Control Valve Hunting: Causes, Tests and Solutions

 

Feature image: a stiction limit cycle from a simulated loop. The process variable traces a square-ish wave while the controller output ramps and the valve sticks, then jumps. ILLUSTRATIVE simulation.

Control valve hunting is the sustained, unwanted movement of a control valve back and forth around the position the loop is asking for, with the process variable swinging around its setpoint in step. It is one of the most common complaints raised about a control loop, and one of the most often misdiagnosed. The instinctive reaction is to retune the PID (proportional-integral-derivative) controller. That is the right answer some of the time. The rest of the time the loop is hunting because of something the controller settings cannot reach: a sticking valve, backlash in a linkage, an unstable positioner, a valve that is too big for the service, or another loop pushing on the same process.

This guide shows how to troubleshoot control valve hunting in a way that finds the real cause first. It covers the two tests that split the problem in half, how to read the trend shape, how stiction, deadband and backlash differ, what a small-step test looks like and how to interpret it, how tuning, the positioner, the air supply and valve sizing each leave a different fingerprint, and how to prove the fix worked. It includes an animation from a simulated loop, a sizing calculation, a recomputed field case study and a short Python listing.

What you will be able to do after reading

• Define hunting and separate it from normal variation, a decaying tuning oscillation and an oscillation that is not coming from your valve.
• Run the manual-mode test and say what each outcome means.
• Read a trend: sine versus square wave and sawtooth, decaying versus constant amplitude, fast versus slow.
• Tell stiction, deadband, backlash and hysteresis apart, and estimate them from a small-step test.
• Decide between a mechanical repair, an air or positioner fix, a sizing change and a tuning change, in that order of suspicion.
• Check your work: acceptance tests, a recurrence trigger and the records that keep the fix from being undone by a retune.

About the numbers. Definitions, procedures and symptom lists come from the cited sources and are labelled SOURCE. The trends and the response figures in the simulation tables come from a small model I wrote (a first-order-plus-dead-time process, a PI controller and a stick-slip valve) and are ILLUSTRATIVE inputs with CALCULATED results. The sizing example uses ILLUSTRATIVE inputs. The field case reuses a published source's numbers and I recomputed them. None of it is plant data, and none of it replaces your valve data sheet, your control narrative or your site test procedures.

Table of contents

1. The short answer: find which layer is doing the oscillating
2. What control valve hunting is, and what it is not
3. The causes of control valve hunting, in one table
4. See it: the animation of a simulated loop
5. Step 1: capture the baseline and confirm the cycle is real
6. Step 2: the manual test that halves the problem
7. Step 3: read the trend shape
8. Stiction, deadband, backlash and hysteresis: the words matter
9. Why stiction makes a limit cycle: the mechanism
10. Tuning or valve? The detune test and a published trial
11. Positioner, booster and air supply
12. Valve sizing and installed gain: when the geometry is the problem
13. Loop interaction, process causes and measurement
14. A field-style case: a solvent feed loop, recomputed
15. Solutions: what to fix, in what order
16. Prove the fix: acceptance tests and a recurrence trigger
17. What hunting costs: stroke counts and wear
18. Prevention: keep valves from becoming hunting valves
19. Common mistakes
20. Safety notes and limitations
21. Frequently asked questions
22. Conclusion: what to do on Monday morning
23. Related reading on this site
24. Sources and further reading

The short answer: find which layer is doing the oscillating

A control loop is a chain: setpoint, controller, controller output, positioner, actuator, stem and trim, process, transmitter and back to the controller. Hunting means some part of that chain is turning a small error into a repeating swing. The fastest way to find which part is to run one test and look at one trend shape.

• Test. Put the controller in manual with a fixed output. If the oscillation stops, the cause is inside the loop (tuning, or a stiction limit cycle that the controller was feeding). If the valve keeps moving on a steady signal, the fault is downstream of the controller: the positioner, the air supply, a booster or the valve assembly (Merobix).
• Shape. Smooth, roughly sinusoidal swings in controller output and process variable point to tuning. A square-ish process wave with a sawtooth controller output points to stiction. A cycle that appears only at low openings points to an oversized valve. Several loops cycling at the same period point to interaction, not to one valve (Merobix).
• Order. Repair the mechanical and pneumatic faults first, check sizing and installed gain, and tune the loop last. Tuning cannot remove backlash, stiction, an air restriction or an incorrectly sized valve (OnOff Valve).

Table 1. First-pass diagnosis (SOURCE: Merobix, OnOff Valve, PIDSnap; author synthesis)

What you see	Most likely cause	Confirming test	Usual fix
Smooth sine in output and PV, gone in manual, calms when gain is reduced	Controller tuning too aggressive	Halve the gain or lengthen the integral time and watch period and amplitude	Retune properly
Square-ish PV, sawtooth output, gone in manual, survives detuning	Stiction limit cycle	Small output steps in manual: the valve ignores several, then jumps	Repair friction source; positioner service; then check tuning
Command reverses and the valve waits before moving	Deadband or backlash	Step up then down by small amounts and count the steps before movement	Tighten linkage or coupling; positioner calibration
Valve keeps cycling in manual at a steady output	Positioner instability, booster interaction or air supply sag	Watch supply pressure during the cycle; conservative positioner tuning set	Positioner gain or tuning set; booster bypass; air supply repair
Cycles only at low openings, worse at low flow	Oversized valve, high installed gain	Compare required Cv or Kv with valve Cv and predicted travel	Smaller trim or a properly ranged valve
Several nearby loops cycle at the same period	Loop interaction or a shared disturbance	Compare periods across loops	Escalate to the process or control engineer

Safety note. Switching a controller to manual, stepping a valve or opening a positioner on a live process can upset the process. Agree the test with the operator, work within the permit and isolation system, know whether the valve is part of a safety instrumented function, and stay within the limits in your site procedure. This article is general education, not a site test procedure.

