THEORY BEHIND EXPERIMENT 20 · DC MOTOR

Predict L and T of a reaction curve by hand

The lab draws the steepest tangent through a recorded step and reads L, T and K from it. This note computes the same three numbers before the run, from the motor's two poles and the 200 ms delay, and shows which physical part of the motor each one measures.

Companion to the lab experiment Read a reaction curve. Run it there; this note predicts what you will see and why.

The model in play

The motor (R = 2 Ω, L = 0.12 H, Kₑ = Kₜ = 0.05, J = 0.002 kg·m², b = 0.0005 N·m·s) starts at rest with no load and an open-loop command of 7 V that reaches it 200 ms late: the engine passes the voltage through a first-in, first-out line of whole 5 ms steps. The lab's panel takes the 20 ms samples, finds the steepest slope by central differences and fits the first-order-plus-dead-time model below.

Li˙=V(t−θ)−Ri−Keω,Jω˙=Kti−bω−τloadL\dot{i}=V(t-\theta)-Ri-K_e\omega,\qquad J\dot{\omega}=K_t i-b\omega-\tau_{load}
G(s)≈K e−LsTs+1,Z–N PI: kp=0.9 TKL,Ti=L0.3G(s)\approx\frac{K\,e^{-Ls}}{Ts+1},\qquad \text{Z–N PI: } k_p=\frac{0.9\,T}{KL},\quad T_i=\frac{L}{0.3}

Simplifications that matter here

  • The fitted model replaces the motor's two poles, at −0.911 and −16.01 s⁻¹, with one lag and a delay; the fast electrical pole turns into a little extra apparent delay.
  • The tangent is read from 20 ms samples, so its point is a sample next to the true inflection.

Predict it by hand

Without the delay, the normalized step of the two poles a and b is y(t) = 1 − (b e^(−at) − a e^(−bt))/(b − a). Its slope peaks at the inflection t* = ln(b/a)/(b − a). The tangent there needs T = 1/y′(t*) to climb from 0 to 1 and leaves zero at t* − y(t*)/y′(t*), the apparent delay of the current building up. The 200 ms input delay shifts the whole curve, so it adds to L and leaves T alone. K is the DC gain, 1/0.07 (rad/s)/V.

t∗=ln⁡(b/a)b−a=0.190 s,y(t∗)=0.111,y′(t∗)=0.766 s−1t^*=\frac{\ln(b/a)}{b-a}=0.190\ \mathrm{s},\qquad y(t^*)=0.111,\qquad y'(t^*)=0.766\ \mathrm{s^{-1}}
T=1y′(t∗)=1.305 sT=\frac{1}{y'(t^*)}=1.305\ \mathrm{s}
L=t∗−y(t∗)y′(t∗)+0.2=0.045+0.2=0.245 s,K=14.29 rad/sVL=t^*-\frac{y(t^*)}{y'(t^*)}+0.2=0.045+0.2=0.245\ \mathrm{s},\qquad K=14.29\ \tfrac{\mathrm{rad/s}}{\mathrm{V}}

Check against the engine

The engine's panel reads L = 0.245 s, T = 1.307 s and K = 14.29 (rad/s)/V. T is 0.002 s longer than the hand value because the slope is a central difference over 20 ms samples: the steepest one is centred on 0.40 s, next to the true inflection at 0.39 s (0.2 + 0.190). With those readings the Z–N PI rule gives kp = 0.336, the gain the next experiment starts from.

What to change next

  1. Double the actual inertia.

    The slow pole moves from 0.911 to 0.446 s⁻¹, so the same formula predicts T = 2.480 s and L = 0.049 + 0.2 = 0.249 s, with K unchanged. The engine reads L = 0.249 s and T = 2.469 s, and the rule's PI gain rises to kp = 0.627 because T/L grows. K reads 14.21, not 14.29, because after 12 s the speed is still at 99.47 rad/s.

  2. Set the dead time to none and choose the right-half-plane zero at s = +5 rad/s instead.

    For slow signals (1 − s/z) ≈ e^(−s/z), a delay of 1/z = 0.2 s, so predict nearly the same L, about 0.045 + 0.2 = 0.245 s, plus a dip below zero first. The engine reads L = 0.242 s, a 9.50% undershoot, and a shorter T = 1.193 s, because the zero's −y′/z term also steepens the climb.

Common mistakes when reading this run

  • Reading L as the input delay alone. With no dead time the panel still shows L = 0.045 s: the current's build-up through the inductance.
  • Expecting a smaller step to change the reading. The model is linear: at 3.5 V the speed ends at 50.00 rad/s and L, T and K are unchanged.
  • Reading a curve that has not settled. K is the final change divided by the step; a run that is still 0.53% short, as with doubled inertia, biases K and every gain derived from it.

Where to go next

Run this experiment in the lab, pin a run and try the two changes above.

Lectures

Terms used in this note

Calculator

PID tuning calculator: Open this reaction curve in the PID-tuning calculator: Z–N PI gives Kp 0.336 with Ti 0.817 s, and SIMC a gentler Kp 0.187 with Ti 1.307 s, each simulated on the fitted model.