Load Cells for Wheelchair Scales: Max Divided by Four Is Not a Rating

1. A legal instrument before it is a scale

I have quoted cells into cash registers, belt weighers, filling nozzles, labellers and truck axles. Almost all of those conversations begin with the same question, which is what the load will be, and almost all of them end with an argument about price. A wheelchair scale is the only platform I quote where the useful argument happens before the mechanics, and it is not about price at all. It is about what the number is legally allowed to say.

The first time this landed on me I was standing in a physiotherapy room in the Netherlands with an integrator who had already built the platform and was shopping for four cells to go underneath it. His requirement list was two lines long: 300 kilograms, 0.1 kilogram divisions, four cells, aluminium, IP65, send the matching data with the delivery. He had the platform welded, the ramp cut, the handrails bent and the junction box bracket made. What he did not have was a reason to believe any of it was legal, because nobody had told him that a scale which weighs a patient is a regulated measuring instrument, and that the regulation is not the one he was expecting.

A load cell for wheelchair scale platform sits inside a machine that is doing three things at once. It is measuring a human being, so a medical device regime reaches it. It is measuring mass with a declared scale interval, so a legal metrology regime reaches it. And it is carrying a piece of equipment that weighs between seven and one hundred and eighty-one kilograms depending on whether it is manual, powered or a heavy-duty bariatric chair, so a mechanical regime reaches it as well. Only one of those three is mine to engineer. The other two are the reason I have to say no to things.

So the shape of this article is: the law first, then the mechanics, then the arithmetic. If you only want the number to give your platform designer, it is 125 kilograms per cell for a 300 kilogram machine with 100 gram divisions, not 75, and section five shows where the extra fifty comes from. If you want to know whether the cell will survive being rolled onto, section ten. If you want to know what a clinic does with the result once it has it, section thirteen, and it is not what most integrators assume.

One warning before the mechanics, because it is the most expensive mistake I see. Wheelchair scales get sold in two flavours that look identical on a quotation. One is a piece of fitness equipment with a display. The other is a verified class III instrument with a marking on it, and its platform, its indicator, its temperature compensation, its test procedure and its paperwork are all different. The cells underneath the two machines can be the same part number. The difference is not in the cell. It is in everything you have to prove once the cell is inside.

2. Seven regulated purposes, and only one of them is a price

In the European Union the instrument that governs non-automatic weighing is Directive 2014/31/EU, usually called NAWI. Its first article limits the directive to instruments used for a list of purposes, and the list is longer than most integrators expect. There are seven entries, and money is only one of them. The fourth reads: the determination of mass in the practice of medicine for weighing patients for the purposes of monitoring, diagnosis and medical treatment.

Read that sentence twice. It does not say trade. It does not say price. It says monitoring, diagnosis and medical treatment, and it puts the weighing of patients in the same article as the weighing of goods, eating side by side with commerce. Price is one of seven heads of regulated use, which makes it about fourteen percent of the reason a scale can become a legal instrument. Every medical weighing article I have written before this one concluded that the trade regime did not apply. This is the first one where it does, and it applies because of four words in a sentence about medicine.

What the marking on the machine actually says

A verified patient scale carries four things next to each other: the CE mark, the letter M, the year it was first put into service, and the four-digit number of the notified body that did the work. The M is the metrology suffix, and it is a different claim from the CE mark on the same label. If you are designing a scale for the European market and you have never seen an M on a nameplate, look at a hospital bed scale or an infant scale the next time you are in a clinic, because the M is the part your cell has to be capable of supporting, and the cell cannot supply it by itself.

There is a second detail that surprises people who come from the trade side of weighing. NAWI gives a presumption of conformity through harmonised standards, and the harmonised standard for non-automatic weighing in Europe is EN 45501. The OIML recommendations, including R76, are the international technical documents that sit behind it, and the OIML’s own published position is that its recommendations do not by themselves confer a presumption of conformity with the directive. The two documents are nearly identical on paper and are not interchangeable in a file. I mention this because a great many quotations I receive have an R76 test certificate in the attachments and a claim of European compliance in the cover letter, and the second does not follow from the first.

