12.4 6-DOF Aircraft Equations of Motion ended with a pitch kick and two motions on two clocks: the nose bobbed and settled in about a second, then the aircraft spent half a minute trading height for speed. The simulator shows them but does not explain them. Linearised about trim, the same model becomes with four states, and the four eigenvalues of give both motions, periods and damping, before any time stepping. It is also how designers check a simulator and how control engineers start an autopilot.
Tools: 7.7 Nonlinear Systems & Linearization, 7.4 Complex & Repeated Eigenvalues, 1.7 Complex Eigenvalues & Rotation.
Transcript
- Last lesson ended with a pitch kick and two motions on two clocks. The nose bobbed and settled in about a second. Then the aircraft spent half a minute trading height for speed. Linearised about trim, the same model has four eigenvalues, and they give both motions before any time stepping.
- Everything starts at trim, an equilibrium where every derivative is zero except steady motion along the path. At 50 metres per second Newton's method gave alpha equal to theta, 2.6606 degrees, elevator minus 2.7452, thrust 931.2 newtons, weight 10 791. Expand about any other point and a constant leftover term makes the modes drift.
- Write the state as trim plus a small change and expand to first order; the Jacobian J does the work. Two facts cut thirteen states to four. At wings-level trim the lateral states appear only in products, so the lateral and longitudinal blocks decouple. And position drops out. That leaves delta u, delta w, delta q and delta theta.
- A simulator programmer does not differentiate forces by hand. Nudge each state up and down inside the same derivative function the simulator flies, and divide by 2 h. The steps are 10 to the minus 4 for u and w, and 10 to the minus 5 for q and theta, so the errors sit near 10 to the minus 8.
- Here is the matrix, with columns u, w, q and theta. The entry in row 2, column 3 is the forward speed 49.94610, the rotating frame term. Row 3, column 2 is minus 0.27371: the pitch stiffness, from C m alpha. Row 3, column 3 is minus 2.86316 per second, the pitch damping. Each entry is a physical sensitivity.
- Now the geometry. If A only stretches a direction v, a start along v stays on that line and decays as e to the lambda t. Any start is a mix of the four, each a mode. A complex pair makes one spiral: the real part is the decay rate, the imaginary part the turning rate. From them come the natural frequency, zeta, the period and the time to half.
- The four eigenvalues form two pairs. The short period, minus 2.6647 plus or minus 3.6964 i, has zeta 0.5848 and a period of 1.70 seconds: the angle of attack changes at nearly constant speed. The phugoid, minus 0.0126 plus or minus 0.2246 i, has zeta 0.0561 and a period of 27.98 seconds: speed and height trade at nearly constant angle of attack. They differ twentyfold, so the system is mildly stiff.
- Over a second the speed hardly changes, so drop u and theta. Omega n squared is Z w times M q minus u zero times M w, and twice zeta omega n is minus the sum of Z w and M q. By hand that gives 4.5533 and 0.5853, good to 0.1 percent. The spring is the weathercock action of C m alpha; the dashpot is pitch damping.
- Lanchester, in 1908, held the angle of attack fixed and kept only speed and path angle. Gravity slows a climb, and lift grows as speed squared, so a speed excess curves the path. The period is root 2 pi V over g: 22.64 seconds. The full model's 27.98 seconds is 23.6 percent longer, a gap of 5.33 seconds. What is the cause?
- Remove effects one at a time. Without drag the period barely moves: 27.85 seconds. Without pitch damping it falls to 22.70, almost exactly Lanchester. The nose must rotate with the curving path, and its damping moment is balanced by a lower angle of attack, so the lift curving the path is weaker. That softens the spring by 1 plus kappa, 0.5101, stretching the period by 1.2289 to 27.83 seconds.
- Now the static margin slider. The short period spring is the margin, but the real part has no C m alpha in it, so the pair moves vertically. At C m alpha minus 3, the imaginary part is 6.79 and zeta falls to 0.365. At minus 0.1 it is 1.23 and zeta rises to 0.908. At the neutral point the pair splits onto the real axis, and past it a positive root appears: the nose simply diverges.
- How small is small? A first-order series should err by the disturbance squared. A pitch kick of 0.2 radians per second leaves gaps of 0.017 degrees and 0.14 metres of altitude. Three times the kick, 0.6, gives 0.158 degrees and 1.27 metres: nine times the gap. The linear model is for small disturbances.
- Handling qualities turn pilot ratings into eigenvalue bounds. From memory, a Level 1 aircraft wants a short period zeta of about 0.35 to 1.30, and a phugoid zeta of at least 0.04. Ours, 0.585 and 0.056, pass, though the derivatives are assumed values.
