Battery Management Systems cover
Interactive book · Elective (Semester 6-8)

Battery Management Systems

Watch a scooter pack's cells drift apart, estimate state of charge with a Kalman filter, size a precharge resistor and see thermal runaway start

22 chapters, 22 interactives, few words. A battery management system is the electronics and software that keeps a lithium-ion pack safe, accurate and long-lived. This book follows it from the cell up: chemistries and datasheets, series and parallel packs and the weakest cell, voltage, current and temperature sensing, equivalent circuit models and HPPC tests, open-circuit voltage and hysteresis, heat generation, coulomb counting, Kalman-filter state of charge, state of health and ageing, state of power, passive and active balancing, protection, contactors and precharge, thermal runaway, liquid cooling, fast charging, BMS architecture and CAN, functional safety and Indian standards, pack sizing, second life and recycling. Two illustrative packs run through the book, a 3 kWh scooter pack of NMC cylindrical cells and a 33 kWh car pack of LFP prismatic cells, so every chapter's numbers describe the same batteries.

  • AUAU3002Batteries and Management System
  • VTUBEE657AEnergy Management in Electric Vehicles
  • SPPU302031MJe-Vehicle Technology (Honors in Electric Vehicles)
  • MUHCEV601 / HCEV801EV Drive and Energy Sources; Electric Vehicle System Design (Honours in EVs)
  • KTUPEEET521Energy Storage Systems

By Skill-Lync · Updated 21 Nov 2026 · 22 chapters · 22 interactives · free, no sign-up

Chapter 1

What a battery management system does: measure, estimate, protect, balance, communicate

Which syllabus covers thisAU AU3002 Unit VVTU BEE657A Module 2SPPU 302031MJ Unit 4MU HCEV601 Module 4KTU not in syllabus

What is a battery management system, and what are its jobs?

A battery management system (BMS) is the electronics and software that keep a lithium-ion pack inside its safe limits and tell the vehicle what the pack can do. Its jobs: measure every series cell group's voltage, the pack current and several temperatures; estimate state of charge (SoC, fuel left), state of health (SoH, ageing) and state of power (SoP, how many kW right now); protect by limiting or cutting current; balance the cells; control heating and cooling; communicate over CAN; and log faults. This book's scooter pack, 14 groups in series of 12 cells in parallel, needs 14 voltage channels, one current sensor and at least 4 temperature sensors.

Why does a lithium-ion pack need a BMS more than a lead-acid battery?

A lithium-ion cell has a narrow, hard window. The LG M50 cell in this book is rated 2.5 to 4.2 V, may be charged only between 0 and 45 °C and discharged between −20 and 55 °C. Above the top voltage the cathode is stressed and lithium can plate; below the bottom one the copper foil of the anode starts to corrode once the anode reaches about 3.4-3.5 V vs lithium. Samsung's 50E specification says the system should be equipped with a device to prevent discharging past the cut-off. One cell outside the window can damage the whole pack, so every series group must be watched.

What happens on a ride when one group in the pack is weak?

Series groups carry the same current, so the group with the least charge empties first. In the interactive, a healthy pack on the book's 120 s city ride lasts 178 minutes and delivers 3,044 Wh, about 101 km at an assumed 30 Wh/km. Give one group 85 % capacity and 1.3 times the resistance: the BMS sees its voltage sag, drops to a 30 A limp mode, and stops the ride at 2.5 V after 2,651 Wh, about 88 km. With no BMS the ride goes on until the controller's 42 V cut, and the weak group is driven to 1.17 V, below empty.

A day in a scooter pack's life: what the BMS measures, estimates and decides, and what goes wrong without one: interactive loads as you scroll

What kinds of BMS are there?

By layout: a centralised BMS has one board wired to every cell (cheap and common in scooters); a modular, master-slave BMS puts a small monitoring board on each module and links them to one master over an isolated daisy chain (most cars); a distributed BMS puts electronics on each cell; a wireless BMS replaces the chain with radio. By function, protection comes in layers: firmware decisions, hardware comparators inside the monitoring chip, a fuse, and a contactor or MOSFET that opens the circuit. Chapter 19 compares the architectures with wire counts for this book's two packs.

What did India change after the 2022 electric-scooter fires?

After electric two-wheeler fires in April 2022, the road transport ministry formed an expert committee and amended the battery standards: AIS-156 for two- and three-wheelers and AIS-038 Rev 2 for cars, buses and trucks. The amendments, issued on 29 Aug 2022 and notified as S.O. 4567(E) dated 28.09.2022, add requirements on cells, the on-board charger, pack design, the BMS and thermal propagation from an internal short; they apply from 01.12.2022. AIS-156 now asks for at least 4 temperature sensors and a warning at least 5 minutes before a fire spreads. Alongside, the 2021 PLI scheme (Rs 18,100 crore) backs 50 GWh of domestic cell factories.

Predict, then check

One of the scooter pack's 14 series groups has only 85 % of the others' capacity. With a BMS stopping the ride at 2.5 V, how much of the healthy pack's range do you get?

Worked examples

The scooter pack's nameplate from the cell datasheet · 14 groups in series, each of 12 cells in parallel; cell: 5 Ah, 3.63 V nominal, 4.2 V full, 2.5 V empty (LG M50 class), 69 g (assumed). Energy use 30 Wh/km (assumed)
  1. Cells = 14 × 12 = 168
  2. Voltage: nominal 14 × 3.63 = 50.82 V; full 14 × 4.2 = 58.8 V; empty 14 × 2.5 = 35 V
  3. Capacity = 12 × 5 = 60 Ah (parallel adds capacity)
  4. Energy = 168 × 3.63 V × 5 Ah = 3,049 Wh = 3.05 kWh
  5. Cell mass = 168 × 0.069 = 11.6 kg; nameplate range = 3,049/30 = 102 km

Answer: 50.8 V nominal (35-58.8 V), 60 Ah, 3.05 kWh, 11.6 kg of cells, about 102 km

Counting the charge one city ride takes · The book's ride is a repeating 120 s cycle of pack current: 0 → 55 A at 5 s, 30 A at 15 s, 22 A at 40 s, 60 A at 50 s, 25 A from 58 to 80 s, 0 at 88 s, −18 A (regenerative braking) at 95 s, −8 A at 102 s, 0 from 108 s; straight lines in between
  1. Charge = area under I(t), trapezoids: 137.5 + 425 + 650 + 410 + 340 + 550 + 100 − 63 − 91 − 24 = 2434.5 A·s
  2. Mean current = 2434.5/120 = 20.29 A, about 0.34C of the 60 Ah pack
  3. Per ride = 2434.5/3600 = 0.676 Ah; at 50.8 V that is about 34.4 Wh
  4. Rides in 60 Ah = 60/0.676 = 88.7, i.e. 177 minutes of riding
  5. Peak 60 A = 1C of the pack = 5 A per cell

Answer: 0.676 Ah per 2-minute ride (mean 20.3 A); a full 60 Ah pack lasts about 177 minutes by charge alone (the simulation stops at 178 minutes, at the voltage cut-off)

A weak group with and without a BMS · Same pack and ride, but group 10 has 85 % capacity (51 Ah) and 1.3 × resistance. The BMS stops the ride when any group reaches 2.5 V under load; without a BMS only the motor controller's 42 V cut (assumed) stops it
  1. Series string: every group carries the same amp-hours, so the BMS must stop when group 10 is empty: about 51 of 60 Ah
  2. Simulated: 2,651 Wh vs 3,044 Wh healthy, 87.1 %; 88 km vs 101 km
  3. No BMS: the pack delivers 54.54 Ah before 42 V, so 3.54 Ah is pushed through the empty group
  4. Group 10's SoC = 1 − 54.54/51 = −6.9 %; its voltage under load falls to 1.17 V
  5. The other 13 groups still hold about 9 % each, so the pack voltage hides the damage

Answer: With a BMS: about 88 km and no damage. Without: 94 km, but group 10 is over-discharged to 1.2 V (−6.9 % SoC)

Chapter 2

Inside a lithium-ion cell: NMC, LFP, NCA, LTO and sodium-ion

Which syllabus covers thisAU AU3002 Unit IVTU BEE657A Module 2SPPU 302031MJ Unit 3MU HCEV601 Module 4KTU PEEET521 Module 2

What is inside a lithium-ion cell, and where does its voltage come from?

Four parts: a negative electrode (anode, usually graphite on copper foil), a positive electrode (cathode: a metal oxide or phosphate on aluminium foil), a porous separator between them, and a liquid electrolyte that carries lithium ions. On discharge, lithium leaves the graphite and slots into the cathode's crystal (intercalation: ions parked between atomic layers, not a chemical rebuild), while electrons go round the external circuit. Each electrode has its own potential against lithium metal; the cell voltage is the difference. For the LG M50 at 50 % SoC the NMC cathode sits at 3.842 V and the graphite at 0.133 V, so U = 3.708 V.

NMC or LFP: why does the shape of the voltage curve matter to a BMS?

A BMS can read SoC from a rested cell's voltage only if the voltage changes with SoC. At 50 % the NMC curve rises 8.96 mV per 1 % SoC; the LFP curve only 0.49 mV, a flat plateau. A 10 mV measurement error is therefore 1.1 % of SoC on NMC but 21 % on LFP, so LFP packs lean on coulomb counting and resets at full charge. LFP pays back in other ways: 2,000 or more cycles against 1,000-2,000, and packs at 81 USD/kWh against 128 (BloombergNEF 2025), but only 109 Wh/kg against 266.

Cell chemistries side by side: NMC, LFP, LTO, NCA and sodium-ion voltage windows, energy, life and safety: interactive loads as you scroll

How do NCA, LTO and sodium-ion cells differ from NMC and LFP?

NCA (nickel cobalt aluminium oxide) is NMC's close cousin: same 4.2 V top and high energy (Panasonic NCR18650B: 243 Wh/kg), but Battery University lists 500 cycles and runaway at 150 °C against NMC's 210 °C; calorimeter tests on 18650 cells put runaway at about 147 °C for NCM811 and 165 °C for NCA. LTO swaps graphite for lithium titanate: only 2.3 V and 89-106 Wh/kg, but Toshiba quotes over 70 % capacity after 20,000 cycles and discharge down to −30 °C. Sodium-ion uses sodium instead of lithium: about 3.0 V nominal in a 2.0-4.0 V window and 105-160 Wh/kg, with CATL claiming over 90 % capacity at −20 °C.

What is a cell's safe operating area?

It is the box of voltage, current and temperature inside which the maker guarantees the cell. For the LG M50: 2.5-4.2 V; charge at 0-45 °C, at most 0.3C below 25 °C and 0.7C above; discharge at −20 to 55 °C, at most 0.5C below 10 °C and 1.5C above 25 °C. Charging below 0 °C plates lithium metal on the graphite; hot cells age fast and approach runaway. The BMS turns this box into thresholds with margins, and into current limits that shrink near the edges, which is exactly what the scooter's limp mode does.

Where do other batteries fit: lead-acid, NiMH, sodium-sulfur, metal-air and solid-state?

Lead-acid: 2.1 V per cell and 35-40 Wh/kg, cheap, still used for 12 V starting and inverters. Nickel-metal hydride: 1.2 V and 46-120 Wh/kg, the classic hybrid-car battery. Sodium-sulfur runs hot at 300-350 °C, so it suits grid storage. Metal-air cells breathe oxygen: zinc-air gives about 470 Wh/kg in practice at 1.45 V but is hard to recharge. Solid-state cells replace the liquid electrolyte with a solid one, aiming at 250-500 Wh/kg and less fire risk; they are still early.

Predict, then check

The BMS reads a rested cell's voltage with a 10 mV error at 50 % SoC. How far off can the SoC read from voltage be?

Worked examples

Cell voltage from the two electrode potentials · LG M50 (Chen 2020 fits): at SoC z the graphite is at x = 0.0279 + 0.8735 z and the NMC cathode at y = 0.9077 − 0.6416 z. Find the open-circuit voltage at 50 % SoC
  1. x = 0.0279 + 0.8735 × 0.5 = 0.4646; y = 0.9077 − 0.6416 × 0.5 = 0.5869
  2. U− = U_graphite(0.4646) = 0.1333 V (vs Li/Li+)
  3. U+ = U_NMC(0.5869) = 3.8416 V
  4. U = U+ − U− = 3.8416 − 0.1333 = 3.7082 V
  5. Same for the LFP cell at 50 %: 3.4005 − 0.1345 = 3.266 V

Answer: 3.708 V for NMC (3.266 V for LFP): the cathode sets most of the voltage, the graphite takes about 133 mV off

SoC error from a 10 mV voltage error · A BMS measures a rested cell's voltage 10 mV high at 50 % SoC. Use the slope of the OCV curve (central difference over ±0.01 %)
  1. NMC slope = 8.96 mV per % SoC; LFP slope = 0.487 mV per %
  2. NMC: ΔSoC = 10/8.96 = 1.12 %
  3. Check by solving OCV(z) = OCV(50 %) + 10 mV exactly: z = 51.09 %, an error of 1.09 %
  4. LFP: ΔSoC = 10/0.487 = 20.5 %

Answer: About 1.1 % on NMC but about 21 % on LFP: voltage alone cannot tell where an LFP cell is on its plateau

Mass and price of a 3 kWh pack in each chemistry · The scooter needs 3,049 Wh of cells. Cell specific energy from datasheets: NMC (LG M50) 18.2 Wh/68.5 g; LFP (A123 26650) 3.3 V × 2.5 Ah/76 g; LTO (Toshiba 20 Ah) 89 Wh/kg. Pack prices: BloombergNEF 2025 averages, NMC 128 and LFP 81 USD/kWh
  1. NMC: 18.2/0.0685 = 265.7 Wh/kg → 3,049/265.7 = 11.5 kg
  2. LFP: 8.25/0.076 = 108.6 Wh/kg → 28.1 kg
  3. LTO: 34.3 kg
  4. Price at survey averages: NMC 3.05 × 128 = 390 USD; LFP 3.05 × 81 = 247 USD

Answer: Cells alone: 11.5 kg NMC, 28.1 kg LFP, 34.3 kg LTO; LFP costs about 37 % less per kWh but weighs 2.4 times as much

Chapter 3

Reading a cell datasheet: capacity, energy, C-rate, cut-off voltages and temperature limits

Which syllabus covers thisAU AU3002 Unit IVTU not in syllabusSPPU 302031MJ Unit 3MU HCEV601 Module 4KTU PEEET521 Module 2

What do rated, nominal and minimum capacity mean, and what is a C-rate?

Capacity is the charge a cell delivers from full to its cut-off voltage under stated conditions. Samsung's 50E sheet gives a minimum standard capacity of 4,900 mAh at 0.2C and a minimum rated capacity of 4,753 mAh at 1C: same cell, faster discharge, 97 %. Nominal (or typical) is what most cells give; minimum is what every cell must give. LG rates the M50 in energy: 18.2 Wh nominal, 17.6 Wh minimum. A C-rate scales current to capacity: 1C empties the cell in about an hour, so for the M50 (1C = 4.85 A) 0.2C is 0.97 A and 2C is 9.7 A.

How do you read a real datasheet such as the Panasonic NCR18650B?

Go line by line and turn each line into a BMS setting. The NCR18650B (an NCA cell): rated minimum 3,200 mAh, typical 3,350 mAh, 3.6 V nominal, so about 12.06 Wh. Charge: CC-CV at 1,625 mA (0.49C) to 4.2 V, about 4 hours. Temperature: charge 0 to +45 °C, and only 0.25C (838 mA) below 10 °C; discharge −20 to +60 °C. Cut-off 2.5 V (in its charts). Mass 48.5 g max, 243 Wh/kg and 676 Wh/L. Every number becomes a threshold, a current limit or a temperature window in firmware.

Why does a cell give less capacity at high current and in the cold?

Internal resistance drops voltage by I × R, so the cut-off arrives earlier while charge is still inside: the lithium is not lost, it is stranded. Resistance rises in the cold, so the effect grows. The 50E sheet: 0.2C 100 %, 1C 97 %, 2C 95 %, and at −10 °C and 1C only 70 %. This book's simple 1RC model gives 99.2 %, 97.9 % and 96.2 %: right at room temperature, far too hopeful in the cold. Fitting the resistance's temperature dependence to the −10 °C point needs an activation energy of 35.7 kJ/mol instead of 20; even then the model has no diffusion limit, which matters most at high C.

Reading a cell datasheet: discharge curves at different C-rates and temperatures, and the usable capacity and energy: interactive loads as you scroll

How do you get the energy from a discharge curve, and what do cut-off voltages do?

Energy is the area under voltage against charge delivered, E = ∫V dQ, so watt-hours equal amp-hours times the mean voltage. In the model the M50 at 1C and 25 °C gives 4.955 Ah at a mean 3.545 V, 17.57 Wh, against LG's 18.2 Wh nominal at 0.2C; 0.79 Wh more turns into heat inside the cell. The cut-off voltage is where the sheet stops counting. Raising it in firmware, say from 2.5 to 3.0 V, gives up a little capacity at the steep bottom of the curve in exchange for less stress and a margin for the weakest cell.

