How We Work
Every finding has a fix. Every fix has an owner. Every owner has a deadline. Every result gets measured.
The Analysis Framework
No finding without a fix. No fix without measurement. No measurement without accountability. Every analysis follows this structure:
| Problem | What the data shows — specific, quantified |
| Raw Signal | The actual MES/floor data that surfaced this |
| Wrong Assumption | What the plant believed before the data said otherwise |
| Analysis Method | How we got from signal to finding |
| Action Taken | Specific corrective action, who owned it, timeline |
| Measurable Result | Before/after with dates and data source |
Case Study: The Downtime That Wasn't Random
Problem
Line 3 losing 14+ hours per week to unplanned downtime. Standard OEE reports showed availability as the primary loss category, but the Pareto chart hadn't changed in months. Same events, same ranking, same morning meeting discussion.
Raw Signal
MES timestamp analysis on Traksys data revealed that 92.6% of major downtime events were preceded by clusters of short stops — those 30-second to 2-minute interruptions that individually look like nothing. Standard OEE bucketed these into "performance loss" where they disappeared into a rate percentage. The raw event stream told a completely different story.
Wrong Assumption
The plant believed downtime was random equipment failure — unpredictable, requiring better spare parts inventory and faster maintenance response. The morning meeting focused on maintenance response time. Nobody was looking at the short stops because OEE said they were a performance issue, not an availability issue.
Analysis Method
Event sequence mining on raw MES timestamps using Python/pandas. Filtered events under 2 minutes, clustered by time window (30-minute rolling) and equipment ID. Correlated clusters with subsequent major fault events within 4 hours. The pattern was consistent across 6 weeks of data: short stop escalation → major fault, with a median lead time of 47 minutes.
Action Taken
Targeted PM on the shrink tunnel — the equipment generating 78% of the short stop clusters. Changed inspection interval from weekly to per-shift. Added operator short-stop awareness to shift handoff (when clusters exceed 3 per hour, call maintenance for inspection before the major fault occurs). One Point Lesson created and trained across all shifts within one week.
Measurable Result
$400K+ in recovered production annually. Unplanned downtime on the shrink tunnel dropped 61% in the first month. Short stop clusters became the early warning system — not the noise everyone had been ignoring. The fix held through recurrence tracking at 30, 60, and 90 days.
Case Study: The Changeover Everyone Accepted
Problem
Changeovers on powder blending lines averaged 50 minutes. The schedule treated this as fixed. Nobody questioned it because "that's just how long it takes."
Raw Signal
Vorne MES timestamp data showed massive variability — some changeovers at 35 minutes, others at 70+. The average was meaningless because it hid a bimodal distribution.
Wrong Assumption
Changeover time was treated as a constant in production scheduling. The real drivers — LOTO procedure design, product sequencing, and equipment adjustment — were never isolated.
Analysis Method
SMED analysis using DMAIC framework. Classified all changeover activities as internal (must be done while stopped) vs. external (can be done while running). Used Python/pandas on MES timestamps to identify which steps drove the variance.
Action Taken
Redesigned LOTO procedure (16 operators → 5 designated). Resequenced the production schedule to eliminate unnecessary product transitions. Upgraded weigh filler equipment for faster calibration.
Measurable Result
38% reduction in changeover time. 50 minutes → 31 minutes average. Sustained across all shifts. Recaptured 19 hours/week of productive time that had been invisible because "planned" downtime is excluded from OEE.