What Actually Goes Wrong Is Never the Wiring
The wiring is the easy part. Positive to negative raises voltage, positive to positive raises capacity, and any electrician on your floor can execute either arrangement correctly on the first attempt. What makes the series versus parallel battery connection decision consequential in motive power is that both paths can deliver the same kilowatt-hours on paper, and the difference between them does not surface during commissioning. It surfaces somewhere in year two or year three, as an unexplained loss of usable runtime, a pack that ages faster than its siblings, or a BMS that starts tripping under a duty cycle it handled fine when the fleet was new.
That delay is the whole problem. By the time the symptoms appear, the topology is welded into the equipment, the cable runs are cut, the chargers are bought, and the cost of reversing the decision is a fleet-wide retrofit. So the useful question is not which configuration adds voltage and which adds capacity. It is which configuration your equipment, your shift pattern, and your maintenance capability can still live with after three thousand cycles.
Five Dimensions That Change, and Only Five
Before going anywhere useful, it is worth compressing the standard explanation into something you can scan once and stop rereading. Everything a series versus parallel battery connection changes falls into five dimensions.
| Dimension | Series connection | Parallel connection |
|---|---|---|
| System voltage | Sums across packs | Unchanged |
| Nominal capacity (Ah) | Unchanged | Sums across packs, in theory |
| Current at a given power | Falls proportionally with voltage | Stays high, set by pack voltage |
| Cable and contactor sizing | Smaller conductors, lower conduction loss | Heavier conductors, more heat at terminations |
| Dominant failure mode | Chain failure: one pack down stops everything | Silent imbalance: the bank keeps running while degrading unevenly |
Total available energy is identical either way for the same number of packs, which is why the arithmetic alone will never tell you what to do. The last two rows are where the decision actually lives, and they are the two rows that consumer-facing comparisons of battery series and parallel wiring tend to hand-wave past. Cell count and chemistry set the rest: a 48V LiFePO4 platform lands at fifteen to sixteen cells in series, which is worth knowing before you start reasoning about how many packs sit around that string. If you are still fixing the voltage platform itself, the voltage and capacity targets behind a custom pack are the upstream decision.
Why Higher Voltage Usually Wins in Motive Power
For material handling and industrial vehicles, the series-weighted answer is right more often than not, and the reason is conduction loss rather than anything about the battery.
The arithmetic below is derived from two relationships rather than measured on a test bench, so the assumptions matter: it holds current and delivered power as the fixed quantities, treats conductor and termination resistance as constant, and ignores the second-order effect of temperature on resistance. On that basis, power being the product of voltage and current means a 9.6 kW draw at 48V pulls roughly 200A, while the same 9.6 kW at 24V pulls roughly 400A. Conduction loss scales with the square of current, so doubling current for the same delivered power roughly quadruples the loss in every conductor, lug, contactor, and fuse holder between the pack and the motor. Move the same load from 48V up to an 80V platform and current falls by about 40 percent, which cuts conductor loss by roughly two-thirds. That energy does not disappear. It becomes heat at terminations, which is exactly where connection resistance lives and where it grows over time.
The practical consequences are unglamorous and expensive. Heavier conductors cost more, weigh more, and are harder to route through a battery compartment dimensioned around a lead-acid brick. Contactors switching higher current wear faster. And voltage sag under peak load bites directly into performance: on a counterbalance truck, a meaningful voltage drop during a heavy lift translates into a slower lift, which the operator experiences as the truck feeling tired long before anything registers as a fault. This is the core reason a series versus parallel battery connection for forklift and AGV systems tends to resolve toward the series side at equal energy. It is also why the 48V and 80V platforms behind our electric forklift batteries are built around long series strings rather than wide parallel banks.
There is a limit, of course. Series strings are constrained by what the equipment's controller, contactors, and insulation system are rated for, by the safety regime that applies above certain DC voltages, and by the fact that a single string means a single point of failure. Which leads to the question nobody in this space answers straight.
The Question Nobody Answers: How Many Packs Can You Parallel?
