He Knew How Many Membranes He Had Bought. Not What the Four Tanks in Front of Them Were For.

A water refilling station owner called us about his membranes. He knew how many he had, he knew the system was sold to him as a three-stage unit, and he knew roughly what the whole thing had cost. What he could not tell me was what the four vessels standing in front of those membranes were for. He knew there were four. He had been backwashing two of them, when he remembered, because the man who installed it had shown him the lever.

That conversation is the most common one we have with refilling station owners, and it is the reason I started a note on it in 2020. I wrote two paragraphs, abandoned it in the middle of a sentence, and it has been sitting unpublished on our own site ever since. I want to finish it properly now, because we published a note recently on when reverse osmosis is actually the right process and it ended by saying that pre-treatment ahead of a membrane is a subject in its own right rather than an accessory. This is that subject. It is deliberately the feed side only. What the membrane itself does, what the reject stream is, and why most buildings that ask us for reverse osmosis do not need it are all in that other note and I will not repeat them here. Nor will I re-argue how a softener gets sized twice, or how we put a water treatment quotation together. This note is about the equipment between the source and the membrane, what each piece is protecting, and the numbers that decide how big it has to be.

What he knew, and what nobody had told him

First-time owners in this business almost always know the same three facts: the number of membranes, the number of stages, and the price. Those are the three things a salesman leads with, so those are the three things that stick. The more knowledgeable ones know what their pre-treatment is, and that is the dividing line between a system that quietly lasts and one that keeps eating membranes.

It is not ignorance and it is not carelessness. The pre-treatment is the part of the plant that has no story attached to it. The membrane is the magic; the four tanks in front of it look like plumbing. Nothing about them changes visibly when they stop working, which is the whole problem and the reason I keep coming back to it. We have made the same observation about sewage plants, where a good operator can carry a plant but cannot carry a bad design, and about the equipment choices that decide how much trouble a plant gives you later. A refilling station is a smaller version of the same arrangement, run by somebody who is also doing the accounting and the deliveries.

We keep our own process description for a refilling station on the site alongside the method of treatment we propose, both written years ago, and the description still lists the train in order: feed tank, feed booster pump, multimedia filter, activated carbon filter, softener, a five micron cartridge, then the membranes, then the product tank, the product booster pump, a polishing carbon, ultraviolet light, and the dispensing bay. Eleven items before and after the membrane, and three of them in front of it are doing work nobody can see.

Three sources, three different plants

The first thing that decides what the pre-treatment has to be is not the production target. It is where the water comes from, and our own description names the three honestly. A deep well is usually high in dissolved minerals and hardness. Surface water from a spring, a river or a shallow well is high in turbidity and suspended solids. A local utility connection already meets the national drinking water standard before you touch it, and it is worth being clear about what a drinking water barrier stack actually has to include.

Those are three different plants wearing the same cabinet, and they draw on different parts of the range of systems we build. On a deep well the softener is the hardest working vessel in the room and the filtration is comparatively easy. On surface water the filtration is the hard part, it varies with the weather, and the quantity of solids arriving in October is not the quantity arriving in April. On a utility connection the main job in front of the membrane is removing the chlorine the utility deliberately put in, plus a polish for whatever the storage tanks and the pipes between the plant and your building have added. One source, one word on a business permit application, and three different sets of vessels.

What follows from that is the reason we ask for a water analysis before quoting rather than after. The number of membranes is the easy decision. The pre-treatment is the decision that needs to know what is in the water, and on the sewage treatment side of our work we have written at length about how rarely anybody has the influent result we ask for. On the drinking water side the position is better, because a station needs a water analysis for its permits anyway, so the document usually exists. It is just rarely read by the person buying the equipment, and almost never read by the person selling it.

The number on the brochure is the smallest of the three flows

This is the part of the conversation where owners go quiet, and it is the only arithmetic in this note. It belongs to a family of mistakes we keep running into, where a number that is real, printed and authoritative turns out to be measuring something other than what the reader assumes.

Our own package comparison rates systems in gallons of product per day: roughly three thousand, six thousand or nine thousand, which the same page translates into about twenty-two, thirty-eight and fifty-eight five-gallon containers an hour. Those are product numbers. They are the only numbers most owners ever see.

Now take the recovery. Our own written process description for one of these systems puts the estimated conversion factor at 0.35, meaning roughly a third of the water entering the membranes leaves as product and the rest goes to drain. That figure is specific to that system, that feed water and that configuration, and I would not transfer it to yours without reading your own proposal, for exactly the reasons set out in the reverse osmosis note. But take it as the worked example it is.

Three thousand gallons a day of product at a conversion of 0.35 means about eight thousand six hundred gallons a day of feed. In metric that is roughly eleven cubic metres a day of product drawn out of roughly thirty-two cubic metres a day of supply, with about twenty-one cubic metres going to drain. Spread over a day of continuous running, the product side is passing something close to two gallons a minute and the feed side is passing six.

