Intelligent Robot Vision Monitoring System
Production deployed monitoring system with ML driven anomaly detection, clip capture, and MSSQL logging, built for real manufacturing constraints.
Problem
Limited visibility into cell behavior and failures.
Manual investigation was slow and inconsistent.
Monitoring needed to be reliable without disrupting cycle time.
Solution
Real time pipeline for detection and state classification.
Anomaly detection and recording engine saving clips on trigger.
MSSQL logging with dashboard friendly schema.
Key Engineering Decisions
Used unsupervised anomaly detection where labeled data was limited.
Separated real time inference from logging and IO paths to avoid blocking.
Optimized hot paths with targeted C++ refactors and SQL indexing.
Results
98% detection accuracy in deployment context.
Latency reduced from seconds to milliseconds through refactors and indexing.
Faster troubleshooting with searchable events and captured clips.