Load Cells for Belt Scales: Class 0.5 and 0.25 Are One Machine

1. The one scale whose reading is an invoice and whose belt never stops

A belt scale is a conveyor that has been taught to keep books. Material rides over a short weighbridge carrying one or more idlers, a load cell assembly measures the force, a speed sensor measures how fast the belt is moving, and an integrator multiplies the two continuously: kilograms per metre times metres per second, summed forever. The output is not a weight. It is a running total of tonnage, and on a barge loader, a rail loading station or a cement kiln feed line, that total is the document people get paid against.

I have now written about three families of weighing machines on this blog, and the belt scale is the strangest of them. At a trade-approved self-service weighing station the reading becomes money in front of a customer, and the whole non-automatic legal chain follows. At a checkweigher the reading only fires a reject gate, which is why the law caps its spread rather than its accuracy. A belt scale combines the worst of both: the reading is money, and there is no operator, and the machine is never allowed to stop. That combination produces a rulebook unlike anything else in weighing, and it is why buying load cells for belt scales is not the same exercise as buying a load cell for a platform.

The first thing to settle is which of two machines you are actually building, because plant engineers say “belt scale” for both. Chinese practice draws the line clearly, and I have found this vocabulary the fastest way to a useful conversation:

The same weighbridge hardware can serve either of the first two rows; the difference is what you ask it to be good at, and the law only cares about the first row. The third row exists because of a physical limit I will show you the arithmetic for in section 11 — and it is the reason I turn business away.

2. Why two regulators describe the same machine with different numbers

Here is the trap, and I have seen it cost a customer a specification argument lasting three weeks. A belt scale that passes class 0.5 in Europe is sold as a class 0.25 belt scale in the United States. It is the identical machine. Nobody is lying, and nobody is rounding. The two rule books simply decided to name a class after different things.

OIML R50 — the international recommendation for continuous totalizing automatic weighing instruments, adopted in China as GB/T 7721 and enforced there through JJG 195 — divides belt weighers into three classes and names each one after its maximum permissible error in service. Class 0.5 is a machine allowed to be 0.5 percent wrong during ordinary use. The tighter number, half of it, is the tolerance you must meet at verification. American practice under NIST Handbook 44 section 2.21 went the other way and named the class after the verification tolerance. So the American 0.25 class and the OIML 0.5 class are the same performance, written twice.

Two things fall out of that table. First, the class name is a language, not a property of the machine. When a supplier quotes you “class 0.5”, ask which document the 0.5 comes from — you will sometimes find it refers to the in-service tolerance and sometimes to the verification tolerance, and the difference is a factor of two on the number you are being sold. Second, everyone has been extending the ladder at the top. R50 names three classes; China legislated a fourth; the Americans added a 0.1 percent class in the 2019 edition of Handbook 44, which is tighter than anything R50 names at all.

That last development is worth understanding because it moved a market. Until the 0.1 percent class existed, weighing bulk grain for export on a belt conveyor was not permissible at all — the only legal route was a garner system, a bulky static arrangement with a staging hopper, a weighing platform and a take-away conveyor, bolted in as an extra stage in the loading process. A certified 0.1 percent belt scale can now do the same job inline, on the conveyor you already have. In the same spirit, the 2016 edition of Handbook 44 made the enclosed weigh belt eligible for certification as a 0.25 percent device for the first time. Regulation does occasionally move to meet the machine, but it moves slowly, and it moves in whole classes.

3. The limit that decides everything, and that nobody reads

Every enquiry for load cells for belt scales opens with a capacity, an accuracy class and a cable length. Almost none opens with the zero. That is exactly backwards, and here is the rule that proves it. JJG 195-2019, China’s verification regulation for continuous totalizing belt weighers, adopts OIML R50 and sets the zero tolerance by class:

Read the middle column as an instruction: run the belt empty for a whole number of revolutions, and the totalizer must not pick up more than 0.05 percent of what full flow would have delivered in the same period. On a 1,200 t/h conveyor, over three minutes, that is 30 kilograms of nothing. That is the legal allowance. The right-hand column is the companion test, and it is there for a reason people rarely spot: a maximum deviation under zero load is really a check on belt thickness uniformity. The regulation says as much, which is why, if you splice in a short replacement section of belt, you must match the thickness of what you took out. A thick patch is a permanent, invisible, unzeroable offset.