What are coulombic and energy efficiency?

Coulombic efficiency is amp-hours out divided by amp-hours in; for a good lithium-ion cell it is very close to 1, which is why coulomb counting works. Energy efficiency is watt-hours out divided by watt-hours in, and it is lower because you charge at a higher voltage than you discharge: the difference is I²R heat. In this book's model, charging the M50 from empty to full at 0.5C takes 18.87 Wh and discharging at 0.5C returns 18.02 Wh: 95.5 % (coulombic efficiency is 100 % by construction). Wikipedia lists 80-90 % for lithium-ion, 66-92 % for NiMH and 50-95 % for lead-acid.

What is a Ragone plot?

It plots specific energy (Wh/kg, how far) against specific power (W/kg, how fast), usually on log axes. Each storage technology is a band: batteries sit at high energy and moderate power, supercapacitors at about 0.5-15 Wh/kg but very high power for seconds. One cell traces a curve: in this book's model the M50 gives 265 Wh/kg at 53 W/kg (0.2C) but 241 Wh/kg at 493 W/kg (2C). LG's 80 W, 10 s pulse rating is 1,168 W/kg. Lead-acid sits near 40 Wh/kg and 180 W/kg.

Predict, then check

Samsung's 50E delivers 4,900 mAh at 0.2C and room temperature. At −10 °C and 1C its datasheet promises about:

Worked examples

Turning the LG M50 current table into scooter limits · LG M50 sheet: 1C = 4.85 A; standard charge 0.3C; maximum continuous discharge 0.5C at −20 to 10 °C, 3.0C at 10 to 25 °C, 1.5C at 25 to 55 °C. Scooter pack: 12 cells per group, ride peak 60 A
  1. Standard charge current = 0.3 × 4.85 = 1.455 A per cell = 17.46 A for the pack
  2. Warm (25-55 °C) limit = 1.5 × 4.85 = 7.275 A per cell = 87.3 A for the pack
  3. Ride peak per cell = 60/12 = 5 A = 1.03C: inside the warm limit
  4. Cold (below 10 °C) limit = 0.5 × 4.85 = 2.425 A per cell = 29.1 A for the pack

Answer: Charge at 17.5 A; discharge up to 87 A when warm, but only 29.1 A below 10 °C: on a cold morning the BMS must halve the ride's 60 A peak

Energy from points read off a discharge curve · Voltages read from the model's 1C, 25 °C curve of the 5 Ah NMC cell every 0.5 Ah, plus the end point at the 2.5 V cut-off: 0 Ah 4.085 V, 1.5 Ah 3.769 V, 3 Ah 3.484 V, 4.5 Ah 3.089 V (every third point shown)
  1. Trapezoids: E = Σ (V_i + V_(i+1))/2 × ΔQ over 10 strips
  2. E ≈ 17.534 Wh (the fine-step model integral is 17.566 Wh)
  3. Mean voltage = E/Q = 17.566/4.955 = 3.545 V
  4. Against the sheet's 18.2 Wh nominal (at 0.2C): 96.5 %

Answer: About 17.57 Wh at a mean 3.55 V; 0.5 Ah strips already get within 0.18 %

A cold Delhi morning: how much range is left? · The scooter pack (12P of 50E-class cells, 4,900 mAh standard) at −10 °C, ridden at about 1C. The 50E table gives 70 % capacity at −10 °C and 1C. Warm range in the bmsday model: 101 km
  1. Pack capacity at −10 °C ≈ 12 × 4.9 × 0.70 = 41.2 Ah (instead of 58.8 Ah)
  2. Range scales at least with capacity: 101 × 0.70 = 71 km
  3. It is less still, because the mean voltage is lower in the cold, so each amp-hour carries fewer watt-hours
  4. The LG limit of 0.5C below 10 °C also caps the pack at 29.1 A, so the BMS must derate power

Answer: About 41 Ah and at most about 71 km instead of 101 km; the charge is not lost, it returns when the pack warms

Chapter 4

Series, parallel and the weakest cell: building a pack from cells

Which syllabus covers thisAU AU3002 Unit IIVTU not in syllabusSPPU 302031MJ Unit 1MU HCEV601 Module 4KTU not in syllabus

What do series and parallel connections do to a pack?

Cells in series add their voltages; cells in parallel add their capacities and share the current. The book's scooter pack is 14S12P: 14 groups in series, each of 12 cells in parallel, so 50.8 V nominal, 60 Ah and 3.05 kWh from 168 cells. Resistance follows the same rules: 14 × 18 mΩ / 12 = 21 mΩ, so the ride's 60 A peak wastes 76 W as heat before counting interconnects. The book's car pack is 104S1P of 100 Ah LFP cells: 332.8 V and 33.3 kWh. The BMS watches each series group, because cells in parallel always share one voltage.

Why does the weakest group set the whole pack's capacity?

Every series group carries the same current, so all of them give the same amp-hours, and the string must stop when the group with the least charge reaches its cut-off. No two cells are identical: a study of 1,100 fresh cells from one production found 0.28 % spread in capacity and 0.72 % in impedance; mixed batches and aged cells spread far more. Parallel cells average out: with 1.5 % cell spread, a 12-cell group varies only 0.43 %, and the weakest of 14 groups typically sits 0.74 % below average. In the interactive you can widen the spread and watch the pack's amp-hours follow its weakest row.

Building a pack from cells: series and parallel, cell-to-cell spread, and why the weakest group sets the range: interactive loads as you scroll

What are cell matching and interconnect resistance, and why do they matter?

Matching (grading or binning) means measuring every cell's capacity and resistance and building each pack from one narrow bin; in the interactive it cuts the spread to a third. Interconnects matter as much: the strips that join parallel cells have resistance. Two layers of 0.2 × 8 mm nickel strip over a 22 mm cell pitch are 0.48 mΩ per step, small next to a cell's 18 mΩ, yet if the load is taken from one end of a 12-cell group the nearest cell carries 7.1 times the current of the farthest. Taking positive and negative from opposite ends, or using copper bars, evens it out. Cell-to-pack designs skip modules and mount cells straight into the pack.

How is a battery pack designed, step by step?

Start from requirements: energy (range × Wh/km), peak and continuous power, system voltage, space, mass, life and cost. Choose the cell (chemistry and format). Pick S from the voltage and P from the larger of the energy and power needs (chapter 21 sizes this book's packs). Then the mechanical and thermal design: holders, busbars, cooling path, enclosure. Then the BMS: channels, sensors, fuse and contactors. Then build and test. In the vehicle the battery compartment must survive a crash without crushing cells, and must give vent gases a path away from the rider: AIS-156 asks for a pressure vent and spacing between cells.

What are formation and pack testing?

Formation happens at the cell factory: the first slow charge and discharge cycles grow the protective film (the solid electrolyte interphase) on the graphite, then cells are aged on a shelf and sorted by capacity, resistance and self-discharge. A pack is tested again at the end of the line: insulation resistance, every group's voltage and the spread between them, resistance under a pulse, BMS functions (over-voltage, under-voltage, over-current, over-temperature and short-circuit protection) and leak tightness. For type approval in India, AIS-156 adds water immersion to IPX7 at full charge and a thermal propagation test from one cell's internal short.

Predict, then check

Cells with 2 % capacity spread go into two packs: 14S1P (one cell per group) and 14S12P. Which loses the larger share of its capacity to the weakest group?

Worked examples

Pack resistance and heat at the ride's peak · 14S12P of 5 Ah cells, each 18 mΩ (assumed typical at 25 °C, 50 % SoC); interconnects ignored. Peak current 60 A
  1. One group: 18/12 = 1.5 mΩ (12 equal resistors in parallel)
  2. Pack: 14 × 1.5 = 21 mΩ
  3. Voltage drop at 60 A: 60 × 0.021 = 1.26 V out of 50.8 V
  4. Heat: 60² × 0.021 = 75.6 W, i.e. 0.45 W per cell (5 A × 5 A × 0.018 Ω = 0.45 W)

Answer: 21 mΩ, a 1.26 V drop and 76 W of heat at 60 A (0.45 W in each cell)

Finding the weakest group from end-of-line test data · Measured capacities of the 14 groups of one scooter pack (Ah): 60.4, 59.9, 60.2, 59.6, 60.8, 60.1, 59.3, 60.5, 60, 59.8, 60.3, 60.6, 59.7, 60.2
  1. Weakest group = group 7 at 59.3 Ah; mean of the 14 = 60.1 Ah
  2. Pack capacity = the weakest group = 59.3 Ah
  3. Lost to spread = (60.1 − 59.3)/60.1 = 1.33 %
  4. Energy lost ≈ 0.8 Ah × 50.8 V = 41 Wh, about 1.4 km at 30 Wh/km

Answer: 59.3 Ah (group 7); 1.33 % of the average group, about 41 Wh: balancing (chapter 13) cannot fix a capacity shortfall, only a charge mismatch

Current sharing in a 3-cell parallel group, with and without strip resistance · Three cells, same EMF 3.6 V, resistances 18, 20 and 24 mΩ, group current 15 A. Then the same group with 1 mΩ of strip between neighbouring cells on each bus, load taken from cell 1's end (a ladder of 2 mΩ steps)
  1. Ideal busbars: I ∝ 1/R. ΣG = 1/0.018 + 1/0.020 + 1/0.024 = 147.22 S; common drop = 15/147.22 = 101.89 mV
  2. I1 = 5.66 A, I2 = 5.09 A, I3 = 4.25 A (sum 15 A)
  3. With strips: nodal equations (E − Vk)/Rk + (neighbour − Vk)/0.002 = 0, with 15 A leaving node 1; solved: V1 = 3.4853 V, V2 = 3.5025 V, V3 = 3.51 V
  4. I1 = 6.37 A, I2 = 4.87 A, I3 = 3.75 A (sum 15 A)

Answer: Ideal: 5.66 / 5.09 / 4.25 A; with 1 mΩ strips the cell next to the load takes 6.37 A and the far one only 3.75 A (1.7 ×)

Chapter 5

Measuring cells: analogue front-ends, ADC resolution, current sensors and thermistors

Which syllabus covers thisAU AU3002 Unit VVTU not in syllabusSPPU 302031MJ Unit 4MU not in syllabusKTU not in syllabus

What does a cell-monitoring IC (the analogue front-end) do?

It measures every series cell through a multiplexer and an ADC, reads the thermistors, switches the balancing resistors and flags over- or under-voltage in hardware, then reports to the BMS microcontroller. One IC covers a few cells: TI's bq76PL536A-Q1 3 to 6 cells (14-bit, ±2.5 mV), the BQ76952 up to 16 (±10 mV at 0-60 °C), NXP's MC33771C 14 (0.8 mV) and ADI's LTC6813-1 18 (2.2 mV). The scooter's 14 series groups fit one chip. The car's 104 cells need 8 MC33771Cs or 6 LTC6813-1s linked in an isolated daisy chain, each chip talking to the next over a twisted pair.

How does a few millivolts of error become a state-of-charge error?

The BMS turns a rested voltage into SoC through the OCV curve, so a voltage error ΔV becomes ΔSoC = ΔV divided by the curve's slope. At 50 % the NMC cell's slope is 8.96 mV per %, the LFP cell's only 0.49 mV per %. Add the LTC6813-1's 2.2 mV, half a 16-bit LSB (38.1 µV) and a 1 °C temperature error in the IR correction: NMC reads within 0.52 %, LFP within 9.7 %. Same chip, 18 times worse. On LFP the voltage cannot pin SoC in the middle; the BMS leans on counting.

Measuring cells: how a millivolt of error or an amp of sensor offset becomes a state-of-charge error: interactive loads as you scroll

Shunt, Hall or fluxgate: how does a BMS measure pack current and voltage?

A shunt is a tiny precise resistor in the current path; the BMS reads its millivolts. It is accurate, but it dissipates I²R and needs isolation on the high side. A Hall sensor clamps around the busbar and senses the magnetic field: isolated and lossless but with more offset. A fluxgate is the low-offset magnetic option. Offset matters because it is counted all the time: LEM's DHAB S/133 Hall sensor allows ±0.23 A, which is 9.2 % of the 60 Ah scooter pack per day of counting. An IVT-S shunt module allows 25 mA, 1 % per day. The same module measures pack voltage up to 1,000 V to ±0.5 %, ±1.66 V on the car pack.

Where do the temperature sensors go, and how accurate must they be?

Most packs use NTC thermistors, whose resistance falls about 4.5 % per °C near 25 °C; a 10 kΩ part with B = 3977 K reads 4,372 Ω at 45 °C. They sit on cell cans, busbars and the hottest spot, never inside a cell, so they read the surface. AIS-156 asks for at least 4 sensors in a two-wheeler pack and a gradual cut of traction above 60 °C. Accuracy also matters for estimation: cell resistance changes 2.7 % per °C, so a 1 °C error shifts the BMS's IR correction by 2.4 mV at 1 C on the scooter cell.

Why must voltage and current be sampled at the same instant?

The BMS compares voltage with current to correct for IR drop and to estimate resistance. If the samples are taken at different moments during a change, the pair is wrong. Suppose the scooter pulls 60 A in a step and the voltage was sampled just before the current: the pack's 21 mΩ makes a 1.26 V mismatch, 90 mV per cell, which looks like 12 % of SoC. AFE chips therefore measure all cells within microseconds (all 18 cells of an LTC6813-1 in 290 µs) and offer pins to trigger voltage and current conversion together.

Predict, then check

The same AFE (2.2 mV error) measures an NMC cell and an LFP cell, both resting at 50 %. Whose voltage-based SoC is worse?

Worked examples

From an AFE's datasheet accuracy to a SoC error · An LTC6813-1 (2.2 mV maximum error, 16-bit ADC over 0-5 V) measures a resting cell at 50 % SoC. Compare the book's NMC cell and LFP cell, and a 12-bit ADC on NMC
  1. 16-bit LSB = 5/65,536 = 76.3 µV; half an LSB = 38.1 µV. Total ΔV = 2.2 + 0.038 = 2.238 mV
  2. NMC slope at 50 % = 8.96 mV/%: ΔSoC = 2.238/8.96 = 0.252 %
  3. LFP slope at 50 % = 0.487 mV/%: ΔSoC ≈ 7.2 % if the reading is high (read through the curve, which bends; about 3.4 % if it is low)
  4. 12-bit ADC: LSB = 5/4,096 = 1.221 mV, ΔV = 2.2 + 0.61 = 2.81 mV: NMC ΔSoC = 0.316 %

Answer: NMC ±0.25 %, LFP up to 7.2 %; resolution matters less than the curve's slope

Coulomb-count drift from a Hall sensor's offset · The scooter's 60 Ah pack uses a Hall sensor with ±0.23 A offset (LEM DHAB S/133) and ±1 % sensitivity error. The BMS is awake 6 h a day and the pack moves 40 Ah a day. No full charge for a week
  1. Offset charge per day = 0.23 A × 6 h = 1.38 Ah = 2.3 % of 60 Ah
  2. After 7 days without a reset: 7 × 1.38 = 9.66 Ah = 16.1 %
  3. Sensitivity error: 1 % of 40 Ah = 0.4 Ah per day, but it partly cancels between charge and discharge
  4. A 25 mA shunt module instead: 0.025 × 6 = 0.15 Ah per day, 0.25 %

Answer: About 16 % of SoC error in a week from the Hall offset alone; a shunt cuts it to 1.8 %

How finely an NTC divider reads 45 °C · A 10 kΩ NTC (Vishay NTCLE100E3103: 4,372 Ω at 45 °C, B = 3977 K) sits under a 10 kΩ pull-up from 3.3 V, read by a 12-bit ADC on 3.3 V
  1. V(45 °C) = 3.3 × 4,372/(10,000 + 4,372) = 1.0039 V
  2. At 46 °C the beta law gives R = 4,204 Ω, so V = 0.9767 V
  3. Sensitivity = 27.15 mV per °C; ADC LSB = 3.3/4,096 = 0.806 mV
  4. Resolution = 0.806/27.15 = 0.03 °C per count

Answer: About 0.03 °C per count at 45 °C; the part's tolerance and where it sits on the cell matter far more

Chapter 6

Equivalent circuit models and identifying R0, R1 and C1 from an HPPC pulse

Which syllabus covers thisAU AU3002 Unit IIIVTU not in syllabusSPPU not in syllabusMU not in syllabusKTU not in syllabus

Rint, Thevenin or PNGV: which circuit stands in for a cell?

The Rint model is an OCV source and one resistor: V = OCV − I R0. It gets the instant drop right but not the slow sag. The Thevenin model adds a resistor R1 in parallel with a capacitor C1, a voltage that builds over tens of seconds. The PNGV model adds a big series capacitor for the OCV falling as charge leaves: Cb = Q ÷ (dOCV/dSoC), 20.1 kF for the 5 Ah NMC cell at 50 %. A second RC pair (2RC) covers the slow diffusion part. BMS code usually runs 1RC or 2RC with OCV from a table. The Electric Vehicles book has a Thevenin interactive.

