Battery Pack Design: The Complete Engineering Guide to Building Safe Lithium-Ion Packs for EVs, E-Bikes, and Robots

A dense engineering guide to lithium-ion battery pack design: series/parallel pack math, nickel-strip and busbar resistance, datasheet cell selection, spot welding vs soldering, BMS topologies and balancing, CC-CV charging, the thermal-runaway chain, UN 38.3 / IS 16046-2 / IEC 62133-2 standards, and a fully priced 48 V 10 Ah e-bike pack build in India.

·

Battery Pack Design: The Complete Engineering Guide to Building Safe Lithium-Ion Packs for EVs, E-Bikes, and Robots

Every e-bike on Indian roads, every drone in a maker's workshop, and every 6,831-cell Tesla Roadster that started it all is the same machine underneath: a handful of commodity cells whose arrangement, interconnect, and protection circuitry decide whether the system delivers rated power for a decade or vents fire on charge cycle 47. A single Samsung INR18650-25R cell stores only 9 Wh — about ₹2.50 of energy at Delhi's ₹8/kWh tariff — but the pack it lives in is where all the physics lives: series connections set voltage, parallel groups set current, nickel strips and busbars set resistance, and the BMS is the only thing standing between normal operation and thermal runaway. Sony commercialized the lithium-ion cell in 1991; the 18650 format followed a few years later, and Tesla's choice to build the Model S 85 kWh pack from 7,104 commodity 18650s (and the Model 3 from 2,976–4,416 21700s) is the reason a ₹15,000 DIY pack today outperforms a ₹40,000 sealed commercial unit of a decade ago. This guide covers the complete engineering of it: cell selection from datasheets, the series/parallel math worked end-to-end, interconnect design, spot welding, BMS topology, charging, the thermal-runaway chain, the UN 38.3-class standards, an India-anchored bill of materials, and the failure taxonomy that follows from getting any of it wrong.

flowchart LR
    A["52 × INR18650-25R<br/>13S4P · 46.8 V · 10 Ah · 468 Wh"] --- B["BMS 13S 30 A<br/>OVP 4.25 V/cell · UVP 2.9 V/cell<br/>passive balance 60 mA"]
    B --- C["Common port P− / P+"]
    C --- D["XT90 anti-spark + 40 A fuse"]
    D --- E["Motor controller / load"]
    F["54.6 V 2 A CC-CV charger"] --- C
    B -. "13 balance leads B0–B13" .- A

1. The Cell: Chemistry, Format, and the Numbers That Matter

Pack design starts at the cell, and every decision downstream — series count, busbar gauge, BMS class — inherits from four cell-level numbers: nominal voltage, capacity, maximum continuous discharge current, and internal resistance.

Chemistry sets the voltage window. The dominant options:

Chemistry · Nominal · Charge max · Discharge min · Character

NMC / NCA (Li-ion) · 3.6–3.7 V · 4.20 V · 2.5–2.75 V · Best energy density; standard for power tools, e-bikes, EVs

LiFePO₄ (LFP) · 3.2 V · 3.65 V · 2.5 V · Lower energy, 3–5× cycle life, thermally stable cathode

LTO · 2.3 V · 2.85 V · 1.5 V · Low voltage but 10 C+ charge rates and 10,000+ cycles

Na-ion (emerging) · ~3.0 V · 3.95 V · 1.5–2.0 V · Sodium cathode, no cobalt; cost-driven, first Indian entrants shipping

Format matters for packaging and current. The 18650 (18 mm × 65 mm) is the commodity workhorse; the 21700 (21 × 70) delivers ~50% more capacity at roughly similar cost per watt-hour; prismatics (EVE, CATL) and pouches scale to 280 Ah and beyond for stationary and traction use.

The datasheet table — the cells you will actually encounter in India, with manufacturer ratings:

Cell · Format · Capacity · Max cont. discharge · Typical IR · Notes

Samsung INR18650-25R · 18650 · 2.5 Ah · 20 A (8 C) · ≈18 mΩ AC · The power-tool standard; 9.0 Wh

Molicel INR18650-P26A · 18650 · 2.6 Ah · 25 A (9.6 C) · ≈15 mΩ AC · Independent testing holds 35 A with 80 °C cutoff

Samsung INR21700-50E · 21700 · 5.0 Ah · 9.8 A (~2 C) · ≈20 mΩ DC · Energy cell — wrong choice for power

Molicel INR21700-P42A · 21700 · 4.2 Ah · 30 A (45 A w/ temp cutoff) · ≈10 mΩ AC · The default high-power 21700

EVE LF280K · Prismatic · 280 Ah · 280 A (1 C) · ≈0.2 mΩ · LFP workhorse for e-rickshaws and storage

The 25R datasheet (charge 0–50 °C, discharge −20 to 75 °C, standard charge 1.25 A CC-CV to 4.20 V with ~100 mA cutoff) is a model of what to read before designing: every cell you use should have its manufacturer charge/discharge temperature windows, standard charge current, and cut-off conditions committed to the design. A pack designer who cannot produce the cell datasheet has no business assembling a pack.

