13.1 Feedback & Block Diagrams ended with a cruise control whose single pole slid left as the gain rose: at , at , at , judged by eye. This lesson puts numbers on what the eye sees. A transfer function packs the linear dynamics of a system into a ratio of two polynomials, and the roots, drawn in the complex plane, say how fast the system moves, whether it rings and how long it takes to settle. The same dictionary serves a spring, a car, a wind-tunnel fan and the short-period mode of 12.5 Trim, Linearization & Longitudinal Modes. The punchline: the cruise loop of 13.1 stayed first order only because it ignored the engine’s lag; include the lag and raising the gain starts to ring.
Transcript
- Take a mass of 2 kilograms on a spring with a damper, pushed by a force u. Newton gives 2 y double dot plus 3 y dot plus 8 y equals u. Laplace transform it with zero initial conditions and each derivative becomes a factor s, so the equation becomes algebra. The transfer function G of s is Y over U, a ratio of two polynomials, here 1 over 2 s squared plus 3 s plus 8. Its poles, the roots of the denominator, are minus 0.75 plus or minus 1.8540 i. There is no zero, and the DC gain is 0.125.
- The poles decide the whole shape. For the mass, omega n is 2 and zeta is 0.375. The real part of the pole, sigma equals zeta omega n, is the decay rate of the envelope, and the imaginary part, omega d, is the ringing frequency. The step response is the final value minus a decaying sinusoid, trapped between two dashed envelopes. Poles on one ray from the origin share one damping ratio. Here the peak is 28.06 percent over the final value, at 1.69 seconds.
- First order is one pole. G of s is k over tau s plus 1, and the unit step response is k times 1 minus e to the minus t over tau. Two timing numbers are exact. The 10 to 90 percent rise time is tau ln 9, and the 2 percent settling time is tau ln 50, the textbook 4 tau rounded. The lab measures them on the cruise plant. Open loop, tau is 10 seconds: rise 21.97, settling 39.12. Closed loop at K equals 4, tau is 2 seconds: rise 4.394, settling 7.824. Moving the pole left shrinks every time by the same factor.
- The peak is where the slope is zero. That first happens at omega d t equals pi, so the peak time is pi over omega d, and the fractional overshoot is the exponential of minus pi zeta over the square root of 1 minus zeta squared. Both are exact for the pure second-order shape, and the overshoot depends on zeta alone. At omega n equals 1 the lab measures 52.66, 25.38, 9.48 and 1.52 percent for zeta 0.2, 0.4, 0.6 and 0.8, the formula's values to the last digit shown. Settling depends on the real part alone: the envelope rule is about 4 over zeta omega n.
- A zero changes the numerator, not the poles. Written as 1 minus s over z, it adds a multiple of the slope: the new response is the old one minus its slope divided by z. On the short-period pair, with omega n 4.5567 and zeta 0.5848, the plain response overshoots by 10.39 percent and peaks at 0.85 seconds. A zero at minus 2 adds the slope: faster, but 68.80 percent overshoot. A zero at plus 2, in the right half plane, subtracts it, and the output first dips to minus 83.58 percent at 0.18 seconds before rising. That is called non-minimum-phase.
- Real systems have many poles, and the response is read from the ones nearest the imaginary axis. Take omega n equal 1 and zeta 0.5. Alone, the pair gives 16.30 percent overshoot and a settling time of 8.076 seconds. Add a third pole at minus 10, twenty times farther than the real part of the pair, and the overshoot is 16.22 percent with settling 8.173: almost the same, so the pair is dominant. Move the extra pole to minus 1, as slow as the pair, and the response changes: 8.15 percent and 6.637 seconds.
- Now the cruise loop of lesson 13.1 with a real engine. The throttle takes about 1 second to deliver the force it is told to, so the actuator 1 over s plus 1 sits in series with the plant. Close the loop and T of s is K over 10 s squared plus 11 s plus 1 plus K. At K equals 4 the poles are minus 0.55 plus or minus 0.4444 i, zeta is 0.7778, overshoot 2.05 percent, peak at 7.07 seconds. At K equals 19 the real part is still minus 0.55, but the imaginary part has grown to 1.3029. Zeta falls to 0.3889, overshoot rises to 26.55 percent and the peak comes at 2.41 seconds. With the first-order model the loop could never ring. With a realistic actuator, raising the gain buys speed at the price of damping.
- So what did the poles tell us? The real part sets the decay, the imaginary part the ringing, and overshoot and peak time follow exactly from zeta. A zero in the left half plane speeds the response and raises the overshoot, and a zero in the right half plane makes it start the wrong way. With the engine lag, going from K equals 4 to 19 takes the overshoot from 2.05 percent to 26.55. And the steady-state error of 1 over 1 plus K is still there. PID control, lesson 13.3, is not yet written. An integral term gives a type-1 loop with zero step error, and a derivative term puts damping back.
