← Back to the atlas
No. 1398Problem statementManufacturing and materials

AI-Driven Anomaly Detection in Component Burn-In & Screening

Source

Source summary

Develop ML screening of electronic-component burn-in time series to detect unusual behaviour within otherwise passing absolute limits. Include lot-relative outlier detection and prediction of 168-hour parameter drift from initial/24-hour measurements, flagging excessive predicted slopes for review/rejection. Evaluate missed defects, prediction MAE and explanations useful to QA inspectors. The source's currents/temperatures illustrate the approach rather than define universal component limits.