Inline Measurement Cycle Time: How Many Stations Do You Need?

POMEAS optical applications team · Reviewed by Tiger Xie · Published 2026-10-14

Quick answer. To size an inline measurement system you need three numbers you already have and one you must choose. You have the target throughput (parts per hour), the per-part time the measurement actually occupies, and the hours the line really runs. You must choose the availability figure. The station count is then N = target parts per hour × per-part time in seconds ÷ (3600 × availability), rounded up. Every arithmetic result in this article is a calculation on stated assumptions, not a measured performance figure. Every POMEAS number quoted is taken from published product specifications or from an already-published POMEAS page. This page is about capacity arithmetic and station count; the upstream decision of whether to measure inline or offline at all is covered in Inline vs Offline Optical Measurement.

1. Three numbers people mix up: takt time, cycle time and throughput

Most cycle-time arguments in a project kick-off are really vocabulary arguments. Fix the three definitions first and the arithmetic becomes uncontroversial.

Term	What it is	Who sets it	Unit
Takt time	The pace the customer demand requires: available production time divided by the quantity the customer wants in that time.	The customer and the shift calendar. It is a target, not a capability.	seconds per part
Cycle time	The time one part actually occupies a resource, from the moment it becomes available to that resource until the resource is free for the next part.	The process design: handling, exposure, processing, handshakes, reject handling.	seconds per part
Throughput	How many good parts the station or line actually delivers per hour once availability is included.	The combination of cycle time, station count and availability.	parts per hour

The relationship that matters for sizing is simple: a station can only deliver 3600 divided by its cycle time parts per hour, and only for the fraction of the hour it is actually available. Takt time tells you what you must deliver. Cycle time tells you what one station can deliver. The gap between them is what you fill with more stations, better overlap, or a redesign of the part flow.

2. What actually consumes time on one part

A single-station figure quoted as an exposure time is not a cycle time. The per-part budget has six terms, and in inline systems the first and last two are usually the ones that decide the answer.

#	Term	What it contains	Who controls it	Typical order of magnitude
1	Handling	Load, locate or clamp, unload. Zero if the part never leaves the production flow.	Mechanics and fixture design	seconds — often the largest term when the part must be moved
2	Trigger and handshake	Part-present signal, trigger latency, PLC exchange, strobe delay.	Electrical and control design	milliseconds to tens of milliseconds
3	Exposure	The time the sensor is actually integrating light.	Light source, lens aperture, motion blur limit	microseconds to milliseconds
4	Readout and transfer	Sensor readout, interface transfer, buffer write. Captured by the acquisition rate.	Camera and interface	milliseconds, or seconds if the software waits for a full file to be written
5	Processing and decision	Locate, measure, tolerance check, log. Usually the true limit of a large-field multi-feature measurement.	Software and CPU or GPU budget	milliseconds to seconds, growing with feature count
6	Reject and rework handling	Divert, mark, re-inspect, operator disposition of a doubtful result.	Conveyor logic and quality procedure	seconds per reject, and it is the term most often forgotten

Terms 2 to 5 can overlap with each other to some degree — a camera can read out while the processor starts on the previous frame. Term 1 overlaps with the production process only if the part never stops, and term 6 can never be assumed away just because the yield is good.

A useful discipline is to write the budget for the worst part, not the average part. Cycle time is set by the slowest part the station has to handle, because the line must not back up when that part arrives.

3. From a throughput target to a station count

Once you have a target throughput and a per-part time, the station count is arithmetic:

N = Q × T ÷ (3600 × A), rounded up to the next whole station.

Symbol	Meaning	Unit	Where the number comes from
N	Number of parallel measurement stations required	stations	The result. Always round up.
Q	Target throughput	parts per hour	The production plan, including 100 percent inspection scope
T	Per-part time the measurement occupies one station	seconds	From the budget in section 2, using the worst part
A	Availability: the fraction of scheduled time the station is actually ready to measure	0 to 1	A decision you make and then verify. See section 4.

Two properties of this formula are worth stating plainly. First, it is linear in both Q and T, so halving the per-part time halves the station count — which is why the measurement and processing terms, not the camera resolution, are usually where the effort pays off. Second, rounding down is never neutral: a fractional result below 1.0 with a single station means the station has no recovery margin at all, and a fractional result of 1.2 means one station will be saturated and the line will wait.

