Experimental results

The PI controller parameters were calculated and the resultant parameters are

presented in Table3.11

Axes Kc1 τI1 Kc2 τI2

Roll 0.2 0.1 7.07 0.283

Pitch 0.5 0.2 7.07 0.283

Yaw 0.7 0.1 n/a n/a

Figure 3.11: Identified inner loop controller parameters

It was experimentally found that the pitch’s inner loop system was produc-

ing large overshoot; therefore, by trial-and-error the pitch’s proportional gain Kc1

Step response tests were performed on the roll angle, pitch angle and yaw rate to evaluate the controller’s performance. The results are shown in Figure

3.12(a), 3.12(b) and 3.13(c) respectively, where the roll and pitch rate data are

plotted in Figure 3.13(a) and 3.13(b).

It is seen from all three step responses that the overshoots are quite high, which can be detrimental in real flights. In addition, the steady-state characteris- tics are also not ideal in all cases and the attitude seems to be slightly oscillatory.

96 98 100 102 104 106 108 Time (s) -10 -5 0 5 10 15 20 25 30 35 40 45 Roll angle ( ° )

(a) Roll angle

203 204 205 206 207 208 209 210 211 212 213 Time (s) -10 -8 -6 -4 -2 0 2 4 6 8 10 Pitch angle ( ° ) (b) Pitch angle

Figure 3.12: Step response of the roll and pitch angles: red (dashed) = refer- ence signal; blue(solid) = measured data

Overall, the autopilot’s performance is considered to be poor. This is mainly because of the common challenge model-based tuning method typically face, which is the negligence of dynamics. In the design of the controller, the higher order dynamics were estimated to first and second order models. Furthermore, in the selection of the gains for the roll and pitch outer loop controllers, the inner loop dy- namics were also simplified. Thus, making both roll and pitch outer loop estimated transfer functions to be identical. However, the flight characteristics between the roll and pitch axes are significantly distinct. Therefore, this approximation may not be the ideal solution for aircraft controller design. Another factor that con- tributed to the sub-par controller performance is the sensor dynamics. Due to the fact that the outer loop feedback are calculated with the complementary filter and low-pass filter, this may also introduce additional dynamics into the system. The negligence of the servo, inner loop and sensor dynamics are believed to be the main causes of the mismatch in the mathematical model and the actual physical system. Resulting in unfulfilled control performance requirements, because the design was done with inaccurate models.

96 98 100 102 104 106 108 Time (s) -300 -200 -100 0 100 200 300 Roll rate ( ° /s)

(a) Roll rate

203 204 205 206 207 208 209 210 211 212 213 Time (s) -200 -150 -100 -50 0 50 100 150 200 Pitch rate ( ° /s) (b) Pitch rate 429.5 430 430.5 431 431.5 Time (s) -20 0 20 40 60 80 100 120 Yaw rate ( ° /s) (c) Yaw rate

Figure 3.13: Response of the roll, pitch and yaw angular rates: red (dashed) = reference signal; blue(solid) = measured data

3.5

Summary

This chapter explored and reviewed the conventional model-based controller design of a fixed-wing UAV. The tranfer functions of the attitude system were identified by estimating the nyquist plot based on the plant’s frequency responses. The inner loop controllers were designed using the root locus method to manipulate the location of the poles and zeros to achieve the desired characteristics. On the other hand, the outer loop controllers were designed using the pole placement method, where the inner closed-loop dynamics were ignored. From the experiment, it has been shown that the unwanted effects from linearisation can be compensated with integrators in the feedback control systems, as the aircraft is able to follow its reference signals. However, the closed-loop performance was found to be sub- par, due to the neglected inner loop, actuator and sensor dynamics. The work performed in this chapter motivates the need for a better PID tuning method that

is able to deal with the neglected dynamics. The novel automatic controller design in the next chapter is, therefore, proposed as a solution to this problem.

Automatic Tuning of the Attitude

Control System

The goal of automatic tuning is to simplify and automate the tuning procedures without the need to determine the nonlinear parameters. Generally, a typical autopilot requires cascaded configuration, which in this case means it is necessary for five PI controllers to be tuned. This can complicate the tuning process and

increase the engineers’ workload (Kada and Ghazzawi, 2011; Luo et al., 2011).

The main idea of the automatic tuning algorithm is to fit the plant’s dom- inant dynamics into an integrating plus time-delay model. This method lies in the use of relay feedback control for generating a set of input and output data that contain the dynamical information of the plant. In contrast to the general approach of relay test, this method puts the system under proportional controlled feedback within the relay loop. This approach, then, is really useful when applied to integrating systems. This automatic tuning algorithm is particularly suitable for UAVs as the linearised models of the UAV’s outer loop systems have integrating dynamics and are unstable by nature

This chapter introduces the four steps of automatic tuning: the relay feed- back structure, estimation of the frequency response, determination of the esti- mated model and controller parameter selection. This is followed by a simulation study to explore the tuning capability of the autotuner. The fixed-wing UAV then

undergoes the automatic tuning process. The flight control system is also vali- dated with wind tunnel experiments with various testing conditions. The closed- loop performance is also compared against the controller design discussed in the previous chapter.

4.1

Automatic tuner design

This section provides the formulation of the automatic tuning algorithm and ex- plains how the PID controller parameters are chosen. The tuning process starts with the system structured in a relay feedback control system. The closed-loop frequency response is then determined based on the oscillatory feedback. This closed-loop information can then be inverted to obtain the plant frequency re- sponse and eventually the estimated integrating plus time-delay model. The con- troller parameters are selected based on the model and a well-established tuning rule.