Why Uptime Flatters Additive Operations
Uptime answers a narrow question: was the machine powered, homed and printing during the shift? It does not ask whether the machine was printing a part somebody needed, whether the build finished with all the parts still good, or whether the chamber was full. A printer that runs one small part for six hours is 100% up and roughly 8% productive.
This matters more in additive than in machining, because the additive equivalent of a workpiece is a build chamber and its utilisation is visible in a way a machining spindle's never is. A chamber loaded to 22% is a machine that will report perfect uptime while consuming 78% of a print window as scrap overhead. Buyers who accept uptime as the supplier's headline metric are accepting a number that cannot fall below roughly 85% in any honest report, which makes it useless as a comparison.
OEE - overall equipment effectiveness - multiplies three factors that each answer a different question. Availability asks whether the cell was ready to produce. Performance asks whether it produced at the rate it should. Quality asks whether what came out was good. The product of three honest numbers in the 70s is a number in the 30s, and that is the correct answer for most additive cells running today.
Defining Availability for a Print Cell
Availability is planned production time minus unplanned stops, divided by planned production time. The controversy is all in what counts as a stop.
For additive, three exclusions are standard and legitimate: no order in the queue, planned maintenance inside the schedule, and material conditioning time that is genuinely outside the operator's control and was budgeted for. Five other events are stops and are routinely excluded anyway, which is where the flattery enters.
| Event | Counted as stop? | Typical share of the gap |
|---|---|---|
| Mid-build failure requiring abort | Yes | 4% - 11% |
| Waiting for a plate change operator | Yes | 3% - 8% |
| Waiting for filament or resin at changeover | Yes | 2% - 6% |
| Recalibration after a failed first layer | Yes | 1% - 4% |
| Firmware or slicer fault | Yes | 1% - 3% |
| No order in queue | No - excluded | 0% - 30% |
| Planned PM inside the schedule | No - excluded | 1% - 3% |
| Material drying already budgeted | No - excluded | 2% - 5% |
The last three exclusions are where a farm can honestly claim 94% availability while producing far less. No-order time is not an availability loss, it is a demand loss, and it belongs in a different conversation about sales and capacity planning. Excluding it is right. Reading 94% and concluding the machinery is healthy is not.
A workable availability definition for additive: (scheduled hours - all stops inside the schedule) / scheduled hours, where scheduled hours are set by the order backlog, not by the calendar. That single change - scheduling to backlog rather than to shifts - makes the metric comparable between a farm running at capacity and one running at half load.
What the answer tells you: if it is calendar hours, availability and demand are mixed into one number, and the metric will look worst exactly when the business is doing well. Buyers comparing two suppliers need the backlog-hour version or the comparison is meaningless.
Defining Performance Without a Nameplate Rate
In machining, performance is measured against a rated cycle time. Additive has no such nameplate - a printer's speed depends on material, layer height, part geometry, nozzle and infill - so performance is usually skipped entirely. That is a mistake, because performance losses in additive are large and specific.
The workable definition uses the slicer's own estimate, corrected by a measured factor. Each part family gets a calibration factor from the first ten builds: actual print hours divided by slicer-estimated hours. Once that factor exists, every subsequent build has an expected duration, and the ratio of expected to actual is performance.
Chamber load below plan is the item that dominates and the one that is invisible in every other metric. Plan called for 14 parts per chamber; the operator ran 9 because the queue was thin or the fixture was not ready. The machine printed for the full duration, uptime stayed at 94%, and 36% of the chamber's value evaporated. This is a performance loss by definition and it is a planning loss in cause.
Slow-mode fallback deserves a note because it hides inside the build. When a printer throttles to protect a failing heater or a marginal flow rate, the build finishes and reports success; the loss only appears as a duration 18% longer than expected. Farms that never compare actual to expected duration never see it.
Defining Quality Including Reprints
Quality is good parts divided by total parts started. The additive-specific complication is that a failed build destroys every part in the chamber, so one event can remove 14 parts from the numerator at once.
Three categories must be counted separately or the number is not actionable:
- Build-loss rejects. The whole chamber is lost - layer shift, thermal event, filament runout mid-build. Typically 3% to 8% of parts in a well-run farm, 12% to 20% in a farm without monitoring.