The German version, because it is the strictest one I sell into

Germany puts the directive into national law through its verification ordinance, and adds obligations that do not exist everywhere else. Medical scales used in hospitals fall into the group that must be verified again on a fixed cycle rather than once, and the cycle for patient scales in a hospital setting is four years. A newly acquired scale has to be declared for verification within six weeks of being put to use, not at the end of the financial year. The penalties for using an unverified instrument in a regulated setting reach fifty thousand euros, and the penalty for failing to declare a new one reaches twenty thousand.

The reason I care about the German numbers is not the money. It is that they explain a purchase pattern I could not otherwise account for. Twice now I have had a German integrator ask for a cell that is more accurate than his own stated specification, with the words, we would rather pay for margin than for a second verification. A four-year cycle means that every drift mechanism in the machine has four years to accumulate before anyone inspects it, and the cell’s zero temperature coefficient accumulates continuously from the day it is installed.

What follows from all of this for a cell supplier is narrow. Your cell has to have the metrological characteristics for the class you are claiming, a temperature behaviour that survives a four-year cycle, unit-to-unit consistency or your corner adjustment will not hold, and replaceability by an identical part after approval. That last one is why I keep the matching data on file and why I tell buyers to ask for it before they freeze a design, not after.

3. What I supply, and whose signature is on the rest

I export weighing components: load cells, force sensors, matched sets with the matching data in writing, summing junction boxes, transmitters and weighing indicators. I do not build scales, I do not hold a medical device registration, I cannot issue an EC type-examination certificate, and I cannot put an M on your nameplate. On a trade scale that boundary is a commercial fact. On a patient scale it is also a legal one, because the person who signs is the person who is responsible for the whole machine, and no component supplier can move that signature onto himself.

Two of the rows in that right-hand column deserve a sentence each, because they are where suppliers quietly overstep. The overload protection row is there because a wheelchair scale can find itself described as having a patient-safe limit, and the moment that limit is used to protect a person rather than a sensor it stops being a weighing component and becomes something I am not permitted to design. The declaration row is there because I have been asked, more than once, whether I could include a note in the shipment that would help a customer’s declaration. The answer is no. A document from a parts supplier describing a machine he has never seen is not evidence, and presenting it as evidence would be worse than sending nothing.

What I can do is narrower and more useful. If you send me the platform drawing, the capacity and the interval you intend to declare, I will tell you in writing which of the three metrology conditions in section six your chosen cell meets, which one it fails, and what capacity or class would satisfy all three. That is arithmetic on your numbers, not approval of your machine, and I will label it that way.

4. Where this sits beside the other patient-weighing jobs

We have published seven articles about weighing people and the fluids attached to them, and none of them is about load cells for wheelchair scales, so the fastest way to waste an afternoon is to read the wrong one. All of them are a human being on a sensor. None of them is the same problem, and the division that matters is whether the reading is a legal quantity, a clinical trend, or a control signal. The table sorts them.

The nearest neighbour is the medical bed, and the difference between them is the whole article. A bed scale is bolted under a bed that stays in one place. Its tare is large and constant, the patient arrives and leaves without the furniture moving, and the interesting error is the load’s own activity. A wheelchair scale has a tare that arrives with the patient and leaves with the patient, is between eighteen and one hundred and eighty-one kilograms, changes whenever someone hangs an oxygen cylinder on the back, and is standing on four feet whose exact positions you do not control because the chair is pushed on by hand. The bed’s tare is a number. The wheelchair’s tare is a piece of equipment.

The second boundary is against the urine article, and it is about direction rather than furniture. A urine bag accumulates. A clinician reads its total as evidence about an organ, no legal instrument is involved, and the sensor’s job is to not drift for three days. A wheelchair scale reads once, in a second, and the number is a legal quantity with an interval and a maximum permissible error attached. Same hospital, same corridor, and one reading is a measurement while the other is a certificate.

5. Four cells does not mean divide by four

Here is the sentence that starts almost every enquiry I answer, and the sentence I have started replying to with a question. A three hundred kilogram platform with four cells needs four seventy-five kilogram cells, because three hundred divided by four is seventy-five. It is arithmetic, it is correct, and it produces the wrong part number every time. The four cells do not carry three hundred kilograms. They carry everything that is standing on the frame as well, and they carry it unevenly.