What control valve hunting is, and what it is not

Tango Valve describes hunting as sustained, unwanted oscillation of valve travel around the position the loop demands, with the process variable swinging around its setpoint while controller output and the actual stem or shaft travel repeatedly reverse direction. It stresses that the motion does not prove the valve body is defective: a controller can drive a healthy valve into oscillation, while stiction, deadband, positioner faults, unstable instrument air, an actuator that responds too slowly or high installed gain can produce a similar symptom (Tango Valve).

Four signals on one time base

Four signals decide most hunting cases: the setpoint (SP), the process variable (PV), the controller output (OP) and the actual valve travel. They must be on the same time base. If OP oscillates and travel follows it closely, look at tuning, measurement and interaction. If OP changes smoothly but travel pauses, jumps or reverses unpredictably, the evidence points to the final control assembly (Tango Valve). The OnOff Valve guide gives the same split as a quick read: a steady command with irregular travel points to stiction, deadband, air restriction, linkage or positioner problems, while a cycling command with accurate travel points to tuning, sensor delay, process interaction or excessive installed valve gain (OnOff Valve).

What it is not

• Normal variation. Small random movement from noise or disturbances is not hunting. The distinguishing feature is that hunting is sustained and does not settle (Allway Valves).
• A decaying tuning oscillation. After a setpoint change, an over-tuned loop may ring for several cycles, each peak smaller than the last. That is a tuning problem. A mechanical limit cycle does not decay: its amplitude stays the same however small the error gets (PIDSnap).
• An oscillation from outside the loop. If the cycle continues with the output frozen and the valve steady, the controller is not the source and the disturbance comes from elsewhere: an interacting loop, an upstream swing or equipment cycling on and off (PIDSnap).

The causes of control valve hunting, in one table

Sources group the causes into a handful of families, and they differ mainly in how they rank them. Merobix lists tuning, stiction, positioner instability and oversizing (Merobix). Allway Valves lists controller tuning, valve mechanical problems, sizing errors and process design issues (Allway Valves). A Valve World paper by a control valve specialist says that about half the time a cycling loop can be traced to the control valve, and that the rest can come from raw material composition, process equipment, control design, mixing and tuning (Valve World, J. Monsen). Treat any ranking as a starting bias, not a verdict.

Table 2. Families of causes (SOURCE-informed: Merobix, Allway Valves, Tango Valve, OnOff Valve; synthesis)

Family	Typical causes	Signature in the trends	Where it is fixed
Controller tuning	Gain too high, integral too fast, a change that removed margin (valve overhaul, process rate change)	Smooth, roughly sinusoidal, in both OP and PV; sustained or growing in auto, gone in manual	Controller
Stiction	Over-tight or aged packing, guide friction, deposits, corrosion, an actuator with little force margin	Square-ish PV, sawtooth OP, constant amplitude, valve holds then jumps	Valve and actuator
Deadband and backlash	Loose coupling or linkage, positioner or actuator lost motion	Delay on reversals; added dead time; limit cycle with integral action	Linkage, coupling, positioner
Positioner and pneumatics	Positioner gain too high for a small fast actuator, booster bypass closed too far, supply pressure sag, restricted tubing or exhaust, worn spool	Faster cycling that continues at a constant controller output	Positioner, booster, air supply
Sizing and installed gain	Oversized valve working near its seat, low valve authority, steep characteristic near closed	Hunting only at low openings or low flow; stable at other positions	Valve selection or trim
Process and interaction	Interacting loops, shared disturbance, long dead time, changing process gain with level or load	Several loops with one period; hunting at one operating point only	Process or control design
Measurement	Slow or noisy transmitter, heavy damping, coarse trend resolution or sample rate	Cycle that looks different at different sample rates; delay in PV	Instrument or historian settings

The rest of the article walks through the tests in the order that eliminates the most causes with the least effort. Before that, the animation shows the three trend shapes you will meet most often.

See it: the animation of a simulated loop

The animation uses a small model I wrote: a first-order-plus-dead-time process (lag 20 s, dead time 3 s), a PI controller and a valve with an adjustable stick band. All values are ILLUSTRATIVE, and the traces are CALCULATED by the simulation. Time is in simulated seconds.

V01 (animated GIF): five scenes from a simulated PI loop with a stick-slip valve. ILLUSTRATIVE model, CALCULATED results; not plant data.

The GIF contains no script and works anywhere an image does. The table lists what each scene shows and the figures I measured from the simulation.

Table 3. The animation scenes (ILLUSTRATIVE inputs, CALCULATED results)

Scene	Setup	What the trend shows	Measured from the simulation
1	Healthy valve, aggressive PI tuning (gain 9, integral time 8 s)	Fast, smooth, sinusoidal swings; output hits its 0 and 100 % limits	Period 13.5 s; PV peak-to-peak 11.4 %
1b	Same, gain halved to 4.5	Oscillation dies out	PV peak-to-peak 0.00 % at the end
2	Stiction (stick band 4 %, slip 3 %), moderate tuning (gain 0.8, integral time 20 s)	Square-ish PV, sawtooth OP, valve flat then jumping	Period 191 s; PV peak-to-peak 3.0 %; OP swing 5.4 %
3	Same stiction, integral time x2.67 (53.3 s)	Slower cycle, same size	Period 358 s; PV peak-to-peak 3.0 %
3b	Same stiction, gain halved (0.4)	Slower cycle, same size	Period 384 s; PV peak-to-peak 3.0 %
4	Stiction, switched to manual	PV stops cycling and holds	PV peak-to-peak 0.00 % after the switch
5	Valve repaired (stick band 0.1 %), same tuning as scene 2	Settles on setpoint	PV peak-to-peak 0.05 % at the end

What the animation teaches

• The shape is the clue. The tuning oscillation is smooth and fast. The stiction cycle has a flat valve, a ramping output and abrupt steps.
• Detuning slows a stiction cycle but does not remove it. With slower integral action the period went from 191 s to 358 s and with half the gain to 384 s, while the PV peak-to-peak stayed near 3.0 %. This matches the retuning trial in the published case study later in the article.
• Manual mode stops both. The manual test alone cannot separate tuning from stiction. It tells you the cause is inside the loop. The shape and the small-step test do the rest.
• Fixing the valve fixed the loop. With the same tuning, removing the friction removed the cycle.

