This paper presents a method for optimization of real hardware with reduced need to access the hardware. It does so with a variable fidelity modeling approach. The models used are a high fidelity hardware model and a low fidelity physics based model. These models are invoked strategically to minimize the number of calls to the hardware. With this lower cost, designers can more fully explore design spaces that would otherwise be too expensive to explore using traditional design of experiments on hardware-in-the-loop systems. We show that the presented method identifies an optimal design more efficiently than a Box-Behnken DOE based optimization. Results are presented based on physical hardware tests with a cantilever beam in bending and a Butterworth filter. These examples provide a step toward use of the presented method on a flapping mechanism.
展开▼