So on a three-thousand-gallon plant the multimedia filter, the carbon filter and the softener are six-gallon-a-minute duties, not two-gallon-a-minute duties. They sit on the feed side of the membrane, so they have to pass the feed flow, which on this worked example is close to three times the number printed on the brochure. Size them on the product number and every vessel in the train is undersized by the same factor, which shows up as channelling in the media, as a softener that has the right capacity but cannot deliver it at the flow it is being asked for, and as a carbon bed with not enough contact time to do what carbon does.

There is a third flow number in the same plant, and it is derived from neither of the first two. The ultraviolet units in those packages are rated at six, eight and twelve gallons a minute, and they sit after the product tank and the product booster pump. They are not sized on production at all. They are sized on dispensing, because dispensing happens in bursts at whatever the faucets can draw while a customer waits, and that has nothing to do with how fast the membranes fill the tank overnight. One plant, three flows: production, feed and dispensing. Three different numbers, each correct for its own vessel, and only one of the three printed on anything the owner is given. When somebody tells me their filters are the right size because the system is a three thousand gallon system, that is usually the mistake underneath it.

Each vessel in front of the membrane is protecting it from something different

This is the part worth writing down, because once it is in this order it is hard to forget. Taking them in the sequence the water meets them:

1. The multimedia filter takes out what you could see if you looked. Sand, silt, rust flakes, whatever the storage tank contributes. Our own description has it holding suspended matter down to around ten to twenty microns with a conventional sand, manganese and anthracite bed, and notes plainly that what gets through causes a premature drop in membrane performance. The glass-based medium we supply instead of sand filters finer than that, which is the reason we changed, and we keep a separate note on filtration ahead of a membrane and another on how the media is layered inside a standard pressure vessel. The same filtration logic turns up at municipal scale in rapid gravity filters, which is a larger version of the same argument about what a bed can actually hold.
2. The carbon filter takes out what you could taste, and the chlorine that would destroy the membrane. Two jobs, and only the first one is visible to a customer.
3. The softener takes out the hardness that would scale the membrane from the inside. Our own description puts it exactly that way: the softener is there to avoid premature clogging of the membranes. The soft water is not a product feature on a refilling station. It is membrane protection. Hardness also does things nobody connects to hardness, such as turning kitchen fat into soap in a grease trap, which is why a softener turns up in commercial kitchens for reasons that have nothing to do with drinking. The question of whether soft water will attack the pipework comes up often enough, and the answer is usually in the pipe schedule.
4. The five micron cartridge catches what the first three missed. It is called a safeguard on our own page and that is the right word. It is not a treatment step.

Notice that three of the four are protecting one component, and the component they are protecting is the only one the owner was sold. That is not an argument for buying more pre-treatment. It is an argument for knowing which of your four tanks is carrying your membrane warranty on its back, which on a deep well is the softener and on surface water is the multimedia filter. If you have hardness and iron together, the sequence matters as well as the sizing, and iron in particular is worth understanding properly because fed to a membrane untreated it behaves as a fouling agent rather than as something the membrane removes for you. We keep notes on iron and on arsenic, ferric iron and manganese for that reason.

The carbon tank fails quietly, and the membrane is what tells you

Of the four, the carbon is the one I would watch. A sand bed that has lost its grading usually announces itself: the cartridge downstream starts loading quickly. A softener that has stopped regenerating announces itself on a hardness test, and hardness tests are cheap. Carbon does neither. It exhausts gradually, it looks exactly the same when it is finished as when it was new, and the water coming out of it tastes fine to a person who has been drinking it for three years.

What tells you is the membrane, several months later, and by then the damage is done rather than developing. The reverse osmosis note made the same point about fouling in general: a membrane fouls slowly and the first evidence is usually a pressure reading nobody was recording. On a refilling station there are two cheap habits that close that gap. The first is writing down the feed pressure and the product conductivity at the same time every day, which takes under a minute and is the same discipline as the daily log we keep arguing for on sewage plants. The second is keeping the dates of the last media change where somebody other than the owner can find them, because the person who knows how the plant runs is usually the person who leaves.

Neither of those is equipment. Both of them are the difference between replacing carbon on a schedule and replacing membranes on a surprise.

The cartridge is a consumable, and it is also an instrument

The five micron cartridge is the cheapest item in the plant and the most informative. Nobody treats it as information. They treat it as a thing you change when the pressure drops.

How fast it loads is a reading on everything upstream of it. A cartridge that runs for weeks is telling you the multimedia filter is doing its job. A cartridge that is brown and tight in four days is telling you the filter upstream has channelled, or lost media, or is being asked to pass more than it can at the real feed flow, which brings us back to the arithmetic above. Changing it faster is treating the symptom, and it is a cheap enough symptom that most owners treat it for years without ever asking what it is a symptom of.

I like it as an example because it costs nothing to use this way. There is a version of the same habit on every plant we work on, and the thing they have in common is that the information was already being generated and nobody was reading it. We found the design flow for a sewage plant sitting in eleven years of water bills for the same reason.