Limitation. The model is deliberately simple: a first-order process, one friction model and no noise. Real loops add noise, nonlinear process gain, valve characteristics and interacting loops. Use the model to understand shapes, not to predict a plant.

Step 1: capture the baseline and confirm the cycle is real

Preserve the evidence before changing anything. Tango Valve advises recording the valve tag, service condition, controller mode, operating point, alarm history, recent maintenance and the time hunting began, because a change in load, product grade, pump operation, instrument air or control strategy may matter as much as the valve condition (Tango Valve). Then trend SP, PV, OP and actual travel on one time base, and add positioner input, supply pressure, actuator pressure, flow or pressure where you have them.

Check that the trend is not lying

The OnOff Valve guide makes the first step of its diagnostic ladder a check that the cycle is real: trend resolution, transmitter damping, sample rate, signal scaling and alarm filtering can all make a stable valve look unstable or hide fast movement (OnOff Valve). The Atlas case study gives a useful rule of thumb in its own audit: a 5 s historian interval is adequate for a 7.5 minute cycle (about 90 samples per cycle) but cannot show sub-five-second dead time or overshoot (Atlas of Engineering). Pick a sample interval from the fastest behaviour you want to see, not from what the historian happens to store.

Table 4. Is the cycle real? (SOURCE-informed: OnOff Valve, Atlas of Engineering; author synthesis)

Check	Why	Action
Sample interval against cycle period	A coarse interval can alias or miss fast cycling	Use at least several samples per cycle and a faster interval when the cycle is quick
Transmitter damping and filters	Heavy damping adds lag and hides fast movement	Record the damping and filter settings with the trend
Signal scaling and ranges	A wrong range makes a small swing look large or the reverse	Confirm engineering units and range for PV, OP and travel
Clock alignment between signals	A skew between travel and PV hides cause and effect	Use signals from one system or document the offset
Alarm and display filtering	A filtered trend can hide the real shape	Use raw or lightly filtered data for diagnosis

Step 2: the manual test that halves the problem

Put the controller in manual at a fixed output and watch the valve and the process variable for a few minutes. Merobix calls this the one move that splits the diagnosis: oscillation that dies in manual lives in the control loop, in the tuning or in a stiction limit cycle the controller keeps feeding, while oscillation that continues on a steady signal is downstream of the controller, in the positioner, the air supply or the valve assembly (Merobix). PIDSnap makes the same point for the other branch: if cycling continues with the output frozen, the controller is not causing it and the disturbance is coming from somewhere else (PIDSnap).

Table 5. Reading the manual test (SOURCE-informed: Merobix, PIDSnap, Control Station; author synthesis)

Manual-mode result	Meaning	Next step
PV and valve settle	The cycle needs the feedback path: tuning, stiction or deadband with integral action, or loop interaction through the controller	Read the trend shape; run the small-step test
PV keeps cycling with the valve steady	A disturbance outside the loop: interacting loop, upstream swing, cycling equipment	Compare with other loops and equipment at the same period
The valve keeps moving on a steady output	Positioner, booster, air supply or the valve assembly	Watch supply pressure and positioner response during the cycle
The valve ignores small output steps, then jumps	Stiction or deadband	Quantify with the small-step test

Read the manual test carefully. Control Station points out that oscillation stopping in manual does not by itself separate stiction from a tuning problem. Both stop because the feedback path is broken (Control Station). The animation shows exactly that in scene 4. Use the next two steps to separate them.

Safety note. Manual mode removes automatic correction. Hold the output at a value the operator agrees is safe, watch the process limits and have a way back to automatic. Do not run the test on a valve that is part of a safety instrumented function without the proper authorisation.

Step 3: read the trend shape

Merobix says shape is diagnosis. Smooth sinusoidal swings in output and process that stop in manual point at aggressive tuning. A square-wave process cycle with a sawtooth controller output that stops in manual points at a stiction limit cycle (Merobix). Valve World describes the limit cycle in the same terms: the process variable oscillates in a square wave and the controller output in a sawtooth wave (Valve World). PIDSnap adds a second clue, decay: an over-tuned loop rings and decays, each peak smaller than the last, while a stiction cycle holds constant amplitude (PIDSnap).

V02: the manual test, three trend shapes and the order of fixes. A heuristic from published sources plus an ILLUSTRATIVE simulation, not a standard.

Table 6. Tuning versus a mechanical limit cycle (SOURCE: Merobix, PIDSnap, Atlas of Engineering; confirmed by the simulation in Table 3)

Feature	Tuning problem	Stiction or deadband limit cycle
PV shape	Smooth, roughly sinusoidal	Square-ish or stepped
OP shape	Smooth, mirrors PV	Sawtooth or triangle (ramping)
Valve travel	Follows OP closely	Flat, then jumps
After a small setpoint change	Decays (maybe slowly)	Constant amplitude
Effect of detuning	Amplitude falls and the period stretches	Period stretches, amplitude stays
In manual	Stops	Stops (feedback broken)
Small-step test in manual	Valve follows small steps	Valve ignores several steps, then jumps

Allway Valves goes further and describes three patterns: deadband-induced oscillation that shows as a square wave in controller output and a triangular wave in process variable, stiction-induced stick-slip that is irregular and asymmetric, and gain-induced instability that appears at one operating point only, for example a butterfly valve near closed or a level loop whose process gain changes with level (Allway Valves). The last pattern is worth remembering: if the loop hunts at one operating point and not others, look at gain there before blaming friction. The distinction between the first two is subtle in practice, so use the small-step test to quantify rather than relying on the shape alone.