Ask a supplier and you will usually get a number, four, sometimes eight, with no explanation attached. That number is doing two jobs at once: it is partly an engineering limit and partly a warranty boundary. Treating it as physics when it is policy, or as policy when it is physics, is how fleets end up with banks that were never going to work.
Start by clearing up a confusion that appears in almost every article on this topic. The figures widely quoted by cell suppliers as pairing criteria, voltage difference within 10 mV, internal resistance difference within 5 mΩ, capacity difference within 20 mAh, describe cell sorting on the production line when selecting which cells go into one pack. They are not a field threshold for connecting two finished packs, and a completed pack will never sit within 10 mV of another after a week in service. Applying a production tolerance to a commissioning task is why some fleets conclude, wrongly, that paralleling is impossible.
The field limit is not a constant at all. It is calculated, and the calculation is short enough to do on a clipboard.
Inrush current at the moment of connection equals the voltage difference between the incoming string and the bus, divided by the total resistance of the loop. Because total resistance falls as more strings are added, the worst case gets worse with every pack you add, which is precisely why the ratings must be checked against the worst case rather than the two-pack case (Orion BMS).
Worked through with numbers, for two 48V 300Ah packs: assume roughly 15 mΩ internal resistance per pack and 5 mΩ across cabling and terminations, giving about 35 mΩ around the loop. If the binding constraint were the contactor, say a 300A inrush rating, the permissible voltage difference would be around 10V, which is a uselessly permissive answer. The binding constraint is almost always the battery instead: cap mutual charging at 0.2C, or 60A for a 300Ah pack, and the permissible difference drops to roughly 2V. Note that 0.2C is a conservative starting point rather than a constant, in the same way the 15 mΩ is a room-temperature estimate. The figure you should actually use comes from the cell manufacturer's maximum charge rate at your pack's operating temperature, and it moves when the pack size, the cable run, or the contactor changes.
Which gives you something more useful than a threshold. Next time a supplier quotes you a parallel limit, you can ask which of those two constraints produced it, and what internal resistance, contactor rating, and charge-rate ceiling they assumed. A number derived from the calculation above can be defended. A number that cannot be traced back to either constraint is a warranty boundary wearing an engineering costume.
Two further limits sit behind the arithmetic. Circulating current, meaning current flowing between packs rather than to the load, driven by differences in state of charge, temperature, internal resistance, and age, can in severe cases exceed the ratings of protection devices sized only for load current, and a DC-side short in a wide bank is fed by every string at once. And a wide parallel system needs layered management: cell-level BMS inside each pack, a coordinating layer above it, and a supervisory function deciding which strings participate. Neither is retrofittable once the bank is already misbehaving.
One correction to the standard framing, since it changes the sizing maths: paralleling does not reliably add capacity in the arithmetic sense. Usable capacity is limited by whichever string first reaches its charge or discharge limit, so every unit of inconsistency comes straight off effective system capacity. Four packs in parallel give you something less than four packs of usable energy, and the gap widens with age.
The behaviour underneath all of this is measurable rather than mysterious. Hot-swap circulating current in a parallel lithium system has been modelled with neural-network methods and validated experimentally within roughly six percent of measured values (MDPI Electronics), and current-sharing dynamics in parallel-connected packs are characterised in detail in the control literature (arXiv preprint 2211.04961). Note also that only the series-connected portion of a system requires active balancing. Directly paralleled cells equalise through their own connection, which is why cell balancing behaviour and inter-string circulating current are two separate problems that get discussed as one.
What the calculation above cannot give you is the resistance values to put into it. Internal resistance varies with cell format, age, and temperature, and the figure on a datasheet is a fresh-cell figure at 25°C. We keep measured resistance curves for the formats we build with, and matching those to a specific duty cycle is the part of this that happens in conversation rather than on a clipboard.
When a Single String Beats Any Parallel Design
We will argue against the wider bank more often than customers expect, so it is worth putting the position on record.
The default preference in battery systems engineering is a single series string wherever the application allows it. It is the lowest-cost configuration, the simplest to manage, and it eliminates the entire category of inter-string problems described above. Multiple parallel strings are justified in a narrow set of circumstances: when redundancy is a hard requirement because the equipment cannot be allowed to stop; when hot-swap capability is needed, as in uninterruptible or continuously-manned operations; and when the only cells available in the required format come pre-assembled into strings that cannot be paralleled at cell level.