What is an HPPC test?

Hybrid pulse power characterisation, as Idaho National Laboratory's manuals define it: a 10 s discharge pulse, 40 s rest, then a 10 s charge pulse at 75 % of the current. It is repeated every 10 % of depth of discharge, after a 1 h rest each time. The instant voltage step gives R0; the sag during the pulse and the recovery after it give R1 and τ = R1·C1. On the book's NMC cell at 50 % and 25 °C, a 5 A pulse drops 90 mV at once, so R0 = 18 mΩ. ΔV/ΔI after 10 s gives DCIR10 = 21.9 mΩ.

An HPPC pulse test: reading R0, R1 and C1 of a cell's equivalent circuit off its voltage response: interactive loads as you scroll

How accurate is reading R0, R1 and τ by hand?

R0 by hand is good to a fraction of a percent. τ is not: the 40 s rest is barely more than one time constant, and taking its last point as the final value moves the 63 % point early. On the model cell (τ = 30 s) the hand reading gives τ = 18.9 s and R1 = 6.06 instead of 12 mΩ. A 120 s rest would give 29.2 s. A least-squares fit of a + b·e^(−t/τ) to the whole relaxation recovers 30 s exactly on clean data. As a check, the model's 30 s DC resistance is 25.6 mΩ; LG's datasheet gives 30 ± 6 mΩ. Our R values are assumed, not measured.

What does the 1RC model look like in a BMS's code?

As a state-space model with state x = [SoC, v1] and input I. Each sample of Δt seconds: SoC drops by I·Δt/(3600 Q); v1 becomes a·v1 + R1(1 − a)·I with a = e^(−Δt/τ). For Δt = 1 s and τ = 30 s, a = 0.9672. The output is V = OCV(SoC) − I·R0 − v1. R0, R1 and τ come from look-up tables in SoC and temperature: R0 rises from 18 to 37.7 mΩ at 0 °C. Engineers build and fit these models in MATLAB/Simulink (Simscape Battery) or in the open-source Python package PyBaMM, which supplies this book's OCV curves.

Where do electrochemical and neural-network models fit in?

Electrochemical models track lithium inside the particles and the electrolyte. The full Doyle-Fuller-Newman model and its simpler single-particle version predict diffusion limits, plating risk and ageing. They need dozens of measured parameters (thicknesses, diffusivities, reaction rates) and far more computation, so they live in design tools more than in BMS chips. Neural-network models learn voltage from current, temperature and history using lots of logged data. They can be accurate inside the conditions they saw and wrong outside them. An ECM is the practical middle: a few parameters per cell, cheap enough to run every 100 ms.

Predict, then check

In the HPPC interactive, the hand-read τ comes out low. If the rest after the pulse were 120 s instead of 40 s, the hand-read τ would:

Worked examples

R0, DCIR and R1 from an HPPC pulse on the car cell · The book's 100 Ah LFP cell at 50 % and 25 °C takes a 100 A, 10 s discharge pulse. Voltage just before: 3.266 V; 0.1 s into the pulse: 3.231 V; at 10 s: 3.2254 V. τ is 40 s (from the relaxation)
  1. R0 = (3.266 − 3.231)/100 = 0.35 mΩ (the 0.1 s step is almost all R0)
  2. DCIR10 = (3.266 − 3.2254)/100 = 0.406 mΩ
  3. The OCV also fell as 1,000 As left: 0.138 mV, worth 0.0014 mΩ
  4. R1 = (DCIR10 − R0 − OCV share)/(1 − e^(−10/40)) = (0.406 − 0.35 − 0.0014)/0.2212 = 0.247 mΩ

Answer: R0 ≈ 0.35 mΩ, R1 ≈ 0.25 mΩ, C1 = 40/R1 ≈ 162 kF; readings rounded to 0.1 mV limit the accuracy

τ, R1 and C1 from three relaxation readings · After a 5 A, 10 s pulse on the NMC cell, the resting voltage reads 3.69138 V at 5 s, 3.69838 V at 25 s and 3.70198 V at 45 s after the current stops
  1. For V = a − b e^(−t/τ) with equal 20 s gaps: e^(−20/τ) = (V3 − V2)/(V2 − V1) = 3.6/7 = 0.5143
  2. τ = −20/ln(0.5143) = 30.08 s (no final value needed)
  3. Height at the end of the pulse b = (V2 − V1)/(e^(−5/τ) − e^(−25/τ)) = 17.02 mV
  4. R1 = b/(I(1 − e^(−10/τ))) = 17.02/(5 × 0.2829) = 12.03 mΩ; C1 = τ/R1 = 2,500 F

Answer: τ = 30.1 s, R1 = 12 mΩ, C1 = 2.5 kF

One step of the state-space model · NMC cell at SoC 0.80, v1 = 20 mV, 25 °C, current 5 A, Δt = 1 s. R0 = 18 mΩ and R1 = 12 mΩ at 50 % scale by 1 + 0.2(2·SoC − 1)² (assumed model)
  1. Scale factor at 0.8: 1 + 0.2 × 0.6² = 1.072: R0 = 19.3 mΩ, R1 = 12.86 mΩ
  2. SoC: 0.8 − 5 × 1/(3600 × 5) = 0.799722
  3. v1: a = e^(−1/30) = 0.9672; v1 = 0.9672 × 20 + 12.86 × 0.0328 × 5 = 21.453 mV
  4. V = OCV(0.7997) − 5 × 19.3 mΩ − 21.45 mV = 4.0215 − 0.0965 − 0.0215 = 3.9035 V

Answer: SoC 0.79972, v1 21.45 mV, V = 3.904 V

Chapter 7

Open-circuit voltage, relaxation and LFP hysteresis

Which syllabus covers thisAU AU3002 Unit IVVTU not in syllabusSPPU not in syllabusMU HCEV601 Module 4KTU not in syllabus

What is open-circuit voltage, and how is the OCV curve measured?

OCV is the cell's voltage at rest, when no current has flowed for long enough that nothing inside is still settling. It depends almost only on SoC and is the BMS's map from voltage to charge. Labs measure it two ways. One is a very slow charge and discharge (C/20 to C/50), so losses are tiny. The other is GITT: short pulses, each followed by a long rest; one study used 150-minute pulses at C/50 and 16-hour rests. The book's NMC cell runs from 2.498 V empty to 4.193 V full. The LFP cell stays between 3.168 and 3.316 V from 20 % to 95 %.

Why can't a BMS read SoC from the voltage right after a drive?

Under load the RC voltages build up, and after the load stops they decay slowly. A fast part (τ about 30 s) is gone in minutes. A slow diffusion part, with τ assumed 1,800 s and the M50 graphite's own time scale at 1600 to 5900 s, takes much longer. After 30 minutes at 1 C, the NMC cell at 45 % reads 43 % after a 10-minute rest and needs 31 minutes to come within 1 %. A one-RC model alone would claim 65 s. That is why test manuals rest cells for an hour before recording OCV.

Open-circuit voltage: relaxation after a load, the flat LFP plateau and hysteresis: interactive loads as you scroll

What makes LFP hard: the plateau and hysteresis

Between 20 % and 95 % the LFP cell's OCV moves only 148 mV, about 0.49 mV per % at 50 %. LFP also shows hysteresis: after charging, the rested voltage sits higher than after discharging at the same SoC. Barai and co-workers measured up to 38 mV in an LFP cell. With an assumed 20 mV gap, the voltage that means 60 % after a discharge means 39 % after a charge. The BMS must remember which way the cell last went, or it can be wrong by tens of percent.

Load voltage, rested EMF or AC impedance: which one tells SoC?

Load voltage: add I·R0 back to the measured voltage and look up the OCV curve. It is instant but leans on the model; a 10 % error in R0 at 1 C is 1 % of SoC on NMC. Rested EMF (the OCV after a rest) is the most accurate reading on NMC, but only after half an hour or more, and weak on the LFP plateau. AC impedance at 1 kHz mostly measures R0, which in this book's model changes 20 % across SoC but 2.5 times between 25 °C and −5 °C. Temperature swamps it, so impedance suits state of health better than SoC.

Does temperature move the OCV curve?

Slightly. The entropic coefficient dU/dT of the LG M50 (O'Regan et al. 2022) is 0.015 mV/K at 50 % SoC but −0.36 mV/K at 10 %. A 20 °C change moves the OCV 0.3 mV at 50 % and 7.3 mV at 10 %, about 0.2 % of SoC there. So OCV tables are often stored at a few temperatures. Far bigger is the effect of temperature on resistance and on how long the cell takes to relax: the cold slows everything.

Predict, then check

After 30 minutes at 1 C, the BMS reads SoC from the voltage after only a 2-minute rest. Compared with the true 45 %, it reads:

Worked examples

How long to rest before trusting the voltage · After 30 min at 1 C the NMC cell is at 45 % with fast RC voltage v1 = 60.1 mV (τ1 = 30 s) and slow v2 = 19.65 mV (τ2 = 1,800 s, assumed). 1 % of SoC near 45 % is 6.98 mV
  1. One RC only: t = τ1 ln(v1/ΔV) = 30 × ln(60.1/6.98) = 65 s
  2. Slow branch: t = τ2 ln(v2/ΔV) = 1,800 × ln(19.65/6.98) = 1,864 s
  3. Both together (solved numerically): 1,864 s = 31.1 min, set almost entirely by the slow part

Answer: About 31 minutes, not the 65 s a one-RC model predicts

Two SoCs for one voltage on the car pack · The car pack's LFP cells rest at an average 3.264 V. The charge and discharge OCV branches are 20 mV apart (assumed; up to 38 mV measured)
  1. If the last move was a discharge, read the lower branch: SoC = 70 %
  2. If the last move was a charge, read the upper branch: SoC = 40.5 %
  3. Difference 29.4 % of 33.3 kWh = 9.8 kWh of range uncertainty

Answer: 41 % or 70 % depending on history: the BMS must track the direction

SoC from a voltage under load, warm and cold · An NMC cell reads 3.620 V while giving 2.0 A. R0 = 18 mΩ at 25 °C; in the cold (5 °C) the true R0 is higher by the Arrhenius factor (Ea 20 kJ/mol, assumed)
  1. Warm: OCV ≈ 3.620 + 2.0 × 0.018 = 3.656 V, SoC from the curve = 43.2 %
  2. Cold factor: exp(20,000/8.314 × (1/278.15 − 1/298.15)) = 1.786: R0 = 32.2 mΩ
  3. Cold: OCV ≈ 3.620 + 2.0 × 0.0322 = 3.6843 V, SoC = 47.1 %
  4. Using the warm R0 in the cold would read 4 % low

Answer: 43 % warm vs 47 % cold: the IR correction needs the right temperature

Chapter 8

Heat in a cell: Joule and entropic heat and the lumped thermal model

Which syllabus covers thisAU not in syllabusVTU not in syllabusSPPU 302031MJ Unit 4MU HCEV601 Module 4KTU not in syllabus

Where does the heat in a cell come from?

Bernardi's equation splits it into two parts. Irreversible heat is the current times the gap between OCV and terminal voltage: Joule heat in R0 plus polarisation heat in R1. At 1 C (5 A) on the scooter cell that is 0.45 W in R0 and, once v1 has built up, 0.3 W in R1. Reversible (entropic) heat, −I·T·dU/dT, comes from the reaction itself and changes sign with SoC. With the LG M50's measured dU/dT it adds 0.54 W at 10 % SoC and takes away 0.098 W at 90 %.

How hot does one cell get? The lumped thermal model

Treat the cell as one temperature T: m·cp·dT/dt = Q − hA(T − Tamb). The M50's 21.1 × 70.15 mm can has A = 53.5 cm². With m·cp = 69 J/K (assumed) and h = 10 W/m²K, the time constant m·cp/(hA) is 21 minutes. A steady 0.45 W would settle 8.4 °C above ambient. In the interactive, one hour at 1 C from 30 °C peaks at 43.7 °C, the heat rising near empty where R grows and entropic heat turns positive. The city ride averages only 0.094 W.

Heat in a cell: Joule and entropic heat, the lumped thermal model and the temperature through a ride: interactive loads as you scroll

Why does a cell heat more in the cold?

Resistance rises as temperature falls: in this book's model R0 at 0 °C is 2.09 times the 25 °C value. For the same current, I²R heat rises by the same factor. The book's city ride makes 0.094 W per cell from 30 °C but 0.214 W from 5 °C. That self-heating helps a cold pack warm up, and some cars pulse current through the pack on purpose to warm it before fast charging. But the voltage sag that comes with high resistance cuts the power available, which the state-of-power chapter covers.

When is one temperature enough? The Biot number

The lumped model assumes the core and surface are at one temperature. The Biot number Bi = h·(V/A)/k checks that; below about 0.1 it is fair. Across the layers, the jelly roll conducts poorly: O'Regan's layer data give k ≈ 1.23 W/mK. With V/A = 4.59 mm, still air (h ≈ 5.3 W/m²K for a lone cell) and h = 10 give Bi = 0.037. A fan at 3 m/s (h ≈ 41) gives 0.15. The core then runs hotter: q'''r²/(4k) is 0.42 °C at 0.45 W and 1.4 °C at 7 A.

What does cell heat mean for pack design?

Multiply by the cell count: 168 scooter cells at the interactive's 1 C average of 0.65 W make about 109 W, in a closed box, often on a hot summer day. The pack must keep every cell inside the datasheet window (LG M50: charge 0-45 °C, discharge −20-55 °C) and keep cells at similar temperatures, because warm cells age faster and drift apart. AIS-156 adds at least 4 temperature sensors and a traction cut above 60 °C. Air gaps, cell spacing, conduction to a cold plate and heating in winter are designed in chapter 17.

Predict, then check

On a 1 C discharge, the entropic (reversible) heat of the NMC cell at 10 % SoC is:

Worked examples

Heat in the scooter cell at 1 C, at 50 % and at 10 % · 5 A from the 5 Ah NMC cell at 25 °C after a few minutes (v1 settled at I·R1). R0 = 18 mΩ, R1 = 12 mΩ at 50 %, × 1.128 at 10 % (assumed). dU/dT from O'Regan 2022
  1. 50 %: Joule 5² × 0.018 = 0.45 W; polarisation 5² × 0.012 = 0.3 W; entropic −5 × 298.15 × 0.0148×10⁻³ = −0.022 W
  2. Total at 50 %: 0.728 W
  3. 10 %: Joule 25 × 0.018 × 1.128 = 0.508 W; polarisation 0.338 W; entropic −5 × 298.15 × (−0.363×10⁻³) = 0.541 W
  4. Total at 10 %: 1.387 W

Answer: 0.73 W at 50 % and 1.39 W at 10 %: the end of a discharge is the hottest part

A cell in a Chennai summer: temperature after 20 minutes · One cell makes a steady 0.6 W in 35 °C air with h = 10 W/m²K. A = 53.49 cm², m·cp = 69 J/K (assumed)
  1. hA = 10 × 0.00535 = 0.0535 W/K; steady rise Q/(hA) = 11.22 °C, so T∞ = 46.22 °C
  2. τ = m·cp/(hA) = 69/0.0535 = 1,290 s
  3. T(1,200 s) = 46.22 − 11.22 × e^(−1,200/1,290) = 41.79 °C
  4. Time to 45 °C: t = −τ ln(1 − 10/11.22) = 2,866 s = 48 min

Answer: 41.8 °C after 20 min; it would cross the 45 °C charge limit after about 48 min and settle at 46.2 °C

Forced air: h, Biot number and the core temperature · Air at 3 m/s blows across a 21.1 mm cell (air at 300 K: k = 0.02624 W/mK, ν = 1.568×10⁻⁵ m²/s, Pr = 0.707). The cell gives 7 A; radial conductivity 1.225 W/mK
  1. Re = 3 × 0.0211/1.568×10⁻⁵ = 4,037; Churchill-Bernstein gives Nu = 32.8, so h = Nu k/D = 40.8 W/m²K
  2. Bi = h(V/A)/k = 40.8 × 4.59×10⁻³/1.225 = 0.153: above 0.1, so the core is noticeably hotter
  3. Heat at 7 A (settled): 49 × (0.018 + 0.012) = 1.47 W; q''' = Q/V = 59.9 kW/m³
  4. Core − surface = q''' r²/(4k) = 59,929 × 0.01055²/(4 × 1.225) = 1.36 °C; surface in 35 °C air settles at 41.7 °C

Answer: h ≈ 41 W/m²K, Bi = 0.15, core about 1.4 °C above the surface

Chapter 9

Coulomb counting: drift, sensor offset, capacity error and reset points

Which syllabus covers thisAU AU3002 Unit IVVTU BEE657A Module 3SPPU 302031MJ Unit 3MU HCEV601 Module 4KTU PEEET521 Module 2

What is coulomb counting (Ah counting)?

