Flight Emulation with Neural Networks for Event Characterization: Center of Gravity

Rotary winged aircraft offer a wide range of capabilities, but several factors limit their handling qualities. One of these limiting factors is the location of the center of gravity. The dry center of gravity is given by CAD models from suppliers, which can be unreliable. Aircraft weight configuration can change for various reasons, such as fuel consumption or releasing payloads during flight. Sudden changes in the center of gravity can negatively impact the stability of the aircraft. Having information on the center of gravity improves situational awareness, enables better feedback control methods, and provides information on changing aircraft properties. This project proposes to use a categorical neural network to estimate the center of gravity location of a SAB Goblin RAW 420 radio-controlled helicopter. A manually changeable center of gravity was installed, allowing one of five possible battery locations; multiple flights have been conducted in each configuration. Flight dynamics and controls data from the inertial measurement unit and flight controller was gathered during flights to use as inputs for the neural network. The network was trained on the data and tested using a series of random data sets to return a center of gravity location, which was compared against ground test data gathered from a static airframe weight and balance test. Results show that the model is capable of processing multiple center of gravity locations.