| Current methods for chemical process simulation and design (flowsheeting) are typically based on the sequential-modular approach, in which the computations for each sort of unit operation are organized into modules and solved sequentially. However, this approach has inherent limitations that make it ineffective in dealing with large and complex processes. One strategy for overcoming the current limitations is the equation-based approach. In this case, the computational modules are eliminated and all the equations describing a process solved simultaneously.; The development of a new process flowsheeting system is discussed. All the equations are linearized and all variables iterated on simultaneously using a Newton-Raphson or quasi-Newton approach. The system utilizes powerful sparse matrix routines permitting the handling of several thousand equations simultaneously in core. Even using efficient sparse matrix methods, the problems are of such size as to, at times, require decomposition. A new decomposition technique is presented. Initial guesses for the equation systems generated through the use of heuristics are found to be superior to guesses arising from simple averaging techniques. A Newton-Raphson solution strategy with step-size relaxation based on norm reduction performs well for both good and poor initial guesses. A simple hybrid solution strategy, blending first and second order methods performs well for some classes of initial guesses. Representation of all thermodynamics as part of the process matrix is explored. A strategy involving the partial nesting of thermodynamic iteration loops is found to be robust and efficient. Several comparisons are made. Linear solution time is seen to dominate the total process solution time. |
