Flow Control Optimization Using Genetic Algorithms with Reduced Order Modeling

The theory of active flow control has been extensively developed and applied to mitigate undesirable flow features of aerodynamic systems. However, active control is often difficult to implement for applications described by complex, nonlinear physics. Genetic algorithms (GA) offer an attractive alternative by mimicking natural selection to converge on an optimal control input for a given objective function. The principal benefit of the GA for our purposes is that it is data driven, i.e., agnostic to the governing equations of the flow and thus does not incur simplifications typically adopted with traditional control approaches. In this work, a real-coded GA is first validated using a two-dimensional, algebraic test function as a surrogate fitness function; this exercise guides the choice of mutation, selection, and crossover parameters for rapid convergence to the optimal solution. The GA is then considered for the problem of a supersonic planar impinging jet to mitigate noise from aeroacoustic resonance modes. The formulation uses a dynamic mode decomposition based reduced order model (DMD-ROM) from a large eddy simulation (LES) database to provide a very economical fitness function evaluation. This allows various combinations of control input forcing variables of notional actuators near the nozzle (amplitude, frequency and phase) to be tested, that would be cost prohibitive without the ROM. Results on the efficiency of the GA in finding the optimal energy forcing gain relative to a brute force parametric sweep are discussed. Ongoing work will exhibit how the GA converges on a control scheme to reduce jet noise with more complex fitness functions.