Load Cells for Liquid Filling Machines: The Tolerance Is the Fill

1. A filling machine never weighs the fill you asked for

A checkweigher weighs a package that already exists. A filling machine has to create the weight it is weighing, and it can only read the result after the fill has already happened. That single sentence is the whole engineering problem, and it explains almost every strange thing about how these machines are specified.

The consequence is that a filler is not accurate. It is repeatable, and it is accurate only because it has learned to stop early by exactly the right amount. The industry has a word for that early amount. In Chinese it is 落差 (luocha) — literally “the fall”. In English engineering literature it is the drop, the in-flight weight, or the pre-act. It is the mass of liquid that is still travelling between the nozzle and the bottle at the instant the controller decides to close the valve, plus whatever the valve delivers during the time it takes to actually close.

Everything else in this article is downstream of that. If you are buying load cells for liquid filling machines, the number you need to understand is not the accuracy class. It is how much of your error the drop-learning loop can hide, and how much it cannot.

If you want the surrounding picture first — the three weighing jobs that every packaging line needs, from infeed to check to outfeed — I wrote that up separately in Load Cells for Packaging Machinery. This article is the narrow one about the fill itself.

Three ways to put a cell on a filler

Filling-machine instruments sold into this market typically advertise three modes, and it matters which one you are in because the load cell’s job is completely different in each. The wording below is the wording on a Chinese filling controller’s own feature list, translated: single-head additive, multi-head additive, and multi-head subtractive.

The fourth row is the one people forget to mention at the enquiry. If your bottles ride in stainless pucks so the line can handle odd shapes, then every weighing includes the puck, and the puck has to come back to the same weight every cycle. That is a tare problem, not a load cell problem, and it is worth about as much as everything else in this article.

Why the industry fills by weight and not by volume

A volumetric filler dispenses a fixed volume. The label promises a mass. The bridge between them is density, and density moves with temperature. For a light organic liquid such as an edible oil or a lubricant base stock, the volumetric expansion coefficient is roughly 0.0007 per degree Celsius. Over a 15 °C shift from a cold morning to a warm afternoon, the same piston stroke delivers 1.05 percent more mass.

On a one-litre fill that is 10.5 grams, against a 12 gram legal allowance at class X(1). A handful of degrees of ambient change therefore eats nearly the whole allowance of a mechanically perfect filler. Water’s coefficient is about three times smaller, which is why water bottlers get away with volume for far longer than oil packers do; a light hydrocarbon’s is larger again, and there volume has no chance at all.

This is also, incidentally, a change of legal chapter. In the European Union a filler that measures volume is a liquid measuring system under Annex VII of the Measuring Instruments Directive; a filler that measures mass is an automatic gravimetric filling instrument under Annex VIII, chapter III. If you are converting a line from one to the other, you are not upgrading a machine — you are replacing the instrument’s whole regulatory identity, and the two do not share a tolerance table.

2. The label decides the class, and where my supply stops

The legal name for what we are discussing is an automatic gravimetric filling instrument, abbreviated AGFI. The European definition is worth quoting exactly, because every word in it does work: “an automatic weighing instrument that fills containers with a predetermined and virtually constant mass of product from bulk.”

Note what is missing. There is no requirement that the machine weigh anything the customer can see. Gravimetric describes how the machine decides when to stop, not what it prints. This produces the single most useful boundary test in the whole subject, and it comes from WELMEC’s own guidance rather than from any manufacturer’s brochure.

I have watched a European packer discover this the expensive way: a line built as a filler, with the actual fill weight printed on every carton because the sales team liked the traceability, which quietly reclassified fourteen machines. The tolerance table did not change. The conformity route, the marking, and the paperwork did.

Which chapter of the law you are in

NIST Handbook 44 §2.24 applies to filling machines that determine package and lot compliance with net content representations, and its exceptions list removes machines used in a plant that has a separate quantity-control programme — statistical process control, suitable weighing instruments, standards traceable to national standards. In plain terms: control net content statistically with traceable equipment and the filler does not have to be a legal-for-trade device. What it does have to be is capable, and the programme has to actually exist on paper.

For completeness, the Chinese pair is JJG 564-2019, the verification regulation that replaced JJG 564-2002 and added the milligram to its units, and GB/T 27738-2011, the product standard, which is currently being revised to adopt OIML R61-1:2017. The Chinese regulation names the same two performance quantities as the international recommendation: the maximum permissible deviation (MPD) and the maximum permissible preset value error (MPSE), and it requires the control instrument used to verify the filler to be better than one third of both.

