GitHub - remontsuri/EV-QA-Framework: ML-powered QA framework for EV battery systems — telemetry validation, anomaly detection, SOH prediction, CAN bus (2.0B + J1939) emulation, DBC parser, Prometheus metrics, Grafana dashboard

ML-powered QA framework for electric vehicle battery systems. Validates BMS telemetry, detects anomalies, predicts SOH degradation, emulates CAN bus traffic, and evaluates thermal runaway risk — MIT licensed.

What it does

Telemetry validation. Pydantic schemas for voltage, current, temperature, SOC, SOH. Catches bad VINs, out-of-range values at the input layer.

ML anomaly detection. Isolation Forest on voltage/current/temperature streams. Configurable contamination, severity thresholds, and number of estimators.

SOH prediction. LSTM-based State of Health forecasting from historical telemetry (TensorFlow optional).

Cell imbalance analysis. Statistical analysis of cell group voltages with configurable thresholds, outlier detection, linear regression trend, and plot export.

Thermal runaway prediction. Standalone ThermalRunawayPredictor with two modes:

rule — configurable heuristic with adjustable weights (dT/dt, temperature, anomaly score)
ml — Isolation Forest on thermal features CRITICAL trigger at >65°C or heating rate >5°C/min.

CAN bus. CAN 2.0B (11-bit ID) and J1939 (29-bit extended) simulation and reception. DBC parser supports Vector CANdb format, SavvyCAN exports, Intel/Motorola byte order, signed/unsigned signals.

Dashboard. FastAPI + WebSocket + Chart.js. Real-time telemetry and Prometheus /metrics endpoint with ready-to-import Grafana dashboard.

CLI. Analyze CSV telemetry, run CAN emulation, train SOH models, start dashboard.

Quick start

# Install from GitHub
pip install git+https://github.com/remontsuri/EV-QA-Framework.git

# Launch dashboard
python -m ev_qa_framework.cli dashboard
# → http://localhost:8000
# → http://localhost:8000/metrics (Prometheus)

# Analyze a CSV
python -m ev_qa_framework.cli analyze -i examples/tesla_model_s_defective.csv -o report.json

# CAN simulation from DBC
python -m ev_qa_framework.cli emulate --dbc my_battery.dbc --duration 60

# Run tests
python -m pytest -v

Examples

Telemetry validation:

from ev_qa_framework.models import validate_telemetry

data = {
    "vin": "1HGBH41JXMN109186",
    "voltage": 396.5,
    "current": 125.3,
    "temperature": 35.2,
    "soc": 78.5,
    "soh": 96.2
}
telemetry = validate_telemetry(data)

Anomaly detection:

from ev_qa_framework.analysis import AnomalyDetector
import pandas as pd

df = pd.read_csv("battery_telemetry.csv")
detector = AnomalyDetector(contamination=0.01, n_estimators=200)
detector.train(df[["voltage", "current", "temperature"]])
predictions, scores = detector.detect(new_data)

Cell imbalance:

from ev_qa_framework.cell_balance import CellBalanceAnalyzer

analyzer = CellBalanceAnalyzer(warning_threshold=0.02, critical_threshold=0.05)
cell_v = [3.30, 3.31, 3.305, 3.312, 3.29]
print(analyzer.compute_statistics(cell_v))
print(analyzer.classify_severity(cell_v))

Thermal runaway (recommended API):

from ev_qa_framework.thermal_runaway import ThermalRunawayPredictor
import pandas as pd

predictor = ThermalRunawayPredictor(mode="rule")
df = pd.DataFrame({"temperature": [35, 37, 42, 58, 62]})
risk = predictor.predict_risk(df)
# {'risk_level': 'HIGH', 'risk_score': 8.3, 'confidence': 0.85, ...}

DBC parsing:

from ev_qa_framework.dbc_parser import DBCParser

dbc = DBCParser("tesla_battery.dbc")
msg = dbc.get_message(0x101)
vals = dbc.decode(0x101, bytes([0x7D, 0x0F, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]))
# {'Voltage': 396.5}

Project structure

ev_qa_framework/
  framework.py         # core QA engine
  models.py            # Pydantic models
  config.py            # thresholds and ML config
  analysis.py          # Isolation Forest, EVBatteryAnalyzer
  soh_predictor.py     # LSTM for SOH (TensorFlow optional)
  can_bus.py           # CAN 2.0B + J1939 simulation
  dbc_parser.py        # .dbc file parser (Vector CANdb + SavvyCAN)
  cell_balance.py      # cell voltage imbalance analysis
  thermal_runaway.py   # thermal runaway prediction (rule + ML)
  metrics.py           # Prometheus metrics
  cli.py               # CLI entry point
dashboard/
  app.py               # FastAPI
  grafana/             # Grafana dashboard JSON
tests/                 # 160+ tests

Development

# Clone and install dev dependencies
git clone https://github.com/remontsuri/EV-QA-Framework.git
cd EV-QA-Framework
pip install -e .[dev,ml]

# Run linting
ruff check .

# Run tests
pytest -v

Changelog

v1.1.0

Thermal runaway deduplicated — ThermalRunawayPredictor is the single API (removed duplicate from EVBatteryAnalyzer)
Fixed risk score calculation: temperature contribution uses deviation from 50°C, not absolute value
CLI analyze now handles both temperature and temp column names
Migrated setup.py → pyproject.toml, added uv.lock
Applied ruff auto-fixes across the codebase
Fixed BatteryCellDataModel import in package __init__.py
Fixed SOHPredictor type hint (Sequential → Any)
Fixed example in framework.py (__main__) — uses pack voltage (396V) instead of cell voltage (3.9V)
Removed stale build/ artifacts

Compatibility

CAN 2.0B (11-bit) and J1939 (29-bit extended)
SavvyCAN / BUSMASTER DBC exports
Prometheus + Grafana
TensorFlow — optional (SOH prediction only)
python-can — only needed for physical CAN hardware; simulation works without it

License

MIT

推荐订阅源

Show HN