The BMS measures pack current many times a second and adds it up: charge out = ∫ I dt. Divide by the capacity and subtract from the starting SoC. One 30-minute ride of the book's scooter draws 10.14 Ah, so a 60 Ah pack falls 16.9 %. It is simple, fast and smooth, which is why every BMS does it. Coulombic efficiency, the share of charge put in that can be taken out again, is above 99.9 % for a healthy Li-ion cell, so the count treats charge in and out alike (lead-acid needs a correction).

Why does a coulomb count drift away from the truth?

Three errors grow differently. A sensor offset adds the same small current all day: 100 mA is 2.4 Ah a day, 4 % of 60 Ah, even while parked. A gain error scales the charge moved: 1 % of a 10.1 Ah ride is 0.1 Ah, and it cancels when the same charge goes back in. A capacity error grows with use: if the cells hold 54 Ah but the BMS divides by 60, after 30 Ah the gauge says 50 % while the truth is 44.4 %. In the interactive's default week with no reset the gauge ends 31 % high, and the rider is stranded on day 7.

Coulomb counting: integrating current, sensor offset and capacity errors that drift, and reset points: interactive loads as you scroll

Shunt or Hall sensor: how good is the current measurement?

A shunt is a tiny precise resistor: the BMS measures the voltage across it. The TI BQ76952 coulomb counter has under 1 µV offset; across an assumed 100 µΩ shunt that is 10 mA, or 0.4 % of 60 Ah per day. A Hall sensor measures the magnetic field around the cable, so it is isolated and loses no power, but its offset is larger and drifts with temperature. Isabellenhütte's IVT-S shunt module states offsets of 8, 25 and 75 mA for its 100, 300 and 500 A versions, with ±0.4 % of reading in total. A bigger range means a bigger offset: pick the smallest range that covers the peak current.

How does a BMS reset its count?

It waits for a moment when it knows the true SoC. At the end of a full charge (constant voltage, current tapering to the cut-off) the pack is at 100 %: set the count to 100 %. After a long rest the terminal voltage equals the OCV, which maps to SoC. For NMC near 50 % a 2 mV reading error moves the SoC only 0.22 %; on LFP's flat plateau the same 2 mV is 6.1 %, so LFP packs lean on full charges. Riders who never charge to full or never park long starve the BMS of resets, and the error keeps growing.

What is the SoC of a pack when its series cells disagree?

A series string carries one current, but its cell groups drift apart. The pack can only discharge until its lowest group is empty and only charge until its highest group is full. Take the scooter's groups spread from 38 % to 46 %: it can give 22.8 Ah and take 32.4 Ah, so only 92 % of the 60 Ah window is usable. A dashboard therefore shows a blend: it follows the lowest cell as the pack nears empty and the highest as it nears full, so it reaches 0 % and 100 % exactly when the BMS cuts off. Balancing (chapter 13) closes the gap.

Predict, then check

The current sensor reads 50 mA too high all the time. The scooter (60 Ah) sits parked for 5 days. The gauge:

Worked examples

A Hall sensor's offset on a parked scooter · The book's scooter pack (60 Ah) uses a Hall sensor with a 30 mA offset. It is parked for a week. Compare a ±100 A shunt module with an 8 mA offset
  1. Phantom charge per day = 0.030 A × 24 h = 0.72 Ah
  2. Per day: 0.72/60 = 1.2 % of SoC
  3. After 7 days: 5.04 Ah = 8.4 %
  4. 8 mA shunt module: 0.008 × 24 × 7/60 = 2.24 % in a week

Answer: The gauge drifts 8.4 % in a week with the Hall sensor and 2.2 % with the shunt module

An aged pack and an un-updated capacity · The scooter's cells now hold 51 Ah (85 % SoH), but the BMS still divides by 60 Ah. Start at 100 %. Energy use 30 Wh/km (assumed); the pack's nameplate energy is 3.05 kWh
  1. After 40 Ah: gauge = 1 − 40/60 = 33.3 %; true = 1 − 40/51 = 21.6 %
  2. At true empty (51 Ah drawn): gauge = 1 − 51/60 = 15 %
  3. The rider believes 15 % × 3,049 Wh = 457 Wh is left
  4. Range promised but missing = 457/30 = 15.2 km

Answer: The scooter dies with the gauge at 15 %, about 15 km short; the BMS must re-learn capacity (chapter 11)

How good is an OCV reset? NMC vs LFP · After a long rest the BMS reads the cell voltage with a 2 mV error and looks up SoC on the OCV curve. Compare the book's NMC and LFP cells at 30, 50 and 70 %
  1. OCV slope, NMC: 10.88, 8.96, 10.54 mV per % SoC
  2. OCV slope, LFP: 4.982, 0.487, 1.764 mV per % SoC
  3. SoC error, solving the OCV curve exactly for a reading 2 mV high (≈ 2 mV ÷ slope where the curve is straight): NMC 0.18, 0.22, 0.19 %
  4. LFP: (the plateau bends, so the exact answer is larger than 2 ÷ 0.487 = 4.1 %) 0.4, 6.1, 1 %

Answer: An OCV reset fixes an NMC count to better than 0.3 %, but on LFP's plateau it can be off by 6 %: LFP packs rely on full charges

Chapter 10

Model-based state of charge: the Kalman filter

Which syllabus covers thisAU AU3002 Unit IVVTU BEE657A Module 3SPPU not in syllabusMU HCEV601 Module 4KTU not in syllabus

Why combine coulomb counting with the cell voltage?

Each is good where the other is weak. Coulomb counting is smooth but drifts and cannot correct a wrong start. The voltage is anchored to SoC through the OCV curve but is noisy and bent by current (I·R0 and the RC voltage). On the NMC cell at 50 % the OCV slope is 9 mV per % SoC, so 5 mV of noise looks like 0.6 % of SoC; on LFP's plateau it is 0.49 mV per %, so the same noise looks like 10 %. A model-based estimator runs the cell model with the measured current and nudges it toward what the voltage says.

How does an extended Kalman filter estimate SoC?

Every sample it does two steps. Predict: count the coulombs to move SoC, update the RC voltage, and grow the uncertainty a little. Update: compute the voltage the model expects, V = OCV(SoC) − I·R0 − v1, compare it with the measured voltage, and move the estimate by the Kalman gain times the error. 'Extended' means the OCV curve is linearised at the current estimate, using its slope. Started 20 % too high with a perfect voltmeter, the filter in the interactive is within 2 % after 3 minutes on NMC and 21 on LFP; plain coulomb counting stays 20 % wrong.

A Kalman filter estimating state of charge: a wrong start, a biased sensor, and the voltage pulling it back: interactive loads as you scroll

What is the Kalman gain, and what do Q and R tune?

The gain says how much to trust the voltage. For SoC alone, K = P·H/(H²P + R): P is how unsure the filter is about SoC, H is the OCV slope and R is the voltage noise variance. Unsure SoC or a quiet voltmeter gives a big gain; a noisy voltmeter gives a small one. With σ = 10 % at 60 % SoC on NMC and 5 mV noise, K = 1.006 per volt, so a 14 mV error moves SoC by 1.41 %. Q, the process noise, says how fast the model's own error grows. Raise Q (or lower R) and the filter follows the voltage harder; lower it and it trusts the count.

When does a Kalman filter give a wrong answer?

When the model is wrong, the filter believes it anyway. If the real OCV sits 10 mV above the filter's table (LFP hysteresis is that size, chapter 7), the SoC estimate is pulled toward the wrong value: about 1.1 % on NMC at 50 %, but 21 % on LFP's plateau. In the interactive this LFP case is still 15 % off after an hour. Ageing changes capacity and resistance, temperature changes R, and a current-sensor bias the model does not know about leaves a steady error. Good BMSs re-identify R and capacity, model hysteresis, and add the bias as an extra state.

What other SoC estimators exist, including neural networks?

The unscented Kalman filter (UKF) pushes a few sample points through the full OCV curve instead of a slope, which helps where the curve bends. Observers (Luenberger, sliding-mode) use a fixed correction gain: simpler, with weaker guarantees. Neural-network and neuro-fuzzy estimators learn SoC from logged voltage, current and temperature with no circuit model. They need many hours of labelled data across temperatures, ages and drive styles, and they only know what they were trained on: a network trained at 25 °C has never seen −10 °C, where this book's cell has 2.9 times the resistance. Production BMSs mostly run a filter, sometimes with a learned correction.

Predict, then check

The filter's OCV table is 10 mV off for both cells. Whose SoC estimate suffers more?

Worked examples

One Kalman update by hand · NMC cell. The filter thinks SoC = 60 % with σ = 10 %. OCV slope there: 0.992 V per unit SoC. Voltage noise σ = 5 mV. The measured voltage is 14 mV above the model's prediction
  1. P = 0.10² = 0.01; R = 0.005² = 25×10⁻⁶ V²
  2. K = P·H/(H²P + R) = 0.01 × 0.992/(0.992² × 0.01 + 0.000025) = 1.006 per V
  3. New SoC = 0.60 + 1.006 × 0.014 = 61.41 %
  4. New P = (1 − K H) P → σ = 0.5 %

Answer: SoC moves to 61.4 % and the uncertainty shrinks from 10 % to 0.5 %: one clean voltage reading is worth a lot on NMC

One predict step · NMC cell (5 Ah, R1 = 12 mΩ, τ = 30 s, assumed typical) at 25 °C, 50 % SoC, v1 = 0. The measured current is 5 A (1C) for one 1 s sample
  1. ΔSoC = I·Δt/(3600·Q) = 5 × 1/(3600 × 5) = 0.0278 %
  2. a = e^(−1/30) = 0.9672
  3. v1 = a·0 + R1(1 − a)·I = 0.012 × 0.0328 × 5 = 1.97 mV
  4. The SoC variance grows by Q; then the update step compares the voltage

Answer: SoC 50 % → 49.972 %, v1 = 1.97 mV; after 30 s at 1C v1 would reach 63 % of 60 mV

Coulomb counting vs the filter after two hours · The 5 Ah NMC cell starts with its SoC guessed 20 % too high; the current sensor reads 20 mA high. Two hours of the ride
  1. Coulomb count: the starting error never changes: +20 %
  2. Bias adds 0.020 A × 2 h = 0.04 Ah = 0.8 % of 5 Ah, in the opposite direction (more charge counted out)
  3. Coulomb count error after 2 h = +20 − 0.8 = +19.2 %
  4. The EKF (noise-free run, no bias) is within 2 % after 3 min and ends less than 0.01 % off

Answer: Coulomb counting: +19.2 %; the Kalman filter corrects itself from the voltage within minutes

Chapter 11

State of health: capacity fade, resistance growth, calendar and cycle ageing

Which syllabus covers thisAU AU3002 Unit IVTU BEE657A Module 3SPPU 302031MJ Unit 6MU HCEV601 Module 4KTU not in syllabus

What is state of health, and how does a BMS measure it?

SoH compares the cell now with when it was new. By capacity: SoH = Q now / Q rated. By resistance: R now / R new, which sets power (chapter 12). An EV pack is usually called worn out for driving at 80 % capacity. For the book's 33.3 kWh car that is 26.6 kWh, and a full-charge range at 130 Wh/km (assumed) falls from 256 km to 205 km. The BMS estimates capacity by counting the Ah between two rested OCV readings: 52 Ah moved while SoC went from 80 % to 20 % means Q = 52/0.6 = 86.7 Ah, SoH 87 % for a 100 Ah cell. It tracks R from voltage steps when the current changes.

What physically wears a lithium-ion cell out?

SEI growth: the solid electrolyte interphase, a thin film on the graphite, keeps thickening and locks up lithium; it runs all the time, faster when hot and at high SoC, and slows as the film grows (roughly with √time). Lithium plating: charging too fast or too cold deposits metallic lithium on the anode instead of inside it, losing lithium and risking shorts (chapter 18). Particle cracking: electrode particles swell and shrink every cycle, more at deep cycles and high current, and cracks expose fresh surface. The knee: after years of slow fade some cells suddenly fade fast; the smooth models here do not show it.

Calendar or cycle ageing: which wears an Indian car pack faster?

Calendar ageing happens while parked; cycle ageing while driving and charging. The interactive adds Naumann's calendar law (√t, Arrhenius in temperature, rising with parked SoC) to Wang's cycle law (Ah throughput to the power 0.55, Arrhenius, C-rate). Each was fitted to one small LFP lab cell, not this car's cells. Parked at 70 % and driven 40 km a day at C/2 in Chennai (29.1 °C mean), the pack is at 85.2 % after 5 years: 10.8 % lost to calendar, 3.9 % to cycling. It reaches 80 % in 9 years; Delhi 11.2, Bengaluru 11.6. Time and heat, not kilometres, dominate.

State of health: capacity fade and resistance growth with temperature, state of charge, depth of discharge and years: interactive loads as you scroll

How does the SoC window affect battery life?

Parking full is the costliest habit. In Naumann's calendar law the SoC factor is 0.959 at 100 %, 0.602 at 50 % and 0.579 at 30 %: a car parked full ages 59 % faster than one parked half full. In the Chennai case, 80 % comes after 4.7 years when parked full every night and 9.5 years at 50 %. Deep cycles also hurt per cycle: the Sony cell's cycle law weights a 100 % cycle 1.35 against 0.83 for a 20 % cycle. That is why makers default to charging to 80-90 % and why storage SoC is kept low.

How much does temperature speed up ageing?

Both laws follow Arrhenius: the rate goes as e^(−Ea/RT). Going from 25 °C to 35 °C speeds calendar fade by 1.25 times (Ea = 17.1 kJ/mol) and Wang's cycle fade by 1.51 times (Ea about 31.5 kJ/mol at C/2). Delhi's 13.8 °C January and 33.3 °C June average out close to its annual mean for these mild activation energies: 5 years at 70 % with the monthly profile gives 10.11 % calendar loss against 9.92 % at the mean. The pack sits warmer than the air while driving, charging and parked in sun, so cooling (chapter 17) buys years.

Predict, then check

Same car, same 40 km a day in Chennai. Owner A parks at 100 % every night, owner B at 50 %. Years to 80 % capacity:

Worked examples

Calendar fade by hand: five years in Chennai · Naumann's calendar law (Sony LFP cell): loss = kT·kSoC·√t, kT = 1.2571×10⁻⁵ e^(−17126/R (1/T − 1/298.15)), kSoC = 2.8575(SoC − 0.5)³ + 0.60225, t in seconds. Mean cell temperature 29.1 °C, parked at 70 %
  1. T = 302.25 K: kT = 1.2571×10⁻⁵ × e^(0.0937) = 1.3806×10⁻⁵
  2. kSoC = 2.8575 × 0.2³ + 0.60225 = 0.6251
  3. t = 5 × 365.25 × 86400 = 157.8×10⁶ s; √t = 12561
  4. Loss = 1.3806×10⁻⁵ × 0.6251 × 12561 = 10.84 %

Answer: About 10.8 % of capacity lost to calendar ageing alone in 5 years

Cycle fade by hand: 40 km a day for five years · The book's car pack (33.3 kWh) at 130 Wh/km (assumed), one charge a day, average C/2, 29.1 °C. Wang's law: loss % = B e^(−(31700 − 370.3·C)/(RT)) Ah^0.55, B = 31,630 at C/2, Ah of the 2.2 Ah test cell
  1. Daily energy = 40 × 0.13 = 5.2 kWh; DoD = 5.2/33.28 = 15.6 %
  2. Full equivalent cycles in 5 years = 0.1562 × 1826.25 = 285
  3. Same cycles on the 2.2 Ah cell: Ah = 285 × 2.2 = 628 Ah; Ah^0.55 = 34.58
  4. e^(−31515/(8.314 × 302.25)) = 3.576×10⁻⁶; × 31,630 × 34.58 = 3.91 %

Answer: About 3.9 % lost to cycling, less than half the calendar loss; SoH after 5 years ≈ 85.2 %

A spare pack in a hot warehouse · A spare LFP pack is stored for years with no use (calendar ageing only, Naumann's law). Warehouse at 40 °C. Compare storing it at 100 % and at 30 % SoC, and at 30 % in a 25 °C room
  1. Calendar only: 0.2 = kT·kSoC·√t, so t = (0.2/(kT·kSoC))²
  2. kT(40 °C) = 1.7502×10⁻⁵; kSoC(100 %) = 0.9594, kSoC(30 %) = 0.5794
  3. 100 %: t = 4.5 years; 30 %: t = 12.3 years
  4. 30 % at 25 °C: kT = 1.2571×10⁻⁵, t = 23.9 years

Answer: 4.5 years to 80 % at 100 % SoC, 12.3 years at 30 %, 24 years at 30 % and 25 °C: store cool and part-charged

Chapter 12

State of power and remaining range

Which syllabus covers thisAU AU3002 Unit IIVTU BEE657A Module 3SPPU not in syllabusMU not in syllabusKTU not in syllabus

What is state of power?

