Andon is a lean manufacturing signaling system that lets an operator or a machine raise a problem the instant it appears, and commits someone to respond. The classic form is a cord over the station and a colored light tower above the line: pull the cord, the light changes, a supervisor or technician comes. The Japanese word means lantern, and the point of the lantern is visibility, making a problem that used to be one operator's private struggle into something the whole area can see and act on within seconds instead of hours.
What andon actually does on a line
The value is not the announcement. It is the compression of the gap between when something goes wrong and when someone competent is standing at the machine. Without a signal, an operator with a jam, a bad lot, a starved feeder, or a spec question has three bad options: fix it themselves and lose time, run anyway and make scrap, or walk off the station to find a supervisor and stop producing while they search. Andon replaces all three with a two-second action that keeps the operator at the station and puts the burden of travel on the person whose job is to help.
The second thing it does is make problems countable. A pull is a timestamped event with a station, a reason, a response time, and a duration. Fifty pulls a week on the same station is an engineering ticket that argues for itself. A plant that has never run andon usually cannot say which station causes the most lost minutes, because nothing was ever recorded except the shift total, and the shift total blames the whole line for one bearing.
Andon signals and what they usually mean
- Green: running normally, no help needed. Most plants treat green as background and rarely look at it.
- Yellow: a call for help without stopping, low on material, a quality question, an operator who needs a second pair of hands. Yellow is the signal that saves the most money because it prevents a stop rather than reporting one.
- Red: the line is stopped or about to stop. Something is wrong that the operator cannot resolve, and production is not continuing until it is fixed.
- Blue or white, where used: plant-specific, often quality hold, maintenance request, or material handling. The exact color matters far less than whether everyone on the floor reads it the same way.
- Machine-generated states: faults, e-stops, starved and blocked conditions the PLC already knows about, which can drive the same tower without anyone pulling anything.
Why most andon systems quietly die
Failure is rarely technical. A plant buys the towers, wires the cords, trains the operators, and within a quarter nobody pulls. The reasons repeat across floors:
- Nobody answered. An operator pulls, waits eleven minutes, solves it alone, and learns the lesson. Two or three of those and the cord is decoration. Response time is the whole system; if you measure one thing, measure that.
- Pulling had a social cost. If a red light triggers a conversation about the operator rather than about the machine, pulls stop and the problems keep happening silently.
- The signal went nowhere but the ceiling. A tower is only visible if someone is in the building looking at it. Supervisors in meetings, techs on the far end of the plant, and the night shift with thin coverage all miss it.
- No record survived. If a pull leaves no trace once the light goes out, the plant loses the pattern data that would have justified fixing the root cause, and keeps paying the same interruption forever.
That last point is the one that turns andon from a floor practice into a data question. The cord tells you about right now. The log tells you which right-now keeps repeating. Plants that get real value out of andon are usually the ones that kept the record, and the record is where digital versions earn their keep. We walk through that shift in more detail in the guide on digital andon systems and pulling the cord without the cord, including what changes when the machine raises the call itself.
Andon versus a stack light, and other confusions
A stack light is hardware. Andon is a process that often uses a stack light. Almost every plant already has towers on its machines; very few have an escalation contract behind them, which is why walking a floor and seeing red lights burning with nobody moving is so common. The same distinction applies to alarm screens on an HMI: an alarm that fires into an empty room is not andon, because nothing is committed to answering it.
Andon is also not the same as stopping the line, though the two are linked. Jidoka, the principle of stopping to build in quality, is what gives an operator the authority to halt production rather than pass a defect forward. Andon is how that authority is exercised in practice. A plant that installs cords but keeps punishing stops has bought the signal without the principle, and it will not work.
What good looks like
On a healthy floor the calls are frequent and short. Frequency is a good sign, not a bad one, because it means people trust that pulling helps. Response is measured in single-digit minutes and tracked by shift, not by quarter. Every call leaves a record with a station, a reason code, a responder, and a duration, and someone reviews the top reasons weekly and actually closes a few of them. Yellow outnumbers red, because most problems are being caught before they become stops. And the reason codes are written by the floor, not by a consultant, so operators pick the right one instead of defaulting to Other.
The honest caveat is that none of this is a technology problem first. A plant with a disciplined supervisor, a whiteboard, and a genuine commitment to answer every call will beat a plant with an expensive system and no accountability. Software helps once the response habit exists, mostly by routing calls to phones instead of ceilings, by capturing machine faults nobody has to notice, and by keeping the history the improvement work depends on.
Where Harmony fits
Most of what makes andon digital is plumbing, and plumbing is what we do on-site. Harmony's engineers connect at the PLC, Allen-Bradley and Rockwell, Siemens, Omron, Mitsubishi, over OPC UA or whatever the machine actually speaks, so faults, starved and blocked states, and cycle stops raise calls without anyone pulling anything. Operator-initiated calls come from a tablet at the station with reason codes the plant writes, and the call routes to a named responder with an escalation timer behind it. Everything lands in a record you can query, so the weekly conversation about the top three interrupters starts with data instead of memory. We are software and hardware agnostic, so the towers and PLCs you already own stay where they are.
This shows up most clearly in high-production plants where one line's downtime is measured in units per minute. In powersports and recreation manufacturing, where mixed-model assembly and long serialized builds make a five-minute unanswered call cascade down the line, the escalation record usually surfaces two or three stations that were quietly eating the shift. Our published pilot is $15–20K one-time over 4–6 weeks with working software by week three, and the pattern throughout is the same one andon has always run on: the system proposes and surfaces, a person decides and answers.