Controller tuning is the process by which a control engineer or technician selects values of user-adjustable controller parameters (for a PID controller these are the bias, gain, integral time, and derivative time) so that the closed loop dynamic response behaves as desired.
Loop tunings are the primary point of contact between an operations/manufacturing engineer and the plant control system. Controller settings determine the system response: a poorly tuned controller may be as bad as no controller at all.
Tuning is a exercise in compromise. Controller objectives, specifications, requirements, and performance always conflict to some degree or another. There are rarely absolute criteria for selecting tunings and so judgement is required.
As you prepare to tune a loop, you must consider a range of concerns and objectives.
All control loops are fundamentally concerned with two objectives: disturbance rejection and setpoint tracking. In the CPI, disturbance rejection is normally the more important concern (despite what the examples in control textbooks may suggest).
An important secondary objective is to minimize the "cost" and variability of your manipulated variables.
Before tuning, you need to have some idea of what to expect. In particular, you want to have an idea of what sort of inputs are likely -- step changes? ramps? impulses? -- and how big they are likely to be. A "tight" tuning designed for small inputs may be the exact opposite of what one would do for a large input.
You also need to understand your system. How much noise do you anticipate? What constraints do safety, the environment, and equipment protection impose on your plans? What constraints do nearby units and equipment impose?
Most methods for obtaining initial tuning sections are based on some sort of model. What type of model are you using? Can you quantify the amount of plant/model mismatch?
You can't tune a loop unless you have some way of deciding whether or not it is "working". Consequently, you'll need to determine how you will measure success. Desired response is often quantified in terms of one of the loop performance specifications discussed earlier in the semester. The specifications used depend on the process, but might include:
Sometimes it is useful to use broad measures of performance that focus less on the specifics of the loop than on the general variability and deviation from desired performance. These types of criteria are particularly important in organizations that attempt to measure "quality" and employ statistical quality control techniques. SQC techniques are primarily designed to reduce and eliminate variability.
The error in a control loop is usually defined as the deviation from setpoint. There are a variety of ways of quantifying the cumulative error:
References:
R.M. Price
Original: 3/24/97; 11/18/93
Modified: 4/13/98; 2/28/97, 4/13/98, 5/26/2003; 7/8/2003
Copyright 1998, 2003 by R.M. Price -- All Rights Reserved