Stiction, deadband, backlash and hysteresis: the words matter

These terms are used loosely on site, and the loose use causes bad diagnoses. The Atlas case study lists boundaries that must not be collapsed: stiction and breakaway concern resistance around the start of motion, dynamic friction acts during motion, deadband is an input range with no observed output change, backlash arises from lost mechanical motion on reversal, and hysteresis is path-dependent output at the same input, and a closed-loop trace may contain several at once (Atlas of Engineering).

Table 7. Terms (SOURCE: Valve World, Atlas of Engineering, Control Station, PiControl Solutions)

Term	Meaning	Where it comes from	How it shows
Stiction	Static friction that must be overcome before the stem moves; higher than the friction while moving	Packing, guides, seat or seal friction, deposits, corrosion (Tango Valve)	Output ramps, valve waits, then jumps
Backlash	Lost motion or slop in a coupling, linkage or stem-to-closure connection	Looseness between parts	No movement after a reversal until the play is taken up
Deadband	The change in control signal needed on reversal before motion begins; equals backlash plus stiction in the Valve World treatment	Both of the above	Added dead time; limit cycles with integral action
Resolution	The smallest change the valve can be positioned to	Stiction sets it	The valve lands in discrete jumps
Hysteresis	Position differs for increasing and decreasing signal at the same value	Friction, lost motion, calibration	Open and close paths do not overlap

A small-step test, with the Valve World numbers

The test for stiction and backlash sends a series of small steps of control signal to the valve, first in one direction and then in the reverse direction (Valve World, J. Monsen). The published example uses 0.1 % steps. It takes three steps before the actuator force breaks the static friction and the valve moves, so the stiction is 0.3 %. On reversing, it takes six steps to take up the play, so the backlash is 0.6 %. The deadband, the change needed on reversal before motion begins, equals the backlash plus the stiction.

Stiction = 3 steps x 0.1 % = 0.3 %

Backlash = 6 steps x 0.1 % = 0.6 %

Deadband = backlash + stiction = 0.6 % + 0.3 % = 0.9 % (SOURCE example, recomputed)

The same paper makes two points worth keeping. First, the deadband does not degrade the resolution: this valve can still be positioned to within 0.3 % even on reversal. Second, the deadband costs time. The controller's integral action has to drive the valve through it, and that shows up as process dead time, which the author calls one of the worst things you can add to a process from the viewpoint of making it difficult to control. As errors get smaller, the integral action gets slower and the delay gets longer, which appears as an increasing time delay (Valve World, J. Monsen).

How to run and read the test

• Agree it first. Get operator agreement and keep the process within safe limits (author advice). The step test itself is run with the loop in manual (Janhen Valve).
• Step small, both ways. Send small commanded changes, for example 0.1 % to 1 %, in the opening direction and then in the closing direction (OnOff Valve, Janhen Valve).
• Measure real travel. Use the positioner feedback if it is independent enough, or a dial indicator or other travel measurement on the stem. A position sensor inside the positioner does not prove the plug or disc moved (Atlas of Engineering).
• Record. The command at which movement starts, the travel delay, the size of the jump, overshoot, settling, and any difference between opening and closing (OnOff Valve).
• Repeat after the repair at the same operating point. The breakaway figure is your stiction estimate and the same test after maintenance proves the fix (Merobix).

A note on terms from the Atlas audit. Resolution, deadband, hysteresis, dead time, overshoot, settling time, repeatability and accuracy are separate properties. One large step that moves successfully can conceal poor small-signal modulation, so test at the step sizes the loop actually uses (Atlas of Engineering).

A short script for the step-test record

The Python function below reads a step-test record, a list of measured positions taken after each 0.1 % command step, and returns how many steps were needed before the position first moved. It is an ILLUSTRATIVE aid for a test log. The threshold is an assumption to set from your travel measurement resolution.

from operator import gt

 

def estimate_deadband(pos, step=0.1, thresh=0.05):

"""Count small command steps before the measured position first responds.

pos[0] is the starting position; pos[i] the position after i steps (percent)."""

base = pos[0]

for i in range(1, len(pos)):

if gt(abs(pos[i] - base), thresh):

return i, round(i * step, 2)

return len(pos) - 1, None

 

# Valve World style record: nothing moves for 3 steps, then the stem jumps 0.3 percent

print(estimate_deadband([50.0, 50.0, 50.0, 50.3, 50.3])) # (3, 0.3)

Run on the Valve World figures it returns 3 steps and 0.3 %, the stiction estimate. Run on the first set of steps after a reversal it returns the deadband (backlash plus stiction), 0.9 % in the published example. The function is not a substitute for a travel measurement of known resolution.

Why stiction makes a limit cycle: the mechanism

The mechanism is the same in every source. Suppose PV is slightly below setpoint. The controller's integral action ramps the output up. The valve, held by static friction, does not move, so the error persists and the output keeps ramping. When the actuator force finally exceeds the breakaway force, the friction falls to its lower dynamic value, the valve jumps, and usually travels farther than commanded. The PV swings past setpoint, the controller reverses, and the same stick-and-slip happens the other way (Control Station, Valve World, Merobix).

Two consequences follow. The cycle exists because of the valve, not because of the controller, so no tuning removes it: Control Station says detuning lowers its amplitude and stretches its period but cannot eliminate it, which is why a retune may help briefly and then the swing returns (Control Station). And it tends to get worse with time: aging packing, corrosion and poor lubrication raise the friction, and an undersized actuator makes the same friction harder to break because it has less force in reserve (Control Station).

What the simulation shows

In the model, the stiction case used a stick band of 4 % and a slip of 3 %, with a PI controller (gain 0.8, integral time 20 s). Table 8 lists what happened when the tuning was changed and when the valve was repaired.

Table 8. Detuning versus repairing the valve in the simulated loop (ILLUSTRATIVE inputs, CALCULATED results)

Case	Gain	Integral time (s)	Stick band (%)	Period (s)	PV peak-to-peak (%)
Moderate tuning, stiction	0.8	20	4	191	3.0
Integral time x2.67	0.8	53.3	4	358	3.0
Gain halved	0.4	20	4	384	3.0
Valve repaired	0.8	20	0.1	none	0.05

Slowing the controller roughly doubled the period (a factor of 1.88 with the integral time and 2.01 with the gain) but left the peak-to-peak nearly where it was. Only removing the friction removed the cycle. That is the quantitative version of the rule: detuning changes how fast a stiction cycle runs, not whether it runs (CALCULATED).

