Kyle Nguyen

Pratik Rathore
Project Report: Twin Rotor MIMO System

Part 1: Introduction

Our objective is to control the position of a beam with two degrees of freedom; in other
words, the ends of the beam are allowed to rotate in such a way that they move along a spherical
surface. The beam is fixed on a pivot (in the middle of the beam) that is holding it in place that
allows the ends to rotate in a spherical manner. One motor fan will be attached on each end of the
beam. The normal vectors to the surfaces of the two motor fans will be perpendicular to one
another in order to simulate two degrees of freedom (i.e. one motor fan will be facing upwards,
while the other motor fan will be facing sideways). A counter-weight will be fixed perpendicular
to the beam and will determine a stable equilibrium position.

Figure 1: Example of our system [1]

The signals that will be measured from this system are the azimuth and elevation angles
associated with the beam. We can use the motors as a way of controlling the azimuth and

1
elevation angles to bring the beam to a desired position. If the measured azimuth and elevation
angles aren’t to our desired position, we can adjust the voltage to the motor fans until we get our
desired position. This is a closed loop system because we are constantly measuring the difference
between the actual and the desired azimuth and elevations angles. This system will be modeled
in Simulink for this report. The equations for the system are derived based on [3]. As explained
below in Part 2, we tried the coupled state space model presented in [2] and found there to be an
issue with the inputs. After more research, we decided to go with a decoupled state space model,
one for pitch and one for yaw. The actual equations for the state space models can be found in
[3].

We also experiment with PID control to design several controllers for this system. In
particular, we analyze the percent overshoot, rise time, and settling time, and quantify tradeoffs
between these various system parameters. These system parameters are found by applying a step
response at the input. The PID controller will be tuned using the sisotool functionality in
MATLAB.

One of our criteria is to test the controller robustness to disturbances. To test the
robustness of our controller, we perform two different experiments. In the first experiment, we
apply band-limited white noise at the controller input, and analyze how the noise power affects
the system response. In the second experiment, we apply a temporary disturbance (i.e. a
time-limited pulse) to the input, and quantitatively analyze the time it takes for the system to
revert back to the steady state. The results of these experiments provide us insights into whether
our controller is suitable for controlling a real-world implementation of the beam-fan system.

Part 2: State Space Model

We originally implemented the state space model described in [2]. This model couples
the pitch and yaw angles together. However, we found that this model, shown in Figure 2.1, was
always unstable, no matter how we tried to tune it.

2
Figure 2.1: Simulink model of the coupled state space system described in [2]. The inputs are
the desired pitch (Psi) and yaw (Phi) angles. The system contains a PID controller to improve the
system characteristics (percent overshoot, rise time, etc…).

As seen in Figure 2.2, the pitch and yaw angles given by the model rapidly diverge
towards infinity. Our difficulties with tuning the system shown in Figure 2.1 led to us looking for
alternate methods to model the twin rotor system.

3
Figure 2.2: Pitch and yaw angles outputted by the system shown in Figure 2.1.

After further research, we found a decoupled state space model with equations shown in
[3]. After implementing this system, we found that this system can be made stable under certain
conditions. We investigate the conditions for stability in Parts 3 and 4. The Simulink model we
used for the decoupled state space system is as below:

Figure 2.3: Simulink model of the decoupled state space systems described in [3]. The inputs are
step functions that represent motor control voltages. The system contains two PID controllers to
improve the system characteristics (percent overshoot, rise time, etc…)

Part 3: PID Controller

In Figure 2.3, the gains in the feedback loop for pitch and yaw are 0.3 and 0.1,
respectively. Assuming the system is stable, this leads to steady-state gains of 1/0.3 = 3.33 and
1/.0.1 = 10 at the pitch and yaw outputs, respectively.

4
 Without any PID control in both the pitch and yaw subsystems, (i.e. P = 1, I = 0, D = 0),
and step amplitudes of 0.02 at both inputs, we obtain an unstable system in yaw, as shown in
Figure 3.1.

