A case study of grey box identification

Abstract Experiment data from a strip steel rinsing process are used as a test case of Grey Box identification, i.e. of designing a nonlinear stochastic dynamic model for the process, when some but not all of the physical phenomena behind its behavior are known, a priori knowledge is uncertain, and the process is subject to unknown disturbances. The purpose of the study is twofold: 1. (1) To find a working procedure for carrying out interactive system identification, in particular using the Grey Box identification tool named IdKit. The procedure comprises a sequence of hypothesis testings and parameter fittings, and involves the designer intimately in the interactive ‘loop’ to contribute ‘engineering sense’. 2. (2) To illuminate the choice between internal-noise, external-noise, or no-noise (deterministic) structures, in particular whether the result may be worth the effort of using optimal state-variable filtering.