Tuning or valve? The detune test and a published trial

Merobix gives a practical test for the tuning branch: halve the controller gain or lengthen the integral time and watch whether the oscillation period stretches and the amplitude shrinks. If detuning calms it, retune properly instead of leaving the loop sluggish. If detuning changes nothing, stop blaming the controller, because no amount of tuning fixes a mechanical cause (Merobix). Tuning often turns out to be marginal after something else changed: a valve overhaul that reduced friction, a process rate change or a well-meant gain increase removes the margin that hid the problem.

The Atlas case study ran this experiment on a real-style loop. Integral time was raised from 90 s to 240 s with gain held constant, a factor of 2.67. The cycle period went from 7.5 min to 15 min, a factor of 2.0, and the flow still moved in ramps and jumps with nearly the same peak-to-peak amplitude (Atlas of Engineering). The case study also notes that the result supports, but does not uniquely identify, stiction: the period did not scale in proportion to the integral time, so process dynamics and the nonlinear actuator both take part, and a single retuning experiment is not a mechanism-isolation test.

Where the sources disagree. Control Station says detuning lowers the amplitude of a stiction cycle as well as stretching its period (Control Station). The Atlas trial and my simulation show the amplitude nearly unchanged. The difference may come from the friction model and operating conditions. Treat constant amplitude as a strong clue, not an infallible one, and confirm with the small-step test.

Leaving slow tuning in service has a price. The Atlas audit warns that slower integral action can lengthen the cycle yet increase sustained quality error, and PiControl Solutions describes detuning as a band-aid that masks a mechanical failure and leaves a loop that is both cycling and sluggish (Atlas of Engineering, PiControl Solutions). The simulation agrees: the stiction loop detuned by half still cycled, with a longer period.

Positioner, booster and air supply

If the valve keeps cycling on a constant signal in manual, the positioner loop itself is unstable. Merobix names high positioner gain on a small, fast actuator, where the positioner overcorrects each tiny error and chases its own tail. The cycling is typically faster than a process oscillation and shows in the stem even when the control room trend looks merely fuzzy. Selecting a lower gain class or the positioner's conservative tuning set usually settles it (Merobix).

The classic field case in the same source is a volume booster fitted for stroking speed with its bypass needle closed too far: the booster amplifies every small positioner correction into a large air move and the pair oscillate together. Opening the booster bypass restores stability at some cost in speed. An air supply that sags and recovers under load causes a similar rhythmic wander, so check the supply gauge during the cycling before condemning the positioner (Merobix). OnOff Valve adds that supply pressure should be measured while the valve is moving, not only at rest, and that the filter regulator, tubing, fittings, positioner output, actuator seals, stem alignment, coupling, travel stops, packing adjustment and guides should be inspected (OnOff Valve).

Table 9. Pneumatic and positioner checks (SOURCE-informed; author synthesis)

Check	What to look for	Source
Supply pressure while the valve moves	Sag and recovery in step with the cycle	OnOff Valve, Merobix
Filter regulator, tubing, fittings	Restriction, leaks, undersized tubing	OnOff Valve
Positioner output and tuning set	Gain too high for the actuator; hunting in manual	Merobix
Booster bypass	Bypass closed too far	Merobix
Positioner spool and seals (older designs)	Spool sticking or worn, seals or tubing damaged by heat	Automation Forum (low authority)
Feedback linkage	Broken or loose feedback link so the position is wrong	Automation Forum (low authority)
Dither or friction compensation settings	Enabled and appropriate for the assembly	Atlas of Engineering
Actuator force margin	Enough thrust or torque to break friction with margin	Control Station, Janhen Valve

Trade-off. Higher positioner gain or dither can improve small-signal motion but may create hunting, air consumption, noise, wear or incompatibility with the assembly. A larger actuator adds thrust margin but can increase overshoot energy and air demand without fixing linkage or trim problems (Atlas of Engineering).

Valve sizing and installed gain: when the geometry is the problem

Merobix says a valve that is too large for the service does all its real work in the first small fraction of travel, where flow gain is steep and every mechanical imperfection is magnified. The symptom is hunting at low rates that vanishes at high ones, with the valve sitting at a small opening whenever the loop is unhappy. No positioner or tuning work cures it because the geometry is the problem (Merobix). Stiction, in contrast, cycles anywhere in the stroke, while oversizing cycles only near the seat. Automation Forum notes that valves are often sized for future higher flow rates, which can buy a valve slightly larger than the application needs and produce inaccurate control (Automation Forum).

Worked example: two valves, one service

Take a service that needs a flow coefficient of 40 (CV, in whatever consistent units you use) at normal flow. Compare a valve with a maximum capacity of 100 against one with 400. Assume a linear characteristic and a constant pressure drop, so flow is proportional to travel. These are ILLUSTRATIVE inputs, and the result is the simplest possible illustration of the effect.

Travel needed = CV_required / CV_max x 100 %

Flow per percent of travel = CV_max / 100

Flow resolution (as % of required flow) = resolution_in_travel x (CV_max / 100) / CV_required x 100 %

Table 10. Oversizing multiplies the effect of friction (ILLUSTRATIVE inputs, CALCULATED values, linear trim assumed)

Quantity	Valve A (max 100)	Valve B (max 400)
Travel at normal flow (%)	40	10
Flow per percent of travel (CV units)	1.0	4.0
Effect of 0.5 % travel resolution (CV units)	0.5	2.0
Effect as % of required flow	1.25	5.00

The oversized valve works at 10 % travel instead of 40 % and turns the same 0.5 % of travel resolution into 5.00 % of the required flow instead of 1.25 %, a factor of 4. The same factor multiplies the loop gain seen by the controller. That is why a mechanical imperfection that is harmless on a right-sized valve can drive a limit cycle on an oversized one. Real valves have inherent characteristics that are not linear and installed characteristics that depend on the pressure drop distribution, so check the actual installed characteristic and predicted travel at minimum, normal and maximum flow (OnOff Valve).

