Roasting processes relied on delayed quality checks and operator experience. Operators often discovered quality issues hours later, after large volumes were already in process. Environmental variability (humidity, seasons), raw material variation (moisture content), and uncontrollable factors led to inconsistent quality, high waste, and elevated energy use.
Leveraged existing GE Historian time-series data, MES event data, and new process sensors. Data was modeled by a data scientist to create predictive analytics around roaster performance and final product color.
FrameworX dashboards delivered predictions in real time, recommending operator adjustments or enabling automated control. The models accounted for controllable variables (burner levels, belt speeds, airflow) while recognizing uncontrollable conditions.
Technical Specifications:
ERP: Microsoft D365 (Cloud)
...
Integration: Process sensors, quality lab results, historian, MES events
Predictive analytics applied to historian + MES event data
...
Continuous improvement through iterative model refinement
Improved first-pass quality by 10–15%
...