DC block

Where the block stops

The defining feature is the terminals. Inside sit battery racks with their contactors, pre-charge circuits and rack fuses, the section and array fusing that combines them, and a DC disconnect; outside sits the equipment that turns the bus into three-phase AC. None of that hardware is unusual — the boundary is what the name marks.

A block's rating is therefore written entirely in DC units: energy in MWh at the battery terminals, a voltage window, a continuous current and a fault duty. Converter datasheets describe the same seam from the other side, as a count of DC inputs with a current limit and a withstand rating attached to each one.

How many blocks can land separately on one converter is an ordering decision rather than a fixed property of the equipment. Power Electronics' gen-3 Freemaq Multi PCSM family publishes the same DC total split two ways: the FP4200M2 frame takes two DC inputs rated 2,295 A each, the FP4200M4 takes four rated 1,148 A each — about 4,590 A either way, with the input count carried as the last digit of the model code.

Other vendors structure it differently. EPC Power's M System documentation allows one BESS DC input per M Inverter alongside up to three PV inputs and states that DC bus configurations are set project by project, while an SMA central-inverter datasheet offering DC coupling of a battery lists six double-pole-fused DC inputs for batteries next to eighteen for PV. Check the count before assuming each enclosure gets an input of its own.

Scale matters more than it looks. Whether a specification means one enclosure or a group of enclosures changes how many independently isolatable units the plant has, how many fuses and disconnects appear on the bill of materials, and how much of the plant a single maintenance isolation removes from service. It also changes what a per-block availability or capacity figure refers to, which is why the definition belongs in the specification rather than in the reader's assumptions.

Sizing the bus from the bottom of the window

Cables, busbars, disconnects and fuses are sized against current, and the current a block draws for a given power depends on where the bus voltage sits when it draws it. A 1500 V-class LFP string of roughly 360 to 416 cells runs nominally around 1,150 to 1,330 V, against a protected floor around 900 to 1,040 V — and that floor is a cold, loaded number, not a sum of open-circuit cell voltages.

Across that span the same megawatt asks for roughly 28% more current at the floor than at nominal. The plant spends most of its life nearer the middle, but the corner is where conductor and fuse selection has to be defensible.

The converter imposes two limits on the same bus, and they bind at different moments. One is the per-input continuous DC current: 1,148 A on a four-input gen-3 frame, which at a nominal 1,150 V works out on the order of 1.3 MW per input before any derating is applied.

The other is the DC range across which the converter will make full power — Power Electronics' gen-3 sheets quote full power from 976 V to 1500 V on the 690 V AC variant and from 1,019 V to 1500 V on the 720 V variant, both with the note that the minimum rises in proportion to AC voltage during a grid overvoltage. Where a block's protected floor sits below the converter's full-power minimum, the bottom of the discharge remains available to the battery but not at the converter's full rating; the VDC window entry owns that knee and the derating below it.

One detail decides where all of this is measured. Voltage at the converter terminals is not voltage at the block terminals once a few thousand amps are flowing through the run between them, so a window check performed on the wrong side of that drop can pass on paper and fail in the field.

Fix the measurement point in the interface document, and check it at the corner that actually threatens it: the lowest state of charge the block is allowed to reach, cold, at the highest current the plant will command. The same sheets also specify DC voltage ripple below 3%, which is as much a limit on what the converter puts back into the block as a description of the bus.

Protection scales with what shares the bus

Paralleling is what turns modest string fault currents into a duty that needs purpose-built DC protection. A single rack feeds something in the range of 1 to 12 kA into a DC fault; once racks and sections are paralleled, a combined system can reach 250 kA or more, and every protective device on the shared bus is rated against that rather than against the string immediately behind it.

Converter inputs publish their side of the same duty — the FP4200 M2 and M4 frames are rated for a maximum DC short-circuit current of 250 kA per input at a 3 ms time constant, other gen-3 frames for 500 kA at 1 ms, and EPC Power publishes a DC short-circuit capability of 230 kA with BESS DC fuses rated at 250 kA interrupting. Read those as withstand and interrupting ratings, meaning what the equipment survives or clears. They are not statements of what the battery contributes, and the two get swapped often enough to be worth naming as a distinct number every time.