From the ODE to the transfer function
Take a linear ODE with constant coefficients and input and Laplace-transform it (6.2 Solving Initial Value Problems) with zero initial conditions: each derivative becomes a factor , and the equation becomes algebra. The transfer function is the ratio of transforms,
the Laplace transform of the impulse response of 6.5 Convolution & Transfer Functions. Zero initial conditions mean the system starts at rest.
The roots of are the poles: the eigenvalues of the system matrix and the exponents in every free response. The roots of are the zeros. Poles decide which modes exist; zeros decide how much of each shows up in the output.
A worked example, used throughout: a mass kg on a spring N/m with damper N s/m, pushed by a force (in newtons), output the displacement in metres (assumed round numbers). Newton gives
The poles are , the ones from 5.3 Mechanical Vibrations: Spring–Mass–Damper. There is no zero. The DC gain m/N is the static deflection per newton: it is the value the displacement settles to for a constant unit force.
Block algebra, again
The rules of 13.1 carry over unchanged. Blocks in series multiply, . Around a negative-feedback loop with forward path and feedback path ,
and with this is the of 13.1. The closed-loop poles are the roots of , and everything below applies to exactly as to .
First order: one pole
The standard first-order form is
The pole is at and the DC gain is . Two timing numbers are exact. The 10 to 90% rise time solves and , so . The 2% settling time is the last time the response is outside a band: gives . The textbook shortcut (from memory) is this rounded.
The lab (below) measures these on the 13.1 cruise plant. Open loop, , s: rise s and settling s, against s and s. Closed loop at , where 13.1 found s and a final value of , so and s: rise s and settling s. Moving the pole left shrinks every time by the same factor.
Second order: poles, wn and zeta
Anything with inertia and a restoring force is second order, with the standard form
with natural frequency (rad/s) and damping ratio . For the mass above, , so ; , so ; and .
The poles are the roots of . For they are complex,
The real part is the decay rate of the envelope ; the imaginary part is the ringing frequency; the distance from the origin is ; and the angle from the negative real axis satisfies , so poles on one ray share one damping ratio. At the pair collides on the real axis; for it splits into two real poles and the response no longer rings.
The unit step response follows by partial fractions on :
The response is the final value minus a decaying sinusoid, trapped between the envelopes (the dashed curves in the lab).
Reading the specifications from the poles
Overshoot. The peak is where . Differentiating gives , which vanishes first at . So the peak time is , and substituting back, . The fractional overshoot is
Both are exact for the pure second-order shape, and depends on alone. At the lab measures overshoots of , , and for , , , , the formula’s values to the last digit shown.
Settling time. The response stays inside the band once the envelope does: , so
The envelope form is an upper bound, since the oscillation is usually inside its envelope at the last band crossing; the right-hand is the textbook rule (from memory, rounded). Settling depends on alone: moving a pole straight left speeds it up; moving it up only raises the ringing.
Rise time has no clean closed form. A rule of thumb (from memory) is , best near .
The first four columns are measured from the simulated response; the last three are the formulas.
| example | (s) | (s) | envelope (s) | (s) | (s) | (s) | |
|---|---|---|---|---|---|---|---|
| mass-spring-damper | |||||||
| short-period mode (12.4) |
The mass has poles . The short-period row takes only the oscillatory mode of 12.4, poles (, ), with unit DC gain: the aircraft’s actual pitch response has extra zeros and a slow phugoid, and its s pitch peak was a different quantity.
The cruise loop with an engine lag
13.1 gave the car’s speed a first-order plant and fed back through an instant throttle. A real engine takes about a second to deliver the force it is told to (assumed s). The actuator block sits in series with the plant:
Matching to the standard form, , , and the DC gain is , the same at as before. Put in (at the poles are the real pair and ):
| poles | (s) | (s) | envelope (s) | ||||
|---|---|---|---|---|---|---|---|
The real part is whatever the gain (once the pair is complex): the gain only moves the poles up, which sweeps the ray toward the imaginary axis. falls from to and overshoot rises from to . The pair first leaves the real axis where the discriminant vanishes, (by hand). With the first-order model of 13.1 the loop could not ring at any gain; with a realistic actuator it does, the same warning as the delay of 13.1 now seen on the -plane.
DC gain, the final value theorem and system type
The unit step response of has . If every pole of has negative real part, the final value theorem (from memory, proved by letting in ) says
So the DC gain is the steady-state value: m for the mass, for the closed loop. The condition matters: a pure oscillator () has poles on the imaginary axis, never settles, and still has a .
