Process Audit for Precision Parts | Yujiaxin Tech

Process Audit on Precision Parts | Cp and Cpk Prove Tolerances | Yujiaxin Tech

A good sample proves a conforming part can be made, not that the next lot will conform. Sample parts get hand-picked, re-machined or run on a machine dialled in that morning; a lot gets none of that. The process control records a process audit checks are where the evidence for a whole lot starts.

The short answer: a process audit confirms whether one operation holds the drawing tolerance after a shift change, a tool change and a new material lot. It does not confirm that one batch passed inspection. What follows: three documents that must agree, the arithmetic behind Cp and Cpk, the gap to Pp and Ppk, seven questions, and what sampling cannot prove.

Where a Process Audit Starts

Pick one part off the machine and read backwards from it. The path is fixed: lot number, production record, one row of the control plan, one failure mode in the PFMEA, the work instruction revision at the machine, the operator qualification, the parameter log for that shift, one point on the control chart. Any break stops the review.

Breaks cluster at the machine, not in the office. The work instruction is one revision behind, the parameter log stops at the first piece, and the control chart point cannot be tied to the lot number. Any one of those makes the capability number a reference value.

From Drawing Callout to Control Plan Row

The control plan, the PFMEA and the work instruction describe the same characteristic three ways: how it is controlled, what happens when control fails, and what the operator does. Each has to point at the other two.

Special characteristic symbols are where this falls apart. A symbol on the drawing needs the same symbol in the control plan, a matching severity in the PFMEA, and the same mark on the work instruction. When it disappears halfway down the chain, one layer stopped treating it as special.

An empty reaction plan column means the characteristic is not controlled. Someone has to write down who reacts, within how many minutes, how much product gets contained, and what triggers a re-check. A capability figure without that has no disposal path. Route matters too, so fix the process route before the process audit starts.

Reading Cp and Cpk Without Fooling Yourself

Cp is (USL - LSL) / 6σ. The numerator is the width of the specification band, the denominator six sigma of the process. Cp answers one question: is the band wide enough. It says nothing about centring.

Cpk takes the smaller of the two sides, min[(USL - μ)/3σ, (μ - LSL)/3σ]. The nearest specification limit decides. Cp is always greater than or equal to Cpk, with equality only at mid-spec, and the difference is the cost of off-centre running.

The sigma here is the within-subgroup estimate, usually R̄/d2, and it excludes drift between subgroups. A study commonly uses 25 subgroups of 4 or 5 parts, about 100 to 125 data points, across shifts and material lots. Last week's data measures last week.

One-sided characteristics have no Cp at all. Flatness, runout and position carry a single limit, so USL minus LSL does not exist and the index cannot be defined. Report Cpu or Cpl and say which.

Against a normal distribution, Cpk 1.33 corresponds to roughly 63 ppm two-sided and 1.67 to roughly 0.6 ppm. Common thresholds are 1.33 for general characteristics and 1.67 for special ones. The six process families differ in critical characteristics, and what a process audit examines in each of the six processes is a separate review.

Short Term, Long Term and the Gap That Matters

Pp and Ppk use the same formulas with the overall sample standard deviation s, which includes drift between subgroups. When the process is stable and sampling is wide, the two sets sit close. A large gap says drift is significant: shift changes, new material lots, tool changes.

Cpk 1.72 against Ppk 1.31 is not an arithmetic error. The 0.41 gap means two curves do not overlap, and it points at the class of edge condition that was never controlled.

The precondition, and the first thing a process audit checks, is a process in statistical control; a Cpk from out-of-control data cannot support a release. Provenance is separate: measurement capability tells you where the numbers come from, not whether the process is stable. Material and density windows for MIM and powder metallurgy are published in the MPIF standards and resources.

A Process Audit Asks for Documents, Not Opinions

Question one: which row of the control plan covers this tolerance. Answer: the current revision, with a characteristic number, sampling frequency, gauge and reaction plan on that row. Question two: which PFMEA failure mode does it map to. Answer: the same line number, with prevention and detection controls.

Question three: which revision of the work instruction is at the machine. Answer: the paper copy plus the controlled issue record. Question four: which lots produced this Cpk. Answer: raw SPC data, including excluded points and the reasons.

Question five: which sigma was used. Answer: the calculation sheet or the software setting. Question six: what happened when that point went out of control. Answer: the reaction plan record, the containment scope, the root cause and the closure evidence.

Question seven: was this characteristic measured after the last tool change, material lot change and shift change. Answer: the setup record and the cross-shift data. Evidence has grades: a signed, dated record tied to a lot number outranks an unsigned printout, then a screenshot, then a verbal explanation. "We trained everyone" scores partial conformity at best, and a process audit counts that as a gap.

What Sampling Can and Cannot Prove

AQL is an acceptable quality level. A lot sampling plan returns accept or reject and controls the risk of accepting a bad lot. By definition it tolerates a proportion of nonconforming parts inside an accepted lot, so acceptance is not a statement about capability. A first piece that passes proves the setup, not the capability.

Capability needs measurement data that is across lots, ordered in time and unselected. Even 100 percent inspection is not zero escape: inspection has its own error rate, and fatigue plus insufficient gauge resolution both leak. Control limits come from process history, specification limits from the customer, and the two are not interchangeable. Standard numbers and conformity frameworks are published through NIST Standards.gov.

Conclusion

Any one of three findings voids a capability figure: the chain does not connect, the arithmetic is wrong, or the reaction plan is blank. At Yujiaxin Tech a process audit reads whether records match the parts on the floor. It cannot see every real fluctuation, records get written after the fact, and a short visit samples the operation. Passing an audit is not the same as having a capable process.

The next step is not a nicer report but walking the floor with a critical characteristic list and the seven questions in order. Yujiaxin Tech builds its process control on the ISO 9001 and IATF 16949 model, lists special characteristics as separate control plan rows and holds them to Cpk 1.67; how Yujiaxin Tech controls its processes describes that practice.

The fastest test is to bring your critical characteristics and assess the process with us.