What are the Alternatives to Reduce Noise?

July 5, 2012
Nearly all loops have noise. Whether you see the noise is a matter of amplitude, frequency, resolution, and data compression. Signal filters and damping can reduce noise but the penalty is a measurement lag and deterioration in the ability to reject disturbances. Here are alternatives for noise reduction to decrease process variability and increase valve life.

Nearly all loops have noise. Whether you see the noise is a matter of amplitude, frequency, resolution, and data compression. Signal filters and damping can reduce noise but the penalty is a measurement lag and deterioration in the ability to reject disturbances. Here are alternatives for noise reduction to decrease process variability and increase valve life.

If the noise is really fast, signal filtering in the I/O system, minimum damping in transmitters, and sensor lags may keep the oscillations within resolution limits and exception reporting and data compression settings. If the noise is really slow, the fluctuations will be hidden in the response to disturbances. Noise frequencies close to the natural frequency of the loop will be amplified from resonance. Let’s narrow the consideration to noticeable noise with frequencies less than the natural frequency.

The Control Talk Blog “What is the best PV Filter and Transmitter Damping Setting” provides guidance on how to set the filter and damping adjustments to keep fluctuations within the deadband of the control valve or variable speed drive. The blog also provides a simple equation in a book excerpt to estimate the attenuation and consequential deception if the measurement lag becomes larger than the primary process lag.

The Control Talk Blog “What is the best PID Execution Time” discusses how to improve the signal to noise ratio by insuring the true process change reported is larger than the quantization error (e.g. analog to digital converter and sensor resolution). The PID execution time is increased until the maximum PV rate of change computed online per last week’s blog multiplied by the PID execution time is larger than quantization error.

Of course, you should improve the sensor type and location to reduce measurement noise as discussed in the Dec 01 2012 Modeling and Control blog “How to Succeed – Part 4”.

What else can one do? The most common alternative for processes with slow rates of change at the chemical companies I spent my career with was the simple insertion of a rate limit function block on the PV signal. The rate limit was set to be faster than the fastest actual process PV rate of change (identifiable online per last week’s blog). The rate limit may be set to be faster in one direction than the other. PV rate limits were particularly effective for level loops since signal lags have a greater detrimental effect on integrating processes and these loops are prone to noise due to turbulence. In one case rotating gear teeth of a finisher passed through the beam path to a nuclear level detector. There is essentially no delay or detrimental effect on the PID response to true processes changes. There is a critical implementation detail that can make you a hero or a villain. The rate limit must be turned off whenever the controller is not in auto or is in output tracking so that operator, sequential, startup, and process actions are immediately seen.

For severely deadtime dominant processes (process time constant << loop deadtime), the PID tuning becomes nearly integral only control because the PID reset time and gain are proportional to the process time constant in most tuning methods.  Integral action averages out the positive and negative fluctuations about the mean value. Severe deadtime dominant loops are more prone to process noise because the process time constant that would filter out process noise is so small. Incorrectly tuned loops on these processes will propagate noise downstream.

Noise can originate from inadequately sized actuators, and excessive stem, seal, or seat friction.  Erratic control valve response becomes process noise and may be seen in the PV response for fast or deadtime dominant processes. The solution is to increase the actuator thrust or torque and if necessary use a lower friction valve design. In some cases, the solution could be as simple as changing the internal seal or stem packing and not over tightening the packing.

Overshoot can be caused by overly aggressive digital positioner tuning (high gain and rate settings). Limit cycling can be caused by the introduction of integral action in the positioner. The solution is to tune the positioner for a smooth response and remove integral action or set an integral deadband to stop limit cycles.

As a side note I am not sure why integral action was added to positioners in the last decade. The positioner gain is so high (e.g. > 50) that the offset between desired and actual positioner is negligible plus any offset is eliminated by the process controller in automatic.  I suspect the addition of integral action was more gamesmanship from the comparison of step test results looking at just the valve with no process feedback correction. The fact the actuator pressure response is an integrator insures a limit cycle from backlash (deadband) when integral action turned on. Some digital positioners have an inner feedback loop on pressure to make the pressure response self-regulating. Here limit cycling doesn’t occur from backlash until the process PID goes into automatic with integral action (2 or more integrators cause a limit cycle from deadband).

A threshold sensitivity limit also known as a dead zone can be used to screen out insignificant changes in the PV. As a minimum this limit should be set to screen out quantization errors. Care must be taken to not set the limit too large because a deadtime is introduced that is on the average ½ of the limit divided by the PV rate of change. For slow processes the increase in controller gain from the reduction in noise actually reduces deadtime from the limit and valve stick-slip and deadband by increasing the PV rate of change from greater proportional mode action. The primary limitation to higher controller gains in processes with large process time constants or slow integrating process gains is the amount of measurement noise. Slide 134 in the presentation ISA-Edmonton-2012-Effective-Use-of-Measurements-Valves-PID-Controllers-Rev1.pdf has equations to compute whether the effect of a sensitivity threshold limit is beneficial or detrimental. The PID integral deadband should be set equal to the limit or an enhanced PID for wireless used to prevent limit cycles. The InTech Oct 1996 article “Eliminate noise by using self-adjusting dead zone” shows how the limit can be adapted online.

To reduce noise, the rate time should be decreased or the derivate filter time expressed as a fraction of the rate time should be increased. This is similar to increasing the lag in a lead-lag block. In DeltaV the alpha parameter default is 0.125 so the derivative filter time is 1/8 the rate time.

Finally, higher order and more intelligent filters can be designed to provide more attenuation with less measurement lag or delay. This methodology can be especially effective for noise inherent in a sensor measurement (e.g Coriolis or vortex meter). The algorithms do not work as well for chemical processes since noise characteristics vary with time, phases, equipment and operating conditions, interactions, and disturbances. 

About the Author

Greg McMillan | Columnist

Greg K. McMillan captures the wisdom of talented leaders in process control and adds his perspective based on more than 50 years of experience, cartoons by Ted Williams and Top 10 lists.