Process Capability for Beauty Appliances: Reading Cpk and SPC Reports

2026-11-04 · By the Aolemon export team

A 200-piece sample order arrives in perfect condition, so the buyer releases a 20,000-piece purchase order. Four months later the returns report shows 6 percent of units heating unevenly, and the supplier answers that every shipment passed inspection. Both statements are true, and that is exactly the problem. Conformity, acceptance and capability are three different questions, and a sample order only answers the first two. The third one decides what the next ten containers look like.

Acceptance sampling is a filter, not a forecast. Under a routine general inspection level II plan at AQL 2.5, a lot containing several percent nonconforming units still has a realistic chance of being accepted, because the plan is built around a consumer risk the buyer agreed to. Passing inspection therefore means this lot was probably not catastrophic. It says nothing about whether the process sat close to a specification limit or far from it. A factory running with the mean one sigma inside the limit and a factory sitting three sigma inside the limit can produce identical inspection results on the same day, and very different results across a year.

Capability work starts by naming the characteristics that actually drive returns and giving each one a number. For a home IPL that means fluence at the window in joules per square centimetre with a tolerance, contact temperature after ten minutes of operation, runtime to eighty percent of initial output, and the step at the handset seam. For a high-speed dryer it is outlet temperature, airflow and sound pressure at thirty centimetres. For a percussion massager it is stroke amplitude and stall torque. Every line needs a nominal value, an upper and lower limit, a measurement method and the gauge used. A supplier spec that reads 36 W tells a buyer almost nothing, because power draw is an input, not a characteristic the customer experiences.

Then split the list by data type, because the statistics do not transfer between them. Continuous characteristics such as temperature, fluence, runtime and sound pressure can carry a capability index. Attribute characteristics such as scratches, missing accessories, moulding flash or a dead unit cannot, and a Cpk printed next to an appearance defect is a reasonable sign the report was assembled rather than measured. For cosmetic and functional go or no-go items, ask for defects per million by category plus limit samples signed by both sides, and agree in writing which defect class counts as critical, major and minor.

Reading the index itself takes two minutes once you know what each one is for. Cp compares the tolerance width against six sigma of the process and ignores where the mean sits. Cpk also accounts for centring, which is why a process can show a healthy Cp and a failing Cpk when the mean has drifted toward one limit. One-sided characteristics, such as a maximum surface temperature or a maximum noise level, must be reported with CPU or CPL rather than a two-sided formula. The payoff is easy to state: a centred process at Cpk 1.33 expects about 63 nonconforming parts per million, Cpk 1.00 about 2,700, and Cpk 0.67 roughly 45,000. On a 100,000-unit order, moving from 1.00 to 1.33 removes roughly 260 defective units before anyone reaches the returns desk.

Numbers are only worth the data behind them, so ask how the study was built. A credible study covers at least 100 readings, usually 25 subgroups of four or five pieces, collected across shifts, operators and material lots rather than in one quiet afternoon. The process has to be in control before the index is quoted, because a capability number from an unstable process describes a moment rather than a process. The measurement system must be qualified first as well: if gauge repeatability and reproducibility consumes more than 30 percent of the observed variation, the Cpk is partly a measure of the gauge. Watch the gap between Ppk, which uses overall variation, and Cpk, which uses within-subgroup variation. A wide gap means the process drifts between subgroups, and that is usually the finding worth acting on.

Finally, put all of it into documents the buyer can enforce. The quality agreement annex should list each critical characteristic with its limits, method, gauge, check frequency and capability target; the control plan should name the reaction when a point falls outside the limit, including who has authority to stop the line. A workable threshold set is Ppk at or above 1.67 for a new programme and Cpk at or above 1.33 in routine production, with anything below 1.00 triggering containment and sorted inspection at the supplier's cost. Add monthly capability and control chart submission, written notice before any change to components, process, machine or production site, and the right to audit the raw data rather than the summary slide. Asking for this before the purchase order costs a paragraph of work. Asking for it after the returns start is a negotiation the buyer usually loses.

← Back to all articles