How Capacitor Units Fail and What the Fuse Sees

Inside a capacitor unit

A capacitor unit is a sealed can holding many wound elements, connected in series sections with several elements in parallel in each section. Units are then connected in parallel groups to make up kvar and in series groups to make up voltage. The fuse question starts at the element.

Dielectric technology changed how elements fail. A widely cited protection paper explains that with older kraft paper or PCB dielectric, a failed element did not weld; the cellulose kept arcing, the paper charred, and gas built up inside the sealed can until, in many cases, it ruptured. Present-day units use as few as two or three layers of impregnated polypropylene film, and failures now cause the foils to weld together, forming a solid connection between the foils without arcing or charring.

How a failure progresses inside an externally fused unit

A relay maker's paper on capacitor bank fundamentals describes the sequence. An element failure welds the foils together and short circuits the other elements connected in parallel in the same section. The remaining series sections now share the unit voltage among fewer sections, so each sees more. If a second element fails, the process repeats, with an even higher voltage on what remains. Successive failures within the same unit eventually make the external fuse operate.

Each early step adds only a little current. A utility and relay maker paper notes that with all-film units the external fuse does not melt with the additional current drawn by the unit after a single element failure, so the unit stays in service until further series failures push the current high enough. The same paper series observes that element failures put the highest voltage stress on the rest of the failed unit, so the failure tends to progress inside that unit to a complete failure before it spreads to the rest of the bank.

Internally fused units behave differently. Each element has its own small fuse, and the failed element is disconnected, failing open, within a few cycles. Less capacitance is lost per failure, and the unit stays in service. That is why internally fused banks need more sensitive unbalance protection.

What the fuse sees when a unit fails

The protection paper identifies two components of the current through the failed unit's fuse: a higher power frequency current, because the shorted unit has lower reactance, and a transient current from the discharge from the healthy parallel elements or units. Both components must be considered when selecting a fuse: the fuses on the healthy units must not blow while they discharge into the faulted one, but the fuse on the failed unit must remove it quickly.

The discharge carries real energy. One manufacturer's capacitor fuse guide calculates it with the parallel capacitance charged to 1.1 times the crest of rated voltage, giving 3.19 joules per kvar. The arithmetic behind that figure: E = 0.5 C V^2 with V = 1.1 x sqrt2 x rated voltage and C = kvar / (2 pi f V^2) works out to 1.21 x 1000 / (2 pi f) = 3.21 J per kvar at 60 Hz, the same within rounding.

Worked example from that guide: a bank with ten 200 kvar, 13,280 V units per group. When one fails, the other nine discharge into it: 3.19 x 9 x 200 = 5,742 J. For scale, a single 50 kvar, 7,200 V unit (2.56 uF) charged to its 10,182 V crest holds 0.5 x 2.56 x 10^-6 x 10,182^2 = 133 J. How that energy is shared depends on the fuse: the same guide says an expulsion fuse conducts to a natural current zero so most of the energy is absorbed in the failed capacitor, while a current-limiting fuse with back voltage absorbs a large portion of it.

Voltage on the healthy units after a fuse operates

When a unit fuse opens, the parallel group it belonged to has less capacitance, so it takes a larger share of the string voltage. A relay maker's published unbalance paper gives exact formulas using P = units in parallel per group, S = series groups per string, R = strings in parallel per phase, F = failed units, and kOV = voltage on the most stressed healthy unit divided by its normal voltage. For a grounded single wye, fused bank: kOV = SP / (SP - F(S - 1)).

Worked example with that formula, S = 4 series groups and P = 10 units per group: one failed unit gives 40 / (40 - 3) = 1.081; two give 40 / 34 = 1.176; three give 40 / 31 = 1.290. As S grows, the expression approaches P / (P - F), so with ten units in parallel one open fuse gives 10 / 9 = 1.111. That is the origin of the published design rule to choose at least ten units in parallel so that one open fuse does not leave more than 10% overvoltage on the remaining units.

The same paper works an ungrounded single wye example: 138 kV, externally fused, four series groups of fourteen 21.6 kV units. Normal unit voltage is 138 / (1.732 x 4) = 19.92 kV, so the permissible factor is 1.1 x 21.6 / 19.92 = 1.192. One failed unit gives 1.063, two give 1.135 and three give 1.217, so the bank can tolerate two failed units in the same group of fourteen, but not three. System overvoltage stacks on top: two failures plus 10% system overvoltage gives 1 + 1.1 x (1.135 - 1) = 1.149.

