| The sequential modular approach to chemical process simulation has been the most common for industrial applications. Because its calculation sequence is fixed by the flowsheet, efficient optimization methods that alter the calculation procedure are difficult to implement. Often, slow and inefficient direct search methods are used to optimize simulated processes.; This thesis introduces a family of methods that use the efficient successive quadratic programming (SQP) methods recently developed for nonlinear optimization. The individual algorithms differ in the way gradients are calculated and tear streams are handled. All of them are faster than present sequential modular optimization methods.; The first algorithm, Quadratic/Linear Approximation Programming (Q/LAP), constructs linear block models and combines them in a large, sparse linear system after the flowsheet has converged. Gradients with respect to design or decision variables are then evaluated by perturbation of the linear model.; The infeasible path algorithm (IPOSEQ) does not require flowsheet convergence. Instead the tear equations are linearized and solved in the quadratic programming step. Gradients for the objective function and constraints are calculated by perturbing variables and passing through the calculation sequence, leaving the simulator executive undistributed.; The last two algorithms are feasible variants of the infeasible path algorithm. Here a converged flowsheet is required for each objective function evaluation but gradients are calculated as in the infeasible path algorithm. The converged Feasible Variant (CFV) includes tear variables and equations in the quadratic program while the Reduced Feasible Variant (RFV) calculates reduced gradients by eliminating these equations. Subtle ways in which the quadratic program handles these formulations affect their algorithmic performance.; All of these methods were tested on small flash optimizations to study their performance and the effect of algorithmic tuning parameters. A case study on an ammonia synthesis optimization as presented for Q/LAP and a comparison of all four algorithms is given in a complex propylene chlorination case study.; Disadvantages and advantages are given for each algorithm and improvements are suggested for future research in this area. |
