1
Learning from Experience

An old Jewish philosopher once said, “Ask me any question, and if I know the answer, I will answer it. And, if I don’t know the answer, I’ll answer it anyway.” Me too. I think I know the answer to all control questions. The only problem is, a lot of my answers are wrong.

I’ve learned to differentiate between wrong and right answers by trial and error. If the panel board operator persistently prefers to run a new control loop that I’ve designed in manual, if he switches out of auto whenever the flow becomes erratic, then I’ve designed a control strategy that’s wrong. So, that’s how I’ve learned to discriminate between a control loop that works and a control strategy best forgotten.

Here’s something else I’ve learned. Direct from Dr. Shinsky, the world’s expert on process control:

• “Lieberman, if it won’t work in manual, it won’t work in auto.”
• “Most control problems are really process problems.”

I’ve no formal training in process control and instrumentation. All I know is what Dr. Shinsky told me. And 59 years of experience in process plants has taught me that’s all I need to know.

LEARNING FROM PLANT OPERATORS

My first assignment as a Process Engineer was on No. 12 Pipe Still in Whiting, Indiana. This was a crude distillation unit. My objective was to maximize production of gas oil, as shown in Figure 1‐1. The gas oil had a product spec of not more than 500 ppm asphaltenes. The lab required half a day to report sample results. However, every hour or two, the outside operator brought in a bottle of gas oil for the panel board operator. The panel operator would adjust the wash oil flow, based on the color of the gas oil.

Figure 1‐1 Adjusting wash oil based on gas oil color

While plant supervision monitored the lab asphaltene sample results, plant operators ignored this analysis. They adjusted the wash oil rate to obtain a clean‐looking product. The operators consistently produced a gas oil product with 50–200 ppm asphaltenes. They were using too much wash oil, and the more the wash oil used, the lower the gas oil production.

I mixed a few drops of crude tower bottoms in the gas oil to obtain a bottle of 500 ppm asphaltene material. I then instructed the panel board operators as follows:

• If the sample from the field is darker than my standard bottle, increase the wash oil valve position by 5%.
• If the sample of gas oil from the field is lighter than my standard, decrease the wash oil valve position by 3%.
• Repeat the above every 30 minutes.

The color of gas oil from a crude distillation unit correlates nicely with asphaltene content. The gas oil, when free of entrained asphaltenes, is pale yellow. So, it seems that my procedure should have worked. But it didn’t. The operators persisted in drawing the sample every 1–2 hours, not every 30 minutes like I had instructed.

So, I purchased an online colorimeter. The online colorimeter checked whether the gas oil color was above or below my set point. With an interval of 10 minutes, it would move the wash oil valve position by 1%. This never achieved the desired color, but the gas oil product was mixed in a tank. The main result was that gas oil production was maximized, consistent with the 500 ppm asphaltene specification.

One might say that all I did was automate what the operators were already doing manually and that all I accomplished was marginally improving an existing control strategy by automating the strategy. But, in 1965, I was very proud of my accomplishments. I had proved, as Dr. Shinsky said, “If it does work on manual, we can automate it.”

LEARNING FROM FIELD OBSERVATIONS

Fifty‐three years ago, I redesigned the polypropylene plant in El Dorado, Arkansas. I had never paid much attention to control valves. I had never really observed how they operate. But I had my opportunity to do so when the polypropylene plant was restarted.

The problem was that the purchased propylene feed valve was too large for normal service. I had designed this flow for a maximum of 1600 BSD, but the current flow was only 100 BSD. Control valve response is quite nonlinear. Nonlinear means that if the valve is open by 5%, you might get 20% of the flow. If you open the valve from 80 to 100%, the flow goes up by an additional 2%. Nonlinear response also means that you cannot precisely control a flow if the valve is mostly closed. With the flow only 20% of the design flow, the purchased propylene feed was erratic. This resulted in erratic reactor temperature and erratic viscosity of the polypropylene product.

The plant start‐up had proceeded slowly. It was past midnight. The evening was hot, humid, and very dark. I went out to look at the propylene feed control valves. Most of the flow was coming from the refinery’s own propylene supply. This valve was half open. But the purchased propylene feed valve was barely open. The valve position indicator, as best I could see with my flashlight, was bumping up and down against the “C” (closed) on the valve stem indicator.

The purchased propylene charge pump had a spillback line, as shown in Figure 1‐2. I opened the spillback valve. The pump discharge pressure dropped, and the propylene feed valve opened to 30%. The control valve was now operating in its linear range.

Now, when I design a control valve to handle a large reduction in flow, I include an automated spillback valve from pump discharge to suction. The spillback controls the pump discharge pressure to keep the Flow Recorder Control (FRC) valve between 20 and 80% open. Whenever I sketch this control loop, I recall that dark night in El Dorado. I also recall the value of learning even the most basic control principles by personal field observations.

Figure 1‐2 Opening spillback to keep the FRC valve in its linear operating range

LEARNING FROM MISTAKES

Adolf Hitler did not always learn from his mistakes. For example, he once ordered a submarine to attack the Esso Lago Refinery in Aruba. The sub surfaced in the island’s harbor and fired at the refinery. But the crew neglected to remove the sea cap on the gun’s muzzle. The gun exploded and killed the crew.

I too had my problems in this refinery. The refinery flare was often very large and always erratic. The gas being burned in the flare was plant fuel. The plant fuel was primarily cracked gas from the delayed coker, supplemented (as shown in Fig. 1‐3) by vaporized LPG. So much fuel gas was lost by flaring that 90% of Aruba’s LPG production had to be diverted to fuel, via a propane vaporizer, to maintain refinery fuel gas pressure.

I analyzed the problem based on the dynamics of the system. I modeled the refinery’s fuel consumption versus cracked gas production as a function of time. The key problem, based on my computer system dynamic analysis, was the cyclic production of cracked gas from the delayed coker complex. My report to Mr. English, the General Director of the Aruba Refinery, concluded:

1. The LPG vaporizer was responding too slowly to changes in cracked gas production from the delayed coker.
2. The natural log of the system time constants of the coker and vaporizer was out of synchronization.
3. A feed‐forward, advanced computer control based on real‐time dynamics would have to be developed to bring the coker vaporizer systems into dynamic real‐time equilibrium.
4. A team of outside consultants, experts in this technology, should be contracted to provide this computer technology.

Six months passed. The complex, feed‐forward computer system was integrated into the LPG makeup and flaring controls shown in Figure 1‐3. Adolf Hitler would have been more sympathetic than Mr. English. The refinery’s flaring continued just as before. Now what?

Figure 1‐3 Unintentional flaring caused by malfunction of the LPG makeup control valve is an example of split ‐ range pressure control

Distressed, discouraged, and dismayed, I went out to look at the vaporizer. I looked at the vaporizer for many hours. After a while, I noticed that the fuel gas system pressure was dropping. This happened every 3 hours and was caused by the cyclic operation of the delayed coker. This was normal.

The falling fuel gas pressure caused the instrument air signal to the LPG makeup valve to increase. This was an “Air‐to‐Open” valve (see Chapter 13), and more air pressure was needed to open the propane flow control valve. This was normal.

But, the valve position itself did not move. The valve was stuck in a closed position. This was not normal.

You will understand that the operator in the control room was seeing the LPG propane makeup valve opening as the fuel gas pressure dropped. But the panel board operator was not really seeing the valve position; he was only seeing the instrument air signal to the valve.

Suddenly, the valve jerked open. The propane whistled through the valve. The local level indication in the vaporizer surged up, as did the fuel gas pressure. The flare valve opened to relieve the excess plant...