Continuous Level Measurement; Noise Filters

“Ask the Experts,” on, is moderated by noted process control authority Béla Lipták. Béla and his cadre of leading experts in process automation, recruited from among the co-authors of the Instrument Engineer’s Handbook 4th Edition, are “in the box” all month answering process control questions from all comers.

Share Print Related RSS

Q: I want to specify an instrument for continuous level measurement in a small cylindrical process vessel with an internal agitator. The internal diameter of the vessel is 632 mm, height is 800 mm, and the measured range is 700mm. There’s a 25-mm nozzle available on top of the reactor for the level transmitter. The process medium is liquid dimethyl sulfoxide (DMSO), having a dielectric constant of 46.4. The conductivity is 3.10 x 10-6 ohm-1 m -1 at 25°. I initially thought about a radar level transmitter, however the dead band on a typical radar transmitter is about 200 mm to 250 mm, and the transmitter would be in close proximity to the vessel wall (multiple echoes). Would an insulated capacitance probe be an option? The agitator is mounted centrally in the vessel and would not come into contact with a probe.

Christopher Beader
CEL International Ltd, Coventry, U.K.

A: You have at least four options: nuclear, load cell, radar and capacitance, each with its own issues.

Nuclear and load cell are both expensive. Also, nuclear requires a NRC license, and weighing requires flexible joints in the connecting piping.

Capacitance measurement has no dead band, but needs to be calibrated for the fluid in the tank. You also have to check if side-loading forces are high enough to bend or damage the probe. Therefore, anchoring the probe is desirable with both types of probes (capacitance or radar).

You should check if the process fluid includes other liquids with different dielectric constants or with different conductivitivities. Otherwise, the capacitance/RF admittance probes have similar limitations as guided wave radar (GWR) probes when it comes to conductive coating.

Radar is probably the best possibility, but because of the small connection size, you cannot use the non-contact type, because it requires a minimum nozzle size of 1.5 in.

The GWR using a time-domain reflectometry (TDR) design would also be my choice. If the nozzle height is more than 50 mm, the dead band should be no problem, because for near-range measurements, the dead band can be as low as 50 mm, particularly if the liquid is reflective. The dead band requirement is slightly more for a single-rod TDR than for the coaxial, but if you include the false signal suppression function, it should enable closer measurement. Your dielectric constant is high enough for GDR measurement if no coating will occur. Proximity to the wall is no problem with the coaxial version because the signal is contained within the tube.

Agitation and the turbulence it causes can cause vibration and mechanical failure, so the probe should probably be secured to the side or bottom of the tank. Check the side loading on the probe tube and make sure the process fluid is clean and free-flowing. The viscosity of 2.14 cP at 20 ˚C should not be a problem. You should also check if the viscosity is constant, or if it changes during the process cycle. As to the type of probe (see Figure 1), I would use the coaxial, because it is the most efficient. It’s efficiency is similar to that of a 75-ohm coaxial cable. If you expect coating, the single-element probe design is the most forgiving.


Figure 1. Probes for continuous level measurement.

Béla Lipták

A: You may be able to use a guided wave radar with a coaxial probe in the DN25 nozzle; the force from the agitator needs checking. Vortex-shedding vibrations can give a cause of failure, and the level probably will be meaningless while the agitator is running. An alternative non-contact method is to hang the vessel on a weigh-beam sensor. That can keep working while the agitator is running, though piping connections must be flexible.

Ian H. Gibson

A: Capacitance or guided radar should work, but there are things to keep in mind for either. In the case of guided radar, a coax would work better in such a short distance, but is this a clean fluid? If it’s thick or tends toward coating  there will be some issues because there is only a 25-mm (1-in.) opening, and we would need the smaller coax.

Capacitance will allow measurement all the way to the top and bottom.

Kris Worfe

A: A similar application I did a few years ago left me with no option to measure flow and container level, since it was a toxic/lethal fluid with cancer-causing agents. Using tension load cells and signal conditioning modules for loss of weight and metering pump, we were able to measure both level and flow successfully.

G. ”Ram” Ramachandran
Systems Research Int’l, Inc

Q: How do I set the noise filter on a noisy measurement in a digital system? Do I filter out any change that occurs faster than a particular period of oscillation? What should that period of oscillation be? How does the filter setting relate to the sampling period of the loop?

Walter Hopkins

A: Filtering is counterproductive to control, and therefore the minimum amount of filtering is usually best in a control loop. Sampling introduces dead time equivalent to half the sample interval. Sampled measurements are usually also filtered in the A/D converter, with the minimum filtering adding another half-sample interval of dead time. Any further filtering adds that much more dead time. The integrated error of the control loop in rejecting load disturbances with optimum PID settings varies directly with dead time squared.

Greg Shinskey
Process Control Consultant, Wolfeboro, NH

A: If it’s a control loop, then the clock of your process is the dead time, if the controller is tuned to reject disturbances. If the controller is tuned for SP changes, then the clock is the closed-loop time constant.

If it’s measurement only, it depends on what you need to observe.

Some rules of thumb for control loops where the goal is to reject disturbances:

  1. Filtering reduces higher frequencies; they’re useless and create too many changes on the controller output.
  2. Scan time should be smaller than the filter time constant, and the filter time constant should be smaller than derivative time. Derivative time should be smaller than dead time, etc. tscan < t filter < derivative time < dead time< integral time < damped oscillations. Time-scale (valid for ideal and series controller):
    • tscan ~0.1 * Dead Time
    • t filter ~0.2 * Dead Time
    • derivative time ~0.5 * Dead Time
    • integral time ~3 * Dead Time
    • period observed: 
    • natural period ~ 2 to 4 * Dead Time
    • damped period ~5 to 8 * Dead Time
  3. Using software tools, you adjust the filter corner frequency to reduce noise above natural frequency of the closed loop. This is roughly Dead Time/5.
    See my white paper at

Michel Ruel, P.E.
TOP Control Inc.

Share Print Reprints Permissions

What are your comments?

You cannot post comments until you have logged in. Login Here.


No one has commented on this page yet.

RSS feed for comments on this page | RSS feed for all comments