Top Ten Limitations – Noise

Loops with the highest potential for performance suffer the most from noise. Measurements and final control elements (valves and variable speed drives) with the best ability to respond to small changes are most at risk. If the noise is more than the measurement resolution-sensitivity-backlash, the loop will see the noise as a disturbance. If the noise multiplied by the controller gain or amplified by the controller rate setting creates PID output fluctuations greater than the final element resolution-sensitivity-backlash, the valve will move self-inflicting a disturbance that possibly upsets other loops. Thus, we have the irony of noise rarely being an issue for pneumatic loops, mechanical sensors, and on-off valves posing as throttling valves. Do we want to go backwards in time or do we want to address the growing noise problem as our automation system capability improves? Do we want to go back to transmitting rotameters (0.5% sensitivity) or go forward with Coriolis meters (0.002% sensitivity)? Do we want filled system temperature transmitters (0.5% sensitivity) or RTD transmitters (0.02% sensitivity)? Do we want level float gages (0.5% sensitivity) or radar level gages (0.02% sensitivity)? Do we want rotary isolation valves with a piston actuator and spool positioner (5% sensitivity) or sliding stem globe valves with a diaphragm actuator and digital positioner (0.1% sensitivity)?

Loops with the best dynamics (smallest deadtime to time constant ratio) can have an exceptionally high controller gain. For example the controller gain that would cause instability is over 100 for most level and reactor temperature loops. Even if we use less than half the gain for stable maximum disturbance rejection, we are talking about gains of 40 or more. The higher the controller gain, the lower the integrated and peak errors. In fact the error for vessel temperature and level control can be less than the measurement sensitivity if the controller gain is used that the loop is entitled to. The limit to how much gain is used is measurement noise and the possible upset to other loops from large rapid movements of the control valve. For reactor control, the movement of the utility valve may be disconcerting but not as important as tight temperature control. For distillate level control by manipulation of reflux, tight level control enforces the material balance so the temperature controller manipulating distillate can do its job, and provides some internal reflux compensation for cold rain storms on a hot summer day.

All loops have noise on the measurement. You may not see it on trend plots due to scale ranges, measurement resolution-sensitivity-backlash, and data historian compression.

The best solution is to eliminate the source of the noise. Use best practices for shielding, grounding, and separation and the best type of heater power controller and variable frequency drive to eliminate electrical noise. Use wireless pH transmitters to eliminate the spikes from the starting of agitators and pumps. Improve the mixing in reactors and neutralizers and locate sensors in the middle of the pipe and sufficiently downstream of inline equipment to eliminate temperature and concentration gradients. Eliminate bubbles from hitting sensors.

Given that you are stuck with noise, you can add a filter or transmitter damping to reduce the noise level to be less than the valve resolution-sensitivity-backlash. For wireless measurements, the noise should be less than the “default trigger level” (wireless sensitivity setting for exception reporting) to increase battery life.

I favor the use of a noise band for sensitive measurements and valves and the wireless PID. The process variable (PV) is not updated for the PID unless the change in the PV since the last update exceeds the noise band. The raw PV with the noise should be available for process and measurement assessment and diagnostics. The noise band introduces a deadtime, but this additional deadtime is on the average less than the deadtime created by the filter time lag in loops with a large process time constant. I wish wireless transmitters allowed a very large default update rate so I could simply use the “default trigger level” to function as a noise band. When used in conjunction with the wireless PID, the noise band eliminates the limit cycle from valve resolution-sensitivity-backlash and split range points and interactions between valve position controllers and process controllers. An integral deadband setting can do some of this but you need to know the valve resolution-sensitivity-backlash that is all over the map varying with stroke and age. You would then also need to use a notch gain setting to prevent valve dither from controller gain or rate action.

The use of a noise band creates an offset. Offsets do not wear out valves or create process variability. The exact match of the process variable to the setpoint is an obsession enhanced by digital displays that does more harm than good as discussed in Can’t Get No Satisfaction and Are We Misleading Our Operators?

The noise band should be set slightly larger than the peak to peak noise amplitude. The noise band setting is in reality a PID sensitivity setting that allows the PID to ignore and not spread and amplify meaningless variability. The enhanced PID for wireless (PIDPlus) will inherently prevent a limit cycle but with the traditional PID, the integral deadband should be set equal to the noise band to prevent a limit cycle from the PID sensitivity.

The stomping of your cowboy boots to white noise and the failure of a valve from dithering and cycling shutting down a plant are nights to remember.