Understanding Machine Downtime Tracking
Every plant knows its machines stop. Far fewer know exactly how often, for how long and why — and without that, the same losses repeat shift after shift.
Ask a supervisor why a machine produced less than planned and the answer is usually "it was down for some time". Ask how long, how many times and for what reason, and the answer becomes uncertain. That uncertainty is where most improvement opportunities are hidden.
Machine downtime tracking is the practice of recording when a machine stops, for how long and why, so that losses can be measured, compared and reduced. This article explains what to track, why records often fail to be useful, and how a plant can start with a simple, reliable method before investing in automatic monitoring.
What counts as downtime
Downtime is any period during which a machine was scheduled to produce but did not. That definition already contains an important distinction: time the machine was not scheduled to run — a weekly off, a shift with no planned work — is not downtime. It simply was not available to lose.
Downtime is commonly divided into planned stops (scheduled maintenance, planned changeovers, breaks) and unplanned stops (breakdowns, material shortages, waiting for quality approval). In the OEE framework, losses during planned production time reduce availability, while short stops and slow running are usually counted under performance.
Three categories deserve separate attention on most shop floors:
- Planned downtime: setups, changeovers, preventive maintenance and scheduled breaks. These are expected, but they still have a target duration. A changeover planned for 20 minutes that regularly takes 45 is a real loss. In OEE terms, scheduled breaks and planned maintenance are usually excluded from planned production time, while setups and changeovers still count as availability loss, even when they are planned.
- Unplanned downtime: breakdowns, tool failures, material not available, power cuts, operator absence. These are the stops most plants think of first.
- Micro-stoppages: stops of a minute or two — clearing a jammed part, a quick adjustment, waiting for a crane. Individually they seem trivial and are rarely written down. Collectively, they can add up to a significant part of a shift.
Micro-stoppages are the hardest to capture manually. A paper register realistically records stops longer than a few minutes; shorter ones show up only as a gap between expected and actual output. That gap is a useful signal in itself, as discussed in why production visibility matters.
Why stoppage reasons get lost
Many plants do record downtime. The problem is that the record often cannot answer the question that matters: what should we fix first? Common patterns include:
- Vague reasons. Entries such as "problem", "maintenance" or "machine issue" describe that something happened, not what. "Maintenance" might mean a spindle bearing failure, a coolant top-up or waiting for the maintenance technician to arrive.
- Totals instead of events. Writing "downtime 65 min" at the end of the shift hides whether it was one long stop or ten short ones — two very different problems.
- Recording from memory. When the register is filled at shift end, durations are rounded and smaller stops are forgotten.
- Free text that cannot be totalled. "Tool broke", "insert change", "tool problem" and "tooling" may all mean the same thing, but they cannot be added up without someone re-reading every entry.
- No visible use. If nobody discusses the record, operators reasonably conclude it is paperwork, and the quality of entries drops.
A practical reason-code list
A reason code is a short, agreed category for why a machine stopped. The aim is not to describe every situation perfectly, but to make stops comparable. A list that is too long will not be used consistently; a list that is too short ends with everything under "other".
A starting list for a machining or assembly area might look like this:
| Code | Reason | Typical examples |
|---|---|---|
| BD | Breakdown | Mechanical, electrical, hydraulic or control failure. |
| SU | Setup or changeover | Job change, fixture change, first-piece setting. |
| TL | Tool change or tool failure | Worn insert, broken drill, offset correction. |
| MT | Material not available | Raw material, blanks or previous-operation parts not ready. |
| QC | Quality hold or inspection | Waiting for first-piece approval, rework decision. |
| OP | Operator not available | Absence, operator moved to another machine. |
| PW | Power or utilities | Power cut, compressed air pressure, coolant supply. |
| OT | Other (describe) | Anything else — reviewed weekly to see if a new code is needed. |
Two rules make such a list work in practice. First, review the "Other" entries every week; if the same description appears repeatedly, it deserves its own code. Second, keep the list identical across shifts and machines in the same area, otherwise comparisons become meaningless.
