Experimental Comparison of Two Integer Valued Iterative Learning Control Approaches at a Stator Cascade

Abstract

Experimental and simulative investigations have shown that active flow control (AFC) is an effective method to influence flow conditions within a compressor. This can be used for different cases like mitigating flow separation or to ensure a uniform flow throughout a compressor stage. Control performance can be improved by making use of a cyclic character found in the rotor/stator interaction or found in new gas turbine setups exploiting cycling combustion. To this end, iterative learning control (ILC) is applied. To achieve a fast actuation, irrespective of the implemented control method, solenoid valves should be installed instead of proportional valves. Unfortunately, the binary character of these valves does not allow the application of conventional control methods, e.g., real-valued ILC. This contribution presents two options to handle the binary control domain in the context of an ILC. Both approaches are tested in a simulation study first to analyze the behavior. Then they are applied to a real test rig featuring a linear stator cascade.

Norm-optimal iterative learning control in an integer-valued control domain

“…Axehill et al (2010) solved a discrete control problem in the context of model predictive control (MPC). Bürger et al (2019) discussed an MPC setup for a nonlinear switching system, and Arnold et al (2020) compared two approaches for binary iterative learning control (ILC).…”