Skip to content

Quickstart

This walks the whole load → spec → analysis flow on a small worked dataset, with the actual printed output so you know what to expect.

1. Load a frame and attach metadata

mfgQC analyses consume a QCData object. You build one by load-ing a tidy DataFrame and naming the measurement column, then attaching spec limits fluently:

import numpy as np, pandas as pd, mfgqc

rng = np.random.default_rng(7)
df = pd.DataFrame({
    "width": np.round(rng.normal(1.52, 0.12, size=100), 3),
    "lot":   np.repeat(np.arange(1, 21), 5),     # 20 subgroups of 5
})

qc = (mfgqc.load(df, measure="width", subgroup="lot", subgroup_size=5)
           .spec(lower=1.0, upper=2.0, target=1.5))

measure= names the value column; subgroup= names the column that groups measurements into rational subgroups; .spec(...) attaches the engineering tolerances. Everything is immutable — each call returns a new object.

2. Capability

print(qc.capability())

Process Capability (method=normal)
==================================
n = 100   mean = 1.4992
sigma (within)  = 0.1105
sigma (overall) = 0.10556
Cp/Cpk sigma    = within (R-bar/d2)

Cp  = 1.508  95% CI (1.3, 1.72)
Cpk = 1.506  95% CI (1.29, 1.73)   (Cpu=1.51, Cpl=1.506)
Pp  = 1.579    Ppk = 1.576   (Ppu=1.581, Ppl=1.576)
Cpm = 1.579

Assumption checks:
  [PASS] normality (Anderson-Darling): AD=0.301, p=0.572; est. Cpk impact 5.0%; n=100
  [FAIL] subgroup_sufficiency (subgroup count >= 25): subgroup count 20; n=20

Recommendations:
  - Only 20 subgroups; >=25 recommended for a stable within-sigma estimate.

Notice three things that make this more than a Cpk calculator:

  • Both σ families are reported — within-subgroup (Cp/Cpk) and overall (Pp/Ppk) — with the estimator named (R-bar/d2). See Capability.
  • Confidence intervals are shown because small-sample point estimates are overconfident.
  • The assumptions are checked and reported. Here normality passes but the subgroup count is flagged as low — mfgQC tells you, and recommends a fix, but does not silently change the calculation. See Reading the assumption report.

3. Control chart

With no kind=, mfgQC infers the right chart from the subgroup size (here 5 → X̄-R):

print(qc.control_chart())

Control Chart: xbar_r (inferred); rules=nelson
==============================================
Xbar: CL=1.4992  UCL=1.6475  LCL=1.3509
R: CL=0.25705  UCL=0.5434  LCL=0

Out-of-control signals: none (process in control)

Assumption checks:
  [PASS] independence (lag-1 autocorrelation): r=0.193, p=0.387; n=20 [low power]

See Choosing a control chart for the inference rule and how to override it.

4. Every result has the same surface

cap = qc.capability()

cap.report()              # the full text above
cap.summary()             # {'Cp': 1.508, 'Cpk': 1.506, 'Pp': 1.579, ...}
cap.to_dict()             # full JSON payload (fields + assumptions + provenance)
cap.view(save="cap.png")  # the canonical capability histogram + spec/limit overlay

Consume to_dict() / summary() from code — never parse report() text.

Next steps