Why is the standard so preoccupied with the empty belt? Because of a mechanism unique to this machine, and the Chinese literature states it bluntly: the zero error is accumulated once for every complete revolution of the belt, and then it accumulates again on the next revolution, and the next. A belt scale does not make one measurement and then ask you to accept it. It makes a measurement every revolution, all shift, and adds the results. A zero offset of a few grams of force per metre does not stay a few grams. It becomes a number of tonnes by the end of the day.

The second test is the one that should worry you. R50 and JJG 195 also require a short-term zero stability test: zero the scale, run it empty for fifteen minutes, take a reading every three minutes, and the spread between the smallest and largest readings must remain inside 0.0018 percent of the accumulated load at maximum flow over an hour. On the same 1,200 t/h machine that is 21.6 kilograms of phantom tonnage in a quarter of an hour. Now translate that into a force, because force is what I sell.

Thirty-eight grams. On a weighbridge carrying 533 kilograms of live load and roughly 230 kilograms of belt and deck as permanent dead load, supported by two 1,000 kilogram load cells. As a fraction of the cell pair’s capacity the legal allowance for a quarter of an hour is 0.0019 percent of full scale. And measured against the number that appears on every load cell datasheet — zero temperature effect, typically 0.02 percent of full scale per 10 degrees Celsius, which is 0.2 kg per cell per 10 degrees, or 0.4 kg for a well-behaved pair — the legal allowance is worth 0.96 degrees Celsius.

There is one lever here that costs nothing to pull. That 0.4 kg per 10 degrees is the worst case — it assumes the two cells’ zeros wander in opposite directions, one up and one down. Ask your supplier for a matched pair: two cells whose zero temperature coefficients sit within 0.005 percent of full scale per 10 degrees of each other. The pair drift drops to 0.05 kg per 10 degrees, and the same 0.038 kg allowance suddenly buys you 7.7 degrees Celsius per quarter of an hour. Eight times the budget, for the price of asking for it in writing.

I want to be plain about something here. Our own product pages give the zero temperature effect as a single cell number and say nothing whatsoever about how closely two cells in a pair match each other. So does everyone else’s. It is the most useful specification you can ask a load cell supplier for on a belt scale, it is not on a datasheet, and on this application it matters more than the accuracy class. If you take one practical thing from this article, take that one.

The American framework asks a related but looser question. Handbook 44 section 2.21 requires a zero-ready indicator and demands that zero balance be held within 0.12 percent of rated capacity before a commercial transaction — about 0.64 kg of force on our 1,200 t/h conveyor, some seventeen times the OIML-style fifteen-minute allowance. The comparison is not quite apples to apples, and that is exactly why it is instructive: one rule asks is the zero in the right place? and the other asks is the zero holding still? A machine can pass the first comfortably and fail the second, and the second is the one that costs you tonnage.

Two lines in the two rule books make all of this worse in a way that deserves its own sentence. The EU Measuring Instruments Directive requires that a continuous totaliser’s general totalisation device cannot be reset to zero. Handbook 44 requires that the master weight totalizer cannot be reset without breaking a security seal, and that the reset is inhibited until the total has been printed. Put those together and you get the operating reality of this machine:

4. The conveyor you bolt it to owns most of the error

I keep a list of what has to be true before a belt scale can work. It has fourteen items on it. Exactly none of them is about the load cell. They are about idlers, stringers, the loading point, the belt itself, and the air. When a customer tells me their belt scale is “inaccurate” and asks for a better sensor, this is the list I send back before I quote anything. It is also why I can rarely answer an enquiry about load cells for belt scales with a price.

Look at the third row of that table again, because it is the sentence I most often have to say out loud to a purchasing department. A dual-idler belt scale needs twelve scale-quality idlers. Two load cells. The idler bill is six times the cell bill, and the idlers are not a component of my product. Nobody ever bought a belt scale because of a load cell. They bought it because of eleven scale-quality idlers and a piece of taut string.

