Improve control loop performance

Specification of control valves doesn’t adequately emphasize the very basic requirement that valve position respond in a timely manner or even at all — leading to process variability.

By Gregory K. McMillan, Emerson Process Management

1 of 3 < 1 | 2 | 3 View on one page

Nearly all control loops in the chemical industry depend upon the manipulation of flow by the use of a final element such as a control valve. It’s generally taken for granted that when a controller changes its output, there’s an actual change in the position of the closure member of the valve (plug, ball or disk). However, the specification of control valves doesn’t adequately emphasize the very basic requirement that the position respond in a timely manner or even at all — and this has resulted in shortcomings that introduce variability into the process.

Before the advent of smart HART and fieldbus positioners, feedback measurements of position were rare because a separate position transmitter had to be purchased, installed and wired. So, the user generally wasn’t aware that differences in valve, actuator and pneumatic positioner design were the source of cycling in the process.

Typically, besides traditional factors such as size and materials of construction, control valve specifications have focused on minimizing leakage through the valve at shutoff and emissions to the environment from packing. Too often, to reduce project costs, plants pick on/off valves to address requirements. This can create performance problems that can’t be fixed simply by adding a smart positioner. While installing a smart positioner always is beneficial, an incorrect feedback mechanism in the valve design can give a false indication of performance.

To avoid problems, always consider five basic valve requirements — linearity, dead time, response time, resolution and dead band. They can give crucial guidance and justification for a final element that leads to better control. Rangeability and sensitivity also are important but, as we’ll see, properly meeting the other requirements will address them.

Linear in a nonlinear world

To get on a common basis, we need to define process gain for a self-regulating process as the final percent change in the controlled variable divided by the percent change in valve position. Note that the calibration span of the transmitter for the controlled variable is a factor. Because the changes seen in data historians for process variables are in engineering units, they must be converted to percent of scale. The maximum allowable controller gain is inversely proportional to the process gain.

The process gain for flow is the slope on a plot of percent flow versus percent valve position (travel). The plot should reflect the installed flow characteristic, not the inherent trim characteristic. This accounts for the reduced pressure drop available to the control valve at higher flows because of the increase in pressure drop in the rest of the system from frictional losses and a decrease in pump discharge pressure. The changing valve drop makes an equal-percentage trim more like a linear characteristic and a linear trim more like quick-opening characteristic. The effect increases as the valve pressure drop as a percent of the total system pressure drop is decreased.

Figure 1. Process gain becomes too low when travel of sliding stem valve exceeds 80%. Source: Ref. 1.

Figure 1. Process gain becomes too low when travel of sliding stem valve exceeds 80%. Source: Ref. 1.

In Figure 1, we see the process gain gets too low for travel above 80% of a sliding stem valve. The control loop must make very large changes in position to change the flow. For similar conditions a ball or butterfly with a 60° maximum rotation would see a corresponding excessive loss of sensitivity at about 60% travel, a typical problem for high capacity valves [1].

A linear installed characteristic is particularly desirable for flow and liquid pressure loops. For critical loops, software programs can generate the installed characteristics with normal fluid data used for sizing the valve if the system frictional loss and pump curve also are known. You then can set the controller gain per the maximum process gain on the plot. You can obtain better performance by computing the controller gain as a function of controller output per the plot. Of course, this depends upon a fixed installed characteristic, so in general the controller gain is reduced. An adaptive controller is now available that can automatically identify the process gain online for more accurate scheduling of the controller gain and tighter process control [2].

Another option is to put a signal characterization block between the controller output and the command signal for valve travel. The signal characterizer computes the percent stroke needed to obtain a percent desired flow (abscissa from the ordinate of the installed valve characteristic). This signal characterization will decrease the effect of resolution and dead band on the flat portion of the curve because it magnifies the change in signal to the valve. The opposite is true for steep portions of the curve. If control valve positions are maximized as plants are pushed beyond their design capacity, the greater concern is the loss in sensitivity at higher valve positions as shown in Figure 1.

If the pressure drop across the control valve is large compared to the pressure drop in the rest of the system — e.g., as in pressure letdown, reagent, surge and vent valves — the installed characteristic essentially is the inherent characteristic. For an equal-percentage trim, the nonlinearity is extreme (process gain can change by a factor of fifty) because the slope of the characteristic is proportional to flow. If a pH loop directly throttles a reagent valve on a static mixer, this change in slope on the valve characteristic compensates for a change in process gain for pH that is inversely proportional to flow.

For a temperature controller that directly throttles a coolant to an exchanger, the equal-percentage characteristic compensates for a temperature process gain that also is inversely proportional to flow. In either case, once a secondary flow loop is added so there’s a cascade loop of pH to reagent flow or temperature to coolant flow, we’re back to the nonlinearity considerations for the flow loop. Major reasons to add these secondary flow loops are to better reject pressure disturbances and provide a more accurate implementation of flow feedforward control [1].

1 of 3 < 1 | 2 | 3 View on one page

Join the discussion

We welcome your thoughtful comments. Please comply with our Community rules.
All comments will display your user name.

Want to participate in the discussion?

Register for free

Log in for complete access.


No one has commented on this page yet.

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