Warranty Trend on a Redesigned Power Supply
A 14-month-old production line builds a 300 W switching power supply. Field returns are tracked by ship month. Returns per thousand shipped were 6 in months 1-4, 9 in months 5-8, and 14 in months 9-14. The reliability group pulls the returned units: 62% show a failed electrolytic capacitor, and the rest show no reproducible fault. Customer duty cycles cluster at 60-80% load in 45 C ambient. A design change in month 6 lowered the capacitor voltage derating from 80% to 65% of rated voltage to reduce cost and shrink board area. Manufacturing says the change was validated with a 168-hour burn-in at 25 C. Your manager wants a root-cause statement and a projection of returns for the next 12 months before committing to a field campaign.
1. Is the month-6 design change a plausible cause of the rising return rate? Construct the argument and state what evidence would confirm or refute it.
Read the worked answer
Plausible, and the timing supports it. Returns rise in the cohort that shipped in months 5-8 and again in 9-14, which is consistent with a change introduced around month 6 entering the field. The dominant failure mode (electrolytic capacitor) is exactly the part whose derating was relaxed.
Engineering basis. Electrolytic capacitor life scales strongly with applied voltage and temperature. Reducing derating from 80% to 65% of rated voltage raises the field stress on the dielectric and typically shortens life. The 45 C ambient at 60-80% load is a harsher condition than the 25 C burn-in, so the validation did not exercise the field stress at all.
What would confirm it:
- Returns stratified by build date: the failure rate should be near zero for units built before the change and elevated after.
- Capacitor lot and date-code traceability linking failures to the post-change part.
- Bench life test of pre- vs post-change units at 45 C and 75% load showing a measurable life difference.
- Ripple-current and hot-spot temperature measurements on returned units vs the derating assumption.
What would refute it:
- High return rates in pre-change cohorts too (points to a common cause such as a supplier lot or a thermal design issue).
- Failed capacitors showing mechanical damage or a manufacturing defect rather than wear-out.
- Field duty cycles far milder than assumed.
The 38% with no reproducible fault should be treated as a separate no-fault-found stream and not used to explain the capacitor trend.
2. Given the observed return rates and the change timing, outline how you would project returns for the next 12 months and what caveats you would state.
Read the worked answer
Approach.
- Split the population into pre-change and post-change cohorts by build date using traceability, not ship date alone, since ship date lags build date.
- Compute returns per thousand for each cohort at matched age (e.g., returns within 6 months of ship). This removes the age-mixing that inflates the raw monthly numbers.
- Fit a simple reliability growth/decay view: if the post-change cohort shows a rising hazard with age, project forward using the fitted hazard rather than a flat rate.
- Multiply projected age-specific rates by the expected shipment volume for each future month to get expected return counts.
Caveats to state explicitly:
- Return data are truncated: recent cohorts have not had time to fail, so naive month-over-month rates understate the mature rate for the newest shipments.
- Reporting lag: returns arrive weeks after failure, so the last 1-2 months are incomplete.
- The no-fault-found fraction may hide additional capacitor or thermal failures that were not diagnosed.
- Field population is still mostly pre-change units, so the full effect of the change is not yet visible; the projection should be presented as a range, not a point estimate.
- If a field campaign or retrofit is launched, the projection changes and should be re-based.
Decision framing. Present two scenarios (no action vs. campaign) with expected return counts and cost, and recommend the campaign if the projected post-change return rate exceeds the warranty accrual assumption.