Green Hydrogen Electrolyser Ultrapure Feedwater: RO, EDI and Polishing Design

Electrolysers are unforgiving about water. A PEM stack fed with water carrying a few hundred µg/L of chloride will see accelerated membrane degradation and catalyst poisoning; an alkaline unit fed with hard water will scale its electrodes and diaphragms. The specification is simple to write — conductivity below 1 µS/cm, TOC below 50 µg/L, chloride and sulphate in the low µg/L range — and surprisingly easy to miss in operation, because the failures show up months later as stack efficiency loss rather than as an alarm the day it happens.

What the Stack Actually Needs

Most PEM manufacturers specify ASTM Type II or better, often effectively Type I: resistivity at or above 1 MΩ·cm at the point of use, with tighter limits on specific ions than the standard resistivity figure implies. Alkaline electrolysers are more tolerant of conductivity and far less tolerant of hardness and carbonate, because they concentrate KOH in the cell and precipitate calcium where the concentration is highest.

The parameter that surprises people is TOC. Resistivity says nothing about uncharged organics, and a resin bed exhausted of organics will happily deliver 18.2 MΩ·cm water with 300 µg/L of TOC in it. That organic load oxidises at the anode. It also fouls the catalyst layer. Much of this mirrors the situation set out in Emulsified Oil, pH and Solids.

Silica matters for the same reason and is harder to monitor. Reactive silica passes through RO reasonably well and is only weakly ionised, so it neither shows on conductivity nor is removed by a standard polishing resin as effectively as chloride. A plant drawing on a groundwater source with 20 to 40 mg/L silica needs to address it explicitly upstream.

Pretreatment: Where Most Systems Are Under-Designed

The polishing train gets the attention and the pretreatment decides whether it survives. A two-pass RO is the standard core: first pass at 70 to 80 percent recovery, permeate to a second pass at 85 to 90 percent, with the second-pass concentrate recycled to first-pass feed because it is cleaner than the raw water.

Ahead of that, the feed needs suspended solids removed and the scaling ions managed. Multimedia filtration or ultrafiltration for turbidity, antiscalant dosing and — where hardness is above about 150 mg/L as CaCO3 — softening. Softening ahead of RO is one of those investments that looks unnecessary on the capital line and pays back within two years in membrane life and cleaning frequency.

Chlorine is the other killer. Polyamide membranes tolerate perhaps 0.1 mg/L free chlorine continuously and effectively zero over months. Where the municipal supply is chloraminated, which is increasingly common, activated carbon alone is not sufficient and a dedicated catalytic carbon or a bisulfite dosing stage with proper contact time is needed. We have seen stacks damaged by chloramine break-through on plants that had carbon and assumed it was enough.

Electrodeionisation has become the default polishing step for continuous plants, and for good reason. It produces 10 to 18 MΩ·cm water continuously, needs no acid and caustic regeneration, and generates no hazardous regeneration waste. For a hydrogen plant running 24/7 with a small operating team, that is a compelling case.

EDI does have a feed requirement: it wants RO permeate with conductivity below about 40 µS/cm, hardness below 1 mg/L, and silica and CO2 managed. Feed hardness above specification precipitates inside the dilute compartments and the damage is permanent. A well-designed plant therefore puts a small mixed bed guard polisher downstream of the EDI, sized to catch transients rather than to do the work.

Conventional mixed bed polishing is still the right answer where the plant operates intermittently, where feed quality is variable, or where the specification includes boron or silica at very low levels. It costs more in chemicals and operator attention. It is also more forgiving of everything else. The process logic sits close to what is described in Sugar, Dairy and High-Strength Organics, particularly on the pretreatment side.

Distribution Loop and the Point-of-Use Problem

Water that leaves the polisher at 18.2 MΩ·cm does not arrive at the stack that way unless the loop is designed for it. Carbon dioxide from any tank vent dissolves and drops resistivity within minutes — atmospheric CO2 alone will take ultrapure water to about 1 MΩ·cm. So the loop runs at a continuous velocity above 1 m/s, uses PVDF or electro-polished 316L, avoids dead legs entirely, and is UV-conditioned at 185 nm for TOC control.

Removal Rates Across Our Installed Base

Across 42 commissioned plants where we hold complete COD records, influent COD ranged from 200 to 172,000 mg/L and treated effluent from 50 to 5,000 mg/L. Average removal across that set is 87.1% — not a marketing figure, but the measured mean.

