The Centrifugal Governor: How a Spinning Machine Regulates Its Own Speed

The core physics: a rotating pendulum finds its cone

Strip the governor down to one flyball on a rigid arm of length ℓ, hinged at a pivot on the spinning spindle. When the spindle turns at angular velocity ω, the ball swings out to a steady cone angle θ measured from the vertical spindle. Two real forces act: gravity mg straight down, and the tension T along the arm toward the pivot. In the rotating frame there is no acceleration once θ is steady, so the horizontal and vertical components balance a net inward (centripetal) requirement:

• Vertical: T cos θ = mg (the arm holds the ball up).
• Horizontal: T sin θ = m ω² r, where the orbit radius is r = ℓ sin θ.

Divide the second by the first to kill both T and m: tan θ = ω² r / g = ω² ℓ sin θ / g. Cancel sin θ (θ ≠ 0) to get the governor's master relation:

• cos θ = g / (ω² ℓ)

Everything the device does follows from this. The vertical drop of the ball below the pivot is h = ℓ cos θ = g/ω² — a beautifully simple result independent of arm length and independent of mass. That height h is the signal the machine reads: a sleeve threaded on the spindle rides up as the balls rise, and its position tracks 1/ω².

Why the balls rise: the 1/ω² signal and the minimum speed

The relation cos θ = g/(ω²ℓ) says the cone opens as ω grows. Because cos θ ≤ 1, a real solution with θ > 0 exists only when ω²ℓ > g, i.e.

• ω > ω_min = √(g/ℓ)

Below this critical rate the balls simply hang straight down (θ = 0); the governor is asleep. For ℓ = 0.30 m, ω_min = √(9.81/0.30) ≈ 5.7 rad/s ≈ 55 rpm — which is exactly why real engine governors were built to run in the 60–150 rpm band, comfortably above cutoff. This threshold is identical to the natural frequency of the same arm swinging as an ordinary pendulum, √(g/ℓ), which is no coincidence: the governor is a pendulum that has been spun up past its own swing frequency.

The height signal h = g/ω² is what makes the governor a speed sensor. Note its nonlinearity: doubling ω quarters h. Near ω_min the ball height changes steeply with speed (high sensitivity, good for control); at high ω the balls are nearly flung horizontal and h → 0, so the device loses resolution — you can see this in the table, where dθ/dω is steep at 60 rpm and shallow at 120 rpm. Designers chose ℓ and the running speed to keep the operating point in the responsive part of the curve.

Closing the loop: from ball height to throttle valve

A speed sensor alone regulates nothing. Watt's genius was mechanical feedback. The rising sleeve is linked, through a bell-crank and rod, to the engine's steam throttle valve. The wiring is deliberately negative: balls up → sleeve up → valve closes → less steam → engine slows → balls fall → valve reopens. The loop drives the system toward a single equilibrium speed.

Trace a disturbance. Suppose the mill it drives sheds load — fewer looms engaged. With the same steam, net torque now exceeds resistance, so the flywheel accelerates, ω rises, and by cos θ = g/(ω²ℓ) the balls climb. The sleeve lifts, the throttle pinches, steam torque drops, and ω settles back. A load increase runs the same story in reverse: ω sags, balls drop, throttle opens, torque climbs. The engine, in effect, senses its own load and answers it — the defining act of a control system, and the reason Norbert Wiener later cited the governor as the origin of the word cybernetics (from Greek kybernḗtēs, 'steersman').

Crucially the equilibrium is a balance of torques, not speeds: the engine stops changing speed when driving torque equals load torque, and the valve opening that produces that torque corresponds, through the linkage, to one particular ball height and therefore one particular ω. That coupling of a mechanical equilibrium to a target speed is the whole trick.

Droop, offset, and the limits of proportional control

The simple Watt governor is a proportional controller: valve position is a fixed function of speed. That has a built-in defect — droop (also called offset or steady-state error). Because a heavier steady load needs the valve held more open, and 'more open' corresponds to a lower ball height and hence a slightly higher ω, the governor holds a different equilibrium speed at each load. It cannot restore the exact original set-point; it trades a permanent small speed error for stability.

• Typical droop for a good 19th-century governor was a few percent — full-load to no-load speed change of order 2–5%.
• Sensitivity is set by how much θ changes per unit ω: differentiate cos θ = g/(ω²ℓ) to get sin θ · dθ = (2g/ω³ℓ) dω, so dθ/dω = 2g/(ω³ℓ sin θ). Larger arms and lower speeds give a bigger swing per rpm — a more sensitive, but potentially twitchier, governor.

Engineers reduced droop by adding springs (Porter governors loaded the sleeve; Hartnell governors used a compression spring so running speed could be tuned) and, in modern terms, by adding integral action — a mechanism whose output depends on the accumulated error — which drives steady-state offset to zero. But integral action introduces phase lag, and that is where the mathematics turns subtle.