Two accuracy classes, not one

The directive requires the manufacturer to declare two classes, and this trips up almost every quotation I receive. The reference accuracy class, Ref(x), is the best the instrument type can do, and it applies to static loads. After installation the individual machine is designated one or more operational accuracy classes, X(x), having taken the actual product into account.

The class factor x must be 2 or less and must be of the form 1, 2 or 5 times a power of ten — so 0.2, 0.5, 1 and 2 are the usual working values. X is a regime relating accuracy to load weight, and x is simply a multiplier applied to the limits given for class X(1). A class X(0.5) machine is allowed exactly half of what X(1) is allowed. Nothing else changes.

The static requirement for the reference class is easy to trip over: the maximum permissible error for Ref(x) on static loads is 0.312 of the per-fill deviation limit, multiplied by x. For a 1,000 gram fill at X(1) that is 0.312 times 12 grams, or 3.744 grams — the same figure that appears later as the setting error, and section 3 shows why.

What I supply on a filling machine, and what I refuse to

I want this on the record early, because the boundary is the honest part of the offer. I make the weighing layer. I do not make, certify or approve filling machines, and I have no interest in pretending otherwise.

One clarification that saves arguments later. An OIML R60 certificate is a component certificate: it says a cell may be used inside a verified instrument. It is not a machine approval, it does not transfer to the machine, and a filler carrying nothing but R60 test sheets from me has no metrological status at all. The machine’s class is earned by the machine, on the customer’s own product, in the customer’s own hall, by whoever holds the approval. My job is to ensure the cell is never the reason that fails. If you are still choosing between cell types for load cells for liquid filling machines, start with How to choose the right load cell for your project.

3. The staircase that sets your tolerance, and the average rule

Here is the table everything else is built on. It is Table 5 of MI-006 chapter III in the current directive, and it gives the maximum permissible deviation of each fill from the average — not from the target — for class X(1). The other classes are this table multiplied by their own factor.

Read the staircase carefully, because it is not smooth. Between 500 and 1,000 grams the allowance is a flat 12 grams; just above 1,000 grams it becomes a percentage again, so a 1,100 gram fill is allowed 13.2 grams while a 1,000 gram fill was allowed 12. Pack sizes that straddle a band edge can therefore have different compliance difficulty for a 50 gram change in target weight, which is a real effect and not a curiosity.

The average is four times tighter than the individual fill

The per-fill table above is the generous half of the law. The demanding half is the setting error: where a fill weight can be preset, the maximum difference between the preset value and the average mass of the fills is 0.312 of the per-fill allowance. The international recommendation and the Chinese regulation express the same requirement as 0.25 of the in-service allowance. Both roads arrive at the same number.

The apparent disagreement between the two rules is arithmetically empty, and seeing why is worth thirty seconds. The directive’s per-fill allowance is 0.8 times the in-service allowance the recommendation uses — 12 grams against 15. And the directive’s setting-error factor is 1.25 times the recommendation’s factor — 0.312 against 0.25. Multiply those two factors and you get 1.0. 0.312 times 0.8 is 0.25. Both documents are describing one requirement, against two different baselines, and the apparently different numbers are the same number.

The floor under the staircase: rated minimum fill

The deviation table only applies at or above the instrument’s rated minimum fill. Below it, the machine is not out of tolerance; it is outside the scope of its own accuracy claim. The directive makes the manufacturer declare this value. The recommendation builds it from the scale interval and the class, and the resulting staircase is one of the more useful tables in the subject, because it converts a clever purchasing question into a hard floor.

Those values come from the recommendation’s own certificates of conformity, and they are why a weigh hopper with a 50 gram interval can never certify a 1 kilogram addition at class X(1): the floor is 3,350 grams. The machine is not faulty. It is the wrong machine for that ingredient, and the only fixes are a finer interval, a looser class, or a separate station. In divisions the floor runs from eleven at fine intervals to a hundred at coarse ones.

Two footnotes. First, these floors assume the fill is produced in one weighing; where a fill is made of several loads the floor is higher, because each load has to be large enough to be weighed honestly. Combination weighers take their class from the fill, not from the individual drops. Second, Minfill rises when the environment is bad, and the recommendation names temperature effect on the no-load indication, zero-setting accuracy, disturbances, warm-up time, the product and the interval. Four of those six are environmental, which is why the same platform carries a higher floor beside a vibrating conveyor than on a test bench. The same logic governs process vessels, which is why I keep the three weighing jobs on a vessel separate in Load Cells for Hopper Scales.

Sixty fills, and the instrument you verify with

The average needs statistics. For fills up to 10 kilograms the recommendation requires 60 test fills to establish it; 32 fills for 10 to 25 kilograms, 20 for 25 to 100 kilograms and 10 above that. Sixty consecutive bottles is a real production run, and it is the reason a machine that looks fine over five bottles is not fine.

