Sunteți pe pagina 1din 4

Sampled-data Control of Double Integrator Systems

Richard T. OBrien, Jr., Eric P. Boemke, Lora M. Gorsky Systems Engineering Department, The United States Naval Academy, Annapolis, MD 21402, USA riobrien@nadn.navv.mil

Abstract-

A sampled-data control system with bounded controls is introduced for double integrator systems. The control system uses samples of the output of a continuous-time, linear controller to determine the amplitude of a predetermined, piecewise constant (banghang), control signal. The predetermined control signal defines the inter-sample behavior of the system and generalizes the constant inter-sample behavior associated with a zero-order hold. Simulation results of the nominal and robust performance are presented.
I. INTRODUCTION
The double integrator system is one of the most fundamental models used in control theory. The double integrator model represents objects moving in low-friction environments or moving with single degree-of-freedom translation and rotation. Specifically, double integrator control systems have applications in spacecraft attitude control and ground vehicle steering control. While many controllers exist for the double integrator system, most require measurements of position and velocity. Furthermore, the position-only controllers do not provide acceptable performance or robustness. In this paper, a bang-hang control algorithm is developed for a double integrator system using position feedback only. This work builds on the preliminq results Previous work on controllers for double in [I]. integrator plants has been summarized in the survey paper by Rao and Bemstein 121. Among these papers, several explicitly address the issues of bounded controls [3,4] and position-only feedback [5]. Furthermore, in 121, the authors consider a variety of approaches to double integrator and establish benchmarks for the evaluation of potential controllers. In this paper, a sampled-data control system with bounded controls is introduced for double integrator systems. The control system uses samples of the output of a continnous-time, linear controller to determine the amplitude of a predetermined, piecewise constant (bang-hang), control signal. The
0-7803-7697-8/03/$17.00 0 2003 IEEE

pre-determined control signal defines the intersample behavior of the system and generalizes the constant inter-sample behavior associated with a zero-order hold. This control scheme is an example of the discretelcontinuous control in [61. The structure of the control signal converts the continuous-time, double integrator control problem into a discrete-time, single integrator control problem. Furthermore, implementation of the algorithm is straightforward and does not require a fast sampling rate. The paper is organized as follows. Section I1 contains an overview of the bang-bang control strategy. Section III contains the design guidelines and examples of nominal performance. Section IV contains an investigation of the robustness of the controller using the benchmarks from 121. Section V contains conclusions and directions for future work. 11. BANG-BANG CONTROL STRATEGY

Figure 1: Closed-loop block diagram of the bang-bang control system

Figure 2 Signals in bang-bang control system

The bang-bang control strategy incorporates minimum-time control concepts in a closed-loop system. Figure 1 shows the closed-loop system block diagram where the control signal generator (CSG)

413

produces a piecewise constant signal proportional to the sampled error. One period of the control signal is shown in Figure 2A. (If the output of the double integrator is position, the input is acceleration.) The resulting velocity and position responses over one period are shown in Figures 2B and 2C. It should be noted that the signals in Figure 2A, B, and C are the same as the responses for open-loop, minimum-time control of a double integrator. Figure 2D shows the closed-loop response using the bang-bang control system. Note that it resembles a first order response. The proposed hang-bang control system offers several performance and implementation features. First, a finite settling time can be achieved as discussed in Section 111. Second, the control system does not require a fast sampling rate because the inter-sample behavior is specified. Third, the implementation of the signal generator is straightfonvard because the signals are piecewise constant and only the amplitude is varied. Fourth, the implementation is modular because the control signal generator and the linear controller are adjusted separately. Finally, the control system demonstrates robustness to unmodeled dynamics as discussed in Section IV. From the survey of controllers for double integrators in [2], there were only two controllers than used position feedback only: a LQG controller with output feedback and the trap door controller from [51. Furthermore, both of these controllers performed poorly in off-nominal situations [2]. As discussed in Section IV, the bang-bang controller is fairly robust and, therefore, has potential to be a successful controller for double integrator plants.
1 1 DESIGN GUIDELINES 1.

where p is the output (position) of the system and U is the input generated by the proportional gain.

B. Closed-loop solution
After some manipulation, the closed-loop system has a closed-form solution for a unit step command p(nT) =l-(l--d),p(O) = O The quantity (2)

(3)
is the design parameter.
C. Design parameters

Using the closed-form solution in (2), it follows that the settling time of the closed-loop step response decreases as a increases from 0 to 1. Specifically, if a=l , the closed-loop response reaches the desired value in one pulse cycle and, as a result, a finite settling time can be achieved. In addition to the design parameter a , a second criterion is required to specific both K and T . Given a bound on the control signal, the gain m must be bounded by K S where e,, is the e,, maximum tracking enor. It is reasonable to assume that the maximum tracking error tracking occurs initially and, therefore, the gain is bounded by

m,

K Ir

There are two design parameters in the bang-bang control scheme: the proportional gain, K , and the pulse period, T . A. Open-loop model

(4)

To develop a design procedure, the effect of the control signal in Figure 1 on the double integrator system must be examined. Since the inter-sample behavior is specified, it is easier to represent the controlled double integrator system as a discrete-time system with a period equal to T . The difference equation for the open-loop controlled system is

where e- = r - O = r . The design parameters are chosen using the expressions in (3) and (4). Given and I , the gain can chosen using (4) and, specifically, it is reasonable to choose

to maximize the control input usage. Given K , the cycle can be computed from (3) as

