What Deadband and Hysteresis Are For

Why this matters

Deadband is the first setting a tech narrows when a customer complains about swing, and it is the setting that quietly writes the equipment's early retirement. Every degree you take out of a differential multiplies the number of starts the machine makes over its life, and the starts are what wears it out, not the running. Worse, the two settings people mean by "deadband" are not the same thing and do not do the same job, so half the adjustments made in the field are made to the wrong parameter. Understanding what these settings buy, and what they cannot buy at any width, is the difference between solving a swing complaint and creating a compressor failure two winters out.

Before you narrow a differential

Narrowing a differential increases starts per hour, and starts are the loading condition motors and compressors are least tolerant of. Get the following straight before you touch it.

• Read the equipment documentation for a maximum cycles-per-hour or minimum-off requirement and treat it as a hard bound, not a guideline. A differential narrow enough to violate it will produce a failure the control setting gets no blame for.
• Never defeat, jumper, shorten or bypass an anti-short-cycle timer or a lockout to make equipment restart. A lockout that is tripping is reporting a condition. Establish why it tripped before you touch the timing, because replacing or defeating a correctly-operating protective function reaches the same end state as jumpering it and leaves the underlying condition running.
• Where a setting must be reached inside an energized control enclosure, 29 CFR 1910.333(a)(1) permits energized work only where de-energizing introduces additional or increased hazards or is infeasible due to equipment design or operational limitations; use a meter and leads rated CAT III at or above the circuit voltage and work to the boundaries and protective equipment NFPA 70E-2021 assigns. Where the panel can be dead, open the disconnecting means, lock and tag under 29 CFR 1910.333(b)(2) in general industry or 29 CFR 1926.417 in construction, and prove dead per NFPA 70E-2021, 120.5.
• If the cycling you are being asked to fix is on combustion equipment, wear a personal carbon monoxide monitor on your body before it fires. Changing cycle length changes how much of the appliance's run is spent in startup, which is the part of the cycle where combustion is least settled.

They are two different settings

Deadband is a zone around the target in which the controller does nothing. Inside the zone, the output does not change no matter which way the measurement moves. It exists on modulating loops as much as on switching ones.

Hysteresis, usually called the differential on a switching device, is the difference between the value at which a device changes state going one way and the value at which it changes back going the other way. A device that closes at 68 F on a fall and opens at 71 F on a rise has a 3 F differential. It is a property of switching, not of a zone.

Actuator hysteresis is a third thing entirely and it is not a setting. It is mechanical slop: linkage backlash, a positioner that does not land on the same position for the same signal depending on which direction it approached from. You cannot adjust it in software and it will not appear in any parameter list. It is a repair.

The practical consequence: a modulating loop that swings has a deadband parameter; a switching device that swings has a differential; a valve that will not repeat has slop. Adjusting a differential to fix slop, which is the most common version of this mistake, produces no change at all and burns a visit.

What they are actually buying

Both settings buy the same two things and neither buys accuracy.

They buy immunity to measurement noise. Every measurement has some jitter, from electrical noise, from turbulence, from the sensor's own resolution. Without a zone or a differential, a controller acts on that jitter and the output chatters. A differential wider than the noise makes the noise invisible to the switching decision.

They buy cycle life. This is the one that matters commercially, and it is arithmetic rather than judgment.

For an on/off output on a process that behaves as one lumped mass, with a constant load through the cycle and no lag between the heat source and the sensor, cycle time is the differential divided by the rate of rise, plus the differential divided by the rate of fall. Call the equipment's gross capacity C, expressed as the rate it would drive the process with no losses, and call the loss rate at the current load L. Then the rise rate is C minus L and the fall rate is L.

Two things about that model before it gets used. It assumes the sensor sees the change immediately, which no real system does; real lag adds overshoot on both ends, so the true swing is wider than the differential and the true cycle is longer than the calculation. And C is not a constant on most equipment, because capacity moves with outdoor condition, entering condition or fuel supply, so C has to be taken as the rate measured at the condition you are working in, not lifted from a nameplate.

Worked example: what a degree of differential is worth

Take an on/off heating output where the measured rate of rise with the equipment on and the space at a mild load is 4 F per hour, and the measured rate of fall with the equipment off at that same load is 2 F per hour. Those two numbers give you C directly: the loss rate L is 2 F per hour, and gross capacity C is the rise plus the loss, or 6 F per hour. Both are measured at that one load and change when the load changes.