State of power is the largest power the pack can give (or take) for the next few seconds, typically 2, 10 or 30 s, without any cell crossing a voltage, current or temperature limit. With the 1RC model from rest, the voltage after a pulse of current I for Δt seconds is OCV − I·R0 − I·R1(1 − e^(−Δt/τ)); setting it equal to the 2.5 V cut-off gives the current limit. For the book's NMC cell at 50 % and 25 °C over 10 s that is about 56 A, 141 W per cell at 2.5 V. The voltage limit alone would let the scooter pack give 23.2 kW.

What limits the power: voltage, current or temperature?

Usually the cell's current rating. The LG M50 datasheet allows 1.5C (7.275 A) discharge from 5 to 60 °C but only 0.3C from −20 to 5 °C; charging is 0.3C below 25 °C, 0.7C above, and not allowed below 0 °C or above 45 °C. For the scooter pack at 50 % and 25 °C over 10 s, the BMS allows 4.34 kW discharge (87.3 A) and 2.16 kW regen. On a −10 °C morning at 15 % SoC: 0.81 kW and no regen at all. The voltage limit takes over near empty and, for regen, near full. The model's honest catch: it allows 164 W per cell at 80 % SoC where the datasheet promises up to 80 W.

State of power: the peak current and power a pack can give now within its voltage, current and temperature limits: interactive loads as you scroll

How is peak power measured on a real cell?

By pulse tests at several SoCs and temperatures. In an HPPC-style test (hybrid pulse power characterisation, chapter 6) the cell takes a 10 s discharge pulse, rests, then a 10 s charge pulse; the voltage drop gives the 10 s resistance and P = Vmin(OCV − Vmin)/R. On the book's cell a 20 A, 10 s pulse at 50 % drops the voltage 438 mV, so R = 21.9 mΩ and P = 138 W. A constant-current method steps to a fixed current and records the voltage; a constant-voltage method holds the cell at Vmin for 10 s and records the current, measuring the limit directly. The BMS stores these as SoC-temperature tables.

Why does the BMS send power limits to the motor controller, and what limits regenerative braking?

Only the BMS knows the cells; only the controller sets torque. So the BMS sends allowed discharge and charge power (or current) over CAN many times a second, and the controller caps torque to stay under it. Regen is charging, so it gets the charge limits. Braking a 150 kg scooter and rider from 40 km/h in 4 s releases 9.3 kJ, an average 2.31 kW. The pack accepts 2.16 kW at 25 °C, 0.92 kW at 15 °C and nothing below 0 °C or near 100 % SoC; the mechanical brakes take the rest.

How does a BMS estimate the remaining range?

Range = energy left ÷ recent energy per km. At 50 % SoC the scooter pack holds about 1.44 kWh at OCV (the area under the OCV curve from 0 to 50 % times the charge), so at 30 Wh/km (assumed) it shows 48 km; if the last trips averaged 36 Wh/km it shows 40 km. Displays smooth the Wh/km over the last 10-50 km so the number does not jump, and subtract a reserve. Cold, hills, speed and an aged pack all shrink it, which is why the estimate keeps changing on a trip.

Predict, then check

A −10 °C winter morning, scooter at 80 % SoC. How much regenerative braking power can the pack accept?

Worked examples

10 s power limit of one cell by the textbook formula · Book's NMC cell at 50 % SoC, 25 °C, rested: OCV = 3.708 V, R0 = 18 mΩ, R1 = 12 mΩ, τ = 30 s (assumed typical). Cut-off 2.5 V; datasheet limit 1.5C = 7.275 A
  1. 10 s resistance: R0 + R1(1 − e^(−10/30)) = 18 + 12 × 0.2835 = 21.4 mΩ
  2. Imax = (3.708 − 2.5)/0.0214 = 56.5 A (exact model with SoC drop: 55.2 A)
  3. Power at the cut-off: 56.5 × 2.5 = 141 W per cell
  4. The datasheet caps continuous current at 7.275 A, so the BMS allows 87.3 A for 12 cells in parallel: 4.34 kW for the pack

Answer: The voltage would allow about 56 A per cell, but the BMS sends 4.34 kW: the current rating limits this pack

A cold, low, aged pack · Scooter pack at −10 °C, 20 % SoC, cell resistance 50 % above new. 10 s pulse
  1. Resistance factor at −10 °C, 20 % (kernel law): 3.13; with age × 1.5 = 4.7
  2. Voltage-limited current per cell (exact model): 9.4 A; for 12 cells 112 A
  3. Datasheet at −20 to 5 °C (LG M50 specification, 2016 issue): 0.3C = 1.455 A per cell, 17.5 A for the pack. The 2017 tentative sheet used in chapter 3 allows 0.5C below 10 °C: check the issue you design to
  4. Pack power = 17.46 A × 46.2 V = 0.81 kW

Answer: Only 0.81 kW, 19 % of the warm pack's 4.34 kW: the scooter feels sluggish until the cells warm up

Regen just after a full charge · Scooter pack at 98 % SoC, 25 °C, 10 s regen pulse. OCV = 4.156 V per cell; limit 4.2 V; datasheet charge limit 0.7C = 3.395 A per cell
  1. Headroom = 4.2 − 4.156 = 44 mV
  2. 10 s resistance at 98 % (R is 18 % higher near full): 25.35 mΩ; simple formula I = 44 mV ÷ 25.35 mΩ = 1.75 A per cell
  3. Exact model (the OCV rises during the pulse): 1.69 A per cell, well below the 3.395 A rating
  4. Pack: 20.3 A × 58.8 V = 1.19 kW

Answer: Only 1.19 kW of regen near full, against 2.16 kW at 50 %: the voltage limit sets it, so the first stops after a full charge use the friction brakes

Chapter 13

Cell balancing: passive bleeding vs active energy transfer

Which syllabus covers thisAU AU3002 Unit IVTU BEE657A Module 3SPPU 302031MJ Unit 3MU HCEV601 Module 4KTU not in syllabus

Why do the series groups of a pack drift apart, and what does it cost?

No two cells are identical. Small differences in self-discharge, a warmer spot near the motor controller, or slightly different capacity make the series groups drift to different states of charge over weeks. Charging must stop when the fullest group reaches its limit and riding must stop when the emptiest group is empty, so the whole string loses the gap. In the book's scooter pack (14 groups of 60 Ah), a 3 % spread leaves 58.2 Ah usable instead of 60, about 91 Wh of the 3.05 kWh gone. Balancing does not add capacity: it lines the groups up again.

How does passive balancing work, and how fast is it?

Passive balancing bleeds the fuller groups through a resistor until they match the lowest. TI's BQ76952 monitor has internal balancing switches of about 28 Ω (15 to 46 Ω) in series with two 20-100 Ω input resistors, so at 4.0 V the current is 4.0/(28 + 40) = 59 mA, under its 100 mA limit. That is slow: a 3 % spread on a 60 Ah group is 1.8 Ah, which takes 30.5 hours at 59 mA. All the extra charge becomes heat, about 0.25 W per group and 62 Wh for the pack. It is simple and cheap, so most packs use it.

Passive bleeding vs active transfer balancing: time to balance, energy wasted and capacity recovered: interactive loads as you scroll

What is active balancing, and when is it worth it?

Active balancing moves charge from full groups to empty ones instead of burning it. A switched capacitor or a small inductor or transformer converter shuttles energy; TI's EMB1499Q, for example, swaps energy between one cell and the whole stack. In this chapter's model one converter moves 2 A with an assumed 85 % efficiency: the same 3 % spread is gone in about 1.5 hours and only about 2 Wh is lost, against 62 Wh burned by passive bleeding. The price is more parts, switches and control. It pays off in big, expensive packs, second-life packs with very mixed cells, and packs that must deliver every last amp-hour.

When should a BMS balance: at the top, at the bottom, while charging?

Most packs balance at the top: they align the groups near full charge, where the charger tapers and the voltage says the most about state of charge (TI's note on cell balancing recommends balancing near the end of charge for this reason). NMC voltage changes steeply near full, so a few millivolts mean something; LFP is flat in the middle, so its BMS balances near the top too. Bottom balancing aligns the groups at empty instead, for packs that are often run flat. TI's BQ76952 design example only balances above 100 mV of imbalance. If the scooter balances for just 1 hour per daily charge, a 31-hour job takes a month.

What fixes imbalance besides the balancing circuit?

Prevention matters more than the bleed resistor. Grade and match cells by capacity and resistance before building groups, keep every group at the same temperature with good cooling, and avoid putting a few cells next to a heat source. Bigger monitors balance faster: NXP's MC33771C balances at up to 300 mA and TI's BQ79616 at 240 mA, four to five times the BQ76952's internal 59 mA. External transistors and resistors can go higher still, at the cost of heat on the board. And one honest limit: balancing aligns the state of charge, but a group with lower capacity still limits the pack.

Predict, then check

One 60 Ah group sits 1.2 Ah above the others. The BMS bleeds it at 60 mA. Roughly how long until it matches?

Worked examples

Passive balancing time and heat on the scooter pack · One 60 Ah group of the book's scooter pack is 2 % fuller than the rest. The BMS uses the BQ76952's internal switch (28 Ω typical) with two 20 Ω input resistors; the group sits at about 4.0 V
  1. Charge to remove: ΔQ = 0.02 × 60 = 1.2 Ah
  2. Balancing current at 4.0 V: I = 4.0/(28 + 2 × 20) = 58.8 mA (the BQ76952 limit is 100 mA)
  3. Time: t = ΔQ/I = 1.2/0.0588 = 20.4 h
  4. Heat: P = V × I = 4.0 × 0.0588 = 0.235 W; inside the chip 28/68 of it = 0.097 W
  5. If the BMS balances only during 1 hour of each daily charge, it needs about 20 days

Answer: About 20 hours of balancing at 59 mA, dissipating 0.24 W

What a 3 % spread costs the scooter · The 14 groups of the scooter pack (60 Ah each, 50.8 V nominal, 3.05 kWh) end a charge between 97 % and 100 %
  1. Charging stops when the top group reaches 100 %; riding stops when the bottom group reaches 0 %
  2. Usable charge = 60 × (1 − 0.03) = 58.2 Ah
  3. Energy lost ≈ 1.8 Ah × 50.8 V = 91 Wh, or 3 % of the range
  4. Passive balancing at 59 mA recovers it in 1.8/0.059 = 30.5 h of balancing

Answer: 58.2 Ah usable instead of 60 Ah: 91 Wh lost until the groups are balanced

Active transfer vs passive bleeding for four groups · Four 60 Ah groups at 100 %, 99 %, 98.5 % and 97 % SoC. Option A: bleed the top three down to 97 % at 100 mA each. Option B: one converter moves 2 A from the fullest to the emptiest group at 85 % efficiency (assumed). Energies use the NMC open-circuit voltage curve
  1. A: charge burned = (3 + 2 + 1.5) % × 60 = 3.9 Ah, energy 16.2 Wh; time = 0.03 × 60/0.1 = 18 h
  2. B: the groups meet at a level L where 0.85 × (charge taken above L) = (charge given below L): L = 98.55 %
  3. Taken: 1.135 Ah; delivered: 0.85 × 1.135 = 0.965 Ah
  4. Energy lost in B = 4.74 − 4.01 = 0.73 Wh; time = 1.135/2 = 0.57 h

Answer: Active: 0.7 Wh lost in 0.6 h; passive: 16 Wh burned in 18 h

Chapter 14

Protection: over- and under-voltage, over-current, temperature and short circuit

Which syllabus covers thisAU AU3002 Unit VVTU BEE657A Module 2SPPU 302031MJ Unit 4MU HCEV801 Module 4KTU not in syllabus

What must a BMS protect a lithium-ion pack against?

The cell maker sets the limits. LG's M50 specification (the book's 5 Ah NMC cell) allows 4.20 ± 0.05 V at full charge, 2.50 V at empty, 1.5C (7.3 A) continuous discharge, charging only from 0 to 45 °C and discharging from −20 to 60 °C. It lists the functions a pack must have: over- and under-voltage, over-charge and over-discharge current, short circuit, over-temperature, a second over-voltage stage, FET failure and imbalance protection, plus balancing. India's AIS-156 (Amendment 3) checks five of these on every L-category (two- and three-wheeler) pack in type approval: over-charge, over-discharge, over-temperature, over-current and short circuit.

Why does a pack have several layers of protection, and how fast is each?

Each layer covers the one above failing or being too slow. Firmware compares measured voltages and temperatures with limits and filters them over milliseconds to seconds. The monitor chip's hardware comparators catch current faults on their own: TI's BQ76952 trips on a short in 15-450 µs and on over-current in 10-425 ms. A fuse melts when its I²t is reached: a 100 A Mersen EV fuse needs 910 A²s. A pyrotechnic fuse is fired by a crash signal. In the interactive's default short (1787 A), the chip opens the MOSFETs at 70 µs, before the fuse would melt at about 332 µs.

Protection: over- and under-voltage, over-current, temperature and short circuit, with detection times and actions: interactive loads as you scroll

What is debounce, and how do you avoid both false trips and missed trips?

A signal must stay past its threshold for a set time (the debounce) before the BMS acts, so a spike or noise does not cut power. Too short and the scooter stops on every pothole; too long and a real fault does damage. Thresholds have the same trade-off. The book's ride peaks at 60 A, so an over-current threshold below 60 A trips on every hard start, while a motor stall at 80 A slips under an 87 A threshold and is never seen. Serious faults latch: the BQ76952 has latched short-circuit and over-current versions that stay off until reset, so a pack does not keep switching back onto a fault.

How does a BMS actually disconnect the pack, and what other pack-safety features are required?

Low-voltage packs like the scooter's use MOSFETs: the BQ76952 drives separate charge and discharge FETs, which switch in microseconds, can interrupt a short and can block charging while still allowing discharge. High-voltage car packs use contactors and a pre-charge circuit (next chapter). India's AIS-156 adds pack-safety rules beyond the BMS (Annex 8K): at least 4 temperature sensors, an alarm above 60 °C with a gradual cut of motor power, an extra fuse or breaker, a pressure-release vent and cells certified to IS 16893. Safety is a pack design job, not only a chip.

Why are temperature limits different for charging and discharging?

Charging is the risky direction in the cold: lithium can plate as metal on the anode instead of going into it (chapter 18). LG limits charging to 0-45 °C and to 0.3C below 25 °C (0.7C above), so the scooter charger must drop from 42 A to 18 A in a cool room and stop below 0 °C. Discharge is allowed from −20 to 60 °C. TI's design example sets the same lines: charge blocked below 0 °C or above 45 °C, discharge stopped at 60 °C. Over-discharge has a hard rule too: LG says a cell that fell below 1.0 V must never be recharged.

Predict, then check

The scooter pack is hard-shorted: about 1.8 kA flows. The chip's short-circuit delay is 60 µs and a 100 A fuse needs 910 A²s to melt. What opens the circuit first?

Worked examples

Setting the over-current threshold from the cell datasheet · The scooter pack has 12 LG M50 cells in parallel (1.5C = 7.275 A each), a 0.5 mΩ shunt, and a BQ76952 whose over-current threshold is set in 2 mV steps. A motor stall draws 150 A, rising with a 2 ms time constant; the delay is 80 ms
  1. Pack limit: 12 × 7.275 = 87.3 A
  2. Shunt voltage: 87.3 × 0.5 mΩ = 43.65 mV → next 2 mV step 44 mV = 88 A
  3. The stall crosses 88 A at t = −2 ms × ln(1 − 88/150) = 1.77 ms; MOSFETs open at 81.8 ms
  4. Heat in the cells: ∫i² dt × 21 mΩ = 37.2 J, a negligible temperature rise for 11.6 kg of cells

Answer: Threshold 44 mV (88 A); the stall is cut at 82 ms after 37 J

Short circuit: chip comparator against the fuse · The scooter pack at 80 % SoC (56.3 V open circuit, 21 mΩ cells) is shorted through 5 mΩ; wiring 5 mΩ, shunt 0.5 mΩ, loop inductance 1 µH (assumed). SCD threshold 500 A, delay 60 µs. Fuse: 100 A, pre-arcing I²t 910 A²s
  1. Steady short current: 56.3/31.5 mΩ = 1787 A; rise time constant L/R = 31.7 µs
  2. Shunt voltage at 500 A: 500 × 0.5 mΩ = 250 mV, inside the 10-500 mV range
  3. Crossing at 10.4 µs; MOSFETs open at 70.4 µs
  4. Fuse: a flat 1787 A would melt it in 910/1787² = 285 µs; with the rise, 332 µs
  5. Energy into the short before the BMS opens: ∫i²dt × 5 mΩ = 0.47 J

Answer: The chip opens at 70 µs, about 4.7 times sooner than the fuse; only 0.47 J reaches the fault

Over-voltage when the charger fails to stop · The weakest group of the scooter pack is at 95 % SoC when the charger sticks at 18 A (12 cells share it). Model: NMC open-circuit curve extended linearly above 100 %, R0 and R1 of the cells at 95 % SoC, 25 °C. OV threshold 4.25 V (LG's 4.20 + 0.05 V) with 0.5 s debounce
  1. Time to reach 100 % SoC: 0.05 × 60 Ah/18 A = 10 min
  2. Per cell 1.5 A, so IR adds 31.4 mV plus up to 20.9 mV of polarisation
  3. Voltage reaches 4.25 V at 10.48 min (bisection on the model); charger stopped 0.5 s later
  4. Charge pushed past 100 %: 12 mAh per cell; at rest the cell settles near 4.198 V

Answer: Trip after 10.5 min with about 12 mAh per cell of over-charge

Chapter 15

High-voltage switching: contactors, precharge, fuses and insulation monitoring

Which syllabus covers thisAU not in syllabusVTU not in syllabusSPPU 302031MJ Unit 4MU HCEV801 Module 4KTU not in syllabus

Why can't a car simply close its main contactors onto the inverter?

