Local RCA engine (4 layers)

How RCA hypotheses are built locally and made auditable.

IntelFactor closes the loop locally:

defect → root cause → corrective action → verified outcome

Layer 1: Defect pattern accumulator

• Stores inspection events locally (rolling window).

• Detects anomaly spikes per defect type, per station.

Layer 2: Process parameter correlator

• Compares anomalies against process parameters (targets, ranges, drift).

• Outputs explainable correlations.

Layer 3: On-device explanation model

• Converts structured context into RCA hypotheses.

• Runs on anomaly triggers, not per-frame.

Layer 4: Closed-loop action recommender

• Produces SOP-linked corrective actions.

• Captures operator feedback.

• Produces verified training triples.