The formula sizes a parallel-station architecture. There is a cheaper architecture available whenever the measurement can happen while the part keeps moving, and section 6 deals with that case separately.

4. Availability: the term most projects leave out

Availability in this context is not a vendor specification. It is the fraction of scheduled line time during which the measurement station is ready to accept a part: no tooling adjustment, no software restart, no waiting for an operator to clear a questionable result, no blocked flow because the downstream diverter is full.

It is a project input, and the honest way to use it is to state the assumption in writing and then plan to measure it during ramp-up. The worked example in section 5 uses 0.85 for a manually loaded offline station and 0.90 for an automated inline station. Those are stated assumptions for the arithmetic, not POMEAS performance figures, and they should be replaced with your own data as soon as you have four weeks of running history.

The practical consequence of A is easy to underestimate because it appears in the denominator. Dropping from 0.95 to 0.85 increases the required station count by about 12 percent for the same throughput. On a multi-station cell that is an entire extra head, and it is cheaper to design out the cause than to buy the head.

5. Worked example: 2,000 parts per hour

Take a line that must inspect 2,000 parts per hour, one shift of 7.5 productive hours. The same target is applied to three architectures so the difference in station count is visible. All arithmetic below is a calculation on the stated assumptions.

Architecture	Per-part time T on one station	Assumed availability A	Stations required, N = Q × T ÷ (3600 × A)	What sets the result
Offline, operator loaded	≈ 20 s (about 12 s load, locate and unload plus about 8 s measurement)	0.85	2000 × 20 ÷ (3600 × 0.85) = 13.1, so 14 stations	Handling dominates; the head count needed is not commercially realistic
Inline, part stops at the station	≈ 2.0 s (handling reduced, exposure and processing unchanged)	0.90	2000 × 2.0 ÷ (3600 × 0.90) = 1.23, so 2 stations	Two stations with load balancing, or one station plus a buffer
Inline, measurement overlaps the process	≈ 0.45 s (no handling term; exposure plus acquisition plus decision only)	0.90	2000 × 0.45 ÷ (3600 × 0.90) = 0.28, so 1 station	The measurement is not the bottleneck; line takt decides throughput

Three readings of this table are worth keeping.

• The offline row is the reason inline exists. At the 2,000 parts per hour target the offline architecture is not a cost comparison, it is infeasible — which matches the conclusion of the inline versus offline selection guide from the other direction.
• The middle row is the one that catches projects out. A single station sized at 0.28 in the last row looks comfortable until the part has to stop, at which point the requirement doubles. If the fixture design is still open, the overlap question is worth more engineering time than the camera choice.
• None of these rows is a throughput promise. Real throughput is verified by timestamping the trigger and the result output on the actual line, at production speed, for at least thirty parts, and taking the worst part rather than the mean.

6. When measurement overlaps the process, and when it cannot

The single-station case in the last row of section 5 is available only under specific conditions. It is worth stating them as a checklist rather than a principle.

Condition	Overlap is achievable	Overlap is not achievable
Part motion	The part passes the field of view at constant speed and is measured in motion	The part must stop, be clamped, or be lifted into position
Number of views	One view contains all features, or the additional views are on separate parallel heads	One station must image several faces in sequence, so the part is occupied for each view
Exposure versus speed	The required exposure at line speed still meets the motion blur limit	The exposure needed for the light budget smears the image at line speed
Result timing	The decision is available before the part reaches the diverter	The part passes the diverter before the software decides, so the part must be held or the station duplicated downstream
Reject policy	A reject can be diverted without stopping the line	A doubtful result requires an operator, and the operator becomes the bottleneck

Reading the table column by column is the fastest way to find out which architecture you are actually designing. A single failing entry in the right-hand column does not automatically force a parallel station, but it does mean the per-part time budget in section 2 must include the term that the failure introduces.

7. What published acquisition figures do and do not tell you

Camera and system datasheets give acquisition rates, and acquisition rates are only one of the six terms in section 2. The table below shows the published POMEAS figures and what has to be added before they can be used in the formula.

