Control plan for process control: scope and limits

A control plan converts a drawing into a monitored process. For each step it names the characteristic, the specification, the gauge, the sample size, the frequency and the action taken when a value leaves the band. Drop the frequency and the reaction, and what remains is an inspection list — and an inspection list does not stop a press.

The number that decides whether a control plan works is not how many checks it contains. It is how many parts are produced between two consecutive checks, because that is the quantity you will quarantine when one fails. A plan that checks a bend every two hours on a line running 240 parts an hour accepts 480 parts of exposure that nobody wrote down.

This is written for the quality engineer approving a supplier's plan, or writing one for a bracket that goes into a lift frame, a genset canopy or a railway coach. It sets out the fields that carry the load, the arithmetic that sizes the frequency, the reaction plan that still works at 3 a.m., and the line where ISO 9001 leaves off. The standards and compliance hub holds the rest of this cluster.

What a control plan controls, and the four fields that do the work

A control plan controls the process that makes a characteristic, not the characteristic itself. That distinction decides where the engineering effort goes, because three of its seven columns are copied from the drawing — step, characteristic, specification — and four are decisions the supplier has to make and defend: method and gauge, sample size, frequency, reaction.

Two suppliers holding the same drawing can produce plans of very different strength and both be technically conforming. The difference sits entirely in those four columns, which is why a buyer reviewing a plan should read column four onwards and skim the rest.

ISO 9001:2015 is the requirement the document answers to. Clause 8.5.1 requires production under controlled conditions, including the availability and use of suitable monitoring and measuring resources and the implementation of monitoring and measurement activities at defined stages. It does not name a control plan, does not set an interval and does not require a capability index. The familiar column layout comes from automotive advance quality planning practice, and it becomes mandatory only when a purchase order or a sector scheme calls it up. That matters at audit: a supplier without a control plan is not automatically non-conforming to ISO 9001, but a supplier that agreed to one on the order and does not have it is.

The gauge is part of the specification

A tolerance of ±0.1 mm checked with a vernier is not the same characteristic as ±0.1 mm checked with a plug gauge, even though the two plans read identically. The working rule is a tenth: the instrument should discriminate at least one tenth of the tolerance band. For a ±0.1 mm callout the band is 0.2 mm, so the gauge needs 0.02 mm resolution. A digital vernier reading 0.01 mm is at the edge; a plug gauge or bore comparator is comfortably inside it, and a go/no-go gauge removes the reading judgement altogether.

Measurement uncertainty then eats into whichever band is left. A gauge with 0.02 mm uncertainty on a 0.2 mm band consumes 10% at each end, so the process has to run to ±0.08 mm to guarantee the callout. That is the hidden cost of a coarse gauge: it quietly tightens the process.

Sample size is not frequency

A plan that says "5 pieces, 2 hourly" has set two independent decisions and only one of them is about statistics. Five pieces describe variation at one moment. The interval describes how long the process is allowed to run unwatched. Most disputes about control plans are actually disputes about the second number.

Control plan values for a laser-cut, formed and spot-welded bracket

The table below is the operating section of a plan for a 2 mm CRCA bracket of the kind that lands in a genset base frame or a rack chassis. Values are typical; the Ganesh column records what the route can actually hold.

Two rows carry more weight than the rest. Incoming thickness is the check that pays for itself, because a 0.10 mm coil variation moves the flat pattern by 1.5708 × 0.33 × 0.10 = 0.05 mm on every 90-degree bend, so a six-bend part drifts 0.31 mm before anything on the press has changed. And the nugget size scales with the square root of thickness: 4√t gives 4.4 mm on 1.2 mm sheet and 6.3 mm on 2.5 mm sheet. A plan that copies the nugget number from one part to another is controlling a different joint.

Frequency is an exposure decision, not a habit

The arithmetic is one multiplication. Exposure E equals production rate r multiplied by check interval i. At 240 parts an hour with a two-hourly check, E is 480 parts. At one hour it is 240. At 30 minutes it is 120. At 15 minutes it is 60. The reaction plan's containment boundary is the last conforming check, so E is literally the size of the pile you will be sorting at 6 a.m.

The cost side is equally easy to state. A dimensional check on this part takes about 90 seconds at the fixture. Sixteen checks in an eight-hour shift at a 30-minute interval consume 24 minutes, roughly 5% of the shift. That is the trade: 5% of capacity buys a fourfold reduction in the quarantine.

There is a second limit on the interval, and it is set by drift rather than risk appetite. If a punch wears 0.02 mm per 1,000 strokes and the tolerance band is 0.1 mm, half the band is consumed after 2,500 strokes. At 240 strokes a minute that is about ten minutes of press time, far shorter than any sampling interval. This is why tool life on a progressive die belongs to a counter and a planned tool change, not to a sample. The progressive die tolerance article works through how much of that band the die consumes at each station. Sampling cannot control wear that outpaces the sample.

Sampling finds drift, it does not find fallout

Capability and inspection answer different questions, and the numbers separate them cleanly. For a centred, normally distributed characteristic, Cpk 1.00 puts the specification limits at three standard deviations and leaves 2,700 parts per million outside them. Cpk 1.33 puts the limits at four sigma and leaves 63 ppm. Cpk 1.67 puts them at five sigma and leaves 0.57 ppm.

