The Nelson rules are a set of eight pattern tests applied to a control chart to decide when a process has gone out of statistical control, published by Lloyd Nelson in 1984 as a refinement of the earlier Western Electric zone rules. Each rule looks for a specific non-random signature in the plotted data (a single extreme point, a run on one side of the centerline, a steady trend, an oscillation) that is unlikely to occur by chance alone. When a rule triggers, it points to an assignable cause worth investigating rather than the normal common-cause noise every process carries.
A control chart plots a measured characteristic over time against a centerline (the process mean) and control limits set at plus and minus three sigma. Points that stay inside the limits and scatter randomly indicate a stable process. The Nelson rules give you objective criteria for the word "random" so that two engineers reading the same chart reach the same verdict. They are the interpretation layer that sits on top of statistical process control, turning a picture into an alarm. Note that a stable process is not automatically a capable one: whether it meets specification is a separate question answered by process capability (Cp and Cpk).
The chart is divided into zones on each side of the centerline: zone C is within one sigma, zone B between one and two sigma, and zone A between two and three sigma. The eight rules read those zones like this:
Reading a chart well means translating the geometry into a physical hypothesis. A single point beyond three sigma (rule 1) usually means something abrupt happened: a wrong material lot, a broken tool, a slipped fixture. A long run on one side (rule 2) or a creeping trend (rule 3) is the fingerprint of gradual change such as heating, wear, or fatigue, which connects directly to the reliability behavior described by the bathtub curve. Alternation (rule 4) and mixture (rule 8) often mean two things are being plotted as one: two spindles, two operators, two shifts. Once you name the pattern, you have a shortlist of causes to chase.
Nelson's set is a tidied, extended version of the four Western Electric zone rules from the 1956 handbook. Western Electric used one point beyond three sigma, two of three in zone A, four of five in zone B, and eight in a row on one side. Nelson kept the same spirit, adjusted the run length for the "same side" rule to nine, and added the trend, alternation, and stratification tests. In practice most SPC software lets you switch either rule set on or off, and many plants adopt a subset rather than all eight so that the chart is not permanently flashing.
Suppose a machined shaft has a target diameter of 20.000 mm. From baseline data the process mean is 20.000 mm and the standard deviation (sigma) of the subgroup average is 0.010 mm. The control limits and zones are therefore:
Now take these ten subgroup averages in time order (mm): 20.002, 19.996, 20.021, 20.024, 20.008, 20.022, 20.019, 20.023, 20.017, 20.026. Every point is below the 20.030 upper control limit, so rule 1 never fires. But look at points 3, 4, 6, 7, 8, and 10: they sit at 20.021, 20.024, 20.022, 20.019 (zone B), 20.023, and 20.026. Among points 6 through 10, four of the five (20.022, 20.023, 20.026 in zone A or B, plus 20.019 in zone B) fall in zone B or beyond on the upper side, which trips rule 6 (four of five consecutive on the same side in zone B or beyond). The chart is warning of an upward shift of roughly two sigma even though no single reading breached the outer limit. That is exactly the kind of early signal the zone rules exist to catch. Feeding this back into a Pareto analysis of shift causes helps you attack the biggest contributor first.
Every rule you switch on raises sensitivity but also raises the false-alarm rate. For a stable process, rule 1 alone produces a false signal roughly once every 370 points. Each additional rule adds its own chance of a false positive, so running all eight can push the combined false-alarm interval down toward one in the low hundreds of points. On a chart sampled every hour that means spurious alarms several times a shift, which trains operators to ignore the chart entirely. The discipline is to enable only the rules that map to failure modes you actually have, document that choice in your control plan, and revisit it as the process matures.
Nelson rules are a charting and statistics discipline: Fabrico does not run SPC software, compute control limits, or apply these rules for you, and you should not expect it to. What every control chart depends on first is trustworthy, time-stamped production data, and that is exactly what Fabrico provides. As a real-time OEE and production-monitoring platform (including computer-vision monitoring for machines without a PLC), Fabrico captures cycle counts, stops, and quality events at the source so your quality team has a clean, continuous stream to chart. Pair that data foundation with the maintenance context from a CMMS, and a rule-triggered shift can be cross-checked against the MTBF and MTTR history of the asset that produced it. Accurate input is the precondition for any rule to mean anything.
No. Most teams enable a subset, typically rule 1 plus rules 2 and 3, and add the others only when a specific failure mode (such as alternating streams or stratification) is a known risk. Enabling every rule maximizes sensitivity but multiplies false alarms, which erodes trust in the chart.
Both divide the chart into A, B, and C zones and test for the same kinds of non-random behavior. The Western Electric handbook defined four rules; Nelson extended the set to eight, adding tests for trends, alternation, and unusually low variation, and set the "same side" run length to nine points.
Not necessarily. Control limits describe the process voice (its natural variation), while specification limits describe the customer requirement. A rule can trigger while every part is still in spec, which is precisely the early warning value: you can react to a shift before it produces scrap.
Fabrico gives your SPC program the accurate, real-time production data it runs on. Book a demo to see how live OEE and quality-event capture feeds cleaner charts and faster reactions.