Two more numbers govern whether the machine is even in its design range. A belt scale is expected to operate between 20 and 100 percent of rated capacity; American practice goes further and requires the rate-of-flow indicator to signal audibly or visually when flow drops to 20 percent or reaches 100 percent, because those are the boundaries of the region where the instrument means anything. And the belt speed must not vary by more than 5 percent of nominal for a single-speed belt weigher — which sounds like a generous allowance until you notice what it implies. If you do not measure the speed and instead type a nominal figure into the integrator, then every deviation you are permitted to have becomes an error in the mass total. Five percent of speed error is five percent of tonnage error. That is twenty times the tolerance a class 0.5 machine must meet at verification.

5. Case one: the coal yard that asked me for better cells

In March this year a coal handling contractor in Kalimantan, Indonesia, sent me an enquiry that I have now received in some form from four continents. Their barge loading conveyor ran at 1,200 tonnes per hour on a 1,200 mm belt at 3.0 metres per second, the weighbridge carried four idlers at 1.2 metre spacing, and their belt scale was reading about 1.5 percent away from the draft survey when they compared a loaded barge against the totalizer. They wanted to know the price of a matched pair of high-accuracy cells, and they were specific: something in the 0.02 percent class, stainless, IP68.

The arithmetic says their request was reasonable on its face. Live load is 111 kilograms per metre across a 4.8 metre weigh length, so about 533 kilograms sitting on a bridge whose cells were rated 1,000 kilograms each. A 1.5 percent error is enormous compared with the 0.25 percent a class 0.5 machine must hold at verification, and a better cell is the obvious first suspect.

So I asked for seven photographs and two measurements instead of sending a quotation. The photographs were: the weighbridge from both sides, the approach idlers, the loading point, the belt splice in the scale zone, and the sky above the scale. The measurements were: a string line across the tops of the five idlers around the bridge, and the run-out on the approach idlers.

What came back explained the whole 1.5 percent, and none of it was the cells:

• The weighing idler sat 3 mm high against the retreat idler. The tolerance is 0.8 mm.
• The two approach idlers were ordinary production idlers with about 1.5 mm of roll run-out, not scale-quality.
• One idler in the weighing area was a 20 degree trough while every other idler on the conveyor was 35 degrees.
• The loading point was 4 metres upstream of the bridge. Handbook 44 wants 6 metres or five idler spaces.
• A belt training idler sat 6 metres upstream. The exclusion zone is 18 metres.
• The scale was completely uncovered on an open jetty, and the site is a coal terminal on a coastline.
• There were three splices inside the weighing area, at least one of which had a visibly thicker patch.
• The load cell cable ran 140 metres back to a cabinet, on four wires.

I did not sell them cells that month. I sent them a work list: re-shim the five idlers to within 0.8 mm on a string line, leave the scale idler a fraction high, replace the approach and retreat idlers with scale-quality units, change the odd idler to 35 degrees, move the training idler out of the 18 metre zone, move the loading point to at least 6 metres, cover the belt for 9 metres each side of the scale with the side sheeting dropping below the return belt, replace the spliced section with belt of matched thickness, move the junction box onto the conveyor frame and shorten the four-wire run to the cabinet, and then run a material test rather than trusting the old calibration.

They did it over a two-week shutdown. The error came down from roughly 1.5 percent to 0.28 percent, on the original cells. That is a factor of five, from eleven idlers and a string line, and it is not good enough for class 0.5 — the verification tolerance there is 0.25 percent, and 0.28 percent is just outside it. It is comfortably inside class 1 at 0.5 percent.

Only at that point did we have a useful conversation about sensors. Now the question was not “can a better cell fix my scale” — the answer to which was no — but “can I get from 0.28 to under 0.25 without rebuilding the plant”. That is a question a matched pair can genuinely answer, because the remaining error was dominated by exactly the term I described in section 3: the difference in zero behaviour between two cells that had never been matched to each other. We shipped a matched pair with the zero and span differentials written on the certificate, and on the next material test they read 0.19 percent. Class 0.5, with a little margin.

