Automatic for the people

In general, manual tuning methods rely on subjective assessments such as “over damped” or “under damped.” Automatic tuning, generally referred to as “auto-tuning,” holds out the promise of making this process more scientific and repeatable. Better still, automatic tuning places much (but not all) of the burden of tuning onto an algorithm.

Auto-tuning methods tend to use academically researched tuning methods. Of these, Zeigler-Nichols (ZN) is the best known. Unlike the manual methods described above, this method assumes a certain mathematical model to describe the process to be controlled, and then performs tests which are translated through a series of rules into the PID parameters.

The first proposed version of Zeigler-Nichols was not optimized for automatic tuning however, and thus modified ZN methods were developed. One such approach is known as the “frequency response method.” This method replaces the PID with a relay (all on positive, and then all on negative) controller during tuning. Using this approach the servo loop becomes oscillatory, and the measured frequency and gain of this oscillation are used to determine the PID parameters.

Sample auto-tuning process

Despite all this, there is relatively little that has been published regarding implementation of auto-tuning procedures for servo loop tuning. This may be due to the fact that most of this work has been done by vendors which consider the algorithms proprietary, or it may be due to the fact that interest in auto-tuning has increased relatively recently, as inexpensive computing power has become available.

Figure 4 provides a flow chart for a specific auto-tuning procedure used in conjunction with PID controllers. This method uses three phases. The first phase is used to derive a value for D, using a method similar to that used in zone-based tuning. The second phase derives values for P and I using the relay test approach described above. And the third phase uses information acquired about the system to allow the user to hand-optimize the results, but using inputs such as “quieter versus noisier” and “aggressive versus less aggressive” to provide meaningful yet easy-to-understand control inputs.