2. Series and Parallel: The Pack Design Equations

The entire pack design collapses into four equations. With n_s cells in series and n_p in parallel:

Series adds voltage; parallel adds capacity and divides resistance; energy is the product of both. Series groups must be capacity-matched (a weak series group hits its cutoff early and the whole string stops), while parallel groups must be voltage-matched at assembly (cells wired in parallel equalize instantly — a 4.2 V cell paralleled with a 3.8 V cell dumps tens of amps into the lower one).

Worked example — a 48 V e-bike pack from 25Rs. With n_s = 13, n_p = 4 (52 cells):

Cell-level maximum current is n_p \times 20 = 80 A — far above what any 48 V controller in this class draws. The cell resistance alone is 13 \times 18/4 = 58.5 mΩ; interconnects add 10–15 mΩ more, so the finished pack sits near 70–80 mΩ. At a realistic 30 A load:

The pack sags from 46.8 V to ~44.4 V under throttle — 5% of bus voltage vanishing as heat, P = I^2 R = 30^2 \times 0.080 = 72 W dissipated across strips, welds, and cells. That sag number, not the label, is what a controller sees; it is why cheap packs with high-resistance interconnects "lose power" on hills.

Common series counts (BMS vendors name their products by them): 10S = 36 V (42 V max), 13S = 48 V (54.6 V max), 14S = 52 V, 16S LFP = 51.2 V nominal ("48 V class" in LFP chemistry), 20S = 72 V. E-rickshaws run 16S LFP; power tools run 5S–10S; drones run 3S–6S at absurd C-rates.

C-rate is the load normalized to capacity: I = C_{rate} \cdot Q. A 20 A draw on the 10 Ah pack is 2 C — mild. A 25R cell at its 20 A rating is running 8 C, at which it gives roughly 70–80% of its 0.2 C capacity and runs warm; the datasheet's 2.5 Ah is rated at 0.2 C and you must not design around it at 8 C.

3. Current Paths: Nickel Strip, Busbars, Wire, Connectors

Interconnect resistance is pure geometry, and it is unforgiving. Resistance of any conductor:

Pure nickel: \rho = 6.99 \times 10^{-8}\ \Omega\cdot\text{m}. A 0.15 mm × 8 mm strip, 20 mm long:

At 7.5 A (30 A pack current divided across 4 parallel cells) that strip dissipates I^2 R = 7.5^2 \times 0.00117 = 0.07 W — nothing. Push the full 30 A through a single 20 mm strip and it becomes 1.05 W in a piece of metal the size of a matchstick — a hot spot, a weld anneal, a failure. The design rule follows: parallel-group cells each carry I/n_p through their own strips; group-to-group series links must be sized for the full pack current, usually by stacking strips or moving to copper busbar.

Community-standard continuous ratings for pure nickel strip (derate these — published tables vary by 2×, and the honest engineering method is to compute R from the equation above and measure temperature rise):

Strip · Cross-section · Typical continuous rating

0.10 × 8 mm · 0.80 mm² · ~5 A

0.15 × 8 mm · 1.20 mm² · ~10 A

0.20 × 8 mm · 1.60 mm² · ~15 A

0.30 × 8 mm · 2.40 mm² · ~20 A

Copper busbar (\rho = 1.68\times10^{-8}\ \Omega\cdot\text{m}, 4.2× better than nickel) is the correct choice above ~50 A: a 20 × 3 mm copper bar (60 mm²) carries 120–180 A at the 2–3 A/mm² conservative design rule. Every series link in a 100 A-class pack should be copper with nickel tabs welded to the cells.