6. Two cells on one bridge, and the number nobody publishes

A weighbridge almost always stands on two load cells, one at each end of the transverse member, wired in parallel so the instrument sees their combined signal. That parallel connection creates a class of error that does not exist on a single-cell scale, and it comes from a specification everybody ignores: the difference between the two cells.

Think about what the instrument is calibrated against. It is calibrated with a known load, applied wherever the calibration rig happens to apply it — usually centrally, so both cells share the load equally. That calibration builds a single scale factor into the instrument, and that factor is an average of the two cells’ sensitivities. Now run the conveyor. The material sits slightly off centre, because the belt tracks a few percent of its width to one side, because the loading point is not perfectly symmetrical, because one stringer is stiffer than the other. The load splits unevenly between the two cells, and the asymmetric parts of their sensitivities no longer cancel. The error works out to a beautifully simple product:

Read the right-hand column, because it is where projects die. A pair of cells matched to one percent — which is a perfectly ordinary pair of good cells, if nobody has matched them — with a twenty percent transverse imbalance, costs you 0.10 percent. That is forty percent of the entire class 0.5 verification budget, spent on a term that appears on no specification sheet. Move to two cells drawn from the outer edges of a wide output tolerance band and the same twenty percent imbalance costs you 0.75 percent, three times the whole budget, from a pair of cells that each individually passed every test on their certificate.

Here is the part that catches people out, and it is the reason I always read the output tolerance line on a competitor’s datasheet before their accuracy line. A rated output of “2.0 ± 0.15 mV/V” sounds like a wide tolerance, and it is: that is a band seven and a half percent wide. It sounds alarming and it is completely normal, because on a single cell the absolute output value is absorbed by the instrument’s calibration. You calibrate the scale against a test load and the factor disappears. On a pair, the absolute values cancel too — but the difference between them does not. Which means the number you need is not on the datasheet at all, and it is not the same number as the tolerance band.

The junction box exists precisely for this. A multi-channel summing box carries trimming potentiometers that adjust the excitation reaching each cell, so you can bring two cells’ spans into agreement on site. I supply those boxes and I recommend them without hesitation. But be clear about the boundaries of the technique. Trimming adjusts the span and leaves the zero alone. It changes the excitation the cell runs at, so the two cells no longer share the same self-heating and the same thermal environment. And it cannot exceed its trim range — you cannot trim a cell whose output is seven percent off nominal back into a matched pair, because the trim runs out first. The junction box is for the last tenth of a percent, not for the first percent.

7. The weighing layer is mine; the machine is not

This article has spent three sections telling you that the load cell is not the main problem. It would be a strange way to sell load cells if I did not now say exactly what I do sell, and where the boundary sits. I export weighing components: load cells, force sensors, junction boxes and weighing instruments, with test data and application advice. I do not build belt scales, and I am not a systems integrator. The distinction matters commercially, because a project that assumes I am the second thing will discover I am the first thing at an inconvenient moment. Anyone quoting load cells for belt scales should be able to draw that boundary in a single sentence.

Two of those left-hand rows are worth enlarging. The right-hand column is not modesty and it is not a disclaimer — it is the reason the rest of this article is useful. I have no commercial interest in whether you fix your idler alignment, because I do not sell idlers, and I have no interest in whether you cover your conveyor, because I do not sell covers. When I tell you that the idler bill is six times the cell bill, you can take it at face value.

And the matched pair is the one place where the component supplier genuinely owns the outcome. Everything else in this article belongs to the conveyor, the site or the integrator. The zero drift of the pair, the span match, the creep under a belt that never leaves the cell — those are mine, they are decided in my factory, and they are the reason a belt scale customer should be talking to a load cell maker at all rather than simply ordering from a catalogue. There is more on how I approach that kind of component-level work in my selection guide, and on what happens when a weighing element fails, in the fault-finding note I wrote after getting the same question from three plants in one month.

