Learning-Driven Optimization with Agentic AI

A closed-loop industrial optimization and decision framework combining machine-learning surrogate models, MILP production scheduling, and a supervisory AI agent for energy-intensive production. The system supports quality-aware and energy-aware production planning under operational constraints, tariff variation, and outages.

Energy-intensive production has to hold product quality while operating inside hard constraints that keep moving: shifting energy tariffs, unplanned outages, and process behaviour that resists exact analytical modelling. Planning built on static assumptions degrades as soon as those conditions change.

Process behaviour is learned as machine-learning surrogate models, which feed a MILP scheduling model that plans production against quality and energy objectives. A supervisory AI agent closes the loop — reading operational state, re-triggering the models, and framing the resulting trade-offs as decisions a planner can act on.

1. 01

   Industrial System

   Energy-intensive production process under operational constraints.
2. 02

   Operational Data

   Process, quality, and energy signals collected from operations.
3. 03

   ML / Prediction

   Surrogate models approximate process and quality behaviour.
4. 04

   Optimization

   MILP scheduling under quality, energy, tariff, and outage constraints.
5. 05

   Supervisory Agent

   Monitors state, re-runs the loop, and explains the trade-offs.
6. 06

   Decision

   A quality-aware, energy-aware production plan a planner can act on.

Led the initiative as R&D Manager — framing the operational problem, defining the modelling and optimization approach, architecting the agentic layer, and directing the work through to a running system.

A live system exists and is operated privately. Access is arranged on request.

No verified outcome figures are published for this project yet. Nothing is estimated or reconstructed here — verified results will be added once they are confirmed.

• Machine Learning — surrogate models
• MILP — production scheduling
• Agentic AI — supervisory agent
• Operations Research

A research output is associated with this project. Its publication status is not yet confirmed and is therefore not stated here.