The correct expression has five terms, and OIML R76 writes it in a form you can use directly. The sum is the maximum capacity you must declare, plus the dead load of the platform itself, plus the initial zero-setting range, plus the allowance for non-uniform loading, plus the tare you intend to place on it. Then you divide by the number of cells, and then you correct for the fact that the load does not sit in the middle.

Five hundred divided by four is one hundred and twenty-five kilograms per cell. The seventy-five in the enquiry is too small by a factor of 1.667, and the failure mode of believing it is not a scale that reads wrongly on day one. It is a scale that passes its factory test, ships, and then starts going non-linear whenever a heavy patient sits on one side, because the cell that was supposed to be at 50 percent of its range is at 83 percent and the overload margin that the data sheet promised has quietly gone.

The same expression at three platform sizes

The last column does not pick the cell just above the calculated number, because those numbers are not catalogue sizes. It picks the next size up, for the reason that applies to every weighing job: the calculated figure is the load the machine sees on its worst day, and the cell has to be comfortable there. Four 150 kilogram cells give 600 kilograms of cell capacity against a declared 300, a factor of 2.0, and 1.20 against the 500 kilogram sum. That is the margin you are buying, and it is the number I put in front of anyone who tells me a 300 kilogram scale only needs 300 kilograms of cells.

There is one more check that people forget, and it is the reason I ask for the ramp drawing. If the ramp is carried by the same frame as the platform, the ramp is part of the dead load and you have counted it. If the ramp is a separate structure that a wheelchair rolls across on its way in, it is not, and the transition at the joint is where the worst instantaneous load arrives. Section ten runs that number.

6. Three formulas that decide whether your cell can be legal

Everything in section five tells you what the cell has to carry. None of it tells you whether the cell is allowed to be there, and that is a separate set of conditions with a separate vocabulary. OIML R76 and EN 45501 express the relationship between a whole instrument and its individual load cells through three inequalities. I run all three on every wheelchair scale enquiry, and the third one is the one that changes people’s designs.

Take the enquiry from section one: 300 kilograms, 100 gram divisions, four cells. The declaration gives n = 3000 intervals, which sits comfortably inside the range a class III instrument is allowed to occupy. The third condition gives 125 kilograms, which is 150 in catalogue sizes. So far everything is easy and the buyer is happy. The first condition is where the enquiry dies, and it is worth doing slowly.

For a set of several identical cells the reduction ratio is taken as one, which is the conservative choice, and with four cells the square root is two. So the limit on the cell’s minimum verification interval is one hundred grams divided by two, which is fifty grams. The 150 kilogram cell is rated class C3, which means three thousand intervals over its range, and its minimum verification interval is therefore 150 kilograms divided by 3000, which is exactly fifty grams. It satisfies the condition with nothing left over. Zero margin, and the reason is not a coincidence: a C3 cell of 150 kilograms and a class III instrument of 3000 intervals with 100 gram divisions are the same instrument seen from two sides.

The table has an unpleasant second reading: a bigger cell is not more capable, it is less. If the frame is heavy enough that the five-term sum pushes the per-cell figure to 200 kilograms, no class C3 cell of that size can support a 100 gram interval in a four-cell arrangement, however good it is. The way out is not a better cell of the same size. It is to buy a cell with a finer class, or to reduce the number of cells, or to accept a coarser interval. That is a sentence I have now written into three separate load cells for wheelchair scales quotations, and it has never once been the answer the buyer wanted.

The same inequality read backwards

Turn the first condition around and it tells you what the cell requires of you rather than the other way. For a 150 kilogram cell at 100 gram intervals and four cells, the cell’s own rating has to reach 3000 intervals, which is C3 and is what the data sheet says. Push the capacity up and the requirement climbs: 200 kilograms needs 4000 intervals, half a tonne needs 10000, a tonne needs 20000, two tonnes need 40000. Every one of those is a finer class than the one most people assume when they say, I will just use a bigger cell to be safe. On a verified scale, being generous with capacity is a way of failing the metrology.

That last row is the reason I am willing to discuss single-cell designs at all, and section eight explains why they do not survive the geometry. One cell has no square root and no reduction, so a 300 kilogram class C3 cell with 100 gram divisions is mathematically legal on paper. It is the platform that refuses.

