Lesson 3 of 14 · Module A: Foundations

Linearisation and deviation variables

Process Dynamics and Control · EIE306

By the end of this lesson you can

  • Linearise a nonlinear model about a steady state using a Taylor series
  • Write a model in deviation variables
  • Judge when the linear model is accurate

Notes

Transfer functions and most controller design methods need linear models. We get one by approximating the nonlinear model close to a steady operating point.

Taylor series

For a function $f(x)$ near the steady state $\bar x$, keep only the first-order term:

$$f(x) \approx f(\bar x) + \left.\frac{df}{dx}\right|_{\bar x}(x - \bar x)$$

Deviation variables

Define $h' = h - \bar h$ and $q' = q_{in} - \bar q_{in}$. Deviation variables are zero at the steady state, which makes initial conditions zero and Laplace transforms simple.

Tank example

Linearising $\sqrt{h}$ about $\bar h$: $\sqrt{h} \approx \sqrt{\bar h} + \dfrac{1}{2\sqrt{\bar h}}h'$. Subtracting the steady-state balance gives

$$A\frac{dh'}{dt} = q' - \frac{c_v}{2\sqrt{\bar h}}\,h' \quad\Rightarrow\quad \tau\frac{dh'}{dt} + h' = K q'$$

with $K = 2\sqrt{\bar h}/c_v$ and $\tau = 2A\sqrt{\bar h}/c_v$. Both depend on the operating point: the tank responds more slowly and more strongly at high levels.

The linear model is good for small changes. A common rule of thumb is to trust it for changes of up to about 10–20 % around the operating point.

Lecture

🎬
Lecture video
Coming soon. The recorded lecture for this lesson will appear here.

Exercises

  1. Linearise $f(T) = k_0 e^{-E/RT}$ about $\bar T$.
    Show solution
    $f \approx k_0 e^{-E/R\bar T}\left[1 + \dfrac{E}{R\bar T^2}(T - \bar T)\right]$.
  2. For the tank of Lesson 2 at $\bar h = 4$ m, find $K$ and $\tau$.
    Show solution
    $K = 2\sqrt4/0.5 = 8\,\text{min/m}^2$, $\tau = 2(2)(2)/0.5 = 16\,\text{min}$.
  3. Why do we use deviation variables before taking Laplace transforms?
    Show solution
    Because they make all initial conditions zero at the steady state, so the transform of a derivative is simply $sY(s)$ and the transfer function can be read directly.