I got a plant $G(s)=\left(0.13s+1\right)/s^2$ to design a compensator which provides below demands:

  • Settling time : max 2s
  • Overshoot : max %35
  • Gain margin : min 10 dB
  • Phase margin : min 30 deg
  • Controller effort (r to u) : max 0.9
  • Bandwith : min 10 rad/s

The best architecture so far was the one below but I couldn't reach the demands.

enter image description here

assigning $C1=0.03\cdot\left(\left(s+80\right)\left(s+10\right)\right)/\left(\left(s+0.12\right)\left(s+1\right)\right)$, $C2=17.5\text{m}$ and $H=1$ results as below:

enter image description here

Can anyone explain or guide a design approach on how to handle $\left(s+a\right)/s^2$ type plants when designing compensators or mention some tips/shortcuts for architecture selection? How do we select the order of the controller?

  • $\begingroup$ Is this a homework question? Or like a job? Could you show what your compensator could accomplish in a plot? What it does manage, and also doesn‘t manage? What don‘t you understand when it comes to design and choosing an order? $\endgroup$ Oct 23, 2019 at 8:07
  • $\begingroup$ it is a job like question and I editted the question to show how my compensator affects. As to order, I know that we should select lower order compensator as much as possible and there is no exact prescription for the type of compensator selection. what I want to learn is that is there a relation or at least an approach between plant order with compensator order when we start to design a compensator. for example my plant is second order and so should I start with 2th order compensator (like lead-lag) in order to save time by skipping working on 1th order compensators. $\endgroup$
    – lsn
    Oct 23, 2019 at 9:09
  • $\begingroup$ Is there a reason why you choose that specific architecture, instead of for example a single feedback loop? $\endgroup$
    – fibonatic
    Oct 23, 2019 at 9:20
  • $\begingroup$ this architecture performed better than the others. I tried the others also. $\endgroup$
    – lsn
    Oct 23, 2019 at 9:40
  • $\begingroup$ Maybe this will help, $-\frac{8.20546 \left(1. s^2-3.84499 s+118.343\right)}{s}$ $\endgroup$ Oct 23, 2019 at 21:50

1 Answer 1


Consider the traditional control diagram below.


If you set $C=(2\zeta\omega_ns+\omega_n^2)/(0.13s+1)$, then you'll get the following closed-loop system transfer function: $$ T=\frac{GC}{1+GC}=\frac{2\zeta\omega_ns+\omega_n^2}{s^2+2\zeta\omega_ns+\omega_n^2}. $$

This can be achieved through zero-pole cancelation, which is doable since $G$ has a zero in LHP.

Finally, you can easily attain your design requirements with $\omega_n \approx 10\,\text{rad/s}$ and e.g. $\zeta=\sqrt{2}/2$.

The requirement on the maximum controller effort cannot be attained with a pure LTI controller though. You may thus consider LTV and/or nonlinear approaches to this end for controlling a double integrator (for example relying on optimal control).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.