What the accuracy class actually promises

Class III does not mean one accuracy. It means three, depending on where in the range you are, and the boundaries are expressed in intervals rather than kilograms. The first band runs from zero to five hundred intervals and carries a maximum permissible error of half an interval. The second runs to two thousand and carries one interval. The third carries one and a half. On our 300 kilogram machine with 100 gram intervals, those boundaries land at 50 kilograms and 200 kilograms, so the permitted error is fifty grams, then one hundred, then one hundred and fifty — not one hundred, which is what almost everyone assumes the last band is.

The in-service column is double the verification column, and it is worth understanding why rather than memorising it. The error a scale is allowed doubles once it leaves the workshop, because a verified instrument is permitted to drift through normal use and is expected to be brought back within the verification limits at its next inspection. On the German four-year cycle that means an instrument may legally sit two divisions and a half away from truth at the end of its cycle. Your cell has to have spent those four years not using up that allowance.

7. The chair is the load

On a bathroom scale the person is the load and the tare is the plastic housing. On a wheelchair scale the tare is a steel and aluminium vehicle that arrives under its own power and leaves again, and on a powered chair it is heavier than the patient. Here are the ranges I work to, assembled from what chair manufacturers publish and from what I have seen on customers’ drawings.

Two things follow from that table and neither is about accuracy. The first is that a bariatric machine can be filled to two thirds of its capacity by a person who is not on it yet. In the 300 kilogram example the bariatric chair alone is 1810 divisions of a 3000 interval instrument, so the useful measuring range above it is only the last 1190 intervals, which is where the coarsest permissible error lives. That is an argument for a 500 kilogram platform even when the patients do not need one, and it is the argument I make when a customer wants to save money by declaring 300.

The second is that the tare is not stable. Oxygen cylinders, infusion poles, drainage bags, pressure cushions and footrests are added and removed by clinical staff all day, and they are added to the chair rather than to the patient. Five of them, one of each, come to 7.7 kilograms in the example I use with customers. Fifty of those are 77 intervals of a 100 gram instrument, and at a hundred kilogram load the permissible error is one hundred grams, so the accessories alone are seventy-seven percent of the permitted error at the working point. A single oxygen cylinder is thirty percent of it. The tare you entered when the machine was installed is stale before the first patient of the afternoon.

8. Why four cells, and why one is impossible

Every few months someone asks why a wheelchair scale needs four cells when a single-point cell would need no corner adjustment, no summing box and no matching. It is the right question, and there is a single number that answers it. A single-point cell is specified for a maximum platform size, because the whole point of its construction is that it compensates for where the load sits. The big single-point cell in our own catalogue is rated for a platform of six hundred millimetres, and its smaller sibling for four hundred.

A wheelchair scale platform is not six hundred millimetres. Chairs are between six hundred and ten and six hundred and eighty-six millimetres wide, and a heavy-duty chair is seven hundred and sixty-two or more, and the platform has to be bigger than the chair with clearance on both sides because a patient is being pushed onto it by a person who cannot see the castors. The platforms I see in this market are eight hundred and fifty millimetres square, a metre square, or somewhere near a metre by a metre. Against a metre square, the largest single-point cell we make is two and a half times too small, and no amount of careful mounting changes that.

The castor is not a test weight

This is the part of the design that has caused me the most argument, and it is worth putting a number on because the number is startling. A castor touches the platform over a contact patch of roughly twenty by thirty millimetres. A fifty-five kilogram corner load over that patch is a contact pressure of about 899 kilopascals. The twenty kilogram test weight that will be placed on the platform during verification sits on a two hundred millimetre disc and produces 6.24 kilopascals. The castor presses about one hundred and forty-four times harder than the weight that certifies the scale.

The consequence is not that the cell fails. It is that the platform surface behaves differently under the two loads. A one metre wide plate in six millimetres of steel deflects about 3.1 millimetres under a mid-span corner load, and doubling the thickness to eight millimetres only brings that down to 1.3, which is still a visible dip and still a stiffer plate than most of the platforms I am shown. That is why every wheelchair scale I have seen that survives has a channel or box frame under the plate rather than a plate alone, and it is why the frame’s weight ends up in the dead load term in section five and costs you cell capacity. The plate and the cell are the same decision.

