What is Truly Important for Measurements and Valves

Sept. 9, 2015

Control is all about dealing with change whether it is disturbances or moving setpoints. The controller can only do as well as the changes it sees and the changes it is able to make. Here we look at what is truly important realizing what is on the specification sheet may not be addressing what we really need.

Control is all about dealing with change whether it is disturbances or moving setpoints. The controller can only do as well as the changes it sees and the changes it is able to make. Here we look at what is truly important realizing what is on the specification sheet may not be addressing what we really need.

First let’s digress to get a perspective. If there were no change, we could set valves to give us the flows on a Process Flow Diagram (PFD). This steady state type of mentality is sometimes propagated from chemical engineering courses focusing on process design. When I first taught a course on process control to chemical engineering students, the question asked was what is the big deal with all this stuff you are talking about in terms of dynamic responses? You just need to set the flows per the PFD.

Correspondingly dynamic modeling fidelity is most often defined to be how well the process flows in the model match the flows in the plant at a given operating point (process setpoint), a carryover in thinking from steady state modeling. For me, dynamic modeling fidelity is how well the change in the process variable, change in measurement and change in valve position match the changes in the plant in terms dead time, time constant and gain (see the last blog for a greater understanding of these terms).  While seeking to match the dynamic response of the model to the plant should be the objective and criteria for fidelity, this is kind of rare in first principle models and even missing in many neural network models. Model Predictive Control (MPC) realizes this is important using powerful identification techniques to model the changes observed for the changes made. This is one of the many reasons why MPC has been successful. Even if you are going to do PID control, I would use the identification software for MPC to get the open loop process gain, primary time constant, secondary time constant and total loop dead time.

The change is best noted and denoted by the details that define the response. Here we first look at problems in the response that can be corrected, such as offset, drift, nonlinearity, span errors, and deadband. While valve and measurement offset is not nice, these effects can largely be eliminated by calibration adjustments and the changes in the setpoint either manually or automatically by the feedback action of upper loops. Drift can be considered to be an offset if slow enough. Smart measurements and valves have reduced offset and drift by an order of magnitude or more. Nonlinearity is more problematic but can be dealt with by internal compensation in transmitters (e.g.  thermocouple and RTD sensor matching),  signal characterization (e.g., titration curve and installed valve characteristic) and adaptive control.  Span errors such as pH electrodes efficiency can be eliminated by calibration and smart instrumentation. Remaining span errors can result in a change in measurement gain but this effect is usually quite small compared to other sources of nonlinearity and could be theoretically corrected based on the size of the change in measurement seen. Deadband does not exist in non-mechanical sensors or variable frequency drives but may be introduced in the configuration setup. Deadband is a common problem in valves due to backlash in connections and linkages. Deadband can be minimized by control valve design and by configuration settings.  Remaining valve deadband can be compensated for by adding a delta equal to the deadband to the change in PID output to jump through the deadband whenever the PID output changes direction by an amount greater than a noise band.

What is left that cannot be readily corrected in the control system design in terms of its response (R) are the five Rs of measurements and valves - resolution, repeatability, 86% response time, rangeability, and reliability. Not commonly recognized is that the effect of Rs besides reliability can be reduced to some extent for measurements by middle signal selection of three separate sensors and transmitters. Middle signal selection can inherently protect against a single failure of any type and does a great job of largely ignoring a slow sensor.

Resolution is the smallest change that a measurement can detect or a valve can respond to.  Once the change occurs, the response is a step whose size is the resolution limit. For a threshold sensitivity limit, the change would match the final change whereas with a resolution limit there is a stair step response that almost by definition would not match the final change.  Valve slip equal to valve stick is essentially a resolution limit.

Repeatability is the difference in final responses for the same change in the process variable for measurements and for the same change in controller output for valves. Noise can appear to be a repeatability error. 

86% response time is the time for a measurement or valve to reach 86% of its final response. For a linear first order approximation, this corresponds to the sum of the dead time and two time constants. The 86% response time is the response criteria per the ISA Standard for Valve Response Testing and should be the standard for sensors particularly pH electrodes because this is the response time of greatest interest in terms of the correction by the controller and because there is a long protracted and variable time to reach a 95% or 98% response time due to various non-ideal effects.  

Measurement rangeability is the ratio of the maximum to minimum process variable where the response at the minimum process variable has the same resolution, repeatability, 86% response time and reliability. Valve rangeability is the ratio of the maximum to minimum flow where the response at the minimum flow has the same resolution, repeatability, 86% response time and reliability.

What can I say about reliability except it is a metric of availability of the measurement or valve (e.g., time to failure)? I would take an expanded view of failure as the inability of a measurement or valve to continue to respond with about the same resolution, repeatability, rangeability, and 86% response time as it normally does.

While you should mind your Ps and Qs (Process Variables and Quality Variables) to achieve the best values of these, you need to pay attention to the 5 Rs in the response of measurements and valves to change (resolution, repeatability, 86% response time, rangeability, and reliability). Unfortunately, these rarely appear on specification sheets.