Maxwell, hunting, and the birth of control theory

Real governors sometimes hunt: instead of settling, the speed oscillates, the balls bob, the engine surges. In 1868 James Clerk Maxwell published On Governors, treating the governor plus engine as a coupled dynamical system and linearizing about equilibrium. The stability question became: do small perturbations decay or grow?

Maxwell wrote the linearized equations of motion and asked whether the roots of the resulting characteristic polynomial have negative real parts — because a small deviation δ evolves like a sum of e^(λt) terms, and any root with Re(λ) > 0 means growing oscillation (hunting), while all-negative real parts mean the system self-corrects. He solved the low-order cases explicitly and posed the general problem to mathematicians; it was answered by Routh (1875) and independently Hurwitz (1895), giving the Routh–Hurwitz criterion still taught today. So the humble flyball governor is quite literally the seed of modern control theory.

The physical culprits are inertia and damping. The flywheel and the flyballs both store kinetic energy and resist changing state, introducing time lags between a speed error and the valve's response. If the loop gain is high and the lag large, the correction arrives out of phase and pumps the oscillation instead of quelling it. Adding friction/damping at the sleeve (dashpots) bleeds energy from the oscillation and stabilizes the loop — the same reason you damp any resonant system near instability.

Scales, numbers, and where governors live today

Put numbers on it. Take arms ℓ = 0.30 m running at 120 rpm (ω = 2π·2 = 12.57 rad/s). Then cos θ = 9.81/(12.57²·0.30) = 9.81/47.4 = 0.207, so θ ≈ 78° and the balls sit only h = g/ω² = 9.81/158 = 0.062 m below the pivot. The centripetal acceleration on each ball is ω²r = 12.57²·(0.30·sin 78°) ≈ 46 m/s² — nearly 5 g — which is why the arms, pins, and sleeve had to be stout forged steel; a 2 kg ball pulls ~90 N outward, and it does so continuously for years.

The principle long outlived steam:

• Gasoline and diesel engines used mechanical flyweight governors for decades to limit maximum rpm and idle stably; small engines (generators, tractors, chainsaws) still do.
• Music boxes, gramophones, and film projectors used tiny air- or friction-damped centrifugal governors to hold constant playback speed — a fan-brake governor is a direct descendant.
• Elevator overspeed governors are safety-critical centrifugal devices: if the car exceeds rated speed by a set margin, flyweights trip the safety brakes.
• Modern turbines and engines replaced the flyballs with electronic governors — a tachometer plus PID software driving a fuel actuator — but the control problem, and Maxwell's stability analysis, are unchanged.

Subtleties and common misconceptions

'Centrifugal force flings the balls out.' In an inertial frame there is no outward force at all; the balls rise because the arm tension's horizontal component provides exactly the inward centripetal force m ω² r needed to keep them orbiting, and the geometry that satisfies both force balances happens to be a wider cone at higher ω. The 'centrifugal' name is a rotating-frame bookkeeping convenience, not a new force — see the related note on centripetal vs centrifugal framing.

'The governor keeps speed perfectly constant.' No — a pure proportional governor cannot, because of droop; it holds a slightly different speed at each load. Perfect regulation requires integral action. Isochronous (zero-droop) governors exist but are prone to hunting precisely because zero droop means very high loop gain.

'Bigger balls give tighter control.' The equilibrium cone angle is mass-independent — cos θ = g/(ω²ℓ) has no m in it. Mass matters for the forces and dynamics: heavier balls give more sleeve force to overcome valve stiction and friction, and more inertia (slower, more stable response). Choosing ball mass and arm length trades responsiveness against stability, not the set-point.

'It works because of energy.' The set-point comes from a force/torque balance, not an energy minimum — though you can also read cos θ = g/(ω²ℓ) as the stationary point of the ball's effective potential in the rotating frame, U_eff(θ) = mgℓ(1−cos θ) − ½mω²ℓ²sin²θ, whose minimum shifts to larger θ as ω increases. Both pictures give the same governing equation.

The conical pendulum at two rotation rates (arm length ℓ = 0.30 m, so ω_min = √(g/ℓ) ≈ 5.7 rad/s ≈ 55 rpm)

Quantity	Slow (ω = 6.3 rad/s, 60 rpm)	Fast (ω = 12.6 rad/s, 120 rpm)
Cone angle θ = arccos(g/ω²ℓ)	≈ 34.5°	≈ 78.1°
Ball height rise below pivot h = g/ω²	0.247 m	0.062 m
Radius r = ℓ sin θ	0.170 m	0.295 m
Centripetal accel a = ω²r	6.7 m/s² (0.68 g)	46.9 m/s² (4.8 g)
Sensitivity dθ/dω near this ω	steep (responsive)	shallow (near flat-out)