9. The low-profile cell you need is the one that drifts most

Section eight ended with a geometric requirement that decides which load cells for wheelchair scales you can actually use: the platform is a metre square, the underside clearance is a few centimetres, and the cell has to live in that gap without lifting the plate. In practice that means a planar beam cell, screwed flat to a channel frame, with a body a few millimetres taller than the plate. Our planar beam family has exactly the right form factor and exactly the wrong temperature coefficient, and the two facts are connected.

The two low-profile cells drift two and a half times more than the cells that will not fit. That is not a defect in them; a planar beam bonded flat to a frame has a different thermal path from a free-standing single point with air around its body, and moving the strain gauge geometry closer to the mounting surface tends to move it closer to whatever temperature the platform frame happens to be. It is also not a reason to avoid them. It is a number to carry into the error budget, and the number is larger than most people expect.

Work it through on the 300 kilogram machine. The full-load band permits one hundred and fifty grams, and the R76 allocation from section six gives each of the four cells seventy-five grams. A 0.05 percent coefficient on a 150 kilogram cell is 75 grams for every ten degrees at rated load, which is the entire allowance in one step. At the actual working point of 125 kilograms per cell it is 62.5 grams, and four cells contributing in the same direction come to 250 grams for ten degrees, which is 1.667 times the permissible error of the instrument. A 0.02 percent cell gives 25 grams at the same point, or one hundred grams across four cells, which stays inside the one hundred and fifty.

Twelve degrees is not a laboratory condition. It is the difference between a treatment room that has been at twenty-two all morning and a chair that has just come in through an ambulance entrance in February, or between a platform under a window and the same platform after the air conditioning has been running for an hour. Thirty degrees is the difference between a heated building and an unheated one. That is the whole distance between the two families, and it is why I ask about the building before I ask about the load.

10. What voids a reading, and what breaks a cell

A load cell on a wheelchair scale does not see the weight of a chair and a patient. It sees whatever the castors deliver at the transition, and the transition is an impact. The size of that impact depends on one thing the platform designer controls, which is how far the load falls, and one thing he does not, which is how long it takes to stop. I model it the simple way, because the simple way gets the decision right: the castor falls a height, arrives with a velocity, and the platform and the castor’s own rubber bring it to rest in a short time. The load that appears is the static load multiplied by one plus the deceleration expressed in gravities.

The first row uses a 220 kilogram gross weight, which is a powered chair of 113 kilograms with a 107 kilogram patient in it. The peak is 882 kilograms and each of the four cells sees 220.6 kilograms, which is 1.47 times its 150 kilogram rating but only 98 percent of the 225 kilograms its 150 percent safe overload allows. The cell survives a ten millimetre lip with about two percent of margin. The second row is the same chair crossing a twenty millimetre step that nobody chamfered: 289 kilograms per cell, which is 28.5 percent past the safe overload and still inside the 200 percent ultimate overload of 300 kilograms, which means the cell will not break and will probably not come back to its calibration either.

The third row is the one that changes machine designs, and it is not a machine event at all. A patient lowering himself onto a chair that is standing on the platform, over a drop of eighty millimetres and arrested in sixty milliseconds, produces 313 kilograms of equivalent load from a hundred kilogram body. Add the eighteen kilogram chair standing there and the instrument displays a peak of about 331 kilograms, which is 10.3 percent above a 300 kilogram maximum, and the reading at that instant is out of range and therefore invalid. But look at the fourth column: each cell sees 82.7 kilograms, which is 55 percent of its rating. Nothing is stressed. Nothing is close to being stressed.

I use these three rows as a design conversation, and it usually ends with the platform person revising the ramp rather than the buyer revising the cell. A chamfered transition of three or four millimetres instead of a step of twenty takes the worst realistic event from 289 kilograms per cell down below the 225 kilogram safe overload, and the fix costs a weld. Two other habits are worth writing into the drawing: a mechanical stop that prevents a castor from landing on a cell’s own mounting rib, and a handrail that is stiff enough that a patient does not use the platform edge as a grab handle on the way in. And keep a record of the first reading after any event of this kind, because the symptom of a cell that has been taken past its overload is rarely a dead channel; we publish a checklist for deciding whether a load cell is damaged and the first item on it is a change in zero return, not a loss of signal.