- To recap. Linearise about trim with a central-difference Jacobian. Each eigenvalue is a mode: the real part decays, the imaginary part turns. The short period is minus 2.6647 plus or minus 3.6964 i, and the phugoid takes 27.98 seconds. Lanchester's 22.64 is 23.6 percent short, through pitch damping rather than drag.
- Next, lateral modes: the other block of four, v, p, r and phi. It has three modes: a fast roll subsidence, a slow spiral that may diverge, and the Dutch roll wobble that brought the first yaw dampers.
Trim: the point we linearise about
A trim state is an equilibrium: every derivative is zero except steady motion along the path. 12.4 found level trim at m/s by Newton’s method: , , N, with N. A Taylor expansion needs this exact base point: about a state that is not an equilibrium a constant leftover term appears, and the “modes” drift for reasons unrelated to stability.
Small perturbations: from thirteen ODEs to
Write the state as trim plus a small change, , and expand to first order (7.7 Nonlinear Systems & Linearization):
since (apart from the steady , which feeds nothing back). Two facts cut thirteen states to four.
Longitudinal and lateral decouple. At wings-level trim . In the , , equations these appear only in products (, , ) or squares ( inside ), or through , so their first derivatives vanish. The Jacobian splits into a longitudinal and a lateral block; 12.6 takes the lateral one.
Position drops out. Nothing depends on , , , and with sea-level density everywhere does not feed back. (A real atmosphere adds a weak fifth state.) That leaves and ; altitude is an output, .
The Jacobian by finite differences
A simulator programmer does not differentiate forces by hand. They nudge each state in the same derivative function the simulator flies:
with for , and for , . The sim builds 12.4’s 13-state vector (quaternion ), calls its deriv, and keeps , , ; the last row is , exact at . Columns :
Reading the entries. Each is a physical sensitivity.
- , : 12.4’s rotating-frame terms , .
- , : nose up tips the weight backward.
- s: more means more and more lift opposing ; mostly s.
- : , divided by because . s: pitch damping, .
- s: drag rises with speed; small, as drag is a tenth of lift.
- : raising lowers ; a body-axis artefact.
Eigenvalues as modes
Geometrically: suppose is a direction in the 4-D state space that only stretches, . Start exactly along ; then , the state stays on that line, and . Any start is a mix of the four eigenvectors, each evolving on its own:
Each term is a mode. Complex eigenvalues come in pairs , and a pair makes one real motion, with (7.4 Complex & Repeated Eigenvalues): a spiral in the plane of and . The real part is the decay rate; the imaginary part is the turning rate. Then
is the cosine of the angle from to the negative real axis: on the axis no oscillation, on the imaginary axis no damping.
The two modes of this aircraft
The characteristic polynomial is , with roots (sim readout, matched by numpy in the lab cell):
| mode | (1/s) | (rad/s) | period (s) | (s) | |
|---|---|---|---|---|---|
| short period | |||||
| phugoid |
The phugoid period is exactly the altitude-peak spacing 12.4 measured on the nonlinear model (peaks at and s). The pairs differ in size twentyfold: mildly stiff (4.5 Stability Regions & Stiffness).
Mode shapes (sim readout, per radian of ): the short period has m/s but m/s, an angle-of-attack change at nearly constant speed. The phugoid has m/s and m/s, a speed change at nearly constant angle of attack, leading by . These are the sim’s “pure mode” disturbances.
The short period
Over a second the speed hardly changes, so drop , and drop (it enters only through ). The block remains:
Worked example: the short period by hand
With , , , :
- , , so rad/s.
- : , .
- rad/s.
The gives and : good to . The spring is mostly , the weathercock action of ; the dashpot is plus lift opposing .
Reconciling 12.4’s pitch peak. The kick sets rad/s, , so with . Pitch peaks where : , s. Both sim runs peak at s, above trim: 12.4’s .
The phugoid and Lanchester’s approximation
Lanchester (1908, from memory) held the angle of attack fixed, let thrust cancel drag, and kept only speed and path angle . Gravity slows a climb, ; lift at fixed grows as , so a speed excess curves the path up:
The full model’s s is longer (ratio , gap s). The sim’s model menu removes effects one at a time, re-trims and re-linearises:
| model | phugoid period (s) |
|---|---|
| full | |
| drag removed (, no induced drag) | |
| pitch damping removed () | |
| Lanchester |
Drag barely moves the period. Its job is damping: Lanchester with drag gives , near the full , and without drag the phugoid here is slightly unstable ().