The instrument used to weigh those 60 fills must be accurate to better than one third of both the MPD and the setting error. A Chinese verification paper works the example through with a 60 kilogram, 20 gram platform as the control scale for a 50 kilogram X(1) machine, and for class X(0.2) the same literature points at balances with more than ten thousand divisions. When a customer tells me his filler has been verified, my first question is what the control instrument was and when it was last verified itself.

One more clause that almost nobody quotes, and it matters if you have variable-frequency drives in the same cabinet. The directive sets the critical change value due to a disturbance as the static error calculated at the rated minimum fill, rounded up to the next scale interval — judged at Minfill, not at maximum capacity. With a one gram interval and an eleven gram floor, that collapses to a single division, which is a real argument for surge protection rather than the cheapest summing box on the market.

4. The drop, the dribble, and why one speed cannot work

The drop is the product in the air plus the product delivered during valve closure, and to first order it is a multiplication: the flow rate at the moment of the decision, times the time the valve takes to stop delivering. A machine pouring one litre per second through a pneumatic valve that takes 45 milliseconds to close has a drop of 45 grams — 4.5 percent of a one litre fill, and the reason the valve cannot simply be told to shut at the target.

The trick is to stop the fast fill early and finish the last few percent slowly. On a machine that fills about 85 to 90 percent of the target at full rate and then throttles to a trickle of roughly one twentieth of the fast flow, the same 45 millisecond valve closure produces a drop of 2.25 grams instead of 45. Twenty times smaller. That much everyone in the industry knows.

What fewer people work out is what happens to the scatter of the drop, because the mean is not the problem. If the closing time varies by five milliseconds and the flow varies by five percent, those two uncertainties combine to about twelve percent of the drop. That twelve percent is what the learning loop cannot remove, because it changes from fill to fill.

Read the third row and the whole architecture becomes obvious. At one speed, the scatter of the drop alone is one and a half times the entire legal allowance on the average — before a single load cell error is added, before the product temperature moves, before anything else. No cell on earth rescues that machine. The dribble stage is not a refinement for fussy customers; it is the only reason the number exists at all.

The line speed sets the flow, and the flow sets the drop

Here is the coupling that catches people when they scale a line up. A cycle is not all filling. Indexing, clamping, taring, valve actuation, post-fill settling and verification all take time, and on the machines I have measured that overhead is on the order of 400 milliseconds. Everything left over belongs to the pour.

At 60 bottles per minute on one head the flow has to reach about 1,667 grams per second and the drop becomes 75 grams, whose scatter at that same twelve percent is 9.1 grams — 244 percent of the legal allowance on the average. So a 60-per-minute one-litre line is not a single-head line. More usefully, the same arithmetic sets the dribble stage: throttle to five percent of the fast rate and the drop is 3.75 grams, whose scatter is 0.46 grams, or twelve percent of the allowance.

One consequence for the machine builder: an adaptive drop loop needs to know when not to learn. Update the constant on every fill and it chases the tail of the distribution and inflates its own scatter, so most controllers learn slowly, or only on fills inside a window, or on a schedule. When a customer’s average is wandering, the first thing I ask for is the machine’s own drop-correction log and the interval at which the loop may update. With no log, the next best thing is to rule the cell out honestly, and I wrote the procedure for that in How to determine whether the load cell is damaged.

5. What a learned advance cannot absorb

A constant error and a drifting error look identical on a data sheet and behave completely differently on a filling line. The rule is short enough to put on a wall: anything constant during the measurement, and constant with respect to whatever reference you last set, cancels. What survives is anything that changes during the measurement, anything that has changed since the reference was set, anything that is not a straight line, and noise.

Applied to each of the three machine types from section 1, that rule sorts the load cell’s specifications into two groups — and the group that survives has one thing in common.

Look at the middle column, which is how most high-speed machines actually work. Three of the seven rows survive in full and all three are expressed as a percentage of the cell’s rated capacity. The bold entries in every column have the same property: they are quantities that nobody re-measures. That is the whole mechanism, and it produces the most useful sentence in this article.

The rule, worked out

Take our standard analog family, whose repeatability is quoted as ±0.02 %F.S — that is the figure on the WST201, WST216, WST217 and WST202 pages. The cell’s contribution to one fill, in grams, is 0.0002 times the rated capacity. The legal allowance for that fill, in grams, is a percentage of the fill. Divide one by the other and the capacity cancels out of the interesting part:

The pattern is worth stating as a number, because it removes the need for a spreadsheet in most meetings. The cell’s share of the allowance is 1.67 percent of full scale divided by the fill percentage, which simplifies to a single ratio: the cell’s share is about 1.67 percent of the quantity, capacity divided by fill. It reaches one hundred percent — the cell alone eating the entire class — at a capacity-to-fill ratio of 60 for class X(1), 30 for X(0.5) and 12 for X(0.2).

