Reduced dimension state feedback anfis controller design for quadrotors

Quadrotor, one of the Unmanned Aerial Vehicles, consists of propellers mounted on four identical rotors, positioned symmetrically on a frame. Thanks to these rotors, it can take off and take off vertically, stay suspended, and can create forward-backward movements, pitch, roll and yaw angles by changing the rotational speeds of the engines. It is very important to design an effective control system because they are used in many risky and dangerous works such as defense, security, search and rescue, and because of their non-linear behavior. In this study; After giving brief information about the definition, types and characteristics of unmanned aerial vehicles, the mathematical model of quadrotor in the form of state space has been derived out. Since quadrotors are a non-linear system, the control system is usually designed on the linearized model. In this thesis, ANFIS (Adaptive Neuro-Fuzzy Inference Systems) is used in the form of a state feedback controller to control all four rotors of the quadrotor, which is a four-input and 12th order system. The training data for ANFIS were obtained by applying the LQR (Linear Quadratic Regulator) control method to the linearized model of the quadrotor. However, since the quadrotor is a 12th order system, the ANFIS used as a state feedback requires 12 inputs, which will take a long time to train ANFIS and it will be difficult to implement in practice because the processing load of ANFIS will be too high. For this reason, generally reduced-dimensional ANFISs have gained importance in high-order systems. In this study, 12 state variables are passed through the Fusion Function and reduced to two inputs representing the error and the derivative of the error for each rotor. Thus, four rotors are controlled by four ANFIS with two inputs instead of 12 inputs. The study was carried out in a software environment, since there were no experimental implementation possibilities. In the control system established with MATLAB/Simulink, the results obtained from both the LQR control method and the reduced size ANFIS controller were evaluated and the effectiveness of the control performance was demonstrated.