The mechanism is pitch damping. The path keeps curving, so the nose must rotate with it, , and rotating costs a damping moment . The aircraft balances it by flying at a slightly different angle of attack, :
While the path curves up, is lower and the lift curving it is smaller. In the normal equation:
The spring softens by , the period grows by : s, leaving s for drag and dropped terms. As and , is speed-independent; the trim speed menu gives ratio at both m/s ( against s) and m/s ( against s).
Static margin and the short-period frequency
Extra lift acts chord behind the centre of gravity, at the neutral point; that distance is the static margin , and (holding and fixed). The short-period spring is , so
On the slider:
- The pair moves vertically. has no : the real part stays near ( at , at ) while falls from to () to ().
- Less margin, more damping ratio: , , at those three. Forward centre of gravity: a stiff, quick, lightly damped nose; aft: sluggish and well damped.
- At the neutral point the pair splits. A few thousandths from both pairs fall onto the real axis; at a root sits at ; beyond, a positive real root ( s at ). Not a growing oscillation: the nose simply diverges, doubling every s. That is static instability in eigenvalue language.
The phugoid lengthens too, through : s at , s at . Lanchester is the stiff-pitch limit.
How small is small?
A first-order Taylor series should err by the disturbance squared. The sim’s “nonlinear − linear” readout over 60 s:
| pitch kick | max gap | max gap | max gap |
|---|---|---|---|
| rad/s | m/s | m | |
| rad/s | m/s | m |
Three times the kick, nine times the gap. At rad/s (a hard yank) altitude is off by over a metre in . Nonlinear terms also couple the modes: from the pure short-period vector ( m/s), the linear altitude stays within m, but the nonlinear one reaches m at s, half a phugoid cycle later.
Handling qualities
Pilots rate aircraft on the Cooper-Harper scale (1 excellent, 10 uncontrollable), and specifications turn good ratings into eigenvalue bounds. From memory, MIL-F-8785C asks Level 1 aircraft for a short-period of about to in precision tasks and landing, and a phugoid of at least . Ours, and , pass. The sim’s dashed rays are ; at (static margin ) the pair sits on them. Too much margin is a problem too: a stiff, under-damped nose needing large elevator.
Computing it
This is the 4-state restriction of 12.4’s model (), so it reproduces the sim’s matrix to the printed digits. Set Cmq = 0 (trim does not change, since there) and the phugoid falls to s, next to Lanchester’s s.
Graded: the short period ( rad/s, ) and phugoid ( s, half-time s) from the 5-digit matrix; the sorting on a second matrix with known eigenvalues; and Lanchester’s s with the full model times longer.
Concept checks
Problems
Use the block with , , , . (a) Find and . (b) After a kick rad/s, , when does the pitch angle first peak? (c) Why before a quarter of the damped period?
Show worked solution
(a) s; , so rad/s.
(b) . With , , and differentiating gives , so . peaks where : , , s. The sim shows s for both models.
(c) A pure decaying cosine would reach zero at s. Pitch damping makes fall faster from the start (), so it gets there sooner.
(a) From and , derive . (b) With quasi-steady pitch balance, show , , and that the period becomes , . (c) Evaluate both at m/s. (d) Show does not depend on .
Show worked solution
(a) (using ), so and .
(b) No pitch acceleration (the short period has settled): , so . With nearly constant, . The extra lift moves to the left: , so is divided by .
(c) s. s, , , s: under the full s, where Lanchester alone is under.
(d) , , so is constant. The sim’s ratio: , , , at , , , m/s.
Ballast moves the centre of gravity aft to a static margin of chord. (a) Find . (b) Predict the short-period and . (c) Predict the phugoid period with the pitch-damping correction. (The sim gives , , s.)
Show worked solution
(a) .
(b) , rad/s; : slower, better damped.
(c) s, , s, within . Less margin, more lag of behind the path, a longer phugoid.
Key Takeaways
- Linearise about trim; a central-difference Jacobian of the simulator’s own derivative function gives with no hand differentiation.
- At wings-level trim decouple from the lateral states; position and (constant density) altitude drop out.
- Each eigenvalue is a mode: real part decay rate, imaginary part turning rate; , , .
- Here: short period (), phugoid ( s), matching 12.4’s s pitch peak and s altitude cycle.
- The block gives the short period to ; its spring is static margin, , its real part independent of it.
- Lanchester’s s is short, through pitch damping, not drag; recovers s.
- Past the neutral point a positive real root makes the nose diverge.
- Next, 12.6 Lateral-Directional Modes: the other block, , and its three modes: fast roll subsidence, a slow spiral that may diverge, and the Dutch roll wobble that brought the first yaw dampers.