Inline Measurement for Battery Cell and Tab 100% Inspection

A battery line does not fail slowly. A coating die wears, a tab punch dulls, a stack drifts in the fixture, and the parts keep coming at the same rate while the dimension that has just moved out of tolerance continues to be produced. Sampling finds that drift late, and what it finds is not a bad part but a bin of bad parts. An inline measurement head changes the arithmetic: it is mounted over the transfer path, it measures the silhouette of every part as the part passes, and it reports a result per cycle rather than per sample. The output is not a report for the quality file. It is a signal that can stop the line, reject a part or feed an SPC chart while the drift is still one part wide.

This page is about the specific case of a measuring head on a cell, tab or busbar line, and it is written to be usable before a quotation exists. It sets out what a silhouette head can actually see on a cell, how field size decides the model, where the published accuracy figures come from and what they do not include, what the exposure figure implies at real line speed, and the mounting faults that break inline measurement without any optical cause at all. Where a job needs thickness, internal structure or a whole-cell drawing, the page says so and points to the instrument that owns that job, because an inline head sold into the wrong measurement is worse than no instrument at all.

Why a cell line moves from sampling to every-part measurement

Sampling is an economic compromise, and on a battery line the compromise is unusually expensive. A lot is measured by counting: if a thousand parts are made between two samples, then a drift that begins just after a sample can contaminate up to a thousand parts before it is seen, and every one of those parts has to be contained, reworked or scrapped once it is seen. Every-part measurement collapses that window to a single cycle. The instrument does not have to be more accurate than the sampling instrument for that to matter; it only has to be repeatable enough to see a step change, and fast enough not to become the bottleneck.

There is a second reason that has nothing to do with defect detection. Cell dimensions are process outputs, not independent properties. Tab width responds to punch wear, tab pitch responds to the indexing of the tab feeder, can diameter responds to the drawing die, seal width responds to the heat-seal platen. When those dimensions are measured every cycle, the trend is a maintenance signal: a tab-width mean that walks in one direction over a shift is a punch telling you its remaining life. A sampled measurement cannot produce that curve, because between samples the process is unobserved, and a trend fitted through a handful of points per shift cannot separate drift from noise.

The counter-argument is cost and floor space, and it is a real one. An inline head needs a place on the machine, a trigger, a data path and a maintenance routine for its window. On a line that already has a downstream inspection station, the honest question is whether the downstream station is early enough. If the answer is that a defect found at end-of-line still leaves a full buffer of suspect cells, a head at the tab station is worth its position on the machine; if the buffer is two parts, the argument is weaker and the money may be better spent on a bench system used for release and first-article work. The decision is about where the measurement sits in the process, not about which instrument is better in isolation.

What a silhouette head can see on a cell, and what it cannot

An inline head of the HM type does not photograph the part in the way a vision camera does. A collimated green LED emitter on one side of the transfer path throws light past the product, and a receiver on the opposite side images the shadow. What the software measures is the outline: the boundary between light and dark. That design decision determines the entire application list, because a boundary is measurable and a surface is not. Diameters, widths, gaps, pitches, edge distances and corner radii are boundaries and belong to this instrument. Thickness, step height, flatness, porosity and subsurface structure are not boundaries seen from one side, and no amount of software makes them measurable from a silhouette.

The consequence on a cell is a split that is worth drawing explicitly before a specification is written. Most of the dimensional callouts on a cell drawing are outline features in the plane of the part, which is exactly the strength of the head. The items that are not are the ones that usually cause the argument: cell thickness on a prismatic or pouch cell, cap height, the step between the cap plate and the can, the depth of a vent score, and anything inside a weld. Those items have their own instruments, and mixing them into an inline head specification produces a project that cannot be accepted.

