AOI defect report optimization
Python reporting for MIRTEC AOI that cut false positives and tied defects to accurate BOM lines for faster analysis.
Role: Process Engineer / Software Developer — Integrated Test Corporation (ITC), 2023–2025
Stack: Python, pandas, AOI machine exports, BOM cross-reference
// Highlights
- Filtered noisy multi-failure reports so engineers saw real defects first.
- Cross-referenced data to append correct BOM line numbers to each finding.
- Persisted only substantive errors to support deeper failure analysis.
// Problem
AOI machines produced dense, multi-failure reports where false positives and unclear line attribution slowed triage and obscured real defects.
// Approach
Implemented Python pipelines that classified and filtered report noise, then joined AOI output to BOM structure so each remaining defect pointed to the correct assembly context.
// Outcome
Triage became faster and more trustworthy; stored errors supported trend analysis without drowning analysts in duplicate or phantom hits.
// AI & orchestration
AOI data is high volume and noisy; the system needed predictable filtering that preserved true failures and stayed explainable to engineers.
Pattern: Define defect classes with engineers → encode stable rules → validate on historical runs → iterate thresholds based on false-positive review.
// Technical notes
- Rule layers tuned with floor feedback to avoid over-pruning edge cases.
- Stable identifiers for correlating repeated failures across batches.
// Metrics
Faster triage by collapsing false positives
Report-to-BOM attribution for each defect