A standard PID uses one fixed proportional gain no matter how far the process is from setpoint. Nonlinear gain breaks that assumption: it lets the controller be gentle when the measurement is already close to target and firmer when the error grows. Combined with an error deadband, a small zone around setpoint where the controller simply stops reacting, this technique quiets down loops that would otherwise nudge the valve back and forth all day chasing measurement noise.
Nonlinear PID gain in one line: Nonlinear gain in a PID means the proportional gain changes with the size of the error, typically low near setpoint and higher farther out. An error deadband is a related idea in which errors smaller than a set threshold produce no output change at all. Both reduce needless valve movement and cycling on noisy or lightly loaded loops without giving up firm control when a real upset arrives.
In a linear PID the controller multiplies the error by a single constant gain. That is elegant, but it means the controller treats a tiny wiggle of a few tenths of a percent with exactly the same aggressiveness, per unit of error, as a large genuine deviation. On a noisy measurement the small wiggles are mostly sensor and process noise, not information worth acting on, yet the linear controller dutifully translates every one into a small output move and, through it, a small valve stroke. Multiply that by thousands of cycles a day and you get wear with no control benefit.
Nonlinear gain reshapes the response so the effective gain depends on how big the error is. The most common form uses a low gain inside a defined region around setpoint and a higher gain outside it, sometimes called gap gain because there is a low-sensitivity gap centered on target. The opposite shape, occasionally called notch gain, does the reverse and is used where fine control right at setpoint matters more than reacting to large excursions. Either way the controller is no longer a single straight line from error to output but a curve or a set of line segments.
The payoff is that near setpoint, where the residual error is mostly noise, the low gain barely responds, so the valve holds still. When a real disturbance drives the error outside the low-gain region, the higher gain takes over and the loop reacts with authority. You get the calm of a detuned loop when things are quiet and the muscle of a tightly tuned loop when they are not, which a single fixed gain can never deliver at the same time.
An error deadband is the simplest nonlinear behavior: a zone around zero error in which the controller freezes its output. As long as the measurement stays within, say, a small band of setpoint, the controller does nothing at all; only when the error crosses the band edge does it resume acting. This is the limiting case of gap gain with the inner gain set to zero. It is extremely effective at stopping valve chatter, because the actuator simply stops moving while the process floats harmlessly near target.
It is important not to confuse this error deadband, which lives inside the controller and acts on the error signal, with a control deadband or hysteresis applied to the process variable or to the actuator itself, such as valve stick-slip or an on-off dead zone. The error deadband is a deliberate control feature you configure to reduce action; actuator deadband is usually an unwanted mechanical imperfection you fight against. They can even interact: a well-chosen error deadband keeps the controller from constantly trying to overcome valve backlash, which both saves the valve and avoids the limit cycle that backlash plus an eager controller tends to produce.
The tradeoff is honest and worth stating. Any deadband gives up a little steady-state accuracy in exchange for far less movement, because the controller tolerates errors up to the band edge without correcting them. On a loop where being within a small band of setpoint is good enough - many level, some pressure, and buffer loops qualify - that is a bargain. On a loop that must sit exactly on target, such as a tight custody or blend control, a deadband is the wrong tool and a smooth low nonlinear gain is usually the better choice.
The loops that benefit most are noisy, lightly loaded, or slow-moving ones where constant small corrections buy nothing but wear. Flow and pressure loops on liquids with turbulent, hunting measurements, tank and vessel level loops that only need to stay in a range, and any loop driving a valve or actuator whose maintenance cost is high are all good candidates. Reducing valve travel directly extends packing and seat life, cuts air or hydraulic consumption on the actuator, and lowers the odds of a stick-slip limit cycle wearing a groove in a trim.
A frequent symptom that points toward nonlinear gain is a valve whose position trend looks like a fuzzy band even when the process is on target and nothing is upsetting it, paired with a measurement that is visibly noisy. The likely causes are a fixed gain that is appropriate for handling upsets but too eager for the quiet periods, plus a measurement with real high-frequency content. The diagnostic path is to trend valve travel and the process variable together over a calm period, confirm the valve is moving without the process actually needing correction, and then decide between more measurement filtering, a small error deadband, or a gap-gain profile depending on how much steady-state accuracy the loop can spare.
Because the nonlinear behavior is configured and executed in the field controller, a cloud SCADA platform such as Merobix is where you see whether it is working. Trending valve position or output alongside the process variable makes reduced travel obvious after the change, and totalizing valve movement over a day or a week turns wear reduction into a number you can report. Monitoring also guards the downside: if a deadband is set too wide, the trend will show the process drifting persistently to one edge of the band without correction, which is the signal to tighten it. That before-and-after visibility is what turns a nonlinear-gain change from a guess into a documented improvement.
Error-squared control is one specific way to make gain rise with error, by multiplying the error by its own magnitude so the effective gain grows continuously as the deviation widens. Nonlinear gain is the broader family of techniques for making gain depend on error size, which includes error-squared but also gap gain, notch gain, and piecewise segmented gains. Think of error-squared as a particular curve within the general nonlinear-gain toolbox.
It trades a little accuracy for much less valve movement. Because the controller ignores errors smaller than the band, the process can float anywhere inside that band without correction, so you give up tight positioning right at setpoint. On loops that only need to stay within a range that is an excellent deal, but on loops that must sit exactly on target a deadband is the wrong choice.
Not quite. Gain scheduling changes the tuning based on an operating condition such as production rate, valve position, or product grade, so the same error can produce different action depending on where the plant is running. Nonlinear gain changes the response based on the size of the current error itself, regardless of operating point. They can be combined, but they solve different problems.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.