That average is worth pausing on. It sits well below the 95%-plus numbers most suppliers quote, because the set includes genuinely difficult streams. On some electroplating and municipal duties the installed configuration only reaches about half the influent COD, and saying that up front is more use to you than a number the plant will never hold.

Storage, where it exists, is nitrogen-blanketed or has a vent scrubber. Many plants dispense with a tank altogether and run the polisher directly into a recirculating loop, which removes an entire class of problems. The same selection criteria reappear in Sugar Refinery and Starch Processing Wastewater Treatment system, which is worth opening alongside this one.

Monitoring belongs at the point of use, not at the skid. Resistivity, TOC and — where the specification is tight — online chloride or sodium as the sentinel for ion breakthrough. A sodium analyser is the cheapest early warning available: it is the first ion to slip when a cation resin exhausts, and it gives hours of warning rather than days. Plants dealing with comparable loads face many of the same trade-offs covered in POME Polishing, Pond Upgrades and Biogas Capture.

Hydrogen plants are often designed for a future nameplate capacity that will not be reached for two years, and the water system gets sized on the same optimistic basis. An RO running at 20 percent of design flow for a year will foul; an EDI at very low flow loses its ability to regenerate resin electrically and its product quality degrades.

Either size the water system for the actual first-phase duty and add skids later, or build in a recirculation and periodic full-flow flush that keeps the equipment at a healthy operating point. The second option is cheaper in capital and worse in everything else.

Water consumption itself is worth checking. Stoichiometrically, a kilogram of hydrogen needs about nine litres of water, but the real figure including RO reject, EDI reject, backwash and CIP is more like 20 to 30 litres. In a water-stressed region that number can decide the site. Recovering reject to cooling make-up or to a second-stage RO is usually worth doing, and it is much easier to include at design stage than to retrofit.

Integrated Treatment Strategies

Where several streams need joint treatment, Coffee, Tea and Cocoa Processing Wastewater Treatment system is the more detailed reference. Most facilities do not run a single clean stream, and a shared equalisation and biological stage is usually the economical answer once the streams are chemically compatible.

Why Choose Baihuipu as Your Manufacturer

Picking a supplier for this kind of plant is mostly about who carries the risk when something goes wrong. We have built this equipment for more than twenty years, and we are the ones who fabricated it, programmed the controls and will still answer the phone three years from now.

In-House Fabrication

Our Guangdong works produces everything from standard modular units to fully bespoke lines. Building in-house keeps cost, schedule and quality under our own control rather than a subcontractor's goodwill.

20+ Years Across Dozens of Industries

Experience matters most when your waste stream behaves nothing like the textbook example. We have dealt with toxic organics, heavy metals, extreme salinity and refractory COD in food processing, chemical manufacturing, electroplating, mining and municipal projects.

We Engineer the Whole Train

Screening, equalisation, biological or chemical treatment, membranes, evaporation, brine handling — the complete line comes from one organisation. Process design, mechanical, electrical and PLC programming sit under the same roof, so there is no gap for finger-pointing at handover.

Spares, Service and Commissioning

Membranes, dosing pumps, diffusers, instrumentation and blowers are stocked for the systems we sell. Our control systems support remote diagnostics, and we can put an engineer on site for commissioning, training or an unplanned shutdown.

Let's Look at Your Numbers

Share your water analysis and site constraints and we will give you a straight assessment of the process route and the realistic operating cost before you commit to anything. Contact our engineers at WhatsApp: +86 13631765076, or find us at hkbhp.com.

WhatsApp: +86 13631765076

Frequently Asked Questions

What does it cost to run, per cubic metre?

Energy dominates, then chemicals, then sludge disposal. We give you a per-cubic-metre breakdown during proposal so you can compare it against your sewer charge — treating too aggressively is a common and expensive mistake.

What information do you need to quote?

Water analysis for your worst-case stream, peak and average flows, discharge limits, available footprint and utilities. If you do not have a full analysis yet, we can start from a partial one and flag the gaps.

How long does a electrolyser ultrapure feedwater system take to deliver?

Standard units ship in about 11 to 11 weeks. Custom trains with pretreatment: where most systems are under-designed included run longer — plan for 16 to 25 weeks once detailed engineering starts. Site installation and wet commissioning add another few weeks.