Published figure	Source	Arithmetic	What is still missing
Full-field acquisition at approximately 30 fps	HM series system specification, as published on the HM inline measurement system page	1 ÷ 30 ≈ 33.3 ms per frame	Handling, trigger and handshake, processing and decision, reject handling
Local 1/8-field acquisition at up to approximately 200 fps	HM series system specification	1 ÷ 200 ≈ 5 ms per frame	The same terms, plus the time to move between the eight sub-fields if they are not imaged in parallel
Exposure time 35, 70 or 150 μs depending on model	HM series system specification	0.035 to 0.15 ms	This is the exposure term only. It is the smallest term in the budget unless motion blur forces it.
Repeatability from ±0.05 μm to ±0.2 μm across four measuring ranges	HM-1015, HM-1040, HM-1065, HM-1120 published tables	—	Repeatability is a quality figure, not a time figure. Increasing it usually costs time, not saves it.

The trap is the comparison between 33.3 ms and a per-part time of two seconds. The difference is not camera performance, it is handling, communication and decision logic. Any project that tries to shorten the cycle time by buying a faster camera while term 1 and term 6 are untouched will spend money and move the number very little.

Where the measurement is genuinely limited by the software rather than the sensor, POMEAS HM systems publish more than 100 measurement tools and a high-precision area with positional accuracy up to ±0.3 μm. Feature count drives the processing term, so a large-field measurement of dozens of features is a processing problem before it is a camera problem — the same distinction made in Magnification Is Not Resolution.

8. How to measure the per-part time instead of estimating it

The formula is only as good as T. Estimating T from a datasheet is the usual source of a project that is late and over budget. Four steps replace estimation with evidence, and none of them requires new hardware.

1. Timestamp the two ends. Record the trigger edge and the moment the pass or fail decision is written. The difference is the true measurement residency for one part, at production speed, with production lighting.
2. Take the worst part, not the mean. Run at least thirty parts and keep the maximum. A station sized on the mean fails whenever the difficult part arrives, and the difficult part always arrives.
3. Repeat after a shift change and after a material change. If T moves by more than a few percent, the availability term is absorbing a variability that belongs in T.
4. Time the reject path separately. Divert, mark, re-inspect and operator disposition are invisible in a normal takt measurement because they do not happen on every part, yet they set the worst-case cycle.

Where the part cannot be moved to a test bench, the same four steps can be applied to a recorded sequence. The motion blur and conveyor speed guide gives the exposure-side half of this measurement, and the ten-cycle repeatability method gives the pattern for turning a one-off measurement into a verified figure.

9. Six common cycle-time mistakes

#	Mistake	Why it fails	What to do instead
1	Using the camera frame rate as the throughput	Frame rate is one of six terms in the budget and is not the largest one	Build the six-term budget, then measure T on the real line
2	Using availability equal to 1.0	It appears in the denominator, so a 10 percent error becomes more than 10 percent more stations	State A in writing, use a conservative value, verify it during ramp-up
3	Substituting exposure time for per-part time	Exposure can be 0.1 percent of the real per-part time in a handled architecture	Time the trigger-to-result interval, not the exposure line on the datasheet
4	Counting the measurement and forgetting the movement	Multi-view parts, flips and tray exchanges add whole seconds that appear nowhere in the optical specification	Write the movement terms into the budget explicitly, per view
5	Rounding the station count down	The line must never wait; a fractional result means saturation	Round up, then check whether the last station is better spent as a buffer or a second head
6	Sizing on the average part	The line is paced by the slowest part it must handle	Size on the worst part and record which part that was

10. Frequently asked questions

What is the difference between takt time and cycle time?

Takt time is the pace your customer demand requires: available production time divided by the quantity required in that time. Cycle time is how long one part actually occupies a resource. Takt time is a target set by demand. Cycle time is a property of your process. If cycle time is longer than takt time, one station cannot meet demand and you either add stations or reduce the cycle time.

How do I calculate how many measurement stations I need?

Multiply the target throughput in parts per hour by the per-part time in seconds, then divide by 3600 multiplied by the availability fraction, and round up. For 2,000 parts per hour, a two-second per-part time and an availability of 0.90, the result is 2000 × 2.0 ÷ (3600 × 0.90) = 1.23, which means two stations. The same target with a 0.45 second per-part time gives 0.28, which means one station.

Does a faster camera always shorten the cycle time?