That is the design rule. If your weighing vessel is more than sixty times the fill, an ordinary class C3 cell has already spent the whole per-fill allowance on its own repeatability, before you add the product, the machine or the ambient. Moving the fill onto a 25 kilogram intermediate metering vessel cuts the cell’s share from 83 percent to 8.3 percent — tenfold, achieved with plumbing. A class C6 cell on the big vessel instead halves the error and leaves it at 45 percent. Both are legitimate engineering; only one of them is cheap, and that trade-off runs through nearly every decision about load cells for liquid filling machines.

6. Live tare or preset tare: the decision that sizes your cell

Two machines can have the same valve, the same platform, the same cell and the same product, and need cells of completely different quality, because of one engineering decision taken years earlier and rarely revisited: does the machine weigh the empty container on every cycle?

A live tare costs time — roughly 40 milliseconds on the machines I have measured, which is four percent of a 1,000 millisecond cycle and a much larger fraction of a fast one. A preset tare costs nothing in time: the average container weight is typed in at changeover, and the machine simply subtracts it. What the preset tare costs is everything in the middle column of the table in section 5. Every slow error in the cell, and every variation in the container, now arrives in full at the fill.

How much the container actually varies

The plastic bottles are not the problem, which is worth saying plainly because the instinct is to blame them: a blow-moulded litre bottle varies by a gram and a half, one tenth of the prepackage tolerance on a 1,000 gram fill. The container that hurts is glass. A 700 millilitre glass bottle weighs about 400 grams, and although a single batch is beautifully consistent the spread within a batch is still around eight grams — 1.14 percent of the fill, and more than half the 15 gram tolerance for that size. Fill glass by weight with a preset tare typed in from the last delivery and you have given away half your allowance before the filler does anything. The fix is a bottle-weighing station, a batch tare re-measured at every delivery, or a container supplier who will hold a tighter mass tolerance.

And here is where absolute accuracy comes back into a machine that otherwise does not need it. If you re-measure the tare at every batch, the number you type in is read off a scale, and the error in that one reading applies to every bottle in the batch. A zero offset of 0.05 percent of full scale on a 5 kilogram platform is 2.5 grams, and 2.5 grams of tare error is 2.5 grams on every one of a hundred thousand bottles. Take that reading once a week and it barely matters; take it once a shift and it is the largest single term in your budget. I hit the identical wall on smart LPG cylinders, where the tare has to be assumed for years at a time.

7. Case one: the one-litre line that ran heavy every afternoon

A personal-care contract packer in Malaysia — two heads, 60 bottles a minute, a one litre shampoo-style surfactant product, in PET. The email was short and came with a spreadsheet attached: fills drifted light through the day, about 1.8 grams by four in the afternoon, and back on nominal every morning. The spreadsheet also showed that the operator had been nudging the preset up each afternoon and the QA manager had been nudging it back each night. A machine of theirs had been sold, so the first hypothesis on the factory floor was the load cells.

Whenever I hear “it drifts through the day” I know the culprit is a slow accumulated force, and I know I will be told it is temperature. So I asked for two things before any drawing, and neither of them was a datasheet. I asked for the drop-correction value logged with a timestamp across a full day, and I asked for photographs of the weigh deck and the area under the nozzles at seven in the morning and at four in the afternoon.

First, clear the cell

I asked them to put their certified one kilogram test weight on the platform at the start of the shift and again at the end, with the deck wiped clean both times, and to send me the readings. They came back 0.4 grams apart. On a 5 kilogram cell with a two-hundredths-of-a-percent repeatability — one gram — that is inside the cell’s own noise. The cell was not the problem, and I said so in writing before proposing anything. It is worth the two days it takes, because it converts the conversation from a supplier defending a product into two people reading a log.

Then, read the log

The drop value did exactly what the complaint said. It started each morning at 68 grams, fell steadily to 66 grams by the end of the shift, and reset at the next start — and here is the part that mattered: the machine takes its zero once per shift, at startup, not per bottle. A live tare would have cost it 40 milliseconds of a two-second cycle and the machine builder had taken that decision years earlier to make the rate.

