FMEA, SPC and MSA
Why these three
Inspection finds defects after they exist. That is expensive — you have already paid for the material, the labour and the machine time.
These three tools work earlier:
FMEA asks what could go wrong, before you start.
SPC watches the process while it runs, so you see a problem developing rather than a defect produced.
MSA checks that your measurements are trustworthy — because the other two are worthless if they are not.
Required in automotive and aerospace. Useful anywhere with repeated production.
FMEA — Failure Mode and Effects Analysis
A structured way of asking: what could go wrong, how bad would it be, and what are we doing about it?
For each step in a process or each part of a design:
What could fail? The failure mode.
What happens if it does? The effect.
Why might it fail? The cause.
How likely is it? How serious? Would we catch it?
Those three judgements combine into a priority. High priority items get action; low priority ones get recorded and left.
The point is not the number. The point is the conversation — a group of people working through a process asking what could go wrong. That discussion finds things nobody had considered, and it finds them before production.
Two kinds:
Design FMEA — what could go wrong with the product itself.
Process FMEA — what could go wrong in making it.
Where it fails: done once, filed, never revisited. An FMEA that has not been updated since the process changed is a historical document, not a control.
FIGURE 1: THE THREE TOOLS
FMEA — before production
- What could go wrong, how bad, what are we doing about it.
SPC — during production
- Watch the process, catch drift before it becomes a defect.
MSA — underneath both
- Are your measurements trustworthy? Everything else depends on it.
SPC — Statistical Process Control
Watching a process while it runs, to see whether it is behaving normally.
The core idea, without the statistics:
Every process varies. Two identical parts are never exactly identical. That is normal.
Some variation is inherent — the ordinary noise of the process working correctly.
Some variation is a signal — something changed. A tool wearing, a material batch differing, a setting drifting.
SPC distinguishes the two.
How it works: measure samples as production runs, plot them, and set limits based on how the process normally behaves. Points inside the limits are noise. Points outside, or patterns like a steady drift, are signals.
What it gives you: a warning before defects. The measurement drifting towards the limit tells you to act while everything is still good.
The critical distinction: control limits come from how the process actually behaves. Specification limits come from what the customer requires. They are different things, and confusing them is the most common SPC mistake.
A process can be in control and still not meet specification — behaving consistently, consistently wrong.
FIGURE 2: TWO DIFFERENT LIMITS
Control limits
- From how the process actually behaves
- Tell you whether something changed
- Calculated from your own data
Specification limits
- From what the customer requires
- Tell you whether the part is acceptable
- Set by the drawing or the contract
MSA — Measurement System Analysis
Checking that your measurements can be trusted.
The question it answers: if two people measure the same part twice, do they get the same answer?
If not, everything else falls apart. You cannot control a process using measurements that disagree with each other, and you cannot judge a part against a specification if your gauge is unreliable.
What it examines:
Repeatability. The same person, the same part, twice. Same answer?
Reproducibility. Different people, same part. Same answer?
Bias. Does the instrument read consistently high or low?
Stability. Does it stay accurate over time?
Why it is skipped: it is unglamorous, and it feels like checking something that obviously works.
Why it matters: measurement variation is often a surprisingly large share of the total variation being blamed on the process. Companies chase process problems that turn out to be gauge problems.
Calibration is not the same thing. Calibration confirms the instrument reads correctly against a reference. MSA checks whether the whole measurement system — instrument, method, people — produces consistent results in your actual conditions.
How they work together
MSA first. Confirm you can measure reliably. Without this, the other two are built on sand.
FMEA next. Work out what could go wrong and where the risk concentrates.
SPC on what FMEA identified. Control-chart the characteristics that matter, not everything.
That sequence matters. SPC applied to everything is expensive and dilutes attention. SPC applied to what FMEA flagged as high risk is targeted.
FIGURE 3: THE SEQUENCE
MSA
- Can we measure reliably?
FMEA
- What could go wrong, and where
SPC
- Watch the characteristics that matter
Act on signals
- Before they become defects
Where they are required
IATF 16949 — automotive. All three, explicitly, along with control plans and PPAP.
AS9100D — aerospace. Risk management including FMEA-type analysis.
ISO 9001 does not name them. It asks for risk-based thinking and process control — and these are how many organisations satisfy that.
Which means: if you supply automotive or aerospace, these are not optional. Elsewhere, they are tools you use where they pay.
Where they pay outside those sectors
FMEA pays anywhere a process is being designed or changed. The discussion alone is worth the time.
SPC pays where you have repeated production and measurable characteristics. It does not fit low-volume or one-off work.
MSA pays wherever measurements drive decisions — which is more places than people assume.
And a general point: the value of all three is prevention. Their return is invisible, because it consists of defects that did not happen. That makes them easy to under-invest in and easy to cut.
Common mistakes
FMEA done once and filed. It should be revisited when the process changes.
FMEA as a paperwork exercise. Scores assigned to satisfy a customer, without the discussion. The discussion is the value.
SPC charts nobody looks at. Data collected, plotted, filed. If nobody acts on a signal, the collection is waste.
SPC on everything. Expensive, and it buries the characteristics that matter.
Confusing control and specification limits. Covered above, and it leads to both false alarms and missed problems.
Skipping MSA. Then chasing process variation that was measurement variation all along.
The short version
FMEA asks what could go wrong before you start. The discussion is the value, not the score.
SPC watches the process while it runs and distinguishes normal variation from a signal that something changed.
MSA checks your measurements are trustworthy — and everything else depends on it.
Do MSA first, FMEA next, and SPC on what FMEA identified. Applying SPC to everything costs more and finds less.
Chasing quality problems that keep coming back?
Get in touch. It is worth checking your measurement system before your process — a surprising share of “process variation” turns out to be gauge variation.