A Criterion to Identify Multivariable Processes Sensitive to Modeling Errors

A simple multivariable control design procedure is first to decouple the multivariable system and then design each controller separately. Decoupled systems may show high sensitivity to process modeling errors. A criterion indicating the sensitivity of the control structure to modeling errors will be presented. The criterion is based on the variation of the eigenvalues of the Gp*Gp (Gp is the process transfer matrix) with frequency. Stiff process models are highly sensitive to modeling errors. Simulation results demonstrate that the criterion provides a good estimate of the sensitivity of processes with or without time delays.