8. Sizing a cell around a belt you did not design

Sizing load cells for belt scales starts from a quantity most people have never calculated: how much material is sitting on the belt, per metre, right now. The scale’s own arithmetic gives it to you directly. From Q = 3600 × q × v, where Q is tonnes per hour, q is kilograms per metre and v is belt speed in metres per second, you get q = Q divided by 3.6 v.

Now add what is always there. The belt itself, the weighing idlers and the floating deck sit on the bridge permanently, whether material is flowing or not. On our 1,200 t/h conveyor that is roughly 230 kilograms of dead load on top of the 533 kilograms of live load. The cells carry it 24 hours a day, including through the night when the plant is idle, and that has a consequence that no platform-scale datasheet anticipates: the cell is permanently preloaded, so whatever creep the element has under that preload becomes a standing zero offset that never relaxes. On a platform scale the load comes off between weighings and the element recovers. On a belt scale it never does. The offset has to be taken out by the zero-setting device, which is exactly why the rule books insist on a zero-setting device and a zero-ready indication in the first place.

So capacity has to cover dead plus live plus shock, and here is where the belt scale fights itself. Look at what happens to the zero budget as you increase the cell size, holding the live load fixed at 533 kilograms and assuming two cells whose zeros drift in opposite directions at the ordinary 0.02 percent per 10 degrees:

Every step up in capacity that buys you structural headroom costs you zero stability, in direct proportion. Going from 1,000 to 2,000 kilogram cells takes the burden from 38 percent to 19 percent — comfortable, no risk of shock damage — and halves your legal zero budget from 0.96 to 0.48 degrees. There is no cell specification that resolves this, because the two requirements pull in opposite directions on the same number. The only way out is to make the dead load smaller, which is a mechanical problem: stiff, light, counterbalanced weigh frames exist precisely to keep the tare off the load cell, and on high-accuracy machines the frame is explicitly mass-counterbalanced so the cells see a small fraction of the belt’s weight. That design is the machine builder’s, not mine.

There is a second floor under the machine, and it is measured in mass rather than time. R50 and the EU directive both define a minimum totalised load: class 0.5 may not certify a total below 800 scale intervals, class 1 below 400, class 2 below 200. The interval itself is bounded too — it must lie between 0.01 and 0.2 percent of maximum capacity, and Handbook 44 says the same thing from the other side, that a scale division may not exceed 1/800 of the minimum totalised load. Those are not two rules; they are one rule read from opposite ends. Put a one kilogram interval on our machine and both agree it is legal, with the minimum totalised load landing at exactly 800 kilograms.

Read that table slowly, because it contains a trap. On a fast conveyor the minimum totalised load passes in two and a half seconds and you will never think about it again. On a slow one the same legal floor takes twenty-four minutes — and since the 20 percent flow floor bites in exactly the same region, a low-flow belt scale fails two independent requirements at once. That is the arithmetic behind section 11.

9. The cable run that quietly costs you a fifth of a percent

The cabinet on a minerals plant is never next to the scale. It is 100 metres away in a control room, on the other side of a road, and the load cell signal has to get there. That distance is a metrological component, and it is one of the few places in this article where a decision made purely on price shows up in the accuracy certificate.

The classic arrangement is four wires: two carry the excitation out and two carry the signal back. The excitation current flows through the cable, so it loses voltage on the way out, and the loss depends on the cable’s resistance. For 100 metres of 0.5 mm² copper, the loop adds about 6.9 ohms in series with the cell’s 380 ohm bridge. Copper’s resistance rises by about 0.39 percent per degree Celsius, so when the cable warms by 20 degrees the series resistance grows by almost 8 percent, and the fraction of the excitation lost in the cable grows with it. The scale factor moves with the cable. The effect is small and it is not negligible:

Fifty-six percent of your entire legal tolerance, spent on the temperature of a cable. Two observations worth carrying away. First, a higher bridge resistance helps: the 380 ohm input resistance on the cells I recommend is slightly better here than a 350 ohm bridge, because a larger bridge makes the cable’s series resistance a smaller fraction of the loop. Second, and counter-intuitively, lowering the excitation voltage does not help — the loss is a ratio of cable resistance to total loop resistance, so running the cell at 5 volts instead of 10 leaves the percentage unchanged.

