Why your electrolyser uses more kWh per kg than the datasheet says

There are five common causes, and they are worth checking in this order: the measurement boundary is not the same as the datasheet boundary; rectifier efficiency is lower than assumed at the load you actually run; auxiliary loads are consuming a larger share than expected because production is low; the operating point differs from the reference conditions; and the stack has genuinely degraded.

In practice the first three explain most reported gaps, and more than one is usually contributing at the same time. Degradation is the cause people reach for first and the one that least often accounts for a sudden change, because it is slow by nature.

This is the most common cause and the least interesting one. Specific energy can be quoted at three different places: direct current into the stack, alternating current into the electrolyser package, or alternating current at the plant boundary including all utilities. The numbers differ by several kilowatt hours per kilogram and all three are legitimate.

Before investigating anything else, establish which boundary the datasheet figure was quoted at and which boundary your meter sits at. If they are not the same, a large part of the gap is definitional rather than physical, and no amount of process investigation will close it.

Worth doing once and documenting, because this question recurs every time a new person looks at the number.

The transformer and rectifier that convert grid alternating current to stack direct current are efficient at their design point and less efficient away from it. Efficiency typically falls as load falls, and the fall accelerates towards the bottom of the range.

A plant following a solar profile spends a large share of its hours well below rated load. Every one of those hours carries a higher conversion loss than the commissioning test did, and the effect shows up entirely in the plant-boundary figure while the stack behaves exactly as before.

This is a frequent source of misattributed degradation claims, and it moves in both directions. It also means a plant-level efficiency number is not comparable between two sites unless their load profiles are comparable.

Electrolyte or water circulation, thermal management, gas separation and drying, instrument air, control power and nitrogen systems all consume energy. Some scale with production. Many do not.

A chiller holding a setpoint, a dryer on a fixed regeneration cycle and circulation pumps running at fixed speed consume close to the same power whether the plant is making hydrogen at full rate or at a quarter of it. Their contribution per kilogram therefore rises roughly in inverse proportion to output.

The practical consequence is that specific energy at 25 per cent load can look alarming for reasons that have nothing to do with the electrochemistry. Separating auxiliary consumption from stack consumption in the metering is what makes this visible rather than suspected.

Cell voltage depends on current density, temperature, pressure and electrolyte or membrane condition. A stack running cooler than its reference temperature has a higher cell voltage and therefore a higher specific energy at the same output, which is a normal response rather than a fault.

There is a useful and often surprising consequence. Lower current density generally means lower cell voltage and better stack-level efficiency per kilogram, while lower load means worse rectifier efficiency and a higher auxiliary share. The two effects run in opposite directions, so system specific energy against load is typically a shallow U rather than a straight line, with an optimum somewhere in the middle of the range.

Knowing where that optimum sits for your plant is worth more than knowing the rated figure, because it is the operating point you can actually choose. It also moves as the stack ages.

Once the first four are accounted for, what remains is stack condition. Ordinary ageing shows up as a rising cell voltage at a given current density and temperature, and it is slow: a trend measured over thousands of hours. Damage is different. Contamination, a membrane or separator breach, a seal failure or a thermal excursion can move the true condition in days.

That timescale is diagnostic, though not on its own. A number that has drifted steadily over a year, normalised for operating point, is ordinary ageing. A number that moved sharply is more often a boundary, load-profile or auxiliary effect, but the alternative is stack damage rather than stack ageing, so rule that out with cell voltage distribution, crossover and water quality before settling on a measurement explanation.

Attribution below the stack requires cell-level measurement, because stack voltage is the sum of hundreds of cells and a small number of degraded cells barely move that total. Whether the rise is uniform across the population or concentrated in a few cells points at different causes and different remedies.

Confirm the datasheet boundary and your metering boundary are the same, or quantify the difference. Pull the load duration curve for the period in question and compare it with the commissioning test conditions. Separate auxiliary consumption from stack consumption, at least approximately. Normalise the remaining stack figure for temperature, pressure and current density. Only then compare against the guaranteed degradation curve.

Most investigations that start at step five and work backwards end up repeating all of it. Most that start at step one finish early.

This is also the sequence that matters if the conversation is heading towards a guarantee claim, because a supplier will apply exactly these steps in exactly this order, and it is considerably better to have done it first. The same figures reappear later in reporting to lenders after commercial operation.