Neurocontrol design for an aerodynamics system: simple backpropagation approach

Abstract

This paper proposes a Neurocontrol (NNC) for a Twin Rotor Aerodynamics System (TRAS) by a simple backpropagation approach to improve the pitch position accuracy. A concept known as gradient descent method is applied to adjust the weights adaptively. The approach has several notable merits namely low computational cost, simple and promising controller. The viability of NNC is verified by using MATLAB to analyze the tracking performance and control effort. PID control is benchmarked against the proposed NNC to determine the effectiveness of the controller. From the simulation work, it was discovered that NNC was superior then PID controller by reducing about 14%, 23% and 97% in the value of the overshoot, settling time and steady-state error respectively. The promising part of NNC was the improvement shown in the controller effort by significantly eliminating the fluctuation and chattering in the control signal. By looking into the future, this work will be a foundation for future improvement due to the fact that there are numerous types of approaches could be embedded in the Neural Network algorithm.