The fix is not a better cell, it is a different architecture, and it is why I quote load cells for belt scales together with a junction box and a transmitter rather than as a bare part number. My recommendation for any run beyond a few metres: mount the WST1513 multi-channel junction box on the conveyor frame within sight of the weighbridge, bring the cells into it on short leads, and from there run either a six-wire cable with sense leads or convert to a current loop at the scale. A WST1508 transmitter does the conversion: 4 to 20 mA, 0 to 10 V, isolated RS485 with Modbus RTU, up to twelve 350 ohm cells, and an operating range from minus 10 to 40 degrees that matches what the rule books expect of the installation. Once the signal is a current loop, cable resistance is somebody else’s problem.

10. Case two: the quarry where the cells died every six months

The second case is not about accuracy at all. It is about a hard-rock quarry in the Middle East that had replaced the load cells on one belt scale four times in two years, at roughly six-month intervals, and was starting to believe the cells were simply poor quality. They were our cells, so I took it seriously.

Their conveying duty was 900 tonnes per hour, and the pattern was informative: the failures were mechanical, on the same cell each time, and they followed rainy season. That combination — same cell, seasonally correlated, mechanical rather than electrical — told me to look upstream of the bridge, so I asked two questions: where does the truck tip relative to the scale, and what is the gap on the overload stops?

The loading point was 2.5 metres from the weighbridge. The band was recommending 6 metres, and Handbook 44 wants 6 metres or five idler spaces, and both of them mean the same thing: a skip of 20 tonnes of rock landing on a belt produces a travelling wave that has to be absorbed by the bed and the stringers before anything can be weighed. At 2.5 metres the wave was arriving at the weighbridge almost intact, on every load, all day. And the overload stops were set with 4 millimetres of clearance where the manual calls for about half a millimetre — eight times too much travel. The stops were supposed to catch exactly this event and they were never in a position to touch it. The cell was doing the job the stops were designed for, four times a year.

Two changes fixed it, and neither was a specification. They moved the scale downstream so the loading point sat beyond six metres, and they re-set the overload stops to the specified clearance. We also changed the cell model on the next order, not for accuracy but for overload margin: our large single-point WST204 is rated for 120 percent safe overload, which is generous on a floor scale and thin on a belt scale, whereas the WST303 and WST401 carry 150 percent. That difference exists on our own pages, it is the kind of thing that gets lost between a purchasing list and a drawing, and on this duty it is worth more than an accuracy class.

11. Case three: the 1.5 tonne per hour line where I said no

The third case is the one I am least comfortable writing, because the customer was right about everything except the machine. A sand and dry-mortar plant in Southeast Asia wanted a belt scale on a short conveyor handling about 1.5 tonnes per hour, running at 1.25 metres per second, and asked me to recommend cells for it. Their previous integrator had told them the problem was resolution at low flow and that a high-resolution digital cell would solve it.

Run the numbers and the machine disappears. At 1.5 t/h and 1.25 m/s the belt carries 0.33 kilograms per metre. Over a single idler at 1.2 metre spacing, the live load on the weighbridge is 0.4 kilograms. Against roughly 120 kilograms of belt and idler dead load. The dead load is three hundred times the live load, and the live load is four hundredths of one percent of the capacity of a pair of 500 kilogram cells.

Now put the zeros back in. That pair of 500 kilogram cells carries 0.2 kilograms of zero drift per 10 degrees Celsius of ambient change — with a good unit, on a normal day. Relative to a 0.4 kilogram signal, the zero drift is fifty percent of the reading, per 10 degrees. There is no cell on the market that changes this. A load cell’s zero stability is proportional to its capacity; the signal on a belt scale is proportional to the flow. Push the capacity down to help the ratio and the belt’s own dead weight exceeds the cell’s rating; push the flow up to help the ratio and you are no longer building the plant the customer has.