Feature on the cell	How a silhouette head sees it	Limit to state in the specification
Can outer diameter (cylindrical)	Two outline edges across the can, measured while the can passes	The whole diameter must fit inside the field with margin on both sides
Cap and terminal diameter	Outline circle across the terminal, if the terminal is presented to the receiver	A bright plated terminal against a bright cap needs edge contrast check on a real sample
Tab width	Distance between the two long edges of the tab	Requires clear backlight through the gap beside the tab, so the fixture must not mask it
Tab pitch	Centre-to-centre distance between two tab edges in one field	Both tabs must be inside one field, otherwise pitch becomes a two-head or a fixture-datum question
Busbar and plate width	Outline width of the conductor at the measured section	Edge quality of a stamped or laser-cut busbar sets the practical repeatability, not the head
Pouch seal width	Edge distance from the seal boundary to the pouch outline	A translucent seal edge is a contrast problem before it is an accuracy problem
Jelly-roll outer diameter	Outline diameter of the wound roll before insertion	Soft, springy edges move between presentations, so repeatability has to be proven on real rolls
Cell thickness	Not measurable in silhouette	Belongs to a height channel or a flash system with a height sensor, not to an inline head
Weld penetration, porosity, internal voids	Not measurable in silhouette	Belongs to radiographic or CT inspection; a silhouette sees the weld outline only

Two of the limits in that table deserve a sentence each, because they recur in every cell project. The first is tab pitch. If the two tabs sit at opposite ends of a long cell, they cannot be in one field of a single head, and the measurement becomes a two-head arrangement or a fixture that presents the cell twice. Neither is wrong, but both change the cost and the cycle-time plan, and discovering it after the head has been ordered is the expensive version of the same conversation. The second is the pouch seal. A heat-sealed aluminium laminate edge is translucent and slightly rounded, and the boundary that looks crisp to the eye is a gradual ramp in the image. That is a lighting and threshold question that has to be settled on a real pouch sample, not on a drawing.

Choosing a head: field size first, accuracy second

The HM family is a set of five heads that differ in field size, and the published figures move in a consistent direction as the field grows: a larger field is bought with a looser repeatability and a looser position accuracy. The pattern is not a manufacturing compromise, it is geometry. The same receiver resolves a smaller field into finer divisions, so the small head is the accurate head and the large head is the wide head. Selecting a model therefore starts with the largest feature that must be seen, not with the tightest number on the data sheet. The family is catalogued at system level on the HM telecentric measurement system product page.

Head	Measuring range	Repeatability	Position measurement accuracy	Cell features that fit the field
HM-1006	φ6 mm	±0.05 µm	±0.2 µm	Tab width on a small tab, terminal diameter, fine busbar section
HM-1015	φ15 mm	±0.05 µm	±0.3 µm	Tab pair on a narrow pitch, cap feature groups, small connector pads
HM-1040	φ40 mm	±0.08 µm	±1.0 µm	Wide tab, busbar width, terminal spacing on a small cap
HM-1065	φ65 mm	±0.1 µm	±1.2 µm	Two tabs on a medium cell, short busbar span, pouch seal band
HM-1120	φ120 mm	±0.2 µm	±2.5 µm	Full can diameter of a large cylindrical cell, module bracket features

The field rule is one line long: the largest dimension that must be inside a single measurement, multiplied by a margin for pass-line wander and fixture variation, has to be smaller than the measuring range. A φ46 mm cylindrical can inside a φ120 mm field leaves a linear margin of 74 mm across the diameter, a factor of 2.6 on the cell diameter (derived), which is generous enough that the can does not have to be perfectly centred. The same can inside a φ65 mm field leaves 19 mm, a factor of 1.4 (derived), which is workable only if the pass line is well controlled. Inside a φ40 mm field the can does not fit at all, and the honest answer is a larger head, not a cleverer algorithm.

The reverse case is where most cell projects actually sit. A prismatic cell of 150 mm in length does not fit any head in the family, so the head is not measuring the cell, it is measuring a feature on the cell: a tab, a terminal, a busbar section, a seal band. That is a legitimate and very common arrangement, and it is the one that should be written into the specification, because it changes what the accuracy claim applies to. A head that is quoted as holding ±0.1 µm repeatability is holding that figure on the feature inside its field, not on the 150 mm cell outline, and the difference is not a detail when the drawing tolerance is on the outline.

One published limitation has to be recorded rather than assumed. The depth of field of each HM head is not stated in the material we hold, so the tolerance to pass-line wander in the depth direction must be obtained from the supplier as a number and validated on the actual fixture. What can be said without a number is the mechanism: the transmitter and receiver are both telecentric, so a part that moves toward or away from the optics within the depth of field does not change apparent size. Outside that range the edge softens and the measurement drifts, and the failure appears first on the small head with the tight tolerance, which is why a fixture that is acceptable for a φ120 mm head is not automatically acceptable for a φ6 mm one.

