shewhart 0.0.1

Statistical process control (SPC) for Python: control charts, process capability, and measurement systems analysis. Validated, pandas-native, automation-first.

shewhart

Statistical process control (SPC) for Python — validated, pandas-native, automation-first.

Named after Walter A. Shewhart, the father of statistical process control.

Why

Quality engineering is the most under-served statistical vertical in Python. R has had qcc since 2004; Python has a graveyard of abandoned fragments. Nothing does proper ANOVA Gauge R&R, capability indices ship without confidence intervals, and none of it is built for the way modern process data actually lives: in pandas, in databases, in pipelines that should produce Monday-morning reports without anyone clicking a mouse.

shewhart is being built to close that gap:

• Control charts that respect Phase I / Phase II separation, with Nelson and Western Electric rule engines
• Process capability (Cp/Cpk/Pp/Ppk) with honest confidence intervals and non-normal methods
• Measurement systems analysis — full ANOVA Gauge R&R the way the AIAG manual defines it
• One-call HTML reports — your weekly control chart review as a cron job
• A public validation suite against NIST/SEMATECH reference datasets

Status

v0.0.1 — first working primitives while the full v0.1 is under active development (target: 2026). The API below is real and tested; everything else is on the roadmap.

import shewhart as sw

limits = sw.imr_limits([10.2, 10.4, 10.1, 10.5, 10.3, 10.2, 10.6])
# {'i_center': 10.329, 'i_lcl': 9.531, 'i_ucl': 11.126, ...}

flags = sw.beyond_limits([10.2, 10.4, 12.9, 10.5], limits)
# [False, False, True, False]

Roadmap

Version	Scope
v0.1	Shewhart chart family (I-MR, X̄-R, X̄-S, p/np/c/u), EWMA, CUSUM, rule engines, capability with CIs, HTML reports, NIST validation suite
v0.2	Measurement systems analysis: crossed/nested ANOVA Gauge R&R, Type 1 studies, attribute agreement
v0.3	monitor — drift and change-point detection with control-chart semantics (ARL-calibrated) for sensor streams and ML-model outputs
v0.4	Multivariate: Hotelling T², MEWMA, PCA-based monitoring

Out of scope by design: DOE (see pyDOE3), reliability (see reliability), general statistics (see statsmodels, pingouin), GUIs.

Author

Built by Bertan Ucar — PhD researcher in AI-driven quality engineering at Tsinghua University.

License

MIT