Worked 8D case: oval, burred holes and punch breakage in stamping

Scenario

A press shop stamps a steel seat-frame bracket in a progressive die on a 250 t mechanical press: strip 2.5 mm hot-rolled high-strength steel (S420MC), five stations, two pierced holes Ø 8.5 mm +0.10 / 0 (LSL 8.50, USL 8.60 mm), roundness 0.03 mm, burr height max. 0.15 mm. The holes take M8 fasteners at the customer’s assembly line.

In week 41 the customer rejects 37 brackets at assembly: the bolt does not enter, holes look oval and have a heavy burr. In the last four months the tool shop has replaced a broken Ø 8.5 punch five times, each time found only when parts with a missing hole were already in the box.

D1–D3 Team and containment

D1 Team. Emma Clarke (quality engineer, 8D leader), Lucas Martin (tool shop), Sofia Rossi (press shift leader), Mark Chen (maintenance), Anna Weber (supplier quality, for the coil supplier), Daniel Kim (metrology). Sponsor: plant manager.

D3 Containment, started within 24 hours:

• Stock at the customer, in transit and in the plant blocked: 18 400 parts from 9 production lots.
• 100 % sorting of every hole with a go / no-go pin gauge (go Ø 8.50, no-go Ø 8.60) and a visual burr check against a limit sample; an oval hole whose small axis is below 8.50 fails the go pin.
• Clean parts marked with a green dot on the label; certified material to the customer from the next shipment, with the 8D number on the label.
• Press: new punch, die clearance checked, first-off five parts measured with the bore gauge before restart.

Sorting result: 412 of 18 400 parts rejected (2.24 %), with 470 defects in total (one part can have several). Containment and the full 8D are documented in the 8D Report; the CAPA and NCR Tracker holds the actions with owners and dates.

Note the limit of the containment: a pin gauge checks the hole only at its smallest size. It does not measure roundness or burr height, which is why the sorting is backed by the visual burr check and why the permanent control (below) uses a bore gauge.

D2 Problem description: is / is not

Problem statement: Pierced Ø 8.5 mm holes in bracket station 3 are oval, undersize and burred, and the Ø 8.5 punches break in service (5 breakages in four months); defects found by the customer at assembly since week 39.

Pareto of the sorting result

The defects from the sort, as a Pareto chart. Three categories carry about 90 % of the defects, and all three point to the cutting condition of the punch and die rather than to handling.

MSA: can we trust the hole measurement?

Before trusting any capability number, the team checks the gauge: a digital two-point bore gauge, resolution 0.001 mm, which is 1 % of the 0.10 mm tolerance (rule of thumb: at most 5 %).

Type 1 study

50 repeated measurements on a setting ring Ø 8.550 mm by one operator. Mean 8.55078, bias 0.00078 mm, s = 0.00115 mm. With 20 % of the tolerance: Cg = 2.90, Cgk = 2.68, both above 1.33. The bias is statistically significant (50 readings detect even 0.8 µm) but small against the tolerance.

Gage R&R (ANOVA), 10 parts × 3 operators × 3 trials

Ten parts covering the range seen in production, three operators (A, B, C), three blind trials each.

%GRR of 12.3 % of tolerance is in the 10–30 % band that AIAG calls conditionally acceptable; against the process variation it is 8.8 % and ndc = 15, so the gauge separates parts well. The operator effect is statistically significant (p < 0.001) though small, the part × operator interaction is not (p = 0.70). The team accepts the gauge for this characteristic and adds a short instruction on how to centre the bore gauge in the hole. Roundness is measured on the CMM, which has its own MSA.

Capability before

25 subgroups of 5 consecutive parts, one subgroup per hour over three shifts, before any change. Nelson rules 1–3 switched on.

The diameter story is in the gap between Cp 1.89 and Cpk 0.75: the spread would fit the tolerance almost twice, but the mean sits only 0.02 mm above the lower limit and drifts down as the punch wears (a punched hole takes the size of the punch). Expected out of tolerance from the overall sigma: about 16 906 ppm. Because the process is not stable, these indices describe what happened, not a prediction.

D4 Ishikawa (6M)

Brainstorming on the Fishbone Diagram with the press operator and the toolmaker at the press. Causes in bold were confirmed by checks or data; the others were checked and ruled out or found minor.

D4 5 Why: occurrence and non-detection

Two chains in the 5 Why Analysis: why the defect was made, and why it was not caught.

Occurrence: why is the hole oval and the punch breaking?

