This study aims to examine how embedding interpretable machine learning (ML) within a Lean Six Sigma Define, Measure, Analyse, Improve, Control (DMAIC) programme can accelerate and standardise root-cause analysis from event-level manufacturing execution system (MES) data, addressing the gap between conceptual “DMAIC 4.0” proposals and demonstrable plant-level routines and impact.
A single-case, explanatory DMAIC intervention was conducted on the pacemaker filler of a high-volume beverage packaging line using one year of MES exports. A practitioner-oriented graphical user interface (GUI) operationalised data ingestion, cleaning, merging and repeatable diagnostics. Analyse combined stratification and Pareto logic with interpretable supervised ML (logistic regression, decision trees, random forest, XGBoost) to prioritise loss drivers. Improve translated prioritised mechanisms into an action plan, and Control reused the same evidence backbone through monitoring triggers and structured re-entry.
Event-level MES data were operationalised into repeatable loss and variability indicators, and interpretable supervised ML produced convergent driver rankings to prioritise RCA across operating contexts. Loss exposure concentrates in short-duration disturbances and varies materially by product and crew. Verification suggests an estimated overall equipment effectiveness uplift of 3.3–4.6 percentage points, equivalent to +286 units/h and ∼1.76 million units/year capacity potential at design speed.
Evidence is single-site and MES-only, with limited programmable logic controller/sensor diagnostics, and several post-improvement effects are estimated rather than observed longitudinally.
The study provides a replicable, practitioner-ready pattern, where a GUI lowers the analytics skills barrier and converts MES traces into RCA outputs, action bundles and control triggers.
Unlike prior largely conceptual accounts, the work operationalises artificial intelligence within DMAIC as auditable, repeatable governance, moving beyond isolated model building toward deployable improvement routines.