The exception that gets misjudged most often is the first one. Redundancy is frequently assumed rather than specified. A fleet manager reasons that a truck stopping mid-shift is unacceptable, and concludes redundancy is required. But a parallel bank only delivers redundancy if the system can detect a failed string, isolate it, and keep operating at reduced capability, which is a control requirement rather than a wiring one. Wiring packs in parallel without that capability buys the complexity of a parallel-heavy industrial series versus parallel battery connection and none of the availability. Outside these cases, a wide bank is usually a workaround for a packaging constraint rather than a design choice, and it should be recognised as such.
There is also a quiet cost that rarely gets counted. Every additional termination adds resistance, and a bank built from many packs multiplies terminations quickly: a four-cell arrangement alone presents eight contact interfaces, each contributing resistance and each a candidate for degradation (Battery University BU-302). Scale that logic to a sixteen-pack bank and the connection budget stops being a rounding error. Our own internal target for a completed motive-power pack is total resistance under 50 mΩ, with individual connections in the 0.5 to 2 mΩ range, and wide parallel banks make that target progressively harder to hold.
The Year-Two Problem
Everything above assumes the packs are consistent. They are, at delivery. The part almost nobody writes down is what happens after that.
Packs that leave a well-controlled production line matched to tight tolerances drift apart in service, and roughly a year of real duty is enough for the spread to become operationally relevant. Ambient temperature varies by where a truck parks. Duty cycles vary by which operator draws which vehicle. Charge opportunity varies by shift. The result is a bank whose strings no longer sit at the same state of charge or the same internal resistance, connected by a very low-resistance path, which is the exact condition that produces sustained mutual charging between packs.
The failure mode this produces is counterintuitive enough to state directly: in a mismatched parallel bank, the strong packs degrade fastest. They carry more of the load current because their resistance is lower, they absorb circulating current when others sag, and they accumulate throughput at a rate the weak packs never reach. The pack you would have bet on fails first, and because the bank keeps delivering usable power throughout, nothing announces the problem until capacity has already gone.
This is also why "just add a new pack to the bank" is the single most common field mistake we see. A fresh pack introduced into a bank that has drifted for eighteen months is not a capacity upgrade. It is a large potential difference across a very low-resistance path, and it belongs in the inrush calculation above before anyone touches a contactor. If a pack must be reintroduced, bring it to matched open-circuit voltage at rest first, and apply the same discipline whether it is a new unit or a previously isolated string coming back online. Watching for the end-of-life signals in an aging pack matters more in a parallel bank than in a single string, precisely because the bank hides them.
Layout and Thermal: The Constraint Behind the Schematic
A topology chosen on paper becomes a physical object, and physical objects have temperature gradients.
Air-cooled packs use either series or parallel ventilation paths, and the terminology collides confusingly with the electrical vocabulary while describing something entirely separate. In a series airflow path, air enters one side and exits the other, carrying heat picked up from early modules into later ones and producing a temperature spread across the pack. A parallel airflow path routes air between modules so it rises more evenly, holding the spread tighter. Modules in the middle of an enclosure accumulate heat; modules at the edges shed it. Since internal resistance and capacity both vary with temperature, an uneven thermal layout manufactures the very inconsistency your electrical design was trying to avoid. That is the reason thermal management architecture belongs in the topology conversation rather than after it.
Two layout rules follow. Every parallel branch must contain an identical series configuration, without exception, because asymmetric branches guarantee circulating current before the pack has even been commissioned. And cable lengths between paralleled packs and the common busbar should be equal, so path resistance does not silently designate one pack as the workhorse. Both are cheap to get right during design and effectively impossible to fix afterwards.
Temperature also changes the numbers you thought you had. A food-distribution customer in Mexico runs our LiFePO4 pallet jack packs in cold storage down to around -18°C, and the packs hold output without a noticeable drop. But current sharing between parallel strings at that temperature does not behave the way it does at 25°C, because internal resistance rises and the resistance spread between packs widens with it. The 15 mΩ figure used in the earlier calculation is a room-temperature figure. In a freezer aisle it is optimistic, and the permissible voltage difference at connection tightens accordingly. A bank that shares current acceptably in a mixed-temperature warehouse may share it poorly in cold store.
