A self-regulating loop, given a step in its output, settles to a new steady value, and most tuning rules are built on that behaviour. An integrating process does not settle; give a tank level loop a step in output and the level ramps forever. That single difference means the standard tuning rules do not apply, and using them can give a badly behaved loop. This guide explains why integrating loops need their own approach, the integrator-plus-dead-time model that describes them, and the lambda and IMC tuning forms adapted for integrators.
Integrating-process tuning in one line: Tuning an integrating process means using tuning rules built for loops that ramp rather than settle, because a step in output produces a continuously changing output with no steady state. The process is described by an integrator-plus-dead-time model rather than the first-order-plus-dead-time model used for self-regulating loops, and lambda or IMC tuning has specific forms for integrators that set the gain and reset from that model and a chosen closed-loop speed.
The defining trait of an integrating, or non-self-regulating, process is that it has no natural resting point for a given output. The classic example is level in a tank with a fixed outflow: if inflow is even slightly greater than outflow, the level rises and keeps rising; if slightly less, it falls and keeps falling. Only when inflow exactly equals outflow does the level hold, and any output that does not achieve that exact balance sends the level ramping away without limit. Pressure in a fixed volume and some temperature and composition situations behave the same way. There is no self-correcting tendency that pulls the process back toward a steady value.
This is fundamentally different from a self-regulating process, which does have a natural resting point. Change the output on a self-regulating loop, such as a flow or a well-behaved temperature, and the process moves to a new steady value and stays there, because the process itself resists further change. That settling behaviour is what the common tuning rules and process models assume, and it is what lets them characterise a process by a steady-state gain, how far it settles for a given output change. An integrating process has no steady-state gain in that sense, because it never settles, so those rules have nothing to attach to.
The practical danger is that applying self-regulating tuning rules to an integrating loop tends to produce sluggish or oscillatory control, because the rules are fitting a model the process does not obey. A loop tuned as if level would settle, when in fact it ramps, is being tuned on a false premise. Recognising early that a loop is integrating, and reaching for the appropriate rules, is therefore the first and most important step in tuning it well. The test is simple: step the output in manual and watch whether the process settles or ramps.
Just as self-regulating loops are commonly described by a first-order-plus-dead-time model, integrating loops are described by an integrator-plus-dead-time model, and getting the right model is the foundation of the tuning. The integrator-plus-dead-time model captures two things: an integrating gain, which is the rate at which the process variable ramps per unit of output change, expressed as a slope rather than a settling distance, and a dead time, the delay before the process begins to respond at all. These two parameters describe an integrating loop the way the gain, time constant, and dead time describe a self-regulating one.
Identifying the model from a test is correspondingly different. For a self-regulating loop you step the output and read the final settled change and the time to get there. For an integrating loop you step the output and measure the slope of the resulting ramp, since there is no final value to read, and the dead time before the ramp begins. The integrating gain is that slope divided by the size of the output step. Because the process never settles, the test has to be watched carefully and often kept short, and the output may need to be stepped back to keep the process variable within a safe range while the ramp is being measured.
The integrating gain and dead time from this model then feed the tuning calculations directly, which is why an accurate model matters. A poor estimate of the ramp slope or the dead time propagates straight into the controller gain and reset, so time spent getting a clean integrating-plus-dead-time fit is repaid in a well-behaved loop. This modelling step is exactly where integrating-process tuning diverges in procedure from self-regulating tuning, and it is the step most often skipped by engineers who try to force a settling-based test onto a ramping process.
Lambda tuning and the closely related IMC approach both have specific forms for integrating processes, and these are what turn the integrator-plus-dead-time model into controller settings. As with self-regulating lambda tuning, you choose a closed-loop time constant, the lambda, that sets how fast you want the loop to respond, and the method then computes the controller gain and reset time from that lambda together with the integrating gain and dead time. The larger the lambda, the slower and more robust the loop; the smaller the lambda, the faster and more aggressive, subject to the dead time. The appeal is the same as for self-regulating lambda tuning: a single, meaningful knob, the desired speed, with the rest following from the model.
There is a subtlety specific to integrators that the integrator forms are built to handle: because the process ramps, an integrating loop can be prone to a slow, rolling oscillation if the reset action is not set with care relative to the loop's response, and the integrator-specific formulas are designed to avoid that. This is one reason it matters to use the integrating form of the rule rather than borrowing the self-regulating form, since the self-regulating form does not account for the ramping behaviour and can leave the loop oscillatory. Averaging duties on buffer vessels, discussed elsewhere, push this even further by choosing a very slow lambda on purpose so the level is allowed to swing while outflow stays smooth.
In a cloud SCADA operation, integrating loops are everywhere, since level is one of the most common measured variables across tanks, separators, and vessels on distributed field assets, and getting their tuning right at scale is easier with good remote trending. The step test needed to identify the integrating model, watching the level ramp after an output change, can be reviewed on the trend history rather than requiring someone at the vessel, and the resulting loop behaviour, whether it holds level cleanly or drifts into a slow oscillation, is visible remotely. Being able to see many level loops across an operation lets an engineer confirm they were tuned with integrating rules rather than mis-tuned with settling-based ones, and catch the slow rolling oscillation that is the tell-tale sign an integrating loop was tuned as if it were self-regulating.
Put the loop in manual and step the output, then watch the process variable. If it moves to a new steady value and settles, the process is self-regulating. If it ramps continuously and does not settle, as tank level does when inflow and outflow are unbalanced, the process is integrating. This simple test tells you which family of tuning rules to use.
Most common tuning rules assume the process settles to a steady value after an output change, and they characterise it by how far it settles. An integrating process like level never settles; it ramps. Applying settling-based rules fits a model the process does not obey, which typically gives sluggish or oscillatory control. Integrating loops need rules based on the integrator-plus-dead-time model instead.
It is the model used to describe integrating processes, defined by an integrating gain, the rate or slope at which the process ramps per unit of output change, and a dead time, the delay before the process starts responding. It replaces the first-order-plus-dead-time model used for self-regulating loops, and its parameters feed directly into integrating forms of lambda or IMC tuning.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.