Robustness of multivariable linear controllers to process nonlinearities

Many nonlinear processes have highly structured model mismatches which make a controller, based on an inverse of a linear nominal model, more robust than is predicted from the theories that assume unstructured, norm-bounded model uncertainty. The structure of model mismatch due to process nonlinearities is analyzed via singular value decomposition of the locally linearized steady-state models at different operating conditions. It is shown that certain constraints on the decomposition matrices, for example one of the decomposition matrices being unaffected by the nonlinearities, have important consequences on robust stability. The implications of this analysis for the identification of ill-conditioned processes for robust controller designs are also discussed. Binary distillation columns and catalytic reactors are treated as examples of processes with such structured model mismatch