Skip to main content

Open data · CC-BY 4.0

Tesla Model 3 2170 cell dataset

Four cylindrical Tesla/Panasonic 2170 NCA cells from a Model 3 pack, tested at six temperatures with characterization tests and drive cycles generated by a validated Tesla Model 3 vehicle model. The open portion is yours to train on; the blind portion only ever runs inside the evaluator.

Four cells, four vehicle configurations

Each cell was cycled with a different vehicle configuration so the dataset covers payload and HVAC extremes.

CellPayloadHVACCycler channelRole
m8080 kgOnCh 1Open + blind cycles
m448448 kgOnCh 2Blind — evaluation only
m448-N448 kgOffCh 3Open + blind cycles
m10001000 kgOnCh 4Open + blind cycles

Vehicle model: 1612 kg curb mass + payload, EPA road-load coefficients, 96s31p pack (2,976 cells, 50.4 kWh), 250 W accessory load, temperature-dependent HVAC power. Predicted EPA range 346 km vs 354 km official.

Test equipment

  • Cycler: Arbin LBT, 8 × 60 A channels, ±2 mV / ±24 mA control accuracy
  • Chamber: Envirotronics SH16C, 16 ft³, ±0.3 °C
  • Sampling: resampled to 1 Hz
  • Capacity at start: ≈ 4.7 Ah (4.5 Ah nominal)
  • Chemistry: NCA (presumed)

Tests by Fauzia Khanum and Mina Naguib with Dr. Phillip Kollmeyer, McMaster University.

Open versus blind data

Tests run at −20, −10, 0, 10, 25 and 40 °C, hottest first. Before every test the cell is returned to 25 °C and CC-CV charged to 4.2 V / 0.1 A so all tests start at 100 % SOC.

Testm80m448m448-Nm1000
C/20 discharge capacity (at 40 °C)OpenBlindOpenOpen
C/3, C/2, 1C dischargeOpenBlindOpenOpen
C/20 charge and dischargeOpenBlindOpenOpen
HPPC (four pulse magnitudes)OpenBlindOpenOpen
UDDS, HWFET, LA92, US06BlindBlindBlindBlind
Eight reordered drive cyclesOpenBlindOpenOpen
HWCUST 1 / 2Open / BlindBlindOpen / BlindOpen / Blind
HWGRADE 1 / 2Open / BlindBlindOpen / BlindOpen / Blind

Drive cycles repeat until pack power capability (from HPPC, scaled to 2,976 cells) drops below 60 kW — the point a vehicle would enter a power-limited mode. At −20 °C this removes roughly a third of the usable capacity. Reference SOC uses the C/20 capacity measured at 40 °C before each temperature block, linearly interpolated across the tests in between.

File format

Every .mat file loads as a struct named meas with these columns.

TimeStampMM/DD/YYYY HH:MM:SS AM
Times
VoltageV, cell terminal
CurrentA, negative = discharge
AhAh counter (reset each test)
SOCreference SOC from interpolated 40 °C capacity
PowerW
WhWh counter
Battery_Temp_DegCcase temperature, mid-cell
Ambient_temp_degCchamber set-point
Cycle_Labele.g. UDDS
Test_Celle.g. m80

Files are named <order>_<test>_<date> Tesla_<temp>_<DriveCycles|Char>_Ch<n>.mat, e.g. 10_REORDERED1_11-08-21_15.03 Tesla_n10C_DriveCycles_Ch1.mat.

What's in the download

Download individual archives from Borealis.

  • 0-Documentation.zip

    Submission guidelines, evaluation description, output-data reference, example-model walkthroughs and aiding-tool guides (five PDFs).

    5.6 MB
  • 1-Open Data.zip

    All open characterization tests and drive cycles for m80, m448-N and m1000 at six temperatures, as MATLAB .mat structs (1 Hz).

    705 MB
  • 2-Aiding Tools.zip

    Model Submission Test Tool, file Concatenation Tool and Figure Data Extraction Tool (MATLAB live scripts).

    10.4 MB
  • 3-Neural Network Training Example.zip

    FNN and LSTM training scripts with instructions for extracting weights into a Model.m.

    64.5 MB
  • 4-SOC estimation Model Examples.zip

    Four complete submission packages: coulomb counter, EKF, FNN and LSTM — the reference for the required format.

    54 KB

Known data notes (from the dataset README, v2, 2026-06)

  • HWGRADE2 was identical to HWGRADE1 at 40 °C and 25 °C; it differs at the remaining temperatures.
  • The −10 °C C/20 discharge/charge test and the −20 °C HPPC test were re-run after the characterization block; their timestamps differ.
  • 2026-06-26: m1000 −20 °C standard cycles and HWCUST2/HWGRADE2 moved from open to blind for consistency with the paper; two missing m80 25 °C characterization files were added.

Use of the data requires citing the ITEC 2022 paper and the Borealis dataset — see the methodology page.