DMAIC case study: leak test failures from 4.2 % to 0.8 %
A worked Green Belt project through Define, Measure, Analyze, Improve and Control: charter, MSA, Pareto, IS / IS NOT, multi-vari, proportion test, DOE, control plan and savings.
This worked example follows one Green Belt project through all five DMAIC phases, with the numbers at every tollgate. The company, the people and the data are invented, but the course of the project is typical for a machining plant. Every step links to the free tool used.
Define: 4.2 % of housings fail the leak test
Since March, 4.2 % of die-cast aluminium housings from line 3 fail the final leak test. Failed housings are re-impregnated or scrapped. The project charter states the goal: below 1.0 % by the end of December. With 400,000 housings a year and €4.50 cost per failure, the expected saving is about €57,600 a year. A SIPOC sets the scope from the casting receipt to the leak tester; the casting process at the supplier stays out of scope.
Measure: first the leak tester, then the baseline
Before trusting the failure rate, the team checks the leak tester. Thirty housings, ten of them known leakers, are tested three times by three operators. The attribute agreement analysis gives a kappa of 0.93 against the standard: the tester is reliable. For the measured leak rate a Gage R&R gives 14 % of the tolerance, acceptable.
The baseline over four weeks confirms 4.1 % failures (412 of 10,050). A Pareto chart of the leak locations shows 71 % at the sealing face, 18 % at the threads and 11 % elsewhere.
Analyze: the clamping, not the casting
The team first suspected porosity from the supplier. An IS / IS NOT analysis points elsewhere: the failures are on fixture 2 of the machining centre, not on fixture 1, and started after a fixture repair in February. A multi-vari study shows the flatness of the sealing face varies most between fixtures, not within a part or over time.
A two-proportion test confirms it: 6.9 % failures on fixture 2 against 1.3 % on fixture 1 (p < 0.001). Flatness on fixture 2 averages 0.038 mm against 0.017 mm on fixture 1 (Welch t-test, p < 0.001). The repaired clamp presses one corner too hard and distorts the face during machining.
Improve: a new clamp and a fixed torque
A small 2² factorial experiment with clamp design (old, new floating) and clamping torque (20, 30 Nm) shows a strong effect of the clamp design and an interaction with torque: the floating clamp at 20 Nm gives the best flatness. The PFMEA is updated with the new cause and controls. A two-week pilot on fixture 2 gives 0.9 % failures; the power calculation showed that two weeks are enough to detect the expected difference.
Control: make it last
Both fixtures get the floating clamp. The clamping torque is set by a fixed-torque tool, and the flatness of the sealing face goes into the control plan with a control chart, one part per hour. The work instruction and the LPA questions are updated. After eight weeks the failure rate is stable at 0.8 %; controlling confirms €58,000 of yearly savings. The control phase checklist is complete and the sponsor closes the project.
What made the project work
- Checking the leak tester before trusting the numbers.
- Letting the data overturn the first opinion (casting porosity).
- Proving the cause statistically before buying the new clamp.
- Putting the result into the control plan, the work instruction and the audit, not only into a presentation.
See the full roadmap in DMAIC step by step and test yourself with the Green Belt practice test.