Block-scale DC fusing is high-speed semiconductor work, and where it physically lives varies by vendor. Power Electronics' GEN3 skid bills of materials list Littelfuse POWR-SPEED PSX square-body fuses, class aR, 1500 V DC, across an 80 A to 1400 A range, with one 1500 VDC fuse per battery input at the skid's BESS-1 to BESS-4 positions.

EPC Power lists BESS DC fuses per M Inverter — 900 A at 250 kA interrupting in the January 2026 datasheet edition, where the April 2025 edition of the same sheet stated 800 A. One SMA central-inverter datasheet, by contrast, ships DC inputs as terminal lugs without fuses, with fused inputs offered as an option at 750 A, 900 A or 1250 A. Whether the converter arrives with the block's fuse inside it is a purchase-order question that has three different answers on three vendors' sheets.

Two coordination questions sit exactly on the seam, which makes them easy for each supplier to assume the other has handled. The first is grading: the fuse protecting the shared bus and the fuses inside each enclosure have to be selective against one another, and those curves normally live in two different vendors' documents, so ask for both on a single time-current plot instead of accepting each in isolation.

The second is ground-fault detection. Grid-scale DC sides are commonly floated with continuous insulation monitoring, and a block and a converter can each arrive carrying a monitoring device; where two of them end up on one galvanically continuous bus they can interfere, so the interface schedule should name which device is active and who receives its alarm.

The seam is commercial before it is electrical

Buying a DC block means buying a guarantee that is referenced at DC terminals. Energy, retention and any round-trip figure in that contract are measured on the battery side of the boundary; conversion losses, transformer losses and auxiliary draw all sit outside it, and so does whatever the meter finally records.

That is a workable arrangement, and it is how a great deal of grid-scale storage is procured — but it leaves the number the project is actually paid on unowned unless somebody takes it deliberately, whether that is an integrator carrying a wrap, an EPC, or the buyer's own engineering. Splitting the supply does not by itself split the responsibility for the plant-level result; it removes it from both sheets.

What closes the gap is an interface schedule that both suppliers sign.

The items that belong on it are the DC voltage window including the full-power minimum, the continuous and peak DC current per input, the number of inputs and which enclosure lands on which, the fault-current withstand and interrupting duty on both sides, the fuse ratings and who supplies them, the disconnect and lock-out arrangement used for maintenance, ownership of insulation monitoring and ground-fault alarms, ripple limits, and the control handshake that decides which device commands contactor closing and which trips whom.

Add a revision reference to every line of it. A published BESS DC fuse rating moved between two editions of the same vendor datasheet within a year, and an interface document that names a value without naming the sheet it came from is a value nobody can re-derive later.

Commissioning is where an unmanaged seam usually surfaces. A DC-side acceptance test proves energy at the block terminals; a plant test proves output at the point of interconnection. Run on different days at different states of charge and temperatures, the difference between them is not attributable to anything, and each supplier can hold up a passing test. Where the DC and AC scopes sit in separate contracts, the cheapest fix is a commissioning sequence that instruments both boundaries in the same run, so the conversion chain is measured rather than inferred from two tests that never met.

Common pitfalls

The first is sizing DC gear at nominal bus voltage. The current that matters lives at the bottom of the loaded window, and the difference across a 1500 V-class LFP string is roughly 28% more current at the protected floor than at nominal — enough to move a fuse or a conductor a size. A related error runs the calculation from the block's power rating instead of the converter input's current limit, which are two different constraints that happen to be expressed in different units.

The second is assuming the converter arrives fused, and then assuming last year's rating is the one that ships. Some vendors fuse every battery input as standard, at least one offers unfused terminal lugs with fusing as a priced option, and at least one has revised the published rating between datasheet editions. All three facts are visible on current sheets, which means none of them can be inferred from the category of equipment being bought.

The third is treating a withstand figure as a contribution figure. A 250 kA per-input rating describes what the converter's DC input can tolerate at a stated time constant; it says nothing about the fault current a particular block will deliver into a particular fault, which comes out of that block's own configuration and its protection study.

Reading one as the other overstates duty in one direction and understates it in the other, and both errors stay invisible until a fault arrives. The fourth is subtler: agreeing to a specification that says "DC block" without saying which scale it means, then discovering at the layout stage that the isolation, fusing and maintenance-outage arithmetic was done against the other reading.