Control valve hunts in AUTO but runs smoothly in MANUAL

This is one of the most useful observations you can make about a loop, and it is worth understanding exactly what it proves.

In manual the controller stops moving the output. If the cycling stops too, then the cycle was travelling around the feedback path: controller, valve, process, measurement, back to controller. Something in that path is generating it. If the cycling had continued, the cause would have been outside the loop entirely.

So this observation eliminates external disturbances, interacting loops, and upstream process swings in one step. What remains is the controller, the valve, or the measurement — and those three can be separated with one more look at the trend.

Trend showing valve stiction: the controller output ramps smoothly while the process variable moves in abrupt steps, producing a limit cycle that does not decay. A sawtooth or square-cornered cycle on the process variable against a smooth, slowly ramping output. The cycle amplitude stays constant instead of decaying. Put the loop in manual and the cycle stops — that single test separates stiction from every tuning problem.

What this rules out

An external disturbance does not care what mode the controller is in. Neither does an interacting loop, an upstream pressure swing, or a pump cycling on level control somewhere else. If any of those were the cause, the process variable would keep moving with the output frozen.

It also rules out most measurement problems. A noisy or glitching transmitter shows its noise in manual as well as in automatic — what changes in automatic is that the controller starts reacting to it.

What it leaves, and how to split them

Three candidates remain. The trend tells you which.

If the process variable moves in steps while the output ramps smoothly, and the cycle amplitude never decays, it is stiction. The valve is sticking and slipping.

If the process variable goes flat every time the output reverses direction, and then follows once the output has moved past a certain amount, it is backlash — lost motion in the linkage.

If the oscillation decays after a disturbance and the output swings as hard as the process variable, it is tuning. This is the only one of the three you can fix from the faceplate.

Trend showing backlash: the process variable stays flat while the controller output reverses, then follows once the deadband has been taken up. Flat spots on the process variable exactly where the output changes direction. The width of the flat spot is the size of the deadband. This is lost motion in the linkage or the actuator coupling, not a tuning problem.

The one trap in this test

A loop with derivative action on a noisy measurement also goes quiet in manual, and it can look like stiction at a glance because the output is busy. The difference is frequency: derivative-amplified noise is fast and irregular, while a stiction cycle is slow and regular.

If the output is chattering many times a second, look at the rate setting before you blame the valve.

Trend showing noise amplification: a modest amount of noise on the process variable appears on the controller output magnified many times over. The output is noisier than the measurement, not quieter. Every wiggle on the process variable appears larger on the output. The valve is being asked to chase turbulence. On a flow, pressure or speed loop this is almost always derivative action that should not be there.

What causes it, most likely first

Valve stiction

most common tuning cannot fix this

The classic cause of this exact symptom. The stem sticks, the controller winds up against it, the stem breaks free and overshoots, and the cycle sustains.

How to confirm it: Sawtooth or square-cornered process variable against a smooth ramping output, constant amplitude.
What to do: Check packing friction and positioner calibration.

Trend showing valve stiction: the controller output ramps smoothly while the process variable moves in abrupt steps, producing a limit cycle that does not decay.

Backlash in the linkage

common tuning cannot fix this

Lost motion between actuator and stem, or in the positioner feedback linkage. The stem does nothing until the play has been taken up.

How to confirm it: Flat spots on the process variable exactly at output reversals. The width of the flat spot is the size of the deadband.
What to do: Check the stem connector and the positioner feedback linkage.

Trend showing backlash: the process variable stays flat while the controller output reverses, then follows once the deadband has been taken up.

Tuning too aggressive

common tuning can fix this

The feedback path is over-correcting, which by definition only happens with the controller in automatic.

How to confirm it: Oscillation decays. Period matches the loop's own dynamics.
What to do: Reduce the proportional action, or slow the reset.

Derivative on a noisy measurement

less common tuning can fix this

Derivative amplifies measurement noise into output activity, which stops the moment the controller stops writing to the output.

How to confirm it: Fast, irregular output activity rather than a regular cycle.
What to do: Set rate to zero on flow, pressure and speed loops.

Questions that come up

Does this mean the tuning is definitely wrong?

No. It means the cycle is being generated inside the feedback path. A sticking valve produces exactly this symptom with perfectly good tuning, and it is the more common of the two.

The valve moves fine when I stroke it in manual. Is stiction ruled out?

Not at all. Stroking a valve through its full range in manual uses large signal changes that easily overcome stiction. Stiction bites on the small movements a controller makes around a steady operating point. Test with small output changes — a fraction of a percent — and watch whether the process responds to each one.

Symptoms

Tuning guides

Free tools

Bump test procedure

Glossary

Last reviewed 2026-08-01.