Adaptive low-order posi-cast control of a combustor test-rig model

S. Park, D. Wee, A. M. Annaswamy, A. F. Ghoniem

Research output: Contribution to journalConference articlepeer-review

5 Scopus citations


Recently, an adaptive posi-cast controller has been developed for dynamic systems with large time-delays. In this paper, we evaluate its performance in the context of a 4 MW combustor test-rig model that mimics many of the dynamic characteristics of an actual engine including a significant time-delay. Using closed loop input-output data and system identification, a model of the test-rig was derived. Using this model, adaptive posi-cast controllers were designed and detailed numerical simulation studies were carried out. These studies consisted of (i) the closed-loop performance of the adaptive controller, (ii) comparison of the adaptive controller with an empirical phase-shift controller, (iii) robustness with respect to parametric uncertainties, (iv) robustness with respect to unmodeled dynamics and uncertain delays, (v) performance in the presence of noise, and (vi) effect of saturation constraints on the control input amplitude. These studies show that the adaptive posi-cast controllers decrease pressure oscillations much faster than the phase-shift controller without generating peak splitting. It also showed that the adaptive controller was capable of stabilizing the plant in the presence of 20% change in the resonant frequency, an order of magnitude change in the damping ratio, and unmodeled dynamics. The studies showed that the performance of the adaptive controller improved as the magnitude of the saturation constraints on the control input increased.

Original languageEnglish
Pages (from-to)3698-3703
Number of pages6
JournalProceedings of the IEEE Conference on Decision and Control
StatePublished - 2002
Event41st IEEE Conference on Decision and Control - Las Vegas, NV, United States
Duration: 10 Dec 200213 Dec 2002


Dive into the research topics of 'Adaptive low-order posi-cast control of a combustor test-rig model'. Together they form a unique fingerprint.

Cite this