Constrained minimum crest factor multisine signals for »plant-Friendly» identification of highly interactive systems

Highly interactive systems are ill-conditioned and highly sensitive to model uncertainty, which imposes limitations to achievable closed-loop performance. In this paper, the goal is to develop an identification testing framework meaningful to highly interactive systems based on the application of constrained minimum crest factor multisine signals. »Plant-friendliness» in the design procedure is accomplished by imposing constraints on the overall span, move size, and variability of both input and output signals. A modified 'zippered» power spectral density is proposed which contains both correlated and uncorrected harmonics to simultaneously excite low- and high-gain directions in the data. This signal design procedure is applied to a simple 2-by-2 multivariate system that exhibits challenging interaction and gain directionality issues. Validation of these signals in a closed-loop setting is accomplished by evaluating the performance of decentralized PI with sleady-state decoupler controller designs based on models estimated from noisy open-loop experiments. The modified »zippered» spectrum signals present a clear advantage over standard zippered designs for short data sets under noisy conditions.