Alarm hysteresis is the practice of using two different values for a single alarm: one value at which the alarm becomes active and a separate, less severe value the signal must recover past before the alarm is allowed to clear. Without it, an alarm sets and clears at the exact same threshold, so a signal that hovers right at that limit crosses it over and over, and the alarm chatters on and off dozens of times a minute. Hysteresis breaks that loop by making the clear condition deliberately harder to reach than the set condition, so once a value has recovered it has to recover meaningfully before the system agrees the problem is over. This page is about how hysteresis is applied specifically to alarm reset, how to size the band against signal noise, and how it differs from simply delaying the alarm in time.
Alarm hysteresis in one line: Alarm hysteresis means an alarm sets at one threshold but only clears after the value recovers past a separate, less severe threshold, so the set point and the reset point are not the same value. This gap stops a signal hovering right at the limit from chattering the alarm on and off, because after the alarm activates the value has to move a real distance back into the normal range before the alarm resets. The size of that gap is chosen to be larger than the ordinary noise on the signal.
The core idea of alarm hysteresis is that the alarm uses two thresholds instead of one. Consider a high alarm on a pressure. The alarm activates when the pressure rises above its set threshold, but it does not clear the moment the pressure dips back below that same value. Instead it stays active until the pressure falls to a distinctly lower reset threshold, and only then does the alarm return to normal. The difference between the set threshold and the reset threshold is the hysteresis band. For a low alarm the geometry flips: the alarm sets when the value falls below the set threshold and only clears once it rises past a higher reset threshold. In both cases the reset condition is deliberately more demanding than the set condition, which is what makes the alarm reluctant to declare the problem over.
The problem hysteresis solves is chattering, and chattering happens because real process signals are never perfectly smooth. A pressure sitting right at its alarm threshold is not holding a single value; it is jittering up and down by small amounts because of turbulence, pump pulsation, electrical noise, and the finite resolution of the measurement. With a single threshold, every one of those tiny jitters that crosses the line sets the alarm, and every one that crosses back clears it, producing a rapid stream of set and clear events for a condition that has not really changed. That flood of transitions clutters the alarm summary, buries genuine alarms, and trains the operator to distrust the annunciation entirely.
By separating the set and clear thresholds, hysteresis guarantees that once the alarm is active, small jitters around the original set threshold cannot clear it, because clearing now requires reaching a value further away. The signal has to actually recover, not merely wobble, before the alarm resets. This gives the alarm a stable memory: it commits to being active until there is convincing evidence the excursion is genuinely over, rather than flipping its state on every crossing. The result is one clean set event and one clean clear event for what an operator would naturally regard as a single episode.
The right size for the hysteresis band comes from the noise on the signal, not from a habit or a default. The band has to be wider than the peak-to-peak amplitude of the ordinary fluctuation the signal shows while it sits near the threshold, because if the band is narrower than that noise, the jitter can still span both thresholds and the alarm chatters anyway. A practical way to size it is to look at how much the signal bounces around under normal conditions and set the band comfortably larger than that swing, so that clearing the alarm requires a real recovery rather than a lucky dip. A signal that is inherently noisy, such as a flow on a pulsating line, needs a wider band than a slow, well-damped signal such as a large tank level.
There is a cost to making the band too wide, which is why it is a sizing decision rather than a matter of using the largest value possible. A very wide hysteresis band means the alarm stays active well after the process has substantially recovered, because the value has to travel a long way back before the alarm agrees to clear. That can leave a stale-looking alarm on the summary long after the operator has fixed the condition, and it can mask a second, genuine re-excursion that happens while the alarm is still latched open by an oversized band. The goal is the smallest band that reliably stops chatter for that particular signal, so the alarm clears promptly when the condition truly ends but never flickers.
In practice the terms hysteresis and alarm deadband describe the same gap between the set and reset points, and many systems configure it as a single deadband value applied to the alarm. The useful distinction is one of emphasis: talking about hysteresis foregrounds the behavior, the fact that the alarm follows a different path clearing than it did setting and therefore remembers its state, while talking about a deadband foregrounds the configuration, the numeric width you enter. Framing it as hysteresis is helpful precisely when the question is about reset, because it keeps attention on the asymmetry that clearing is harder than setting, which is the whole reason the alarm stops chattering.
Hysteresis is an amplitude technique, and it is important not to confuse it with a time-based filter that does a superficially similar job. Hysteresis controls where in value the alarm sets and clears; a time filter, such as an on-delay or off-delay, controls how long a condition must persist before the alarm changes state, regardless of value. The two attack different causes of nuisance alarms. Hysteresis attacks a signal that is genuinely sitting at the threshold and jittering across it, where the fix is to require a real recovery in amplitude. A time filter attacks a signal that briefly spikes across the threshold and comes back on its own, where the fix is to require the condition to last long enough to be worth an operator's attention.
Because they address different failure modes, hysteresis and time filtering are complementary rather than substitutes, and a well-configured alarm often uses both. A noisy signal that both hovers at the threshold and occasionally spikes past it benefits from a hysteresis band to stop the hovering chatter and a short delay to swallow the spikes. Using a time delay to paper over a hovering signal tends to work poorly, because the condition really is true for extended periods and no reasonable delay suppresses it, while using hysteresis to catch a momentary spike does nothing because the spike may clear well past the reset threshold on its own. Matching the technique to the behavior is the key.
In a cloud SCADA environment, where alarms are evaluated on live telemetry from many remote sites and pushed to operators by notification rather than watched on a board, unstable alarms are especially expensive, so both techniques matter. A chattering high alarm on a remote wellhead does not just clutter a screen; it can fire a stream of text messages or emails that the on-call operator quickly learns to mute. Configuring hysteresis on the reset condition so the alarm commits to one set and one clear per genuine episode is what keeps those notifications trustworthy. A platform like Merobix that trends the raw signal alongside the alarm state also makes the band easy to tune, because you can see exactly how much the signal jitters at the threshold and choose a reset gap that clears just past that noise.
Hysteresis is an amplitude gap: the value must recover past a separate reset threshold before the alarm clears, which stops a signal jittering at the limit from chattering. A delay time is a duration: the condition must persist for a set period before the alarm changes state, which swallows brief spikes. They fix different problems, so a noisy alarm that both hovers at the limit and occasionally spikes often uses both together.
Wide enough that it comfortably exceeds the normal peak-to-peak noise on the signal near the threshold, so ordinary jitter cannot span both the set and reset points. If it is narrower than the noise the alarm still chatters, but if it is much wider than needed the alarm lingers after the process has recovered and can mask a genuine re-excursion. The aim is the smallest band that reliably stops chatter for that particular signal.
In most systems they refer to the same gap between the set and reset thresholds, often entered as a single deadband value. The difference is emphasis: hysteresis describes the behavior, that clearing follows a different path than setting so the alarm holds its state, while deadband describes the numeric width you configure. Talking in terms of hysteresis is helpful when the focus is specifically on how and when the alarm resets.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.