The Atlas audit adds a caution about sizing fixes: oversizing provides nominal capacity but compresses normal operation into low travel, increases installed gain and reduces useful resolution, while undersizing loses extreme-point capacity (Atlas of Engineering). Interim relief sometimes comes from operating changes that move the valve to a healthier opening. Treat that as a workaround with a date on it (Merobix).

Loop interaction, process causes and measurement

• Several loops at one period. If nearby loops oscillate at the same period, suspect interaction or a shared disturbance, not this valve. Escalate to a process or control engineer with the trends (Merobix).
• One operating point only. A loop that is stable at most valve positions but oscillates at one may have a gain peak there, for example near closed on a butterfly valve or at certain levels in a level loop (Allway Valves).
• Disturbance checks. The Atlas case checked pump discharge pressure, upstream level and transmitter zero and span before blaming the valve, and compared signal noise with the size of the cycle (Atlas of Engineering).
• Measurement. A slow or noisy transmitter, heavy damping or coarse trend resolution can create or hide a cycle. Confirm the measurement before acting on the shape (OnOff Valve).
• Sensor and valve in different planes. Controller output, positioner feedback, stem travel, closure member position, flow and the process variable are different planes. Agreement on one plane cannot be silently assigned to another (Atlas of Engineering).

A field-style case: a solvent feed loop, recomputed

The Atlas case study follows a solvent flow loop feeding a batch reactor. After a valve actuator service, operators saw a slow cycling flow trend, and the first suggestion was to detune the PID. I recomputed its numbers (Atlas of Engineering). The case is a published teaching example, so the values are SOURCE figures and my arithmetic is CALCULATED.

Table 11. The Atlas case numbers, recomputed (SOURCE figures, CALCULATED checks)

Quantity	Value (SOURCE)	Recomputation (CALCULATED)
Controller output during one ramp	47.0 % to 55.5 %	Breakaway command span = 8.5 %
Valve travel over the same ramp	47.1 % to 47.3 %	Travel change = 0.2 %
Flow before and after the jump	111 and 130 m3/h	Jump = 19 m3/h
Travel after the jump	54.9 %	Travel jump = 7.6 %
Installed gain during the jump	stated as 2.5	19 / 7.6 = 2.5 (m3/h per % travel)
Flow swing expected from the delayed command	stated as 21.3 m3/h	2.5 x 8.5 = 21.25 m3/h, against an observed 24 m3/h peak-to-peak
Companion feed and ratio band	96 m3/h, ratio 1.25 plus or minus 0.05	Acceptable solvent flow 115.2 to 124.8 m3/h
Ratio at the cycle extremes	108 and 132 m3/h	1.125 and 1.375, both outside 1.20 to 1.30
Integral absolute error estimate, before and after	peak error 12 and 2 m3/h	360 and 60 (m3/h x min), a 83 % reduction

The numbers hold together. The command moved by 8.5 points while travel moved by 0.2, and then travel jumped by 7.6 points while the command moved by only 0.3. That reversal, from almost no response to a large response at nearly constant command, is more informative than the flow waveform. The estimate of the flow swing from the delayed command (21.3 m3/h) is close to the observed swing (24 m3/h), so the trend is consistent with a valve-positioning fault, not just aggressive tuning. A note on the source: its ratio fractions are printed with the companion feed on top (for example 96 over 120), but the stated values 1.25, 1.13 and 1.38 match the solvent flow divided by the companion feed, which is what I used.

The case study is careful about what the numbers do not prove. The signature is consistent with packing friction but does not prove it: an I/P converter threshold, a sticky positioner relay, a restricted pneumatic path, linkage play, sensor coupling or a closure member problem can look the same externally. If the travel sensor is inside the positioner, agreement at that sensor does not prove the plug moved. And the installed gain of 2.5 is a local secant for that pressure-drop state, not the valve characteristic (Atlas of Engineering).

The maintenance decision in the case was to inspect packing load, actuator linkage, positioner calibration, air supply, I/P converter behaviour and stem travel, repeat the manual small-step test after maintenance, and restore tuning only after the valve response was acceptable. After service the breakaway command span was 0.7 % instead of 8.5 %, the flow peak-to-peak 4 m3/h instead of 24, and operator interventions per batch fell from 6 to 10 down to 0 to 1 (SOURCE, Atlas of Engineering). The case study also reduces a risk priority number from 140 to 28 (7 x 5 x 4 and 7 x 2 x 2) and itself cautions that this is an 80 % reduction in an ordinal product, not a measured 80 % reduction in probability or risk.

Solutions: what to fix, in what order

Repair what is mechanical, then what is pneumatic, then check sizing, and tune last. OnOff Valve states the principle directly: repair mechanical faults, tune the loop only after the final element responds correctly, and reselect the valve when sizing or trim architecture is fundamentally wrong (OnOff Valve).

Table 12. Solutions (SOURCE-informed: Merobix, OnOff Valve, Atlas of Engineering, Control Station; author synthesis)

Cause	Corrective action	Check afterwards	Watch out for
Packing friction (stiction)	Adjust packing load or replace packing; check stem condition and lubrication	Small-step test: breakaway falls	Reducing packing load can worsen leakage; follow leakage and emissions requirements
Guide, seat or trim friction	Overhaul the valve; clean deposits, repair corrosion	Breakaway, travel repeatability	Escalate when packing adjustment does not shrink the breakaway (Merobix)
Backlash in coupling or linkage	Tighten or replace the coupling, linkage or connection	Reversal test: steps before movement	Check after reassembly in both directions
Positioner instability	Lower positioner gain or use the conservative tuning set	Hold steady output; no cycling	Slower response
Booster interaction	Open the booster bypass slightly	Cycling stops in manual	Slower stroking
Air supply sag or restriction	Repair filter regulator, tubing, fittings; supply capacity	Supply pressure steady during moves	Measure while the valve moves
Insufficient actuator force margin	Review thrust or torque against friction; larger actuator if justified	Breakaway and response times	Overshoot energy, air demand
Oversized valve	Smaller trim or a properly ranged valve	Travel at normal flow; no cycling at low flow	Capacity at the extremes
Aggressive tuning	Retune from a measured process response, not trial and error	Decaying response, no cycle	Do not leave the loop sluggish
Interaction or process disturbance	Process or control engineering review; decoupling or different strategy	Loops no longer share a period	Needs the process engineer