What a downtime record should capture
A useful downtime entry answers six questions. Each one exists for a reason:
- Machine: which asset stopped, using a consistent machine ID rather than a nickname.
- Start time: when the stop began. This reveals patterns such as stops clustering at shift start or after lunch.
- End time: when production resumed. Duration is calculated from start and end, not estimated.
- Reason code: from the agreed list, plus a short note when needed.
- Who recorded it: so questions can be clarified while memory is fresh.
- Action taken: what was done to restart — and whether a permanent fix is still pending.
This can be done on a printed sheet next to the machine just as well as on a tablet. The format matters less than the discipline of recording each event when it happens.
Finding the largest loss
Once stops are recorded with consistent reasons, they can be totalled and ranked. This is often called Pareto analysis: sorting causes from largest to smallest, on the observation that a few causes typically account for a large share of the total loss.
A CNC cell records 65 minutes of downtime in one shift. Logged as a single total, it offers no direction. Logged as events with reason codes, it might look like this:
- Material not available — one stop, 25 minutes
- Tool change — three stops, 18 minutes in total
- Setup or changeover — one stop, 15 minutes
- Quality hold — two stops, 7 minutes in total
The material wait is the single largest loss and is not a machine problem at all — it points to coordination with stores or the previous operation. The three tool changes suggest checking whether tool life is being planned or tools are run until they fail.
One shift is not enough to draw conclusions. The same ranking repeated over two or three weeks, however, usually makes the priority obvious. The useful habit is to pick only the largest repeating loss, act on it, and check whether its total falls — rather than trying to address every category at once.
Also look at frequency, not only minutes. Ten five-minute stops and one fifty-minute stop have the same total but require different responses: the first often points to a process or tooling issue; the second to a breakdown or supply failure.
MTTR and MTBF
Two maintenance metrics become possible once breakdowns are recorded as individual events with start and end times.
MTTR (Mean Time To Repair) is the average time taken to restore a machine after a failure: total repair time divided by the number of failures. It reflects how quickly breakdowns are diagnosed and fixed, including, as used here, the time spent waiting for a technician or spare part.
MTBF (Mean Time Between Failures) is the average operating time between one failure and the next: total operating time divided by the number of failures. A rising MTBF indicates the machine is failing less often.
Both metrics depend entirely on the quality of the underlying record. If breakdowns are mixed with setups under "maintenance", or durations are estimated at shift end, MTTR and MTBF will look precise while being unreliable. For most small plants, consistent event recording should come before any attempt to report these figures.
Getting started on one machine
Pick the machine whose output most often decides whether the shift meets its plan, and start there. None of these steps needs new software:
- Choose one bottleneck machine — the one whose output most often limits dispatch.
- Print a simple log sheet with columns for start time, end time, reason code, note, name and action.
- Agree on the reason-code list with operators and supervisors of all shifts, and display it beside the machine.
- Record every stop above five minutes as it happens, not at shift end.
- Compare logged downtime with the output gap. If the machine produced 80 parts less than planned but logged downtime explains only half of that, the rest is likely short stops or slow running.
- Review the top reason weekly and agree one action against it.
If recording turns out to be the bottleneck, that is a useful finding too. It is one of the reasons plants move from paper to digital entry.
Hypotheses and open questions
The points below describe what Shopfloor Insights is investigating. They are not conclusions.
- In small and mid-sized plants, downtime records are more often limited by vague reasons than by missing entries.
- A reason-code list of about eight to ten items, reviewed weekly, is recorded more consistently than a detailed list of twenty or more.
- A meaningful share of downtime in machining cells is caused by waiting — for material, inspection or a technician — rather than by machine failure itself.
- What is the shortest stop duration that operators can realistically record by hand without disturbing production?
- How much of the gap between planned and actual output is explained by recorded downtime in typical Indian machining units?
- Does recording downtime on a phone or tablet at the machine improve reason quality compared with paper, or only speed?
If downtime records in your plant have been useful — or frustrating — your experience would help shape this work. See the full machine downtime challenge, or try the machine downtime reporting prototype, which demonstrates reason-coded logging using sample data.