Figure 3.1: Plot of pitch and yaw outputs with no PID control. Simulation time is 200 s.

By tuning the PID controllers using sisotool, we can make the system stable and also
control system characteristics such as percent overshoot, rise time, and settling time.

5
Figure 3.2: Sisotool applied to the pitch subsystem PID controller.

When sisotool is applied to the PID controller in the pitch subsystem, as shown in Figure
3.2, the percent overshoot, rise time, and settling time are 6.52 %, 3.89 s, and 74.5 s,
respectively. Despite our best efforts, sisotool was unable to give us a faster response for the
pitch subsystem; when we tried to decrease settling time at the expense of having a higher
percent overshoot, the system would always become unstable. Interestingly, the I term in the PID
controller was the only non-zero term provided by sisotool. We settled on P = 0, I = 2.944, D = 0,
and N = 100 for the pitch PID controller.

6
Figure 3.3: Sisotool applied to the yaw subsystem PID controller.

When sisotool is applied to the PID controller in the yaw subsystem, as shown in Figure
3.3, the percent overshoot, rise time, and settling time are 56 %, 0.223 s, and 3.93 s, respectively.
Notice that the PID controller has now made the yaw subsystem stable, unlike in Figure 3.1. In
addition, the yaw subsystem with PID has a much smaller settling time than the pitch subsystem
with PID; however, this comes with the cost of having a much higher percent overshoot. We
settled on P = 43.447, I =1.673, D = 94.688, and N = 554.276 for the yaw PID controller.

7
Figure 3.4: Pitch and yaw outputs of the entire system with PID controllers described above.
The inputs are step functions with amplitude 0.02.

As seen in Figure 3.4, the PID control for yaw stabilized the yaw subsystem. We can also
see that the yaw subsystem has gain 10, and the pitch subsystem has gain 3.33.

Part 4: Stability and Instability — Feedback Gains

While working with the system, we found that changing the feedback gains would affect
the overall system gain. Assuming that the feedback gains for the pitch and yaw subsystems were
Kp and Ky, respectively, we would expect overall steady-state gains of 1/Kp and 1/Ky,
respectively.

However, certain feedback gains could lead to instability in the subsystems.

8
 Kp 0.01 0.03 0.1 0.3 1

Stable? Yes Yes Yes Yes No

Figure 4.1: Stability of pitch subsystem for various feedback gains.

Figure 4.2: Example of unstable system in pitch. The feedback gain is 1, the input is a step
function with amplitude 0.02, and the PID controller is the one found in part 3.

Ky 0.01 0.03 0.1 0.3 1

9
 Stable? No Yes Yes Yes Yes

Figure 4.3: Stability of yaw subsystem for various feedback gains.

Figure 4.4: Example of unstable system in yaw. The feedback gain is 0.01, the input is a step
function with amplitude 0.02, and the PID controller is the one found in part 3.

As seen in Figures 4.1, 4.2, 4.3, and 4.4, changing the feedback gains for the two
subsystems can create an unstable system. For the pitch subsystem, increasing the feedback gain
creates an unstable system; for the yaw subsystem, decreasing the feedback gain creates an
unstable system. Interestingly, the changes that produced instability across the subsystems were
not the same — pitch became unstable while decreasing feedback gain, but yaw became unstable
while increasing feedback gain. These findings suggest that the stability of the system is highly
sensitive to both the PID controller and feedback gains.

10
Part 5: System Robustness

Next, we are going to test the robustness of our system by performing two experiments:
applying a band-limited white noise as a disturbance, applying a temporary disturbance
(time-limited pulse) and seeing how long it will take for the system to return to steady state.
After applying the band-limited white noise as a disturbance, we get the following model:

Figure 5.1: Simulink model as in Figure 2.3 with a band-limited white noise added as a
disturbance. The noise power of the band-limited white noise is set to 0.001 and the sample time
is set to 0.1.