When to escalate

Merobix recommends escalating to a valve overhaul when the manual step test shows breakaway steps large enough to explain the process cycle and packing adjustment does not shrink them, because the friction then lives in the trim or guiding where only a teardown reaches it. Escalate to a process or controls engineer when several loops oscillate together at a common period (Merobix). Bring data: the trend shapes, the manual-mode result, the breakaway step size and when the cycling started relative to maintenance or process changes. A hunting complaint with that package attached gets solved, while one without it gets retuned again and comes back next month (Merobix).

Prove the fix: acceptance tests and a recurrence trigger

The Atlas case study lists the evidence a defensible closeout should hold. The table adapts it. The principle is that a smooth one-hour trend, a positioner calibration certificate or a lower risk number cannot substitute for a failed gate.

Table 13. Closeout evidence (SOURCE: Atlas of Engineering, adapted)

Evidence	Why it matters
Before and after trends with PV, SP, OP and travel on one time base	Shows the change at the same operating point
Manual small-step test before and after	Confirms actuator response independent of closed-loop tuning
Positioner diagnostics	Documents air supply, travel calibration, friction and alarms
Tuning record	Stops a hidden tuning change being mistaken for the mechanical repair
Process requirement check (for example ratio or quality limits)	Shows the process need is actually met
Maintenance work order	Links the symptom to the physical correction
Recurrence trigger	Defines when the issue must be reopened

The recurrence trigger should be quantitative. The case study's example is to reopen the action if the valve breakaway exceeds 2 % of command span, or if flow oscillation exceeds 8 m3/h peak to peak for more than three cycles, or if operators must place the loop in manual during normal production (Atlas of Engineering). Use your own limits from your process requirement. The same source warns that one quiet hour covers only about eight of the former 7.5 minute cycles, so it is persuasive evidence but not proof against recurrence across packing temperature, stem direction, fluid viscosity, pressure drop and ambient conditions, and an unguarded point estimate of breakaway should not own the decision without the measurement uncertainty.

What hunting costs: stroke counts and wear

Allway Valves puts the wear in numbers: a valve cycling 10 times per minute for 8,760 hours per year makes about 5.25 million stem strokes per year, against a few hundred strokes for a well-controlled valve (Allway Valves). The arithmetic checks: 10 x 60 x 8,760 = 5,256,000 cycles per year (CALCULATED). A slow cycle is much less severe: the 7.5 minute cycle in the Atlas case gives (60 / 7.5) x 8,760 = 70,080 cycles per year (CALCULATED). The mechanism of damage matters too: packing wear from repeated stem movement, seat wear from repeated throttling at one opening, and fatigue in piping and instruments from repeated pressure cycling are listed in the same source, along with off-spec product and pressure excursions (Allway Valves). Those source statements come from a valve distributor and are not quantified further.

Prevention: keep valves from becoming hunting valves

• Run periodic small-step or travel tests on important loops, and trend the breakaway figure so deterioration shows before the loop cycles.
• Use positioner diagnostics where available and keep the as-found and as-left records (Atlas of Engineering).
• Specify packing and its adjustment procedure with leakage and friction in mind, and tell maintenance why over-tightening a gland to stop a small leak can create a control problem (Merobix).
• Review valve sizing and installed gain at design and after any process change that moves the normal operating point.
• Document tuning and keep the history, so a retune is never used to hide a mechanical fault.
• Consider loop performance monitoring. Vendors describe closed-loop methods that detect stiction without taking the loop out of automatic (PiControl Solutions). Treat these as tools to find candidates, not as proof of a physical finding. A diagnostic flag is not a repair.
• Control changes: a valve, positioner, controller equation, scaling or tuning change should go through management of change, and the loop should be re-checked afterwards.

Common mistakes

Table 14. Common mistakes (author synthesis)

Mistake	Why it hurts	Better practice
Retuning first	Detuning a stiction cycle slows it and leaves a sluggish loop	Manual test, trend shape and small-step test first
Reading manual-mode stopping as proof of tuning	A stiction cycle also stops in manual	Add the small-step test
Trusting a coarse trend	A slow sample rate or heavy filter hides the shape	Check resolution and sample rate
Testing in one direction only	Backlash shows on reversal	Step both ways
Using the positioner sensor as proof the plug moved	The sensor may agree while the plug does not move	Measure stem travel independently where possible
Cranking down the packing to stop a leak	Raises friction and creates stiction	Address the leak with leakage limits in mind and re-test the response
Ignoring sizing	Oversizing multiplies the effect of every imperfection	Check travel and installed gain at minimum, normal and maximum flow
Changing tuning and hardware at the same time	You cannot tell which change worked	One change at a time, with a tuning record
Declaring victory after a quiet hour	One hour covers a handful of cycles	Use a defined acceptance test and a recurrence trigger

Safety notes and limitations

• Source authority. Most sources are valve suppliers, distributors, software vendors and practitioner sites, plus one trade-magazine paper by a control valve consultant and one teaching case study. None is a standard or a regulator document. Several sell related products. Treat ranking claims (which cause is most common) as opinion.
• Claims I did not adopt. One source gives an equation for the period of a deadband limit cycle, and another says that over 30 % of valves in operation have performance-degrading mechanical issues. I could not verify either, so neither is used. The Atlas case study lists ISA control valve test and terminology standards (for example ISA-75.25.01 for step response and ISA-75.13.01 for positioner performance); I did not access them and do not quote them.
• Simulation. The trends come from a deliberately simple model with illustrative parameters. They show shapes, not plant behaviour.
• Illustrative numbers. The sizing example and the tuning values are assumptions, not plant data.
• Safety. Manual mode, step tests and positioner work affect a live process. Follow your permit, isolation and authorisation procedures, and treat any valve in a safety instrumented function under that system's rules.
• Visuals. The images are conceptual or simulated and should be reviewed by a qualified control engineer before being used in training or procedures.

