What fails first on a hydrogen compressor under variable duty

Hydrogen compression on a green hydrogen plant is usually specified against a steady feed rate, because that is how compression is normally bought. What it gets is the output of an electrolyser following a solar or wind profile: ramping through the morning, varying through the day, ramping down and stopping.

The machine therefore spends much of its life away from its design point, starts and stops far more often than a process compressor would, and sees suction conditions that move with upstream production rather than sitting at a setpoint.

That gap between specified and actual duty is where the failure modes come from, and they are not the ones on the datasheet. Datasheets describe steady-state performance; the problems here are transitional.

Reciprocating machines dominate at the pressures and flows typical of a plant feeding storage or a tube trailer. They tolerate varying suction pressure reasonably well and manage flow turndown through unloaders or speed, but every start and stop cycles the valves, rings and packing, and those are the wear parts.

Diaphragm machines are used where gas purity has to be protected absolutely, since the process gas never contacts the hydraulic fluid. They are excellent at that and unforgiving about the things that break diaphragms, which include overpressure events and operation outside the intended stroke conditions. A diaphragm failure is usually abrupt rather than gradual.

Ionic and other liquid-piston designs offer good turndown and fewer conventional wear parts, at the cost of a more complex auxiliary system whose own reliability then matters.

Across all of them the pattern is the same: the machine handles the varying flow, and the wear concentrates in whatever component absorbs the transitions.

Hydrogen is a small molecule at high pressure, so sealing is inherently harder than it would be with a heavier gas, and it is the characteristic maintenance item on reciprocating hydrogen compression.

Packing wear is progressive and observable. Leakage past the packing shows up in vent or recovery system flow, in the temperature of the packing case, and in gradually falling volumetric efficiency. All three are trendable, and on many plants none of them is trended.

The consequence of ignoring it is not usually a failure. It is a slow loss of capacity and a rising leak rate, which on hydrogen is a safety consideration as well as an efficiency one because the escaping gas has to go somewhere the design accounted for.

Material compatibility sits underneath this. Components exposed to hydrogen at pressure have to be selected for it, and a substitution made during a repair from what was available rather than from what was specified is a failure mechanism introduced by maintenance rather than by operation.

Valves are the classic reciprocating compressor wear item and they fail from fatigue, from impact and from any liquid or particulate carryover. Cycling loads them more than steady running does, so a maintenance interval expressed in operating hours systematically mistimes their replacement on a plant that starts daily.

Rings and packing behave similarly. The transitions, where clearances and temperatures are moving, do more of the work than the steady hours in between.

The practical implication is to count starts and load changes alongside hours, and to set intervals against whichever correlates with what is actually being found at inspection. Most plants have the hours and not the starts, which is a recording decision rather than a measurement one.

This is the same shape of problem as start-stop cycling on the stack upstream, and the two are usually driven by the same daily profile, which means they can be addressed by the same operating change.

A machine that heats up and cools down every day moves. Thermal growth and contraction work on alignment, on piping strain at the connections, and on hold-down arrangements.

Piping strain is the one most often missed, because it is designed once and assumed to stay correct. Repeated cycling can move a support or fatigue a connection, and the resulting nozzle loading shows up as vibration or as an alignment change rather than as anything that names itself.

Vibration monitoring is the conventional technique and it works here unchanged, which makes this half of the plant the part where classical predictive maintenance applies directly. Baseline signatures taken when the machine is known-good are what make later comparison meaningful, and they cannot be recovered afterwards. The same reasoning applies to the pumps around the machine, where the readings that make a vibration number interpretable are already in the historian: pump condition monitoring from the signals a plant already has.

Suction and discharge pressure and temperature per stage, which give volumetric efficiency and stage-wise performance. Vibration at the conventional locations. Packing case temperature and vent or recovery flow, for sealing condition. Valve temperatures where the machine is instrumented for them. Auxiliary system health, including lubrication and cooling. And a start and load-change count.

The stage-wise view is what separates a valve problem from a ring problem from a cooling problem, because each moves a different combination of stage pressure ratio, discharge temperature and volumetric efficiency. A single overall capacity number cannot do that.

Almost all of this instrumentation belongs to the compressor package and is therefore behind an OEM control system, which makes the acquisition question the same one that arises everywhere else on the plant: read through an interface that already exists, without touching control logic. That is covered in reading plant data without modifying the control system.

The most effective fix for a compressor being cycled by a variable upstream is frequently not mechanical. Buffer storage between the electrolyser and the compressor decouples the two, letting the compressor run closer to a steady duty while the buffer absorbs the variability.

The sizing question is what fraction of the daily profile the buffer has to absorb to keep the compressor inside an acceptable operating band, and it trades capital cost against a reduction in starts and in time spent at part load.

It is worth evaluating explicitly against the mechanical alternatives, because the arithmetic often favours it and because it is a decision that has to be made at design when it is cheap rather than after two years of maintenance experience when it is not.

Where the compressor is already installed and cycling, the remaining levers are operational: minimum run times, coordinated ramping with the electrolyser, and accepting a slightly higher electrolyser minimum load to keep the compressor above its own. The same decoupling question returns further downstream, where an ammonia synthesis loop under a variable hydrogen supply has considerably less tolerance for the variation than the compressor does.