Analysis Software Answers Performance Questions

Several software packages help end-users design and tune their control loops for "optimal" performance. Some tuning tools work better than others, but it's not always easy to tell the difference because it's not always clear whether one design works any better than another. Controller A with one set of tuning parameters may eliminate errors faster than controller B with a different set of...

By Vance J. VanDoren, Ph.D., P.E. April 1, 1998

Several software packages help end-users design and tune their control loops for “optimal” performance. Some tuning tools work better than others, but it’s not always easy to tell the difference because it’s not always clear whether one design works any better than another. Controller A with one set of tuning parameters may eliminate errors faster than controller B with a different set of tuning parameters, but controller B may cause excessive fluctuations in the process variable.

The problem is defining “optimal” loop performance and assessing whether or not a particular controller has been able to achieve it. That’s where ProcessDoc-PA from Matrikon Applications (Edmonton, Alberta, Canada) comes in. It uses process data collected during normal closed-loop operations to compare the performance of an existing controller to a user-specified benchmark.

ProcessDoc supports three benchmarks—settling time, minimum variance, and relative performance. Settling time is the simplest measure of closed-loop performance. It shows which controller is better by determining which one lines out first after a disturbance or a setpoint change.

Variance is a measure of how much the process variable fluctuates around its setpoint. For control applications requiring the smoothest possible process variable, variance should be minimized. ProcessDoc can compute the actual variance present in the process variable and compare it to the theoretical minimum variance. This feature allows users to test several controllers for variance and select the one that comes closest to achieving the theoretical minimum.

Can the controller do better?

Perhaps more importantly, the variance calculation also allows users to determine if any controller can do better than the last one tested. If the existing controller successfully achieves the theoretical minimum variance, then any further reductions in variance would have to come from changes in the process (such as reducing the transport delays that cause deadtime). Further refinements of the controller would be futile.

The relative performance benchmark allows users to directly compare the performance of two controllers (or one controller with different settings). This feature is particularly useful for determining if retuning a controller will lead to any significant improvement in performance.

By comparing the variance before and after a tuning operation, it also allows users to quantify the relative performance improvement achieved by retuning. Several time- and frequency-domain graphs are provided to help users decide which controller (or which settings) performed better, and why.

ProcessDoc handles most of the analysis automatically. Users supply a few parameters including a value for the process deadtime (or a range of possible values). Knowledge of the deadtime allows ProcessDoc to observe the controller’s effects on the process variable correctly. Without it, ProcessDoc would be unable to identify the control efforts reponsible for specific changes in process variables.

Although knowledge of the deadtime is certainly critical for assessing a controller’s performance, it seems to me that ProcessDoc should be able to estimate the deadtime automatically, using just the process data it already has. The fewer in- puts required from the users, the better. Fortunately, the relative performance benchmark does not require knowledge of the deadtime to evaluate the performance improvement based on retuning operation.

ProcessDoc runs under Microsoft Windows 95 or NT 4.0 on a 133 MHz Intel Pentium or better. It requires 32 MB of RAM and 20 MB of available disk space. The list price for ProcessDoc-PA is $10,000.

For more information on ProcessDoc-PA, www.controleng.com/info .

Author Information
Consulting Editor, Vance J. VanDoren, Ph.D., P.E., is president of VanDoren Industries, West Lafayette, Ind.