Interested in linking to "Time after time"?
You may use the Headline, Deck, Byline and URL of this article on your Web site. To link to this article, select and copy the HTML code below and paste it on your own Web site.
Greg: Ed Giugliano weighed in with a correct answer to the January Puzzler that asked why the load cell indication would shift at high noon regardless of the time in the batch sequence.
Ed: We had a similar problem during plant startup a few years ago with our limestone day bin. (I work at a power plant and we have a 500-ton pulverized limestone bin that feeds an SO2 control system). It was instrumented with strain gauges for use with a loss-in-weight feed system. However, each sunny afternoon the sun would shine on the bin, heat it up, and expand the steel enough to drive the system bonkers. It did, however, work fine at night, in the morning, and on cloudy days, so it was about a 75% success. We have abandoned this system and gone to volumetric feed control, which works well enough. I think something similar may be happening to the system in your Puzzler--some kind of ambient temperature or radiant heating effect.
Stan: I remember encountering this problem in the 1960s. It happened to one of Greg's former bosses. Some of the problems came from pipes that expanded and contracted due to temperature changes. After a great deal of time and work, they abandoned the system. Just like the man said.
Greg: The weigh tank was on the top of the structure. At high noon, the direct sunlight would cause the vessel to deflect enough to shift the load cell output. Needless to say, the cost to correct the piping and vessel support design “after the fact” was considerable.
Stan: In another application, it was noticed that a shift in the weigh-tank trend record correlated with wind velocity. Here, the solution was cheap. A tarp was hung on the side of the open air structure. This plant had the wind velocity recorded since it used hazardous raw materials.
Greg: We also had a funny but true reply from Mike Larocca.
Mike: The top of the load cell tank happens to provide a great view of the surrounding plant's beautiful scenery. One of the plant mechanics takes his lunch at precisely noon each day, and prefers to sit on the top of the weigh tank while eating it.
Greg: This actually happened on one of my startups.
Stan: Obviously, we need skinnier mechanics or less scenic views.
Greg: If you are doing a flow-rate calculation based on loss in weight, you can end up with huge flow spikes. In fact, to maximize the signal-to-noise ratio, the signal should first be filtered and the calculation time interval should be large enough so that the true change in weight is large. To prevent the addition of a large time delay, the calculation time interval should not be achieved by the use of a large execution time or scan time but by the use of a dead time block to delay the new signal to create an old signal. The old signal is subtracted from the new signal and divided by the block time-delay setting. Note that if the filtering was done after the calculation instead of on the input signals, the attenuation of noise is reduced by the ratio of the block delay time to the input scan time. Needless to say, the flow calculation should be forced to zero to prevent erratic or negative values when the tank is empty or a valve is closed.
Stan: The same principal is important to minimize the noise from any rate of change calculation.
Greg: Most temperature loops cannot use the full extent of derivative action needed to compensate for thermal lags because the combination of small calculation intervals, measurement noise and resolution limits cause valve dither. The solution is to use transmitters with spans narrowed to the region of interest, good wiring practices, and calculation time intervals large enough to ensure the true temperature change is significant. Ideally, this would be achieved by the use of a calculation time delay block. In practice it is achieved by an increase in the execution time, which adds a dead-time to the loop that is on the average one half of the execution time, assuming the output is sent at the beginning of the execution time. The actual time delay varies from zero to the full execution time depending upon when an unmeasured disturbance arrives so that the increase in error from the additional dead time cannot be negated by dead time compensators or model predictive control. This poses an interesting dilemma. These compensators and predictors are less sensitive to modeling errors if the execution time approaches the process dead time but this increases the peak error by about 1.5 and the integrated error by about 2.25 on the average for unmeasured upsets to dominant lag systems. For changes in set points and feedforward or disturbance variables, everything is hunky dory, which only adds to the confusion as to what is the best execution time.
Stan: I have a cat named Dory. Maybe I should get one named Hunky.
Greg: Projects also have shorter execution times that decrease the signal to noise ratio. The only way to avoid this contraction in schedule is to retire. Many people have asked me if I am retired.
Stan: The following questionnaire can help our readers decide if they are retired and qualified to be contributors to our next book tentatively titled “The Art of Retirement.”
Check all that apply:
Greg: We need to close now to make room for advertisements. We will continue with insightful questions from Stan next month, but time for the Puzzler.
ControlGlobal.com is exclusively dedicated to the global process automation market. We report on developing industry trends, illustrate successful industry applications, and update the basic skills and knowledge base that provide the profession's foundation.