Lecture 1: Preliminary Comments

A control system is a device or process whose behavior is inuenced by external stimuli (the control inputs) according to some set of rules (the system dynamics). Finding a set of mathematical formulas which accurately describe a systems behavior is a central theme in science and engineering referred to as mathematical modeling. A key step in formulating a mathematical model for a system is dening the systems state. The state of a system is a minimal set of variables which completely characterizes a phenomenon of interest. In general, there are many choices of variables which could dene the state; having made such a choice, these variables are then referred to as state variables. The denition of a systems state depends on the problem being solved. For example, consider the problem of controlling a spacecraft. If the goal is to force the spacecraft to assume a particular attitude, then the state variables must obviously include a set of variables which dene attitude, e.g., Euler angles or unit quaternions. But a spacecrafts attitude varies under the inuence of external moments (including the control moments that one would use to inuence the attitude). This variation is captured by a set of dynamic equations that relate the attitude rate to these external moments. Thus, the state variables must also include a set of variables that dene the spacecrafts angular rate. Assuming that the external moments do not depend on other factors, such as the spacecrafts position relative to the earth, then these state variables may be sucient to dene the system state for the attitude control problem. On the other hand, suppose there is gravitational moment that depends on the spacecrafts position relative to the earth. Then position and velocity must also be included among the state variables. The term control system implies that control inputs can be prescribed in order to make the system behave in some desired way. The control design problem then is to dene a set of rules for these inputs so that the complete collection of rules governing the controlled system yield the desired behavior. For a given system, there is typically a special subset of state variables (or functions of the state variables) called outputs which are of particular interest. For an aircraft, for example, one may be particularly interested in speed, heading, and altitude. These variables are only a small subset of the complete aircraft state vector, but it is particularly important to control them well. In summary, a control system is dened in terms of its state, the inputs, the outputs, and the behavioral rules that relate all of these elements. Example #1: A DC Servomotor State: Shaft angle, shaft angular velocity, and eld current. Input: Terminal voltage. Behavioral rules: A dierential equation describing the rotation of the shaft plus another dierential equation describing the eld current in the motor windings. Output: Shaft angle. Example #2: An aircraft State: Position (X), attitude (), velocity (v), and angular rate (). u x and v = v , = , X = y , w z

p = q . r

Inputs: elevator (e), aileron (a), and rudder (r) angles, thrust (T ) Behavioral rules: Six second order dierential equations derived from Newtons second law. These may be restated as six rst order kinematic equations relating position and attitude to body velocities plus six rst order dynamic equations (the force and moment equations). 1

Output: Speed (V ), heading (), and altitude (h) Example #3: A spacecraft with a reaction wheel State: attitude (), angular rate (), and reaction wheel relative angular rate () Inputs: The torque u applied to the reaction wheel. Behavioral rules: Three rst order attitude kinematic equations plus three rst order attitude dynamic equations plus a rst order equation for the reaction wheel rate. (Note that we typically dont care about the rotor angle.) Output: Attitude ().
M k1 /////////// x m k2
zR

z1 b1

z2 b2

Figure 1: An active suspension system. Example #4: Active Suspension System Consider a simplied model for an active automobile suspension system. (This example appears in [1].) The system is described by the following parameters and state variables.

M : the mass of the suspended portion of vehicle (engine, chassis, etc.) m : the mass of the unsuspended portion of the vehicle (tires, axles, etc.) z1 (1 ) : the location (nominal location) of M with respect to the nominal road height z z2 (2 ) : the location (nominal location) of m with respect to the nominal road height z k1 : spring stiness of the passive suspension system b1 : damping coecient for the passive suspension system k2 : spring stiness of the tires b2 : damping coecient of the tires u : force exerted on the suspended body by the unsuspended body (positive upward) zR : roadway height with respect to some nominal height The roadway height is actually a function of the longitudinal position of the vehicle, say zR (x) where x is the longitudinal position coordinate. Suppose that the vehicle translates in the positive x direction at a constant speed V so that x = V t + x0 . Thus, we may write zR (x) = zR (V t + x0 ) =: zR (t) and ignore the (trivial) horizontal dynamics. Considering only the vertical dynamics, we break the system apart and draw free body diagrams for each element; see Figure 2. Using the constitutive relations for the springs and damper and applying Newtons

