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

This paper develops a stacked ensemble EWMA (SE-EWMA) chart for Phase II monitoring of process location. The aim is to give practitioners one calibrated alarm rule that uses both mean based and median based evidence, rather than forcing a prior choice between efficiency under light tailed behavior and stability under heavy tailed or skewed behavior. The two EWMA paths are converted into exceedance evidence scores and combined through a logistic stacking rule. A single threshold is calibrated by Monte Carlo simulation to attain the desired in control average run length while accounting for dependence between the two paths. The simulation study covers shifts, linear drifts, and quadratic drifts under normal, heavy tailed, and skewed reference regimes. The results show that SE-EWMA remains close to the better single EWMA chart across the tested settings, with small average loss relative to the best baseline and stable behavior under distributional changes. After an alarm, a signal conditioned classifier uses a short look-back window of monitoring trajectories to estimate the probabilities of shift, linear drift, and quadratic drift. The industrial screw driving case study shows the full detection and diagnosis workflow. SE-EWMA detects the used workpiece condition, and the post alarm classifier assigns high probability to a shift type disturbance, which agrees with the engineering interpretation.