Every plant I have worked in has a story about the machine that ran beautifully during the acceptance run and fell over in week two of production. The supplier came in, ran it for an afternoon at nameplate rate, produced good parts, packed up and left. Three months later the line delivers two-thirds of promised output and everyone argues about whose fault it is.
The problem is almost never deception. It is that a short run at rated speed is not a capacity verification at all. Thermal growth has not stabilised, tool wear has not progressed, chip evacuation has not been stressed, operators have not fatigued, and the accumulation of small stoppages has not had time to express itself. Sign off capacity on that basis and you have proven only that the equipment is not broken on day one.
Real capacity verification is a sustained run, planned like an experiment, with defined acceptance criteria for rate and quality simultaneously. The word simultaneously matters. Rate without quality is scrap production at speed; quality without rate is a museum. You need both, held together over hours, under realistic conditions, with the failure modes allowed to develop.
Sustained Runs: Duration, Sample Points and What Counts
The duration must be long enough to surface the degradation mechanisms that matter on that specific process. For a machining cell, thermal equilibrium alone can take a couple of hours of cutting; tool life on a finishing insert may only reveal its behaviour after several hundred parts. For an automated assembly line, the relevant horizon is often shift-length, because that is when operator changeover, bin replenishment and maintenance windows intersect. In practice I plan verification runs of at least one full shift, sometimes two, and for bottleneck equipment I have insisted on a full day.
Structure the run with defined sample points. Do not simply count parts at the end. Record cumulative good-part count at fixed intervals — every thirty minutes is a workable default — so you get a rate curve rather than a single average. A process that delivers the target count but shows a declining rate across the shift is telling you something the average conceals. Log every stoppage with a timestamp, a duration and a cause code, even if the cause is unknown. Interstitial losses are the silent killers of capacity.
Good-part count is the only count that matters. Set the acceptance criterion as sustained good parts per hour at defined quality thresholds, measured at the inspection points you would actually use in production. If the run produces the rate but only because inspection was skipped or relaxed, you have verified nothing. Quality gates run during the run, at production frequency, with production-qualified gauges and production-qualified people.

The First-Hour Phenomenon
The first hour of any run is the best hour, and it is the worst possible basis for a capacity claim. Cold or freshly maintained equipment running fresh tooling on a full hopper of material, with an engineer standing beside it, is not the steady state. It is the best case. Capacity verified in the first hour is capacity you will see on a good Monday morning after a thorough weekend service.
What degrades is predictable. Machining spindles and fixtures drift with thermal growth; the first parts can be dimensionally excellent, then shift as temperatures stabilise, sometimes across a tolerance boundary. Moulding tools fill differently as cavity temperatures settle. Welding heads accumulate spatter and contact wear, changing energy delivery. Adhesive nozzles wear or partially clog, changing bead geometry. Vibratory feeders heat up and change part presentation. None of this shows up in hour one.
So when a supplier offers a capacity demonstration, look at the clock. If the demonstration is scheduled to last one hour, decline it. If the rate data comes from a run shorter than the process's dominant time constant — thermal, tool life or shift pattern — treat it as marketing. Insist on the full-duration run and insist that the data be reported as a time series, so the first-hour advantage is visible and quantifiable rather than averaged away.
Bottleneck Behaviour Under Load
Capacity is a property of the system, and the system's capacity is the bottleneck's capacity. But bottlenecks behave differently under sustained load than in isolation, which is where isolated machine capability studies mislead. A station that cycles at the required rate when fed by hand, in a clean state, with a technician clearing every jam, will not hold that rate when integrated upstream and downstream under real load.
Under load, queues form, and queues change behaviour. A buffer that fills changes the failure logic of the line: the bottleneck stops being starved and starts being blocked, and blocking has its own dynamics — accumulation and release, uneven loading, reflow of parts into fixtures designed for steadier presentation. Watch the buffer states deliberately. Run the line until the buffer before the constraint fills and observe whether the constraint maintains rate with a full buffer behind it, or degrades.
Load also stresses the support systems that isolated testing never touches: chip conveyors, coolant filtration, compressed air supply, pallet changers, vision system buffer memory, label applicators. Every one is invisible at low utilisation and becomes a constraint at high utilisation. During the run, deliberately do not baby the line. Let bins empty and get replenished on the normal schedule. Let operators take their actual breaks with the actual relief coverage, or none, as production intends. A capacity number that requires continuous attention from a process engineer is fiction.
