Power fade

What it is (precise)

Power fade is the decline, with age, in the power a battery can deliver or accept at matched conditions — same state of charge, same temperature, same direction. That last clause carries the definition: a cold snap also takes megawatts off the plant, but the weather gives them back, whereas power fade is measured by holding the conditions still and watching the capability at those conditions shrink from its beginning-of-life value.

Degradation runs on two measurable axes — stored energy goes down and internal resistance goes up — and power fade is the second axis expressed at the terminals: what the rising resistance does to the MW number. It is why State of Health carries power- and resistance-based definitions alongside the capacity-based one, and why a degradation review that tracks only retained MWh is reading one axis of a two-axis process.

The boundaries are worth a sentence each. Resistance growth owns the mechanism — SEI thickening, electrolyte depletion, interface aging — and that page covers the physics; power fade is the consequence those mechanisms produce at the system boundary.

Derating is the neighbouring concept on the other side: the capability a healthy machine gives up when it runs off its rating's reference conditions, and recovers when the conditions do — power fade is the capability that no longer exists at the reference conditions themselves. And it is not the converter: the PCS's MVA ceiling is set by silicon and cooling, not electrochemistry, so power fade is the DC side sliding down beneath a fixed AC nameplate.

Why it shows first at the corners

Deliverable power is a headroom calculation. Discharge current is capped where the loaded cell voltage — open-circuit voltage minus the I × R sag — reaches the protection floor, so the sustainable current scales with voltage headroom divided by resistance. At mid state of charge an LFP cell sits on its flat 3.2-3.3 V plateau with the full margin down to the 2.5 V floor, and rated power fits with room to spare.

Near empty, the open-circuit voltage drops off the edge of the plateau while resistance climbs — most sharply near empty of anywhere in the window — so the numerator shrinks and the denominator grows at once. Charging mirrors it against the 3.65 V ceiling: acceptance collapses near full, which is why charge power derates toward the top of the window even at beginning of life. Aging raises resistance across that whole surface, pulling every limit inward — and the corners cross below rated power first because they started closest to it.

Cold compounds the geometry rather than adding to it. Electrolyte ionic conductivity falls by roughly an order of magnitude between +25 C and -20 C, and because the same resistance term sits in the sag, the heat and the current cap, a cold, aged, near-empty rack is the worst case on every axis simultaneously.

The fleet dimension arrives through series-string mechanics: under load the weakest rack reaches its voltage limit first and ends a rated-power discharge early, so the effective full-power window is set by the limiting rack, not the fleet average — and fleets age unevenly. The practical result is a full-power envelope narrower than the energy envelope: in late life a plant can often still sweep, gently, MWh it can no longer sweep at rated MW.

Why energy-duration sizing hides it

A four-hour system discharges at 0.25C, and at beginning of life, mid-window, at moderate temperature, the cells behind it can sustain considerably more current than rated power ever asks of them. That gap is power margin the project never priced and never measures — no monthly report has a line for it. Power fade spends the margin silently: capability decays from ample toward just-enough over years while every dashboard number holds steady, because the daily duty never visits the part of the envelope that is shrinking.

The demand curve is meanwhile rising to meet the falling capability. Holding nameplate MW constant on a faded fleet raises the real C-rate the cells see — a system built at 0.25C operates near 0.3C once retention reaches about 80% — and the same daily megawatt-hours now come out of a smaller pack, so each discharge runs deeper into the low-SOC end, where sag is worst and the current for a given megawatt is highest, at exactly the point in life the cells can deliver least.

When the two curves finally cross, it happens at a corner: the plant fails to hold rated MW into the tail of a discharge, or on a cold morning, and the capacity test books the shortfall as missing MWh. The mechanism was power, but the test sheet reads energy, which is one reason the phenomenon stays misdiagnosed. For short-duration, high-C-rate duty the crossing comes earlier — such a system can reach a practical power end-of-life before its energy end-of-life.

How it shows up in specs, tests and contracts

On a cell or DC-block datasheet, power capability appears as charge and discharge current or power tables against state of charge and temperature — beginning-of-life values, every one. In operation the same surface lives in the BMS as the available-power limits it publishes to the EMS, and those limits pull inward across life as the BMS re-estimates resistance; the plant's true power capability in year ten is whatever that estimator says it is, which makes the BMS's per-rack resistance history worth securing contractual access to.

The annual capacity test, meanwhile, proves power at one point only: the contract C-rate, across the contract window, temperature-corrected. Passing it demonstrates nothing about the corners the test never visits.

The warranty asymmetry is the commercial heart of the term. Capacity warranties are the standard instrument — year-by-year retained-energy tables with a test procedure and a remedy — while explicit power warranties are much rarer, because a power figure is only testable at a stated state of charge, temperature, direction and pulse duration, and no single test point represents the whole envelope. The practical questions follow from that gap: does the warranty track any resistance or power-capability metric, or only retained energy?

Does the vendor publish later-year power surfaces, or only the beginning-of-life table? And when the BMS caps available power below nameplate, does the availability guarantee count that hour as unavailable, derated, or fully available? For plants earning on MW — frequency response, reserves, capacity obligations — settlement is on delivered power, and an unguaranteed power axis is an unhedged revenue risk.

Common pitfalls

The standing confusion is with derating, and the test is reversibility at matched conditions. A 45 C afternoon and a decade of cycling can take the same megawatts off the plant; the afternoon gives them back at sunset. Derating is a healthy machine reading a different row of its rating matrix; power fade is the matrix itself decaying. The distinction decides who owns the shortfall — a derated hour is an accepted design condition, a faded megawatt is degradation the warranty may or may not cover — so logging which one occurred is worth the trouble.

The subtler pitfall is single-axis health tracking. Two fleets at the same retained energy are not the same machine: one can hold rated power to the floor of the window while the other caps out early on cold mornings, and a State of Health dashboard reporting only capacity cannot tell them apart.

The same blindness afflicts vendor comparisons — pulse-power and resistance figures are only comparable at matched state of charge, temperature and pulse duration, and a number quoted at 50% SOC and 25 C says nothing about the corner where the plant will actually feel its age first. Model the corners; that is where power fade lives.