The motor inverter has a large DC-link capacitor across its input; TI's note on pre-charge says these range from several hundred to thousands of microfarads. An empty capacitor looks like a short. Close the book's car pack (332.8 V) straight onto 1000 µF and only the cells' 36 mΩ, some wiring (10 mΩ assumed) and about 2 µH of loop inductance (assumed) limit the current: it peaks near 4 kA within 54 µs. TI warns that closing without pre-charge can weld the contacts or pit them with an arc. A welded contactor can no longer disconnect the pack.

Switching a high-voltage pack: contactor sequence, precharge resistor, inrush without it, and insulation monitoring: interactive loads as you scroll

How does a precharge circuit work, and how is the resistor sized?

TI's sequence: close the main negative contactor, close a pre-charge contactor in series with a resistor, close the main positive when the capacitor reaches 90-95 % of pack voltage, then open the pre-charge contactor. With 50 Ω and 1000 µF, τ = RC = 50 ms and 95 % takes τ × ln 20 = 150 ms. The current starts at 6.65 A instead of kiloamps. The resistor absorbs ½CV² = 55.4 J whatever its value, with a 2.2 kW peak, so it is a pulse-rated part. Closing at 95 % leaves a 16.6 V step: a brief 200 A blip worth only 0.14 J.

What switching and fusing parts sit in a high-voltage pack?

A car pack's junction box holds two main contactors, one in each pole, so the pack can be isolated completely; the pre-charge contactor and resistor; a current sensor; the main fuse; and often a pyrotechnic disconnect fired by the crash sensor. Contactors are electromagnet-driven switches that can break hundreds of volts DC under load. The BMS checks them: TI's note says no contact may be stuck and no insulation fault may exist before the contactors close, so the BMS compares voltages across each contactor to detect welding. Low-voltage scooter packs use MOSFETs instead (chapter 14).

What is insulation monitoring, and what does UN R100 require?

An EV's high-voltage system floats: neither pole is connected to the chassis. A single leak from one pole to the body is not dangerous by itself, but a second leak on the other pole would put pack voltage on the body. UN Regulation 100 requires at least 100 Ω per volt of working voltage for DC buses (500 Ω/V for AC). The book's car pack works up to 379.6 V, so it needs 37.96 kΩ. Its Annex 4A method measures the voltage from each pole to the chassis, adds a known resistor and measures again. The BMS's insulation monitor repeats this kind of check and warns the driver when it falls below the limit.

How do cabling and service disconnects keep mechanics safe?

UN R100 calls anything above 60 V DC or 30 V AC high voltage. Its high-voltage cables outside enclosures must have an orange outer covering, and the pack carries a yellow and black high-voltage symbol. A service disconnect is a plug that breaks the pack's circuit for servicing, often in the middle of the series string so each half is a lower voltage; R100 requires finger-proof (IPXXB) protection when it is opened without tools. High-voltage connectors must either need tools, lock, or drop their live parts to 60 V DC or less within 1 second of being pulled apart.

Predict, then check

You double the precharge resistor from 50 Ω to 100 Ω. What happens to the energy the resistor must absorb?

Worked examples

Sizing a precharge resistor for the car pack · The book's car pack (332.8 V) feeds a 1000 µF DC link. The main positive contactor must close within 200 ms, at 95 % of pack voltage. Loop resistance of cells and wiring 46.4 mΩ
  1. 95 % needs t = τ ln 20, so τ ≤ 0.200/ln 20 = 66.8 ms
  2. R ≤ τ/C − loop = 66.8 ms/1000 µF − 0.046 = 66.7 Ω → choose 56 Ω, the E12 standard value below it
  3. Check: τ = 56 ms, t95 = 168 ms
  4. Peak current 5.94 A; peak power V²/R = 1976 W
  5. Energy in the resistor up to 95 %: ½CV²(1 − 0.05²) = 55.2 J

Answer: 56 Ω: 95 % in 168 ms, 5.9 A peak, 55 J pulse

Inrush without precharge · Same pack and capacitor, main contactors closed directly. Loop: 46.4 mΩ and 2 µH (assumed)
  1. Ignoring inductance, the first-instant current would be V/R = 332.8/0.0464 = 7172 A
  2. With L: α = R/2L = 11600 s⁻¹, ω0 = 1/√(LC) = 22361 rad/s, damping ratio ζ = α/ω0 = 0.519 (under-damped)
  3. Peak at t = atan(ωd/α)/ωd = 53.6 µs, with ωd = 19116 rad/s
  4. Peak current = V/(Lωd) e^(−αt) sin(ωd t) = 3994 A; the energy lost in the loop is still ½CV² = 55.4 J

Answer: About 3.99 kA after 54 µs, against 5.9 A through the 56 Ω precharge resistor chosen above

An isolation check by the UN R100 Annex 4A method · The car pack (332.8 V measured, 379.6 V working) has leakage of 25 kΩ from the positive pole to the chassis and 2 MΩ from the negative pole. The voltmeter has 10 MΩ input resistance
  1. V1 (negative pole to chassis) = 327.9 V; V2 (positive pole to chassis) = 4.1 V; V1 ≥ V2
  2. Add Ro = 100 Ω/V × 379.6 V = 37.96 kΩ from the negative pole to the chassis; V1' = 198.9 V
  3. Ri = Ro × Vb × (1/V1' − 1/V1) = 25 kΩ, exactly the smaller leak (the meter's resistance cancels)
  4. Per volt: 25000/379.6 = 66 Ω/V, below the required 100 Ω/V

Answer: 66 Ω/V: the pack fails the 100 Ω/V rule and the driver must be warned

Chapter 16

Thermal runaway: self-heating, propagation and how packs are tested

Which syllabus covers thisAU AU3002 Unit IVTU not in syllabusSPPU 302031MJ Unit 6; 302033MJ Unit 5MU HCEV601 Module 4KTU not in syllabus

What happens inside a cell as it heads into thermal runaway?

Heat a charged cell slowly in an accelerating rate calorimeter and it passes through stages. Ohneseit and colleagues (2023) measured 21700 cells at full charge: NMC cells began heating themselves at 85.5 °C (the protective SEI layer on the anode breaks down first), vented gas at 121.5 °C, went into runaway at 198 °C and peaked at 591.6 °C. In runaway the separator melts, the electrodes short and the cathode releases oxygen that burns the electrolyte. LFP cells started at 124.4 °C, vented at 168.8 °C, ran away at 256.3 °C and peaked at 498.6 °C. Other studies differ: an NCM811 18650 started self-heating at 62.5 °C.

What is the point of no return?

A cell makes heat from its own reactions, and that heat grows exponentially with temperature, while cooling removes heat only in proportion to the temperature difference, hA(T − Tamb). Plot both against temperature. Where they cross at a low temperature the cell is stable; the upper crossing is the point of no return: above it, heating always wins. For an NMC 21700 in still air (h = 10 W/m²K) at 25 °C, the model puts it at 166 °C. A cell with a 3 W internal short in a 45 °C box with poor cooling (h = 5) has no stable point and runs away in about 1.2 hours.

Thermal runaway: heat generated vs heat removed, the point of no return and propagation to the next cell: interactive loads as you scroll

What triggers thermal runaway, and why is LFP safer?

Triggers are mechanical (a crush or a nail), electrical (over-charge, an external short, or a cell over-discharged and then recharged), thermal (a nearby fire or overheating) and internal shorts from manufacturing defects, which no BMS can prevent. AIS-156's propagation test starts runaway in one cell with a heater, stopping at runaway or 300 °C, and detects it by a voltage drop plus a temperature rise of at least 1 °C per second. LFP is safer because its cathode holds oxygen more tightly: in Ohneseit's tests it started self-heating about 39 °C later than NMC, peaked 93 °C lower, and nail penetration did not trigger runaway at all.

How does runaway spread through a pack, and what does AIS-156 require?

One cell in runaway heats its neighbours, and if they reach their trigger temperature the event spreads. India's AIS-156 (Amendment 3) requires L-category (two- and three-wheeler) packs to pass a thermal propagation test and to give an audio-visual warning at least 5 minutes before fire or explosion. In this chapter's model, NMC cells touching each other pass runaway on in about 30 seconds; a 0.2 mm insulating pad stretches that to 3.3 minutes, and 0.5 mm stops it. Real packs add spacing, mica or aerogel sheets, cooling and cell fusing. The model ignores hot vent gas, so only a test can prove a design. Chapter 20 lists the standards.

Why does a pack need vent management?

A cell in runaway releases hot, flammable gas, so the gas must leave the pack through a planned path, away from people. AIS-156 requires a pressure-release vent on L-category packs so that a single-cell short cannot build pressure inside. Even LFP vents: a study of a 50 Ah LFP cell at 100 % charge measured 2.33 mol of gas, about 57 litres at room conditions, with hydrogen about 40 % of it; at 50 % charge it made 0.68-0.90 mol. Good designs add a one-way vent valve, gas channels and a sensor that warns early.

Predict, then check

You improve the cooling around a cell (higher h). What happens to the point of no return?

Worked examples

Heat made vs heat removed for an NMC 21700 · An NMC 21700 cell (69 g, cp 0.974 J/g K, 21 mm × 70 mm) in still air, h = 10 W/m²K, at 25 °C. ARC data: self-heating 1 °C/min at 142.7 °C and 10 °C/min at 173.4 °C
  1. Area = π × 0.021 × 0.070 + 2 × π × 0.0105² = 53.11 cm²
  2. At 142.7 °C: made = 69 × 0.974 × 1/60 = 1.12 W; removed = 10 × 0.00531 × 117.7 = 6.25 W: cooling wins
  3. At 173.4 °C: made = 11.2 W; removed = 7.88 W: self-heating wins
  4. The crossing between them (bisection on the model) is the point of no return: 166 °C
  5. Tangency (no stable point left) needs an ambient of 122 °C at this h

Answer: Point of no return ≈ 166 °C; the cell is safe in still air unless something heats it past that

Using an activation energy to extrapolate a heating rate · The NMC cell heats itself at 10 °C/min at 173.4 °C. After venting the reactions have an activation energy of 1.14 eV. Estimate the rate at 198 °C, where runaway starts
  1. Ea/kB = 1.14/8.617×10⁻⁵ = 13229 K
  2. Rate ratio = exp[13229 × (1/446.55 − 1/471.15)] = 4.7
  3. Rate at 198 °C ≈ 10 × 4.7 = 47 °C/min
  4. Measured at the start of runaway: 55 ± 2.6 °C/min

Answer: About 47 °C/min, within about 15 % of the measured 55 °C/min

How much gas a venting LFP cell releases · A 50 Ah LFP cell at 100 % charge released 2.33 mol of gas in runaway (0.93 mol hydrogen); at 50 % charge, about 0.90 mol. Room conditions 25 °C and 101.3 kPa
  1. V = nRT/P = 2.33 × 8.314 × 298.15/101325 = 0.057 m³ = 57 litres
  2. At 50 % charge: 0.90 mol → 22 litres
  3. Hydrogen share at 100 %: 0.93/2.33 = 40 %
  4. The vent path must carry this gas out of the pack and away from the rider

Answer: About 57 litres of flammable gas from one cell, 40 % hydrogen

Chapter 17

Pack thermal management: air, liquid cold plates, immersion and heating

Which syllabus covers thisAU CME337 Units I, IIVTU not in syllabusSPPU 302033MJ Unit 5MU HCEV801 Module 5KTU not in syllabus

Why does a battery pack need its temperature managed at all?

Two ageing mechanisms pull opposite ways. In the cold, resistance rises (about 2.1 times at 0 °C in this book's model) and the anode can plate lithium while charging. When hot, side reactions such as SEI growth speed up. The car cell's own datasheet shows it: at least 5,000 cycles at 25 °C, 3,500 at 35 °C and 2,000 at 45 °C. A 2021 review gives 15-35 °C as the usual window; a 2023 review gives 25-40 °C and says cells in one pack should differ by no more than 5 °C, because a warmer cell ages faster and the pack drifts apart.

What cooling options exist: air, liquid, refrigerant, heat pipes, thermoelectric and phase-change material?

Air cooling blows cabin or outside air past the cells: cheap and light, but air carries little heat. Per litre, 50/50 water-glycol stores about 3,151 times more heat per degree than air. Liquid cooling pumps glycol through cold plates under the cells and is widely used in car packs. Direct refrigerant cooling lets the air-conditioning refrigerant boil inside the plate. Heat pipes move heat by evaporating and condensing a sealed fluid. Thermoelectric (Peltier) modules pump heat electrically but are inefficient. Phase-change material, such as paraffin wax, soaks up heat while it melts. Immersion puts the cells in a dielectric liquid.

How does a liquid cold plate set the cell temperatures?

The coolant warms as it collects heat, by ΔT = Q/(ṁ cp). The car pack at 1C makes 624 W (104 cells × 6 W, assumed resistances). At 6 L/min of 50/50 glycol entering at 25 °C, the coolant warms by only 1.66 °C. Each cell sits above the coolant by its heat times the resistance from cell core to coolant, here 0.325 K/W. So the hottest cell is 28.6 °C and the spread is 1.65 °C. Doubling the flow cools the hottest cell by only 0.83 °C: the plate-to-cell resistance dominates. With air at 3 m³/min, the air warms 10.5 °C and the hottest cell reaches 46.7 °C.

Cooling a pack with a liquid cold plate: flow, coolant temperature rise and the spread between cells: interactive loads as you scroll

How is a pack heated in winter, and what does it cost?

Charging a cold pack risks lithium plating, so the BMS warms it first. A PTC heater in the coolant loop heats the glycol, which heats the cells. The car pack's cells hold 206 kJ per °C. Warming them from 0 °C to 15 °C with 5 kW takes 10.5 minutes and 0.88 kWh, about 2.6 % of the pack, with a 15 W/K heat leak (assumed). Without the leak it would take 10.3 minutes. Heat pumps do the same job with less electricity. Preconditioning means heating or cooling on the way to a fast charger, so the pack arrives ready.

What are thermal interface materials, and why does the resistance chain matter?

Between a cell and the cold plate sit gaps that air would fill badly. A thermal interface material, a gap pad or paste, fills them. Heat then crosses a chain of resistances: inside the cell, through the interface, the plate wall and into the coolant. The last step alone, 1/(hA) with h = 1,000 W/(m² K) on the 160 × 50 mm cell base, is 0.125 K/W. At 6 W that is 0.75 °C. Designers shrink each link because the coolant ΔT is usually the small part.

Predict, then check

The car pack runs at 1C with 6 L/min of coolant. You double the flow to 12 L/min. The hottest cell:

Worked examples

Coolant flow for a fast charge · The car pack (104 LFP cells, R0 + R1 = 0.6 mΩ each, assumed) fast-charges at 1.5C. The coolant is 50/50 water-glycol at 25 °C (cp = 3,488 J/(kg K), ρ = 1,076 kg/m³). Allow a 2 °C coolant rise
  1. Cell current = 1.5 × 100 = 150 A; heat per cell = 150² × 0.0006 = 13.5 W
  2. Pack heat Q = 104 × 13.5 = 1,404 W
  3. ṁ = Q/(cp ΔT) = 1,404/(3,488 × 2) = 0.201 kg/s
  4. Volume flow = 0.201/1,076 × 60,000 = 11.2 L/min

Answer: 1,404 W of heat needs about 11.2 L/min for a 2 °C rise

Hottest cell on a Chennai afternoon · The plate's coolant comes from a radiator at 30 °C, 8 L/min, the car pack at 1C. Cell-to-plate resistance 0.2 K/W plus 1/(hA) = 0.125 K/W (h = 1,000 W/(m² K), 160 × 50.1 mm base)
  1. Q = 104 × 6 = 624 W; ṁ = 1,073 × 8/60,000 = 0.1431 kg/s
  2. Coolant rise = 624/(0.1431 × 3,504) = 1.24 °C
  3. Cell above coolant = 6 × 0.325 = 1.95 °C
  4. Hottest cell (last in the flow) = 30 + 1.24 + 1.95 = 33.2 °C; spread 1.23 °C

Answer: 33.2 °C with a 1.23 °C spread: inside 15-35 °C and the 5 °C spread rule; at 40 °C coolant it would need a chiller

Warming the car pack before a winter fast charge · Delhi morning: cells at −5 °C, target 15 °C. Cells: 104 × 1.98 kg × 1,000 J/(kg K) (assumed cp). PTC heater 6 kW; heat leak to the air 15 W/K (assumed)
  1. Heat capacity M = 104 × 1.98 × 1,000 = 205,920 J/K
  2. Without losses: t = M ΔT/P = 205,920 × 20/6,000 = 686 s = 11.4 min
  3. With the leak: t = −(M/G) ln(1 − G ΔT/P) = −(205,920/15) ln(1 − 300/6,000) = 704 s
  4. Energy = 6,000 × 704 = 1.174 kWh, of which 30 Wh leaks away
  5. Share of the 33.28 kWh pack = 3.5 %

Answer: 11.7 minutes and 1.17 kWh (3.5 % of the pack)

Chapter 18

Charging: CC-CV, fast-charge limits and lithium plating in the cold

Which syllabus covers thisAU AU3002 Unit IIVTU BEE657A Module 2SPPU 302031MJ Unit 6MU HCEV601 Module 5KTU not in syllabus

What is the difference between AC and DC charging, and slow and fast charging?

