SPC for Short Runs and Low Volume

Short-run SPC is a set of control-chart methods for job shops and low-volume production that code each measurement against its part's target, so many different part numbers share one chart. Instead of charting the raw dimension, you chart the deviation from nominal (DNOM) or a standardized Z value, which lets you run real statistical process control even when you never make thirty of anything.

Classic SPC assumes long runs of one part number. High-mix, low-volume shops live in the opposite world: a run of 40 pieces, then a changeover, then 25 of something else. By the time a traditional chart has enough data to trust its limits, the job is off the machine. Short-run SPC solves that by charting the process, not the part.

Why does classic SPC break on short runs?

A standard control chart needs a baseline, commonly 20-25 subgroups collected under normal conditions, before its control limits mean anything. That is the fatal mismatch: in a short-run shop the production run is often over before you could ever collect that baseline. You end up with charts that never earn valid limits, so nobody trusts them and the effort dies.

The insight that rescues SPC here is that the machine is usually stable across part numbers even though the dimensions change. A CNC turning center that holds a diameter to a certain spread on part A tends to hold a similar spread on part B. The variation belongs to the process, not to the specific part. If you can strip out the thing that changes between jobs, the target dimension, what is left is the process variation, and that you can chart across many jobs.

What is a DNOM (deviation-from-nominal) chart?

A DNOM chart plots the deviation of each measurement from that part's nominal target: you subtract the nominal (or target) from the reading and chart the difference. Part A's 25.00 mm target and part B's 12.50 mm target both become zero, and a reading 0.02 mm over target plots at +0.02 whether it came from A or B. The chart center line sits at zero and the control limits describe the machine's variation, not any single part's dimension.

DNOM is the simplest and most common short-run method, and it comes with one important condition: it assumes every part on the chart has roughly the same inherent variability (the same process spread). That is a fair assumption when the same machine and setup make parts of similar size and material. Group your parts so that condition holds, and DNOM is clean and easy for operators to plot.

What is a standardized Z chart, and when do you need it?

A standardized (Z) chart goes one step further: it subtracts the nominal and divides by the process standard deviation for that part, so every point is expressed in standard-deviation units. Control limits sit at the familiar plus and minus three regardless of which part is running. Because it normalizes the spread as well as the center, a Z chart can mix parts that have genuinely different variability on one chart.

The trade-off is that a Z chart needs a reliable estimate of the standard deviation for each part before you can plot it, which means some prior data or a trustworthy historical value. DNOM only needs the nominal, which you already have from the drawing. So the rule of thumb is: reach for DNOM when your grouped parts share similar variation, and step up to a Z chart only when you must chart parts whose spreads truly differ.

How do you set up short-run SPC?

Short-run SPC succeeds on the setup work, grouping parts sensibly and coding the data, far more than on chart math. Here is the working sequence.

1. Pick the characteristic that costs money. Same rule as any SPC rollout: chart the dimension tied to real scrap, rework, or customer returns, not everything on the drawing.
2. Prove the measurement first. If the gauge cannot repeat, the chart lies. Settle gauge R&R before a single point is plotted, because coded data hides gauge noise just as well as real signal.
3. Group parts that share a process. Put parts on the same chart only when they run on the same machine and setup with similar variability. This grouping is the decision that makes or breaks a DNOM chart.
4. Choose DNOM or Z. Similar spreads across the group means DNOM. Genuinely different spreads means a standardized Z chart, which needs a sigma estimate per part.
5. Code every reading to its target. Subtract the nominal (DNOM) or subtract and divide by sigma (Z) at the moment of measurement. The operator plots the coded value, not the raw dimension.
6. Build limits from the pooled data. Once you have enough coded subgroups across jobs, compute control limits from that pooled variation, never from the spec. The limits belong to the process now, so they carry from job to job.
7. Write the reaction plan and hold it. A coded point beyond the limits still means investigate: tool wear, a setup miss, a bad material lot. The signal and the response work exactly as they do on any control chart.

What are the limits and gotchas of short-run SPC?

Coding is powerful, and it hides things if you are careless. Watch these:

• Bad grouping poisons the chart. Mixing parts with different true variability on a DNOM chart inflates the limits and blinds you to real signals. If in doubt, split the group or move to Z.
• A part-specific bias disappears. If one part number always runs slightly high because of a fixture quirk, coding to nominal can bury that offset in the pooled noise. Watch each part's points, not just the overall pattern.
• Capability is still per part. The chart tells you the process is stable across jobs, but Cpk is judged against each part's own tolerance. A stable coded chart does not mean every part is capable of its spec.
• Coding by hand invites errors. Subtracting the right nominal for the right part, every time, at the machine, is exactly where manual short-run SPC breaks down. One wrong nominal and the point is meaningless.

For an individual part run so short that even coding does not gather enough points, the individuals and moving range chart is the companion tool: it charts single readings when subgroups are impossible.

Picture a contract machine shop that turns forty different part numbers a week on one lathe, none in runs over fifty pieces. Traditionally that shop has no SPC at all, because no single job lasts long enough to build a chart. With a DNOM chart keyed to each part's diameter target, every job feeds the same chart: the operator plots the coded deviation and the chart accumulates points across Monday's bushings, Tuesday's spacers, and Wednesday's pins. Two weeks in, the shop has a stable, trusted chart on that lathe, and when the tool starts to wear the coded points drift toward a limit and signal before a single part goes out of tolerance. That early warning, on a machine that used to run blind, is the entire payoff.

Where short-run SPC pays off

Short-run SPC is how job shops, contract machining, and high-mix low-volume plants get the benefit long-run manufacturers take for granted: an early-warning signal that the process drifted before the parts go out of tolerance. The method is settled and the statistics are the same century-old Shewhart logic; only the coding is new (ASQ, control charts).

The real barrier to short-run SPC is rarely the math; it is coding hundreds of readings to the right nominal, on the floor, without mistakes, and then trending them across jobs. That is a data-handling job, and it is the one Harmony automates: capture the reading and the part number, and the coding, charting, and cross-job trend happen without an operator doing arithmetic at the machine. See a connected floor in our customer story and pair this with choosing subgroup size to get both the coding and the sampling right.