11. Two and a half microvolts, and where the twelve bits go

Everything so far has been about whether the cell is allowed to be there. This section is about whether your electronics can see it, and the answer contains one surprise that catches people who have only worked with single-point cells. Start with the signal. Four 150 kilogram planar beam cells, each rated at 1.5 millivolts per volt, excited at six volts, share a load of 500 kilograms. Each cell carries 125 kilograms, which is 83.3 percent of its range, so each produces 1.25 millivolts per volt, and the bridge output at full load of the machine is 7.5 millivolts.

One division of a 100 gram interval is 2.5 microvolts at the bridge, and separating 3000 divisions needs 11.55 bits of clean resolution, so the frequently quoted rule of thumb that a verified scale wants sixteen bits is right. What surprises people is the top row. A twelve-bit converter has one code per 1.83 microvolts, which is coarser than the 2.5 microvolt division it is supposed to be counting, so it resolves 1.37 divisions across the whole instrument. A 12-bit design is not marginal on this machine; it cannot display the interval it promises.

The single-point cell would have given you more signal

Here is the surprise. The single-point cell that the geometry in section eight forced us to abandon has a rated output of 2.0 millivolts per volt, and at six volts that is 12 millivolts across its own range. If a single cell could carry the platform, each 100 gram division would be 4.0 microvolts, which is sixty percent more signal per division than the four-cell set produces. The arrangement we are obliged to use is the one with the smaller output, and it is smaller because 1.5 millivolts per volt is what a planar beam is made at, not because anyone chose it. It is not a problem at sixteen bits. It is worth knowing before you specify a four-and-a-half digit display and wonder why the last digit moves.

Four bridges in parallel, and what that does to a battery

Planar beam cells also have higher bridge resistances than the single-point family, and on a mains-powered machine that is irrelevant. On a portable chair scale with a battery in the handrail it is the second largest decision in the design. Four 1090 ohm bridges in parallel present 272.5 ohms, and six volts across that is 22 milliamps and 132 milliwatts. Four 406 ohm bridges, which is what the single-point family has, present 101.5 ohms and draw 59 milliamps, which is 2.68 times as much. On the same 2000 milliamp-hour battery that is the difference between ninety hours and thirty-four.

The second consequence is less obvious and bites in the field. Four 1000 ohm output bridges in parallel present 250 ohms to the indicator, against 87.5 ohms for the 350 ohm family. That is three times the source impedance, so every ohm of cable, connector contact and junction box terminal contributes three times the offset error it would on the low-impedance set. On a machine where the summing box sits in the frame a metre from the cells, that is a real number and it is why I ask for the wiring drawing rather than assuming it. We publish a step-by-step guide to junction box wiring and adjustment that covers the terminals and the corner trim procedure, and the summation hardware I quote for this class of machine is the stainless multi-channel box in section fifteen.

12. The chair that sits on the platform for eight hours

A wheelchair scale takes thirty seconds to do its job, and the creep specification on the cell is quoted over five minutes or thirty. The temptation is to conclude that creep is irrelevant here, and for a single weighing it nearly is. Thirty seconds is one tenth of a five minute window, and the 0.02 percent of full scale that a cell is allowed to creep over five minutes is 30 grams on a 150 kilogram cell, which is 0.3 of a division and sits inside the 0.5 division allocation from section six.

The assumption that breaks is the one about how long the load stays on. In a day room or a physiotherapy suite the chair is not removed between patients; it is parked on the platform while the therapist does something else, and eight hours on the platform is ninety-six of the five minute windows the data sheet is describing. A specification expressed as a limit over five minutes does not tell you what the reading will be after eight hours, and I will not extrapolate it, because creep is not linear in time and the return-to-zero after the load comes off is the number the next patient’s reading depends on.

Two practical points from real machines. First, the creep windows differ between the families in section nine: the planar beam cells quote five minutes, the shear beam quotes fifteen, and the single-point cells quote thirty. That is not a quality ranking, it is what each family’s manufacturer chose to publish, and it means a five minute figure and a thirty minute figure are not comparable numbers even when they look identical. Second, a chair left on a cold platform and then brought into a warm room makes the platform itself move, and on a frame that is bolted at four points that movement appears in the reading. Neither of these is a reason to change the cell. Both are reasons to define what settled means in your own procedure, and to write it into the operating instructions rather than leaving it to the operator’s judgement.