The four o’clock photograph explained the rest. The weigh deck carried a faint damp film of product, with a standing bead along the front lip, fed by a drip tray whose drain had been positioned a few millimetres inside the deck’s footprint. The film was on the order of 0.09 millimetres thick over a deck of about 0.02 square metres. That is roughly 1.8 millilitres — about 1.8 grams of product standing on the weighing platform, unaccounted for, and getting deeper every hour.

The magnitude is worth putting against the law. 1.8 grams on a 1,000 gram fill is 48 percent of the 3.744 gram allowance on the average, and the sign is the dangerous one: the fills were coming out below nominal, which is not a giveaway problem, it is a short-weight problem. The company was not losing money on that line. It was accumulating a compliance failure.

What we did, in the order that mattered. First, the drain: move the tray’s outlet clear of the deck footprint and add a lip so that anything landing on the deck runs out of the weighing zone instead of pooling in it. That is a sheet-metal change and the machine builder’s job, not mine. Second, a five-second wipe at every break, written into the shift sheet with a place to initial it, because a fix that does not survive a shift change is not a fix. Third, move the zero from once per shift to once per hour. Fourth, and the one they had not considered, a live tare on every bottle would have made the whole class of problem disappear, at four percent of the cycle. I put the number in front of them and let them decide, because a rate loss to buy back part of a legal allowance is a business decision, not a weighing one.

I sold nothing on that visit. The cells on that machine are ours and they were innocent, the drift was a drip tray and a cleaning interval, and the fix cost the price of two stainless brackets and one line on a checklist. They came back eighteen months later for a second line, which is the arithmetic I actually care about.

8. The nozzle, the sleeve and the drip tray

Everything that touches the container, or touches the machine while the container is being weighed, is inside your weighing system. On a dry filler this is mostly theory. On a liquid filler it is the main mechanical risk, because liquids arrive through pipes and leave through nozzles, and both of those have to be somewhere.

I have a lot of respect for the attention section of a Chinese filling-machine manual, which lists its own precautions in the plainest possible language. Two lines from one of them are worth more than a page of theory: the nozzle must not touch the wall of the drum during filling, and the weighing tray must not touch the side of the machine. A machine builder who writes that down has been bitten by it.

The nozzle is a load path

If the nozzle rests on the neck of the bottle, some fraction of the nozzle assembly’s weight and of the supply pressure’s reaction goes through the container and into the cell. The magnitude surprises people. A drip tray or a nozzle guide that exerts the equivalent of 20 grams on the container adds an apparent 20 grams to a 1,000 gram fill — two percent, against a per-fill allowance of 1.2 percent. That is not a calibration error you can trim out, because the contact is intermittent: it appears and disappears as bottles index, as the nozzle wears, and as the container moulding varies.

The same argument applies to the suck-back systems used to stop drips. A suck-back piston strokes upward when the fill ends, and its reaction lands on whatever the cylinder is bolted to. If that is the weighing frame, the cell sees it. If it is the machine frame, everything is fine. Ask where the cylinder is bolted. It is a one-line question with a two-thousand-dollar answer.

The supply vessel’s pipe is not a spring, it is a short circuit

On a subtractive machine the supply vessel hangs on the cells, and it is connected to the rest of the plant by a pipe. This is the failure I see most often on loss-in-weight installations, and it fails in a way that makes people doubt the sensors, because the weight readings stay clean and repeatable and are simply wrong in a way that changes with temperature and with level.

The numbers are instructive. A vessel on four cells that deflect 0.3 millimetres at rated load moves less than a tenth of a millimetre between full and empty. Three metres of one-inch schedule-40 steel pipe, clamped at both ends, has an axial stiffness of roughly twenty-one kilonewtons per millimetre. Multiply by that tenth of a millimetre and you get about two kilonewtons, or two hundred kilograms, of force through a pipe that was supposed to be a conduit.

The number is absurd on purpose, because it tells you the truth: a rigidly clamped pipe on a weighed vessel does not behave as a stiff spring, it behaves as a structural short circuit. In practice the pipe, the bracket or the vessel yields somewhere and the weighing system is simply no longer measuring what you think. That is why every loss-in-weight vessel in serious service has a flexible sleeve or bellows in the connection, and why a bracket that “looked fine” is the first thing I ask a customer to photograph. And a sleeve has its own behavior: its stiffness changes with pressure and with age, so it is a term in your budget rather than a solution to it. It is a small term, and it is not zero.

Three rules for the mechanical envelope of a weighed container on a liquid line. The nozzle must not touch. The sleeve must flex rather than resist. And the drip tray must drain somewhere that is not the weighing zone. All three are cheap at the design stage and expensive afterwards, and none of them can be fixed by the part I sell. If you are still choosing the mounting arrangement for the vessel itself, the module-by-module version of this argument is in Tank and Silo Weighing Systems.