Frequently asked questions

What is control valve hunting?

Control valve hunting is sustained, unwanted oscillation of valve travel around the position the control loop demands, with the process variable swinging around its setpoint. It does not by itself prove the valve is defective: tuning, stiction, deadband, positioner faults, air supply problems, valve sizing and loop interaction can all produce it.

How do I tell valve stiction from bad controller tuning?

Put the controller in manual and run small steps, and look at the trend shape. Tuning problems give smooth, roughly sinusoidal swings in both controller output and process variable, which decay or calm when the gain is reduced. Stiction gives a square-ish process wave with a sawtooth controller output at constant amplitude, and in manual the valve ignores several small steps and then jumps. Both stop when the controller is placed in manual, so manual mode alone does not separate them.

Why does my valve keep moving when the controller is in manual?

A moving valve on a steady controller output means the oscillation is downstream of the controller. The cited sources name an unstable positioner loop (for example gain too high for a small fast actuator), a volume booster with its bypass closed too far, and an air supply that sags and recovers under load. Watch the supply pressure during the cycle before condemning the positioner.

Will retuning the PID fix hunting caused by stiction?

Not on its own. Detuning slows a stiction limit cycle and may make it look cured, but the friction is still there and integral action still winds up across it. In the simulated example, slower integral action roughly doubled the period and left the peak-to-peak size almost unchanged. Repair the mechanical cause, then retune if needed.

What step size should I use for a small-step test?

Use steps small enough to resolve the valve behaviour and in the range the loop actually uses. A published valve paper uses 0.1 % steps, and a published case study applies steps of about 1 % and reports the command at which movement starts. The right size depends on your travel measurement resolution and your site procedure, so do not copy these numbers without checking them.

Can an oversized valve cause hunting?

Yes. An oversized valve does its real work in the first small fraction of travel, where flow gain is steep and every mechanical imperfection is magnified. In the illustrative example, a valve with four times the capacity needs a quarter of the travel and makes each 0.5 % of travel resolution worth four times as much flow. Hunting that appears at low flow and vanishes at high flow is the signature.

What is the difference between stiction, backlash and deadband?

Stiction is static friction that must be overcome before the stem starts to move. Backlash is lost motion in a coupling or linkage that must be taken up on reversal. In one published treatment deadband, the signal change needed on reversal before motion begins, equals backlash plus stiction: 0.6 % plus 0.3 % gives 0.9 % in the worked example. Hysteresis is path-dependent position at the same input.

Is tightening the packing a safe fix for a leak on a control valve?

It can create a different problem. Over-tightened packing raises static friction, and the cited sources say it is a common cause of stiction after a leak is chased by cranking the gland down. Reducing packing load can lower breakaway force but may worsen leakage, so decide the packing load with the leakage and emissions requirements in mind and test the valve response afterwards.

How do I prove the hunting is really fixed?

Repeat the same small-step test and a trend at the same operating point as the diagnostic, and compare before and after. Record the breakaway figure, the peak-to-peak variation, the cycle and any operator interventions, keep the tuning record so the fix is not hidden by a tuning change, and define a recurrence trigger so that the problem is reopened if the breakaway or the oscillation exceeds a stated limit.

Conclusion: what to do on Monday morning

Control valve hunting is a symptom, not a diagnosis. The loop is a chain, and the job is to find the link that turns a small error into a repeating swing. Run the manual test to learn whether the cause is inside or downstream of the controller. Read the trend shape: sinusoidal and decaying for tuning, square and sawtooth at constant amplitude for stiction. Quantify friction with a small-step test in both directions. Check the positioner, booster and air supply, then the sizing, and tune last.

For your next hunting loop, capture the four signals on one time base, record the operating point, run the tests in order, and bring the data when you escalate. After the repair, repeat the same tests, keep the tuning record, and set a recurrence trigger. Download the images, step through the animation scene by scene against Table 3, and try the small-step script on your own test records.

Related reading on this site

• How to troubleshoot loop check vs function test: causes, tests and solutions
• Flow control loop example with animation: how a PI loop measures, decides and moves the valve
• How to troubleshoot proof test interval and PFDavg: causes, tests and solutions
• Reading instrument tags and control loops on a P&ID

Sources and further reading

• Valve World, Control valves and cycling control loops (J. Monsen, consultant, Nov 2023)
• Atlas of Engineering, Control valve stiction limit cycle case study (teaching case, updated Jun 2026)
• Merobix, How to diagnose control valve hunting and oscillation and What is valve stiction in control loops (automation integrator)
• Tango Valve, Control valve hunting: causes and troubleshooting (valve supplier)
• OnOff Valve, Control valve hunting: causes, tests and fixes (valve supplier)
• PIDSnap, Control valve hunting: causes and how to diagnose it (software vendor)
• Control Station, Valve stiction: why it undermines PID control and how to detect it (software vendor)
• Allway Valves, Why is my control valve hunting and Control valve hunting and instability (valve distributor)
• Automation Forum, Valve hunting in control valves: causes, diagnosis and fix (practitioner site)
• Janhen Valve, Troubleshooting control valve stiction (valve manufacturer)
• PiControl Solutions, Control valve diagnostics in PID loop performance (consultancy and software vendor)
• Mascot Valves, Common control valve problems and troubleshooting tips (valve manufacturer)

Educational content. Product and company names appear only as sources. Not a substitute for valve data sheets, manufacturer instructions, site procedures or review by a qualified control and instrumentation engineer.