Mg

Fk 1

Fb 1

k1

b1

Fk 1

Fb 1 mg

Fk2

Fb2

k2

b2

Fk2

Fb2

Figure 2: Free body diagram of the active suspension system. second law to each mass, we obtain M z1 = Fk1 + Fb1 M g + u = k1 ((z1 z1 ) (z2 z2 )) b1 (z1 z2 ) M g + u m2 = Fk2 Fk1 + Fb2 Fb1 mg u z = k2 ((z2 z2 ) zR ) + k1 ((z1 z1 ) (z2 z2 )) b2 (z2 zR ) + b1 (z1 z2 ) mg u. Now, z1 and z2 , are the nominal positions of the two masses with respect to the nominal road height. Thus, by denition, 0 = k1 (1 z2 ) M g z 0 = k2 z2 k1 (1 z2 ) mg. z In words, the suspension system perfectly balances the weight of the various components when the system is in its nominal conguration. The equations of motion are M z1 = k1 (z1 z2 ) b1 (z1 z2 ) + u m2 = k2 (z2 zR ) + k1 (z1 z2 ) b2 (z2 zR ) + b1 (z1 z2 ) u. z Note that zR = the road.
dzR dx V

; we only require zR as a function of x in order to compute disturbance forces due to

In this course, we will study systems modeled by linear time-invariant, ordinary dierential equations. Although most engineering systems are nonlinear, we will learn how to develop linear approximations to nonlinear dierential equations. For the active suspension system problem, as it turns out, the system is already linear time-invariant. There are two general approaches to control of linear, time-invariant control systems: 3

 the frequency domain (classical control) approach, and the time-domain (or state-space control) approach. In the frequency domain approach, we would take the Laplace transform of all of the signals and solve for the transfer functions of interest. For example, if we were interested in the passive (zero input) vehicle response to variations in road height, we would solve for the transfer function from ZR (s) to X1 (s). In the state space approach, we would start by transforming the two second order ODEs into a set of four rst order ODEs to obtain a control system in the form x = Ax + Bu To do so, dene the additional state variables v1 = z1 and v2 = z2 and replace zi with vi and zi with vi . Doing so, one obtains z1 = v1 z2 = v2 1 v1 = (k1 (z1 z2 ) b1 (v1 v2 ) + u) M 1 v2 = (k2 (z2 zR ) + k1 (z1 z2 ) b2 (v2 zR ) + b1 (v1 v2 ) u) . m These equations can be rewritten in the following matrix form 0 0 1 0 z1 z1 0 z2 0 z2 0 0 0 1 b1 k1 b1 v1 = k1 v1 + 1 M M M M M 1 k1 +k b1 +b v2 v2 m k1m 2 b1m 2 m m

0 0 u+ 0

k2 m

b2 m

0 0 0

zR zR

Classical approach. In the frequency domain approach, one takes the Laplace transform of the governing LTI system equations and designs a compensator to obtain a desirable frequency response, typically a low-pass lter response similar to that of a mass-spring-damper. We will learn all about the classical, frequency-domain approach to compensator design in the rst part of this course. At the outset, we note some of the advantages and disadvantages to this approach. Advantages. It is relatively easy to design a compensator to meet a specic set of performance specications (e.g., desired settling time and percent overshoot for a stable, underdamped response). Classical techniques include some well-developed tools for studying robustness to disturbances, sensor noise, and mathematical model uncertainty. The language of classical control is widely used and understood. Disadvantages. Classical control focuses on single-input, single-output (SISO) systems. Many of the control systems of interest to aerospace and ocean engineers have multiple inputs and multiple outputs. Classical techniques can be extended to MIMO systems, but the process is tedious and often ad hoc. The frequency domain approach is only appropriate for time-invariant systems. Time-varying systems can arise in practice and must be dealt with using the time-domain state-space approach. 4

State space approach. In the time domain approach, one writes the LTI equations in the standard rst order form x = Ax + Bu y = Cx + Du and designs a compensator using one of a number of techniques for state-space systems. Here x represents the system state, u is the input, and y is the output. The square matrix A is the state matrix, B is the input matrix, C and D dene the output in terms of the state and input. (The system above is a SISO system because u and y are scalars. A MIMO system can easily be treated by simply allowing a vector of inputs u and a vector of outputs y.) There are advantages and disadvantages to the state-space approach, as well. Advantages. One may easily treat MIMO systems in the state-space setting. There are straight-forward extensions to deal with time-varying systems. Optimal control problems, in which one chooses a control to minimize some cost (or maximize some value) function, are formulated and solved in the state-space setting. Important system properties, such as controllability and observability, are dened in the statespace setting. The state-space setting is becoming the new standard setting for control. Disadvantages. It is a bit more dicult to describe how performance specications should inuence the compensator design than it is in the classical setting. Feedback control in the state-space setting typically requires knowledge of the full system state, including components which may not be directly measured. The latter problem is often addressed by designing a companion system which estimates the true state given sensor measurements and the input history. This state estimator adds complexity to the control design, but can improve performance. State estimators are often used in guidance and navigation systems, for example, to determine position and attitude from inertial and other sensors.

References
[1] P. R. Blanger. Control Engineering: A Modern Approach. Saunders College Publishing, Philadelphia, e PA, 1995.