9. Settle time, filtering and the honest answer about resolution

Everything up to here assumed the machine can capture a weight. A liquid line is where that assumption costs money, because you are not weighing a solid object. You are weighing a container whose contents are free to move, and they keep moving after the valve shuts — which is the first thing to check when you are sizing load cells for liquid filling machines.

The first sloshing mode of liquid in a vessel has a period you can estimate before you build anything. For a cylindrical container, the square of the angular frequency is 1.841 times gravity divided by the radius. Take the square root, divide by two pi, and invert.

Read the table in the opposite direction to your instinct. The small bottle is not the easy one. It rings at just over three hertz, so it finishes more than three full swings inside a one second window and averages out nicely. The drum and the IBC swing so slowly that a one second window contains less than one and a half periods, and the mean of a partial period is not the mean of the signal. The decay column is worse: with the light damping of a bare PET bottle, the slosh is still visible at ten percent after two and a quarter seconds. At 60 bottles a minute each head has one second.

Three things change that, and only one of them is electrical. The first is geometry: a baffle, a narrower neck, a taller liquid column, or a shallower vessel all reduce the amplitude. The second is damping: at a damping ratio of fifteen percent the same bottle settles in 0.76 seconds, and at thirty percent in 0.38 seconds, both of which fit. The third is the weighing window, and that is where a controller earns its keep.

Now the resolution question, answered plainly. The fill is set by when the valve closes, not by what the display shows, so a finer display does not produce a tighter fill. What matters is the controller’s internal resolution and the repeatability of the trip point. Our WST1201 samples its bridge 1,200 times a second with a 300,000 count converter, and the WST1502 indicator runs a 5,000,000 count converter. Internal resolution is abundant and cheap. The reproducibility of the moment the valve shuts is neither, and it is a valve, pump and flow question. I would rather a customer spend the money there.

10. Case two: the drum line where I stopped replacing cells

A blending plant in the Gulf making cutting oils and hydraulic fluids. Two sizes off one machine: a 20 litre pail, about 20 kilograms net, and a 5 litre can, about 5 kilograms net. Both are filled subtractively from a weighed 250 kilogram supply vessel on four cells, with the fills taken off the vessel as a loss in weight.

Their complaint was precise and I liked it immediately. The pails held fine, inside half a percent. The cans were all over the place, plus or minus one and a half percent. The cells under that vessel were ours, they had been in service four years, and they had already replaced the set twice. The request was a quotation for a third set, and a question about whether a class C6 cell would fix it.

I asked for one thing before quoting: the fill weight distribution for both sizes, over one shift. It came back as two histograms. The pails were centred and narrow. The cans were centred and wide. The same cells, the same vessel, the same product, the same afternoon — and one size passed while the other failed.

That is the most useful kind of evidence a customer can send me, because the machine has already run the experiment for me. Two fills of four times the mass difference, one vessel, one cell set. If the cells were the cause, both sizes would fail. So I did the arithmetic I do on every loss-in-weight enquiry, and it explained the histogram in one line: capacity divided by fill is 12.5 for the pail and 50 for the can.

The fourth row is the whole answer and it contains no exotic sensor. A 25 kilogram intermediate vessel fed from the 250 kilogram day tank does the filling; the big vessel becomes a buffer with a level switch. The cells are three WST217 single point cells at 20 kilograms each, IP66 and six wire, chosen because the vessel sits in a wash area and because a six wire cell lets the junction box and the cable resistance stop mattering. That is what I quoted, and what they bought.

What I declined to quote was the third set of cells for the big vessel, and the C6 upgrade. Not because C6 does nothing — the last two columns show it halves the temperature term and cuts repeatability by four — but because at a ratio of fifty it still spends nearly half the legal allowance on the load cell alone, on a machine that already had two sets of perfectly good cells in it. Twice is a diagnosis. Three times would have been a habit. The metering vessel cost more than the cells and it was the cheaper purchase.

Two side effects are worth recording, and they are why I now raise the metering vessel early on every loss-in-weight enquiry. The day tank no longer has to be emptied to change product. And the refill blind spot goes away: while the operator refills the big vessel with a hose, that vessel cannot be weighed, and the zero it takes afterwards is taken with a hose leaning on it and a fresh surface of foam. Decoupling the two roles removes the worst possible moment to establish a tare.

11. Choosing the capacity, and the smallest fill the law will let you certify

There are two limits on capacity and people usually only know one of them. The upper limit is mechanical: the vessel or deck, plus product, plus eccentricity, plus impact, inside the cell’s safe overload with margin. The lower limit is legal, and it comes from the division.