Why the 110% line matters, and how failures cascade

IEEE's 2005 interpretation of IEEE 18-2002 says the 110% maximum continuous overvoltage capability is provided so the bank can stay in service after the failure of at least one unit or element, and that it was intended for contingencies, not as a design basis. An informal subcommittee survey reported in the same document found users typically alarm at 105% of rated voltage and trip on unbalance above 110%.

The relay maker's paper explains why the limit is enforced: a failure causes an internal overvoltage inside the bank, which may cause more failures, which create even higher overvoltage, and eventually a cascading failure. It also notes that after a unit fails open in a group, the next unit failure is most likely to be in the same group, because those units are now the most stressed. A university dissertation on unbalance protection adds that capacitor units most often fail from continuous overvoltage stress, and that fusing is the first line of protection against unit failure.

Fuseless strings spread the stress differently. With E elements in series, losing one raises the voltage on the others by about E / (E - 1): published examples give 48 / 47, about 2%, and 100 / 99, about 1%. A published 230 kV fuseless bank table shows 1.41% overvoltage with one failed element, 4.35% with three and 9.09% with six, with alarm at three and trip at six.

Other ways a bank faults

Unit failures are not the only events the fuses and relays must cover. Published protection papers list these as well:

• Arc-over within the capacitor rack, and rack-to-rack flashovers between series groups where connections are not insulated.
• Bushing failures or faulty connections at a capacitor unit.
• Lightning: the RUS bulletin notes a surge charges the capacitor to the surge voltage by passing enormous charging current, which can cause the unit to fail from internal heat.
• Animals and contamination: one large utility's field experience, published with a relay maker, is that arcing and case ruptures are no longer the main causes of bank trips; most are caused by animals climbing on the bank, contamination or human error.
• Discharge current from parallel units into an external fault, and inrush when the bank or a neighboring bank is switched.

Case rupture: what fusing has to prevent

If a failed unit carries fault current long enough, internal arcing builds gas pressure until the case bursts. Capacitor makers publish tank rupture curves for that limit. One maker's curves for standard and heavy duty units carry a minimum of 2,000,000 A2s for fault currents of 10,000 A or less; its extreme duty curve 5,600,000 A2s up to 15,000 A; a second maker states 3,500,000 A2s minimum for 100 kvar and larger units below 10,000 A. To put 2,000,000 A2s in time: at 10 kA it is 2 x 10^6 / 10,000^2 = 0.020 s, about 1.2 cycles at 60 Hz.

The curves label a safe zone, usually no greater damage than slight swelling of the case, and an unsafe zone where the case may rupture. A fuse coordinates when its maximum clearing curve falls to the left of, or below, the rupture curve at and below the available fault current. One unit maker's instruction sheet is candid that fusing reduces the risk but does not remove it: explosive case rupture is possible in any application, even with proper fusing. It adds that partial fusing on delta or ungrounded wye systems, such as a two-bushing unit fused on one terminal, can let an internal fault from the unfused side to the case rupture the unit.

The case rupture article covers reading these curves against a fuse.

How to recognize a failed unit

Modern all-film units usually do not bulge from internal gassing when they fail, so visual signs are weaker than they were. In an externally fused bank the spring-loaded fuse or its indicator shows which unit dropped out. The RUS bulletin says slight bulging or blooming of a tank is not necessarily a failure; failure is usually shown by a severely bulging or ruptured case, or a blown fuse.

The reverse is also true. A utility and relay maker paper warns that intact fuses do not mean the units are healthy, and a blown fuse does not necessarily mean the unit failed; finding partially failed units needs a capacitance measurement of all units. The RUS bulletin says a unit whose measured capacitance varies significantly from the rated value should be retired.

Common mistakes

• Assuming a unit with an intact fuse is healthy. Element failures can accumulate below the fuse's operating current.
• Sizing a unit fuse for the failed-unit current only. Healthy units' fuses also carry the discharge into the failed unit and must not operate.
• Ignoring parallel energy. Every added unit in a group adds stored energy that the failed unit and its fuse must absorb or hold off.
• Leaving a bank in service after a blown fuse without checking the overvoltage on the remaining units in that group.
• Treating a slightly swollen case as proof of failure, or a flat case as proof of health. Measure capacitance.

Frequently asked questions