Blog / Connectivity
Connectivity · Guide

Machine State Detection: Running, Idle, Stopped

OL By OEE Lab Editorial |Updated 2 November 2026 |7 min read
Short answer

Combine a signal that shows power or motion with one that shows output, and add short timers so that noise does not create false states. Running means motion plus parts in the last few cycles, idle means power with no motion, stopped means motion with no output, and any fault signal overrides the rest.

Why one signal gets the state wrong

Every machine signal answers a narrow question. A motor current says the motor is drawing power, and a run bit says the logic asked the motor to run.

Neither one says that good work is leaving the machine. A conveyor can carry nothing, a press can stroke with no strip, and a filler can wait on an empty infeed with its motors turning.

Example: a cutting machine is watched for one hour. Its motor current stays high for 53 minutes, so a current-only monitor shows 53 minutes of running.

A part sensor on the same machine tells a different story. For 8 of those minutes the motor turned and nothing came out, so true running time was 45 minutes.

Current only: 53 min runningCurrent + count: 45 min running03060 minRunStopIdle
Hypothetical hour: the count exposes 8 minutes that current alone reports as running (75% against 88% of the hour).

Define the states before you read a signal

Write one sentence for each state, in words the line team would accept. If two people could read a definition differently, the data will drift.

  • Running: the machine is producing output, at any speed.
  • Idle: it is powered and ready, with no motion and no fault.
  • Stopped: it should be producing, and output has halted without a fault signal.
  • Fault: an alarm, fault bit or red lamp is active.
  • Off: no power, or no shift scheduled.

Some teams merge idle and stopped, and that is fine for a first pass. Signals rarely tell you whether idle time was planned, so setup and breaks usually need a mode input or an operator tag.

The wider picture of states and what each one feeds is in what is machine monitoring. This page stays on the logic.

A rule table you can copy

Read the table from the top and stop at the first row that matches. Fault comes first, because a fault outranks every other reading.

Fault lampMotor currentParts in last 3 cyclesState
OnAnyAnyFault
OffAbove 6 AYesRunning
OffAbove 6 ANoStopped (motor on, no output)
Off1-4 ANoIdle
OffBelow 1 ANoOff or powered down
Example rules for one machine. The current bands are hypothetical: measure your own, because every machine differs.

The third row is the valuable one. It catches starved, blocked and running-empty time, which a single signal books as production.

The gap between 4 A and 6 A is deliberate, and the next section explains why.

What you can detect with what you have

You rarely get every signal on day one. Here is what each set can honestly tell you.

  • Run signal only (a PLC bit, a motor contactor or current): on or off, with no view of output. Use it only as a start.
  • Part count only: running or not running. It cannot split idle from a fault, but it is honest about output.
  • Run signal plus count: running, idle and stopped. This is the minimum worth building on.
  • Run, count and fault: all of the above, and faults split out from waiting. A stack light or a PLC fault bit can supply the fault input.
  • Add mode: setup and manual time separate from unplanned stops, where the controller or an operator input provides it.

Reading these inputs takes different hardware on different machines. See PLC signals for OEE where a controller is readable, and clamp-on current sensors where it is not.

Anything wired inside a cabinet is work for a qualified electrician, with the panel made safe under your site rules.

Three filters that stop false states

Raw signals flicker, so raw states flicker too. A log with hundreds of state changes a shift is often noise, so check the signal before you blame the machine.

1. Two thresholds, not one

Example: a single 5 A threshold on a current that wanders between 4.8 A and 5.2 A flips the state again and again. Set the state on at 6 A and off at 4 A, and the wandering stops mattering.

This gap is called hysteresis. Many sensor and gateway setups can do it, or your software can.

2. A hold timer

Require a new reading to last before you accept it. Example: current must stay above 6 A for 10 seconds before the state moves to running, which ignores the short surge when a motor starts.

3. A no-output timer based on the cycle

Choose the time without a part that ends running. Three ideal cycles is a sensible starting point, and it scales with the machine's rate.

Too short, and normal variation creates false stops. Too long, and short stops vanish.

