SPC for Precision Machining

• Control charts and capability

  X-bar/R and X-bar/s, Phase I limits from the first 25 subgroups, Cp, Cpk, Pp, Ppk and observed ppm per characteristic, two-sided and upper-limit-only (flatness, true position, surface roughness).

• Run-rule and anomaly detection with labels

  the six Western Electric/Nelson flags are in the file; the true shift is too, so detection rate, false-alarm rate and average run length can be measured rather than assumed. A benchmark for ML detectors (CUSUM, EWMA, isolation forests, sequence models) against a known truth.

• Root cause analysis

  join a signal to the cause active on its machine, the material lot, the operator, the hours since the last tool change and the days since maintenance.

• Tool wear and predictive maintenance

  shaft diameters and surface roughness grow and bores shrink with the hours on a tool, and reset at each change; within-subgroup spread widens as a machine goes longer without maintenance.

• Measurement system analysis

  each gauge has a crossed GR&R study (ANOVA or range method) and a calibration history; readings sit on the gauge's own resolution, so a coarse gauge shows up as it does on a real chart.

• Teaching and dashboards

  realistic enough for an SPC course, a Six Sigma project, a Power BI/Tableau quality dashboard, or a MES/QMS demo database.

• subgroups.active_cause_idThe assignable cause in force when the subgroup was sampled, if any.285 distinct, 95% empty
• subgroups.true_cause_shift_sigmaThat cause's shift of the mean, in within-subgroup sigma.-4.36 to 4.62, mean 0.021
• subgroups.true_tool_wear_sigmaTool wear since the last change, in sigma.-0.6 to 0.6, mean 0.03
• subgroups.true_lot_shift_sigmaThe material lot's hardness effect, in sigma.-0.17 to 0.29, mean 0.0099
• subgroups.true_operator_shift_sigmaThe operator's setup bias, in sigma.-0.32 to 0.21, mean 0.00061
• subgroups.true_driftThe process mean's own slow drift (AR(1)), in sigma.-0.3 to 0.32, mean 0.00016
• subgroups.true_process_meanThe true process mean at that moment, in the characteristic's unit.-0.038 to 156.5, mean 32.57
• subgroups.noise_factorHow much wider the within-subgroup spread is as maintenance ages.1 to 1.2, mean 1.06
• assignable_causes.magnitude_sigmaEach cause's size, with its category, start, end and direction.0.45 to 4.62, mean 1.55

• 5.0% of subgroups were sampled while an assignable cause was active; 94% of the Phase II subgroups under a cause of 1.5 sigma or more trip at least one rule, against 20% of those under none (tool wear, lots and drift still move the mean there, and are labelled).
• Rule 1 false-alarm rate, stable Phase II: 0.78% (theory 0.27%); estimated limits from 25 subgroups and an autocorrelated process put it above the textbook 0.27%, as they do in practice.
• 0.28% of readings are out of specification. 9% of characteristics have Cpk below 1.0 and 65% are at 1.33 or better; 25% are marked critical to quality.

• every key resolves and every primary key is unique; each characteristic's machine and gauge are on its part's line, each operator on the subgroup's line, each lot belongs to the part and was received before it was used;
• every subgroup has exactly 5 readings; X-bar, max, min, range and s re-add exactly from the readings;
• control limits, Cpk and Ppk recompute from the subgroups and readings; all six rule flags recompute with 0 differences;
• each feature is measured on a gauge that suits it, and every reading sits on its gauge's resolution;
• a cause is only recorded on a subgroup sampled inside that cause's window on that machine;
• the gauge R&R trials recover each gauge's stated %GR&R.

• readingsEvery individual measurement, on its gauge's resolution, with out-of-spec flag414,750 rows · 8 cols
• subgroupsOne row per n=5 sample: X-bar, R, s, Western Electric flags, the lot, operator, hours since tool change, and the TRUE cause of any shift82,950 rows · 36 cols
• gauge_rr_trialsA crossed gauge R&R study per gauge: 10 parts x 3 appraisers x 3 trials3,240 rows · 7 cols
• tool_changesEvery insert, drill or wheel change and why2,200 rows · 5 cols
• material_lotsBar-stock heats per part: supplier, received date, hardness536 rows · 6 cols
• assignable_causesEvery special cause: machine, category, start, end, size in sigma, direction, corrective action, CAPA300 rows · 10 cols
• maintenance_eventsPreventive and corrective maintenance with downtime240 rows · 5 cols
• gauge_calibrationsCalibration history with as-found error and result216 rows · 7 cols
• characteristicsThe drawing's controlled features: spec limits, gauge, Phase I control limits, Cp/Cpk, Pp/Ppk, observed ppm158 rows · 40 cols
• operatorsOperators by line and home shift, experience, certification, setup bias72 rows · 7 cols
• partsMachined parts: family, material, sector, drawing revision, annual volume40 rows · 8 cols
• gaugesOne CMM, air bore gauge, micrometer and profilometer per line: resolution, %GR&R, repeatability and reproducibility36 rows · 10 cols
• machinesCNC machines, each with its own capability factor and PM interval36 rows · 7 cols
• production_linesMachining lines and their process family9 rows · 5 cols
• plantsThree plants and their region3 rows · 3 cols

Is this real data?

No. This is synthetic data generated by software. No row describes a real person, company, patient, store, machine or transaction. The patterns are modelled to be realistic and the statistics quoted are measured on these files, but they do not describe any real population or market. Use it for learning, testing, demos, benchmarks and prototyping, not as evidence about the real world. Provided as is, without warranty.

What do I get?

One zip of 8.9 MB: 15 linked tables and 504,786 rows as CSV, a README of what each table holds and how the data behaves, INTEGRITY.json with the 78 audit checks, and RECIPE.json, the Misata blueprint that made these exact rows (seed 20250106).

Can I try it before buying?

Yes. The free preview (110 KB) is a slice of the same data with the keys intact, plus the README, so you can load it and check it fits before you pay.

How was it checked?

78 of 78 checks pass, re-run on the delivered files by an independent script with plain pandas: every key resolves and every primary key is unique; each characteristic's machine and gauge are on its part's line, each operator on the subgroup's line, each lot belongs to the part and was received before it was used; every subgroup has exactly 5 readings; X-bar, max, min, range and s re-add exactly from the readings; control limits, Cpk and Ppk recompute from the subgroups and readings; all six rule flags recompute with 0 differences; each feature is measured on a gauge that suits it, and every reading sits on its gauge's resolution; a cause is only recorded on a subgroup sampled inside that cause's window on that machine; the gauge R&R trials recover each gauge's stated %GR&R.

Can I use it commercially?

Yes: in any project, course, benchmark, demo or product, commercial or not. You may not resell or redistribute the dataset itself as a dataset.

What should I know before using it?

Each characteristic is sampled about three times a day (525 subgroups over the six months) at the machine's own times, not on an exact clock. A rule flag is true on every subgroup where its condition holds, not only on the first: eight points on one side flag the eighth, ninth and so on. design_cpk is what the process was designed for; cpk is what the Phase I data shows. A subgroup sampled before its part's first lot arrived in the window has no lot id (the lot in use came before the window); days_since_maintenance and hours_since_tool_change are empty before a machine's first event in the window. Plants, parts, suppliers and people are invented; no real company, product or person is represented.

Can I get a bigger or different version?

Yes. RECIPE.json runs in Misata Studio or through Misata's MCP server to make a variant, or ask us to build one to your spec.

Want it bigger, in another setting, or with your own columns? Have us build it or make it in Studio.