You are on page 1of 14

 

 
 
 
 
 
Lab 12: Maglev Control 
12/8/2010 
Frank Laritz II 
Maxim Markov 
Alex Prog 
Part A: Pre-lab

Linearized Maglev hardware transfer function:

3734
 
2180

Specifications:
1. ess w.r.t. step = 0
2. ts ≤ 3 seconds
3. % overshoot < 10%
4. Maximum gain = 4.5

Design Methodology:
1. Because the characteristic equation is missing a damping term, a differentiator is needed
2. In order to meet the ess w.r.t. step = 0 spec, an integrator is needed
3. In order to fix the free term -3734 a gain is needed

Therefore, a PID controller of the form KD*s + KP + KI/s was selected.

   
Alternatively, the controller can be expressed as C(s) =

Design Steps:
1. Determine the admissible domain

% overshoot = 100 * < 10%


> 0.591

.
ts = < 3 seconds

α= < 0.9375
2. Root Locus of P(s) with a numerator = 1 instead of -3734

Based on the root locus, a pair of complex conjugate zeros were chosen in the admissible
domain.

S1 = ‐1.75 + j0.5  
S2 = ‐1.75 – j0.5 

Those poles yield the quadratic equation: s2 + 3.5s + 3.313

This equation can be equated to the controller C(s) by factoring out the KD coefficient.

   
C(s) =

s2 + 3.5s + 3.313 =    
Relating the coefficients:

= 3.5

= 3.313

3. In order to determine the coefficient KD, a root locus of P(s) * C(s) was performed with
the numerator of P(s) set equal to 1. Therefore, gain in the root locus is equal to
KD * -3734

KD =   = -0.87

Therefore:

.   .   .
C(s) =
Relating the coefficients:

KP = -3.05
KD = -0.87
KI = -2.88

These gains provide a = 0.778 and an wn = 1.81 which within the admissible domain.
The selected gains are expected to meet all design specifications and reject input
disturbances.

The final PID controller is described by the equation:

C(s) = -3.05 - 0.87s - 2.88/s

4. Precompensator

The system will not meet the specifications without a precompensator because the
additional zeros introduced by the controller will have adverse affects on the system (i.e.
an increase in percent overshoot). Most PID controllers assume that the zero and the pole
introduced by the PI part of the controller cancel each other out because they are typically
very close. However, this was not the case for the selected controller. Since the
controller introduces two zeroes that are complex conjugates, it is necessary to cancel
both of them.

In the general form:

Pre_C(s) =
   

Substituting in the values for KD, KP, and KI:


.
Pre_C(s) =
  .   .
Step Response of the closed loop system with a precompensator:

The step response has the following characteristics:

1. ess w.r.t. step = 0


2. ts = 1.79 seconds
3. % overshoot = 2%

Therefore, the controller was a great success and the system meets all of the design
specifications.
Part B: In-lab Data & Post-lab Analysis

Controller Performance Assessment:

Asymptotic Stability:

The controller successfully stabilized the system. Additionally, the system noise was minimized
by selecting the KD coefficient to be less than 1. Also, the controller was designed to take into
account the nonlinear restriction on the input to the plant (i.e. controller gains could not exceed
4.5). Lastly, no controller parameters needed to be tuned for the system to be asymptotically
stable.
Step response with an amplitude of 0.4 cm:

ess w.r.t. step ≈ 0


ts = 2.47 seconds
% overshoot ≈ 0%

The controller met all specifications without additional tuning.


Square wave response with amplitude of 0.4 cm and a period of 8 seconds:

ess w.r.t. step ≈ 0


ts = 2.67 seconds
% overshoot ≈ 0%

The controller met all specifications without additional tuning.


Square wave response with amplitude of 0.45 cm and a period of 8 seconds:

ess w.r.t. step ≈ 0


ts = 2.61 seconds
% overshoot ≈ 0%

The controller met all specifications without additional tuning.


Square wave response with amplitude of 0.5 cm and a period of 8 seconds:

ess w.r.t. step ≈ 0


ts = 2.54 seconds
% overshoot ≈ 0%

The controller met all specifications without additional tuning.


Square wave response with amplitude of 0.6 cm and a period of 8 seconds:

ess w.r.t. step ≈ 0


ts = 2.39 seconds
% overshoot ≈ 0%

The controller met all specifications without additional tuning.


Impulsive disturbance response:

In order to reject impulsive disturbances without introducing much noise, the controller gain KD
was decreased from -0.87 to -0.65. However, the method of disturbance for the previous gain of
KD = -0.87, which was a horizontal disturbance, was found to be an incorrect form of
disturbance. Therefore, it is expected that the original KD gain would have successfully rejected
impulsive disturbances without introducing a large amount of noise, if a vertical disturbance had
been introduced instead. The new controller was able to stabilize the system when an impulsive
disturbance with magnitude 1 cm was introduced. The maglev could be operated on a ‘shaky’
base if the magnitude of the disturbances was equal to or less than 1 cm.

Applications:

If a controller can be designed to levitate a metal ball, it would not be too difficult to design a
controller that levitates any rotor. Magnetic bearings are already common in the market, and,
due to Earnshaw’s theorem, are inherently unstable without an active controller. Magnetic
bearings have many advantages over traditional bearings, most importantly reduced friction. In
Meco’s Turbomolecular Pump1, this allows for a lubricant-free design, eliminating the need to
contaminate the body with oil, which could have disastrous biochemical effects.

                                                            
1
 http://www.mecos.com/en/applications/turbomolecular_pump.php 
In the case of a rotor, the problem no longer is one-dimensional. Not only is it necessary to keep
the rotor near the desired y position, but there is also a desired x position. This would be a
simple extension of the problem addressed in lab. The simplest solution would be to use a second
controller and electromagnet for the position on the x axis.
However, the technique used in this lab required the linearization of the plant around an
equilibrium state. While this was acceptable for the lab when we placed the ball near the
equilibrium, it would not work for larger disturbances. Since the object would be spinning at a
variable rate, it would be possible to create large disturbances from torque applied to the object.
If the disturbances are significant, the linearized model would fail. This was noticed this in lab
when too large of a disturbance was applied, causing the ball to fly up to the electromagnetic or
drop down to the ground. Here, the linearized model was no longer valid, and a more advanced
controller would be needed to compensate for large disturbances.

You might also like