Automotive is where SPC stopped being a technique and became a contract. Your customer does not ask whether you use control charts; they ask for the chart, the capability study, the reaction plan and the evidence that somebody acted on the last signal. IATF 16949 requires process studies on manufacturing processes to verify capability (9.1.1.1), the statistical tools to be identified in the control plan (9.1.1.2), and the concepts behind them — variation, stability, capability, over-adjustment — to be understood and applied by the people using them (9.1.1.3). Meanwhile the control plan you signed says which characteristics get charted, at what frequency, by whom.
So the question on a supplier's shop floor is never "should we do SPC". It is "is the SPC we are doing worth anything between audits".
A machined bore, charted properly
Subgroup (5 consecutive parts, sampled hourly)
Limits frozen on the first 18 subgroups. The boring tool starts wearing around subgroup 20: five consecutive averages past the upper control limit, and the six-point rising trend flagged and shaded across subgroups 20 to 25.
Exactly one of these 125 parts misses the tolerance — the fourth part of subgroup 25, at 12.0216 mm. The chart called the process at subgroup 21, twenty parts earlier, on a run of parts that every gauge in the building would have passed. That gap is what you are buying.
Illustrative data — generated to behave like this process, not taken from anyone's plant. The limits, the signals and the capability indices above are computed at page load by the same engine that draws a customer's chart.
This is the whole argument in one picture. The tolerance is ±0.020 mm and the baseline capability is Cpk 1.75 — a process most customers would sign off happily. Of the 125 parts here, exactly one misses the drawing: the fourth part of subgroup 25.
A final-inspection gauge finds that part at subgroup 25 and starts a containment. The control chart puts a boundary around what this process does when nothing unusual is happening, and it says the process stopped doing that at subgroup 21 — four hours and twenty parts earlier, on parts every gauge in the building would have passed. The boring tool is wearing, and you had four hours to change an insert.
Two things to notice, because they are the two things spreadsheet charts get wrong:
- The limits are frozen. They were computed once, from the first 18 subgroups, and they do not move afterwards. If they were recomputed over all 25 subgroups, the drift would widen the limits and drag the centre line up with it — and the chart would show nothing at all. This is the single most common silent failure in a spreadsheet SPC system.
- The signal is a pattern as well as a point. Here the limit rule fires first, at subgroup 21 — but the six-point rising trend is flagged and shaded across subgroups 20 to 25, and that is what tells you the failure mode is a drift rather than a jump. On a slower-wearing tool the trend rule fires first and buys you the extra warning; a chart that only implements "point outside the limits" never gets either piece of information.
What is actually charted
On a typical tier-1 or tier-2 line:
- Machined dimensions — bore diameters, shaft journals, face runout, hole position. X̄-R with subgroups of 3 to 5, sampled hourly or per pallet.
- Torque and force — press-fit force curves, bolt torque, staking load. Usually individuals, because every part is measured and there is no rational subgroup.
- Weld and joining — nugget diameter, spatter counts, ultrasonic amplitude.
- Injection moulding — shot weight, cavity pressure, part weight per cavity. One chart per cavity, always: a 16-cavity tool is 16 processes wearing a single tool number.
- Paint and coating — film build, gloss, orange peel.
- Leak and functional test — flow rate, decay pressure, end-of-line current draw.
The characteristics that carry all this are the special characteristics: the ones with the inverted delta, the customer symbol, the "critical" tag on the drawing. Those are the ones that appear on the control plan with an SPC requirement, and they are where a chart is not paperwork.
The AIAG & VDA manual changed the ground under this
On 1 July 2026 AIAG and VDA published the first harmonised AIAG & VDA SPC manual, replacing the AIAG SPC 2nd edition that had stood since 2005 and consolidating the parallel VDA guidance. If you supply into both North America and Europe, it is the end of keeping two glossaries.
What matters practically:
- Normality is no longer assumed quietly. The manual expects the distribution to be tested before capability indices are quoted, and gives routes for non-normal and time-dependent data instead of pretending the problem away.
