Skip to content

shewhart

Statistical process control for Python. Control charts with the standard run rules, process capability analysis with confidence intervals, and (coming in 0.2) measurement systems analysis. Every released number is checked against published reference values, including NIST-certified datasets, on every CI run.

pip install git+https://github.com/bertanucar/shewhart

Sixty seconds

import shewhart as sw

r = sw.imr(df, value="torque", rules="nelson")
r.ok           # False if any rule fired
print(r.summary())

shewhart imr - n=10 - rules=nelson - fitted (Phase I) - v0.1.0.dev0
  i_center=10.7  i_lcl=9.046  i_ucl=12.35  mr_center=0.6222  mr_lcl=0  mr_ucl=2.033  sigma_within=0.5514
verdict: OUT OF CONTROL - 3 signal(s):
  - nelson_1 (i): point 9 - 1 point beyond 3 sigma
  - nelson_2 (i): points 0-8 - 9 in a row on one side of center
  - beyond_limits (mr): point 9 - moving range beyond UCL

Fit limits once, monitor forever:

sw.imr(df_baseline, value="torque").baseline.save("line3_baseline.json")

# in the weekly job:
r = sw.imr(df_new, value="torque", limits="line3_baseline.json")
sys.exit(0 if r.ok else 1)

Why this library

  • Correct. Chart constants are computed from their defining integrals; the published tables, NIST StRD certified values, and NIST/SEMATECH handbook examples are reproduced in the test suite. See Validation.
  • Phase I and Phase II are different things. Estimating limits and monitoring against frozen limits are separate operations, and baselines serialize to JSON you can commit next to your code.
  • Built for pipelines. Rule violations are structured data, r.ok is an exit code, and reports are single self-contained HTML files.

Where to start