How much is too much control valve cycling?

Question

In a textbook on process controls, I read that when tuning a controller there must be a balance between performance and excessive control element cycling. My question is, what corresponds excessive cycling?

To narrow down the question, assume we have a control valve with linear characteristics. What is the minimum time the controller should take to move the valve 5% of the controller output? For a typical set point change, how many cycles (up and down) of the valve are too much prior to reaching the final valve position?

I suspect there is a process specific part of the answer (minimize disturbance, smooth operation criteria), but what I am after is: what is too much cycling to avoid damaging the valve itself or to prolong the valves lifetime?

Answer 1

Assuming you have correctly selected/sized the control valve and actuator assembly, you could expect valve motion to act within 10s of seconds (you could make it open slower by fiddling with the actuating mechanism). Cycling usually isn't a problem with respect to the valve lifetime (typically stated in the years timeframe) when the valve is properly sized, but it can become an issue if the valve's operating temperature or pressure is exceeded.

I don't have any experience with valves that have internal packing, so I won't comment on them further than saying cycling may affect the lifetime of valve packing. It would be worth talking with a controls equipment company to get a more detailed understanding on this area.

Answer 2

I would assume that most of the time a feedback gain is limited by how big of an input you can give, so when it starts to saturate. You would like to avoid is breaking the system, for example if a too big of a force is applied, or the system might overheat if too much energy is supplied in a too short of a time frame.

Your sensor sampling rate also limits the fastest possible rate at which you can control your system. So applying a feedback gain which would result is an analytically faster response (when modeling it as an continues system) would also not improve the performance.