- Post-processing rejects. The build succeeded, the support removal, sanding or curing ruined the part. Usually 2% to 5%, and invisible if quality is measured at the printer.
- Reprints on customer return. Escaped both inspections. Usually under 1%, but the most expensive per unit because it carries freight and goodwill.
Quality measured only at the printer typically reads 96% to 98%. Quality measured through post-processing and first-article inspection reads 89% to 94%. The difference is not measurement rigour, it is scope: the printer cannot see what the bench does to a part.
What the answer tells you: at the printer, it excludes everything the bench breaks. Ask for the reject count by stage - build, post-processing, final inspection - and the sum is rarely below 6% even in a strong operation.
A Worked Month: 94% Uptime, 41% OEE
Twelve printers, one shift plus a night skeleton, 26 working days, 8,736 scheduled machine hours in the month. The farm reports 94% uptime and 97% quality.
| Factor | Calculation | Result |
|---|---|---|
| Availability | (8,736 - 742 stop hours) / 8,736 | 91.5% |
| Performance | slicer-expected 5,180 h / actual 6,340 h printed h | 81.7% |
| Quality | good parts 38,940 / started 42,110 | 92.5% |
| OEE | 0.915 × 0.817 × 0.925 | 69.1% |
| Chamber-utilisation adjusted | × mean chamber fill 42% / plan 70% | 41.5% |
The 41.5% figure is not a reprimand, it is a description. Of every scheduled hour, 41.5% of its potential output reached the shipping bench. The gap decomposes into 8.5 points of stops, 18.3 points of rate and duration losses, 7.5 points of quality, and 27.6 points of chamber under-fill.
The ordering matters for what to fix first. Chamber under-fill is the largest single bucket and the cheapest to address: it is a queue and nesting problem, not a capital problem. Stops are next and are mostly a staffing and changeover problem. Quality third. Rate last, because it is bounded by physics on a given machine.
A programme that chases purchase of a faster printer before fixing chamber fill is paying capital to solve a scheduling problem, which is the most common misallocation in additive production.
Six Losses, Mapped to Additive
The classical six big losses translate directly once the machine's product is a chamber rather than a workpiece.
| Classical loss | Additive equivalent | Typical magnitude |
|---|---|---|
| Equipment failure | Hotend, heater, board, motor failure mid-build | 2% - 6% of scheduled |
| Setup and adjustment | Plate change, Z calibration, material swap | 4% - 11% |
| Idling and minor stops | Paused build, filament tangle cleared, clog purge | 3% - 7% |
| Reduced speed | Slow-mode fallback, thick layer compensation | 2% - 6% |
| Process defects | Layer shift, stringing, delamination, warp | 5% - 12% |
| Startup losses | First-article tuning, new material qualification | 1% - 4% |
Two of these are additive-specific in their cost structure. Setup loss is larger than in machining because plate changes are manual on most machines and the chamber sits idle throughout. Process defects are larger because a single event consumes the whole chamber rather than one workpiece - which is precisely why build-level monitoring pays for itself in this category and rarely does in machining.
Startup losses are worth tracking separately for a second reason: every new material a farm adds buys a fresh startup loss, so a catalogue of 30 materials carries 30 qualifications per year. Farms that track this number start consolidating material ranges, which improves both OEE and inventory.
Measuring This Without Buying Software
A cell can be measured with the timestamps every printer already records: job start, job end, material used, and result. The minimum viable dataset is four fields per build, exported weekly.
The discipline that makes this work is not the spreadsheet, it is the reason code. Every stop over 15 minutes needs one of six codes, and every rejected part needs one of four. Two weeks of coded data will identify the largest loss bucket reliably; most farms find it is chamber fill, and most had assumed it was machine failures.
The related operational machinery is covered elsewhere: reliability metrics and MTBF for the failure side of availability, job scheduling and queue software for the chamber-fill side, MRB and nonconformance disposition for what happens to the rejects, and pilot-lot capability validation for the statistics behind the quality factor. Where the question is whether the capacity is cheaper in-house or at a bureau, the make-or-buy arithmetic uses the same loaded hour rate.
Operational Review
Want a Baseline OEE Figure for Your Cell Before You Buy More Machines?
Send one month of build records - start, end, estimate, parts good. You will get the three factors, the loss buckets ranked by size, and the capacity that is already paid for but not yet being produced.
