Two numbers decide how long an aluminum electrolytic bank survives in a power conversion system: the temperature inside the capacitor, and the voltage across its terminals. Capacitance is not one of them. A 20-year service target is reached by setting a core temperature budget first, then picking a series whose published ratings survive that budget.
The first two articles in this series, on the DC-link bank and on the DC bus stage by stage, worked out where the capacitors sit and what each position is sized against. This one runs the calculation the other way, starting from the service life the project has to deliver and working backwards to a part number.
Why the Capacitor Is the Wear Part in a 20-Year System
Energy storage assets are financed over 20 to 25 years, and the transformers, switchgear, enclosures and power semiconductors inside a PCS are all specified against that horizon. Aluminum electrolytic capacitors usually are not, and the arithmetic shows why.
Take the 20-year inverter target the US Department of Energy has used for photovoltaic systems. It works out to roughly 80,000 hours of operation. The longest endurance published for aluminum electrolytics has historically sat near 20,000 hours at 85 °C, and a careful ripple and thermal design pushes a good bank toward 40,000 hours, which is about ten years of continuous running.
The way to close that gap is to treat the bank as a wear part and design the service interval around it. A replacement at year ten, planned and reachable, still lets a 20-year system reach its design life. A bank specified as though it will outlast the switchgear delivers its first failure as an unplanned outage.
The Lifetime Equation, and Where It Stops Working
Manufacturers publish endurance as a rated life at a rated core temperature and a rated ripple current. The usual extrapolation to a different operating point is:
L = L0 × 2(T0 − Tc) / 10 × (V0 / V)n
L0 and T0 are the rated life and rated core temperature, Tc is the core temperature actually reached, V0 and V are rated and applied voltage, and n is a voltage exponent that runs from about 3 to 5 depending on the series. The first term is the doubling rule familiar to anyone who has ever sized a power supply. The second is the term most energy storage designs never fill in.
Apply the doubling rule alone to a part rated 8,000 hours at 105 °C, and assume the inverter runs eleven hours a day:
| Core temperature | Extrapolated endurance | Years at 11 h/day |
|---|---|---|
| 105 °C | 8,000 h | 2.0 |
| 95 °C | 16,000 h | 4.0 |
| 85 °C | 32,000 h | 8.0 |
| 75 °C | 64,000 h | 15.9 |
| 65 °C | 128,000 h | 31.9 |
Read the table backwards. A 20-year target over that duty cycle needs about 80,000 hours, which is ten times the rated endurance. Ten doublings means three and a third ten-degree steps, so core temperature has to sit at roughly 72 °C or below for the whole service life.
The equation has edges worth knowing. It models electrolyte loss, and it holds only while electrolyte loss is the mechanism limiting life. Below ambient it is meaningless. At the far end, seal degradation and oxide growth set their own floor. Manufacturers will also decline to warrant an extrapolation beyond some multiple of rated endurance, commonly three to five times, so a design leaning on a twenty-fold boost from temperature needs checking against the series datasheet rather than assuming.
Core Temperature Is Not Ambient Temperature
The most common mistake in these calculations is reading the cabinet temperature into Tc. Core temperature is ambient plus self-heating:
ΔT = Irms2 × ESR / thermal dissipation
A cabinet sitting at 50 °C ambient can easily run its capacitors at 60 to 65 °C once ripple losses are added, and on the doubling rule that difference is worth two to three times the service life.
ESR moves the loss in proportion, which is why low impedance series exist and why they matter more in a continuously loaded application than in a switching supply that idles. Ripple margin moves it quadratically: running a part at 70% of rated ripple instead of 95% cuts the loss by nearly half. Layout moves the dissipation path, since mounting above a heat source or blocking airflow across a group raises the effective ambient even when the cabinet reading looks fine.
When the estimate is too uncertain to design against, measure it. A thermocouple bonded to the can of a representative unit, read at full load inside the real enclosure, replaces a guess with a number for a few hours of work.
Working Voltage Is the Second Knob
Energy storage is unusual in holding its bus at high voltage for most of the day. A 1,500 V system sits with its DC link somewhere around 700 to 800 V continuously, rather than idling at a low rail the way a consumer supply does. The voltage term in the lifetime equation therefore stays loaded, and it deserves the same attention as temperature.
Derating to 80% of rated voltage is the normal starting point, and strings used at 1,500 V typically carry 20 to 30% voltage margin across the group. The effect is measurable. With n = 4, a 10% voltage derating multiplies calculated life by about 1.5.
Series connection brings a lifetime problem of its own. Unequal leakage current drives the DC voltage across a string out of balance, and whichever capacitor ends up with the largest share ages fastest, then takes the rest of the string with it. Balancing resistors are not optional beyond two units in series, and their own dissipation has to be counted in the thermal budget.
Working Through a Real Selection
Take a 400 kW string PCS with a DC link held at 700 V, a cabinet ambient of 50 °C, and eleven hours of operation a day for twenty years. Twenty years at that duty is 80,300 hours of operation, which is the endurance the bank has to deliver.
Against a series rated 8,000 hours at 105 °C, reaching 80,300 hours needs a ten-fold extrapolation, which puts the core temperature ceiling at 72 °C or below. At 50 °C ambient that leaves 22 °C of headroom for self-heating. A well-chosen low ESR part running at 70% of rated ripple typically contributes 5 to 10 °C, so the headroom holds, provided the ripple current is measured rather than assumed from the load.
Voltage comes next. A 700 V bus cannot be held by a single aluminum electrolytic, since standard units top out in the 400 to 500 V range. Two 450 V parts in series put 350 V across each, which is 78% of rated voltage and inside the usual derating target. Balancing resistors go across both.
That leaves the endurance class. The long-life series rated at 8,000 and 12,000 hours exist for duty cycles like this one, and choosing the higher rating buys margin against a core temperature that ends up warmer than the calculation predicted.
Xuansn manufactures radial, snap-in and screw terminal aluminum electrolytic capacitors across these endurance classes, with rated life, ESR and ripple current published for each series.
FAQ
Can an aluminum electrolytic bank really last 20 years in a PCS?
Yes, if a replacement is planned rather than assumed away. Very few aluminum electrolytics will run 80,000 hours with a core temperature above 70 °C. Most 20-year designs get there by holding a low core temperature with a long-life series, or by documenting a mid-life change-out. Splitting the duty with film capacitors is the other route when the high-frequency ripple content is high, since the electrolytic group then carries only the bulk energy.
What core temperature should I design to?
Work backwards from the service target instead of picking a round number. Eighty thousand hours against an 8,000-hour part lands at 72 °C or below; a ten-year target relaxes that to about 82 °C. The number that matters is the can temperature in service, not the cabinet ambient.
Does a 10 °C reduction really double the life?
It is a good working approximation, and it is what the standard extrapolation uses, but it holds in the middle of the range. At the edges the relationship flattens, and the manufacturer’s warranty limit on extrapolation applies. Treat the doubling rule as a design guide and the endurance curve, where a series publishes one, as the authority.
How do I measure core temperature in service?
A thermocouple bonded to the can of a unit in a representative position, read at full load inside the real enclosure. Some designs also fit a surface temperature sensor to one capacitor in the group for condition monitoring. Either is more reliable than calculating from cabinet air, which misses self-heating entirely.
If you are working through this calculation for a specific design, send the bus voltage, the ripple current and the cabinet ambient to coco@xuanxcapacitors.com, or contact us here, and we will match a series to the numbers.
