Joined: 23 July 2012
Why does the industry prefer to use PI rather than PID or PD? Is it because of the reasons below? Please let me know if my theory / finding is wrong.
- Easier to tune
- Easier to implement
- Save time
- D is difficult to find
- D slow down performance
- D creates slight oscillations.
Joined: 05 September 2004
1. Project plans tend to underestimate the time required for commissioning, and in particular controller tuning. So as soon as you start people are normally asking when you will be finished. So keeping things simple is the best approach for bespoke control systems, especially as the small overshoots you get with PI controllers are never normally a problem. In a bespoke system the final set point can always be changed to compensate for the maximum overshoot or undershoot if this becomes a problem. It is easier to do that than spend a lot of time tuning the full PID controller.
2. Using a PI controller with integral de-saturation gives good stability even when controlling non-linear systems with varying frictional or resistive loads. In practice the small unavoidable overshoot given by a PI controller, is never normally a problem in bespoke systems. However instabilities arising from adding a differential term to try to control systems with varying loads and inertias that can't all be tested in advance during the commissioning phase is a real problem.
De-saturating the integral to the maximum or minimum output achievable can avoid the controller locking up with integral runaway. This can happen if controller remains enabled and the measurement input sensor is disabled or disconnected for some reason.
I don't think it works if you use integral de-saturation on a full PID controller, however I am not completely sure on that.
For a system with variable inertia and variable loads like an end of line or simulation dynamometer using the right constant integral term means you can cope with changing inertia by altering the proportional term in a linear fashion. The exact way the proportional term needs to vary with changing inertia can be calibrated for in advance (even for inertia choices not used during commissioning). This saves a lot of time.
I don't know the best way of doing the maths to prove this, but it definitely works.
3. Adding differential terms increases the noise sensitivity of the controller, so if the noise floor on the measurement input changes over time the controller performance can be degraded, and instabilities may result. Noise on measurement inputs may change over time in ways which can't be easily tested for in advance during a short commissioning period. The integral term alone insulates the controller from variations in the measurement noise floor.
There are probably some other reasons, but I can't remember. Someone designing a controller for a product which is a well behaved second order system with low noise levels and with measurement sensors that can't easily be disconnected or disabled might have a completely different opinion to me. I am specifically taking here about relatively slow response bespoke systems which need to be tuned on site during a very short commissioning phase.