13. What the clinic actually does with the number

The reason a wheelchair scale is a legal instrument is not that a regulator enjoys paperwork. It is that the number ends up in a decision, and the decision has a threshold on it. The clearest example is dialysis. A haemodialysis session removes fluid, the amount removed is prescribed, and the rate at which it is removed is expressed per kilogram of the patient’s body mass per hour. The clinical term is ultrafiltration rate, and it is computed as the volume removed divided by the product of dry body mass and treatment time in hours.

Take a sixty kilogram patient and three litres over four hours. That is 12.5 millilitres per kilogram per hour, and the published quality threshold that dialysis units work to sits at thirteen. It is not a soft threshold: cohort studies associate rates above it with roughly seventy percent higher cardiovascular mortality, and the practical rule for a single session is to remove no more than three to five percent of body mass, or about five kilograms, whichever is lower. Dry weight is then adjusted in steps of 0.2 to 0.3 kilograms between sessions, which on our machine is two to three divisions.

Read that block again, because it is the reason I am willing to spend a whole article on a platform that looks trivial. A verified instrument contributes an uncertainty of 6.67 percent to a three litre fluid shift. The clinical convention that surrounds it, the plus or minus half a kilogram that a ward works to, is 16.67 percent. The legal instrument is more accurate than the decision it feeds, and the decision is what the patient experiences. Buying a better cell than section six requires will not move the outcome. Weighing without the chair, on the same platform, in the same session as the previous reading, will.

I am not a clinician and I do not give clinical advice. What I can say is what falls out of my own arithmetic. The value of a wheelchair scale to a dialysis unit is not its resolution; it is that the reading is taken the same way every time, by staff who did not have to decide where the patient was standing. That is a design property, and it is bought in section seven and section fourteen, not in the accuracy class.

14. Three projects, and the two quotes I did not send

Case one: the Netherlands, and a platform that was already welded

This is the physiotherapy room from section one. The integrator had built everything and needed four cells. His requirement was 300 kilograms, 100 gram divisions, four cells, low profile, IP65. He had asked three suppliers and received three quotations for 4 × 100 kilogram cells, which is what three hundred divided by four rounds up to in most catalogues. He sent me the same request and I sent back a question about the frame instead of a price.

The frame weighed sixty-eight kilograms including the ramp that was bolted to it, and the platform was 1050 by 972 millimetres. Adding the maximum, the dead load, the initial zero range, the corner allowance and a ten kilogram distribution term gave 498 kilograms, and 498 divided by four is 124.5, so the cell is a 150 kilogram part and not a 100 kilogram one. He had the drawings in front of him while we spoke and he re-did the sum himself, which is the version of this conversation I prefer.

The interesting part came next. Low profile pointed at the planar beam cells, which are the only ones that fit in the sixteen millimetres he had left under the plate, and those carry a temperature coefficient of 0.05 percent of full scale per ten degrees on both output and zero. At 124.5 kilograms per cell that is 62 grams for ten degrees on the slope, and four cells adding the same way is 250 grams against a permitted error of 150 grams at the heavy end. He asked me to send the 0.02 percent cells from the single-point family anyway. I explained that the largest single-point cell we make is rated for a 600 millimetre platform, and his platform is over a metre, so those cells would have been correct on temperature and wrong on geometry by a factor of more than 1.6.

What he did about it is worth copying. He kept the planar beam cells and attacked the thermal path, because the temperature problem was not the cell’s coefficient but the frame’s thermal time constant. The channel frame and the plate were lagging the room by more than an hour, so a machine that had been in a cold corridor read differently from one that had been in the treatment room all morning. He put a temperature sensor on the frame, characterised the lag with the platform loaded, and compensated in firmware with a first-order model. That is his work and his file, not mine, and I told him so in writing. The instrument verified at class III with 3000 divisions and shipped.

