AI Pattern Detection

Western Electric and Nelson rules are binary — a pattern either triggers a rule or it doesn't. AI pattern detection works on a continuous scale, identifying emerging patterns before they cross rule thresholds. A gradual upward drift might take 20 subgroups to trigger Nelson Rule 5 (6 consecutive trending points) — AI can flag the drift at subgroup 10 with a confidence score, giving you a head start on investigation.

SigmaResolve detects five pattern types: **Trends** (gradual drift using rolling linear regression), **Level shifts** (sudden mean changes using change-point analysis), **Cyclical patterns** (periodic oscillation via autocorrelation), **Mixture** (bimodal distributions suggesting two process streams), and **Stratification** (unnaturally low variation suggesting data issues). Each detection includes the statistical method used, confidence score, and probable process causes.

Every AI finding includes a High/Medium/Low confidence rating. High-confidence findings appear as primary alerts alongside rule violations. Low-confidence findings go to a separate 'Observations' section — visible but not alarming. The system displays the statistical method for each detection: 'Trend detected using linear regression on 20-subgroup rolling window, slope significance p < 0.002.' Quality engineers don't trust black boxes.

Unlike generic anomaly detection, SigmaResolve's AI provides manufacturing-relevant probable causes. A detected upward trend suggests: tool wear, fixture loosening, thermal expansion, material degradation. A sudden shift suggests: new material lot, operator change, machine adjustment, fixture replacement. Causes are drawn from a manufacturing knowledge base — not generic ML vocabulary.