Optimization for Mars Rotorcraft Blade: Weighted Ensemble Neural Networks and Spatial Discretization

Compared to the Mars rover, the Mars rotorcraft’s ability to take off and land vertically makes it more capable of performing missions on Mars, as its detection missions are not restricted by terrain and its detection efficiency is higher. This paper introduces a Mars quadcopter capable of conducting sampling detection on the planet’s surface. This paper optimizes the chord length distribution of the blades of Mars quadcopter by using flight power consumption as a constraint to enhance the aircraft’s overall performance. The thin and cold atmosphere of Mars forces the blades to operate under low-Reynolds-number and high-Mach-number conditions, leading to reduced thrust and increased drag. This paper proposes an optimization method that combines weighted ensemble of neural networks with spatial discretization to optimize the blade of Mars quadcopter. This method employs 4 optimizers to train neural network models, each designed to fit the mathematical mapping relationship between 14 variables for blade chord length parameter distributions and performance values. This paper integrates the aforementioned 4 models using a weighted ensemble method. Building on the concept of spatial discretization, this paper simplifies infinite parameter combinations into finite ones. Subsequently, the neural network is used to predict the performance values of various parameter combinations, enabling the selection of high-performance blade. This paper conducted a single-rotor lift–drag characteristic test on the optimized blades in an environment with an equivalent Martian surface atmospheric density of 0.016 kg/m³.