A load cell has a published quantity called the minimum verification interval, and it is simply the rated capacity divided by the manufacturer’s Y factor. Our digital WST1207 states its Y outright — 10 000 for class C3 and 15 000 for class C6 — which lets you write the whole legal envelope on one line:

That number is the finest legal division the cell can support. It is not an accuracy figure and it is not a resolution figure; it is a permission. The rated minimum fill then comes from the staircase in section 3, applied at that division, and the staircase is steep. At a division of ten grams the floor is 330 grams. At twenty grams it is 1 340 grams. So the three hundred kilogram cell in the last row, working at a thirty gram division, cannot be declared for any fill below roughly two kilograms no matter which class plate the machine carries.

Read the two together and a design rule falls out that no datasheet states. Capacity buys you mechanical safety and it costs you legal permission, and the exchange rate gets worse the bigger you go. At the fine end of the staircase the floor runs at about eleven divisions, so a 60 kilogram cell with a four gram interval is usable down to somewhere between forty and ninety grams; at coarse divisions the same table moves to sixty-seven and a hundred divisions, and the floor climbs out of reach.

Put together, the procedure fits on one page. One, write down the net fill, the container tare, any carrier, and the eccentricity. Two, add them and add margin for impact and off-centre load — about one and a half times the static gross is a reasonable start for a machine that drops containers onto the deck. Three, divide by the number of cells. Four, compute capacity divided by fill and compare it with 60 for X(1), 30 for X(0.5) and 12 for X(0.2). Five, if step four fails, do not change the cell yet — change the vessel or the number of weighing points, which is where the leverage is. Six, check the smallest fill the chosen division will certify.

For the sizes we see most often the answer lands in one of three places. Small doses on a bench: the digital WST1201. Bottles and small pails up to about twenty kilograms: the analog single point families, WST201, WST216 or WST217 depending on deck size and washdown. And any weighed vessel: the digital WST1207, where a sealed stainless body makes the temperature and cable arguments mostly go away. Choosing load cells for liquid filling machines is arithmetic before it is catalogue, and section 15 gives each of them a card.

Before any of that, though, the honest first question is not which cell. It is whether the vessel you are weighing has to be that large.

12. Case three: the five gram sachet line I turned down

A enquiry from a contract packer in South East Asia, forwarded to me by a machine builder we have worked with. Five gram sachets of a seasoning blend, four hundred a minute on one lane, target class X(0.2), and they wanted a quotation for load cells for the filler.

It is a good enquiry to answer carefully, because everything about it is at the edge. I started with the legal envelope and worked outward, and the envelope closes fast.

The penultimate line is arithmetic, not opinion: five grams delivered in one hundred and fifty milliseconds is a flow of about thirty-three grams per second, and a valve that takes twenty milliseconds to close passes 0.667 grams of product after the command. That in-flight mass is the drop from section 4, and at this scale it is nine times the entire per-fill allowance. Learning it away handles the mean. It cannot help the scatter, and at this size the scatter is the problem.

So the machine has to dribble, and the dribble itself is not the problem. At five percent of the main rate, about 1.7 grams per second, the residual drop is 33 milligrams — under half the allowance, and a mean that the drop-learning loop can carry. The problem is time. Delivering the last three quarters of a gram at that rate takes about 450 milliseconds, and on a machine running four hundred packs a minute there is no lane with 450 milliseconds to spare.

That is why I did not quote load cells for the filler they described. The architecture is wrong, not the sensor, and no cell I sell fixes a cycle that is shorter than the filling sequence it has to contain. What works at this dose is a multihead weigher, where many small hoppers are weighed at once and the machine combines three or four of them to hit the target — the weighing and the filling happen in parallel instead of in sequence, which is the only way the time budget closes. The small cells inside those hoppers are a product family we do cover, so the conversation did not end there.

One thing I took away from it and now use as a first filter on any small-dose enquiry. At five grams, the control instrument the law asks for needs an accuracy of twenty-four milligrams. That number is the answer to the project. If a customer cannot put a balance of that class next to the line, they are not going to be able to demonstrate compliance no matter which load cell they buy, and somebody should say so before the purchase order rather than after the first audit.

13. Washdown, corrosion and what an IP rating does not tell you

On a dry filling line the second-biggest killer of load cells is mechanical. On a liquid line it is water, and specifically the water that arrives with caustic in it at eighty degrees. The two-digit number on the datasheet is a real test result, and it is much narrower than most people assume.