Wire follows the classic conservative chassis-wiring table (silicone-insulated, ~30 °C rise):

AWG · mm² · Chassis rating · AWG · mm² · Chassis rating

18 · 0.82 · 10 A · 12 · 3.31 · 23 A

16 · 1.31 · 13 A · 10 · 5.26 · 33 A

14 · 2.08 · 17 A · 8 · 8.37 · 46 A

For a 30 A pack, 12–10 AWG output leads are correct; use 10 AWG for anything past 40 A. Connectors: XT60 (60 A), XT90 (90 A, the anti-spark variant adds a series resistor that pre-charges controller capacitance before the main contacts make — without it, connecting a charged pack to a controller's ~1 mF of input capacitance produces a 100 A-class spark that pits connectors and pops BMSes), Anderson SB50 for industrial. A pack-level fuse of 1.25–1.5× maximum continuous current is non-negotiable — an ANL or blade fuse in a holder, never a bare inline splice.

4. Interconnects: Spot Welding, the Magnet-Test Myth, and Why Soldering Is Banned

Cell-to-strip joints are made by resistance spot welding: two electrodes press the strip onto the terminal while a 0.8–2 kA pulse flows for 2–5 ms, and the contact resistance (a few hundred µΩ at the interface) localizes Joule heat Q = I^2 R t into a molten nugget that freezes in milliseconds. Capacitive-discharge welders (the kWeld and Malectrics class, driven off a 3S LiPo) put 30–80 J into a weld for 0.15–0.2 mm nickel; cheap transformer welders deliver the same current with worse pulse control. Weld quality is verified by pull-testing (a good 0.15 mm weld holds several kilograms) and by inspecting for full-surface nuggets — a weld that is merely cosmetic adds milliohms of series resistance exactly where the sag math of Section 2 says it hurts.

Soldering cells is banned practice. A soldering iron idles at 350 °C against a can whose polyethylene separator shuts down at ~135 °C and melts at ~165 °C, and whose SEI layer begins decomposing above 90–120 °C. Ten seconds of tip contact is enough to locally melt the separator and plant an internal short that may declare itself weeks later as a fire. If a design genuinely requires solderable cells, buy factory tabbed cells — never solder bare cans.

The magnet-test myth. DIY forums have long repeated that pure nickel strip is non-magnetic while nickel-plated steel attracts a magnet. This is wrong physics: nickel is ferromagnetic — it is one of the four ferromagnetic elements, with a Curie point of 358 °C, and a neodymium magnet sticks to it firmly. The magnet test cannot distinguish them. Nickel-plated steel has ~2.3× the resistivity of pure nickel (steel \rho \approx 1.6\times10^{-7}\ \Omega\cdot\text{m}), rusts through the plating within days of humidity, and welds differently. Distinguish them by: (a) the salt-water corrosion test (steel rusts in 24–72 h), (b) abrading the surface (steel shows a gray core under the plating), (c) measuring strip resistance against the \rho math above, or (d) XRF. Buy strips only from known suppliers and verify.

Fusible links. On high-current packs, the series connection of each cell is a deliberately necked-down strip section sized to act as a fuse: the neck cross-section is chosen so the strip melts open at roughly 2× the cell's continuous rating before the cell itself becomes the fuse. Cell-level fusing converts a single-cell internal short from a pack fire into a disconnected cell — cheap insurance at zero extra parts count, applied at the same time as welding.

5. The BMS: Protection and Balancing

The battery management system is a series of FET switches plus a monitoring MCU, and it enforces a hard boundary around every operating parameter:

Balancing corrects the slow drift of series-group voltages. Passive (resistive bleed) balancing shunts the high cells through a resistor — cheap BMSes bleed at 30–60 mA, smart units (JK, JBD/ANT class) at 0.6–2 A. Active balancing moves charge between groups with switching converters at 1–2 A. The sizing equation is simple:

With cells matched to ±1% capacity, a 10 Ah pack starts with ~0.1 Ah of imbalance; a 60 mA passive balancer needs ~1.7 h per 1% of imbalance — fine for nightly charging on a well-matched pack, hopeless for rescuing a badly mismatched one (which is why grading at assembly matters more than balancing ever can).

Topologies. Common-port BMSes share one FET pair for charge and discharge — simpler, cheaper, standard for e-bikes; separate-port units have independent charge/discharge paths and suit solar and regeneration. Wiring order is safety-critical: connect B− first, then balance leads B0 through B13 in ascending order, then B+; never reverse a balance lead, and never connect the pack to load/charger until the BMS is wired — a BMS with reversed balance leads dies instantly on first balance cycle.

Parallel groups need no balancing (cells hard-paralleled equalize by physics); only series groups get balance leads. One balance lead per series node — 14 leads for 13S. Check balance-lead insulation routing for chafe against the nickel strips; a chafed lead is a short waiting on vibration.