At Cpk 1.33, one part in roughly 15,900 is outside the band. A plan taking five parts per shift sees twenty parts a day. It will not find a 63 ppm escape in any realistic production run; what it detects is the mean moving, which is a different and genuinely useful job. ISO 7870-2 covers this use of the Shewhart chart — its scope is the statistical method of charting values from rational subgroups taken at regular intervals. It does not tell you what to do when a point falls outside a limit, and that omission is deliberate: the reaction is a manufacturing decision.

The consequence for a buyer is blunt. For any characteristic where one escape stops a customer line or creates a safety exposure, the control has to be 100% check or error-proofing, not a sample. A fixture that physically rejects a part bent the wrong way is a 100% control with a failure rate of zero by construction, and it is cheaper than an inspector over a year of production. Capability reduces how often you need to look; it does not license not looking.

The reaction plan: five lines, not one sentence

A reaction plan that says "inform supervisor" is not a plan. It has no authority, no boundary and no restart criterion, so at 3 a.m. the line keeps running. Five lines make it operable.

1. Who stops what. The operator stops the machine. The authority is written into the plan and trained, not delegated upward at the moment of a failure.
2. What is contained. Everything produced since the last conforming check, plus anything in the downstream buffer. The coil or batch identity is recorded, because that boundary is what makes the sort finite.
3. What is measured on contained parts. The same characteristic with the same gauge, at a stated sample size. A different method at containment produces a second argument.
4. What is corrected. The parameter that moved — ram depth, weld current, bath concentration, gas flow — and the person permitted to change it. "Operator retrained" is not a correction; it is a note.
5. How production restarts. A stated number of consecutive conforming parts, three in normal practice and five for a safety characteristic, plus a re-check after the first hour of running.

Where the standard stops, and what substitutes for it

ISO 9001:2015 clause 8.5.1 requires controlled conditions and defines what controlled means; it stops short of prescribing the document, the interval, the sample size, the capability target or the reaction. ISO 22514-1 sets out the general principles of capability and performance and defines the indices, but it does not set an acceptance threshold either: Cpk 1.33 or Ppk 1.67 are customer requirements, and they belong in the purchase order if they matter.

What substitutes for a full control plan depends on the part.

• One-off or first article. A first article inspection with a recorded value against every dimensioned characteristic, signed, beats a plan with frequencies that will never elapse twice.
• Repeat part, no customer format. A route card with in-process checks at the operations that can create the defect: the pierce, the form, the weld, the coat. Fewer checks, placed where the failure originates.
• Safety or fit-critical characteristic. Error-proofing or 100% gauging. Sampling is the weakest of the three options and the most commonly specified.
• Capability question in disguise. When a customer asks "can you hold this", the answer is a study on 30 consecutive parts under production conditions, not a control plan. A plan full of 100% checks is a sorting operation wearing a quality document's clothes.

What Ganesh actually runs

Ganesh manufactures under ISO 9001:2015 and ISO 14001:2015 across five units in Mumbai (Vasai) and Nagpur. Presswork runs on mechanical presses from 20 T to 300 T with 250 T punching, and the press tools, progressive dies, welding fixtures and checking fixtures are built in the in-house tool room — which matters for a reaction plan, because a plan that ends in "modify the tool" needs a tool room that can act inside the containment window rather than a purchase order to a third party.

Forming runs on the 200 T, 5-axis CNC press brake with a 4,450 mm bed and Delem control, to 10 mm. Welding runs on three CLOOS robotic MIG cells and more than 30 manual MIG/MAG, spot and SPM stations. Coating runs seven-tank pretreatment, a 16 ft × 16 ft shot blast chamber and an 18 ft × 14 ft electrostatic powder booth, with DFT read on a gauge and adhesion proved by cross-hatch. PPAP and first article inspection are available, with ±0.5 mm as the typical sheet metal tolerance and tighter on critical features by agreement.

The practical effect is that the checking fixture and the manufacturing fixture come from the same drawing and the same datum. That single fact removes the most common cause of a plan and a first article disagreeing: a checking fixture built to the nominal model rather than to the process.

What to put on the drawing and the purchase order

• The critical characteristics, marked, with the reason — fit, safety, coating — so control effort lands on them.
• The default tolerance standard and its class where nothing is called out; ISO 2768-1 class m is the usual fallback.
• The gauge or method where only one is acceptable, such as a plug gauge, CMM or dedicated fixture.
• The characteristics that need recorded values rather than a tick.
• The traceability unit: coil, heat, batch or serial.
• Whether a control plan is required, in what format, and at what revision.
• The change clause: any change of process, material, sub-supplier or tool re-opens the plan and requires re-qualification.

Dimensioning practice sits in ISO 129-1, which covers the presentation of dimensions and tolerances on technical drawings. A characteristic without a dimensioned value cannot enter a control plan with a number, which is why the drawing review comes before the quality plan and not after it.

When a control plan is the wrong instrument

Below roughly 50 pieces, the plan costs more than the parts and a first article with recorded values serves better. When the defect is created by the design — a hole too near a bend, a flange shorter than the tooling needs, a corner with no relief — no frequency catches it, because every part is wrong in the same way. And when the customer's real question is whether the process can hold a tolerance at all, the deliverable is a capability study, not a document with an interval.

Send the drawing and the characteristics you intend to control, and the manufacturability review and quotation will say which of them the route can hold without 100% inspection. The welding inspection and traceability guide covers the record side once the parts are welded, and the inspection article for repeat stampings covers the sampling frequency on a press. If you would rather start from the part, request a quote with the drawing revision and annual volume attached.