Degradation Mechanisms to Measure Specifically
Cumulative good-part rate at intervals is the headline. Beneath it, measure the process variables that drive the degradation you expect. For machining: dimensional drift on a critical feature traced against part sequence, spindle load trend, tool wear offsets applied and when. For moulding: cycle time per shot, cavity pressure curves, part weight trend — part weight is a cheap integrator of process stability and often the first thing to move. For joining: torque-angle signatures trended across the run, electrode dress frequency, tip life consumed.
Stoppage analysis deserves its own rigour. Classify every stop into adjustment, material refill, jam, quality check, unplanned maintenance and other. The other category starts large; it shrinks as you learn the machine. What you are looking for is the pattern of small stops — five seconds here, twelve seconds there — that individually look trivial and collectively consume the margin between demonstrated and sustained rate. A line can lose the equivalent of an hour of production per shift to micro-stops without a single stoppage appearing in the maintenance log.
Quality degradation must be measured on the same time axis as rate. Plot rejects and rework per interval alongside good-part count. If rejects climb in the back half of the run while rate holds, the process is holding speed by consuming quality. Conversely, if rate falls while quality holds, the process is protecting quality by slowing — which some controls do automatically and quietly. The simultaneous plot tells you what neither number tells you alone.
Instrumenting a Capacity Verification Run
- 01Define criteria firstRate and quality thresholds written and agreed before the run starts.
- 02Run at realistic loadReal staffing, real replenishment, buffers allowed to fill and drain.
- 03Log continuouslyCumulative good parts per interval plus timestamped stoppage causes.
- 04Trend process variablesDrift, tool offsets, signatures — plotted on the same time axis.
- 05Judge the final halfRolling average over the settled portion, not the honeymoon.
Acceptance Criteria That Hold Up
Write the acceptance criteria before the run, and make them bind rate and quality together. A well-formed criterion looks like: sustained good-part output of not less than X parts per hour, measured as a rolling average over the final half of the run, with total rejects not exceeding the production-quality plan limits and no adverse quality trend across the run duration. The rolling-average-over-the-final-half formulation is deliberate: it prevents the first-hour advantage from carrying the result and forces the process to demonstrate settled behaviour.
Build in a repeat requirement. One good run can be luck; two consecutive shifts meeting criteria is evidence. Define a requalification trigger as well: if the process undergoes significant change — new tooling design, control software update, relocation, a new material lot source — the sustained run repeats. Capacity claims have a shelf life, and it expires with the first significant change.
Document the conditions under which the run passed, honestly: ambient temperature, material lot, operators by name or shift, maintenance performed immediately before. A capacity number without its conditions is not a number; it is a slogan. When the line later fails to reproduce the rate, the first question is always which condition differed, and if you did not record them, you cannot answer.
A capacity number without its conditions is not a number; it is a slogan.
What To Do When It Fails
Most verification runs fail. That is the point of them. Treat failure as diagnostic data, not as a negotiation. When rate degrades across the run, walk the degradation curve backwards to its onset and identify what changed at that timestamp — a tool offset, a temperature threshold, a buffer state, a shift change. The time-stamped stoppage log and the process-variable trends are what make this possible; without them you are guessing.
Distinguish three failure types, because they demand different responses. Rate failure with stable quality is a speed problem: tool life, thermal drift, micro-stops — engineering work on the constraint. Quality failure at rate means the process window is too tight at speed: slow it down or widen the window through process work. Simultaneous degradation means the process is fundamentally marginal at the claimed rate, and no amount of tweaking will hold it — re-rate the equipment or redesign the process step.
Resist the pressure to accept conditional sign-off with a list of open actions to be closed before production. In my experience those actions are never resourced once the equipment is paid for, and the capacity shortfall becomes the plant's problem instead of the supplier's. The sustained run exists precisely to force these issues into the period when someone still has an incentive to fix them. Hold the line on it.
Demonstration Versus Verification
Capacity demonstration
- One afternoon at nameplate rate
- Single average part count
- Engineer attending the machine
- Relaxed or skipped inspection
- Conditions undocumented
Capacity verification
- Full shift minimum, realistic load
- Rate curve at fixed intervals
- Production staffing and replenishment
- Quality gates at production frequency
- Criteria and conditions written beforehand
