Write-ups on the real engineering problems behind process monitoring: the bugs, the statistics, and the architecture.
A common, easy-to-miss bug, using population variance instead of sample variance, quietly makes process capability indices look better than they…
Read case study →Most tutorials show EWMA control limits as fixed values. That's wrong for the early samples in any run. Here's the…
Read case study →The engineering discipline that let SPC, anomaly detection and influence scoring share one codebase without turning into spaghetti as it…
Read case study →Moving from historical CSV exports to live PLC tags via OPC UA breaks more assumptions than it seems like it…
Read case study →Industrial process data is heavily autocorrelated, which silently breaks ordinary cross-validation and inflates confidence in feature-influence scores.
Read case study →No single method should be trusted alone for ranking which variables drive an outcome. Here's how a weighted ensemble hedges…
Read case study →All the statistical rigor in a monitoring system is wasted if a floor operator can't act on it in three…
Read case study →