414

To achieve a finite settling time, set a=l

D.Nominal pe?formance

0 '

DI

02

03

01

05

06

07

OB

09

T,me @er,

Figure 5: Closed-lwp responses for a=1 and various values o f K

Figure 3: SIMULINK implementationof the ocntorl signal

generator

Figure 4 shows the nominal response for a=l and various gains K=10,8,6,4. Since a = l , each response has a finite settling time but the settling time is inversely proportional to the gain. To completely test the nominal performance, the response is simulated for various initial conditions on position and velocity. While the bangbang controller can handle non-zero initial positions, the controller can handle only small initial velocities. This fault is due to the design of the control signal in Figure 1. Note that the velocity in Figure 2B has the same value at each sample. Therefore, the system cannot react to non-zero velocity. Currently, the control algorithm is being modified to include a second control pulse to account for initial velocity.

N. ROBUST PERFORMANCE
The benchmark experiments in [2] test the double integrator controllers under a number of scenarios including mass variation, imaginary or real poles instead of two poles at the origin, and measurement delay. For these tests, the nominal design was used and the robustness of the controller was evaluated. The bang-bang controller demonstrated robustness to mass variation from 80% reduction to 100% increase. Specifically, as mass is increased, the settling time increased. However, a problem did arise when the mass was very small because the system was being under-sampled. This situation could be resolved by lowering the value of the design parameter a,which in tum would lower the value of pulse (sampling) period, T. To simulate computation errors in the controller implementation, the poles of the double integrator were moved from the origin to the imaginary axis or to the real axis (as repeated roots). Again, the bangbang controller demonstrated robustness. When the

00

05

I5

rme

25

35

45

Figure 4 Closed-loop respanses for K = 10 and various values of

To investigate the role of the design parameters, simulations of the closed-loop system in Figure I are run for the scenario where m = l and r=0.1. The SIMULINK implementation of the control signal generator is shown in Figure 3. Using (5). the gain is computed to be K=10. Figure 3 shows the closed-loop response for four values of a =1,0.8,0.6,0,4. As expected, a finite settling time is achieved for a=l and the settling time increases as a decreases. Also, for a < l ,the change in position for each pulse cycle (as in Figure 2C) is visible.

41 5

poles were moved out along the real axis, the system became noticeably slower, but the percent overshoot remained negligible as when the both poles of the double integrator were at zero. When the poles were moved out along the imaginaty axis, on the other hand, the system oscillated much more, especially as the distance from the origin was increased. The settling time also increased significantly. The bang-bang controller demonstrated robusmess to measurement delays of 0.1 to 0.8 seconds. The results confirmed the findings in [2] that when a small delay is added, e.g. 0.1, the settling time and percent overshoot are actually better than when there is no delay at all. When the delay on this project was increased to 0.8, the settling time became very large, and the percent overshoot jumped to almost eighty percent. These findings concur with the results in [2] that when the delay was larger than 0.2, the transient specifications rapidly grow. These results provide possible avenues for refinement in the design guidelines. Furthermore, the relative success of the bang-bang controller in these off-nominal scenarios distinguishes it from the other position-only controllers in [2]. V. CONCLUSIONS A sampled-data control system with bounded controls was introduced for double integrator systems. The control system uses samples of the output of a continuous-time, linear controller to determine the amplitude of a pre-determined, piecewise constant (bang-bang), control signal. The pre-determined control signal defines the intersample behavior of the system and generalizes the constant inter-sample behavior associated with a zero-order hold. Simulation results of the nominal performance demonstrated the roles of the two design parameters and the controllers ability to achieve finte settling time. The simulation results for offnominal conditions demonstrated the robustness of the controller and the potential avenues for refinement of the design guidelines. In the future, the sensitivity to the initial velocity will be address. Furthermore, hardware tests will be performed to validate the discrete-time modeling results and the closed-loop performance. Finally, the controller will be implemented as part of an autonomous vehicle, steering control system. VI. REFERENCES

[2] V.G. Rao and D.S. Bemstein Naive Control of the Double Integrator, Control Systems Magazine, 21(5), October 2001, pp. 86-97. [3] A.R. Tee1 Global Stabilization and Restricted Tracking for Multiple Integrators with Bounded Controls, Systems C? Conrrols Leners, 18, 1992, pp. 165-171. [4] S.P. Bhat and D.S. Bemstein Continuous finitetime Stabilization of the Translational and Rotational Double Integrators, Transacrions on Automatic Conrrol, 43(5), May 1998, pp. 678-682. [SI R.T. Bupp, D.S. Bemstein, V.S. Chellaboina, and W.M. Haddad Finite Settling Time Control of the Double Integrator using a Virtual Trap Door Absorber, Transactions on Automatic Control, 45, Apr 2000, pp. 776-780. [6] C.D. Johnson Improved Discrete-Time Controller Performance using the ControlDimension Multiplier Effect of DiscreteIContinuous Control Theory , Proceedings of the Southeastem Symposium on Systems Theory, Huntsville, AL, March 2002,484490.

[ I ] M.G Vasek. and R.T. OBrien, Jr. Bang-Bang Control of Double Integrator Systems, Proceedings of the Southeastem Symposium on Systems Theory, Huntsville, AL, March 2002,275-278.

416

S-ar putea să vă placă și