Inspection Tells You What Happened. SPC Tells You What Is About to Happen.

Final inspection is a filter. It sorts good parts from bad after the money has already been spent on material, machine time, and labor. Statistical process control does something different: it watches the process while it runs and signals that the output is drifting toward the spec limit before a single part crosses it.

The economics are blunt. A molder running a 30-second cycle in a four-cavity tool makes 480 parts an hour, so a hot runner tip that starts degrading in the morning and is caught at the end-of-shift inspection has produced thousands of suspect parts. Catching the same drift at the next sample point costs 60 parts and 20 minutes of press time. That gap is the business case for SPC.

The Control Chart and How to Read It

The core tool is an X-bar and R chart. At fixed intervals you pull a small subgroup, typically 3 to 5 consecutive parts, measure one critical dimension on each, and plot two things: the subgroup average (X-bar) and the range within the subgroup (R, the largest minus the smallest).

The chart carries two sets of lines, and confusing them is the most common mistake on the shop floor:

  • Specification limits come from your drawing. They say what the customer will accept.
  • Control limits come from the process itself, calculated from the observed variation, and normally sit at plus or minus three standard deviations of the subgroup average. They say what this process does when nothing is wrong.

Control limits are never drawn from the tolerance. A process can sit inside its control limits and still produce parts outside spec, meaning it is stable but not capable. It can also run every part inside spec while the chart says something changed. The two situations demand different responses.

You are looking for patterns, not single bad points. A point outside a control limit is one signal. Seven consecutive points on the same side of the center line is another, usually tool wear or a slow temperature drift. A jump in the R chart with a stable X-bar chart typically means a fixture came loose or one cavity started behaving differently. Cavity-to-cavity variation hidden inside a pooled average is the classic way a molding line fools itself, the same failure that shows up in first mold trial inspection when parts are not bagged by cavity.

Cp and Cpk: Is the Process Even Capable?

Capability indices compare the width of your process to the width of your tolerance. Cp is the ratio of total tolerance to six standard deviations of process variation. Cpk adds centering: it measures the distance from the process mean to the nearer spec limit, divided by three standard deviations.

The practical readings:

  • Cpk below 1.0. The process is producing out-of-spec parts right now. Sorting is your only short-term option.
  • Cpk 1.00 to 1.33. Marginal. Any drift produces scrap. Most automotive and medical customers will reject this.
  • Cpk 1.33. The common contractual minimum for a critical dimension, roughly 63 defects per million.
  • Cpk 1.67 or higher. Comfortable. You can run the process unattended between sample points.

A high Cp with a low Cpk is the most actionable result you can get: the process is tight enough, it is just aimed at the wrong target. That is usually a one-parameter fix, a shim or an offset, and it is free. A low Cp means the variation itself is too wide, and no amount of centering saves it. Then you are choosing between a better process, a better tool, or opening the tolerance, which is where an honest tolerance stack-up analysis earns its keep. Engineers routinely specify plus or minus 0.002 in (0.05 mm) on features where 0.010 in would assemble identically, and every one of those over-tight callouts becomes a capability problem someone pays for.

Which Dimensions Get Charted

Not all of them. Charting 40 dimensions on every part is how SPC programs die: the operators stop measuring, then start backfilling numbers, and the data becomes fiction.

Pick 3 to 6 characteristics per part: the ones that failed in prototype, the ones carrying a load or a seal, the ones with the tightest tolerance relative to capability, and anything the customer flagged as critical to quality. The rest get periodic layout inspection. Marking those features clearly on the print, using the datum structure in GD&T basics, lets a factory in another time zone chart the right things without a phone call.

Add one free variable to every plastic and die cast part: shot weight or part weight. It takes two seconds on a scale, needs no fixture, and moves when material, cushion, or cavity fill changes. It is the cheapest early-warning signal in the plant.

When a Small Manufacturer Actually Needs It

SPC pays when volume is high enough that a delayed detection is expensive, when the process has a physical drift mechanism such as tool wear, electrode erosion, or bath depletion, or when a customer contractually requires capability data. At 200 units a year of a machined bracket, full inspection is cheaper and simpler than charting, and nobody should pretend otherwise.

It also becomes non-negotiable the moment you sell into automotive, aerospace, or medical supply chains, where a PPAP or a validation package requires documented capability studies from a run of at least 30 consecutive parts. That data usually comes out of the pilot production run, not out of prototypes, because prototypes are made under conditions nobody will reproduce.

Starting Without Expensive Software

A spreadsheet is enough for a first program. Run 25 subgroups of 5 parts under normal conditions, calculate the control limits from that baseline, print the chart, and hang it at the machine with a pencil on a string. Operators plotting by hand notice trends that a dashboard hides.

Three things make it stick. Write a reaction plan on the chart: what to do when a point goes out, who to call, whether to quarantine. Run a gauge R&R, because a measurement system consuming more than 30 percent of your tolerance generates false alarms until people stop believing the chart. And never adjust the machine because one in-control point looked off center, which is overcontrol and reliably doubles variation.

SPC pairs with, not replaces, acceptance sampling. Keep your AQL inspection at the container level and your first article inspection at every setup and material change. Attribute defects that are not dimensional, such as the cosmetic issues cataloged in injection molding defects, are better handled against a physical golden sample than against a number.

Set Up a Control Plan That Fits Your Part

Projects House builds control plans for clients entering serial production: selecting the characteristics worth charting, sizing subgroups and sample intervals, running the initial capability study, and writing the reaction plan the factory will actually follow. Send your part drawing and target volume through our contact form.