Measurement Attenuation and Deception Tips

Dec. 2, 2014

Sensor lags, transmitter damping, and PID signal filters can make oscillations look better but is this really a good a thing? Here we look at how these dynamics affect what you see and how much of a problem it can be. We also offer means to find out what the process variable is actually doing.

Sensor lags, transmitter damping, and PID signal filters can make oscillations look better but is this really a good a thing? Here we look at how these dynamics affect what you see and how much of a problem it can be. We also offer means to find out what the process variable is actually doing. A time constant can be beneficial or detrimental. A single large time constant in the process can reduce the variability of process inputs to the point of being negligible in the process output. This effect is rarely included in the analysis of potential benefits in reducing variability in upstream loops. For a back mixed liquid volume, the process time constant is essentially the residence time (volume/flow). This time constant is so large that limit cycles and damped oscillations and peak errors from control loops disappear. Furthermore, the process time constant can greatly slow down step disturbances giving the PID ample time to correct for them before there is an appreciable error. Plus, the maximum allowable PID gain is proportional to this time constant for near integrating process tuning rules. The downsides of a large process time constant occur for disturbances downstream of the time constant, large setpoint changes, and tuning tests that are looking for a steady state. The process time constant slows down the correction from the manipulated flow and the time to steady state. A feedforward signal with a lead time equal to the process time constant can help. The time to reach setpoint and to complete a tuning test are mitigated by the use of a high PID gain, proportional action on error, and an integrating process tuning test that needs to only see the initial ramp rate.

These same benefits of a large time constant in terms of slowing down disturbances and attenuating oscillations from the process inputs can mislead practitioners into using time constants in the measurement resulting is a delayed and attenuated view of the real world. If the measurement time constant becomes the largest time constant within the loop, the deception is incredibly destructive. The amplitude of an oscillation and the peak error from a disturbance decrease and the PID gain can be increased (if the reset time is less than 4 dead times) as the measurement time constant is increased. Note that the equations for peak and integrated error and tuning do not know where the time constant is in the loop. As the largest time constant increases, the PID gain can be increased and the observed errors decreased. Equation 6.7a or 6.7c in Effect-of-Measurement-Time-Constant-Equations needs to be used to convert the filtered peak error as measured to the actual process variable errors. Four actual examples show the spectrum and severity of the deception.

An automation engineer presenting his paper at an ISA Conference said he almost was not allowed to do the paper because his company thought the technical advantage he discovered was so important it should be kept secret as a proprietary knowledge. The presenter had increased the measurement time constant to be by far the largest time constant in the loop. Since he was only trending the filtered measurement, the amplitude of the observed variability was drastically reduced.

A biochemist partially withdrew a temperature sensor in a bioreactor thermowell. The considerable resulting air gap in the tip made the temperature trend smoother. He was so proud he showcased his discovery and decided to run all of his batches this way.

A process engineer noted that a recently installed temperature sensor in a massive block of metal on the extruder outlet showed a dramatic reduction in the temperature variability. The other extruder temperature installations were accordingly modified to move the temperature sensors from the polymer melt to a block of metal. The trend charts looked better but operations eventually realized something was wrong when customers complained about product quality.

The trend charts of temperatures during the startup of a new plant were incredibly smooth but plant performance was horrible. Upon removal of one temperature sensor, sand was found in the tip of the thermowell. The E&I construction crew had installed open thermowells before the piping was sand blasted.

Even seemingly small measurement time constants can be problematic. Figure 6.3 in Effect-of-Transmitter-Damping-on-Surge-Detection shows that a time constant of 1.7 seconds will hide compressor surge oscillations that typically have a period of 1 to 2 seconds. Several surges will have occurred before the suction flow has dropped enough to indicate a problem. Each compressor surge cycle reduces compressor efficiency from high axial thrust, radial vibration, and temperature damaging seals and the rotor. The compressor may trip before the anti-surge control system opens the vent or recycle valve sufficiently.

Measurement time constants including damping adjustments in pressure and differential pressure transmitters should be minimized (e.g. < 0.2 sec) for compressor, furnace, and liquid polymer pressure control. All of the transmitters in the warehouse should have their damping settings minimized. One automation engineer lamented that the transmitters from one supplier had a default damping setting of 1.2 seconds. He found he had to reduce the setting to 0.2 seconds to prevent a compressor trip but whenever a transmitter was replaced on the weekends or night shift, a compressor would shut down.

People are not the only source of an excessive measurement time constant. A new clean glass pH electrode on a static mixer or inline pH control system has a measurement time constant larger than the process time constant. As the electrode ages or gets coated, the pH trends may look smoother.

Temperature sensors in fluidized gas reactors, reformers, and furnaces have a measurement time constant from the thermowell that is larger than the process time constant for the gas volume, which has a small residence time and little back mixing. As thermowell design and installation gets worse, the peak temperatures from hot spots and disturbances appear smaller.

The key to knowing whether a measurement time constant is too large is spotting the increase in dead time and settling time. A slow reset cycle (oscillation that is about 10 times the dead time) may develop if the reset time was set based on a fast sensor.

The installation of faster measurement can be discredited because the amplitude of the variability on a trend chart will be larger after the improvement in measurement design or installation. Such was the case for the installation of a much faster pressure transmitter on a phosphorous furnace. Even though the number of electrode seal blows and furnace trips had decreased, operations were concerned about the appearance on the trend charts. Fortunately, the old transmitter was kept for indication only that showed the actual process variability had decreased. The lesson here is to keep the old measurement or make the calculation of the filtered measurement based on the old measurement time constant using Equation 6.7c to show a “before” and “after” trend.

Even if attenuation is not a factor, a slow sensor can appreciably increase the integrated absolute error (IAE) for an unmeasured disturbance. As rule of thumb, the sensor lag, transmitter damping, or PID signal filter time should be less than 5% of the reset time.