With AC charging, the car's own onboard charger turns grid AC into DC; its size limits the power. India's Bharat AC-001 gives three 3.3 kW sockets. Filling 90 % of the car pack at 3.3 kW takes about 9.7 hours (94 % charger efficiency assumed). With DC charging, a big charger outside the car feeds the battery directly. Bharat DC-001 gives 10 or 15 kW at 48 or 72 V for two- and three-wheelers; CCS chargers can reach hundreds of kW for cars. Slow overnight charging is gentler on cells; fast charging trades some life and heat for time.

What are CC-CV and pulse charging, and why does charging slow down above 80 %?

CC-CV means constant current until the cell reaches its maximum voltage, then constant voltage while the current fades. LG's M50LT sheet: 0.3C to 4.20 V, then hold until the current falls to 50 mA, about 4 hours. During CC the terminal voltage is OCV plus the I·R drop; near full that sum hits the limit, so the charger must cut current. In the model, the scooter's 10-80 % at 0.7C takes 60 minutes, the last 20 % another 30. Pulse charging adds short rests or reverse pulses; its claimed benefits are still debated. The Electric Vehicles book has a simpler CC-CV interactive.

What limits fast-charge current, and what is a charge current map?

The BMS sends the charger a current limit from a map of temperature and state of charge. The scooter cell's datasheet allows 0.3C below 25 °C and 0.7C from 25 to 45 °C, with no charging outside 0-45 °C. So 10-80 % takes 60 minutes at 25 °C but 140 minutes at 5 °C. The car cell allows 1C, giving 42 minutes in the model. Tata quotes 56 minutes for the 35 kWh Punch.ev on a 50 kW charger, an average of 26.2 kW or 0.75C: real cars taper earlier to protect the cells.

Charging: CC-CV, fast-charge current limits by state of charge and temperature, and the 10-80 % time: interactive loads as you scroll

What is lithium plating, and why is charging in the cold dangerous?

While charging, lithium ions must slide into the graphite anode. The anode sits only about 92 mV above lithium metal when full. Current through the anode's resistance pulls it lower; if it dips below 0 V, metallic lithium plates on the surface instead. Plated lithium is lost capacity and can grow needles that short the separator. Cold raises resistance, so the safe current falls. In this book's teaching model, a scooter cell at 90 % can take 4.3 A at 25 °C but only 2 A at 0 °C. That is why BMSs block cold charging or preheat first.

Which connectors and protocols do chargers use, and what is smart charging?

CCS2 (Combo 2) adds two DC pins to the Type 2 AC plug and is the car standard in India and Europe; CCS1 extends the North American J1772 plug. China uses GB/T, Japan CHAdeMO, and India's Bharat DC-001 uses the GB/T connector for low-voltage vehicles. During a DC charge, the BMS keeps telling the charger the voltage and current it wants. OCPP links chargers to a central system for billing and load control. Smart charging shifts or slows sessions to suit the grid, from the cloud or a local controller. Wireless charging uses coupled coils, with no plug.

Predict, then check

The scooter pack charges at 25 °C from 10 % with the datasheet's 0.7C limit. Which part takes longer?

Worked examples

The car pack's 10-80 % time, compared with a real car · The book's car pack (100 Ah LFP cells, 33.28 kWh) charges at the cell's 1C continuous limit (EVE LF100LA datasheet) from a 50 kW charger. Tata quotes 56 min for 10-80 % on its 35 kWh Punch.ev at 50 kW
  1. Cell current = 1C = 100 A; pack power ≈ 332.8 V × 100 A ≈ 33 kW, below the charger's 50 kW
  2. CC time = ΔSoC × Q / I = 0.7 × 100 Ah / 100 A = 0.7 h = 42 min
  3. Energy into the pack ≈ 0.7 × 33.28 = 23.3 kWh; from the grid at 94 % ≈ 24.8 kWh
  4. Punch.ev average power = 0.7 × 35 kWh / (56/60 h) = 26.25 kW = 0.75C

Answer: 42 min in the model vs 56 min claimed: the real car averages 0.75C, well under 1C

What the BMS lets a scooter charger do · The scooter pack (14S12P of 5 Ah NMC) at 50 % SoC, pack voltage ≈ 14 × 3.708 V. A 3 kW charger is connected. LG M50LT limits: 0.7C at 25-45 °C, 0.3C at 0-25 °C
  1. Charger current = 3,000 / 51.9 = 57.8 A
  2. BMS limit at 30 °C = 0.7 × 5 A × 12 = 42 A; at 10 °C = 0.3 × 5 × 12 = 18 A
  3. 20-80 % = 0.6 × 60 Ah = 36 Ah: at 42 A 51 min; at 18 A 120 min
  4. Energy ≈ 0.6 × 3.05 kWh = 1.83 kWh; from the grid at 94 % ≈ 1.95 kWh

Answer: The BMS cuts 58 A to 42 A on a warm day (51 min for 20-80 %) and to 18 A on a cold one (120 min)

Plating check on a cold morning (teaching model) · One scooter cell at 90 % SoC and 0 °C, charging at 3.5 A (0.7C). Graphite OCP from the LG M50 fit; R0 = 18 mΩ, R1 = 12 mΩ at 25 °C, scaled by the book's temperature law; half of the drop is in the anode
  1. Anode stoichiometry x = 0.0279 + 0.9 × 0.8735 = 0.8141; graphite OCP = 92 mV
  2. At 0 °C: R0 = 42.5 mΩ, R1 = 28.3 mΩ
  3. Anode drop = ½ × 3.5 × (42.5 + 28.3) mΩ = 123.9 mV
  4. Anode potential = 92 − 123.9 = −31.9 mV vs Li
  5. Current for a 20 mV margin = (92 − 20)/(½ × 70.8) = 2.03 A

Answer: The model anode sits at −32 mV, below the 20 mV margin: limit the current to about 2 A or warm the pack (the datasheet's 0.3C = 1.5 A agrees)

Chapter 19

BMS architecture and communication: centralised, modular, wireless and CAN

Which syllabus covers thisAU AU3002 Unit VVTU not in syllabusSPPU not in syllabusMU HCEV801 Module 4KTU not in syllabus

What are the main BMS architectures?

A centralised BMS has one board with sense wires running to every cell group: simple for the scooter's 14 groups (23 wires with 4 thermistors), messy for the car's 104 (137 wires crossing the pack). A modular or master-slave BMS puts a small monitoring board on each module and links them to a master. A distributed BMS goes further, with electronics on every cell. A wireless BMS replaces the links with radio. Choosing one trades cost, wiring weight, assembly time, isolation and what happens when one link fails.

How do modular and wireless BMSs connect their monitoring chips?

Each module's analogue front-end (AFE) chip measures its cells: TI's BQ79616 handles 6 to 16 cells to ±1.5 mV, ADI's LTC6813-1 up to 18. The chips pass data along a daisy chain: a two-wire isolated link (ADI calls it isoSPI) from board to board up the stack, so high-voltage modules can talk to a low-voltage master. For the car pack's 8 modules that is 16 wires across the pack instead of 137. Wireless BMSs drop even those: TI's CC2662R-Q1 is an automotive-grade 2.4 GHz radio chip made for this job. The cost is radio design and security.

How does CAN carry BMS data?

CAN is a two-wire bus shared by all the car's controllers. A classical frame carries an 11-bit ID, a 4-bit length and up to 8 data bytes, plus a 15-bit CRC, an acknowledge slot and framing: 108 bits before stuffing. After five equal bits the sender inserts an opposite stuff bit, so a frame can grow to 132 bits. The ID is also the priority: when two nodes start together, a 0 bit overrides a 1, so the lower ID wins without losing time. Powertrain buses often run at 500 kbit/s. CAN FD carries up to 64 bytes and can speed up the data phase.

How is a signal packed into a CAN frame and decoded?

A signal is a group of bits with a start bit, length, byte order, a factor and an offset, usually listed in a DBC file. The value is raw × factor + offset. Intel byte order puts the low byte first. In this book's original layout, bytes 60 0D are pack voltage: 0x0D60 = 3424, × 0.1 = 342.4 V. Bytes 18 FC are current, signed: 0xFC18 = 64,536, which as a 16-bit two's complement number is −1,000, × 0.1 = −100 A, a charge. Temperature uses an offset: raw 71 − 40 = 31 °C.

BMS architecture and CAN: centralised, modular and wireless layouts, and decoding a BMS message: interactive loads as you scroll

What are CANopen and FlexRay, and where do they fit?

CANopen, maintained by CAN in Automation (CiA 301), adds a standard object dictionary on top of CAN, with process data objects for fast values and service data objects for settings; industrial packs, forklifts and e-rickshaw controllers often use it. FlexRay is a time-triggered automotive bus at up to 10 Mbit/s with two channels for redundancy, used where timing must be guaranteed. A BMS might log 860 frames per second on CAN in a busy car pack, about 23 % of a 500 kbit/s bus in the worst case.

What tools do BMS engineers use: data loggers, cyclers and model-based design?

A CAN data logger such as CSS Electronics' CANedge1 records two CAN or LIN channels, including CAN FD, to an SD card for road and test-bench data. A battery cycler such as Arbin's LBT series charges and discharges many cells at once, 8 to 256-plus channels, to run capacity, HPPC and ageing tests. Model-based design builds the BMS algorithms as Simulink-type models, tests them against cell and pack models (model, software and hardware in the loop) and generates embedded C from them.

Predict, then check

The BMS (ID 0x2B4) and another controller (ID 0x0A0) start sending at the same instant. What happens?

Worked examples

Decode the car's charging message · Message 0x2B4 data bytes (byte 0 first): 60 0D 18 FC 5A 47 EC 08. Layout (Intel byte order): pack voltage u16 × 0.1 V; current s16 × 0.1 A (+ = discharge); SoC u8 × 0.5 %; max temperature u8 − 40 °C; min cell voltage 12 bits + 1,000 mV; 4 fault flags
  1. Voltage: low byte first, 0x0D60 = 3,424 → × 0.1 = 342.4 V
  2. Current: 0xFC18 = 64,536; above 32,767 so subtract 65,536: −1,000 → × 0.1 = −100 A (charging)
  3. SoC: 0x5A = 90 → × 0.5 = 45 %; temperature: 0x47 = 71 → 71 − 40 = 31 °C
  4. Min cell: byte 6 = 0xEC plus the low nibble of byte 7 (8) as the high bits: 0x8EC = 2,284 → + 1,000 = 3,284 mV
  5. Fault flags: top nibble of byte 7 = 0, no faults

Answer: 342.4 V, −100 A, 45 %, 31 °C, weakest cell 3.284 V, no faults

How busy does the BMS make the CAN bus? · The car pack's BMS sends 104 cell voltages, four 16-bit values per frame, every 100 ms, and 6 status frames every 10 ms, all 8-byte classical frames on a 500 kbit/s bus. Use the worst-case frame of 132 bits plus 3 bits of interframe space
  1. Cell frames = 104/4 = 26, × 10 per second = 260 frames/s
  2. Status = 6 × 100 = 600 frames/s; total 860 frames/s
  3. Bits = 860 × 135 = 116,100 bit/s
  4. Load = 116,100/500,000 = 23.2 %

Answer: 23.2 % of the bus in the worst case, from the BMS alone: one reason cell data moves to CAN FD, which packs 64 bytes (8 times more) per frame

Counting harness wires: centralised vs modular · Car pack: 104 cells in series as 8 modules of 13, 2 thermistors per module. Scooter: 14 groups, 4 thermistors. Each cell group needs a voltage tap (cells + 1 taps); each thermistor needs 2 wires; a daisy-chain link is 2 wires
  1. Centralised car: 105 taps + 16 × 2 = 137 wires from the modules to one board
  2. Modular car: taps stay on each module board; 8 links × 2 = 16 wires cross the pack
  3. Scooter centralised: 15 + 4 × 2 = 23 wires, all short; one 16-cell AFE covers 14 groups
  4. A centralised car board would need 7 AFE chips of 16 cells anyway

Answer: 137 vs 16 wires for the car (wireless: 0), so cars go modular; the scooter stays centralised with 23

Chapter 20

Functional safety and standards: ISO 26262 ASIL, AIS-156, AIS-038, IEC 62619, UN 38.3

Which syllabus covers thisAU not in syllabusVTU not in syllabusSPPU 302036MJMU HCEV801 Module 4KTU not in syllabus

What is functional safety, and how does a hazard get an ASIL?

Functional safety means the electronics fail in a safe way. ISO 26262 starts with a hazard analysis: for each malfunction, rate Severity (S0 no injury to S3 fatal), Exposure (E0 incredible to E4 almost every drive) and Controllability (C0 to C3, where C3 means under 90 % of drivers avoid harm). A table turns the three into an Automotive Safety Integrity Level: QM (normal quality work), then A to D. Of the table's 36 cells, 1 is D, 3 are C, 6 are B, 8 are A and 18 are QM; S+E+C = 10 gives D, 9 C, 8 B, 7 A. A BMS missing an over-voltage while charging can rate S3, E4, C3: ASIL D.

Functional safety: rating a battery hazard by severity, exposure and controllability to get its ASIL: interactive loads as you scroll

What are a safety goal, a safe state and the fault-tolerant time?

Each rated hazard gets a safety goal, such as "stop charging before any cell passes 4.25 V" (the LG M50LT's over-voltage limit). The safe state is what the system falls back to: charger off, contactors open. The fault-tolerant time interval (FTTI) is how long a fault can last before it becomes dangerous. Detection plus reaction must fit inside it. A BMS sampling every 100 ms, confirming over 3 samples, deciding in 10 ms and opening a contactor in 50 ms needs 360 ms. A charger stuck at 3.36 A overfills a 5 Ah cell by only 0.0067 % in that time.

What changes as the ASIL rises, and what is ASIL decomposition?

A higher ASIL demands more rigour: stricter reviews, more testing and hardware that catches its own faults. A BMS meant for ASIL C or D usually measures cell voltage twice by independent paths, for example the AFE plus a separate over-voltage comparator, and checks one against the other. NXP states its MC33771C cell monitor (14 cells per chip) supports systems up to ASIL D. Decomposition splits a requirement over two truly independent parts at a lower level: D can become B(D) + B(D) or C(D) + A(D); two headlights at A(B) make an ASIL B function. Independence must be shown, with no shared power or clock.

Which Indian standards govern EV batteries, their BMS and chargers?

AIS-156 covers two- and three-wheelers (L category) and AIS-038 Rev 2 cars, buses and trucks; both were adapted from UN Regulation 100 (high voltage above 60 V DC, insulation at least 100 Ω per volt). After the 2022 scooter fires and recalls of about 6,700 scooters, Amendment 3 (Phase 1 from 1 Dec 2022, Phase 2 from 31 Mar 2023) added: cells certified to IS 16893, an IPX7 test at 100 % SoC, a BMS proven to stop over-charge, over-discharge, over-temperature, over-current and short circuit, a thermal propagation test with a warning at least 5 minutes before fire, and for two-wheelers 4 or more temperature sensors with an alarm above 60 °C. Chargers need cut-off, soft start and earth-leakage detection.

What do UN 38.3 and IEC 62619 test?

UN 38.3 must be passed before lithium cells or packs are shipped: T1 altitude (50,000 ft), T2 thermal cycling (12 h at −40 °C and at 72 °C, 10 times, so 240 h of dwell), T3 vibration, T4 shock, T5 external short circuit at 57 °C, T6 impact (a 9.1 kg mass from 61 cm, about 54 J), T7 overcharge at twice the recommended current and T8 forced discharge. IEC 62619 is the safety standard for industrial and stationary lithium batteries, such as telecom and home storage: short circuit, impact, drop, thermal abuse, overcharge, forced discharge, internal short, thermal propagation and the BMS's control of voltage, current and temperature.