At a 1.0 F differential and this load: on-time is 1.0 divided by 4, or 0.25 hours, which is 15 minutes. Off-time is 1.0 divided by 2, or 0.5 hours, which is 30 minutes. The full cycle is 45 minutes, so 60 divided by 45 gives 1.33 cycles per hour.

Now find the worst case, because it is not the design day. Cycle rate peaks when on-time equals off-time, which happens when the loss rate is half of gross capacity, that is at L of 3 F per hour. There the rise rate is also 3 F per hour, on-time is 20 minutes, off-time is 20 minutes, the cycle is 40 minutes, and the rate is 1.5 cycles per hour. That is the highest this equipment will ever cycle at a 1.0 F differential, and it happens at half load - a mild shoulder-season afternoon, not the coldest morning of the year. This is why short-cycling complaints cluster in spring and fall and why a machine that behaves all winter starts complaining in April.

Now double the differential to 2.0 F. Every on-time and off-time doubles, so every cycle time doubles and every rate halves. At the original load the cycle is 90 minutes and the rate is 0.67 cycles per hour. At worst case the rate is 0.75 cycles per hour. One degree of differential bought a 50% reduction in starts across the entire load range.

What it cost. The controlled swing goes from 1.0 F to 2.0 F, plus whatever overshoot the real lag adds on each end. That is the trade in full: half the starts for twice the swing, at every load.

What would flip the recommendation. If the process is one where the swing is the product - a tempering bath, a curing space, anything with a specification on the variable itself - the swing is not available to trade and the correct answer is capacity control or a modulating output, not a wider differential. If the equipment's documentation sets a maximum cycles-per-hour figure, that bound decides the differential and the swing is whatever falls out of it.

The failure mode. Narrowing the differential to 0.5 F to satisfy a comfort complaint doubles the rates again, to 3.0 cycles per hour at worst case. The customer is satisfied for a season and the equipment accumulates roughly twice the starts per year for the rest of its life. Nothing on the ticket connects the two events, so when the failure comes it reads as a bad compressor.

What a differential cannot buy at any width

This is the part that keeps techs from over-using the setting.

It does not improve accuracy; it guarantees a floor on inaccuracy. A 2 F differential means the controlled value is at least 2 F wide by definition. Any deviation specification written tighter than the differential is unmeetable, and writing one is a promise that generates a callback.

It does not fix noise, it hides noise smaller than itself. If the jitter is 3 F peak to peak and the differential is 2 F, the differential does nothing at all. Noise that large is a wiring, grounding, placement or connection problem and it gets fixed at the source.

It is not a substitute for a minimum-off timer, and neither substitutes for the other. A differential governs how far the process must move before the state changes. A minimum-off timer governs how much time must pass regardless of what the process does. A large load step can satisfy a wide differential in seconds, and only the timer stops the restart. Systems that need protection need both.

It does not settle a modulating loop that is hunting on delay. If the oscillation's amplitude is larger than the deadband, the deadband is invisible to it and nothing changes. If you widen the deadband until it is larger than the amplitude, you have not stopped the oscillation, you have converted it into a slower limit cycle at the edges of the band.

It is not capacity control. Equipment substantially oversized for its load cycles hard at every reasonable differential, because C is far larger than L, so on-time is short at every load. Widening the differential trades comfort for cycle life without touching the sizing, which is sometimes the only available answer and should be stated to the customer as exactly that rather than as a fix.

Confirming you set it correctly

Time three consecutive full cycles at the current load and compare against your calculation. If the measured cycle is much longer than predicted, lag is adding overshoot and the true swing is wider than the differential you set. If it is much shorter, the load or the capacity is not what you measured, and the rate figures need to be retaken.

Check the cycle rate at half load, not at the condition you happen to be standing in. The peak is at half load and that is the number the equipment lives with. A differential validated on a design day is validated at the easiest point on the curve.

Log the differential you left, with the date and the measured rates it was based on. The next person to see a swing complaint needs to know whether the current setting is a commissioning value or the third narrowing in two years, because those two situations call for opposite actions.

References

• 29 CFR 1910.333(a)(1) and (b)(2), OSHA general industry work practices for energized and de-energized electrical work
• 29 CFR 1926.417, OSHA construction lockout and tagging of circuits
• NFPA 70E-2021, 120.5, verifying an electrically safe work condition
• Manufacturer documentation for maximum cycles per hour and minimum off-time requirements on the specific equipment
• See related: Why a Control Loop Hunts; How to Tell a Deadband Problem From a Sensor Problem