Intelligent Production Optimization

Manufacturing inefficiency costs come from two places: quality failures caught too late, and inventory imbalances that cause production stoppages. This project addressed both simultaneously with an integrated AI system.

The computer vision module handles visual inspection on the production line: cameras capture components or assemblies at key checkpoints, and the CV model flags anomalies in real time. Unlike a separate QC station downstream, inline inspection means defective components are caught before they consume additional manufacturing resources.

The predictive inventory module uses a Bidirectional LSTM (BiLSTM) network trained on historical consumption data, production schedules, seasonal demand patterns, and supplier lead times. BiLSTM processes time-series data in both forward and backward directions, capturing temporal dependencies that unidirectional models miss — particularly useful for inventory patterns with both trend and cyclical components.

The system generates daily inventory forecasts with confidence intervals, surfacing reorder recommendations before stockout risk materializes. The MLOps framework built around both models enables continuous retraining as new production data arrives — the models improve over time without manual intervention.

All components are deployed on Microsoft Azure with a centralized data management pipeline. The React dashboard gives plant managers a unified view of inspection results, inventory forecasts, and production optimization recommendations — replacing multiple disconnected spreadsheets and manual processes.