The accuracy chain on a moving part

Two published numbers describe an inline head, and they are not interchangeable. Repeatability is how tightly the same edge repeats on the same part in the same position; on the HM family it ranges from ±0.05 µm on the two small heads to ±0.2 µm on the φ120 mm head, model dependent. Position measurement accuracy is how close the reported position of a feature is to its true position in the field, and it ranges from ±0.2 µm to ±2.5 µm across the same five heads. The ratio between the two widens as the field grows: 4:1 on the HM-1006 and 12.5:1 on the HM-1120 (derived), which means the large head is a repeatable instrument whose absolute position claim is carried mainly by calibration.

Both figures are manufacturer specification values, and the test conditions behind them are not stated in the material we hold: environment, sample count, artefact, integration time and averaging are all unpublished. That does not make them wrong, and it does make them a capability statement rather than an acceptance criterion. The distinction matters most on a cell line, because the achievable tolerance there is set by a chain that includes the head but is not limited to it: pass-line wander against the depth of field, fixture repeatability, vibration transmitted from the tab punch or the sealer, thermal drift of the mounting frame over a shift, and contamination of the protection window by electrolyte mist or metal dust. A quotation that promises a production tolerance equal to the published repeatability is promising something the instrument alone cannot deliver.

The exposure figure is where the arithmetic becomes concrete. The HM family uses a 35 microsecond exposure, and the point of that number is that the shadow is captured before the part can move appreciably. Whether it moves appreciably is a calculation, not a claim: motion blur equals line speed multiplied by exposure time, so at 100 mm/s the smear is 3.5 µm, at 300 mm/s it is 10.5 µm, and at 1 m/s it is 35 µm (all derived). Read against a repeatability of 0.05 to 0.2 µm, that arithmetic says something useful: the exposure is fast enough to freeze a slow-moving cell line, and it is not fast enough to make a fast one invisible. On a high-speed transfer the blur term must be either reduced by a shorter strobe or removed from the measurement direction by the geometry of the presentation, and either way it has to be settled with the measured line speed rather than with the word frozen.

There is a compensating property that should not be overlooked. Because the optics are telecentric on both sides, in-plane motion along the optical axis does not change the scale, so a part that drifts toward the receiver keeps its measured size until it leaves the depth of field. The failure mode of a poorly presented cell is therefore not a gradual size error, it is a sudden loss of edge quality followed by a wrong answer. That is easier to detect in software - an edge-strength or contrast gate will catch it - and it is another reason to insist on a trigger and a presence check rather than a free-running measurement.

Cycle time: what the line rate actually demands

Inline measurement is usually sold on speed and usually limited by something other than the exposure. The budget per part is fixed by the line: a line running 60 parts per minute gives 1,000 ms per part, and one running 300 parts per minute gives 200 ms per part (both calculated as 60,000 divided by parts per minute). Against those budgets the 35 microsecond exposure is three to four orders of magnitude smaller, so the exposure is never the constraint. What consumes the budget is the handshake with the machine, the time allowed for the part to be inside the field, the data path to the PLC or the SPC system, and the decision about accept, reject and re-measure.

Line rate	Cycle budget per part	Exposure share of budget	Practical note
30 parts per minute	2,000 ms (calculated)	0.0018 percent (calculated)	Time for two heads or a re-measure on the same part is available
60 parts per minute	1,000 ms (calculated)	0.0035 percent (calculated)	Comfortable for a single head with an SPC write per part
120 parts per minute	500 ms (calculated)	0.007 percent (calculated)	Handshake latency starts to matter more than measurement
300 parts per minute	200 ms (calculated)	0.0175 percent (calculated)	Reject actuation and data logging define the design, not the optics

The other capability that shapes a cell job is what one capture returns. The family is specified to output up to 100 items from a single silhouette, covering outer diameter, width, height, corner radii and other drawing-indicated dimensions, and the same capture can also flag defect-type features that change the outline, such as burrs, chips and foreign matter. That is a software capability statement, and it should be read as one: it says how many dimensions the software can evaluate from a shadow, not that any hundred tolerances can be held to the same uncertainty. On a cell, the practical reading is that one tab capture can return tab width, tab pitch, edge straightness and burr presence together, which is why a single head can cover a feature family that would otherwise need several gauges.

