the model equations then in steady-state simulations (Hunter et al, 1989).

Most currently available process simulators perform steady-state simulations and utilise the sequential modular approach for solving the unit model equations, with options for simultaneous modular convergence. Examples include ASPEN-Plus, DESIGN-II and PROCESS (Biegler, 1989) which are available in PC versions.

Process simulators capable o f dynamic simulation generally employ the equation- orientated approach. Examples o f dynamic simulators are SPEEDUP, developed in the 1960's at Imperial College (Pantelides, 1988; Sargent and Westerberg, 1964), QUASILIN (Smith and Morton, 1988) and more recently gPROMS (Pantelides and Barton, 1993).

Many o f the available simulators capable o f steady-state simulation have been extended to allow dynamic simulations o f some unit operations. New "hybrid" simulators are also being developed that combine steady-state and dynamic versions o f a simulator in the one package (Fouhy, 1991). Major differences between the available process simulators also lie in their libraries o f unit model equations and physical property data.