Hysteresis: why an alarm chatters

When temperature hovers right around an alarm threshold, the alarm switches on and off and every activation sends a new notification. Even if the real temperature sits steady at 8.0 °C, measurement noise can make the sensor read anywhere between 7.9 and 8.1, and short events such as a door opening do the same.

Hysteresis separates the switch-on threshold from the switch-off threshold. In the example below the alarm switches on at 8.0 °C and has to fall to 7.5 °C before it switches off.

The second setting is the delay: how many minutes the alarm waits after the threshold is crossed. A short door opening stays under the delay and raises nothing, while a real deviation still alarms once the delay runs out. Band and delay together lengthen the time a product can sit above its limit, so both have to fit inside the response time the product owner has approved.

Pick the band width from records, not by guessing: take a week of raw data, look at the amplitude of normal fluctuation and the sensor resolution, then replay candidate band and delay settings over past records and compare when the alarms would have started. A wide band delays a real deviation and a narrow one multiplies notifications.

The product limit is independent of hysteresis and stays with quality or the product owner; the band is never set wide enough to let the product stay above its limit. When everyone can reach the settings, someone tired of the noise widens the threshold quietly. So every change is recorded with its reason, old and new value, test data, approver and date.

The number of notifications shows whether the setting is right: if every alarm gets looked at, the count is correct, and if not, it is too high. A team that receives dozens of notifications a day will not look at the next one either, and a real deviation disappears in that traffic. If alarms are still missed, review shift ownership and the escalation path too.