- New indices. Cw / Cwk describe within-subgroup variation, and machine capability (Pm / Pmk) is treated properly rather than being borrowed from the process family. Low-volume and equipment-validation cases finally have a named method.
- CUSUM and EWMA are in scope for processes where a Shewhart chart is a blunt instrument — slow drifts, tight tolerances, high-cost parts.
Read that as: the bar for "we have a chart" went up, and the bar for "we understand what our chart assumes" went up further.
Why the usual tooling hurts here
The spreadsheet. It works until the person who built it changes job, the limits recalculate every time a row is added, and nobody is looking at 02:00 on a Sunday. Automotive volumes mean an unwatched drift is a containment exercise, a sort, and a customer complaint — not a scrapped part.
Enterprise SPC bolted onto the MES. Capable, complete, and typically a six-figure programme with a consultant attached. It is the right answer for a plant that has decided to standardise everything, and a poor answer for a supplier who wants one press line charted before the next audit.
Minitab and friends. Superb for a study — capability analysis, gauge R&R, a designed experiment. They are analysis tools, not monitoring tools: nothing in them is watching your line right now, and nothing alerts.
Where this platform fits
Between the spreadsheet and the programme. Concretely:
- Baseline once, freeze, and keep the history. Every limit set is stored with the window it came from, who set it and when. Re-baselining after a legitimate process change is a recorded event, not an edit to a formula.
- All eight Nelson rules, evaluated on arrival, with severity per rule and an e-mail when an alarm fires. Rule 2 (nine on one side) and rule 3 (six rising) are exactly the tool-wear patterns above.
- Signals have a lifecycle. A flagged run is acknowledged, given a cause, and closed by a named person. That log is the reaction-plan evidence an auditor asks for, and it is the part spreadsheets never have.
- Data arrives on its own — CSV export from the CMM or gauge software, or
a
POSTfrom the PLC gateway or test rig. The ingest API ships with a Postman collection generated from the live routes, so a controls engineer can wire it up in an afternoon. - A screen on the line. A read-only shop-floor view for the cell, and a share link for the customer quality engineer who wants to see it without an account.
What we do not do
The other core tools are a different job, and we would rather point at them than blur the edges:
- PPAP and the core tools. We compute Cp, Cpk, Pp and Ppk, but gauge R&R, MSA studies, DOE and the PSW package are not here.
- CUSUM and EWMA. The new manual points at them for drift-sensitive processes; we draw Shewhart charts (X̄-R, X̄-s, I-MR) and c-charts today, so a customer-mandated EWMA scheme needs something else.
- Normality testing and transformations. If a characteristic is bounded at zero, that is yours to know for now.
- MES functions. No scheduling, dispatch or traceability.
If you want all of it under one roof, an enterprise suite is the honest answer. If you want your special characteristics genuinely watched by Thursday, this is the shorter route.
What earlier detection is worth
Do the arithmetic with your own numbers, because the industry averages are worthless to you. Take the cell above: 60 parts an hour, five of them measured, sampled hourly. Scrap value including labour and overhead, €18.
- Found at final inspection. The first out-of-tolerance part appears in subgroup 25. Everything made since the process moved is suspect — four hours of production, 240 parts, sorted, with whatever has already gone downstream to be chased.
- Found by the chart. Subgroup 21. One hour of production is suspect, 60 parts, and the action is an insert change.
At a 15% scrap rate on the sorted parts, the difference is roughly €650 against €160, plus a sort that took somebody half a shift. Scale it by how long your parts sit between the machine and the gauge — in plants where final inspection happens on the next shift, the first line is a whole shift of production, and one event a quarter pays for the line plan several times over.
That is before the containment you did not have to declare, and the customer scorecard you did not have to explain.
Start with one characteristic
Export the last few hundred readings for one special characteristic as CSV and upload it. You map two columns, you pick the window that looks stable, and you have the chart above with your own numbers on it. If you would rather check the maths before trusting anything, the Cp/Cpk calculator and the control-chart rules checker run the same engine, with no signup and nothing stored.