Remote tuning of controllers

01 February 2007

Poor performance or unstable behaviour of a critical controller can lead to a significant deterioration of the operation of an entire unit. However, bringing troubled loops back to sound performance quickly is easier said than done.

Troubleshooting controllers can be a demanding task. It is comparable to detective work that requires experience, analytical skills, and a well organised approach. Yet, in many cases even with sound expertise and a solid approach, a solution might not be possible. Without the help of suitable tools, finding a solution may take too much time.

And time is a critical factor. The longer the problem persists, the higher is the economic loss for the plant. And secondly, in today’s tight manpower situation any unplanned extra effort poses a real problem.

So if just one of these key factors— expertise, time, or tools—is missing, the chances for success are low and then the most obvious action is to call for outside help. This is exactly what happened in the following case.

The (not so unusual) problem
One of our customers had a serious problem in an important loop: a distillation tower temperature was
swinging. The experienced staff had no tuning or trouble-shooting tool at their disposal and attempts to resolve the situation had failed. Time was a major problem because of deadlines for other projects. So they called the author for help.

In such a case I would normally go to the plant and solve the problem, but this customer was 1,000 kilometres away and off the main routes. The time needed to get there plus the travel cost made a visit clearly the last option to consider.

As an alternative, I suggested the staff collect a set of data that would allow me to study the loop and extract the needed process information. Of course, an open loop test would be preferable, but data from a closed loop test or even ‘normal’ operating data would do as well, provided there was enough information content.

The approach
A key task in curing a troubled loop is to identify the process parameters, because once they are known the PID settings can be easily calculated. The sequence of steps to be taken is to first analyse the data for usefulness, next to obtain the process parameters, then calculate the PID tuning parameters, and finally to test the results.

This approach includes several different tasks that require suitable tools. We used our own software toolkit TOPAS as it addresses all these different tasks and combines a comprehensive data analysis facility, several identification methods, a novel PID tuner, and an integrated simulation environment.

Of course, no tool can make up for poor or missing information, so our main concern was the quality of the plant data. Unfortunately for this case an open loop test was out of the question, but a closed loop setpoint
step could be carried out. The data were transferred from the DCS into an Excel spreadsheet and sent to us by e-mail. The results are shown in figure 1. The setpoint is the yellow curve, PV the blue and the controller output the red one.

Our turn
Now it was our turn. The data were imported into TOPAS and a first inspection showed that the test results were certainly not ideal. There were unexplained movements in the PV and the output. But overall the step was big enough to give a clear reaction from the process.

First we would try a direct identification. Using the least squares approximation in TOPAS, the process parameters were quickly identified.

Had a direct identification not been possible, we would have resorted to a more heuristic procedure, although at the expense of extra time. Since the controller configuration and tuning were known we could have simulated the loop and adjusted the process parameters until its behaviour would closely enough
match the real loop. But a test run with the simulation proved that the fit was adequate (the green curve in figure 1).

We now proceed to the next step, the PID tuning. Based on the process and controller data and the specification (setpoint or load tuning, active or smooth control action, or minimum use of the resource), TOPAS evaluates 25 different methods and presents the three best results graphically.

These settings were tried again with the simulation. If necessary, manual adjustments could have been made in order to reflect influences not accounted for in tuning calculations like measurement noise, non-linear process behaviour, etc. But the tests showed that no further adjustments were necessary.

Finally just 10 minutes after the e-mail was received, new PID settings were sent back to the customer and a very short time after they were implemented, the loop stabilised—an achievement we have not heard about before. The key factors for this success were the power of the tool, especially the ease of the data import and analysis, and the range of powerful methods for process parameter estimation and PID tuning.

The consequences of poor performance
With the controller now back to normal operation let’s take a look at the benefits achieved for the plant. That the process needs to be stable is obvious, but that is not sufficient for a profitable operation: truly sound
control performance is required. Therefore let us briefly look at the negative effects poor control can have
on the operation, and by extension, profitability.

The setpoint is by definition the optimal value for the variable in question. In our case it is the temperature in the distillation tower. Any deviation from the setpoint means an economic loss which can occur through various mechanisms. We will discuss just one here.

In our case the temperature has a major, nonlinear effect on the product quality. If the temperature would be on target and perfectly constant then the product quality would have a certain average value. However, variations in the temperature will cause the average quality to shift towards a lower value. This deficit requires corrections, e.g. blending in a high quality (and thus costly) component by raising the temperature setpoint—which of course means higher energy consumption.

Secondly, there is also a nonlinear effect between the temperature and the energy input into the tower. Energy savings during time periods where the temperature is below the setpoint are more than outweighed by the higher input needed in the case when it is above the setpoint. The consequence is a higher-than-optimal energy consumption. The fact is, the larger the variance of the temperature, the higher the economic loss.

Studies (which were done at a time when the oil price was $40 per barrel) have shown that smoother temperature control can save about €1,000 per year per 1 MW heat capacity—just from energy savings. Thus for a big exchanger or a furnace the tuning and optimisation effort and the cost for a tuning tool pays
back in shortest time, just for one loop!

What have we learned?
So, what is the moral of this story? A key prerequisite is the quality of the process information, and thus the ability of the local staff to make a solid test or to select suitable operating data. That implies that they must have proper, practical education and experience, even if the rest of the analysis work is being done off site. If the right tools are available on-site then the plant staff can solve the problem themselves in a short time. But
if they don’t have the tools on-site, then today’s Internet communications means they can easily make up for such a deficit.

Of course, in our case a quick solution was possible because we were approached by a long-time customer and there were no questions on payment terms or confidentiality issues. However in cases where the help seeker and the help provider do not know each other, these issues can be addressed quite quickly. The real question is: how much money is lost by the plant during the time needed to sort out its problems compared to the price of a suitable tool?

This case has shown us that remote help can be of great value where local resources lack either the time or the proper tools, or where travel cost and time are too expensive in relation to the total effort needed. As a consequence of this experience, we have included remote assistance in our service offer.

Hans H. Eder; ACT, hans.eder@act-control.com, www.act-control.com


Print this page | E-mail this page