DATA-DRIVEN REDUCED ORDER MODELING FOR THE INTEGRATED ACTIVE ADAPTIVE ROTOR FRAMEWORK FATIGUE REDUCTION APPLICATION

Adaptive structures have become a topic of interest for improving performance across many aerodynamic domains, such as drones and other aircraft, due to the variety of potential beneficial effects made possible by manipulating the many aerodynamic properties of a structure. One type of morphing that is of interest is that of out-of-plane twist, where local twist values are modulated along the span of a wing or blade. This benefits the overall aircraft or system through the introduction of new methods and axes of control with which to manipulate the system. However, implementation of such a technology is faced with several challenges, ranging from difficulties imposed on materials used for implementation, to development of controls schemes utilizing the system, to the high computational cost of modeling a continuously morphable twist distribution. Faced with such challenges, this research effort takes on one at a time to develop an integrated Active Adaptive Rotor (AAR) framework featuring out-of-plane twist morphing. Previously, this was advanced through multiple phases of development. Groundwork was laid out through explicit analytic optimization of twist configurations for a specific system using a Genetic Algorithm (GA) and wind tunnel testing for validation of results. Work has also been done to bridge the materials concerns through the application of topology optimization and investigation into utilizing modular stiffness. Now, to approach the issue of computational expense for modeling the AAR system, in order to utilize this technology in real-time controls applications, a methodology for the rapid development and deployment of specific data-driven models, modeling specific aerodynamic response parameters, was developed. This paper presents an application of that methodology, simultaneously furthering the integrated AAR framework through the reduction in computational expense of modeling the aerodynamic responses, as well as exploring the range and depth of potential applications for the AAR technology. Specifically, this paper develops a data-driven Neural Network (NN) model which correlates environmental conditions, such as inflow wind speed, to AAR twist configurations to frequency-domain aerodynamic thrust response, then utilizes the spectral model in the minimization of fatigue damage by the system during a discrete operating time. Time-domain response data is simulated for the AAR system, through varying the span-wise out-of-plane twist distribution under otherwise identical conditions. Time-domain data is converted to frequency-domain through application of the Fast Fourier Transform (FFT). This data is then used to train a NN model correlating the twist inputs with the spectral response. The developed model is then used in a GA-based Pareto Front investigation, optimizing based on the fatigue damage calculated from the modeled spectral response. The optimization results show an 10.5%, 20.7%, and 29.4% improvement in fatigue damage reduction for steel, aluminum, and composites, respectively. This promising result is twofold: not only does the AAR system show potential in the ongoing effort to improve system health through the application of morphing structures, this result also assists in the overall reduction in computational expense of system controls through the success of the involved model.