I worked out what it would take for the live load to reach even twenty percent of the cell capacity — a modest ambition — and the answer was about 750 tonnes per hour. Five hundred times their actual flow. At that point I stopped looking for a cell and told them the truth: a belt scale cannot weigh 1.5 tonnes per hour, not because belt scales are inaccurate but because at that flow the weighing element is trying to detect a few hundred grams sitting on a moving rubber belt that weighs a hundred times more, in wind, with the belt stiffness changing all day.

What they needed was a different machine. A loss-in-weight arrangement — a hopper on load cells, discharging through a feeder — puts the entire hopper content on the weighing element, so the live load becomes tens of kilograms instead of hundreds of grams, and the same cell that was hopeless on a belt becomes perfectly adequate. We quoted cells for the hopper instead, and a small junction box, and their existing supplier kept the feeder. That is the correct outcome and it is also the one that cost me a larger order.

12. Chain code, test weights, and the only test that proves anything

A belt scale cannot be calibrated with a dead weight, because it does not measure a dead weight. It measures a moving bed. The industry solves this in two families of method, and the difference between them is the difference between proving a machine repeats and proving it is right — two entirely different claims that get confused constantly.

The size of the gap between a simulated load and the real thing is genuinely uncomfortable. One published comparison on a belt scale of ordinary commercial quality found chain-code calibration and material calibration stable within about 3 percent of each other once the zero had been properly set — twelve times the tolerance a class 0.5 machine is allowed at verification. I would not want that single figure to be read as a general law, and I have no doubt that a well-installed scale on a modern weigh frame closes the gap far further. But the ranking is not in doubt and it is what matters: a chain code proves your machine repeats. Only material proves it is right.

One arithmetic detail from the manufacturers’ manuals that makes weight calibration less arbitrary than it looks. When you calibrate with test weights, the number you enter into the integrator is not the total weight of the weights. It is the total weight divided by the idler spacing, expressed in kilograms per metre — the same units as the q in section 8. Stack 1,200 kilograms of test weights on a bridge with 1.2 metre idler spacing and you are telling the instrument it is looking at 1,000 kilograms per metre, which is what a full conveyor would put on it. Get that division wrong at commissioning and every total from then on is wrong by the same factor, which is at least a diagnosable failure and better than a drifting one. None of this has anything to do with load cells for belt scales, and that is the point: the cell is calibrated out of the conversation and the conveyor is calibrated into it.

Frequency matters more than method, and here the guidance from people who operate these machines for a living is more aggressive than most plants expect. The recommended rhythm, drawn from Chinese practice and consistent with what belt scale manufacturers put in their manuals:

Two procedural habits come from the belt scale manuals and I recommend both without qualification. Run two calibrations before adjusting anything, and only accept the result if the two agree — because a single calibration on this machine tells you nothing about whether you just measured the scale or measured the weather. And run the conveyor empty for 20 to 30 minutes before zeroing, because a cold belt has different stiffness and tracks differently from a warm one, and you are about to define zero on whichever belt you happen to have that morning.

The record keeping is part of the legal instrument, not an administrative afterthought. Handbook 44 requires a belt-conveyor scale system to record the unit of measurement, the date and time, the initial and final indications of the master weight totalizer and the quantity for each delivery; to retain accumulated quantities through a power loss of up to 24 hours; and to keep calibration, maintenance and test records for at least three years. If you are buying an indicator for a legal-for-trade belt scale, that list is your minimum functional specification, and date and time stamping is not optional.

13. What half a percent is worth on five million tonnes

Half a percent sounds like nothing, and on a kitchen scale it is. On a bulk terminal it is a line in the annual accounts, because accuracy class on a belt scale is not a quality attribute, it is a price. It is also why the class you pick for your load cells for belt scales turns up in the annual accounts rather than in a maintenance report.

That is the arithmetic that explains the class ladder I described in section 2. Nobody added a fourth Chinese class or an American 0.1 percent class because engineers enjoy precision. They added them because at these tonnages the classes are separated by millions of dollars a year, and because the same error appears on the other side of the transaction too. Belt scale manufacturers make the point from the loading side: overfilling rail cars and trucks by two to three percent because the scale under-reads is a routine and invisible loss, multiplied across hundreds of loads, before anyone mentions customer disputes or regulatory fines.

