An evidence based logistic stacking EWMA chart with post alarm drift type classification

A stacked ensemble EWMA (SE-EWMA) chart is developed for Phase II monitoring of process location. Mean and median EWMA charts are run in parallel, each path is converted to a smooth exceedance evidence score, and the evidence is combined through a logistic stacking rule. A single threshold is calibrated by Monte Carlo simulation to achieve the specified protection while accounting for the dependence between the two base paths. Run length performance is studied under step shifts, linear drifts, and quadratic drifts, across normal, heavy tailed, and skewed reference regimes. Baseline comparisons are reported for the mean EWMA and median EWMA charts. After an alarm, a signal conditioned diagnostic model is applied to a short look-back window of monitoring trajectories and returns class probabilities for shift, linear drift, and quadratic drift. The diagnostic model is trained on simulated labeled signals and uses slope and curvature summaries to keep computation light and to support interpretation. A case study on an industrial screw driving process illustrates the full workflow using the maximum rotation angle at the torque target as the monitored characteristic. The SE-EWMA detects a small upward change associated with a used workpiece condition, and the post signal classifier assigns high probability to a shift type disturbance.