Food manufacturing runs two quality systems side by side that mostly do not talk to each other. There is food safety — HACCP, critical control points, prerequisite programmes, the audit standards (BRCGS, IFS, FSSC 22000) — and there is process control, which is everything about weight, fill, seal, colour, texture and cost. SPC belongs in both, and is usually properly implemented in neither.
The reason is prosaic. Most CCP monitoring is a paper checksheet on a clipboard by the line, transcribed later by someone in QA, and most net-content control is a set of averages typed into a spreadsheet at the end of the shift. Both "work" in the sense of surviving an audit. Neither tells you anything while the shift is still running.
Net content: the chart that pays for itself
Subgroup (5 consecutive packs, every 30 minutes)
Limits frozen on the first 14 subgroups. A nozzle change before subgroup 15 drops the mean fill by about 3 g: four averages below the lower control limit, and a sustained run below the centre line that completes at subgroup 21.
Nothing here is a legal problem: the T1 limit for a 500 g pack is 485 g and no pack comes near it. The chart answers the other question — how much you are giving away above the declared weight, and whether that number is under control. The centre line is the answer, and it moved.
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 a 500 g pack, sampled five at a time every half hour. The legal question in the EU average-weight system is easy to satisfy: the T1 tolerable negative error for a 500 g pack is 15 g, so the T1 limit is 485 g, and nothing here is remotely near it.
The expensive question is the other one. The centre line of this chart is 501.6 g, so on average every pack contains 1.6 g you are not being paid for. At 1,200 packs per hour on a 16-hour day, that is 31 kg of product given away per line per day. Put your own ingredient cost per kg on that number — it is usually the single largest controllable loss on a packing line, and it is completely invisible to a system that only checks against 485 g.
Which is why the control chart is the right instrument: it is watching what the process does, not whether the pack is legal. On this chart a nozzle change before subgroup 15 drops the mean from 501.6 g to 499.2 g. If you were running to the legal check, it is a non-event. If you are running to a target overfill, your process has moved 2.4 g without anyone deciding to move it — and the chart says so at subgroup 16, half an hour later.
The general rule holds here as everywhere: the tolerance is not on this chart. These points are subgroup averages, and averages of five vary about 2.2× less than individual packs — the lightest average on this chart is 497.5 g while the lightest individual pack behind it is 491.6 g. Drawing 485 g on a chart of averages would make the process look far safer than it is. If you want the legal question answered, ask it of individual packs — that is what Cp/Cpk is for, and it is a different calculation.
What gets charted
- Net content and fill — pack weight, fill volume, drained weight, headspace.
- Thermal process — pasteuriser hold temperature, retort come-up time, F₀, cooking centre temperature.
- Cold chain — chill and freezer temperatures, time above threshold.
- Product attributes — moisture, pH, water activity, Brix, salt, fat, colour, viscosity, texture.
- Packaging integrity — seal strength, torque on caps, leak test results, date-code verification.
- Micro and hygiene — environmental swabs, ATP results, TVC counts (counts, not measurements — so a c-chart, not an X̄ chart).
- Foreign body — metal detector and X-ray rejects per hour.
SPC inside HACCP
The audit standards ask for CCP monitoring, corrective action, and verification that the system works. A control chart is a much better fit for that than a checksheet with a pass/fail box:
- A CCP limit is a critical limit — a cliff edge with a legal or safety consequence. It is not a control limit, and putting them on the same axis confuses two different things.
- What SPC adds is the early warning inside the safe zone: a pasteuriser drifting toward its critical limit over four hours is a maintenance job today and a product hold tomorrow. The checksheet records a series of passes right up to the failure.
- Making the trend visible turns "corrective action" into something you did before the deviation instead of after, which is exactly what a verification auditor is trying to establish.
More on that split on the HACCP and SPC page.
Why the usual tooling hurts here
Paper and clipboards. No trend visible until someone types it up, which is after the shift and often after the product has shipped. Transcription errors are routine. The record is genuinely hard to audit.
Checkweigher software. Very good at the legal question — average weight, T1/T2 counts, reject rate — and usually blind to everything else on the line. It also generally cannot chart your pH, your seal strength or your swab counts, so you end up with one system per instrument and no common view.
Excel. The usual failure mode: limits recomputed each time data is added, one author, no alerts. See SPC charts without Excel.
Where this platform fits
- One place for every characteristic on the line, not one system per instrument. Fill weights from the checkweigher, temperatures from the pasteuriser log, lab results from the QA sheet — same charts, same rules, same screen.
- A line-side display. The shop-floor view is a read-only wall screen that survives being left on for a week. Operators see the chart they are affecting.
- Frozen limits with history, so seasonal or recipe changes are recorded re-baselines rather than quiet drift.
- Alerting when a rule trips, which on a night shift is the entire value.
- Share links for a retailer's technologist or a group QA manager, with no account and no edit rights.
What we do not do
- Food safety management. No hazard analysis, CCP register, corrective-action workflow, traceability or supplier approval. Your food safety management system stays where it is; this charts what it monitors.
- p, np and u charts. Rejects per 1,000 packs with a varying batch size is a u-chart, which is on the list; straight counts per equal-sized interval (c) work today.
- Statutory average-weight records. We chart net content; the packers' legal paperwork stays with the system that produces it now.
- Checkweigher integration out of the box. A CSV export or an HTTP post is all it takes; a proprietary-only protocol needs something in between.
- Lab results management. This is not a LIMS.
Where to start
Take one filling line's checkweigher export for the last month and upload it — subgroup it as you sample it, freeze the limits on a good week, and read the average overfill straight off the centre line. Most people find the giveaway number before they find the quality problem, and that is usually the number that buys the subscription.