STUDIES ON THE DESIGN OF ROBUST CHEMICAL PLANTS (PROCESS, CONTROL)

In chemical industry, dynamic operability problems frequently arise due to poor process designs, inaccurate process models and the control systems that are unable to compensate for modeling uncertainties. Traditional methods like empirical overdesign cease to be an alternative approach for alleviating the dynamic operability problems. In this thesis, the uncertainty occurring in chemical plants is characterized to formulate pertinent dynamic operability measures, assessing the uncertainty tolerance of feedback control systems. Examples show that the conventional feedback structure is inadequate from process design standpoint because it fails to identify dynamic operability problems due to poor process designs alone and is biased by the given controller.; The model reference scheme (MRS) enables the designer to formulate dynamic operability within the process design context. Based on this representation, robustness indices are developed to aid the designer in quantifying the stability and performance characteristics of a chemical plant facing modeling uncertainties. This development, together with the evaluation of magnitude bounds on the uncertainty, remedies the conservatism associated with other process design methods. Moreover, a sensitivity analysis is established which systematically develops plant designs with improved robust dynamic operability. This methodology makes use of singular value decomposition concepts and a recent theory of singular value sensitivities. Examples indicate that this approach is very valuable in identifying critical design parameters and directions for further design decisions.; A classification of uncertain parameters is utilized based on their duration of variation and economic impact. This prompts the formulation of the design of robust chemical plants with multiregime capabilities. Two cases are identified where the process faces either predetermined operating regimes or totally uncertain operating regimes, and the resulting problem is solved within a multiobjective optimization framework. A computer software is developed that handles this problem as a nonlinear two-stage programming problem. A case study demonstrates its interactive use and effectiveness in solving a broad class of process design problems in the presence of modeling uncertainties.