After running the simulation, we get the following results:

11
Figure 5.2: Response of the system with band-limited white noise added as disturbance. Noise:
0.001, Sample Time: 0.1.

We can also test with decreasing noise power for the band-limited white noise:

12
Figure 5.3: Response of the system with band-limited white noise added as a disturbance. Noise:
0.00001, Sample Time: 0.1

As you can see from above, adding a band-limited white noise as a disturbance only
increases oscillation of both pitch and yaw. We can also see that decreasing the noise power not
only decreases the oscillation of yaw, but also decreases the minimum settling time for the
signal.

To experiment with pulse disturbances, we added a pulse that drops the input to 0 at t =
30 seconds, which is well after the system has reached steady state. The simulink model is as
follows:

13
Figure 5.4: Simulink model used to test the effect of a pulse disturbance added to the system

From the above model, we get the following results:

Figure 5.5: Response of the system with a pulse disturbance added at t = 30 secs

14
 As you can see above, when applying a pulse disturbance at t = 30 secs, it takes the
system approximately 7.5 seconds to return to the steady state. Therefore we can conclude that
our system is robust to pulse disturbances.

We can now run the simulation for longer so that we can calculate the percent overshoot,
rise time, and settling time of our system with a pulse disturbance:

Figure 5.6: Response of the system with a pulse disturbance added at t = 30 secs.

From Figure 5.6, we can tell that the percent overshoot, rise time, and settling time for
yaw are 65.189%, 0.609 seconds, 8.352 seconds respectively. We can also see that the percent
overshoot, rise time, and settling time for pitch are 5.32%, 7.77 seconds, 63 seconds (estimate).
Comparing these to our results above, we can see that the two responses are similar and that
adding a pulse disturbance shows the robustness of the system.

Part 6: Conclusion

15
 From our findings above, we found that a decoupled state space system works best for
our Twin Motor MIMO System with a feedback gain for the pitch and yaw to be 0.3 and 0.1
respectively. After using sisotool to tune the PID controller to tune the characteristics of the
system (percent overshoot, rise time, settling time), we obtained PID controllers with
characteristics P = 0, I = 2.944, D = 0, and N = 100 for pitch and P = 43.447, I =1.673, D =
94.688, and N = 554.276 for yaw.

After checking the stability of our system, we found that increasing the feedback gains
for the two subsystems can create instability. For the pitch subsystem, increasing feedback gain
creates instability while for the yaw subsystem, decreasing feedback creates instability. From this
data, we ultimately came to the conclusion that the stability of the system is highly sensitive to
both the PID controller and feedback gains.

After adding in a band-limited white noise as a disturbance to our system, we found that
an decreased noise power will result in less oscillations for the system. We can therefore
conclude that increasing the noise power will result in more oscillations. By performing the pulse
disturbance experiment, we found that our system is robust to pulse disturbances. The percent
overshoot, rise time, and settling time for both the pitch and yaw subsystems were similar with
and without the pulse disturbance.

In conclusion, the Twin Rotor MIMO System was an interesting system to experiment
with. It is highly reliant on both PID control and feedback gains, and is also robust to pulse
disturbances.

16
References

[1] Chalupa, Petr, et al. “Modelling of Twin Rotor MIMO System.” Procedia Engineering, vol.

100, 2015, pp. 249–58. DOI.org (Crossref), doi:10.1016/j.proeng.2015.01.365.

[2] Chaudhary, S., & Kumar, A. (2019). Control of Twin Rotor MIMO System Using

1-Degree-of-Freedom PID, 2-Degree-of-Freedom PID and Fractional order PID Controller. 2019

3rd International Conference on Electronics, Communication and Aerospace Technology

(ICECA). https://doi.org/10.1109/iceca.2019.8821923

[3] Cajo, R., & Agila, W. (2015). Evaluation of Algorithms for Linear and Nonlinear PID

Control for Twin Rotor MIMO System. 2015 Asia-Pacific Conference on Computer Aided

System Engineering. doi:10.1109/apcase.2015.45

17