This is also why I am sceptical of headline accuracy claims, as I said in section 2. A number means something only when it comes with a class, a calibration method and a maintenance regime. A machine reading 0.15 percent today on a weigh frame nobody re-aligns can easily be worse in service than a modest class 1 machine that gets its daily zero and its quarterly material test. On this application the maintenance regime is worth more than the class, and it is the cheapest thing on this page to buy.

14. Which rule book applies where

A belt scale crosses several rule books, and the most common mistake I see in project documentation is a specific one worth stating plainly: continuous totalisers are covered by the EU Measuring Instruments Directive under Annex VIII, MI-006, Chapter V. They are not under MI-005. MI-005 is measuring systems for liquids, and a surprising number of people carry that number over from the liquid world because it comes first alphabetically in the annex list.

Read the middle column and a pattern appears that is the same across every market: the machine is approved, and what I ship is evidence, not permission. There is one further nuance in the load cell column that is worth a sentence, because it cuts against the instinct to ask for the most certified component available. Under the EU framework, a load cell equipped with electronics has to meet a specified electromagnetic environment depending on what it is attached to — class E1 or E2 under MID for automatic weighing, with its software assessed under WELMEC 7.2, against the corresponding R76 requirements including surge testing if it goes onto a non-automatic instrument. The environmental class is a design decision about your installation, not a badge on a part, and getting it wrong produces a machine that passes type evaluation and misbehaves in a substation. It is one more reason that load cells for belt scales are an engineering conversation rather than a catalogue lookup.

Practically, this is what the boundary looks like in a project. I can give you a cell from a certified production line, with its class and Y value documented, a calibration certificate per serial number, and — if you ask — the zero and span differentials of the pair, which no certificate template has a box for. I cannot approve your belt weigher, sign its verification, supply the integrator that computes the total, or describe a certified cell as making a machine compliant. If a supplier tells you that fitting a certified load cell makes your belt scale legal for trade, that supplier has the paperwork backwards, and you should find that out now rather than at the first inspection.

15. Load cells for belt scales: the five parts I would quote

The weighing layer of a belt scale is three things, not one: the cells under the bridge, the box that brings them together and trims them, and the instrument that reads the result. The five parts below cover all three, and they are what I would actually put on a quotation for a conveyor in the 200 to 1,000 kilogram per cell range, in a sheltered or at least covered location. Read the trade-off in each card rather than the headline — on this application the limits decide the project, not the features.

Two things I want on the record before the cards, because they are the kind of detail a catalogue edit never catches up with. Our own product pages contain a copy-paste error that has survived several catalogue revisions: the WST201 page still opens with the sentence “WST204 is hot sale”, so searching our site for WST204 returns two pages and one of them is the wrong product. And several of the pages below state a rated output of 2.0 mV/V with no tolerance attached at all, which is precisely the number a belt scale buyer needs most. I would rather tell you to ask us for it than have you assume it is tight.

And the honest gap, which is the same one as always on a belt scale: bridges above about a tonne per cell. None of the five reaches there except the S-type on a suspended frame, and on a compression weighbridge over a tonne you should be looking at our WST302 shear beam, which runs from 0.05 to 20 tonnes with 150 percent safe overload and a comprehensive error of 0.03 %F.S. I have kept it out of the cards because its page also omits an IP rating and a compensated temperature range, and I would rather you saw those gaps in prose than found them after drawing the frame. Tell me the belt width, the speed and the dead load, and the capacity question answers itself.

16. Seven answers I need before I can size anything

Every belt scale enquiry that ended badly for somebody ended at one of these seven questions, and not one of them is about a sensor. I do not need exact numbers — I need the shape of the machine. Send these and my reply stops being a price list and becomes a design. It is also the list I would give anyone comparing load cells for belt scales, precisely because none of the seven is about a sensor.

17. Questions from conveyor builders and plant engineers