Intelligent Production Optimization

Factories lose money in two places: defects caught too late, and stock levels that are either too high or run out and stop the line. This project tackled both with one AI system.

The vision module inspects parts on the line. Cameras photograph components at key points and the model flags anything abnormal on the spot. Because the check happens inline rather than at a QC station at the end, a bad part is removed before more work goes into it.

The stock module uses a BiLSTM network trained on past consumption, production plans, seasonal demand and supplier lead times. It reads the history in both directions, which helps with stock patterns that have both a trend and a cycle.

Every day the system produces a stock forecast with a confidence range and suggests what to reorder before a shortage happens. Both models retrain automatically as new production data comes in.

Everything runs on Microsoft Azure. A React dashboard puts inspection results, stock forecasts and suggestions in one place for plant managers, replacing several separate spreadsheets.