Apply the theorem to the error in a loop with a unit-step reference:
This is the of 13.1, derived rather than measured. The number of free integrators in is its system type. A type-0 loop (such as ) has a finite and a nonzero step error. A type-1 loop has , so and the step error is : the loop can hold a nonzero output with zero error, the thing 13.1 found missing. It follows a ramp with finite error , . Integral action is how PID adds that integrator.
Adding a zero
A zero changes the numerator, not the poles, so the settling envelope stays but the shape changes. Write it as , a zero at that keeps the DC gain. Since means differentiation, the new response is the old one plus a multiple of its slope:
For negative (left half plane) the slope is added: quicker, more overshoot. For positive (right half plane) it is subtracted, and since starts out rising the output first goes the wrong way. On the short-period pair above the lab gives:
| zero | overshoot | peak time (s) | other |
|---|---|---|---|
| none | s | ||
| s | |||
| dips to of the final value at s first |
The nearby left-half-plane zero is faster but at more than six times the overshoot. A right-half-plane zero is called non-minimum-phase. Aircraft show it: pulling the elevator up first pushes the tail down, so the normal acceleration at the centre of gravity first moves the wrong way (from memory; 12.4 does not model it). Feedback cannot remove it, since a forward-path zero stays a zero of the closed loop.
Dominant poles
Real systems have many poles, and the response is read from those nearest the imaginary axis; the rest decay before the slow modes finish. Set and in the lab (); alone, the pair gives overshoot , peak s, 10 to 90% rise s, settling s. Add a third real pole at , twenty times farther than the pair’s real part (DC gain kept at ): overshoot , peak s, rise s, settling s: almost the same, so the pair is dominant. Move the extra pole to , as slow as the pair, and the response changes: overshoot , peak s, rise s, settling s. The rule of thumb (from memory) is that poles five to ten times farther left can be dropped.
The same logic holds when . At , the poles are and , and the measured settling is s, near the s of the slow pole alone. Critical damping is the fastest response with no overshoot: s at .
Worked examples
Worked example: the loop from the block diagram. Forward path , . Closed loop: . So , , . Poles: , . ; s; envelope settling s; rule of thumb s. Lab: , s, and a measured s. The envelope bound is above the measured settling: at small the last band crossing happens at a peak well inside the envelope.
Playground: the loop with an engine lag
The exercise
Turn a pole, a numerator and a loop into specifications.
Concept checks
Problems
A system obeys . Find , the poles, , , , and the predicted overshoot and peak time for a unit step.
Show worked solution
Zero initial conditions: , so . Then gives , gives , and . The poles are (). , i.e. of the final value , so the peak is . s, and s from the rule of thumb.
A forward path is closed with unit feedback. Find the closed-loop , , , the step overshoot, the system type, and the steady-state error for a unit step and a unit ramp.
Show worked solution
. So , , . , i.e. , with s. There is one integrator in , so the loop is type 1: the step error is . For a ramp, and the error is .
. Write the reduced model that keeps the pair and the DC gain, and compare its settling with the full response’s measured s.
Show worked solution
The third pole is twenty times farther left than the pair. Replace by its DC value : , DC gain . Then , and s against s: an error of about .
(a) For the loop with the engine lag, , use the final value theorem to find the step limit and steady-state error, and compare with 13.1. (b) Why does the theorem fail for ?
Show worked solution
(a) The poles are strictly in the left half plane, so and the error is : the same as 13.1; the actuator changes the transient, not the steady state. (b) has poles at on the imaginary axis: the response oscillates forever, and is not a limit.
What comes next
The cruise loop still has the error , and raising to cut it costs damping (overshoot at with the engine lag). PID control, lesson 13.3 (13.3 PID Control), is the standard remedy: an integral term gives a type-1 loop with zero step error, and a derivative term puts damping back. It will judge its tuning with the vocabulary of this lesson.
Key Takeaways
- A transfer function comes from the ODE by Laplace-transforming with zero initial conditions; poles are the roots of , zeros the roots of , and is the DC gain. Series blocks multiply; a negative loop is .
- First order : rise , settling ( is the rule of thumb): s and s open loop for the cruise plant, s and s at .
- Second order: poles ; the real part sets the decay, the imaginary part the ringing, the angle the overshoot.
- Exact for the pure shape: and . Rules of thumb (from memory): and , good to about .
- A left-half-plane zero speeds the response and raises overshoot ( to at ); a right-half-plane zero makes it start the wrong way ( dip at ). A far pole barely matters; a pole as slow as the dominant pair does.
- The final value theorem gives and ; a type-1 loop has zero step error.
- With a s engine lag the cruise loop has poles : overshoot grows from at to at . Next: PID (13.3) trades these off.