Process control engineers treat final control elements in the same way they treat measurement devices - with absolute indifference. To most of them, a valve is a valve is a valve. It's job is to open and close according to what the controller tells them and they do just that. The problem is, in a shocking number of cases, they don't.
Control valves and dampers, being mechanical devices are subjected to a lot of mechanical issues like wear and tear, deterioration with time. Measurement devices these days are very robust and most would not deteriorate drastically with time. When they fail, they just go. The controller running inside the modern distributed control system (DCS) is even more reliable with a hot standby ready to take over in case of failure. The same cannot be said about final control elements. Therefore the weakest link in the process control loop is frequently the final control element.
The problems of control valves usually manifest themselves as stiction, deadband and hysteresis.
Deadband is a general phenomenon where a range or band of controller output values fails to produce a change in the measured process variable. This is bad for process control. Process control systems these days execute at a rate of about 3 times per second. On top of that, each time it executes, the output changes in the magnitude of usually less than 1%. But most relevant in the case of deadband, the changes can occur is either direction.
If a control valve is suffering from a deadband problem, when the controller output reverses direction, the control valve does not respond. Therefore the process variable also does not respond to the command of the controller. The controller does not know it, it thinks that its previous command is not good enough and so issues another (sometimes more drastic) command. When the control valve finally comes out of its deadband, the controller command has caused it to overshoot.
The controller then tries to go back the other direction only to be faced with the same situation. And the process will be driven to overshoot in either directions and cycles continuously forming what is called a limit cycle.
This is somewhat similar to deadband except that it does not only happen when the controller changes direction. Again stiction (also known as 'sticky valve') can be due to a variety of reasons, a common one which is packing friction.
As far as process control is concerned, the effect of stiction is also like deadband whereby the valve fails to respond when required and when it does respond, overshoots the setpoint. The controller then tries to bring it back the other way.
Hysteresis occurs when the same change in the controller output in both directions results in a different change in the process value. For example, when the controller output is 20%, the process variable is 30°C. when the controller output increases to 25%, the temperature increases to 35°C. However, when the controller goes back down to 20%, the temperature only goes down to 33°C.
This results in different process gains in both directions and will confuse the controller, which has been tuned for only one process gain. We have to remember that industrial controllers are linear. The following figure illustrates a case of valve hysteresis.