Station count follows from geometry rather than from speed. One head covers one measurement plane on one side of the part. A cell with tabs at both ends needs either two heads or a presentation that turns the cell, and a cell family that spans more than the largest field needs more than one head or a smaller feature set. The planning arithmetic for that decision - stations, cycle time, and the utilisation that follows - is set out in the cycle time and station count calculation, and it is worth running before the head count is fixed, because the cost of a second head is usually smaller than the cost of discovering the need after the fixture is built.

Presentation and mounting: the failures that are not optical

Most inline measurement problems on a cell line are mechanical, and they appear as optical symptoms. The first is backlight clearance. A silhouette needs the emitter on one side of the part and the receiver on the other, with an unobstructed path through the gap that defines the edge. On a tab measurement the edge of interest is at the boundary of the cell, which is exactly where the fixture, the conveyor and the hold-down all want to be. A fixture that grips the cell across the tab masks the edge, and the resulting measurement is stable, repeatable and wrong.

The second is pass-line wander. A cell that is presented 0.5 mm higher on one cycle than the next moves within the depth of field and measures the same; a cell that is presented with a tilt changes the projected width, and the error is proportional to the tilt and to the feature size. The fix is a datum, not a filter: a hard stop that the part is presented against, a compliant hold-down that repeats, and a presence sensor that refuses to measure a part that is not seated. On a line where the transfer is a walking beam or a vacuum belt, the vertical repeatability of the beam is part of the measurement system and belongs in the acceptance test.

The third is contamination. Electrolyte mist, metal dust from the tab punch and the general atmosphere of a cell room all deposit on the protection window, and a contaminated window degrades edge contrast before it changes any number, so the first symptom is usually an increase in edge-failure counts rather than a dimensional drift. A positive-air-purge window and a cleaning interval defined by measured edge strength, rather than by calendar, is the practical answer. The fourth is thermal and structural drift: a head on a bracket that grows with the room temperature over a shift will move relative to the pass line, and the artefact check that would catch it is the same one that catches calibration drift, run at the start of a shift and after any crash or tool change.

None of these faults can be designed out by choosing a better head. They can be designed out by treating the head, the fixture, the trigger and the window as one measurement system with one acceptance test, which is also the only way the published repeatability figure has any chance of surviving into production. The error-budget method for that acceptance test is described in the inline diameter error budget guide, and it applies unchanged to a tab or a busbar.

Where an inline head is the wrong instrument

An inline head measures an outline in transit. Every job that is not an outline in transit belongs somewhere else, and saying so early is cheaper than a failed acceptance. Cell thickness, cap height and the step between cap plate and can are height measurements: on the IMAGE 3 series the height channel is a point-spectral confocal sensor with a ±3 mm range, a 55 mm working distance and a 17 µm spot, standard on MAX and optional on MAX2, and a silhouette head cannot substitute for it. Weld penetration, porosity and internal alignment are radiographic or CT jobs, and a silhouette of a weld seam shows the seam outline and nothing about what is under it.

The boundary with a flash measurement system is the one that matters most on a battery project, because both instruments are optical and both report dimensions. The split is in how the part is presented. A flash system measures a stationary part in one calibrated field: the whole cell or a group of small parts is placed, the field is exposed once, and every feature inside it is measured at once. That suits a lab, an incoming inspection station, or an end-of-line cell where the part is already stopped. An inline head measures a moving part through a fixed small field and returns one feature family per pass. A cell that is stopped for other reasons and needs a full drawing checked belongs to the flash route, described in the battery cell and tab flash measurement page; a tab station that needs width and pitch on every part at line rate belongs to a head. Using a flash system as a line gauge wastes its field, and using a head for a whole-cell drawing produces a fixture with a travel axis.

Two further cases are worth naming. A transparent or reflective part - a separator film, a shiny nickel tab against a polished cap - fails on contrast before it fails on accuracy, and the countermeasures are lighting and geometry rather than a different head class; the practical catalogue of what works is in the guide to inline measurement of transparent and reflective parts. And a job that needs the average diameter of a continuous strand rather than the dimension of discrete parts is a laser gauge problem, covered by the wire, cable and tube gauging page, where the LM-1030 covers 0.3 to 30 mm with ±2 µm accuracy. Discrete cells and continuous foil are different measurements even when the numbers look similar.

