Solder Defect Reduction: Pareto by Count Versus Pareto by Cost
Over five weeks a contract electronics plant logged 600 solder-related defects on 42,000 boards. Counts and average rework cost per event: bridging 212 ($6), insufficient solder 148 ($6), tombstoning 96 ($42), missing component 71 ($18), wrong component 38 ($55), damaged component 22 ($30), other 13 ($20). The plant manager wants the improvement team to "kill the bridging bar" because it is the tallest bar on the Pareto chart. The team has one engineer and eight weeks. Production runs three shifts; the night shift builds 30% of the volume but accounts for 150 of the 212 bridging events. Inspection is 100% visual by operators at each bench, and there is no defect-location data beyond defect type.
1. Rank the defect types by count and by cost. Which single defect should the eight-week team attack, and what is the risk in using the count-based Pareto alone?
Read the worked answer
By count (total 600; defect rate = 600/42,000 = 14,286 ppm):
| Defect | Count | % | Cum % |
|---|---|---|---|
| Bridging | 212 | 35.3 | 35.3 |
| Insufficient solder | 148 | 24.7 | 60.0 |
| Tombstoning | 96 | 16.0 | 76.0 |
| Missing component | 71 | 11.8 | 87.8 |
| Wrong component | 38 | 6.3 | 94.2 |
| Damaged component | 22 | 3.7 | 97.8 |
| Other | 13 | 2.2 | 100.0 |
By cost (total COPQ = 212(6)+148(6)+96(42)+71(18)+38(55)+22(30)+13(20) = $10,480):
| Defect | Cost | % | Cum % |
|---|---|---|---|
| Tombstoning | $4,032 | 38.5 | 38.5 |
| Wrong component | $2,090 | 19.9 | 58.4 |
| Missing component | $1,278 | 12.2 | 70.6 |
| Bridging | $1,272 | 12.1 | 82.7 |
| Insufficient solder | $888 | 8.5 | 91.2 |
| Damaged component | $660 | 6.3 | 97.5 |
| Other | $260 | 2.5 | 100.0 |
Judgement: attack tombstoning. It is only third by count but is 38.5% of the cost of poor quality and costs seven times more per event than bridging, because it scraps the component and usually forces rework on a reflowed joint. The count Pareto answers "where are most of the events?", not "where is most of the loss?" Ranking by count alone misallocates the only engineer for eight weeks, and the top-count item (bridging) is fourth by cost at 12.1%.
Caveat to state in the answer: cost per event is an assumption. Confirm it with the accounting owner, and confirm tombstoning's count is not inflated or deflated by detection differences between benches before committing resources.
2. The manager says the bridging problem is obvious - "it's the night shift, retrain them." Evaluate that conclusion using the data given and name the specific data you would collect next.
Read the worked answer
Test the claim with denominators. Bridging rate by volume:
- Night: 0.30 x 42,000 = 12,600 boards; 150/12,600 = 11,905 ppm.
- Day + evening: 29,400 boards; 62/29,400 = 2,109 ppm.
The night shift rate is about 5.6 times higher. That is a real, sizeable stratification signal - not proof of a training deficiency.
Why "retrain" is not yet a conclusion:
- Confounded variables. Shift co-varies with operator, machine, nozzle set, paste lot, humidity at night, and supervision. Any of these could produce the same split.
- Detection bias. Inspection is 100% visual at each bench. If the night shift inspects more strictly or has better lighting, more bridging is recorded without more bridging existing. You cannot separate occurrence from detection with defect counts alone.
- No location data. Defect type alone cannot distinguish a paste-print problem from a placement or reflow problem.
Data to collect next (one week, one check sheet):
- Board serial, line, machine, shift, operator, and defect location (component designator or grid zone).
- Paste-print volume and stencil-cleaning log by shift; reflow profile/oven zone temperatures by run.
- A short masked audit: one inspector re-inspects 100 night-shift and 100 day-shift boards from the same lots, so you can quantify detection agreement between shifts.
Then run the comparison again on location-coded data. If bridging clusters on one component family, it is a design/print issue, not an operator issue.
3. Resource is limited to one engineer and eight weeks. State how you would sequence the work and what evidence you would accept that the action worked.
Read the worked answer
Sequence (cost-weighted, not count-weighted):
- Weeks 1–2 - measurement check. Confirm the inspector-to-inspector agreement from the masked audit. If disagreement is large, all defect counts, including the Pareto, are suspect and must be re-baselined first.
- Weeks 2–6 - tombstoning. Highest COPQ ($4,032, 38.5%). Typical causes are uneven pad heating, paste offset, component placement offset, or a profile issue; these are diagnosable in a few designed trials rather than by brainstorming.
- Weeks 5–8 - bridging on the night shift only. Second action runs in parallel because it is already well stratified; focus on the specific component/location once the location data are in.
Deliberately defer wrong component ($2,090, 19.9%): high cost per event but low count, and usually a setup/changeover control issue that is cheaper to fix as part of a changeover checklist later.
Evidence of effectiveness (pre-committed, not after the fact):
- A stated metric and baseline: tombstoning events per 1,000 boards, currently 96/42,000 = 2,286 ppm.
- A target and a run-in period: at least 4 weeks or 25,000 boards post-change.
- A before/after comparison on the same lines with the same inspection method, plus a p-chart or two-proportion test rather than comparing two raw counts.
- A verification audit 60–90 days after closure to show the gain held, since the team has eight weeks but the effect must persist.