Vertical take-off and landing (VTOL) aircraft have become important aerial vehicles for various sectors, such as security, health, and commercial sectors. These vehicles are capable of operating in different flight modes, allowing for the covering of most flight requirements in most environments. A tail-sitter aircraft is a type of VTOL vehicle that has the ability to take off and land vertically on it elevators (its tail) or on some rigid support element that extends behind the trailing edge. Most of the tail-sitter aircraft are designed with a fixed-wing adaptation rather than having their own design. The design of the tail-sitter carried out in this work had the particularity of not being an adaptation of a quad-rotor system in a commercial swept-wing aircraft, but, rather, was made from its own geometry in a twin-rotor configuration. The design was performed using ANSYS SpaceClaim CAD software, and a numerical analysis of the performance was carried out in ANSYS Fluent CFD software. The numerical results were satisfactorily validated with empirical correlations for the calculation of the polar curve, and the performance of the proposed tail-sitter was satisfactory compared to those found in the literature. The results of velocity and pressure contours were obtained for various angles of attack. The force and moment coefficients obtained showed trends similar to those reported in the literature.

Rocket Thrust Vectoring Attitude Control based on Convolutional Neural Networks

En aplicaciones de control inteligente, uno de los problemas es determinar el número de capas y de neuronas en cada capa. Esto se vuelve más complejo con ciertos tipos de redes neuronales como las basadas en wavelets donde las traslaciones y dilataciones son parámetros adicionales. Este artículo presenta un estudio comparativo para determinar el tipo de wavelet y número de neuronas que muestran el mejor desempeño para controlar un helicóptero Quanser de dos grados de libertad (GDL). Se presenta un controlador tipo PID-WaveNet-IIR el cuál se compone de controladores PID discretos en el tiempo, con ganancias auto-sintonizables por una red neuronal de base radial cuyas funciones de activación son wavelets y un par de filtros de respuesta de impulso infinito (IIR) para ``podar'' algunas neuronas. Por medio de simulaciones numéricas, usando LabVIEW, se presenta el desempeño del sistema en lazo cerrado para: diferentes condiciones de operación, tipos de familia wavelets donde se fijan valores mínimos de los errores de seguimiento, tipo de wavelet, el número de neuronas de la red y número de coeficientes de adelanto y atraso de los filtros IIR.