One consequence came back six months later and it is the best illustration of section six I have. He wanted to launch a 500 kilogram bariatric version of the same machine and declare 100 gram divisions on it, because the two models would then share an indicator setting. That is 5000 intervals on a machine whose five-term sum is 800 kilograms, which needs 200 kilogram cells, and a class C3 cell of 200 kilograms has a minimum verification interval of 66.67 grams against a limit of 50. The 500 kilogram model cannot declare 100 grams in a four-cell class III arrangement. No cell exists that would let it. He declared 200 gram divisions on the bariatric model, which gives 2500 intervals and lets a 250 kilogram C3 cell pass with room to spare, and the two models now have different indicator settings and a note in the manual explaining why.

Case two: Germany, and the twenty gram resolution I would not homologate

A hospital group in Germany asked for a chair scale that would read to twenty grams so that the 0.2 to 0.3 kilogram dry weight adjustments in section thirteen could be seen as ten or fifteen counts on the display. The engineering is not the problem; a 300 kilogram platform with a 20 gram interval is 15000 divisions, and the classes a non-automatic weighing instrument may occupy put a ceiling on that number. Fifteen thousand is above the ceiling for class III, and the machine cannot be a legal instrument at that resolution.

I quoted two things instead of one. The first was a 300 kilogram, 100 gram verified instrument that satisfies section six and can carry an M. The second was a separate trend channel on the same platform: an unverified internal resolution of ten grams, clearly labelled in the manual and on the screen as a trend and not a measurement, with no legal claim attached to it and no verification of it. The medical physics department accepted that split and it is the honest one. The number a clinician uses to see a trend and the number a regulator verifies do not have to be the same number, provided the machine never confuses the two. What I would not do is put 20 grams on the nameplate of a verified instrument, and no arithmetic in this article would have made that legal.

Case three: a machine that had already shipped, and failed at the heavy end

A manufacturer in Turkey had built a 300 kilogram, 100 gram wheelchair scale and sent it for verification. It passed at the light end and failed at the heavy end, and the failure appeared as a non-linearity that grew with load. His cells were 200 kilogram class C3 parts bought from another supplier, and his first message to me asked for a matched set of the same capacity with tighter matching data, because he had concluded that the cells were inconsistent.

The matching data would not have helped, and the reason is one line of section six. A 200 kilogram class C3 cell has a minimum verification interval of 66.67 grams, and a 3000 interval instrument with 100 gram divisions and four cells imposes a limit of 50 grams. His machine was outside the condition before any cell was mounted, and no amount of matching changes a class. He had two ways back inside: a finer cell class, or a per-cell figure below 150 kilograms.

He chose the second, which is the cheaper one, and he found it by weighing his own frame. It came to 94 kilograms with the ramp, against the 70 I had assumed for a platform of that size, and taking forty kilograms out of the frame brought the five-term sum from 524 to 484 and the per-cell figure from 131 to 121. He re-declared with 150 kilogram cells, re-tested, and passed. The lesson he wrote back to me is the one I would put on a wall: the frame you design decides which cells you are allowed to use, and it is much easier to move the frame than to move an accuracy class.

15. Five load cells for wheelchair scales, five different jobs

These are the five parts I would put in front of a wheelchair scale builder, and each one carries the limitation that matters as well as the specification. Read the small print under each card before the specification list, because on this application the limitation is usually the deciding factor and the specification list is not. Note what none of them is: none of these is a medical device, none carries a medical certification, and none of them can put an M on your nameplate. What they carry is a millivolt output, a temperature coefficient, an overload figure and a bridge resistance, and those are the four numbers your design will actually argue about.

Whichever cell you pick, three components around it are mine as well and are worth quoting at the same time, because buying them separately is how a machine ends up with a summing box that has never been opened. The WST1513 stainless multi-channel junction box is the summation and corner-trim part for a four-cell platform, and it is the piece that makes the difference between a scale and four independent sensors. The WST1501 weighing indicator is the display and the zero logic that section seven is about. And if you would rather not have four analogue bridges at all, the WST1208 digital platform cell moves the conversion into the cell, which changes the cabling, the summing question and the noise budget all at once.

16. The twelve questions, and the habit that matters more

This is the list I work through on every wheelchair scale enquiry before I send a part number, in the order I ask it. Most of it is not about the cell, and that is the point of it. If this is your first weighing project rather than your tenth, the complete guide to load cells covers the vocabulary in sections five, six and eleven from first principles.

17. Questions I get asked about load cells for wheelchair scales