This paper extends a recently suggested approach for the steady-state process optimization with guaranteed robust stability and flexibility under parametric uncertainty. The approach is based on measuring the distance of a candidate point of operation to so-called critical manifolds. Critical manifolds locally separate regions of the space of the process and controller design parameters with desired process traits from those regions with undesired traits. Pairs of desired and undesired traits may, for example, be feasibility and infeasibility of operation, stability and instability, or more generally, desired dynamical behavior and undesired dynamical behavior. While in previous applications knowledge on the existence and location of critical manifolds were assumed to be available before the optimization was attempted, the present paper presents an algorithm in which critical manifolds are automatically detected as the optimization proceeds. This algorithm, which is the conceptual contribution of the paper, allows the application of the critical manifold-based approach to processes for which no a priori information on the existence and location of critical manifolds exists. As a proof of concept the algorithm is applied to the reaction section of the HDA process. An analysis of the critical manifolds of this process model is not available. Since 12 uncertain parameters exists, analyzing the critical manifolds would be tedious. While an analysis of the 12 uncertain parameters is not practical, the critical manifoldbased optimization approach can be applied to models with this number of parameters and beyond.
IntroductionA new approach to taking constraints on the process dynamics into account in steady-state process optimization has recently been suggested. [1][2][3] The key features of this new approach are that (i) it can be applied to a broad class of dynamical features, among them stability and disturbance rejection; (ii) it guarantees the desired dynamical behavior despite parametric uncertainty in the process model; and (iii) it can be applied to ensuring the desired dynamics and process feasibility simultaneously. While the new approach is more general, a typical application is to guarantee stability and feasibility of the optimal point of operation despite parametric uncertainty of the process model when optimizing the process with respect to an economic objective function. 1 The two central concepts behind the new approach are the notion of a critical manifold and the normal space to that manifold. Readers not familiar with the notion of a manifold may think of the nonlinear generalization of a linear subspace of a vector space. Just as a plane defined by a single linear equation is a linear subspace of the linear vector space R 3 , nonlinear equations or sets of equations can define sets of points in R n . These so-called manifolds can locally be approximated by linear subspaces. Globally, however, nontrivial nonlinear manifolds may bend and fold to form nonlinear geometrical objects. The stability bound...