By applying the following guidelines, a control engineer can tell whether a loop is tuned appropriately or if it may be a source of performance problems, and then make or recommend the appropriate changes.
Note: Tuning is given in the format “gain/integral/derivative,” where gain (or proportional control) is on a percentage-of-span basis, integral (or reset) is in minutes per repeat, and a derivative is in minutes. The term “valve” may refer to any final control device.
1. Don’t waste time on flow loops. Set the tuning to 0.25/0.25/0.0—and move on. Spending more time is usually pointless because process gain will normally vary with valve position itself. The relatively small gain—overall ideal flow controller gain is 1.0—reflects operation at low valve opening, where actual flow response is usually greatest. At higher valve openings, the short integral time will make up the difference, typically bringing flow to setpoint in a matter of minutes if not seconds, which is usually much faster than the rest of the process. This is often the most salient consideration when tuning any loop—being timely in the context of the process while minimizing the potential for control instability. Large, slow valves may need a longer reset; occasional loops may need an even smaller gain, especially if they’re operating at very low valve opening.
2. Don’t waste time on level loops. Set the tuning to 1.0/response time/0.0 – and move on. Response time, in level control context, is the approximate time it would take for the vessel, beginning at setpoint (typically 50%), to completely fill (or empty), in response to an average disturbance, in the absence of any control response. Typical refining process response times are 1 to 3 minutes for boots, 5 to 15 minutes for drums, and 30 to 60 minutes (or longer) for tanks.
Where severe disturbances are possible, or where potential consequences of overfill or underfill are severe, use a larger gain instead of a shorter integral (excessive integral action on level loops is the most common source of process oscillations). The level loop gain should rarely be more than 2.0. This level tuning guideline applies whether or not the level is cascaded to flow. The level is an integrator and is often nonlinear so that proportional control provides the most reliable overall long-term performance. There are several common alternative strategies, such as gap gain, surge control tuning, and dynamic optimization of vessel inventory under model-based control. However, none of these approaches add value or are more reliable.
3. Don’t waste time on temperature, pressure, composition, etc., unless they are already cascaded to flow (or sometimes pressure). If not cascaded, implementing a cascade is the next best step to take. The cascade will linearize the process response, thereby allowing optimal reliable tuning across the full operating range. Where a flow meter is not available, it may be possible to characterize the controller output based on the valve characteristic; however, this technique carries several uncertainties and vulnerabilities and is a distant second-best practice.
4. Invest time on pressure, temperature and composition loops that are cascaded to flow (or sometimes pressure). The cascade structure, in addition to linearizing the process response, allows loop gain to be accurately determined from process data at two or more operating points. (Exercise: Try this calculation on a common heater temperature/fuel gas flow cascade.)
5. Set gain to roughly ½ to ¾ of the observed gain to maintain long-term stable and reliable performance due to potentially changing process conditions, which can be expected to occur in most loops for many reasons. Set the integral equal to the process response time that is determined by using a step test, process experience, or historical data—erring on the long side. There is a significant long-term performance benefit to a slightly smaller gain and longer integral than traditional error-minimization or quarter-amplitude decay criteria—such as Zeigler-Nichols—will recommend, up to the point of exceeding the process timeliness consideration.
6. Do not use derivative action. The derivative is a way to reduce total error by using more aggressive gain and reset and then relying on derivative to put on the brakes. This is like speeding up and then braking hard as you approach a stop sign, to get there slightly sooner. The derivative action is inappropriate in most industrial process control applications when observing process speed limits, minimizing overshoot and oscillation, and safeguarding process stability at all times are paramount. Many modern tuning software packages that routinely recommend non-zero derivative settings are inexperienced. Model-based multivariable control algorithms do the same thing to achieve “error minimization” or “profit maximization,” but the resulting aggressive behavior, as in single-loop control, usually needs to be back out to provide long-term reliable performance. This caveat is in addition to the traditional concern about large derivative bumps due to transmitter noise or instability.
7. Controller gain is heavily dependent on the span. For example, a pressure controller with a span of 0 to 1,000 per square inch gauge (PSIG) (using a modern smart transmitter) will need a gain that is 10 times larger than the same controller spanned for 900 to 1,000 PSIG (as such loops were often designed in the past, to have greater accuracy within the actual operating region) to provide the same control response to a given error. Without understanding this difference, there is a reluctance to accept the larger, but functionally equivalent, gain values. As a loose guideline, the gain is typically 1.0 for every 100 to 200 units of the span for large-span temperature and pressure controllers.
8. Be aware of loop interactions. When the action of one loop strongly impacts other loops (s), the user will need to decide which loop should be tuned normally and which one(s) reducing gain and increasing reset time. There is often no practical reliable option to tune all loops in a set of interacting loops for the timely response, due to the technical demands of doing so, using techniques such as “decoupling”, and the reality of changing process gains, which undermines any such schemes. The speed guideline for interacting loops has a similar basis to the traditional cascade rule, except in reverse.
For example, the secondary loop(s) should be detuned to be at least three to five times slower than the primary (high priority) loop. This fundamental rule has been historically overlooked and ignored in multivariable control. This frequently results in unstable performance and ultimately contributes to a degraded performance unless all the involved models remain highly accurate, which is rarely realistic.
9. Be bold with gain and cautious with reset when tight control is necessary. There is a perception that too much gain may cause cycling, and that since reset is in units of time, a shorter setting might bring faster control. In reality, the proportional control action is instantaneous and too much reset, especially in combination with too little gain, is the most common cause of process oscillations. So where tighter control is necessary, use more gain (up to the limit of actual average process gain) and tune integral more accurately (being careful not to make it less than actual process response time).
10. Understand when to use, and when not to use feedforward. Feedforward can be advantageous when a major disturbance is well understood when its model (gain, response time, and deadtime) does not change significantly in time for any reason, and where it is warranted to avoid hitting a hard process limit or to capture large earnings or avoid large losses. When these criteria are not met, especially if model dynamics (response time and deadtime) are not reliably known, feedforward should be avoided. Every feedforward model comes with engineering, reliability, and maintenance costs, which the wholesale use of feedforward in model-based multivariable control has emphasized.
Successful loop tuning—defined as minimizing rework, detuning, returning, remodeling, etc.—relies as much on an understanding of process operation performance criteria as knowing traditional single-loop tuning tools and methods such as Ziegler-Nichols. When the process operation perspective is neglected, loop tuning often is caught in a cycle of rework (Figure 1), rather than behaving as it should.
Another important takeaway is that model-based control remains subject to most of the same tuning caveats as traditional loop tuning such as the effects of variable process gains, the implications of interactions, discretion regarding feedforward, and leaving level controls in the base-layer with robust tuning to maximize process reliability.
Originally, it was thought that model-based control transcended most of these concerns, but experience such as degraded performance, model maintenance, and short lifecycle, has revealed how these principles still apply. Using these guidelines can help break the rework cycle, increase success rate, and lead to years of reliable maintenance-free process control performance for most loops.