Predict, then check

A hazard is rated S3, E4, C3, so ASIL D. A redesign makes it simply controllable (C1) without changing S or E. The new rating is:

Worked examples

Rating a contactor that opens at highway speed · The car's main contactor opens unexpectedly while driving at 100 km/h. The team rates it S2 (severe injury, survival likely), E4 (highway driving on almost every trip for this car) and C2 (90 % or more of drivers can coast to the shoulder). Illustrative ratings
  1. Look up S2, E4, C2 in the ISO 26262-3 table: ASIL B
  2. Check by the sum rule: 2 + 4 + 2 = 8 → ASIL B
  3. If the car mostly runs in the city (E3, about once a month at highway speed): 2 + 3 + 2 = 7 → ASIL A
  4. If a rear-end crash at that speed is judged life-threatening (S3): 3 + 4 + 2 = 9 → ASIL C
  5. Decomposition of B: A(B) + A(B), for example two independent contactor drivers that must both fail

Answer: ASIL B; the rating moves to A or C with the assumptions, which is why the HARA records its reasoning

Is the BMS fast enough for a stuck charger? · A 5 Ah NMC cell is charged at 3.36 A, the LG M50LT datasheet's maximum charge current (0.7C of that cell's 4.8 Ah rating), and the charger fails to stop. The BMS samples every 100 ms, needs 3 samples over the limit, takes 10 ms to decide and the contactor opens in 50 ms
  1. Reaction time = 3 × 100 + 10 + 50 = 360 ms (the fault can start just after a sample)
  2. Overcharge rate = 3.36 / (5 × 3600) = 1.8667×10⁻⁴ per second = 0.0187 % of capacity per second
  3. Overcharge before the contactor opens = 0.0187 × 0.36 = 0.0067 %
  4. Time to overfill by 1 % = 1 / 0.0187 = 53.6 s
  5. A slow BMS sampling once a second: 3 × 1000 + 60 = 3060 ms → 0.057 %

Answer: 360 ms and 0.0067 % overcharge: a slow overcharge leaves tens of seconds, so even the slow BMS is in time; a hard short circuit needs a fuse instead

Planning a UN 38.3 run for the scooter pack · The book's scooter pack, 14S12P of 5 Ah NMC cells, must ship to a test lab. Use the UN 38.3 conditions as listed by SwRI and the Samsung 50E's standard charge current of 2.45 A per cell
  1. T2 thermal: 10 cycles × (12 h at −40 °C + 12 h at 72 °C) = 240 h, 10 days of dwell
  2. T6 impact energy = m g h = 9.1 × 9.81 × 0.61 = 54.5 J (applied to cells)
  3. T7 overcharge current = 2 × 2.45 = 4.9 A per cell; with 12 cells in parallel, 58.8 A for the pack
  4. T1 to T5 run in sequence on the same samples, so the pack must survive altitude, heat, vibration and shock before its short-circuit test

Answer: About 10 days for T2 alone, 54 J impacts and a 58.8 A overcharge: plan the test samples and time early

Chapter 21

Sizing a pack and its BMS: energy, power, voltage, cells, mass and fuse

Which syllabus covers thisAU AU3002 Unit IIVTU not in syllabusSPPU 302031MJ Unit 1MU HCEV801 Module 4KTU PEEET521 Module 2

How do you size a battery pack, step by step?

Follow a flow chart. 1: write the requirements (range, energy per km, peak power, system voltage). 2: energy needed = range × Wh/km ÷ usable window. 3: cells in series S = system voltage ÷ cell voltage, rounded up. 4: cells in parallel P = the larger of the energy need and the power need. 5: check the result: current per cell, mass, fuse, BMS channels. The book's scooter needs 90 km × 30 Wh/km ÷ 0.9 = 3000 Wh. With 14 cells in series each parallel column holds 254.1 Wh, so energy asks for 12P and a 5 kW peak asks for 11P: 14S12P.

Sizing a pack and its BMS: range, power and voltage to cells in series and parallel, mass and fuse rating: interactive loads as you scroll

How is the pack voltage chosen?

Higher voltage means less current for the same power, so thinner cables and smaller losses, but more cells in series and stricter safety rules. UN R100 treats anything above 60 V DC as high voltage. Two-wheelers use 48, 60 or 72 V classes: the book's 14S scooter pack reaches 58.8 V when full, just under 60 V. Cars use 300-400 V: the book's 104S car pack is 332.8 V nominal and 379.6 V full, drawing 165 A for 55 kW. An 800 V car would draw 83 A for the same power, which also allows faster charging.

Where does the energy-per-km number come from?

From the vehicle's mass, drag, speed and driving pattern, or from a test. Certified test cycles are gentle: the Ather 450X's 3.7 kWh and 146 km certified range work out to 25.3 Wh/km, and the Tata Tiago.ev's 24 kWh and 315 km to 76.2 Wh/km. Real riding with traffic, heat and air-conditioning uses more, so this book assumes 30 Wh/km for its scooter and 130 Wh/km for its car. The usable window matters too: a BMS that keeps 5 % at each end leaves 90 % of the nameplate energy for driving.

What are specific energy and specific power, and when does power set the pack?

Specific energy is Wh per kg; specific power is W per kg. The 5 Ah NMC cell holds 263 Wh/kg and the 100 Ah LFP cell 162 Wh/kg. Packs add casing, cooling and wiring: with cells at 70 % of pack mass (assumed) the scooter pack is 184 Wh/kg. Small, powerful packs end up power-limited: doubling the scooter's peak to 10 kW lifts P from 12 to 21. A battery-ultracapacitor hybrid lets an ultracapacitor, with far more W/kg but little Wh/kg, cover the peaks so the battery can be sized for energy.

How are the BMS channels, sensors and fuse sized?

Every series group needs a voltage channel. TI's BQ76952 monitors 3-16 cells, so the 14S scooter needs one chip; the 104S car needs 6 ADI LTC6813-1 chips (18 cells each) or 8 NXP MC33771C (14 each). AIS-156 asks two-wheeler packs for at least 4 temperature sensors. The main fuse is usually rated about 1.25 times the continuous current: 12 × 7.3 A = 87.6 A for the scooter, so the next standard rating, 110 A. It must still carry the 98 A peak: a standard fuse may need twice its rating to blow within a second.

Predict, then check

The scooter keeps its 90 km range but its peak power is doubled from 5 kW to 10 kW. The pack becomes:

Worked examples

The book's scooter pack from its requirements · Range 90 km at 30 Wh/km (assumed), usable window 90 %, peak 5 kW for 30 s, system voltage at least 48 V. Cell: 5 Ah NMC 21700, 3.63 V, 69 g, peak 9.8 A (Samsung 50E maximum continuous), continuous 7.3 A (assumed)
  1. Energy needed = 90 × 30 / 0.9 = 3000 Wh
  2. S = 48 / 3.63 = 13.22 → 14 (50.82 V nominal, 58.8 V full)
  3. Energy per parallel column = 14 × 3.63 × 5 = 254.1 Wh → P for energy = 11.81 → 12
  4. Peak current = 5000 / 50.82 = 98.4 A → P for power = 10.04 → 11
  5. 14S12P = 168 cells, 3.049 kWh, cells 11.59 kg, pack ≈ 16.6 kg at 0.70
  6. Fuse ≥ 1.25 × 12 × 7.3 = 109.5 A → 110 A; one BQ76952 AFE

Answer: 14S12P, 168 cells, 3.05 kWh, about 16.6 kg, 110 A fuse: energy, not power, sets this pack

The book's car pack from its requirements · Range 230 km at 130 Wh/km (assumed), usable 90 %, peak 55 kW (the Tiago.ev's 74 hp), about 330 V. Cell: 100 Ah LFP prismatic, 3.2 V, 1.98 kg, peak 200 A (2C for 30 s, as the EVE LF280N), continuous 100 A
  1. Energy needed = 230 × 130 / 0.9 = 33222 Wh
  2. S = 330 / 3.2 = 103.12 → 104 (332.8 V)
  3. One column holds 104 × 3.2 × 100 = 33280 Wh ≥ 33222 Wh → P = 1
  4. Peak current = 55000 / 332.8 = 165.3 A ≤ 200 A → P = 1
  5. 104 cells, 33.28 kWh, cells 205.9 kg, pack ≈ 294 kg; needs 187 W/kg at peak
  6. AFE: 104 / 18 = 5.78 → 6 × LTC6813-1; fuse ≥ 1.25 × 100 → 125 A

Answer: 104S1P, 33.28 kWh, about 294 kg, 6 AFE chips, 125 A fuse

An e-rickshaw at 48 V: big LFP cells or small NMC cells? · 80 km a day at 50 Wh/km (assumed), peak 2 kW, at least 48 V, usable 90 %. Compare the 100 Ah LFP prismatic with the 5 Ah NMC 21700
  1. Energy needed = 80 × 50 / 0.9 = 4444 Wh
  2. LFP: S = 48 / 3.2 = 15; one column = 15 × 320 = 4800 Wh → 15S1P, 4.8 kWh, 29.7 kg of cells
  3. NMC: S = 14; one column = 254.1 Wh → P = 17.49 → 18, so 14S18P = 252 cells, 4.57 kWh, 17.4 kg
  4. Specific energy of the cells: LFP 162 Wh/kg, NMC 263 Wh/kg
  5. Peak current 2000 / 48 ≈ 42 A is easy for both; the LFP pack has 15 cells to monitor, the NMC pack 252 cells (14 channels)

Answer: LFP: 15 cells, 29.7 kg; NMC: 252 cells, 17.4 kg. NMC is lighter, LFP simpler and longer-lived

Chapter 22

Second life, recycling and India's battery rules

Which syllabus covers thisAU not in syllabusVTU not in syllabusSPPU 302031MJ Unit 6MU not in syllabusKTU not in syllabus

What is a battery's second life?

A car pack leaves the car when its range has shrunk too far, yet it still holds most of its energy. Stationary storage needs no acceleration and little space, so the pack can store solar power for a home or back up a telecom tower. The book's 33.3 kWh car pack retired at 75 % SoH, used between 10 % and 90 % SoC (as the EVE LF280N datasheet recommends), offers 20 kWh. A home taking 6 kWh a day cycles 24 % of it daily. With the fade assumed here it lasts 9.6 years before reaching 60 % and delivers 21.1 MWh.

Second life and recycling: a retired car pack as home or grid storage, and the metals a recycler gets back: interactive loads as you scroll

Why does a second-life pack wear out faster as it ages?

Capacity is lost to SEI growth, lithium plating and particle cracking (chapter 11), both with time and with cycles. In second life the daily need stays fixed while the capacity shrinks, so every day becomes a deeper cycle. In the home example the pack loses 1.51 SoH points in its first year but 1.62 in its last. A straight-line estimate that ignores this gives 10 years instead of 9.6. Real aged cells can also reach a knee and fade suddenly, so a second-life BMS must keep measuring capacity and resistance. At the end, India's rules forbid sending collected waste batteries to landfill or incineration.

How are retired cells sorted and graded?

Repurposers first check the pack's logged history from its BMS, then test modules or cells: a full capacity test (a 0.5C discharge of a 100 Ah cell takes about 2 hours), a DC resistance pulse like the HPPC test of chapter 6, and a self-discharge check after a rest. Cells far from the group, with low capacity, high resistance or fast self-discharge, are rejected, because one weak cell limits a whole series string (chapter 4). The rest are regrouped with similar cells and given a new BMS, tuned for slow daily cycles. India's 2022 rules name this "refurbishment": re-purposing a used battery for its second life.

What does recycling recover, and why do recyclers prefer NMC?

Pyrometallurgy smelts cells and recovers nickel, cobalt and copper as an alloy; lithium tends to be lost to slag. Hydrometallurgy dissolves the shredded black mass in acid and recovers each metal, lithium included. From the PyBaMM electrode data, an NMC811 cell holds about 109 g lithium, 843 g nickel and 106 g cobalt per kWh; LFP holds about 83 g lithium and 931 g iron, which is cheap. At assumed yields (80 % Li, 95 % Ni and Co), a 33.3 kWh NMC pack gives 32.9 kg of these metals and the LFP pack only 2.2 kg.

What do India's Battery Waste Management Rules 2022 require?

Producers carry Extended Producer Responsibility: they must collect and recycle or refurbish a share of the batteries they sell, proven by certificates traded on the CPCB portal. For EV batteries, collection starts in 2024-25 for e-rickshaws, 2026-27 for two-wheelers and 2029-30 for cars: 70 % of the batteries sold 3, 4 and 8 years earlier. Recyclers must recover 70 % of a battery's dry weight in 2024-25, 80 % in 2025-26 and 90 % from 2026-27. New EV batteries must contain 5 % recycled material in 2027-28, rising to 20 % from 2030-31. Shortfalls attract environmental compensation.

How will batteries be tracked: passports and swapping?

From 18 February 2027 the EU requires a digital battery passport for every EV battery, light-vehicle battery and industrial battery above 2 kWh: a machine-readable record of its identity, materials, performance and environmental impact, reached through a code on the battery. That data makes second-life grading easier. NITI Aayog's draft battery swapping policy of 20 April 2022, for two- and three-wheelers, likewise asks for a unique ID on every battery and station. Swapped packs are charged off the vehicle and shared by many riders, so their BMS data decides when each one moves to a second life.

Predict, then check

A retired pack supplies the same 6 kWh every day. Compared with its first year, in its last year of second life it loses:

Worked examples

The book's car pack as home storage · The 33.28 kWh LFP car pack retires at 75 % SoH. Window 80 % (10-90 % SoC). A home needs 6 kWh a day. Fade: 20 % over 3,500 equivalent full cycles (EVE LF280N) plus 1 % a year calendar fade (assumed); end at 60 %
  1. Capacity now = 33.28 × 0.75 = 24.96 kWh; usable = × 0.8 = 19.97 kWh
  2. Daily depth = 6 / 24.96 = 24 %; 87.7 full cycles a year
  3. dS/dt = −(a/S + b), a = 365 × 6 × 0.2/3500 / 33.28 = 3.76×10⁻³, b = 0.01
  4. t = (S0 − S1)/b − (a/b²) ln((a + b S0)/(a + b S1)) = 9.62 years from 75 % to 60 %
  5. Energy delivered = 6 × 365 × 9.62 = 21.1 MWh

Answer: 20 kWh usable, about 9.6 years and 21 MWh of second life

A telecom tower: LFP or NMC? · A tower needs 14 kWh a day (assumed). Compare the LFP car pack with an NMC pack of the same 33.28 kWh, both at 75 % SoH. NMC fade: 20 % over 500 cycles (Samsung 50E); calendar 1 % a year for both
  1. Daily depth = 14 / 24.96 = 56.1 %, which fits the 80 % window
  2. Full cycles a year = 365 × 0.561 = 205
  3. LFP: a = 365 × 14 × 0.2/3500 / 33.28 = 0.0088 → 6.52 years to 60 %
  4. NMC: a = 365 × 14 × 0.2/500 / 33.28 = 0.0614 → 1.48 years
  5. The window still holds a day's need down to 52.6 % SoH, so 60 % is the end in both cases

Answer: LFP lasts about 6.5 years, NMC about 1.5: the datasheet cycle life decides second life

Metals in the car pack and the recycling target · Electrode inventories from the PyBaMM parameter sets: per kWh, NMC811 109 g Li, 843 g Ni, 106 g Co, 99 g Mn; LFP 83 g Li, 931 g Fe. Assumed yields 80 % Li, 95 % Ni and Co. Cell masses: 1.98 kg per 100 Ah LFP cell, 69 g per 5 Ah NMC cell
  1. LFP pack (33.28 kWh): Li = 83.2 × 33.28 / 1000 = 2.77 kg; Fe = 31 kg
  2. NMC pack: Li 3.61 kg, Ni 28.1 kg, Co 3.52 kg, Mn 3.28 kg
  3. Recovered Li + Ni + Co: LFP 2.21 kg; NMC 3.61×0.8 + 28.1×0.95 + 3.52×0.95 = 32.9 kg
  4. Rules from 2026-27: recover 90 % of the dry weight. LFP cells 104 × 1.98 = 205.9 kg → 185.3 kg; NMC cells 126.5 kg → 113.9 kg
  5. So the rules push recyclers to recover graphite, copper, aluminium and the cathode's other elements too, not just the valuable metals

Answer: NMC: 32.9 kg of Li, Ni and Co; LFP: 2.2 kg of lithium. Either way 185 kg (LFP) or 114 kg (NMC) of material must be recovered

Keep going

Recommended reading

  • Battery Management Systems, Volume I: Battery Modeling, G. L. Plett
  • Battery Management Systems, Volume II: Equivalent-Circuit Methods, G. L. Plett
  • Battery Management Systems for Large Lithium-Ion Battery Packs, D. Andrea
  • Lithium-Ion Batteries: Basics and Applications, R. Korthauer (ed.)
  • Linden's Handbook of Batteries, K. W. Beard (ed.)

How this book was made

Chapters follow the official syllabi listed above. Every interactive's formulas and every worked-example number were checked by script. Explanations, animations and examples are original to Skill-Lync.

From the exam to a core job

Talk to a counsellor.

Ask how battery and BMS engineers use battery management systems on real projects, and what the Master's Certification Program in Battery Technology covers.

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.