Temperature–Strain Sensor Selection and Dimensionality Reduction for High-Speed Aerodynamic Recovery

This paper presents two fundamental components to enable the implementation of the novel vehicle-as-a-sensor concept. This concept consists of a nonintrusive sensing strategy for airframes that aims to estimate in-flight aerodynamic loads using sensors that are located in the vehicle’s internal structure. These sensors measure strain and temperature at limited locations, and this information is then fed to a neural network inverse model to obtain the desired loads. The first fundamental component relates to determining the number and location of these sensors by using the eigenvalues and eigenvectors of the observability gramian to rank the sensor locations by their importance in the recovery of aerodynamic loads. The second component relates to the ability to recover pressure. It is assessed by a numerical investigation that considers computational fluid dynamics data in high-speed flight conditions. The reduction of the dimensionality of the pressure distribution is addressed by two methods that are compared for efficiency: application of proper orthogonal decomposition versus an autoencoder neural network. The results show that it is possible to obtain a reasonable set of sensors that are capable of inferring all states of interest of the system without performing an optimization, which can be costly due to the combinatorial nature of the sensor selection problem. They also show that the autoencoder neural network is a promising approach to reduce the dimensionality of the recovered pressure distribution for this vehicle-as-a-sensor concept in high-speed flight.