Three Configurations, Worked Through
The following are the three patterns we see most often, and how the series versus parallel battery connection resolves in each.
| Application | Configuration | Reasoning |
|---|---|---|
| Electric pallet jack, single shift | 24V single series string | Modest power draw, tight compartment, no redundancy requirement. Parallel strings would add cost and failure modes for no operational benefit. |
| Counterbalance forklift, two shifts | 48V single string, 15–16 cells in series, high-capacity cells | Higher voltage keeps current and conductor sizing manageable under repeated heavy lifts. Capacity is bought through cell format, not through parallel packs. |
| Heavy-duty fleet or GSE, continuous operation | 80V string with a deliberate, managed parallel arrangement | Redundancy and hot-swap justify the added complexity, but only with matched packs, symmetric branches, layered BMS, string isolation capability, and protection rated for worst-case inrush. |
The pattern across all three is the same: reach the energy target through cell format and series voltage first, and treat parallel strings as something adopted for a specific operational reason rather than as the default route to more amp-hours. Where compartment geometry makes that impossible, the constraint should be named as a constraint in the specification rather than absorbed silently into the design. That is also the fastest way to tell whether a motive power battery range can meet the target without a custom build.
Specifying It to a Supplier
Once the topology direction is clear, the specification you send out determines whether you get a pack engineered for your duty cycle or a catalogue item with your voltage on the label. Six inputs make the difference:
- Equipment nominal voltage and the controller's acceptable voltage window, not just the nominal figure
- Peak current draw during the heaviest operation in the cycle, and its duration
- Battery compartment dimensions and maximum permissible pack weight
- Shift pattern and available charge windows, including whether opportunity charging is expected
- Operating temperature range, with cold-storage or outdoor exposure called out explicitly
- Whether the equipment can tolerate a full stop, which determines whether redundancy is a real requirement or an assumption
The questions you ask back are worth more than the specification you send, because each one filters a different kind of supplier. Asking what the quoted parallel limit is based on separates the vendors who calculated it from the ones who inherited it from a datasheet. Asking what internal resistance and contactor rating went into that calculation tells you whether anyone has looked at your equipment rather than a generic case. Asking how the strings communicate, whether there is a coordinating layer above the pack-level BMS or just several packs sharing a busbar and hoping, is the fastest way to find out whether a supplier has built multi-string systems or only sold them. And asking for the documented procedure for reintroducing an isolated pack will tell you whether they have ever had to support one in the field, because that procedure only gets written after the first time it goes wrong.
We run the same four questions against ourselves before quoting any multi-pack system, and the resistance curves and reintroduction procedure referenced earlier are what that process produces. Getting the series versus parallel battery connection right for a specific duty cycle is where that work happens, which is what custom pack configuration is for.
FAQ
How many battery packs can be connected in parallel?
There is no universal number. Calculate it: permissible voltage difference at connection equals your accepted inrush current multiplied by total loop resistance, with the accepted inrush set by whichever is lower, the contactor rating or the mutual charging rate the cells can tolerate.
Does a parallel battery connection really add capacity?
Not fully. Usable capacity is limited by whichever string reaches its charge or discharge limit first, so inconsistency between strings comes directly off effective system capacity.
Can I add a new battery pack to an existing parallel bank?
Only after matching open-circuit voltage at rest. Connecting a fresh pack to a bank that has drifted in service creates a large mutual charging current across a very low-resistance path.
Is series or parallel better for forklift and AGV batteries?
For motive power, the series versus parallel battery connection usually resolves toward series, because higher voltage reduces current at the same power and cuts conduction loss across the whole electrical path.
What happens if one battery fails in a series connection?
The circuit opens and the system stops. That chain-failure mode is the main reason redundancy-critical applications accept the engineering cost of parallel strings.
If you are sizing a multi-pack system for a specific fleet and want the configuration reviewed before it is committed, talk to our engineering team.