Setting the value on purpose is covered in stop threshold for OEE.

1
Clean the signal
Apply on and off thresholds to analog readings and ignore pulses too short to be real.
2
Apply the rule table
Check fault, then motion plus output, then the rest, from top to bottom.
3
Hold before accepting
The new state must last for its timer, such as 10 seconds or 3 ideal cycles.
4
Back-date the change
Log the state from when it truly began, such as the time of the last part.
The order matters: clean the signal first, decide the state second, and correct the time stamp last.

The time stamp trap

A no-output timer knows a stop has happened only after the timer runs out. If you log the stop at that moment, every stop looks shorter than it was.

Example: the ideal cycle is 12 seconds, so a 3-cycle timer fires after 36 seconds. With 20 stops in a shift, that is 20 x 36 s = 720 s, or 12 minutes wrongly booked as running.

The fix is to back-date. When the timer fires, start the stop at the time of the last counted part.

How to test your state logic

Do not trust a state model because the chart looks tidy. Run this check on each machine type before you use the numbers.

  1. Stand at the machine for one hour with the live state on a phone or tablet.
  2. Note every real stop, with its start and end, on paper.
  3. Have the operator force the states you rarely see: a normal stop, a starved infeed, an alarm that is safe to trigger. Never bridge a guard or an interlock to do it.
  4. Compare your notes with the state log, stop by stop.
  5. Change the rule or the timer, then repeat on another shift.

Look for missed stops and invented stops separately, because they have different causes. Missed stops point to a long timer or a slow read, and invented stops point to a short timer or a noisy signal.

Common mistakes

  • Treating power as production. A machine that is on is not a machine that is making parts.
  • Reusing one threshold across a fleet. Two similar machines can draw different currents, and worn parts change the numbers.
  • Ignoring product changes. A cycle-based timer needs the ideal cycle of the running product. Read how to set the ideal cycle time first.
  • Reading too slowly. A signal read once a minute can jump straight over a 20-second stop.
  • Asking signals for the reason. Signals say when a machine stopped, not why. Give operators a short list of downtime reason codes.

What to do next

Pick your worst-understood machine and list the signals you can reach today. Write its rule table, test it for a shift, and only then copy the pattern to the next machine.

Once the states hold up, running time gives you availability. Put it into the availability calculator as a first check.

See it live in Fabrico

A clean state model still leaves the why open. Fabrico, a light MES for OEE, takes stops, speed and counts live from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line. See how Fabrico reads machine signals.

What people ask next

What is the difference between idle and stopped?

Teams define this themselves, so write it down. A common split is that idle means powered and ready with nothing moving, while stopped means the machine should be producing and output has halted. Both count as lost time when they fall inside scheduled time, as planned vs unplanned downtime explains.

Can I detect machine state from a part counter alone?

Yes, for running versus not running. If no part arrives within a few ideal cycles, the machine is not producing. You cannot tell a fault from waiting without another signal, so add a current sensor or a fault input when you can.

How often should machine signals be read?

Read fast enough to catch the shortest stop you care about, often every second or two. If the device can report changes as they happen, that is better still. PLC signals for OEE works through what slow reads miss.

How long should a machine be stopped before it counts as a stop?

Choose a threshold tied to the cycle, such as a few ideal cycles, and keep it the same on comparable machines. Shorter gaps then show up as a speed loss. Stop threshold for OEE covers how to choose it.

Can a signal tell me whether a stop was planned?

Usually not by itself. A mode input from the controller can separate setup from production, but breaks and meetings often look like any other stop. A schedule or an operator tag fills that gap.

See OEE per line and shift on your own machines

Fabrico is a light MES for OEE. It takes stops, speed and counts from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line.

Book a Fabrico demo

← More OEE Lab guides · Set a stop threshold →

OEE essentials: OEE calculator · What is OEE · Six big losses · Availability · TEEP · Cost of downtime · OEE benchmark · Free OEE tracker · Free OEE software · How to track downtime · How to implement OEE · Light MES vs full MES