Four gaps in that table cause most of the failures I get asked about. Chemistry: every one of those tests uses fresh water, and a nitrile seal that survives an IP66 jet will swell and fail in a caustic circulation. The cable entry, where water actually gets in and which is often the least specified part of the assembly. Thermal cycling: a sealed cavity with air inside breathes as it heats and cools, and after a few thousand clean-in-place cycles the O-ring stops being a seal. And standing water in the mounting recess, which no rating protects against because the water is not trying to get in, it is sitting there.

The temperature gap is the one that shows up as a weighing problem rather than a failure. Our standard cells are compensated from minus ten to plus forty degrees, and a clean-in-place cycle runs at around eighty. The cell is not damaged, but its compensation network is far outside the range it was trimmed for, so the zero moves, and it moves back over the next half hour. That is the honest explanation for a machine that needs a fresh tare every shift after a wash, and it is why I ask customers to let the machine return to ambient before they establish the shift tare rather than fighting the controller.

On the practical side, three rules cover most liquid lines. Point the spray away from the weighing zone rather than relying on the rating, because a shroud costs less than a sealed cell. Drain the mounting recess so no water can stand around a cell foot. And match the junction box to the cells, because a perfectly sealed cell connected through a box with rubber grommets is an unsealed weighing system. Our stainless WST1513 multi-channel junction box has sealed cable entries and surge protection for this reason — and the page does not state an IP rating for it, so I would rather confirm that with the factory than have a customer assume it. For site wiring and trimming, the procedure is in Load Cell Junction Box Wiring Diagram and Adjustment Guide.

14. Signals, cables and the junction box

A strain gauge bridge is a remarkably small signal. Two millivolts per volt of excitation is typical, so a cell excited at five volts returns ten millivolts at rated load. Put a one kilogram fill on a five kilogram cell and the useful signal is two millivolts, and a tenth of a gram is 0.2 microvolts. That is the environment in which the cable, the junction box and the adjacent variable-frequency drive all get a vote.

The excitation line deserves a warning, because turning up the excitation is the first thing people try when they want more resolution. Doubling the excitation doubles the signal and quadruples the heat in the bridge, and that heat lands directly on the strain gauges and their compensation resistors. Our WST201 page recommends five to twelve volts with fifteen as an absolute maximum, and that recommendation exists for this reason. If a machine needs more resolution, buy a digital cell with a better converter rather than cooking an analogue one. On the controller side, the WST1505 process indicator keeps a million counts internally while showing 1 in 3,000, with 4-20 mA and 0-10 V outputs and eight relay outputs for driving a filling sequence.

The choice between an analogue summing junction box and digital cells is worth making deliberately rather than by habit. A junction box sums the cells into one weight, which is all a controller needs to run a filler, and it is blind: a mounting that has come loose, a vessel that has picked up a stray cable, or a foot shimmed wrong all appear as nothing at all until they appear as a weight error. A digital cell reports its own corner, so with Modbus or CANopen the controller can watch all four corners and see the asymmetry before it becomes a fill error.

15. The seven cells I put on filling machines most often

We are a component supplier. We do not build filling machines, we do not quote complete filling lines, and we do not take responsibility for the machine’s declaration — that belongs to the builder whose name is on the plate. What we do is size and supply the weighing core: the cells, the junction box, the indicator or controller interface, and the matched test data that lets the builder prove the rig repeats. These seven are the load cells for liquid filling machines that I reach for, and each one carries the limitation that matters as well as the specification.

If none of the seven matches your machine, that is normal and it is not a problem. We make sixteen product families and we do custom versions regularly; the useful thing to send is the fill size, the container, the number of weighing points and the cycle time, and we will tell you honestly whether the machine as designed can hold the tolerance before we tell you which cell to buy. If your application is the small end — liquid dosing in tenths of a gram rather than kilograms — the closest thing I have written is Load Cells for Smart Coffee Machines.

16. What I ask before I quote, and the routine that keeps the number

Nearly every quotation I send for load cells for liquid filling machines starts with the same ten questions. They take a customer ten minutes to answer and they change the recommendation in about half the cases, which is a good enough ratio to keep asking them.

The routine that keeps the number afterwards is shorter still, and deliberately boring. Certified weight on the platform at the start and end of every shift, deck wiped both times, recorded with the initials of whoever did it. A wipe of the weighing zone at every break, written into the shift sheet. The tare re-measured at every container delivery rather than typed in from last month. Sixty consecutive fills whenever the average is in question, weighed on a control instrument that has itself been verified. And a weekly look at the machine’s own drop-correction log, because a drop value walking in one direction is the earliest warning you will ever get, and reading it costs nothing.

17. Questions I get asked about load cells for liquid filling machines