Key takeaways
Short answer: Control limits and specification limits are the most confused pair in statistical process control, and mixing them up undermines the whole method. Specification limits are the voice of the customer: the boundaries, set by design or the customer, that separate a conforming part from a nonconforming one. Control limits are the voice of the process: boundaries calculated from the process's own data (typically the mean plus or minus three sigma) that describe the range of variation the process naturally produces when stable. They answer different questions — is the part acceptable, versus is the process behaving normally — and they must never be placed on the same chart or derived from each other.
Specification limits define what is acceptable to the customer: the boundaries that separate a conforming part from a nonconforming one. They come from outside the process — from the product design, the engineering drawing, the customer's requirement, or a regulation — and they express what the part must be to be fit for use. A dimension specified as 10.0 plus or minus 0.5 millimetres has a lower specification limit (LSL) of 9.5 and an upper specification limit (USL) of 10.5; anything outside that range is, by definition, a defect. Specification limits are the "voice of the customer." They are fixed externally and have nothing to do with how your particular process happens to behave — the customer requires what the customer requires, whether your process can hold it easily, barely, or not at all. Their job is to draw the line between good and bad product. They say nothing about whether your process is stable, predictable, or in control — only whether a given part meets the requirement. That is a judgement about the product, not about the process.
Control limits describe what the process actually does when it is stable: the range of variation it naturally produces. Crucially, they are not chosen — they are calculated from the process's own measured data, typically as the process mean plus or minus three standard deviations (three sigma). The upper control limit (UCL) and lower control limit (LCL) mark the boundaries within which the process output falls when only common-cause (normal, random) variation is present. Control limits are the "voice of the process": they tell you how the process behaves, derived entirely from how it has actually behaved. A point outside the control limits, or a non-random pattern within them, signals a special cause — something has changed, and the process is no longer behaving as it normally does. Control limits are a property of the process, recalculated when the process genuinely changes, and they exist to answer one question: is the process in a state of statistical control, behaving consistently and predictably? They have nothing to do with what the customer wants — only with what the process does.
The cleanest way to hold the distinction is the pair of phrases: specification limits are the voice of the customer, control limits are the voice of the process. They come from completely different sources and answer completely different questions. Specification limits come from the customer or design and answer "is this part acceptable?" Control limits come from the process's own data and answer "is the process behaving normally?" One is about the product, the other about the process. One is set externally and fixed, the other is calculated internally and changes only when the process changes. This is why the two numbers are generally unrelated: there is no requirement that your control limits fall inside, outside, or anywhere near your specification limits — that relationship is exactly what process capability measures, and it can come out any way. Treating the two as the same thing, or expecting them to coincide, is the root error that SPC beginners make. They are two independent assessments that happen to live in the same units.
The single most important practical rule in SPC is: never put specification limits on a control chart, and never calculate control limits from the specification. A control chart plots process data against control limits to detect special causes; drawing the spec limits onto it invites a destructive habit called tampering — reacting to a point that is outside spec but inside the control limits as if the process had changed, when it has not, and adjusting a stable process that should be left alone. Conversely, treating control limits as if they were spec limits — judging parts good or bad against the UCL and LCL — is meaningless, because control limits say nothing about customer acceptability. The two serve different functions and feed different decisions: control limits drive the decision "should I look for a special cause and intervene?", while specification limits drive the decision "is this part shippable?". Mixing them corrupts both decisions. Keep them on separate analyses: control charts for process stability, capability studies for the relationship between process spread and specification.
Because control (stability) and capability (meeting spec) are independent, a process can sit in any of four states. In control and in spec is the ideal: the process is stable and its natural variation fits comfortably inside the specification. In control but out of spec is the capable-process problem: the process is perfectly stable and predictable, but its natural spread is too wide for the specification, so it predictably produces some defects — this is a process-capability failure (a low Cpk), not an instability, and the fix is to reduce variation or recentre, not to chase special causes. Out of control but in spec is the dangerous illusion: the process is unstable but happens to still be producing acceptable parts for now — a special cause is at work and defects are coming unless it is found. Out of control and out of spec is the obvious crisis. The lesson of the four states is that being in control and being capable are different things: control limits tell you about stability, specification limits and capability tell you about conformance, and you need both pictures.
Suppose a dimension is specified as 10.0 plus or minus 0.5, so LSL is 9.5 and USL is 10.5. The process is centred at a mean of 10.0. In the first case the process standard deviation is 0.1, so the three-sigma control limits are 9.7 and 10.3 — comfortably inside the 9.5-to-10.5 specification. Here the process is both in control (points fall within 9.7-10.3) and capable (its whole natural spread fits inside spec with room to spare); essentially no defects. Now suppose the standard deviation is 0.2. The control limits become 9.4 and 10.6 — wider than the specification. The process can be perfectly in control, every point falling predictably within 9.4-10.6, and yet it is producing out-of-spec parts, because its natural variation spills past 9.5 and 10.5. Same centred process, same in-control behaviour, but the second is not capable. This is the whole point in one example: the control limits describe the process and are calculated from its sigma, while the specification limits are fixed by the customer, and whether one fits inside the other is capability — a separate question from whether the process is in control.
Control charts protect the Quality factor of OEE by catching process shifts before they turn into defects. When a control chart signals a special cause — a point outside the control limits or a non-random pattern — it is an early warning that the process has changed, allowing intervention before out-of-spec parts (and the resulting quality losses) accumulate. Distinguishing special-cause from common-cause variation also drives the right response: hunt down and eliminate a special cause, but reduce common-cause variation through fundamental process improvement rather than tampering. A process that is both in control and capable produces a high, stable Quality factor; one that is out of control produces unpredictable quality losses, and one that is in control but not capable produces a steady, predictable drag on Quality that capability work (improving Cpk) must fix. Trustworthy charts also depend on trustworthy measurement — see Gauge R&R vs calibration.
Fabrico surfaces the Quality losses that SPC is meant to prevent, against live OEE — so when defects start to climb, you see it in the Quality factor and can connect it back to a process that has drifted out of control or was never capable. By making quality losses and their timing visible, it complements the control chart on the floor: the chart catches the statistical signal, and the OEE trend shows the production cost of letting it go unaddressed. Book a demo to connect process control to real quality performance.
Specification limits define what the customer requires — the boundary between a good and bad part, set by design or the customer. Control limits describe what the process actually does, calculated from its own data as the mean plus or minus three sigma. One judges the product; the other judges process stability.
No — this is a classic SPC error. Control charts plot process data against control limits to detect special causes. Adding spec limits invites tampering: reacting to points relative to the spec rather than the control limits, and adjusting a stable process that should be left alone.
Yes. A process can be perfectly stable (in control) yet have natural variation too wide for the specification, so it predictably produces out-of-spec parts. This is a capability problem (low Cpk), not an instability, and the fix is to reduce variation or recentre — not to chase special causes.
Control limits are calculated from the process's own measured data, typically as the process mean plus or minus three standard deviations. They are not chosen or set from the specification — they describe the range of variation the process naturally produces when it is stable.
Control charts catch process shifts before they become defects, protecting the Quality factor of OEE. A process that is both in control and capable yields high, stable quality; one out of control yields unpredictable losses; one in control but not capable yields a steady, predictable quality drag that capability work must fix.
Iniciar sesión
Blog