A five-item selection checklist

Before a head is specified, five answers should exist on one page, and each of them changes the model or the fixture. First, the feature list with tolerances: which dimensions are actually called out, on which feature, in which tolerance class, and which of them are not outline dimensions at all. Second, the field arithmetic: the largest feature that must be inside one measurement, multiplied by the wander margin, compared against the five measuring ranges. Third, the presentation: how the part arrives, whether it stops, what it is presented against, and whether the edge of interest is reachable by the light path. Fourth, the dynamic conditions: line speed, part pitch, the resulting smear from the exposure arithmetic, and the vibration sources within a metre of the head. Fifth, the output path: what the line does with the result, how fast it has to do it, and what happens to a part whose edge fails the contrast gate.

If any of those five is missing, the specification will be filled in by assumption, and the assumption will be the supplier's. The most common missing item is the second: a project is quoted on a head that fits the sample, and then the part family grows by ten millimetres and the field no longer covers it with margin. The second most common is the fifth: a measurement that is correct but arrives 300 ms after the reject gate has passed, which is a measurement that cannot reject anything. Both are cheap to settle on paper and expensive to settle after installation.

Frequently asked questions

Can an inline measuring head check every cell without stopping the line?

Yes, within the limits of the presentation. The head images a shadow during a 35 microsecond exposure while the part passes through a fixed field, so the line does not have to stop or index for the measurement itself. What the line does have to provide is a trigger that fires while the feature is inside the field, a transfer that holds the part within the depth of field, and a data path that returns the result before the reject gate. Every-part measurement therefore depends on the machine as much as on the head: a line with a walking beam and a hard datum will measure every part, while a line with loose transfer and no presence check will measure every part and be wrong about some of them.

Which head size fits a cylindrical cell of about 46 mm diameter?

A φ65 mm head covers the can with 19 mm of linear margin across the diameter, a factor of 1.4 on the cell diameter (derived), and a φ120 mm head covers it with 74 mm of margin, a factor of 2.6 (derived). The smaller field is the more accurate choice on paper, since its published repeatability is ±0.1 µm against ±0.2 µm for the large head, but it leaves less room for pass-line wander. If the can is well guided and the pass line repeats, the φ65 mm head is the better fit; if the transfer wanders, the large field buys tolerance at the cost of accuracy, and the accuracy question then has to be answered against the drawing tolerance rather than against the data sheet.

Can a silhouette head measure cell thickness or cap height?

No. A silhouette is the outline of the part as seen from one side, so it carries width and diameter information and no depth information. Thickness, cap height and the step between the cap plate and the can are height measurements and require a different channel: on the IMAGE 3 series that channel is a point-spectral confocal sensor with a ±3 mm range, a 55 mm working distance and a 17 µm spot. A project that lists thickness and tab width together is a project with two instruments in it, and the specification should say so rather than asking one head to do both.

What tolerance can a cell line realistically hold with a head specified at ±0.05 µm repeatability?

Not ±0.05 µm. The published figure is a capability statement measured under conditions that are not stated in the material we hold, and production adds pass-line wander, fixture repeatability, vibration, thermal drift and edge-quality variation on top of it. The realistic way to answer the question is to build an error budget from the actual fixture and the actual part, then verify it with a gauge study on real cells at line speed. A head with a ±0.05 µm repeatability figure is often the right instrument for a tolerance band of a few micrometres, provided the mechanical chain around it is designed to the same order.

Do I need one head or two for a cell with tabs at both ends?

Two heads, or a fixture that presents the cell twice. One head measures one field on one side of the part, so two tab groups at opposite ends cannot both be inside the same field unless the cell is short enough to fit both within the measuring range. The alternative to a second head is a rotation or a second pass, and both consume cycle time and add a presentation error. On a line with a 1,000 ms budget per part the second head is usually the simpler answer; on a slower line with a wide field the single head and a re-presentation may be enough.

Related reading

The system-level view of the HM heads and the LM-1030 laser gauge, together with the HM telecentric measurement system product page, is set out in the inline dimensional measurement system guide. Head-by-head selection logic, including the reasoning behind the field and accuracy trade-off, is in the HM series five-model selection guide. Presentation and station-count planning is covered by the inline vs offline selection guide and the cycle time and station count calculation, and the error budget method is in the inline diameter measurement error budget. For the stopped-part case use the battery cell and tab flash measurement page; for EV weld seam optics, the zoom lens page for EV battery weld inspection.