Diagnosing membrane thinning in a PEM electrolyser from operating data

The membrane in a PEM cell is a thin proton-conducting polymer separating the two electrodes. It conducts ions, blocks electrons, and keeps the product gases apart. Thinning is the loss of material from that barrier over time, through chemical attack and through mechanical damage, and it is one of the mechanisms that ends a stack's life.

It matters operationally for two reasons that are easy to confuse. A thinner membrane conducts protons slightly better, which nudges cell voltage down. It also lets more hydrogen through, which pushes crossover up. Those move in opposite directions, and a plant watching only efficiency can miss the mechanism entirely.

Membrane degradation chemistry is well characterised at materials level. From the control room the question is narrower: which measurable quantities move, in what order, and how thinning is separated from the other things that produce the same symptoms.

Chemical attack is driven by reactive species formed as an unintended product of the electrochemistry, particularly where hydrogen and oxygen meet at a catalyst surface. That gives it an unpleasant feedback property: crossover produces the conditions that increase crossover.

Metal contamination accelerates it. Ions from feedwater, from corroding balance-of-plant components or from the stack hardware itself can catalyse the attack, which is why feedwater quality specifications for PEM are strict and why a departure from them is expensive rather than merely inefficient.

Mechanical damage is a different route. The membrane swells and contracts with hydration and is loaded by differential pressure, so repeated humidity and pressure cycling works it in a way steady operation does not. A plant that starts and stops daily is doing more of this than one that runs continuously, which is why the two mechanisms have different duty drivers and different mitigations.

Crossover trend at a fixed load usually moves first. Hydrogen content in the oxygen stream, compared at a comparable operating point over weeks, is the most direct available observation of barrier condition, and it responds before the electrical signature does.

Cell voltage moves second and ambiguously. A thinner membrane has lower ohmic resistance, so the voltage contribution from that term falls, while any loss of catalyst activity pushes it up. The net can be flat while both are changing, which is why a single voltage reading is a weak instrument here.

The order is itself diagnostic. Crossover rising while cell voltage holds or falls slightly points towards the barrier. Cell voltage rising while crossover holds points elsewhere, usually towards catalyst or transport. Both rising together suggests more than one mechanism, which is common late in life.

Turndown is the operational symptom that arrives before either is alarming. As baseline crossover rises, the load at which the hydrogen fraction in the oxygen stream reaches its safety margin moves upward, and the plant quietly loses low-load hours. That shows up in the production record before it shows up in a condition report.

Contamination produces rising cell voltage and can produce apparent performance loss without touching the membrane. Feedwater conductivity history and the timing of the change usually separate the two: contamination effects can arrive over hours to days, and thinning does not. Conductivity does not capture every contaminant, so it narrows the hypothesis rather than confirming it.

Flow maldistribution produces a temperature spread across the assembly and a cell voltage pattern that follows the hydraulic layout rather than the age of the cells. If the outlying cells are grouped by position rather than scattered, the problem is more likely hydraulic than material.

Instrument drift produces the most convincing false trend of all. A gas analyser reading progressively high looks exactly like a rising crossover. Calibration history and cross-checks against a second measurement come before conclusions, and this is one of the things a physics layer is good at, because a set of readings can be tested for mutual consistency.

Catalyst loss and membrane thinning can coexist and frequently do late in life, so the useful output is a ranked pair with the measurement that would separate them rather than a single confident answer.

A stack is hundreds of cells in series reporting one aggregate voltage. A single cell drifting by a hundred millivolts changes that aggregate by a fraction of a per cent, which is well inside normal variation, so the aggregate is close to blind to exactly the pattern that matters.

The distribution is the useful measurement. Uniform drift across the population is consistent with expected ageing. A small number of cells departing from their neighbours under identical current and temperature points at something local, and local problems have different causes, different remedies and, under most supply contracts, different owners.

For thinning specifically, the distribution answers a commercial question. A few outliers late in life point at a local cause worth investigating before the population is written off. A population that has moved together is a stack approaching end of life, and that is a reserve and replacement conversation rather than a maintenance one.

Feedwater quality is the lever with the best return, because contamination accelerates the chemical route and is upstream, visible and controllable. Conductivity monitoring with a defined response to a departure is cheap relative to what it protects.

Reducing unnecessary pressure and humidity cycling helps the mechanical route. Where the resource profile is fixed, the adjustable part is usually the shutdown procedure and the treatment of brief trips, not the number of daily starts.

Operating point matters at the margin. Sustained operation deep in the low-load region spends time where crossover is highest, and where turndown has narrowed, the honest response is to raise the minimum operating load rather than to run closer to the safety margin.

None of this reverses thinning. The bonded assembly is not individually serviceable, so the practical question is how much operating margin remains and when replacement should be planned, which is a remaining useful life question with a commercial answer.