Neural Network and Machine Learning Allocation of Redundant Controls for Power Optimization on a Compound Helicopter

For a compound helicopter with control of main rotor speed, auxiliary propeller thrust, and stabilator pitch; a predictive neural network is trained to estimate power as a function of the redundant control settings for a range of flight speeds using a comprehensive database of 2,335 prior Rotorcraft Comprehensive Analysis System (RCAS) simulations. This neural network can be used as a surrogate model for a gradient based optimization to find the redundant control settings that produce a trim state with a minimized power requirement. The study highlights the importance of modeling relevant constraints (main rotor flapping and control limits), as well as having training data in the same locality that the predictions are being made. For producing accurate results with sparse initial datasets (20 trim states), a process of machine learning is demonstrated, wherein the neural network is iteratively updated using the resulting power from each prior optimization. When used appropriately, both the machine learning and fixed models demonstrate ability to allocate redundant controls such that the power requirement is within 1% - 3% of the true minimum.