6. Charging: CC-CV, Charger Sizing, and the ₹-per-Kilometre Number

Li-ion charges constant-current, constant-voltage (CC-CV). For the 13S4P pack: charge at 0.5 C (5 A, or 1.25 A per cell per the 25R datasheet) until the pack reaches 54.6 V, then hold 54.6 V while current tapers; terminate at C/20–C/25 (0.4–0.5 A). There is no trickle and no float — the charger must terminate, and the BMS provides the second layer if the charger misbehaves. Total charge time at 0.5 C is roughly 3–4 h including taper.

Charger sizing: a 54.6 V 2 A unit (₹800–1,500) charges the pack in ~5–6 h; 5 A units halve that. Charge only within the cell datasheet's temperature window — 0–50 °C for the 25R, narrower in practice (below 5 °C, derate; below 0 °C, never).

The energy cost. Coulombic efficiency at 0.5 C is ~92–96%, so one full charge draws:

At ₹8/kWh that is ₹4.1 per full charge. With real-world Indian e-bike consumption of 15–25 Wh/km (throttle-heavy), the pack delivers ~20–30 km per charge — ₹0.14–0.21 per kilometre, against ₹2–3/km for a petrol two-wheeler. The pack's own amortization at ₹18,000 over 500 cycles adds ~₹1.2/km — still a third of petrol. This is the arithmetic that has converted 2 million+ Indian two-wheelers to electric, and it is worth explaining to every buyer.

7. Safety Engineering: The Thermal Runaway Chain and the Standards

Thermal runaway is not one event but a temperature ladder, each rung exothermic and each feeding the next:

flowchart LR
    A["90–120 °C<br/>SEI decomposition<br/>exothermic"] --> B["≈135 °C<br/>PE separator shutdown<br/>impedance spikes"] --> C["≈165 °C<br/>separator melt<br/>internal short"] --> D["180–230 °C<br/>NMC cathode O₂ release"] --> E["&gt;250 °C<br/>runaway: vent + flame"]

Every rung is design-relevant. The SEI (solid-electrolyte interphase) decomposes exothermically from ~90–120 °C — which is why pack interiors must never reach there. The PE separator's shutdown at ~135 °C (pore collapse raises impedance and stops current) is the last passive defense; by ~165 °C the separator melts outright and the cell internally shorts. NMC cathodes release oxygen as they decompose (~180–230 °C, lower for high-nickel chemistries), and the oxygen burns the carbonate electrolyte — whose dimethyl-carbonate fraction has a flash point of only ~18 °C — which is why a runaway cell vents flame, not smoke, with surface temperatures past 400–500 °C. LFP's olivine cathode holds its oxygen to 250 °C+, which is the entire reason LFP packs are the safer chemistry for e-rickshaws and home storage. Propagation to neighboring cells is conduction-driven; pack design mitigates it with cell spacing, fish-paper/ceramic barriers, and per-cell fusing (Section 4).

The standards that make this testable. Any legitimate cell or pack in India traces to:

A maker's pack should at minimum be built from CRS-registered cells, pass the T1–T5 analogies (drop, vibration, short) by construction, and be shipped never by air, near 50% SoC.

8. Worked Build: A 48 V 10 Ah E-Bike Pack — BOM and Assembly

The complete India-anchored bill of materials for the 13S4P pack:

Item · Spec · Qty · Cost (₹)

Samsung INR18650-25R (genuine) · 2.5 Ah, 20 A · 52 · 250–350/cell → 13,000–18,200

BMS 13S · 30 A, common-port, passive balance · 1 · 800–1,500

Nickel strip 0.15 × 8 mm · pure nickel, verified · ~2 m · 300–500

18650 cell holders · 4×13 or equivalent · 52 · 300–500

Fish paper + Kapton · cell-end insulation · rolls · 200–400

XT90 anti-spark + fuse holder · 40 A ANL · 1 set · 300–500

Silicone wire 10–12 AWG · red + black, 300 mm · 2 · 200–300

54.6 V 2 A charger · CC-CV, terminates · 1 · 900–1,500

PVC shrink tube · pack wrap · 1 · 150–300

Spot welder · transformer class (own) · — · 2,500–6,000 (kWeld class: 15,000–25,000)

Capacity/IR tester · ZB2L3-class · — · 400–800 (better: 5,000–8,000)

Total pack · · · ≈17,500–23,000

Assembly sequence, in order of safety:

  1. Grade every cell — capacity and IR, not labels. Reject anything below ~95% of rated capacity, IR outside ±15% of the lot mean, or any cell that self-discharges visibly in 24 h. Match parallel groups within 1% capacity and a few mΩ.
  2. Arrange in holders with the positive ends insulated — fish-paper rings or Kapton discs. The bare can of a 18650 is the negative terminal; an unwrapped can edge touching a positive cap is a dead short.
  3. Weld series first (each cell's bridge to the next group), then the parallel bus, then fusible-link necks. Verify every weld by pull test. Insulate each welded layer with fish paper before the next layer goes on.
  4. Wire the BMS in order: B− → B0 through B13 balance leads (ascending, verified against a voltmeter as you go) → B+ last. Do not connect load or charger.
  5. Fuse, output leads, anti-spark connector, then shrink-wrap with the BMS outside the wrap's heat trap (BMS FETs need air).
  6. First charge at half rate, supervised, measuring every series group against the BMS reading. A group diverging >50 mV from its neighbours during charge means a bad weld or a bad cell — stop and find it.

9. Testing and Commissioning

Before the pack meets a motor, it meets a test log:

10. The Failure Taxonomy

Ten ways packs actually die, in rough order of Indian frequency:

  1. Counterfeit cells — re-wrapped 2 Ah cells labeled 25R/30Q at 20–30% of genuine price, with IRs of 40–100 mΩ. The pack "works" for 50 cycles and sags horribly the whole time.
  2. High-resistance welds — cosmetic welds add tens of mΩ, local heating, and premature pack sag (Section 2 math).
  3. Dendrite internal shorts — from deep over-discharge below ~2.5 V, followed by recharge; declares itself as a fire days or weeks later.
  4. Nickel-plated steel sold as nickel — 2.3× resistance, rust-through, weld inconsistency (Section 4).
  5. Missing cell-level fusing — a single internally shorted cell takes down its parallel group and its neighbours with it.
  6. Balance-lead chafe — lead insulation rubs through against a strip edge under vibration; the resulting short defeats the BMS's single layer of protection.
  7. Charger mismatch — a 58.8 V (14S) charger on a 13S pack, or a charger that never terminates; the BMS OVP becomes the routine stop mechanism and FETs die.
  8. Charge below 0 °C — lithium plating, permanent capacity loss, latent shorts.
  9. No pack-level fuse — controller or harness short turns the entire pack's 80 A capability into the "fuse."
  10. Unsupervised first charge — the one test where a reversed balance lead, a missed weld, or a counterfeit cell announces itself, sometimes with fire.

11. Sourcing in India: Counterfeits, BIS, and the LFP Economics

The Indian market's defining problem is counterfeit cells: "25R-class" cells at ₹80–150 that are genuine 1–1.5 Ah re-wraps. The defenses are procedural — buy from established electronics distributors, insist on CRS-registered product (IS 16046-2), grade every incoming cell yourself (Section 8), and treat a price 40% below market as proof of fraud rather than a bargain. Genuine commodity cells (25R at ₹250–350, P42A at ₹500–700) cost what they cost precisely because the fakes define the floor.

The economics that justify the effort, at the traction end: a lead-acid e-rickshaw (48 V, 140 Ah) carries ~6.7 kWh nominal, ~50% usable at C-rates, and dies in 12–18 months of daily service at ₹25,000–35,000 per replacement. A 16S LiFePO₄ pack from EVE LF280K-class cells — 51.2 V, 280 Ah, 14.3 kWh, ~90% usable — costs ₹50,000–80,000 in cells plus BMS, but delivers 3,000+ cycles: four to six years of daily service, ~80% lower energy cost per kilometre, and a 200 kg weight cut that the rickshaw repays in range and tyre wear. The same 280 Ah LFP cells, in 8S–16S arrays with a proper BMS, are now the default home-backup upgrade over lead-acid inverters. The cell count changes, the chemistry changes, the BMS label changes — the engineering of Section 2 through Section 9 does not.

Conclusion

A battery pack is the most dangerous component a maker will ever build, and the most mathematically tractable. Four equations size it; the nickel-strip resistance equation sizes the interconnects; the balance-time equation sizes the BMS; and a datasheet plus a CC-CV charger closes the loop. What cannot be shortcut is the discipline: genuine graded cells, pull-tested welds, fusible links, a BMS wired in order, a fuse, a supervised first charge, and a test log. The difference between a pack that runs 800 cycles and one that vents at cycle 47 is rarely the chemistry and almost always the discipline. Whether the next pack out of your workshop is a 13S4P e-bike unit or a 16S LFP e-rickshaw bank, the numbers in this guide are the checklist — and the checklist is what separates a battery from an incident.

More FabFlow blog posts