ServicesHow We WorkSectorsAcademyLibraryAboutContactOpen the Toolbox →
Process Control · Diagnostics · Stability

Erratic output: when a signal won’t hold still

Flow that surges, pressure that hunts, current that swings — output that refuses to settle is its own ISO 14224 failure mode (ERO), and it’s one of the most misdiagnosed. The swing might be the control loop chasing itself, a valve sticking, the machine cavitating or surging, or the process itself. The good news: instability has a clean numerical fingerprint — rising variance — that a monitoring system can catch long before an operator complains.

Loop TuningValve StictionVariance / σISO 14224 · ERO
⚡ TL;DR

Erratic output = unstable output. A value that oscillates, hunts or swings instead of holding its setpoint. The three suspects are the control loop, the final element (usually a sticking valve), and the machine/process (cavitation, surge, two-phase flow).

Diagnose by the shape of the swing: a smooth sinusoid points to loop tuning; a square-ish saw-tooth points to valve stiction (a limit cycle); an irregular, broadband swing points to a process or mechanical cause.

Whatever the cause, the signature is the same: the signal’s variance (standard deviation) rises. Trend σ and you catch developing instability early — a detectable P on the P–F curve.

1 · What “erratic” actually looks like

A steady signal sits on its setpoint with only small noise. An erratic one does one of these:

The pattern is the first clue to the cause — so read the trace before you touch anything.

2 · The three suspects

SuspectTypical signatureCause & fix
Control loopSmooth, growing or sustained sinusoid; period set by loop dynamicsToo much gain / wrong tuning → retune (reduce gain, adjust integral). See valve authority.
Valve (stiction)Saw-tooth limit cycle; output moves in jumps, not smoothlyPacking too tight, worn stem, no positioner → service the valve; the loop can’t be tuned around stiction.
PumpErratic flow/pressure, noise, vibrationCavitation, suction recirculation, air entrainment, running near shutoff, worn wear rings.
CompressorViolent flow/pressure swings near the surge lineSurge → anti-surge control, recycle.
Process / sensorBroadband, irregularTwo-phase flow, slugging, entrained gas — or a noisy instrument (rule that out first).

Stiction is the great impostor. A saw-tooth limit cycle is not a tuning problem — no amount of retuning fixes a valve that moves in sticky jumps. Detune and it slows; retune and it grows. Read the shape: smooth = loop, saw-tooth = valve.

3 · The numerical signature: rising variance

Diagnosis is one thing; detection is another. Whatever the cause, an erratic signal shares one property a computer can watch continuously: its variance grows. A steady loop has a small standard deviation (σ); as it starts to hunt, σ climbs — often well before anyone notices the swing by eye.

Play with the amplitude and period below and watch σ and the peak-to-peak swing move. That σ is exactly the kind of derived feature a monitoring platform trends to flag developing instability.

Interactive — Hunting & the variance signature

Live model

A signal sits at a setpoint of 50 with small noise. Add an oscillation and watch how the standard deviation (σ) and peak-to-peak swing rise — the numerical fingerprint a monitoring system trends to catch instability early.

How far the output swings around setpoint
How fast it cycles
Std deviation σ
the trended feature
Peak-to-peak
%
swing size
Period
s
cycle time
Verdict
 
Output signal (30 s window)
Blue = output · green dashed = setpoint · shaded = ±2% steady band
OutputSetpoint±2% band
Model: output = 50 + amplitude · sin(2πt/period) + small noise; σ is the standard deviation over the window. For a clean oscillation σ ≈ amplitude / √2. Real detectors trend σ (and rate-of-change of σ) so a slowly worsening hunt is caught while it’s still small.

4 · Detecting & diagnosing it

Bluestream Predictive Maintenance

Catch instability as a rising trend, not a complaint

Bluestream’s predictive platform computes derived features like rolling standard deviation on any signal and trends them — so an output that’s starting to hunt raises a predictive alert while the swing is still small, with the lead time to fix the loop or service the valve before production feels it.

Talk to us about predictive maintenance →

Key takeaways

Across the Academy