SPC is a method, not a system category: control charts that show whether a process has drifted beyond its normal variation. An MES is software that runs and records production, and it may include an SPC module. The MES supplies context such as order, lot and machine, SPC flags the drift, and the MES can then hold the affected output.
"MES vs SPC" compares two different kinds of thing. You can run SPC on paper with a pencil, as people did for decades, and you can run an MES with no control charts at all.
So the useful question is how they work together, and what you lose by running one without the other.
SPC in one worked chart
Statistical process control separates normal variation, called common cause, from a real change, called special cause. The main tool is the control chart, which Walter Shewhart proposed in a 1924 memo.
Example: a machine turns a shaft to 25.00 mm, tolerance plus or minus 0.15 mm. Every 30 minutes, an operator measures 5 parts, which is a subgroup, and past data gives an average of 25.00 mm and an average range of 0.10 mm.
The constants come from the NIST handbook table for X-bar and R charts, which gives A2 = 0.577 at n = 5 and A2 = 3 / (d2 x square root of n). So d2 = 2.326 checks out: 3 / (2.326 x 2.236) = 0.577.
A range chart runs beside this one, to watch the spread. With D4 from the same table, its upper limit here is about 0.21 mm.
Control limits come from the process itself, and specification limits come from the customer. A process can be in control and still make bad parts, or out of control while every part is in tolerance.
Here is the chart after eight subgroups.
What the MES adds
A chart by itself says "something changed at 11:30". An MES can say what was running at 11:30, which lot of bar stock it used, who was on shift and whether a tool change just happened.
Without that context, finding the cause means walking the floor and asking around. With it, you can filter the chart by tool, material lot or operator, and many special causes show up quickly.
| SPC | MES | |
|---|---|---|
| What it is | A statistical method, often in software | An execution system |
| Main data | Samples of a measured feature or defect counts | Orders, counts, stops, lots, checks |
| Tells you | The process has changed | What ran, when, with what result |
| Who uses it | Quality engineers, trained operators | Operators, supervisors, planners |
| Acts by | Raising a signal | Prompting checks, holding lots, blocking orders |
How the data flows
There are three common setups, and the data is much the same in each.
- SPC module inside the MES. The MES prompts the operator at the right interval, stores the readings and draws the chart.
- Separate SPC software. The MES passes order, part number, machine and lot. Gauges or operators feed measurements, and any rule violation returns to the MES as a signal.
- In-line measurement. A sensor or a vision system supplies every part, and SPC runs on the stream, usually with subgroups formed by time.
Whichever you pick, the signal needs a defined reaction: stop, check, hold or adjust. A chart with no reaction plan is decoration.
Check gauges before you trust any of it. If the measurement system varies as much as the process, the chart mostly shows gauge noise.
Where SPC meets OEE
The quality factor in OEE is a lagging count. It tells you how many parts were bad after the fact, as how to calculate OEE quality shows.
SPC is an early warning that aims to stop those parts being made. Continuing the example, the machine makes 2 shafts a minute and the drift starts just after subgroup 7.
With SPC, the signal comes at the next sample, so at most 30 minutes of output is suspect. With only a final audit at the end of the 6 hour run, 150 minutes of output are suspect.
The MES then makes the sort smaller, since it knows which bins came from that window. See MES traceability for how that link is kept.
Keep the two measures separate. Output rejected later still counts against OEE quality, and if parts are reworked they are not good first time, as first time through explains.
Where the overlap causes trouble
- Limits versus tolerances. An MES quality check often uses pass or fail against the tolerance, which is not SPC. Your 25.07 mm average passes that check and still flags on the chart.
- Reactions in two places. If both an MES rule and SPC software can hold a lot, decide which one wins.
- Sample schedules. Sampling by time, by count and by tool change give different results, so set one rule per feature.
What to start with
- Most of your loss is stops and speed: start with the MES, and add charts later on the features that create scrap.
- Most of your loss is scrap from drift: start SPC on a few critical machines. A spreadsheet or standalone tool can do it if operators are trained.
- Customers want Cpk reports, or you must link drift to lots and tools: you need both.
- Simple assembly with low scrap: you may need neither yet. Track first pass yield first, because it will show when you need more.
Five ways a control chart goes wrong
- Charting everything. Choose features that matter to the customer or that drive scrap, and leave the rest.
- Tampering. Adjusting after every reading that looks high adds variation. Only act on a rule, such as a point beyond a limit.
- Using spec limits as control limits. That removes the early warning, so the process has to be near scrap before anyone reacts.
- Recalculating limits after every drift. The limits then follow the problem, and the chart goes quiet. Recalculate after a confirmed process change.
- Using the wrong constants. The constants above only hold for subgroups of 5, so look up the right ones for your size.
For the wider picture on losses, see cost of poor quality and the scrap rate calculator.
SPC watches one feature, while you still need to know how the whole line performs. Fabrico is a light MES for OEE: it takes stops, speed and counts live from your machines, so availability and performance sit next to your quality count per line and shift. See how Fabrico's light MES for OEE works.
What people ask next
Is SPC part of an MES?
It can be. Many MES products include a quality module with control charts, and others pass data to separate SPC software. The MES modules page shows how quality functions fit in.
What is the difference between SPC and SQC?
SPC watches the process while it runs, so you can react before bad parts appear. Statistical quality control is a wider term that also covers sampling and inspecting finished output. In daily use, many people use the two loosely.
What is a good Cpk?
Your customer's requirement decides, so check the contract or quality manual first. For scale, the NIST handbook puts a centered process at 1.33 at about 64 rejects per million. The 1.16 in the example above works out near 500 per million.
Do I need SPC if I have 100% automatic inspection?
Inspection finds bad parts, and SPC finds the trend before they become bad. If the vision system records a measurement and not only pass or fail, you can chart it and get both. See how to reduce scrap and rework.
Does SPC improve OEE?
It can raise the quality factor by catching drift sooner, and it can cut sort and rework time. It does nothing for stops and slow running, so it pays to know your largest loss first, using the six big losses calculator.
Fabrico is a light MES for OEE. It takes stops, speed and counts from PLCs or IoT sensors, with add-on AI cameras counting output where a machine has no usable signal, and operators tag the cause at the line.