What start-stop cycling actually costs an electrolyser stack

A plant following a solar profile does not run at variable load. It runs a cycle: a start in the morning, a ramp up through the low-load region, several hours somewhere in the middle of the range, a ramp back down through the low-load region, and a stop. Then it does it again the next day.

That distinction matters because the wear that follows tracks the number of transitions rather than the hours accumulated. Two plants running the same average load, one continuously and one on a daily cycle, are not doing the same thing to their stacks, and a maintenance plan built on operating hours mistimes the second.

Wind coupling produces something less regular and no more steady. The transitions are less predictable, which makes them harder to schedule around, and the count over a year is often higher rather than lower.

When current stops, the cell does not immediately become inert. Each electrode sits at a potential determined by the residual gas at its surface rather than by the power supply, and that potential can be somewhere the materials would rather not be. Gas remaining in the compartments continues to permeate through the separator, and the composition on each side drifts.

What the plant does about that is the design decision that matters most. Purging with an inert gas, holding a small polarisation current, depressurising, or simply leaving it are all seen, and they produce materially different conditions inside the assembly overnight.

The practical consequence is that the stop procedure is frequently a bigger lever on cycling cost than the number of stops. A plant that stops well many times can outlast one that stops badly less often, and the procedure is usually adjustable where the resource profile is not.

In PEM systems the characteristic concerns are potential excursions at the electrodes during shutdown and restart, and mechanical fatigue in the membrane from repeated pressure and humidity swings. The assembly is bonded, so mechanical damage is not individually serviceable and presents as rising cell voltage or rising crossover.

In alkaline systems the liquid electrolyte introduces different problems. Circulation through shared manifolds means shunt currents persist while the stack is idle, and the electrode surfaces are exposed to a changing chemical environment as gas clears. Thermal mass also makes the restart slower and more energy-expensive, which is a cost that recurs nightly rather than a degradation mechanism.

AEM sits between the two architecturally and has the shortest operating record, so cycling behaviour is where its degradation assumption carries the most uncertainty. That is a reason to instrument it rather than a reason to avoid it, and the point is developed in alkaline against PEM against AEM under variable load.

Across all three, the low-load region is a second and separate stress. Crossover rises as load falls, and passing through that region twice a day means the plant spends real time near the condition that sets its minimum stable load.

Holding the stack at a low load overnight, or at a small holding current, avoids the transition but consumes energy and produces little or no useful hydrogen. Stopping avoids the energy cost and pays the transition cost. Which is cheaper is arithmetic, and the arithmetic is usually not done because one of the two numbers is missing.

The energy side is straightforward. Holding load multiplied by hours multiplied by the overnight power price, plus any auxiliary consumption that stays on either way.

The degradation side is the missing number, and it is the one to ask the supplier for explicitly: what does a cycle cost in guaranteed stack life, and how is a cycle defined. A supplier who cannot answer is telling you something useful about how well characterised their product is under this duty.

Once both numbers exist the decision usually turns out to be seasonal rather than fixed. Short summer nights at low power prices favour holding; long winter nights at high prices favour stopping.

A supply contract that limits cycles has to define one, and the definitions vary enough to change the annual number substantially. A stop to zero current is unambiguous. A ramp to twenty per cent and back may or may not count. A brief trip followed by an immediate restart may count as one, as two, or as none.

The definitions worth pinning down before signature: what current or load threshold constitutes a stop, whether time below that threshold matters, how a trip is treated differently from a planned stop, whether a partial ramp counts, and whether cycles are counted per stack or per system.

This is the same class of problem as the exclusions in a performance guarantee, and it is settled the same way: map each defined term to a measured signal before anyone needs it. The reasoning is set out in how a performance guarantee is measured.

Count cycles under the contractual definition, continuously, from commissioning. A count assembled retrospectively from a historian that compressed the transitions is not evidence.

Record the state during each stop: potential, pressure, purge status, temperature and how long the stack sat there. This is the variable most likely to be adjustable and least likely to be recorded.

Track cell voltage at a reference current density and temperature, normalised, so the degradation trend can be separated from the operating point. Cell-level distribution where available, because cycling damage is frequently local rather than uniform and the aggregate hides it.

And trend crossover at a fixed load, since the low-load passages are part of the duty and separator condition is what determines how much turndown remains as the plant ages.