A necessary prerequisite for model-based control is a model of the process. Such certainty-equivalence, model-based control schemes rely on an off-line estimated model of the process, i.e., the process is “probed” or excited by a carefully designed input-signal under open-loop conditions and the input-output data are used to generate a suitable model of the process. In a majority of model-based control schemes used in the chemical process industry, the models are generated with little regard for their ultimate end-use, e.g., as in model-predictive control. Almost always, in such cases, reduced-complexity models are generated to capture the most dominant dynamics of the process. Such batch or off-line identification methods represent a major effort and may require anywhere from several hours to several weeks of open-loop tests.
KeywordsSensitivity Function Bias Error Shaping Filter Residual Test Bias Distribution
Unable to display preview. Download preview PDF.