Overview
The Process Optimization Agent (POA) is a key digital tool developed within the TRINEFLEX project to improve industrial energy efficiency and operational performance. It supports process optimization through case-specific functionalities and enables process flexibility by dynamically adjusting optimization objectives in response to external signals, such as electricity price variations.
By re-weighting the objective function in real time, the POA identifies cost-optimal operating strategies while ensuring compliance with process, quality, and regulatory constraints. The tool has been demonstrated in two industrial use cases: VERALLIA and EYDAP.
VERALLIA – Glass Feeder Optimization
In glass container manufacturing, the feeder process conditions molten glass before forming. Temperature control via gas burners is essential to guarantee glass homogeneity and product quality.
For VERALLIA, no significant energy flexibility was identified, as temperature deviations would directly compromise product quality. Therefore, the POA focuses on minimizing energy consumption while strictly maintaining thermal and homogeneity targets.
A high-fidelity CFD-based Digital Twin by LCM represents the glass feeder thermal behaviour, capturing key heat transfer mechanisms and interactions between heating zones, glass flow, and burner operation. Glass homogeneity is evaluated using temperature data aligned with the 3×3 outlet measurement grid.
The POA determines energy‑efficient heating‑zone temperature setpoints by iteratively interacting with the Digital Twin, identifying operating points that satisfy homogeneity and process constraints while minimizing energy consumption. The resulting setpoints are provided directly to the Decision Support System (DSS) as recommended operating conditions for plant operators.
EYDAP – Wastewater Aeration Flexibility
In the EYDAP use case, the POA optimizes the aeration process of a wastewater treatment plant, where micro-organisms decompose organic matter using oxygen supplied by aerators. The long hydraulic retention time (≈20 hours) allows for flexible operation of the aeration system.
A linear model estimates dissolved oxygen (DO) levels and biochemical oxygen demand (BOD) across five aeration zones, ensuring DO concentrations remain above 0.5 mg/L and that legislative limits at the outlet are respected.
In contrast to VERALLIA, EYDAP shows substantial energy flexibility potential, estimated at several hundred kWh per hour (around 20% of nominal aeration consumption). Over a 36-hour horizon, the POA generates alternative energy consumption scenarios by applying penalty factors to specific hours in the objective function, enabling load shifting during high electricity price periods.
The optimization is implemented in Pyomo and solved using Gurobi. Selected aerator setpoints are validated with LCM’s activated sludge model and then transmitted to the DSS via the Energy Management System (EMS).
Conclusions
The POA demonstrates strong adaptability across industrial sectors: it enhances energy efficiency under strict quality constraints at VERALLIA, while enabling meaningful demand-side flexibility at EYDAP. This confirms the POA as a versatile and scalable optimization tool within the TRINEFLEX framework.