No. Acquisition is one of six terms in the per-part budget. In architectures where the part must be handled, the handling term dominates, and in multi-feature measurements the processing term often dominates. A faster camera moves a term that may be a small fraction of the total. Measure the budget before buying acquisition rate.

Can one inline measurement head keep up with a 2,000 parts per hour line?

It depends on whether the measurement overlaps the production process. At 2,000 parts per hour the line produces one part every 1.8 seconds. If the measurement can be made in motion with no handling term and the trigger-to-decision interval is under 1.8 seconds, one head is sufficient and the line takt sets throughput. If the part must stop, and the per-part time rises to about two seconds, the arithmetic returns 1.23 stations, so one head is not sufficient.

What availability figure should I use in the calculation?

There is no published figure that applies to your line. Availability in this calculation means the fraction of scheduled time the station is ready to accept a part, so it must come from your own production data or be stated as an explicit assumption. The worked example in this article uses 0.85 for a manually loaded station and 0.90 for an automated one, and both are labelled as assumptions for the arithmetic rather than as measured performance.

Why is the handling time often the largest term?

Because it is measured in seconds while the optical terms are measured in milliseconds or microseconds. Loading, locating, clamping and unloading a part is mechanical work, and it does not become faster when the camera becomes faster. This is why the single largest cycle-time improvement in most projects comes from removing the handling step — measuring the part while it is still in the production flow — rather than from the imaging chain.

How do I measure the per-part time instead of estimating it?

Record a timestamp at the trigger edge and a second timestamp when the pass or fail decision is written, at production speed and with production lighting. Repeat for at least thirty parts and keep the maximum, not the mean. Then repeat after a shift change and a material change, and time the reject path separately, because reject handling does not appear in a normal takt measurement.

Does 100 percent inspection require a separate station from the production process?

Not necessarily. If the measurement can be made while the part stays in the flow and the decision arrives before the part reaches the diverter, the measurement shares the production station and the line takt decides throughput. A separate station becomes necessary when the part has to stop, when the decision arrives too late for the diverter, or when a doubtful result needs an operator.

What happens to the calculation if a part needs more than one view?

Each additional view adds either a handling term or an occupancy term to the same station, and the station count is linear in the per-part time. Two sequential views therefore roughly double the per-part time on a station that must reposition the part between views. If the additional views can be placed on separate parallel heads on the same trigger, the extra time is absorbed by parallelism instead of by serial occupancy.

Are the numbers in this article measured values?

No. Every arithmetic result in this article is a calculation on the assumptions stated next to it, and is labelled as such. The POMEAS figures quoted — approximately 30 fps full field, up to approximately 200 fps on one eighth of the field, exposure time of 35, 70 or 150 μs depending on model, more than 100 measurement tools and positional accuracy up to ±0.3 μm in the high-precision area — are taken from published POMEAS specifications. Replace the assumptions with your own timestamps before making a purchasing decision.

11. Related POMEAS resources

• Inline vs Offline Optical Measurement: How to Choose the Right Inspection System
• Motion Blur in Machine Vision: Exposure Time and Conveyor Speed
• Motorized Zoom Lens Repeatability: 10-Cycle Test Method
• Magnification Is Not Resolution: What Actually Determines the Smallest Visible Feature
• Object-Side Resolution vs Measurement Accuracy
• Camera Module Die-Cut Measurement: 60 s to Under 5 s per Part
• POMEAS HM Telecentric Measurement System for In-line Inspection
• POMEAS IMAGE 3 Instant Vision Measuring System
• Machine Vision Light Source Selection: Point, Ring, Telecentric Parallel and Coaxial
• FA Industrial Lens Selection Guide: Sensor Size, FOV, WD and Accuracy

If you have a sample part, a cycle-time target and a production speed, send them through Contact and we can size the measurement step with you.

Talk to a POMEAS Engineer About Cycle Time and Station Count

If the line speed and the part are both fixed and the measurement step is the open question, send the target throughput in parts per hour, the number of parts per shift, the part size, the number of dimensions to be measured, whether the part can be measured in motion, and the permitted time between the part arriving and the decision being needed. We can start from the per-part time budget rather than from a product name.

Email: [email protected] | WhatsApp: +86 185 9810 2007

Reviewed by Tiger Xie.