How an electrolyser performance guarantee is measured and defended

Electrolyser supply contracts usually guarantee some combination of four things: specific energy consumption, expressed in kWh per kg of hydrogen or kWh per normal cubic metre; hydrogen output rate at a stated load; hydrogen purity, usually against a fuel-quality standard such as ISO 14687:2025; and a degradation rate, commonly expressed as a voltage rise in microvolts per hour at a reference current density, or as a percentage loss per thousand operating hours.

Each of those is defined at reference conditions. A typical set fixes current density, stack temperature, operating pressure, feedwater quality and, for alkaline systems, electrolyte concentration. The guarantee is a promise about behaviour at that operating point, demonstrated during a performance test that is usually run once, at commissioning, under conditions chosen to be stable.

A plant coupled to solar or wind then spends most of its life somewhere else entirely. That gap between the tested point and the operated reality is where nearly every guarantee dispute originates. What any individual project guarantees, and who owns a shortfall, is set by the executed contract, and reading it is not optional. The measurement problem sits underneath whatever it says.

The owner observes that specific energy consumption at the plant boundary has risen. The supplier responds that the stack is performing within its guaranteed degradation rate, and that the additional consumption comes from how the plant is being run.

Both statements can be true at the same time. Plant-level kWh per kg is the sum of stack behaviour, balance-of-plant losses and the operating regime, and a guarantee usually covers only the first of those. Establishing which one moved is a measurement question, and it is answered before it is negotiated.

Normalisation is the mechanism by which a measurement taken at 40 per cent load and 68 degrees is compared against a guarantee written at rated load and 80 degrees. It is also the part of the contract that is most often underspecified.

Cell voltage depends on current density, temperature, pressure and electrolyte condition, and it does not depend on them linearly. Correcting a measurement across a wide gap in operating point introduces uncertainty that can be the same size as the degradation being argued about. A polarisation curve captured at commissioning, and repeated periodically under controlled conditions, is what makes that correction defensible rather than assumed.

Worth checking in the contract before it is signed: which parameters are corrected for, what correction method is specified, what measurement uncertainty is accepted, how many operating hours constitute a valid sample, and whether the test may be repeated on demand or only at fixed intervals.

Between the electricity meter at the plant boundary and the direct current entering the stack sit the rectifier, the electrolyte or water circulation pumps, thermal management, gas separation and drying, and any compression before storage. Every one of those consumes energy that appears in plant-level kWh per kg while having nothing to do with stack condition.

Rectifier efficiency is the most commonly missed term. It is not constant across load. Conversion efficiency usually peaks somewhere below rated load and falls away at deep part load, and where that peak sits is a design property rather than a rule. A plant following a solar profile spends a large share of its hours at partial load, which raises measured specific energy at the plant boundary even when stack behaviour is unchanged. Attributing that rise to the stack is a common and expensive error, in both directions, and it is among the most frequent reasons a plant reads more kWh per kg than its datasheet.

The separation is straightforward in principle. Measure direct current energy into the stack, measure alternating current energy at the boundary, and account for the difference by subsystem. It requires metering at the stack terminals that many plants do not have, which is why the argument is usually conducted on inference instead.

The supplier's usual position is that degradation has been accelerated by how the plant was operated. That position is often technically reasonable. Start-stop cycling, sustained operation below minimum recommended load, temperature excursions, pressure transients and feedwater quality departures all have documented effects on stack condition, and most supply contracts carve them out of the guarantee.

Answering it requires an operating history, not a snapshot. Cumulative hours in each load band, number and depth of start-stop cycles, time spent outside the recommended temperature envelope, and feedwater conductivity history together describe whether the plant stayed inside the envelope it was sold against.

This history is also worth having for its own sake. If cycling is genuinely driving degradation, that is an operational decision with a measurable cost, and it can be traded against curtailment or grid import on commercial terms rather than argued about after the fact.

Stack voltage is the sum of hundreds of cells in series. A single cell drifting by 100 millivolts moves a total of several hundred volts by well under a tenth of a per cent, so aggregate voltage is a poor instrument for the question being asked.

The distribution across cells is far more informative than the mean. Uniform drift across the whole stack is consistent with expected ageing. A small number of cells departing from the population, under identical current and temperature, points at something local: contamination, a membrane problem, a flow maldistribution. Those have different causes, different remedies and, under most contracts, different owners.

That distribution is also the most direct evidence available that a stack has or has not degraded as promised, because it is measured on the stack itself rather than inferred from a plant-level energy balance with several other terms in it.

A guarantee argument is retrospective. It asks what has happened over the last several thousand operating hours, and no instrument installed today can answer that.

The practical minimum is a commissioning baseline that is genuinely usable: a polarisation curve at defined conditions, a documented reference performance test, and metering at the stack terminals as well as the plant boundary. On an operating plant that record usually has to be built without modifying the control system. Beyond that, continuous logging of load, temperature, pressure, feedwater quality and, where available, cell voltage distribution, retained for the life of the guarantee rather than the retention period of the historian.

Yunify is designed to hold that record and to attribute movement in specific energy to a mechanism rather than to a suspicion. Yunify CVM adds cell-level measurement, which few plants have, and the physics models estimate the terms that no sensor reports directly. The commercial value of that is not the analytics. It is being able to answer a supplier's technical position with a technical answer.