Backwash is water you buy, and water you then have to put somewhere

Two things about backwash that do not appear on a specification sheet.

The first is that a backwash needs a higher flow than service, because the point of it is to lift and tumble the bed rather than to pass water through it. So the supply to the plant has to be able to deliver a rate it only needs for a few minutes a day. On a plant fed from a utility line that is sometimes the binding constraint on the whole station, and it is one of the reasons our own process description puts a feed tank and a booster pump in front of everything: the tank is there precisely because source pressure cannot be relied on. It is doing the same job as the one tank on a sewage plant that treats nothing, which is to decouple what arrives from what the next stage has to pass. An undersized supply does not stop you producing water. It stops you cleaning the filters properly, which is a slower and more expensive failure.

The second is where that water goes. Backwash from a multimedia filter is dirty water in a modest volume, and the softener regeneration is a salty one. Neither is a difficult disposal problem for a small station, but both are a disposal problem somebody has to have thought about, in the same sense that on a sewage plant everything you take out of the plant has to go somewhere. Add the membrane reject on top and the volume leaving the building to drain is larger than the volume being sold, which has a bearing on what the station declares as its discharge. We have written about how a declared flow becomes a ceiling you are held to, and about the compliance position more generally, and a refilling station is not exempt from arithmetic simply because its product is clean.

Whether the backwash is manual, semi-automatic or fully automatic is the main mechanical difference between the packages we offer, and across the stations we have put in it is worth deciding on honestly rather than on price. A manual backwash is not worse engineering. It is a task that depends on a person remembering, and whether that is acceptable depends entirely on who that person is going to be in three years.

What we ask before quoting one of these

These are the questions behind every consultation we take on a station, in the order we ask them, because the answers change what comes after:

1. Where is the water coming from, and is there an analysis. Deep well, surface water or utility, and the laboratory result if one exists. Everything downstream is decided here. If there is no analysis, we say so in the quotation rather than assuming a convenient feed, in the same way we show a sewage plant design at several wastewater strengths side by side instead of picking one and hoping.
2. How many containers a day do you actually expect to sell, and over how many hours. Not the system rating. The business number, which is the one that sets the product tank and the dispensing side.
3. What can the supply deliver, in flow and not just in volume. Including during a backwash. This is the question that most often changes the layout.
4. Where does the drain go. Backwash, regeneration and membrane reject, all of it, and what the building is permitted to discharge.
5. Who is going to operate it, and what else is that person doing. This decides manual against automatic, and it decides how much of the plant can be allowed to depend on somebody noticing something. It is the same question that drives how we set up maintenance and operation on the sewage side.

None of those five is about membranes. That is the point of asking them in that order. The membrane decision is the easiest one in the room once the other five are answered, and the hardest one to get right if they are not. It is the same reason we argue that on a sewage plant the design decides the cost long before the build does, and why we would rather read a drawing somebody already has than price it blind. If you have a proposal in front of you from anybody, including us, the useful exercise is to go through it looking for those five answers. We did exactly that with two sewage plant quotations read line by line, and there are a handful of questions worth asking before signing anything of this kind. We also have no objection at all to you checking the whole thing with an AI tool before you sign it.

Where this stops being transferable

The order of the train travels well. Filtration, carbon, softening, cartridge, membrane is the sequence on almost every station in the country and the reasoning behind each step holds wherever the water comes from.

The sizing does not travel at all. The conversion of 0.35 in the worked example above is one documented case on one system and one feed water. Better pre-treatment and a better feed permit a higher recovery, which is the part of this argument that pays for itself: every point of recovery you gain is water you do not buy twice, and that is the honest commercial case for spending money in front of the membrane rather than behind it. It is the drinking water version of an argument we make constantly about sewage plants, where the figure that decides ownership is the running cost per cubic metre rather than the purchase price. But it means the feed flow on your plant is a number you have to get from your own proposal, not from this note.

Source variability does not travel either. A plant on a stable utility connection can be run by somebody following a schedule. A plant on surface water that turns brown after heavy rain needs somebody who can tell that it has, and a filtration stage with enough margin to absorb it, and that is a different plant even if the equipment list reads the same.

And the whole of this assumes a commercial station selling water as its product. A household unit under a sink is not a small version of this; the economics, the duty and the consequences of a fouled membrane are all different, and nothing above should be scaled down to fit one. The general rule we work to is to treat to the level the end use actually needs and no further. Equally, a station running a specialised source such as a brackish or saline supply is a different engineering problem again, and the rules of thumb here are too small to carry it.

If you are running a station now and are not sure which of your tanks is doing what, the quickest useful thing is to look at how fast your cartridge loads and when your carbon was last changed. We keep a reference page for refilling station owners covering the business side as well as the equipment, and there is more on our water treatment work and on our wider project list. If you would rather talk it through with somebody who will look at your analysis first, book a free 15-minute consultation.