Model reduction for control of infinite-dimensional systems in industrial applications

Thursday, March 17, 2016 - 11:00am - 11:30am
Keller 3-180
Milos Ilak (United Technologies Corporation)
This overview presentation will cover applications of modeling, control, and estimation for infinite-dimensional systems at United Technologies Research Center (UTRC), the research arm of United Technologies Corporation (UTC). The underlying theme is model reduction as key enabler of control, estimation, and sensor placement. Two examples in climate control – reduced-order modeling of transport of contaminants and mixing in buildings (Surana et al, ACC, 2008), and control design for rejection of temperature disturbances in rooms with stratified flows (Ahuja et al, ACC, 2011) represent earlier UTRC work in model reduction. More recent efforts have been focused on the spectral decomposition of the Koopman operator, which allows for a global linear representation of nonlinear dynamics, and is therefore advantageous in comparison with local linearization. Dynamic Mode Decomposition, a method of growing popularity in fluid mechanics, provides a computationally straightforward approximation to the Koopman decomposition. As a first step towards low-order models for combustion control, DMD has been applied to unsteady heat release experimental data for a model combustor with a low-swirl injector (Therkelsen et al, Combustion and Flame, Vol 160 (2), Palies et al, APS DFD, 2015), and the underlying instability mechanism has been identified. Extensions of DMD-based model reduction towards control are an area of ongoing work at UTRC. Current work and open problems related to this topic will be discussed.