NONLINEAR CONTROLLER COMPARISON ON A BENCHMARK ...

NONLINEAR CONTROLLER COMPARISON ON A BENCHMARK ... NONLINEAR CONTROLLER COMPARISON ON A BENCHMARK ...

24.08.2013 Views

a robust design strategy would, in practice, stabilize a physical plant that di ers substantially from its mathematical model. Thus the crudeness of our model only helps to demonstrate the robustness of the control laws that succeed in yielding adequate performance on the actual hardware system. Another advantage of this mechanical model is that it is identical, in form, to the system equations presented in [2] as a nonlinear benchmark problem. Since the object of this research is to compare nonlinear robust control strategies, this model is ideal in the sense that it is already widely used as a standard mathematical system for testing and comparing nonlinear control techniques. The mathematical model in (2.4) is slightly di erent than the standard model of the NLBP in that our input is voltage, not torque. This adds one extra term and a few constants in order to incorporate the voltage to torque transfer function into the model. Many of the papers involving the NLBP express the system equation in a non-dimensionalized form. Since some of the control strategies in this comparison make use of these dimensionless equations, and because they simplify the model by combining all of the physical parameters into a single value, we include them here: 0 1 _z B C B C BzC B C B _ C @ A = Where _ represents d z = y = t = d q M +m I+mc q k M +m 0 B @ v(x) =u(x) m+M k(I+mc2 ) mc = p (I+mc2 )(m+M ) _z ;z+ _2 sin( ); 2 cos( )_z K2 m K2 g Rmmc 1; 2 cos( ) 2 _ cos( )[z; 2 sin( )]+ K2 m K2 g Rmmc 1; 2 cos( ) 2 1 C A + 0 B @ 0 KmKg Rm cos( ) 1; 2 cos( ) 2 0 KmKg Rm 1; 2 cos( ) 2 and the following transformations are used: 12 1 C v(x): (2.5) C A

2.3 Software Set-up All of the control strategies presented in this paper were simulated in MAT- LAB's SIMULINK environment. They were then implemented by way of Quanser Consulting's Multiq3 I/O board which interfaced with the SIMULINK Real Time Workshop. Figure 2.5 shows the SIMULINK model used in the simulations and in the actual exible beam system. Their are two main blocks, one is the control block that takes the four states as inputs, as well as an initial open loop disturbance. The control has only one output, a voltage signal that is fed into the system block. The output of the strain gauge and the motor encoder were run through the Multiq3 board into SIMULINK. These provided a direct state measurement of the angle, , and the strain was a good estimator for the linear displacement, y. The state ve- locities in the system equations were numerically computed using lters of the form s s+ . Although this theoretically makes for a relatively crude observer, in practice these pseudo derivatives actually performed very well. Even though the assumption of full-state feedback is not strictly true, in practice a more complicated observer is unnecessary to achieve good performance by the control laws. The sample rate for in the SIMULINK model for both the simulation and the FBS was 1 kHz. All of the physical parameters for the Quanser system are noted here: k = 30N=m, m = :05Kg, M = :6Kg, c = :285m, I = :0030Kgm 2 , Km = :001Nm=A, Kg = 70, Rm =2:6 Ohms. 13

2.3 Software Set-up<br />

All of the control strategies presented in this paper were simulated in MAT-<br />

LAB's SIMULINK environment. They were then implemented by way of Quanser<br />

Consulting's Multiq3 I/O board which interfaced with the SIMULINK Real Time<br />

Workshop. Figure 2.5 shows the SIMULINK model used in the simulations and in<br />

the actual exible beam system. Their are two main blocks, one is the control block<br />

that takes the four states as inputs, as well as an initial open loop disturbance. The<br />

control has only one output, a voltage signal that is fed into the system block. The<br />

output of the strain gauge and the motor encoder were run through the Multiq3<br />

board into SIMULINK. These provided a direct state measurement of the angle, ,<br />

and the strain was a good estimator for the linear displacement, y. The state ve-<br />

locities in the system equations were numerically computed using lters of the form<br />

s<br />

s+<br />

. Although this theoretically makes for a relatively crude observer, in practice<br />

these pseudo derivatives actually performed very well. Even though the assumption<br />

of full-state feedback is not strictly true, in practice a more complicated observer<br />

is unnecessary to achieve good performance by the control laws. The sample rate<br />

for in the SIMULINK model for both the simulation and the FBS was 1 kHz. All<br />

of the physical parameters for the Quanser system are noted here: k = 30N=m,<br />

m = :05Kg, M = :6Kg, c = :285m, I = :0030Kgm 2 , Km = :001Nm=A, Kg = 70,<br />

Rm =2:6 Ohms.<br />

13

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!