1. Why does the punch break? It is loaded sideways at every hit and fails in bending fatigue at the shank radius.
2. Why is it loaded sideways? The clearance is uneven (0.05 / 0.20 mm), so the punch is pushed towards the wide side; the same deflection makes the hole oval and the burr one-sided.
3. Why is the clearance uneven? The stripper plate guide bushes are worn (0.08 mm play) and no longer guide the punch.
4. Why are they worn? The die is maintained only after a failure; there is no preventive interval based on the number of hits.
5. Why is there no hit-based interval? The die has no stroke counter and the maintenance plan was set by calendar, not by use. Root cause (occurrence), together with the over-hardened punches from the punch supplier.

Non-detection: why did defective parts reach the customer?

1. Why were the oval holes not detected? Only a go pin was used at the first-off check.
2. Why only a go pin? The control plan specified it.
3. Why did the control plan say so? It was copied from an older part with a ±0.2 mm hole and never reviewed in the PFMEA.
4. Why were parts with missing holes shipped? After a punch breakage nobody knew since when the punch was broken: there is no last-off check and no rule to quarantine back to the last good check.
5. Why no such rule? The reaction plan did not define it. Root cause (non-detection).

Weibull: punch life and change interval

The tool shop log has the hit counts at which the last 15 Ø 8.5 punches failed (about 18 months of records) (taken from the press counter at each change, all failures, no censored units):

89 000 · 129 000 · 172 000 · 178 000 · 184 000 · 191 000 · 208 000 · 209 000 · 212 000 · 215 000 · 218 000 · 218 000 · 227 000 · 278 000 · 314 000

Weibull fit by maximum likelihood: β = 4.31, η = 222 000 hits, B1 = 76 500, B10 = 132 000, mean life 202 000 hits. β well above 1 means wear-out: the failure rate rises with use, so a preventive change before failure is effective.

The team sets the change interval at 80 000 hits, close to B1. From this fit, the probability that a punch survives 80 000 hits is 98.8 %. A visual check of the punch and cutting edge is added at 40 000 hits. These lives were measured under the old, misaligned condition; the interval is conservative and will be reviewed after six months with the new punches (punches changed before failure then need a Weibull analysis with suspended data).

D5–D6 Corrective actions

The clearance value is the tool supplier’s call: it depends on material strength, thickness, punch geometry and the hole quality you need. Tight clearance raises punch load and wear; large clearance gives more burr and rollover.

D6 Capability after

Same plan as before (25 subgroups of 5), over two weeks after all actions, including two punch changes at the counter limit.

The diameter is now centred (mean 8.5506 against a mid-tolerance of 8.55) and stable; the Anderson-Darling test does not reject normality (p = 0.74). Roundness is a one-sided, zero-bounded characteristic; here the data pass the normality test (p = 0.28), otherwise a non-normal method would be needed. Both meet the usual requirement of Cpk ≥ 1.33 with a wide margin; the lower end of the 95 % interval for the diameter Cpk is 2.13.

D7 PFMEA update with AP

The PFMEA for station 3 had no line for punch breakage and rated detection of hole defects as if the hole were measured. Updated lines in the FMEA Worksheet with AIAG-VDA action priority:

Severity stays the same (the effect on the customer has not changed); occurrence falls because of the guided punch and the hit-based change, detection improves through measurement and the in-die sensor. The ratings are illustrative: rate them with your own team and tables.

D7 Control plan update

Build it in the Control Plan; the reaction plan is the part that most often fails in audits, so write who does what, not only “inform quality”.

D7 LPA questions

Questions added to the Layered Process Audit (LPA) for the press line:

1. Is the punch hit counter below the change limit, and is the last change recorded with the clearance check?
2. Are first-off and last-off hole checks for the current shift recorded, with the bore gauge, and inside the limits?
3. Is the lubrication roller wet and the lubricant level above minimum?
4. Is the in-die sensor active (test with the test piece at shift start recorded)?
5. After the last stop for a broken punch or sensor alarm: were parts quarantined back to the last good check?

D7 Lessons learned

• Tool wear is a process parameter: manage dies by hits, not by calendar or failure. Stroke counters were added to all progressive dies with pierced holes of class 1 tolerance.
• A Cp much higher than the Cpk says “centre the process”; for a punched hole that means the punch size and its wear, not the press.
• Never copy a control plan from another part without checking the tolerance; a go pin alone cannot check a two-sided tolerance.
• Every detection method needs a defined reaction including how far back to quarantine.
• Weibull on simple tool-shop records gave a defensible change interval in one afternoon; keep recording hits at every change.
• Lessons transferred to the PFMEA family for pierced holes and to two other dies (read-across).

D8 Closure

Closure criteria agreed with the customer: three months without a complaint, no punch breakage, capability after ≥ 1.33 on diameter and roundness, and all actions verified in the CAPA and NCR Tracker. The team was recognised in the plant meeting and the 8D was closed in week 49 (illustrative). Certified-material marking was stopped after 30 clean shipments with the customer’s written agreement.